THE UNIVERSITY OF MANCHESTER

Transcription

1 THE UNIVERSITY OF MANCHESTER SCHOOL OF MATERIALS XCT Analysis of the Defect Distribution and its Effect on the Static and Dynamic Mechanical Properties in Ti-6Al-4V Components Manufactured by Electron Beam Additive Manufacture A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy in the faculty of Engineering and Physical Sciences Author: Sam TAMMAS-WILLIAMS Supervisors: Prof. Philip B. PRANGNELL Prof. Iain TODD 2015

2

3 Contents List of Figures List of Tables Abstract Declaration Copyright Acknowledgements Glossary Introduction 29 2 Literature Review Additive Manufacture Principles and Development of AM Advantages over Traditional Manufacturing Metal AM Processes Ti-6Al-4V Introduction to Titanium and its Metallurgy Machining Titanium Theβ α Transformation and Microstructure Development Influence of Lamellar Microstructure on Mechanical Properties Selective Electron Beam Melting of a Powder Bed Selective Electron Beam Melting Hardware Advantages of SEBM Electron Beam Melting Process Typical Microstructures Static Mechanical Properties Fatigue of Titanium Fracture Mechanics Stages of Fatigue Crack Growth Fatigue of SEBM Ti-6Al-4V Stress Concentrations Quasi-Cleavage Facets Crack Propagation Porosity and Defects in Powder Bed AM

4 CONTENTS Typical Appearance and likely Origins of Internal Defects Influence of Beam Energy Influence of Melt Strategies Quantification of the Defect Population in SEBM Pore Formation in Energy Beam Welding Movement of Gas Bubbles within Melt Pools Surface Roughness Powder Production Techniques Hot Isostatic Pressing Summary of the Existing Literature Experimental Methodology Arcam Operating Procedure S12 Components Build Preparation Electron Beam Control Samples Manufactured Geometry Melt Strategies Hot Isostatic Pressing Material for Mechanical Testing Alternate Powder Feedstock Mechanical Testing Tensile Testing Fatigue Testing Metallography Optical Microscopy Surface Profile Measurement Electron Microscopy β Grain Reconstruction X-ray Computed Tomography Principles Potential Sources of Error Macro XCT High Resolution XCT Data Down-sampling Interrupted Fatigue Testing Capability of Tensile Rig to Maintain Consistent Load D Image Analysis

5 CONTENTS 3.6 Finite Element Modelling Stress Concentrations within Idealised Geometries Modelling of Real Pore Geometries from XCT Data Modelling of Fatigue Cracks Observed by XCT Defects and their Origins in SEBM Defects within a Standard Sample Overview of XCT Data Pore Size Distributions Pore Morphologies Pore Alignment Spatial Distribution of Pores in the x-y Plane Surface Roughness Repeatability of Measurements Influence of Threshold Value on XCT Image Analysis Comparison of XCT Scans of Identical Samples Sample Manufactured with Previous Control Software Overview of XCT Data Pore Size Distributions Pore Morphologies Spatial Distribution of Pores in the x-y Plane Measurements from the Powder Feedstock Current Plasma Atomised Powder Gas Atomised Powder Discussion Gas Pore Formation Lack of Fusion Pore Formation Tunnel Defect Formation Variation Between Modern and Older Arcam SEBM Methodologies Accuracy and Consistency of XCT Results Surface Roughness Summary Melt Strategies and Sample Geometries Modifications to Standard Melt Strategies Electron Beam Melting Settings Effect of Process Modification on Defect Distribution Influence of Beam Speed, Offset and Focus on Pore Volume Fractions

6 CONTENTS 5.2 Effect of Sample Geometry and Location in Build Chamber Influence of Wall Thickness Influence of Sample Geometry and Orientation Influence of Sample Geometry and Location in the Build Chamber Relationship between Defect Population and Microstructure Variation Discussion Influence of Energy Density Effect of the Different Melt Strategies Influence of Sample Geometry and Location Summary External Changes to SEBM-AM Methodology Influence of Powder Feedstock on Porosity Powder Analysis Comparison of Porosity Observed in Built Samples and Powder Feedstock Effect of Post Manufacture HIPing on Porosity Coarse Pores in Whole As-Built Samples Fine Porosity within Machined Cylinders Discussion Influence of Powder Feedstock on Porosity Removal of Porosity by Hot Isostatic Pressing Summary Effect of Porosity on Mechanical Properties Mechanical Properties of the SEBM Samples Tensile Testing Tensile Fracture Surface Analysis Fatigue S-N Data Fatigue Fracture Surface Analysis Curve Fitting to Observed Fatigue Life Prediction of Fatigue Crack Initiation Stress Concentrations within Idealised Geometries Initial Sorting Algorithm FE Analysis of Pores most Likely to Initiate Cracks Comparison Between Predicted and Observed Crack Initiation Effect of Local Stress/Strain Distribution on Total Fatigue Life Characterisation of Fatigue Crack Growth

7 D Characterisation of Crack Morphology Measured Crack Growth Rates Effect of Porosity on Stress-Strain Concentration around Cracks Influence of Grain Orientation on Crack Growth Discussion Comparison to Conventionally Manufactured Ti-6Al-4V Samples Fatigue Crack Initiation Fatigue Crack Propagation Summary Conclusions and Potential for Further Work Conclusions Further Work References 317 Appendix: MATLAB script to sort the XCT data 319 This thesis contains words 7

8

9 List of Figures 2.1 Schematic diagram of the stereolithography process Conversion from a CAD model to slices for AM Relationship between build rate, power, and feature definition Schematic diagram of the selective laser melting process Schematic diagram of direct laser fabrication Schematic of electron beam freeform fabrication (EBF 3 ) Specific strength with temperature Crystal structures of theαandβphases Burgers orientation relationship Effect of alloying elements on theβ-transus temperature Optical micrographs of lammella microstructures in Ti-6Al-4V Continuous cooling diagram for Ti-6Al-4V Martensitic structures in Ti-6Al-4V Bi-modal and equiaxed microstructures in Ti-6Al-4V Influence of slip length on mechanical properties Schematic of SEBM apparatus Electron penetration range Relative electron beam intensity profile Near-infrared thermal images taken during the SEBM manufacturing processing Scanning strategy used for SLM and SEBM Experimental measurements and model predictions of melt pool size Finite difference model predictions of melt pool shape Experimental and modelling results of melt pool in powder Reconstructed β grains and texture within bulk material Reconstructed β grains within walls of various widths Plan view of reconstructedβgrains in SEBM walls of various widths Nucleation of columnar grains from partially melted powder particles Typical microstructure within bulk SEBM Ti-6Al-4V Evidence ofα in SEBM Ti-6Al-4V Variation in α lath thickness and tensile properties with distance from the base plate

10 LIST OF FIGURES 2.31 Diffusion of baseplate alloying elements into sample Build temperature effect on yield strength and microstructure SEBM mechanical properties compared to SLM Modes of fracture Effect of stress intensity factor range on crack growth Schematic of different regimes of fatigue crack propagation Idealised stage II crack growth by plastic blunting and sharpening Fatigue life of SEBM titanium Fatigue life of SEBM titanium Fracture surfaces showing pores at fatigue crack initiation site Fracture surfaces showing microscopically smooth facets at fatigue crack initiation site Crack growth rate in Ti-6Al-4V samples manufactured by SEBM Influence of spheroid geometry on stress concentration Stress concentration due to pores proximity Stress concentration due to pore depth from a free surface The duality of fatigue data found by separating the location of inclusions Cross-colony fatigue crack initiation Slip band model to explain basal facets Schematic comparison of crack growth behaviour for both small and large crack Crack fronts at various numbers of fatigue cycles measured by XCT Schematic of surface crack growth behaviour Development of plastic zone around crack Influence of pores on fatigue crack growth rate in laser welded Ti-6Al-4V Examples of pores in SEBM Experimental and lattice Boltzmann modelling images of tunnel defects Lattice Boltzmann modelling of the formation of tunnel defects Influence of process setting of porosity volume in SEBM Schematic diagram of the influence of focus offset on melt pool geometry Processing window of Ti-6Al-4V processed by SEBM Balling and delamination defects in SEBM Overview of pores in cross sections with different settings Histogram of pore sizes within structurally optimised cantilever beam Histogram of pores sizes with various line offsets and focus offsets Schematic diagram of keyhole and melt pool formed during electron beam welding Calculated growth of bubbles in a melt pool due to hydrogen diffusion External surface of SLM and SEBM samples

11 LIST OF FIGURES 2.67 Images of powder particles containing gas Schematic diagram of a HIP operation Remnants of pores following HIPing Photographs and schematic diagram of Arcam S Photographs of the powder recovery system Photographs of the powder bed during SEBM process Schematic of the melt strategy employed by Arcam Effect of beam current on beam speed for a range of speed functions Increase in speed when starting a new hatch pass Increase in speed when melting overhanging sections Dimensions of samples manufactured to examine effects of wall thickness on defect population Dimensions of the samples manufactured to examine effects of geometry of defect population Dimensions of samples manufactured to investigate effect of geometry and sample location Dimensions of tensile samples Example of tensile sample Dimensions of fatigue samples Schematic diagram of XCT process X-ray generation and energy spectra X-ray photon attenuation XCT reconstruction of 3D volume from projections Image magnification and blurring Nikon Metris custom bay XCT system Zeiss Xradia Versa 500 XCT system Diagram of the rig developed to hold samples under load during XCT scanning Steps taken to load fatigue samples prior to XCT scanning Load and stress relaxation of the tensile rig developed for the project Principal axis of an ellipsoid Example of theoretical pore modelling set up Example of theoretical pore modelling set up Creation of finite element mesh from voxel data Creation of tetragonal elements from orthogonal voxel data Photograph of standard sample used to characterise defects Examples of XCT data sets obtained Pore size frequency distributions

12 LIST OF FIGURES 4.4 Low resolution examples of stitched optical images of porosity Examples of typical pores seen in SEBM imaged by SEM Examples of typical pores seen in SEBM imaged by SEM Pore aspect ratio distributions obtained from the standard sample Pore aspect ratio against equivalent diameters obtained from the standard sample Orientation distributions of irregular pores Variation in porosity volume fraction in the x-y plane Component surface as-built by SEBM imaged by SEM View of upper surface of a standard sample Profile of surfaces of a standard cuboid Histogram of grey values within XCT data Variation in detected volume fraction porosity with threshold value Variation in detected volume fraction porosity with depth for various threshold values Histogram of pore sizes detected in two standard identical samples Variation of pore volume fraction in x-y plane in two identical samples Examples of pores found in sample Gs Pore size distributions in sample Gs Pore aspect ratios against equivalent diameters in sample Gs Variation in pore volume fraction with depth in sample Gs Visualisation of XCT data showing pores within powder particles Size distributions in Arcam supplied powder The effect of melt strategy modification on melt pattern The effect of the turning function on beam velocity and energy density Visualisation of all the pores detected in samples with modifications to melt strategy View of top surface melted layer Influence of the speed function, line offset and focus offset on detectable pore volume fraction Photograph of samples manufactured to examine effect of wall thickness Variation in pore volume fraction with sample width Visualisation of all the pores detected the 10 mm width wall sample including large tunnel defects Spatial distribution of pores with height Periodogram of pore volume fraction with height Effect of wall thickness on surface roughness Tunnel defects observed in wall manufactured with only hatching

13 LIST OF FIGURES 5.13 Visualisation of all the pores detected in samples with various geometries and build directions Variation in pore volume fraction due to sample geometry and build direction Example slice of XCT data showing tunnel defects in proximity to sample surface Visualisation of all the pores detected in samples manufactured to examine the effect of location in build chamber Variation in pore volume fraction from the sample surface for samples of different geometries Effect of an overhang on pore volume fraction Effect of wall thickness on pore volume fraction pore volume fractions plotted against energy density Hatching pattern when melting multiple models Powder size distributions Variation in pore volume fraction due to sample geometry and build direction D visualisation of defects before and after HIPing Slices of XCT data from samples as-built and after HIPing Pore volume fraction in as-built samples and after a HIP cycle Slices of XCT data from samples as-built and after HIPing Engineering stress-strain curves Variation of tensile properties with test temperature True stress-strain curves Tensile fractures surfaces Fatigue cycles to failure against maximum stress Fatigue fracture surface of sample tested at 575 MPa Fatigue fracture surface of sample tested at 760 MPa Examples of fatigue crack initiation Logarithmic plot of fatigue results Logarithmic plot of fatigue cycles to failure against predicted stress intensity factor Fatigue data Visualisation of stress around two pores Tensile stress concentration generated by a void at different distances from the surface Tensile stress concentration due to the interaction of two voids Visualisation of stress around two pores

14 LIST OF FIGURES 7.16 Stress distribution within the fatigue test sample Histogram of pore sizes within fatigue sample Method used to identify surface porosity from the XCT data Fatigue crack initiation location Results of fatigue testing Crack growth imaged by XCT Crack growth projected onto the x-y plane Crack diversion imaged by XCT Crack growth rate against change in stress intensity factor for SEBM samples Stress and strain distribution around a fatigue crack XCT data showing interaction of fatigue crack with residual porosity Stress and strain around a fatigue crack with and without porosity near tip Grain orientations measured by EBSD along fatigue crack profile Comparison of SEBM fatigue data to conventional processing routes

15 List of Tables 2.1 Summary of different metal AM processes Static mechanical properties of SEBM Ti-6Al-4V Default Arcam beam settings Arcam control constants used Theme modifications Metris custom bay XCT settings XCT data acquisition settings used for high resolution scans Summary of average pore statistics Quantification of pores detected in two identical samples Summary of average pore statistics in sample Gs Electron beam settings used to manufacture samples with process modifications Porosity quantification from samples manufactured to analyse the effect of geometry Summary of powder size statistics Quantification of fatigue fracture initiation sites Quantification of the pores that led to fatal fatigue cracks

16

17 Abstract The University of Manchester Doctor of Philosophy (PhD) November 11, 2015 XCT Analysis of the Defect Distribution and its Effect on the Static and Dynamic Mechanical Properties in Ti-6Al-4V Components Manufactured by Electron Beam Additive Manufacture Samuel Tammas-Williams Selective electron beam melting (SEBM) is a promising powder bed Additive Manufacturing technique for near-net-shape manufacture of high-value titanium components. An extensive research program has been carried out to characterise in 3D the size, volume fraction, and spatial distribution of the pores in model samples, using X- ray computed tomography (XCT), and correlate them to the SEBM process variables. The average volume fraction of the pores (<0.2 %) was measured to be lower than that usually observed in competing processes, but a strong relationship was found with the different beam strategies used to contour, and infill by hatching, a part section. The majority of pores were found to be small spherical gas pores, concentrated in the infill hatched region; this was attributed to the lower energy density and less focused beam used in the infill strategy allowing less opportunity for gas bubbles to escape the melt pool. Overall, increasing the energy density or focus of the beam was found to correlate strongly to a reduction in the level of gas porosity. In addition, the volume fraction of pores in bulk material was found be approximately linearly related to the volume fraction of gas pores in the powder. Rarer irregular shaped pores were mostly located in the contour region and have been attributed to a lack of fusion between powder particles. When manufacturing samples with older melt strategies an extra defect type was observed: large tunnel defects that grew through the deposited layers. They develop when capillary and wetting effects overcome gravitational forces, which leads to the melted powder tracks separating by beading up, rather than filling in large voids present in the preceding layer. These samples were also used to confirm that a hot isostatic pressing cycle was able to close all internal porosity to below the resolution limit of the equipment used ( 2 µm), apart from defects with surface connected ligaments. Pores were found to be crucial in determining fatigue crack initiation sites, with those at the surface found to be far more likely to initiate a crack. Finite element modelling demonstrated that pores near the sample surface generated a much higher elastic stress concentration than those in bulk material. Plotting the fatigue cycles to failure against the estimated stress intensity factor generated by the pore at the crack initiation site was found to be more informative than using the global stress, with higher stress intensities associated with shorter fatigue lives. Furthermore, by XCT analysis of machined but untested fatigue samples it was possible to predict with reasonable accuracy (>97.5 %) where fatigue cracks would initiate based on the relative stress intensity factor of all the pores. In contrast, crack growth was found to be insensitive to porosity, which was attributed to the much higher stress concentration generated by the crack in comparison to the pores. Some crack diversion was associated with the local microstructure, with prior β grain boundaries often coincident with crack diversion. 17

18

19 Declaration No portion of the work referred to in the thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning. 19

20

21 Copyright i. The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the Copyright ) and s/he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes. ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made. iii. The ownership of certain Copyright, patents, designs, trade marks and other intellectual property (the Intellectual Property ) and any reproductions of copyright works in the thesis, for example graphs and tables ( Reproduction ), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and commercialisation of this thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available in the University IP Policy (see intellectual-property.pdf), in any relevant Thesis restriction declarations deposited in the University Library, The University Library s regulations (see manchester.ac.uk/library/aboutus/regulations) and in The University s policy on presentation of Theses. 21

22

23 Acknowledgements Despite a PhD being an individual and sometimes lonely affair, it would certainly not be possible to do alone. Many people from the universities of Manchester and Sheffield freely and cheerfully gave both their time and expertise to help me. First thanks must got to my two supervisors, Phil and Iain, who both contributed enormously to my project. Without them, their support and ideas (and occasional criticism), this document would not exist. Numerous other people also contributed to the planning, conducting and thinking about the results of the experiments within this thesis and all deserve my thanks. From the CDT, Claire Hinchliffe and Brad Wynne. In the X-ray team: Fabien Léonard, Chris Martin, Julia Behnsen, Jasmin Stein and Richard Balint all contributed to my knowledge of the theory and practicalities of XCT. In the Mercury Centre using the Arcam: Fatos Derguti, Everth Hernandez-Nava, Lampros Kourtis and Chris Smith all passed on hard earned experience on how to make the thing do want you actually want. James Hunt for his untiring ability to send samples for testing and not being saddened by my questioning about procedures. Meurig Thomas for discussions about actual science. The many members of the sadly departed C7 who s time overlapped with mine, and with who pub curry based discussions were often usefully academic. Jack Donoghue and Alphons Antonysamy who helped with sample prep and EBSD; Jack, along with Jefri Draup, also proof read various articles; and Tom Brownsmith and Hao Zhao who assisted with SEM. Chris Martin, Dave Strong, and Mark and Gaz from the workshop who helped with the design and manufacture of the tensile rig. Outside of academia, it would not have been possible without my friends, many of whom came from one or more of C7, the CDT or the Mercury Centre. My partner Hayley for her cheerfulness, forgiveness when I didn t come home till late and encouragement when times were hard. Finally, my parents, for their constant support and not worrying too much that I hadn t got a proper job. I am very grateful to everyone named above and many more. 23

24

25 Glossary A number of terms are used within this thesis that can be ambiguous to readers not closely involved in the topics discussed. Indeed, many of the terms were defined by manufacturer of the additive manufacturing system under investigation, so are unknown outside of the field. Therefore, a list of terms is provided below detailing all the abbreviations and manufacturer terms used in this document. AM Additive manufacture. AR Aspect ratio, generally here referring to pore morphology. Arcam AB The Swedish company that manufactures the SEBM machines and owns the intellectual property regarding the electron beam control. BCC Body centred cubic. Build A manufacturing cycle within the SEBM machine used to produce components. Build temperature Prior to melting each layer during SEBM, preheating is intended to raise the powder bed to this temperature. CAD Computer aided design. Contouring Melt strategy used to melt the 2D outline of a layer of a sample during SEBM, utilises the MultiBeam setting. CT Compact tension, a common geometry for samples used to measure fatigue crack growth rates. DLF Direct laser fabrication, an AM process. DMD Direct metal deposition, an AM process. EBF 3 Electron beam free form fabrication, an AM process. EBM Control Software used to set up and control the SEBM systems, and also calculates the electron beam current and speed to be used during SEBM to both preheat and melt the powder layers. EBSD Electron back scatter diffraction. FE Finite element. Filtered back projection A method of reconstructing a 3D volume from 2D projections aquired during XCT scanning. Focus offset A current applied to the focusing coils to modify the focal plane of the electron beam. GA Gas atomisation, a powder production technique. 25

26 GLOSSARY Hatch offset Distance between parallel melt passes, also referred to as line offset. Hatching Melt strategy used to melt the central region of a layer of a sample during SEBM. HAZ Heat affected zone. HCF High cycle fatigue. HCP Hexagonal close packed. HDH Hydride dehydride, a powder production technique. HIP Hot isostatic pressing. Layer During AM 3D components are manufactured from layers defined by a 2D geometry, and in powder bed AM it can also refer to the layer of powder deposited. LCF Low cycle fatigue. LEFM Linear elastic fracture mechanics. Line offset Distance between parallel melt passes, also referred to as hatch offset. Log file An automatically generated record of the SEBM build containing much information regarding the process, including the electron beam currents and speeds used during melt steps. MultiBeam Used during the contour strategy, the electron beam moves so fast it appears to he eye as if there were multiple beams, this has the effect of keeping multiple melt pools active during contouring. PA Plasma atomisation, a powder production technique. Preheat The electron beam is defocused and moved rapidly across the layer to heat and sinter the powder layer prior to melting. PREP Plasma rotating electrode process, a powder production technique. PRS Powder recovery system, Arcam equipment used to remove the sintered powder from samples. RA Reduction in area. SEBM Selective electron beam melting, the AM process under investigation here. SEM Scanning electron microscopy. Speed function A function used by EBM control that calculates the electron beam speed for a given current. SLM Selective laser melting, an AM process. STL A file format exportable by most CAD software to define 3D geometries and used by most AM equipment. Themes Settings predefining the beam speed and current, build temperature, layer thickness, etc. for a given material. Thickness function A function used by EBM control to increase the speed of the 26

27 GLOSSARY electron beam when melting overhanging structures during the hatching melt strategy. Turning function A function used by EBM control to increase the speed of the electron beam when it turns back on itself during the hatching melt strategy. UTS Ultimate tensile strength. Voxel A 3D volume element, analogous to a 2D pixel. WAAM Wire and arc additive manufacture, an AM process. XCT X-ray computed tomography. XCT scan The process of collecting XCT data. 27

28

29 1 Introduction Journalists have hailed 3D printing as a new industrial revolution that will result in a step change in the manufacturing sector [1, 2, 3]. 3D printing describes a family of technologies in which material is deposited and consolidated in successive layers, using a focused heat source, to build up a structure from 2D slices [4]. Predictions have been made that in the near future every household will own a 3D printer and be able to produce household items, as and when required [3]. Indeed, there are 3D printers that can manufacture nearly all of their own internal components and thus, to a certain extent, self replicate [5]. Whether the media hype is a triumph of engineering or marketing, is, in the authors mind, yet to be decided. However, from the advantages claimed by manufacturers it is easy to see why there has been such a keen interest. When applied to metallic materials, 3D printing is generally referred to as additive manufacturing (AM). AM techniques offer the capability of manufacturing lightweight components, with topographically optimised geometries, directly from 3D computer aided design (CAD) data, without extensive machining. When combined with shorter lead times, high material utilisation rates, and reduced tooling costs, AM is an attractive proposition for low volume manufacturing industries (e.g. biomedical [6] and aerospace [7]). The AM process that is the focus of this thesis, selective electron beam melting (SEBM), is a promising powder bed-based AM technique for near-net-shape manufacture of high value titanium components. The SEBM system developed by Arcam AB, who is currently the only commercial supplier of such equipment, employs a rapidly scanned electron beam focused by electromagnetic lenses with sufficient energy to melt the precursor powder layers [8]. Unfortunately, and despite this system being sold as a product ready for implementation in large scale manufacturing, there are a number of drawbacks to the SEBM approach. Not least is the propriety nature of the algorithms used to control the electron beam which make it hard to predict how an arbitrary geometry will be melted when using the standard, optimised, settings recommended by Arcam. Further, when SEBM is applied to Ti-6Al-4V, the room temperature grain structure and texture is atypical of that produced by conventional thermomechanical processing. Hence, more research is required to fully understand the implications of the microstructure on the mechanical performance. Likewise, the phase transformations that result from the deposition process are not fully characterised and understood. Another major drawback, and the subject of this thesis, is the pres- 29

30 CHAPTER 1. INTRODUCTION ence of defects (porosity) in components. Fatigue testing of SEBM Ti-6Al-4V samples has shown pores to be ready sites for crack initiation. Despite this, while many authors have noted the presence of various types of porosity in samples manufactured by SEBM, there have been no detailed examinations of the size, morphological or spacial distribution of the pores. Additionally, no thorough examinations of the influence of melt strategies and sample geometries on defect populations been published. Furthermore, there have been no attempts to quantify the effects of the pores to the fatigue life; in other words, to understand why cracks are more likely to initiate at certain pores rather than others. The work in this thesis can therefore be understood as having two aims. Firstly, to address the current lack of fundamental understanding that exists of defect-process relationships with the SEBM process. To achieve this end, X-ray computed tomography (XCT) has been used extensively to quantify the size, morphology, frequency, and distribution, in three dimensional space, of the pores found in titanium test samples. This has allowed statistically valid results to be obtained with far more detailed information than has been previously possible. By using XCT systems with different resolutions, it has been possible to quantify the position of the full size range of pores in relation to the beam scanning strategies, as well as to measure the true sizes and morphologies of the pores. This has allowed valuable insight to be gained into the origin of different types of defect and their location, with respect to the build cycle in the SEBM process. Moreover, strategies to avoid the appearance of defects have been developed based on the more detailed knowledge gained regarding pore formation. The second aim was to investigate and predict the effect of porosity on the fatigue performance of SEBM Ti-6Al-4V samples. A range of techniques was used to predict the fatigue crack initiation site and characterise crack growth. Finite element (FE) analysis of idealised pore geometries revealed possible reasons for the preferential nucleation of cracks at certain pores within a component. XCT was also used to identify pores in fatigue samples prior to loading. Pores could then be sorted based on the predicted local stress from FE modelling and thus the most detrimental defects to fatigue life could be identified. A custom built tensile rig was used to apply load to the fatigue samples whilst collecting XCT data. This opened the cracks and allowed them to be imaged in 3D by XCT. Hence, the influence of pores on both crack initiation and growth could be experimentally observed by using interrupted fatigue test samples. Finally, fracture surfaces and crack paths were investigated using both SEM and EBSD in order to relate some crack diversion to microstructural effects. The thesis presented here documents the way in which the above aims were addressed. Following this introduction, a literature review of the relevant published research is provided. This includes an introduction to AM and titanium, as well as a more in depth discussion of SEBM process, the fatigue of titanium and the defects 30

31 encountered within SEBM samples. The experimental techniques and modelling procedure utilised are described in the proceeding chapter. The results and discussion that follow have been broken down into smaller chapters, concerning specific topics, in order to aid the readability of the thesis. The first results chapter deals with the characterisation of the defects within standard cuboid samples and a discussion of the possible errors in the measurements. The subsequent chapter documents and discusses the experimental results regarding the effect of different sample geometries and melt strategies on sample defect populations. A third chapter documents the way in which external factors can be modified in order to reduce the appearance of defects, namely, hot isostatic pressing (HIPing) or changing the powder feedstock. The final results chapter details the direct observations made with regards to the influence of defects on the fatigue performance of samples. The major deductions are summarised in the final chapter alongside suggestions of further study, to build on the results of this project. 31

32

33 2 Literature Review The literature review in this thesis gives an overview of the current state of the art with additive manufacture (AM) processes when applied to titanium aerospace alloys and selective electron beam melting (SEBM) in particular. In addition, to better understand the fatigue response of Ti-6Al-4V components produced by SEBM, both general titanium metallurgy and the fatigue behaviour of titanium alloys is reviewed. To allow this broad range of topics to be more easily discussed, the chapter is divided into sections. Firstly the principles and types of AM processes are reviewed before an overview of titanium metallurgy is presented. Selective beam melting powder bed AM is then reviewed in more detail, with a focus on the parameters relevant to the particular SEBM process being investigated. Currently available data on the fatigue of SEBM Ti-6Al-4V samples is then reviewed alongside a general discussion of the fatigue of titanium alloys. Finally, the literature on the typical defect types and their likely origins in SEBM and related processes is reviewed along with a discussion of the effect of important process settings and melt strategies on their appearance and distribution. 33

34 CHAPTER 2. LITERATURE REVIEW 2.1 Additive Manufacture ASTM International describes additive manufacturing (AM) as a process of joining materials to make objects from 3D model data, usually layer upon layer, as opposed to [by] subtractive manufacturing methodologies [like machining] [9]. The ASTM standard covering terminology for additive manufacture was updated as recently as July 2012, so it is unsurprising that other terms exist to describe AM. Some of the terms occasionally used include: additive layer manufacturing, rapid prototyping, rapid tooling, rapid manufacture, freeform fabrication, solid freeform fabrication, direct digital manufacturing, stereolithography and 3D printing [4, 10, 11, 12]. In the media, 3D printing has become a common term, generally referring to the production of plastic components. For the remainder of this thesis the process will be referred to as AM as it has become the generally accepted term for the production of metallic engineering components Principles and Development of AM Stereolithography (SLA), often considered the first AM process, was patented by Hull in 1986 [13] and incorporated into a working machine shortly afterwards [4]. During SLA, illustrated schematically in Figure 2.1, a UV light is employed to cure a thin layer of resin, the geometry of which is derived from a CAD model, after which the build platform is lowered to allow another layer of resin to cover the cured layer. The liquid resin is then cured directly onto the previous layer with adhesion between the layers resulting in a solid component. A solvent is then used to remove any excess resin which has not been cured, leaving material only where required [13]. Since Hull s patent was issued, many companies have produced equipment for AM [4]. Early in the development of AM it was decided that developing a method to Figure 2.1: Schematic diagram of the stereolithography process. Reprinted from Al- Bermani [14]. 34

35 2.1. ADDITIVE MANUFACTURE (a) (b) (c) Figure 2.2: Example of the conversion from CAD model to slices for AM: (a) Original solid CAD model; (b) converted to an STL surface of triangles; (c) sliced into layers for manufacture. Adapted from Gibson [4]. slice all available CAD formats for all available AM formats was impractical; a generic CAD format was therefore required [4]. Hull s company, 3D systems, subsequently developed the STL (signifying either stereolithography [4] or standard triangulation language [15]) format and made it freely available. STL files, which represent CAD models as surfaces described by triangles, are now exportable by most CAD packages and used by most AM systems. Each triangle is described by both the position of its vertices and a vector normal to the face to indicate the outer surface. Incorrect conversion by CAD software to STL can result in problems, such as gaps in the surface or incorrect labelling of inner surfaces [4]. Specialist software, such as MAGICS [16], has since been developed to check and repair STLs. All AM processes follow the same principals of dividing a CAD model (STL) into slices that are then deposited sequentially to manufacture the component. The orientation of the STLs is known to affect component accuracy and build time [4], as well as the parts microstructure and mechanical properties (see subsection 2.3.4). Figure 2.2 illustrates the typical steps required to convert a CAD model into slices. Note that the cup defined by the CAD file is truly circular with smooth edges, whereas the STL is an approximation made of planar triangles. Deviations from the desired CAD model can be introduced by either using too few triangles in the STL, or too large a thickness when slicing the part [4]. Following the slicing of the STL, the AM system can be prepared for the build, with the procedure varying depending on the equipment, but usually including practical considerations such as ensuring sufficient feedstock, as well making any required adjustments to the control software [4]. The build cycle itself is a largely autonomous process and the number of different AM processes and materials are too large to describe fully here. For more information, the reader is referred to the relevant literature [4]. 35

36 CHAPTER 2. LITERATURE REVIEW Advantages over Traditional Manufacturing The advantages of using AM to produce metallic components, listed below, make it an attractive proposition for low volume manufacturing in the biomedical [17, 6] and aerospace industries [7]. These include: The production of complex shapes that would be impossible, or very difficult, to machine using traditional methods is relatively trivial using AM. There is a large amount of literature on the production and properties of lattice structures, for example [6, 18]. Near net shape parts can be produced reducing the machining operations required with usually only small amounts of surface finishing required. The difficulty and cost involved when machining titanium (see subsection 2.2.2) make this even more attractive for the production of titanium components. AM avoids the long lead times and high costs associated with the tooling or moulds required for conventional processes such as forging and casting. The additive process leads to greater material utilisation. The buy-to-fly ratio (the weight of the initial material compared to that of the final component) is considerably less than with traditional subtractive operations, which can have ratios as poor as 20:1 and average around 5:1 [19, 20, 21]. Estimates of buy-tofly ratios for AM processes vary from 3:1 to close to 1:1 [19, 22]. Unused material can be recycled and reused leading to materials usage rates of 97 % [19]. Functionally graded components can be manufactured by varying the chemistry [23, 24] or the density and strength of porous lattice structures [11, 25] Metal AM Processes Metal AM processes can be classified based on both the heat source and feedstock. Systems have been developed that provide thermal energy via laser, electron beam and electric arcs/plasma, while both powder and wire can be employed as feedstock. Powder can be delivered by either being blown directly into the melt pool, or as a layer of powder spread uniformly across the whole build area. These two processes are termed blown powder and powder bed respectively, whereas wire is always fed directly into the melt pool. There are advantages and disadvantages to all the techniques used, the most obvious being the trade off between resolution and material deposition rate. Figure 2.3 illustrates the relationship between material deposition rate, feature resolution and power requirement. Table 2.1 gives an overview of some of the metal AM pro- 36

37 2.1. ADDITIVE MANUFACTURE Figure 2.3: Schematic illustration of relationship between build rate, power, and feature definition with metal AM. Reprinted from Frazier [26]. Table 2.1: Summary of different metal AM processes. The numerical data provided here is representative of that available in the literature, but in many cases it would be possible to achieve different values by varying the processing conditions. Data taken from references: [10, 27, 28, 29, 30, 31, 32, 33]. Process SLM SEBM Heat source and Melt pool Layer Build rate Typical maximum method depth (mm) height (mm) (cm 3 h 1 ) size (mm) Laser beam & powder bed < Electron beam & powder bed DLF/DMD Laser beam & blow powder EBF 3 Electron beam & > wire feed Plasma arc & WAAM > wire feed cesses available and the component sizes and build rates that are possible. Feature resolution and dimensional accuracy of the final component is a function of the layer thickness and melt pool size. The best resolution, but lowest build rate, is achieved using selective laser melting (SLM) of a powder bed [4, 10]. SLM can produce complex parts with high accuracy and resolution. A schematic of the process, which takes place in a protective inert atmosphere [34], is given in Figure 2.4. A roller/scraper is used to spread powder uniformly across the build area before a high intensity laser fully melts and consolidates the powder in small local melt pools where required [18]. The build table is then lowered to allow the next layer of powder to be spread. Similar to most AM processes, SLM uses a orthogonal coordinate system with layers melted in the x-y plane and built up in the z-direction. x is often, but not invariably, used as the material de- 37

38 CHAPTER 2. LITERATURE REVIEW Figure 2.4: Schematic diagram of the selective laser melting (SLM) process. Reprinted from Kruth [35]. Figure 2.5: Schematic diagram of direct laser fabrication (DLF). Reprinted from Kobryn and Semiatin [36]. position direction. Selective laser sintering (SLS) is very closely related to SLM, but the temperature is kept lower and the powder is only partially melted [35]. Direct laser fabrication (DLF) [23, 36] (also known as direct metal deposition (DMD)) also employs a laser as the heat source, but rather than spreading powder across the entire build area, powder is injected directly into the melt pool with a carrier gas, which is consequently considerably larger in dimensions. Like SLM, DLF is carried out under a protective atmosphere. Figure 2.5 illustrates the major components and coordinate system used for DLF. 38

39 2.1. ADDITIVE MANUFACTURE Figure 2.6: Electron beam freeform fabrication (EBF 3 ). Reprinted from Taminger and Hafley [30]. Selective electron beam melting (SEBM) AM is essentially identical to SLM, except the laser heat source is replaced by an electron gun and processing takes place in a vacuum rather than an inert atmosphere [37]. The larger melt pool results in both a slightly lower resolution and higher deposition rate. Only conductive materials can be used. Non conductive materials would lead to a build up of charge which would deflect the beam and cause the powder to be repelled. A more detailed review of this process is provided in section 2.3. Higher deposition rates can be achieved by utilising wire rather than powder as the feedstock. Wire and arc (or plasma) additive manufacture (WAAM) can provide a much higher deposition rate than most other metal AM processes of several kilograms an hour but with a much thicker layer height and thus lower resolution [31]. With WAAM, an electric arc provides the heat, into which wire is fed and melted onto the previous layer [31]. Electron beam freeform fabrication (EBF 3 ) is a similar process that can provide relatively high levels of material deposition [30]. EBF 3 takes place in a vacuum and utilises an electron beam to melt wire feedstock. Figure 2.6 gives a schematic of the equipment within the EBF 3 vacuum chamber. The deposition rate, or resolution, can be altered by varying the wire diameter. Alternatively, a laser heat source in a protective argon environment can be used to melt wire feedstock [38]. Metal AM has been used to produce components from a number of different alloys, including various steels [18, 35], nickel based [24, 34], zirconium [24] and aluminium [32, 39]. However, there has been particular interest and much published research in the application of AM to manufacture components made from titanium, and, in particular, the alloy Ti-6Al-4V [4]. Consequently, this is the material used for all the results within this thesis. 39

40 CHAPTER 2. LITERATURE REVIEW 2.2 Ti-6Al-4V Titanium and its alloys are of great importance to aerospace, with the industry utilising around 50 % of all titanium production [40]. This is primarily due to titanium alloys high specific strength and, in the context of aero-engines, good creep resistance (up to about 550 C) [41]. Figure 2.7 shows the specific strength of titanium alloys alongside other common engineering alloys at a range of temperatures, illustrating why titanium is so important to the aerospace industry. The high energy cost of extracting titanium from its ore (around 14 times that of steel), coupled with the difficulty in forming and machining titanium, makes its alloys relatively expensive compared to competing materials such as aluminium or steel [40]. Titanium alloys high cost restricts their application where material properties alone would suggest they are superior to the competition, for example, there is little used in the automotive sector [41]. Outside of aerospace, titanium is limited to other high value applications such as biomedical and chemical processing equipment, both of which utilise its excellent corrosion resistance. Of the titanium alloys in service, the most widely used is Ti-6Al-4V, which accounts for over 50 % of worldwide titanium production [41]. Figure 2.7: Relationship of specific 0.2 % proof stress (ratio of proof stress to relative density) with temperature for light alloys, steels and nickel alloys. Reprinted from Polmear [40]. 40

41 2.2. TI-6AL-4V Introduction to Titanium and its Metallurgy Despite making up approximately 0.6 % to 0.86 % by weight (hereafter given the symbol wt%) of the earth s crust, the element titanium was not discovered until 1791 by clergyman William McGregor [40, 41]. The strong tendency of titanium to react with oxygen and nitrogen subsequently prevented the development of a commercially attractive process to isolate titanium from its ore for over a century [41]. During the second world war, Wilhelm Justin Kroll reduced titanium tetrachloride (TiCl 4 ) with magnesium in an argon atmosphere to produce a porous titanium sponge. Although new reduction methods are being developed [42, 43], Kroll s process remains essentially unchanged as the dominant process for titanium production [40, 41]. The Kroll process cleared the way for commercial exploitation and large scale production of titanium. Titanium s low density (4500 kg m 3 ) and high melting temperature (1678 C) made it very attractive to the military aerospace sector, which has provided much research funding [40]. Ti-6Al-4V was developed in 1954 and remains to this day the most widely used titanium alloy [41]. Ti-6Al-4V s name refers to the major alloying additions of 6 wt% aluminium and 4 wt% vanadium. Alloys developed in the United States have names reflecting their chemical composition, such as Ti-6Al-4V (often shortened to Ti-64) and Ti-6Al-2Sn- 4Zr-2Mo (Ti-6242). Numerical designations can also be used for alloys developed in Britain (e.g. IMI 834), but this refers to a numbering system used by the IMI Titanium company, now owned by TIMET. Titanium s metallurgy is dominated by the fact that it can undergo an allotropic transformation and has a very high solubility with most transitional metals. Depending on temperature, it can adopt one of two crystal structures. The phase transformation between the high temperature body-centre-cubic (BCC) crystal structure β phase and the low temperature hexagonal-close-packed (HCP) α phase occurs at the β-transus temperature, above which the crystal structure is entirely β. The exact β-transus temperature is highly dependent on the presence of substitutional and interstitial elements, but for pure titanium is approximately 882 C [40, 41, 44]. The crystal structures of both the α and β phases are illustrated in Figure 2.8. Figure 2.8 also highlights the most densely packed planes and the lattice parameters of the two phases. The orientation relationship between the two phases during transition through the β-transus is characterised by the Burgers orientation relationship (named after Wilhelm Gerard Burgers, the first person to notice this relationship). The Burgers orientation relationship was initially observed between the α and β phases of zirconium [45]; subsequently, the same orientation relationship was confirmed for the phase transition in titanium [46], with both diffusion controlled nucleation and growth, and martensitic transformations closely following this same habit plane relationship [41]. The Burgers 41

42 CHAPTER 2. LITERATURE REVIEW Figure 2.8: Crystal structures of the low temperature α (left) and high temperature β (right) phases of titanium. Reprinted from Lütjering and Williams [41]. (110) (0001) Figure 2.9: Burgers orientation relationship between the most densely packed planes of the α andβphases. Reprinted from Leyens and Peters [47]. orientation relationship involves alignment between the close packed planes and directions in each phase. This is given both graphically in Figure 2.9 and also below with the Miller/Miller-Bravais indices. {110} β {0002} α 111 β 1120 α The Burgers orientation relationship gives a total of twelve discrete α orientations that can form from an initial β orientation. An inverse transform from α to β gives a possible total of six distinct orientations. Therefore, theoretically, a single heating and cooling cycle through the β-transus could lead to the formation of 72 different variants and a near random texture. Experimentally, this generally does not occur with a preference for certain variants often observed [48]. The β-transus temperature can be altered by the addition of solute alloying elements that stabilise either the α or β phase [41]. Elements that are soluble in the α phase are termed α stabilisers, and raise the β-transus temperature; those that are soluble in the β phase (β stabilisers) lower the β-transus temperature. In addition, the α and β stabilisers either similarly increase, or reduce, the martensitic start temperature. 42

43 2.2. TI-6AL-4V Figure 2.10: Phase diagram illustrating the effect of alloying elements on the β- transus temperature. Reprinted from Lütjering and Williams [41]. Figure 2.10 gives some common α and β stabilisers and provides a schematic representation of their effect on the thermodynamic stability of the phases. The β stabilisers can be further subdivided into isomorphous and eutectoid forming alloy elements. Figure 2.10 also shows elements that are classed as neutral, due to their similar solubility in both the α and β phase, and hence small influence on the β-transus temperature. Non-neutral alloying additions can also extend the temperature range in which both α and β phases are thermodynamically stable; this allows both phases to exist in the same alloy at the same temperature. Titanium alloys can therefore be grouped based on the phases present [41, 44], into: α alloys are single phase alloys based on the α phase. They can be further separated into unalloyed titanium, with small amounts of residual oxygen and iron, and those which are alloyed with solid solution strengtheningαstabilisers. The high corrosion resistance of these alloys means the primary use is in the petrochemical industry. Near α alloys contain small (1 % to 2 %) amounts ofβstabilisers and retain their strength and creep resistance at elevated temperatures. Discs, rings and blades at the hot end of compressors in gas turbine engines are the main application of these alloys. α+ β alloys contain a larger proportion of β stabilisers, typically 4 % to 6 %, which lead to a significant fraction of β being present at ambient temperatures, while other elements strengthen the α phase. These alloys are used in a wide range of applications from hip replacements to fan blades and airframe structures in aerospace. Metastable β and near β alloys contain a high enough fraction of β stabilisers that on quenching from high temperature they can retain a fully β structure. They can then be strengthened by controlled precipitation of the α phase. Structural aerospace applications, such as landing gear, are a common use of these alloys. 43

44 CHAPTER 2. LITERATURE REVIEW Machining Titanium One of the major advantages of AM, in comparison to traditional subtractive manufacturing techniques, is the reduced machining required to produce the final product. The classification of titanium and its alloys as difficult to machine materials [49] makes the use of AM all the more attractive for the production of titanium components. A number of intrinsic properties of titanium have been suggested as the cause of its reduced machinability compared to other commonly used engineering alloys, such as steels and aluminium [40, 41, 50, 51]. During machining, a cutting tool is used to remove material from the workpiece in chips, the formation of which involves large strains and plastic deformation. Machining titanium results in thin serrated chips as well as high levels of heat generation [50]. The high temperatures generated close to the cutting edge of the tool is the primary cause of the high tool wear experienced when machining titanium [50, 51]. A large proportion (about 80 %) of this heat is then absorbed into the tool rather than the workpiece due to the low thermal conductivity of titanium, which is around one sixth that of steel [50, 51]. The thin chips, and corresponding small tool-chip contact area (around one third that of steel), and titanium s high retained strength at the elevated temperatures generated during machining, lead to high stresses at the tool tip. To avoid an uneconomically short tool life, the cutting speeds must thus be kept low [49], ergo increasing the unit cost of a machined component. Another major difficulty when machining titanium is chatter of the workpiece, the principal cause of this phenomenon being titanium s relatively low modulus of elasticity The β α Transformation and Microstructure Development Ti-6Al-4V is an α+β alloy and both the α and β phase are retained at room temperature. However, the microstructure of Ti-6Al-4V is very dependent on the processing conditions used and can be tailored to suit different applications. As discussed in subsection 2.2.1, on cooling through theβ-transus (approximately 995 C for Ti-6Al-4V [20, 52, 53]), titanium undergoes an allotropic transformation, with the microstructure formed dependent on the rate of cooling. Slow cooling gives rise to the α phase precipitating initially on the β grain boundaries to form a layer of grain boundary α. Colonies of α lamellae (also referred to as laths) then grow into the β grains by means of diffusion with both the width of the lamellae and the colony size inversely related to the cooling rate. The maximum colony size is thus limited by theβgrain size. At higher cooling rates, the number ofαvariants nucleated increases and α lamellea can also nucleate autocatalytically at α plates already within the β grain. This reduces the effective transformed microstructural scale [54]. The resultant 44

45 2.2. TI-6AL-4V Figure 2.11: Optical micrographs of lammella microstructures in Ti-6Al-4V formed by cooling from above the β-transus to room temperature at: (a) & (b) 15 C s 1 ; (c) 1.5 C s 1. Reprinted from Ahmed and Rack [52]. microstructure is termed Widmanstätten. Figure 2.11 shows examples of the result of these diffusional transformations at different cooling rates. At fast cooling rates, such as those experienced during quenching, transformation to non-equilibrium phases can occur. Ahmed and Rack [52] characterised these transformations and quantified the cooling rates required, using an adaptation of the end quench Jominy procedure to produce a continuous cooling diagram for Ti-6Al-4V as shown in Figure They found that cooling rates of 20 C s 1 to 410 C s 1 were accompanied by a massive transformation to α m where the high cooling rate suppressed lattice diffusion. Transformations to α m and α are competitive and the volume fraction of α m will increase with cooling rate. Figure 2.13 gives examples of the α m resulting from different cooling rates. As the cooling rates increase still further ( 410 C s 1 ), α m is replaced by an acicular martensite (α ). These martensitic phases will decompose to the equilibrium α and β phases by precipitation of the β phase upon annealing in theα+β phase field [41]. As can be inferred from Figures 2.11 and 2.13, the transition between these transformations is not discrete, with a gradual transition where one process begins to replace another [52]. The continuous cooling diagram in Figure 2.12 shows the cooling rates at which the transformation associated with the slower cooling rate was judged by 45

46 CHAPTER 2. LITERATURE REVIEW Solution treat at 1050 C for 30 minutes β-transus at 994 C β α β α m β α M s = 575 C 525 C s C s 1 20 C s 1 Figure 2.12: Continuous cooling diagram for Ti-6Al-4V after solution treatment for 30 min at 1050 C, redrawn from the diagram proposed by Ahmed and Rack [52]. Figure 2.13: Optical micrographs of martensitic structures in Ti-6Al-4V following cooling from above the β-transus to room temperature at: (a) 410 C s 1 ; (b) 275 C s 1 ; (c) 175 C s 1 ; (d) 20 C s 1. Reprinted from Ahmed and Rack [52]. 46

47 2.2. TI-6AL-4V (a) (b) Figure 2.14: Examples of other microstructures possible in Ti-6Al-4V produced by thermo-mechanical processing: (a) bi-modal, reprinted from Peters and Ritchie [57]; and (b) equiaxed, reprinted from Yuri et al. [58]. Ahmed and Rack [52] to be replaced by the transformations associated with higher cooling rates. Figure 2.12 also indicates a martensitic start temperature (M s ), above which a martensitic transform will not occur, however high the cooling rate. Ahmed and Rack [52] took this value from the literature, where a range of possible martensitic start temperatures can be found. They quote a relatively low value of 575 C from Madjdic and Zeigler [55]; other sources state values from 625 C [56] to 775 C [47]. The transformation microstructures produced on rapidly cooling from above the β-transus are the most similar (of standard processing techniques) to those that are seen in powder bed AM owing to the relatively high cooling rates found in these processes. However most Ti-6Al-4V is used in a wrought form where hot deformation followed by recrystallising during annealing below the β-transus can be used to develop both bi-modal (Figure 2.14a) and fully equiaxed (Figure 2.14b) microstructures [41]. Nonetheless, to date there has been no reported attempts to introduce a deformation step to AM with SEBM. Therefore, for a full review of thermo-mechanical processing of Ti-6AL-4V and the effect on the resultant microstructure, the reader is directed to the literature [40, 41] Influence of Lamellar Microstructure on Mechanical Properties As mentioned in the previous section, SEBM of Ti-6Al-4V produces microstructure that, in comparison to that produced by standard processing routes, is most similar to that of a rapid β transformed material. In brief, SEBM produces a fine lamellar transformation microstructure with a coarse columnar primary β grain structure [27]. The formation of this microstructure is discussed in more detail in subsection A brief introduction to the range of literature investigating the effect of this microstructure on the mechanical properties ofα+β titanium alloys is provided here. This is intended to provide only a background and a more detailed discussion of the mechanical properties 47

48 CHAPTER 2. LITERATURE REVIEW (a) (b) Figure 2.15: Effect of lamellar microstructure on mechanical properties in Ti-6Al-4V: (a) Influence of slip length on mechanical properties; and (b) effect of cooling rate, and thus microstructural size, on yield stress and elongation. Adapted from Lütjering [59]. of Ti-6Al-4V is provided in later sections. In particular, the static properties of SEBM Ti-6Al-4V is reviewed in subsection and the various factors that contribute to the fatigue performance of Ti-6Al-4V are reviewed in section 2.4. Lütjering states that the most influential microstructural parameter on the mechanical properties of fully lamellar microstructures is the α colony size, which determines the effective slip line length [59]. Figure 2.15a illustrates schematically the general relationship noted by Lütjering between colony size and different mechanical properties. Colony size is inversely related to the cooling rate, which is plotted against elongation and yield stress in Figure 2.15b for three different α+β alloys. As the slip length decreases, due to smaller colony sizes with faster cooling, the yield stress and ductility both initially increase. Figure 2.15b assumes a constant primary β grain size and it should be noted that a finer primary β microstructure would also be expected to reduce the colony size. When the cooling rate is high enough to produce martensite the yield strength increases dramatically as, according to Lütjering [59], the effective slip length becomes equal to the width of the individual plates. In addition, the increased lattice distortion associated with martensitic microstructures will increase the dislocation glide resistance. Conversely, the ductility reaches a maximum for fine colonies, after which it decreases with faster cooling and the formation of martensitic structures (Figure 2.15b) [59]. 48

49 2.2. TI-6AL-4V The high cycle fatigue (HCF) strength of lamellar Ti-6Al-4V also increases with decreasing slip line length. HCF is dominated by the time for crack nucleation, which itself depends on the resistance to dislocation motion and the formation of microcracks in persistent slip bands that can then break through initial microstructural barriers such as grain boundaries. The colony size would therefore be expected to strongly influence this process [59]. In comparison, the low cycle fatigue (LCF) strength depends on the resistance to micro-crack propagation, in addition to crack nucleation. Colony boundaries, as well as martensitic plates, provide an effective barrier to microcrack propagation. Thus, the LCF strength will improve with decreased slip line length [59]. Large (macro) crack propagation resistance, in contrast to the previously mentioned properties, increases with increasing slip line length. With high R-ratios (see subsection for definition), the crack front geometry, or roughness, increases with larger colonies and this hinders propagation through its effect on closure (subsection 2.4.6). At low R-ratios, crack closure increases with increasing roughness of the fracture surface and increasing shear displacement at the crack tip, both increasing with colony size. This reduces the crack propagation rate by reducing the plastic strain range at the crack tip [59]. 49

50 CHAPTER 2. LITERATURE REVIEW 2.3 Selective Electron Beam Melting of a Powder Bed To date, only one company, Arcam AB [22], has produced commercially available equipment for powder bed AM with an electron beam heat source, a process similar in principle to SLM. Since SLM is the more prevalent technique, the differences between these two methods are highlighted in this section. It is also important to note that Arcam have released themes (settings predefining the beam speed and current, build temperature, layer thickness, etc.) for Ti-6Al-4V and cobalt-chrome alloys for use with their equipment. As a result, most of the studies on component properties built with the Arcam machine have been with Ti-6Al-4V, although processing conditions for other materials, such as titanium aluminides [37, 60], are being developed. In this section the principles of the selective electron beam melting (SEBM) system are reviewed alongside the typical microstructures produced and the relationship to the mechanical properties of the built parts. The fatigue performance of SEBM Ti-6Al-4V is reviewed in more depth in subsection Selective Electron Beam Melting Hardware One clear difference between SLM and SEBM is that the entire heating process takes place within a controlled vacuum ( mbar) rather than an inert atmosphere [29, 4], since electrons would be scattered by gas atoms. The vacuum chamber is first taken to a higher vacuum (10 4 mbar) before bleeding in helium to allow a consistent low pressure. Figure 2.16 shows a schematic diagram of the major components making up an SEBM system. The electron gun (1) produces the electron beam which is then focused with electromagnetic lenses (2), and deflected to the desired location with electromagnetic deflection coils (3). Within the electron gun a filament with 10 A (dc) current and 60 kv potential between it and the anode reaches temperatures above 2000 C, allowing easier escape of the electrons during operation [28, 12, 62, 60, 63]. These are subsequently accelerated by the voltage potential to velocities between 0.1 and 0.4 times the speed of light [64]. A control electrode between the filament and the anode regulates the beam current by modifying the potential difference. The beam is focused on the work-piece by a series of electromagnetic lenses and has a spot size of approximately 0.35 mm [65]. When combined with typical beam currents of between 5 ma and 15 ma this gives a power density from 3.1 kw mm 2 to 9.3 kw mm 2. Pre-alloyed powder, with diameters from 45 µm to 100 µm, is stored in hoppers (4) and then gravity fed into the path of the rake (5) which moves between the two hoppers to spread each powder layer. After each layer has been processed by the beam the solidified structure (6) is moved vertically downwards by the build table (7) the distance of one layer height [12]. 50

51 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.16: Schematic of SEBM equipment. Reprinted from Murr et al. [61]. The layer thickness is generally larger in SEBM than SLM [62], though Arcam are releasing new themes to try and reduce the difference. Early Arcam themes for Ti-6Al-4V utilised a 100 µm layer thickness [12, 27, 61, 64], whereas the newer themes allow layers of 70 µm, or as fine as 50 µm, to be used [29]. Progress has also been made in the maximum build size available, from the earlier S12 model with a 200 mm 200 mm 180 mm build chamber to the new Q20 model, which has a build space with a diameter of 350 mm and a height of 380 mm [22] Advantages of SEBM The Arcam process, and the use of an electron beam rather than a laser to melt a powder bed, brings with it a number of potential advantages. These include: The electron beam can be moved almost instantaneously with the deflection coils. In contrast, deflection of a laser beam requires the movement of galvanic mirrors where a motor must move an object with inertia. Electron beam melting thus allows complex lattice structures to be built much more quickly [4], or the beam to be split and moved so rapidly the process behaves as if there were more than one beam [29]. The vacuum chamber the process takes place in ensures a clean environment and reduces the requirement for shielding gas, although a small amount of gas is still required to maintain the controlled vacuum [14]. 51

52 CHAPTER 2. LITERATURE REVIEW Electron beams have a high electrical efficiency, approaching 95 % [66]. In contrast, typically only 10 % to 20 % of electrical energy is converted to laser beam energy. However, laser technology is advancing, and beams with efficiency of 70 % to 80 % have been developed, hence this advantage may become less important in the future [4]. The electron beam couples well with any electrically conductive materials [66], and there are no issues with reflectivity as there are with a laser beam. The proportion of absorbed beam power has been calculated to be 0.6 for SEBM [27] and 0.35 for laser AM [67] of Ti-6Al-4V. High power electron beam generation equipment is generally cheaper than high power laser systems [4]. However, this cost saving is reduced by the increased cost of the vacuum system compared to a shielding gas system. The higher power of the electron beam also allows relatively higher deposition rates, in comparison to other forms of powder bed AM [14]. However it must be noted that there are drawbacks to the SEBM process. The process only works with conductive materials, results in a poorer surface finish and has a lower feature resolution compared to SLM Electron Beam Melting Process When the electron beam is moved across the powder bed, the kinetic energy of the electrons is converted to thermal energy in the powder. The electrons gradually giving up energy through a series of collisions with atoms within the material which converts their kinetic energy into lattice phonons [68]. The thermal vibration of the powder atoms occurs by transfer of energy from the accelerated electrons to the outer electron shell of atoms within the powder [69]. Figure 2.17 shows a schematic diagram of the possible paths of electrons when hitting a material. Electrons are unlikely to be absorbed at the surface of the material due to their small size ( nm) compared to the lattice repeat distance (approximately 0.2 nm to 0.4 nm) and will penetrate a distance into the material while scattering off the crystal atoms before dissipating their energy [69]. The atoms at the surface are instead heated by conduction from deeper in the material. The depth (R p ) at which, statistically, the great majority of electrons will be absorbed can be found using Equation (2.1) [69]. R p = V ρ (2.1) Setting the accelerating voltage V and the and the material density ρ, to standard values for SEBM of titanium, 60 kv and 4500 kg m 3 respectively, this results in a penetra- 52

53 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.17: Electron penetration range. Reprinted from Taniguchi et al. [69]. tion depth of 18 µm, a significant fraction of the layer thickness. The maximum absorption of electrons occurs at depth X e, whist most electrons are absorbed at a depth of X d. A significant fraction of the beam energy can be lost by reflected electrons, depending on the accelerating voltage and target material, however, this is typically only about 20 % of the incident energy [68]. In addition to heat generation, X-rays and fluorescence can be produced, further reducing energy absorption [69]. The negative charge of the electrons can lead to a build up of charge in one location as electrons are absorbed, if they cannot be conducted away [4]. A build up of charge, and associated high repelling forces between the beam and negatively charged powder particles, can lead to detrimental effects. The beam will become more diffuse as it is repelled. In addition, if the repulsion between powder particles becomes great enough, it will overcome the gravitational and frictional forces holding them in place and lead to expulsion of particles which can cause a breakdown in the process and create a powder cloud [4]. As well as limiting SEBM to conductive materials, melt strategies must be altered to avoid charge build up even in loosely connected metal powders. Newly spread powder is thus not immediately melted with the electron beam, rather, the entire bed is first preheated and sintered by rapidly scanning a defocused beam with a high current across it to increase its conductivity [64]. Once sintered, the powder is more conductive and can be safely melted with a more focused beam in the desired locations. The beam focus is altered by means of a focus offset, which moves the focal point either above (positive focus offset) or below the powder bed. Preheating is also used to maintain an approximately constant temperature within the powder bed, referred to as the build temperature [27]. Attempts have also been made to add a preheating stage to SLM, but this requires a separate heating system [4]. The focused electron beam in SEBM, in comparison to the laser beam used during SLM, is more diffuse, partially to avoid charging [4]. Figure 2.18 shows the relative intensity of the electron beam generated by a focused Arcam system, which is very similar to the Gaussian distribution often encountered during electron beam melting 53

54 CHAPTER 2. LITERATURE REVIEW Figure 2.18: Intensity profile of the electron beam generated by a focused Arcam S12 machine. Provided by Arcam to, and reprinted from, Al-Bermani [14]. Figure 2.19: Near-infrared thermal images taken during the SEBM manufacturing of a 25 mm square block: (a) during contouring; and (b) during hatching. Reprinted from Gong et al. [70]. [69]. To simplify modelling the thermal interaction of the beam, Juechter et al. [65] used a diameter of 0.35 mm for the Arcam system, which corresponds to approximately the full width at half maximum of the histogram presented in Figure The melt procedure used by the Arcam SEBM system is, in brief, as follows. Firstly contour passes are used to melt and consolidate the outline of each 2D section slice. The contour strategy (also referred to as contouring ) uses a technology known as MultiBeam, which rapidly moves the beam so as to keep several separate melt pools active at one time [29]. Gong et al. [70] used near-infrared imaging to image the contour melting process as shown in Figure 2.19a which clearly illustrates the near instantaneous interaction of the beam with multiple locations on the powder bed. The centre of each section is then filled in by rastering the beam in a snaking melt strategy known as hatching. As can be inferred from Figure 2.19b, the hatching step does not utilise the MultiBeam function and instead employs a single beam with a forwards and backwards motion with a continuous path [27, 29]. Other than the MultiBeam used during 54

55 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.20: Scanning strategy used for SLM and SEBM. For each layer first the outline is melted, followed by infill hatching rotated by 90 each consecutive layer. Reprinted from Santos [71]. contouring, the melt process is similar to that used during SLM. Figure 2.20, despite referring to an SLM process, describes the identical melt sequence employed during SEBM. In the Arcam machine, the hatching direction rotates by 90 every layer, i.e. for a constant melt area every fourth layer melt sequence is identical. The spacing between consecutive hatch passes (termed hatching pitch in Figure 2.20, and also known as hatch offset or line offset) can be varied to alter the heat input. Arcam recommend a line offset of 0.2 mm for a 70 µm layer thickness, although line offsets between 0.1 mm and 0.25 mm have been utilised in previous studies [28, 65, 72]. During hatching, the beam power is decided by the control software in an attempt to maintain a constant average surface temperature for each layer. Once the current is decided, a speed function is utilised to try to preserve a constant melt pool depth by maintaining an approximately constant line energy in J mm 1, given by Equation (2.2) [73]. Line Energy= P v (2.2) where: P and v are the beam power (W) and speed (mm s 1 ), respectively. Al-Bermani et al. [27] used single passes of the electron beam, controlled by the speed function, on a solid Ti-6Al-4V plate to examine the size and geometry of the melt pool with different beam currents and therefore, different beam speeds. In addition to experimental analysis, Al-Bermani et al. used the Rosenthal solution of a moving heat source in a thick plate, given by Equation (2.3), to model the size of the melt pool. T = η P ( ) v (x+r) 2 π k R exp 2 α (2.3) where: T is the temperature change, η is an efficiency factor, P is the beam power, k is the thermal conductivity, R= x 2 + y 2 + z 2, v is the beam velocity, α is the thermal diffusivity and x, y, and z are the coordinates of each location with x aligned with the beam traverse direction. 55

56 CHAPTER 2. LITERATURE REVIEW Figure 2.21: Experimental measurements and model predictions of melt pool size in the SEBM process. Cross sectional views perpendicular to beam direction of melt tracks with various beam currents on a Ti-6Al-4V plate: (a) 6 ma; (b) 8 ma; (c) 10 ma; (d) 12 ma. (e) The Rosenthal predicted melt pool geometry in plane parallel to beam motion. Reprinted from Al-Bermani et al. [27]. 56

57 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figures 2.21a d show micrographs of the single melt tracks taken in the plane perpendicular to the electron beam velocity. The beam current was varied from 6 ma to 12 ma and the speed from 188 mm s 1 to 608 mm s 1. In addition, Figures 2.21a d show the solidus isotherm prediction from the Rosenthal equation. The experimental results indicated that an efficiency factor of 0.6 was appropriate for use in the Rosenthal solution [27]. Both experiment and theory revealed that with a current between 8 ma and 12 ma the speed function was effective in keeping a constant melt pool depth. When a 6 ma current was used, the speed function applied a too low beam velocity that led to a larger melt pool [27]. At all speed/current combinations the melt pool was found to be elongated in the beam travel direction, increasingly so as the speed increased (2.21e), and to have an approximately semi-circular profile normal to the travel direction. Al- Bermani et al. s result is qualitatively similar to the heat profile observed on the nearinfrared image in Figure 2.19b. Prior to the work of Al-Bermani et al., the melt pool size had been modelled using FE by Zäh and Lutzmann [74]. This work used somewhat lower beam speeds and currents than the current Arcam standards, but also predicted an elongated melt pool with elongation increasing at higher beam velocities, up to nearly 8:1 length to width ratio. More recently, modelling was undertaken by Antonysamy et al. [75], who used a thermal finite difference code to predict the size and shape of the melt pool, as well as the location of the β-transus isotherm. At the condition modelled, a melt pool highly elongated in the beam travel direction was predicted. Differences in the depth to width prediction were found with Antonysamy et al. [75] predicting a smaller depth to width ratio in contrast to the prediction from Al-Bermani et al. [27] which was semi circular (width/depth = 1). However, all three of these attempts to model the melt pool geometry were conducted on a solid plate of Ti-6Al-4V. Clearly melting powder would alter the thermal properties of the material and therefore the melt pool size and geometry. Lattice Boltzmann methods have been used to model the melting and solidification of powder particles [73, 76]. When melting randomly placed powder particles, Bauereiß et al. [76] demonstrated that surface tension and wetting effects dominate over those of gravity and viscous forces. When combined with the relatively slow diffusion in titanium, this study has shown that this can lead to powder particles coalescing with the closest surface before they are fully melted. The closest surface may well be another powder particle rather than the previously solidified layer [76]. Furthermore, the model predicted the packing density and beam power can significantly influence the melt pool shape [73]. When the line energy is reduced, such that the melt pool is not much larger than the particles themselves, the melt pool tends to assume a round shape in order to minimise the surface energy (Figure 2.23a). As the 57

58 CHAPTER 2. LITERATURE REVIEW Figure 2.22: Finite difference model predictions of melt pool shape in bulk material with a beam speed of 500 mm s 1 and current of 10 ma. Thermal field temperature isotherms in a plan view surface (a) and the cross section at the maximum melt pool width (b). The centre line melting temperate isotherm illustrates the elongated nature of the melt pool profile. Reprinted from Antonysamy et al. [75]. line energy, and hence melt pool size, increase, the relative contributions of surface tension, gravity and wetting change and the melt pool attains a shape more similar to that predicted by Al-Bermani et al. [27], Zäh and Lutzmann [74] and Antonysamy et al. [75], Figure 2.23b. However, even with an approximately constant packing density and beam power, the melt pool geometry was found to be heavily dependent on the local stochastic packing of the particles, which was predicted to lead to a large variation in melt pool size and geometry [73]. The influence of powder size distribution was not investigated, however, it is likely that this too will influence the melt pool geometry. All of the modelling work described in this section has neglected to include any variation in the temperature around the melt tracks. The snaking motion of the hatching means that one side of the melt pool will have been heated by the previous hatch pass. Thus the melt pool will be asymmetrical due to one half being already fully dense prior to the beam pass and having both a higher temperature and conductivity. In addition, no studies have been conducted to examine the local effect of this snaking, i.e. in the region where the beam turns back upon itself. 58

59 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED (a) Experiment Simulation (b) 0.5 mm 100 cells = 0.5 mm Figure 2.23: Experimentally obtained optical images (left) and modelling (right) results of the influence of electron beam power on melt pool geometry. Line energies of: (a) J mm 1 ; (b) J mm 1. Reprinted from Korner et al. [73]. 59

60 CHAPTER 2. LITERATURE REVIEW Typical Microstructures Published research has shown that SEBM of Ti-6Al-4V produces some unique microstructural features that can be influenced by factors such as the build geometry, raster strategy and heat input [12, 27, 29, 61, 75, 77]. With the small moving melt pool, the high thermal gradient at the solidification front typically leads to a coarse columnar primary phase β grain structure that develops by epitaxial re-growth, up through the deposited layers, with a preferred 001 fibre direction [27, 75]. Figure 2.24 illustrates the strong texture measured by Antonysamy et al. [75] at different heights within bulk SEBM material viewed in the x-y plane. Figure 2.25 also shows columnar grains in the x-z plane of thin walls built up from a thick base. Clearly, the bulk solidification microstructure (Figure 2.24) is different to that near the walls surfaces. Prior to discussing the reasons for this, the origin of the bulk columnar β grain structure must be reviewed. Epitaxial re-growth of the β grains occurs primarily because of the steep thermal gradient and little constitutional supercooling present in a Ti-6Al-4V melt pool [27, 75]. Nucleation of new grains in the melt requires a higher under-cooling than that required for heterogeneous nucleation at a substrate material [78]. In welds and Figure 2.24: Reconstructed β grains and texture at different heights within bulk material: (a) 0.5 mm; (b) 5 mm; (c) 25 mm; (d) 35 mm from the baseplate. Reprinted from Antonysamy et al. [75]. 60

61 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.25: Reconstructedβgrains in SEBM Ti-6Al-4V walls of various widths: (a) example of room temperature ESBSD used to reconstruct the 1.0 mm wall shown in (b); (c) 1.5 mm wide; (d) 2.0 mm wide; (e) 5.0 mm wide. Reprinted from Antonysamy et al. [75]. 61

62 CHAPTER 2. LITERATURE REVIEW in SEBM, where the liquid material is of near identical composition to that of the substrate, the wetting angle will be near zero resulting in almost no barrier to nucleation and very little undercooling required for heterogeneous nucleation [79]. Furthermore, because the heat source is heating the liquid and the substrate is relatively cold, there will be a large thermal gradient at the solid-liquid interface, which makes it difficult to achieve sufficient undercooling for nucleation. In addition, the main alloying elements in Ti-6Al-4V exhibit partition coefficients close to unity, so that insufficient constitutional supercooling occurs ahead of the solidification front to promote the nucleation of grains in the melt pool [27, 75]. The strong texture is established due to the preferred 001 growth direction of cubic metals, such as the BCC β phase of Titanium. Once nucleated, a preferentially orientated grain will grow faster than surrounding grains that are not orientated so favourably. The more random grain structure shown at the very bottom of Figure 2.25, near the base plate, gives way to more strongly textured grains as preferential growth of certain grains occurs while new grain nucleation is suppressed. By using the melt pool modelling reviewed in subsection to predict the thermal gradient (G) and growth rate (R) during solidification, it was shown that conditions are those that are known to produce columnar grains during other Ti-6Al-4V solidification processes. Both Al-Bermani et al. [27] and Antonysamy et al. [75] predicted a thermal gradient from K cm 1 to K cm 1 and a growth rate from 10 cm s 1 to 30 cm s 1. These values have been combined with the work of Kobryn and Semiatin [36], who used various solidification processes, including AM, to plot a solidification map of G against R. Together, this has shown that the conditions experienced during solidification would be expected to be within the range that resulted in either a fully columnar or mixed columnar and equiaxed microstructures (at the rear of the melt pool) [27]. However, Al-Bermani et al. [27] suggested that, as the solidification time is so brief, once columnar grain growth is established, it occurs preferentially over nucleation of new equiaxed grains. While modelling and experimentation has shown that the coarse columnar primary phase β grain structure shown in Figure 2.24 is to be expected in bulk material, the sample geometry and different melt strategies used can make large differences to the local grain structure. When Antonysamy et al. [75] reconstructed the β grains near SEBM sample surfaces they observed a more random texture. Figure 2.26 shows a plan view (x-y plane) of the reconstructed β grain structure for walls of varying thickness shown previously in Figure It is clear that when melting small sections, with only the contour beam strategy (Figure 2.26a & b), the grain structure is very different to that seen in the bulk material, Figure In addition, even when melting larger specimens (Figure 2.26c & d), the result of the contour pass and surrounding powder particles is a skin layer, approximately 0.8 mm wide, where the solidification condi- 62

63 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.26: Plan view of reconstructed β grains in SEBM walls of various widths: (a) 1.0 mm; (b) 1.5 mm; (c) 2.0 mm; and (d) 5.0 mm. Images (a) to (c) show approximate widths of the contour passes and image (d) labels the different types of β grains near sample surfaces. Reprinted from Antonysamy et al. [75]. 63

64 CHAPTER 2. LITERATURE REVIEW Figure 2.27: Nucleation of columnar grains near a SEBM component surface from partially melted powder particles in the x-z plane: (a) optical image showing attached particle; and (b) reconstructedβgrains illustrating epitaxial re-growth. Reprinted from Antonysamy et al. [75]. tions are different. Inclusive of the bulk, a total of five different β grain types were observed near the surface. The process by which these form is complex, and detailed in full by Antonysamy et al. [75]. In summary, moving in from the sample edge, these β grains consisted of: (i) small grains within the attached partially melted powder; (ii) inwardly growing columnar grains as far as the contour centre; (iii) preferentially orientated grains of type (ii) that continue to grow upwards at the contour centre (axial grains); (iv) columnar grains in the inner half of the contour pass; and finally (v) the standard bulk columnar grain microstructure. Grain types (i) and (ii) were observed in walls with thickness 1 mm and under, while types (iii) and (iv) were seen only in walls greater than 1 mm. The final type (v), that of the bulk microstructure, was seen in sections greater than 2 mm in thickness only, that were large enough to require use of the hatching strategy in addition to contouring. Types (ii) to (v) are labelled in Figure 2.26d, whereas type (i) is not labelled, but can be seen at the very edge of the samples. A higher resolution map of the reconstructedβgrains at the very edge of a vertical wall sample surface is provided in Figure 2.27, alongside an optical micrograph of the whole wall width. Figure 2.27b shows more clearly the smaller type (i) grains within partially melted powder particles. Antonysamy et al. [75] also used Figure 2.27b as an example to demonstrate that the grains near the surface nucleate on, and grow from, the surrounding powder particles rather than previously solidified layers. In addition to walls of various widths, Antonysamy et al. [75] reconstructed the solidification β grain structure of various geometries that could be of interest when trying to industrialise SEBM. One such geometry was that of thin walls transitioning to a thick section. β grains nucleated from the powder beneath the thick section, in a similar fashion to those seen on the edge of vertically orientated samples (Figure 2.26), in addition to growing out from the previously solidified wall. Newly nucleated β grains would dominate over those growing from the wall if their orientation was preferential. A similar structure was observed when manufacturing 2 mm walls, inclined 30, 45 64

65 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.28: Optical micrograph of the typical microstructure within a bulk SEBM Ti-6Al-4V sample. Reprinted from Al-Bermani et al. [27]. and 60 to the build direction. Again, after a small region where β grains nucleated from the powder bed underneath the sample, columnar grains grew vertically upward, parallel to the build direction, regardless of the build angle. In the SEBM process with Ti-6Al-4V, upon cooling, the solidified β grains transform into fine α laths, with a small fraction of retained β, the morphology of which is dependent on the cooling rate, as well as being affected by the build time and temperature [27, 41]. Figure 2.28 gives an example of the typical transformed microstructure seen in bulk material. During manufacture it is thought that each layer passes through both the liquidus and β-transus temperature several times due to the thermal cycles caused by the additional layers being deposited [27]. It is also currently unclear whether the initial β α transformation is diffusional or martensitic [75], as while typically there is a very high cooling rate (10 3 K s 1 to 10 5 K s 1 [27]), the build temperature (730 C) is within the range of martensitic start temperatures in the literature. Since both martensitic and diffusional transformations obey the Burgers orientation relationship, the possible α orientations remain the same. Antonysamy et al. [75] demonstrated that there was little variant selection taking place in SEBM with Ti-6Al-4V by comparing the actual α texture with texture generated by randomly transforming the prior β grains according to the Burger orientation relationship. Hence, the resultant texture of the α phase room temperature microstructure was found to be much weaker than that of the parentβ, 3 random in contrast to the 8 random for the β texture [75]. In SLM, microstructural banding can sometimes be observed [36, 80]. These bands, when viewed in the x-z plane, can appear to be outlines of the thermal field from subsequent melt tracks [80]. It has been suggested that the banding is caused by changes to the characteristic size of the α microstructure or precipitation of the intermetallic Ti 3 Al phase [36, 80]. In both cases the reason for the bands has been attributed to the deposition of subsequent layers being deposited generating a new heat 65

66 CHAPTER 2. LITERATURE REVIEW affected zone (HAZ) in layers of previously deposited material. An energy dispersive X-ray spectrometer has been used by Thijs et al. [80] to show periodic segregation of aluminium in SLM of Ti-6Al-4V, with peaks spaced by the layer height and zones with more than 20 atomic % aluminium. In SEBM there has been no reported microstructural banding. If bands are caused by a HAZ forming each layer, the elevated build temperature of SEBM must reduce this effect. Some aluminium segregation has been observed in SEBM, however this was attributed to increased evaporation from larger melt pools when the beam quickly turns back on itself [65]. Evidence of a martensitic transformation has been provided by Al-Bermani et al. [27]. They found that in small SEBM Ti-6Al-4V cuboid samples (<20 mm in x and y, and<10 mm in height), a martensitic layer approximately 500 µm could be observed with both optical micrographs and X-ray diffraction (XRD) intermediately below the top surface (Figure 2.29). The implication of this none uniform microstructure is that different areas of the build experience different thermal histories. Furthermore, when Hrabe and Quinn [81] measured the α lath widths at a number of heights in a 27 mm tall sample, they found a statistically significant upward trend, see Figure 2.30a. A similar trend was reported by Murr et al. [61], who recorded anαlath width of 3.2 µm at the top of a sample and 1.6 µm at the bottom. Larger α laths are associated with slower cooling rates; an apparent contradiction between Al-Bermani et al., who suggest a faster cooling rate at the top of a sample, and Hrabe and Quinn and Murr et al. who suggest a slower cooling rate. However, since martensite was only observed in small samples with a small thermal mass, the build temperature may still increase with height. In addition, Hrabe and Quinn neglected to measure the upper 4 mm of the sample, so it is unclear whether the increase in α width they report continued to the upper surface. The α lath widths given by Hrabe and Quinn [81] in Figure 2.30 is somewhat smaller than those stated by Murr et al. [61] (1.6 µm to 3.2 µm). Karlsson et al. [29] have reported a lower still value, between 0.6 µm and 0.7 µm. Differences inαlath size have been attributed to differences in sample size [29], with larger samples resulting in a slower cooling rate and a larger microstructural size. However, this could also be attributed to coarsening of the microstructure, owing to an increased time larger samples are held at the elevated build temperature, due to the longer build time. Conversely, both Hrabe and Quinn [81] and Puebla et al. [77] noted reductions to the α lath width by increasing the beam speed, while keeping the beam power constant. Finally, a chemical transient region was recorded by Al-Bermani et al. [27] in the bottom 500 µm of SEBM samples (in contact with the baseplate), where melting of the stainless steel baseplate leads to diffusion of Cr, Fe and Ni into the Ti-6Al-4V alloy. These elements stabilise the β phase, and a clear difference is visible in the backscatter electron image presented in Figure The brittle nature of this initial layer has the 66

67 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.29: Evidence ofα in SEBM Ti-6Al-4V: (a) transition from diffusionalα+β microstructure to diffusionlessα in a 5 mm tall sample; (b)α at the top of a 2 mm-tall sample; (c) a fine, diffusional α+β structure at the top of a 18 mm tall sample; and (d) XRD trace of bulk material from a 25 mm tall sample compared to that from a 1 mm tall sample; traces have been shifted vertically for clarity. Reprinted from Al-Bermani et al. [27]. (a) (b) Figure 2.30: Measured variation in: (a) α lath thickness; and (b) static mechanical properties with distance from the base plate in SEBM Ti-6Al-4V. Reprinted from Hrabe and Quinn [81]. 67

68 CHAPTER 2. LITERATURE REVIEW Figure 2.31: Diffusion of baseplate alloying elements into an SEBM Ti-6Al-4V part leading to a brittle region in the vicinity of the baseplate. Reprinted from Al-Bermani et al. [27]. unintended benefit of allowing easier removal of samples from the baseplate without resorting to cutting [27] Static Mechanical Properties Ti-6Al-4V samples manufactured via SEBM typically exhibit static mechanical properties comparable to those of wrought material. Table 2.2 shows a range of the yield (σ y ), ultimate tensile stresses (UTS) and elongations available in the literature, alongside the minimum requirement of ISO for wrought and annealed Ti-6Al-4V. Despite some disagreement as to the absolute values of yield stress and UTS, most exceed that of ISO Material tested without machining, i.e. with rough, as-built, surfaces, exhibited a reduced σ y, UTS and elongation in comparison to most of the literature available for machined material. When comparing the strengths quoted in the same study, the standard deviation is small. Antonysamy [10] noted that there was no statistically significant difference between samples, regardless of orientation and location in the build chamber. Other authors also have noted that the x and y directions are equivalent in properties [7]. Due to the time difference between published work, some discrepancy between studies may have been introduced by the release of new hardware and software by Arcam. Furthermore, as noted in the preceding section, the microstructure is not uniform between builds or even within a single build [81]. Al-Bermani et al. [27] has demonstrated a dependence of the microstructure and yield stress on the build temperature. In general, a lower build temperature implies faster cooling, which leads to a finer microstructure and an increased yield stress, Figure For example, the lower σ y and UTS reported by Antonysamy [10] compared to that by Al-Bermani et al. [27] can be attributed to the higher build temperature used to manufacture the specimens. Unfortunately, not all the literature gives the exact conditions used to manufacture the specimens tested. When measuring the tensile strength of a single sample at different heights in a build, results from Hrabe and Quinn (Figure 2.30) show no significant difference in properties, despite showing statistically significant differences to theαlath thickness with build height. 68

69 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Table 2.2: Static mechanical properties of SEBM Ti-6Al-4V material taken from the literature. If available, the standard deviation is given. Sample description σ y (MPa) UTS (MPa) Elongation (%) Year Ref. ISO > 780 > 860 > 14 Arcam AB data sheet [82] HIPed samples in x/y [26] HIPed samples in z [26] As built (not machined) in x/y direction 783±15 833±22 2.7± [7] As built (not machined) in z direction 812±12 851±19 3.6± [7] x/y direction 899 ± ± ± [83] z direction 869 ± ± ± [83] 15 samples in x direction 832 ± ± ± [10] 15 samples in y direction 817 ± ± ± [10] 15 samples in z direction 802 ± ± 8 15 ± [10] Bulk material [84] 5 samples in x/y direction 830 ± ± ± [85] 5 samples in x/y direction (HIPed) 795 ± ± ± [85] Top of build (coarseαlaths) [61] Bottom of build (fine and coarseαlaths) [61] Figure 2.32: Influence of build temperature microstructure and mechanical properties in SEBM Ti-6Al-4V samples: (a) yield strength; (b) α lath width; (c) α colony scale factor. Standard deviation of results is indicated by the error bars. Examples of micrographs taken as-built and following HIPing, for each temperature. Reprinted from Al-Bermani et al. [27]. 69

70 CHAPTER 2. LITERATURE REVIEW All machined tensile samples from SEBM material show a marginally lower yield stress and UTS in the z direction (parallel to the build direction) as opposed to in either the x or y directions, when comparing data from the same studies [10, 26, 83]. It has already been shown that the microstructure in SEBM is coarser further up in a build from the baseplate. Vertical samples would, by necessity, require a build height at least that of the sample size. Horizontal samples, on the other hand, are likely to be arranged flat, and thus have a lower build height. This coarser microstructure may explain the reduction in mechanical properties. Mechanical property anisotropy has been noticed in many forms of AM with the x-y directions generally marginally stronger [26]. In SLM, this could be attributed to banding of the porosity in horizontal layers [12]. To date, banding of porosity has not been reported in SEBM, although it may also contribute to reduced mechanical properties in the z direction. Conversely, when samples were tested as built (i.e. with rough surfaces) those manufactured in the z direction outperformed those in the x/y direction [7]. This could be due to the different microstructure near the sample surface compared to in the bulk (subsection 2.3.4) or owing to a reduced surface roughness in the z direction compared to the x or y direction (subsection 2.5.7). The values in Table 2.2 also include some tensile data following Hot Isostatic Pressing (HIPing). HIPing involves the application of heat and pressure, with the intention of healing internal porosity [41] (subsection 2.5.9). Al-Bermani et al. investigated the effect of HIPing on both the mechanical properties and microstructure of SEBM Ti-6Al-4V samples. HIPing was observed to result in a coarsening of the microstructure and an associated decrease in yield strength, Figure The same trend was seen in the results of Facchini et al. [85], where the coarser microstructure post HIPing increased the effective slip line length and thus reduced the tensile strength. When the tensile results for SEBM specimens are compared to SLM, most authors report that SLM produces a greater yield strength and UTS, owing to the finer transformation microstructure, but reduced elongation to failure [12, 26, 83]. Figure 2.33 illustrates the differences between the two techniques. Rafi et al. [83] attributed this to the formation of martensite in the SLM process due to the higher cooling rates and much lower build temperature. An increased number of build defects (lack of fusion between layers) in SLM has also been suggested to contribute to the ductility reduction [12]. 70

71 2.3. SELECTIVE ELECTRON BEAM MELTING OF A POWDER BED Figure 2.33: Ultimate tensile strength (UTS), 0.2 % offset yield stress (YS) and elongation for material machined from an SEBM vertical cylinder and a SLM horizontal cylinder. Reprinted from Murr et al. [12]. 71

72 CHAPTER 2. LITERATURE REVIEW 2.4 Fatigue of Titanium The fatigue performance of conventionally produced titanium products has been studied extensively [86, 87, 88, 89, 90, 91, 92, 93]. In contrast, the data available for SEBM titanium is relatively limited. With SEBM, the few studies available have also only tested the high cycle fatigue strength of the material. During high cycle fatigue, the majority of the deformation within the sample is elastic, and testing is load controlled. Conversely, low cycle fatigue is strain controlled and the stress applied is great enough to cause plastic deformation in the majority of the sample [94]. In the published work available, the vast majority of testing has been performed on dog bone type samples under cyclic load, with only one attempt to date being made to study crack growth rates. In this section the fatigue response of SEBM Ti-6Al-4V is reviewed alongside that of traditionally produced titanium specimens, for which there is more available literature. The section begins with some fracture mechanics, which has been particularly successful in the characterisation of fatigue crack propagation [94] Fracture Mechanics Fracture mechanics today originated from the work of Griffith in 1921 [95], who developed an equation to describe when unstable fracture would occur in a thin brittle plate with a central crack of length 2 a. Griffith postulated that for a crack to grow, the energy release rate by reduction in elastic energy must be greater than the energy consumption rate required to form the new surfaces. Griffith s equation was later modified by Orowan to include the effect of plastic work, which allows the critical stress for fast fracture (σ f f ) of metals to be calculated with Equation (2.4). σ f f = E (2 γ s + γ p ) π a (2.4) where: E = E 1 ν 2 E = E (plane strain) (plane stress) and: γ s is the free surface energy per unit area, γ p the plastic work per unit area, E the Young s modulus and ν the Poisson s ratio. Full derivations of Equation (2.4) are available elsewhere [94, 96, 97]. It is clear from this analysis that the presence of a crack in a material can significantly reduce the load it can support before failure. The growth of cracks can be characterised more precisely using linear elastic fracture mechanics (LEFM) to calculate the stresses near the crack tip. Indeed, the char- 72

73 2.4. FATIGUE OF TITANIUM (a) (b) (c) Figure 2.34: Modes of fracture: (a) tensile opening (mode I); (b) in-plane sliding (mode II); and (c) anti-plane shear (mode III). Reprinted from Suresh [94]. acterisation of long fatigue crack growth for a broad range of loading, environmental and microstructural conditions is based on LEFM [94]. Fracture modes can be separated into the three modes shown in Figure 2.34 [94, 96, 97], with most fatigue testing conducted in mode I: tensile opening of the crack. All the following equations are based on mode I opening. For a plate of infinite width with a centre through crack of length 2 a, under mode I loading, the stress around the crack tip is given by Equations (2.5a c). σ xx = σ a 2 r cos θ ( 2 1 sin θ ) 3 θ sin 2 2 (2.5a) σ yy = σ a 2 r cos θ ( 2 1+sin θ ) 3 θ sin 2 2 (2.5b) σ xy = σ a 2 r cos θ 2 sin θ 3 θ cos 2 2 (2.5c) where: σ is the global stress, and r and θ are the distance and angle from the crack tip, respectively. These equations are valid when a>r> ρ, where ρ is the radius of the crack tip. Again, derivations of Equations (2.5a c) are available elsewhere [94, 96, 97]. It was noted by Irwin that the stress around a crack tip is proportional to the square root of the crack length multiplied by the global stress. He then defined a new term, the stress intensity factor (K I ), which is a convenient method to describe the stress around a flaw. Two cracks with different conditions, but a constant value of K I, will have the same stress field [97, 96]. The stress intensity factor is made general by the application of a shape factor (Y ) that depends on the crack and specimen geometry. Thus, for the general case, the stress intensity factor is given by Equation (2.6). K I = Y σ a (2.6) 73

74 CHAPTER 2. LITERATURE REVIEW Figure 2.35: Effect of stress intensity factor range on crack growth in Ti-6Al-4V with an α lamellar microstructure at three different load ratios (R). Reprinted from Nalla et al. [90]. With the stress intensity factor, it is possible to estimate the extent of the plastic zone around the crack in ductile materials [94, 96, 97]. The plastic zone results in the crack behaving as though it were larger, with greater displacements and lower stiffness than in the purely elastic situation [97]. Fracture mechanics also provides methods to approximate other values of importance to predicting crack behaviour, such as crack opening displacement and crack deflection under far-field multi-axial stresses. For brevity they are not displayed here, however equations are available elsewhere [94, 96, 97]. In 1961, Paris et al. [98] hypothesised that fatigue crack growth per cycle (da/dn) was a function of the stress intensity factor range ( K). When da/dn is plotted against K a plot similar to that displayed in Figure 2.35 can be produced. Paris et al. [98] showed that results from different investigations using different specimen sizes and stresses, but the same alloy, temper, and loading range (R=σ max /σ min ), when plotted on a single figure would show very similar rates of crack growth against stress intensity range [94, 96, 97]. The stress intensity range, given by Equation (2.7), is thus often referred to as the driving force for crack propagation. K = K max K min = Y (σ max σ min ) a (2.7) In addition, it was noted that a significant proportion of the data on the double logarithmic plot of K against da/dn fell on a straight line, Figure Fitting an equation to this straight line results in Equation (2.8), known as the Paris equation [94, 74

75 2.4. FATIGUE OF TITANIUM 96, 97]. da dn = C ( K)m (2.8) Where C and m are empirical constants that depend on the material, microstructure, fatigue frequency, load ratio and temperature. The Paris equation is the most common method of characterising fatigue crack growth rates for a range of testing conditions and materials [94]. If the shape factor, Y, is assumed to be approximately constant over the life of the crack, it is possible to integrate the crack growth rate estimation provided by Equation (2.8). Assuming an initial defect size of a 0, the number of cycles, N f, required for a crack to grow to a critical size, a f, is given by Equation (2.9) [94]. 2 N f = (m 2) C Y m ( σ) m π m/2 1 a (m 2)/2 0 1 a (m 2)/2 f (2.9) From Equation (2.9) it can be seen that number of cycles to failure is strongly dependent on the initial defect size, whereas the final crack size is less influential [94]. Ordinarily, Y varies with crack size, thus, integration is performed numerically Stages of Fatigue Crack Growth A schematic illustration of crack growth rate against stress intensity factor range is provided in Figure Initial crack growth is known as stage I crack growth, which occurs when the plastic zone is contained within a few grain diameters [94] (regime A in Figure 2.36). Stage I crack growth is characterised by a zigzag profile due to growth Figure 2.36: Schematic of different regimes of stable fatigue crack propagation. Reprinted from Suresh [94]. 75

76 CHAPTER 2. LITERATURE REVIEW (a) (d) (b) (e) (c) (f) Figure 2.37: Idealised stage II crack growth by plastic blunting and sharpening: (a) No load; (b) onset of tensile loading; (c) peak load; (d) beginning of stress relaxation; (e) zero load; and (f) next loading cycle begins. Reprinted from Dieter [97]. occurring chiefly by single shear in the direction of the primary slip system [94]. New crack surfaces are formed by extreme localised deformation leading to shear decohesion. Cracks with very low stress intensity factors may not propagate at all, as they cannot break through microstructural barriers; the stress intensity threshold range below which they do not propagate is denoted K T h [99]. In contrast, where the local stress is high, such as at stress concentrations, stage I growth may not be observed and cracks will grow initially by stage II crack growth [97]. As the stress intensity range increases the plastic zone will encompass many grains. In this condition crack growth occurs generally in a plane normal to the applied stress. At least two slip systems become active in this stage, known as stage II growth, which takes place in the Paris regime (regime B in Figure 2.36). Stage II crack growth can result in ripples on the fracture surface known as striations. Each striation represents the crack tip position during a loading cycle [97]. Stage II crack growth and striation formation occurs by a plastic blunting and resharpening process outlined in Figure 2.37 [94, 97]. As a tensile load is increased the crack opens, becomes blunt, and extends by a distance comparable to the crack tip opening displacement. Upon relaxation of the load, the crack closes and becomes sharp once again, with an increase in net length [94, 97]. Further increases to the stress intensity factor range, regime C Figure 2.36, results in static failure modes such as cleavage and ductile rupture developing. These modes occur when the maximum stress intensity factor approaches the fracture toughness of the material. Static failure modes will be observed at a lower stress intensity factor if the load ratio is higher [94]. 76

77 2.4. FATIGUE OF TITANIUM Fatigue of SEBM Ti-6Al-4V In contrast to the favourable static properties (subsection 2.3.5) the HCF of SEBM Ti-6Al-4V sample can show large scatter [7, 10, 100]. Figure 2.38 illustrates the results of fatigue testing of polished samples with a maximum stress of 600 MPa and R-ratio = 0.1, by Antonysamy [10]. It is clear that, at the same stress ratio, fatigue life can exhibit a range of several orders of magnitude, making qualification of aerospace components challenging. Antonysamy found that samples tested in the build direction (z) had a significantly lower life than those in either the x or y direction. Frazier s review of metal AM [26] noted a similar trend with samples having an 8 % higher fatigue strength at cycles in the x-y directions (441 MPa) than in the z direction (407 MPa). Figure 2.39 represents an attempt by Edwards et al. [7] to produce S-N curves for SEBM Ti-6Al-4V in a number of conditions. The fatigue life measured is somewhat lower than quoted by Antonysamy and Frazier, although they also reported a reduced fatigue life in the z direction. In addition, they showed that testing samples as built (i.e. not machined), resulted in multiple cracks initiating at the rough surfaces and a shorter fatigue life. When shot peening was utilised to introduce compressive residual stresses at the surface, the crack initiation occurred away from the surface. Antonysamy [10], Frazier [26] and Edwards et al. [7] all recorded that all machined samples had cracks that nucleated exclusively at pores, rather than at other microstructural features. In addition, Antonysamy found that samples with the shortest lives had cracks growing from pores near to their surfaces. Figure 2.40 gives an example of the crack initiation location for a sample that survived less than one million cycles, alongside an example of a longer lived sample, that sustained over two million cycles. All cracks reported by Edwards et al. [7] initiated from pores near the surface unless shot peening was used to introduce compressive residual stress at the surface. Thus, it is Figure 2.38: Number of fatigue test cycles to failure with a fixed maximum stress of 600 MPa, R = 0.1, for Ti-6Al-4V material produced via SEBM. Note that tests were stopped after cycles. Reprinted from Antonysamy [10]. 77

78 CHAPTER 2. LITERATURE REVIEW Figure 2.39: Number of fatigue test cycles to failure for Ti-6Al-4V material produced via SEBM. Reprinted from Edwards et al. [7]. Figure 2.40: Fracture surfaces showing pores at crack initiation sites in SEBM Ti-6Al-4V fatigue samples: (a) the fracture surface of a sample that failed before one million cycles; (b) the fracture surface of a sample that failed after over two million cycles. Reprinted from Antonysamy [10]. 78

79 2.4. FATIGUE OF TITANIUM Figure 2.41: Fracture surfaces showing microscopically smooth facets in an SEBM Ti-6Al-4V fatigue sample: (a) the overall fracture surface; (b) an enlarged view of the area indicated by the arrow. Reprinted from Rafi et al. [83]. clear that pores near the surface are more detrimental to fatigue life than those within the bulk material. Conversely, three-point bend fatigue testing by Chan et al. [101] did not record cracks initiating at pores. However, this work tested as built and electrodischarge machined (EDM) specimens, both of which had a relatively high surface roughness and resulted in cracks initiating from the surface. Porosity has also been recorded by a number of authors as having a negative influence on the fatigue performance of SLM samples [102, 103, 104, 105, 106]. Furthermore, in SLM the fatigue life of samples with cracks initiating from pores near the surface has also been noted as being shorter than those with cracks initiating from subsurface pores [104, 105]. In common with SEBM, the shortest fatigue life has been reported when testing in the z direction [102]. In contrast to the studies mentioned above, Rafi et al. [83] observed crack initiation in some SEBM samples characterized by microscopically smooth facets, shown in Figure They suggest that crack initiation occurred at theαcolony boundaries, with the multiple facets possibly being caused due to shearing across neighbouring α colonies. A similar initiation process has been noted in wrought Ti-6Al-4V and is discussed further in subsection below. Rafi et al. conducted these tests alongside testing of specimens manufactured by SLM, which were found to have a superior fatigue limit, despite cracks initiating at pores within the material. Following HIPing, Arcam claim a fatigue life of > cycles at 600 MPa for SEBM Ti-6Al-4V test samples [82]. The review by Frazier [26] has reported a similar result, with a fatigue strength at cycles of 607 MPa in x/y and 538 MPa in z, after HIPing. Frazier attributes the improvement to the elimination of residual porosity. In contrast, the improvement in the fatigue life of Ti-6Al-4V SEBM samples noted by Facchini et al. [85] was attributed to the microstructural coarsening increasing crack resistance. However, given the dependence of high cycle fatigue on crack initiation, 79

80 CHAPTER 2. LITERATURE REVIEW Figure 2.42: Crack growth rate in CT samples machined from Ti-6Al-4V blocks manufactured by SEBM, R=0.1. Reprinted from Edwards et al. [7]. and the lack of any fracture surface analysis by Facchini et al. [85], it seems more likely that the reduction in pores acting as crack initiation sites is more important than the increase in crack propagation resistance. When comparing the fatigue properties of Ti-6Al-4V samples manufactured with various AM techniques, Brandt et al. [107] found that HIPed SEBM material had a superior fatigue life than HIPed SLM or laser cladding material. Brandt et al. also recognised the detrimental effect that defects, and particularly those near the surface, have on the fatigue life of unhiped samples. In addition to plotting S-N curves, Edwards et al. [7] used compact tension (CT) specimens to plot fatigue crack growth rates against stress intensity factor. Figure 2.42 shows the crack growth rate for loading in the horizontal and vertical direction, compared with a baseline wrought product. While crack growth in the two orientations of the SEBM Ti-6Al-4V showed no noticeable difference, within the Paris regime they deviated slightly from the wrought material. The results indicate that SEBM Ti-6Al-4V actually had a greater resistance to crack growth than the baseline metal [7]. Residual stresses can strongly influence the fatigue performance of samples, with compressive stresses beneficial and tensile stresses detrimental [94]. Extremely large tensile residual stresses have been observed in SLM sample in both the top 1 mm and in the vicinity of the baseplate (within 0.5 mm) [108]. Much lower residual stresses have been measured in material produced by SEBM using X-ray diffraction, with tensile stresses observed near the upper surface and compressive stresses near the lower surface in contact with the baseplate [7]. However, these residual stresses were confined to the outer 0.03 mm of the specimens, thus, residual stresses would have a negligible effect when testing machined samples. 80

81 2.4. FATIGUE OF TITANIUM Stress Concentrations It is well known that stress concentrators, such as notches or holes, severely reduce the fatigue strength of components [97]. The same is true of pores which act to raise the local stress to above that in the bulk material. The stress concentration factor, distinct from the stress intensity factor, is defined as the maximum stress over the global stress (K t = σ max /σ ). Under plane stress and purely elastic conditions, the stress concentration is a function of the geometry only. In contrast, for plane strain conditions, the Poisson s ratio (ν) also influences the stress concentration [109]. Theoretical analysis of the stress concentration around a spherical void in an infinite body under uni-axial loading gives Equation 2.10 [110]. K t = ν ν (2.10) In Titanium: A Technical Guide, by Donachie Jr [20], the reported Poisson s ratio of 0.32 for Ti-6Al-4V at 25 C, when inserted into Equation 2.10, results in K t = Higher values of Poisson s ratio give higher stress concentrations. Larger stress concentrations are generated when the aspect ratio of the void increases [110]. Figure 2.43 illustrates the effect of varying the minimum radius of an oblate spheroid cavity, proportional to the aspect ratio. It is clear that with an increase in the void aspect ratio there is an increase in the stress concentration. Furthermore, the effect of the Poisson s ratio is relatively small [110]. If, instead of a single void, there is a number of voids stacked in the direction of loading, the stress concentration is reduced. The stress reduces further when either, the number of voids is increased, or the distance between them is reduced [110]. In contrast, when two pores are arranged normal to the loading direction, their close proximity can lead to an increased stress concentration [111]. Recently, finite element (FE) modelling of the effect of the proximity of a free surface and other pores on the stress around the spherical pore has been published [111]. Figure 2.44 shows the stress concentration predicted by FE analysis of two identical spherical pores in close proximity under purely elastic conditions. Despite modelling the behaviour in an aluminium alloy, this modelling work by Xu et al. [111], used a Poisson s ratio of 0.3 and is therefore likely to closely represent the behaviour of Ti-6Al-4V. From Figure 2.44, it is clear that the stress concentration is increased only when the two pores are very close, less than a single radius apart, otherwise they have little influence on each other. Figure 2.45 illustrates the predicted effect of pore depth from a free surface on stress concentration. From the elastic line in Figure 2.45, it can be seen that the stress concentration increases in a small region when the cavity is either just inside or outside 81

82 CHAPTER 2. LITERATURE REVIEW Figure 2.43: Influence of spheroid geometry on stress concentration. Reprinted from Pilkey and Pilkey [110]. 82

83 2.4. FATIGUE OF TITANIUM Figure 2.44: Stress concentration due to two spherical voids arranged normal to the direction of loading. Reprinted from Xu et al. [111]. Figure 2.45: Stress concentration due to pore depth from a free surface, made dimensionless by dividing the distance from the surface to the pore centre (D) by the pore radius (r). Negative values indicate that the pore centre was outside the material edge. Reprinted from Xu et al. [111]. 83

84 CHAPTER 2. LITERATURE REVIEW the material. This region of significantly higher stress provides some explanation as to the reduced fatigue life seen when cracks in AM samples initiated from pores just below their surface [7, 10]. Xu et al. [111] also used a stress strain curve of a cast 713 Al-Zn alloy to model the plastic deformation. The elastic-plastic line in Figure 2.45 is less informative, since, with the material model they used, once the UTS was reached, deformation would occur perfectly plastically to accommodate the load and results in a flat trend line. When D/r = 1, the free surface is tangential to the spherical pore. Therefore, at the surface, over an infinitesimally small volume, the area goes to zero and the stress in the purely elastic case should go to infinity (clearly, in reality this would not occur due to plastic deformation). Nonetheless, they have modelled this situation and recorded stress concentrations in both the elastic and plastic cases. The peaks are an effect of the meshing rather than geometry; they have attempted to model the stress in an infinitesimally small volume using elements of a finite size, which the authors have neglected to mention. One study has investigated the effect of porosity on the fatigue life of aluminium samples by utilising X-ray computed tomography (XCT) [112]. The XCT data was used to create FE models of real pore geometries within a fatigue specimen. It was found that fatigue cracks initiated at pores with the greatest local geometrical mean of the stress and strain concentration (k g ), Equation (2.11). k g = K σ K ε (2.11) Where K σ and K ε are the local stress and strain concentrations respectively. Additionally, it has been noted that the geometrical mean of the stress and strain concentration is approximately equal to the elastic stress concentration (k g K t ), this effect is known as Neuber s rule [94]. Although pore size has no influence on the stress concentration for a constant aspect ratio in bulk material, larger pores will result in a larger plastic zone. This will therefore interact differently with a microstructure of a fixed scale. Ergo, pore size, in addition to the stress concentration due to geometry, is an important factor in determining crack initiation [112]. When fatigue cracks in titanium samples initiate at inclusions rather than pores, it has been noticed that crack initiation at the surface also resulted in the shortest fatigue lives [113]. Inclusions generate a stress concentration due to the difference in stiffness between them and the surrounding matrix [110] and can also crack or debond at high stresses [114]. In work by Chandran [113], the position of the inclusion, and thus crack initiation, was found to be more important than variation in the microstructure produced by two separate thermo-mechanical treatments in determining the fatigue life, Figure The effect was so pronounced that it was appropriate to 84

85 2.4. FATIGUE OF TITANIUM Figure 2.46: The duality of fatigue results found by separating the location of inclusions in Ti-10V-2Fe-3Al for two microstructures (designated B and E). Inset images show examples of crack initiation seen on fracture surfaces. Reprinted from Chandran [113]. draw two separate S-N curves depending on the location of the inclusion at the crack initiation location [113]. Chandran showed that Poisson statistics could be used to predict whether a defect would appear in a critical location near the surface. By such a method they demonstrated that defects were expected near the surface approximately 50 % of the time, thus explaining the duality of the S-N curves. It has been suggested that, from a viewpoint of fatigue, inclusions or pores are equivalent to small cracks [115]. Consequently, they can be assumed to generate a stress intensity factor based on their projected area normal to the application of the load. The stress intensity factor generated is then given by Equation (2.12a) for a defect at a surface and Equation (2.12b) for a subsurface defect within bulk material [115]. K max = 0.65 σ π area (Surface) (2.12a) K max = 0.50 σ π area (Subsurface) (2.12b) From Equations (2.12a b), it is clear that defects near the surface will lead to a higher stress intensity factor and thus, crack growth rate, than those beneath the surface of identical size. Nonetheless, treating defects as cracks may not always be appropriate. Irregular porosity in aluminium has been found to initiate non-propagating cracks [116]. Sharp notches are also known to initiate cracks that cease to grow once they reach a certain 85

86 CHAPTER 2. LITERATURE REVIEW size [94] Quasi-Cleavage Facets Although most authors have recorded fatigue cracks in SEBM titanium initiating from pores [7, 10, 26], Rafi et al. [83] observed some cracks in SEBM Ti-6Al-4V samples originating from regions exhibiting microscopically smooth facets. When fatigue testing is conducted on conventionally produced titanium alloys, sites of crack initiation are often characterised by near basal plane faceting, sometimes referred to as quasicleavage facets [86, 87, 88, 89, 90, 91, 92, 93]. Despite the name, these facets are not the result of brittle failure, rather, they are due to gradual failure in a persistent slip band. In fully lamellar microstructures, cross-colony facets occur in near basal orientated colonies [90]. Figure 2.47 shows an example of a faceted fracture surface in lamellar Ti-6Al-4V. With this orientation, where the c-axis is closely aligned with the loading direction, slip on the basal plane is not favoured and, since the facets lie perpendicular to the stress axis, it is difficult to attribute them to a slip deformation model [91]. To explain the appearance of these facets, Evans and Bache [86] proposed a model based on the Stroh pile-up criterion [117]. The model assumes that a strong grain, unfavourably orientated for slip, has a neighbouring weak grain, favourably orientated for slip. Deformation within the weak grain leads to a pile up of dislocations at the grain boundary and a resultant shear and additional tensile stress in the strong grain. The shear stress results in a slip band, which opens up to form a crack under the tensile loading. Figure 2.48 illustrates this process graphically. Bache [91] put forward a second theory to explain the formation of the facets involving a strong and a weakly orientated grain. Since the same stress is applied to both, the weak grain would ideally deform more than the strong grain. However, since Figure 2.47: Cross-colony fatigue crack initiation in lamellar Ti-6Al-4V. Reprinted from Nalla et al. [90]. 86

87 2.4. FATIGUE OF TITANIUM Figure 2.48: Slip band model to explain basal facets formed in titanium alloys. Reprinted from Bache [88]. both are microscopically constrained to the same average strain, the strong grain must support more of the stress to deform uniformly. The increased stress leads to the strong grain failing first Crack Propagation Crack propagation has already been related, through fracture mechanics, to the stress intensity factor in subsection When predicting crack growth with LEFM, it must be assumed that conditions near the crack tip are identical when the driving force, K I, is identical, regardless of specimen size and crack geometry. This concept is known as similitude and occurs when cracks are large relative to the microstructure and the tip encounters many grains of various orientations [94]. However, small or short cracks can propagate at velocities far faster than long cracks subjected to the same driving force, K I. Figure 2.49 illustrates schematically this effect for both small and short cracks in comparison to long crack propagation. The small scale of the crack, relative to the microstructure, means that the use of continuum mechanics is inappropriate, and is a major reason for this apparently atypical growth [94]. Small cracks can be defined as those similarly sized to the characteristic microstructure (microstructurally small) or to the plasticity around the crack tip (mechanically small). In addition, those physically small, i.e. less than a couple of millimetres, where crack closure is not fully developed [118], or chemically small, where the growth of small cracks is accelerated by stress corrosion cracking, can be classified as small cracks [94]. The terms, small crack and short crack are occasionally used interchangeably in the literature, however, in other cases a distinction is made [118]. Short cracks are often defined as those that are completely contained within the strain field of a stress concentrator such as a notch [94]. Alternatively, short cracks can in- 87

88 CHAPTER 2. LITERATURE REVIEW Figure 2.49: Schematic comparison of crack growth behaviour for small, short and large cracks. Reprinted from Ritchie et al. [99]. clude those that are only short in one direction, e.g. after removing material from a cracked component leaving only a wide crack tip [90]. Small/short cracks can be particularly problematic as they often exist below the threshold level for detectable cracks and the threshold stress intensity range below which long cracks will not propagate ( K T h ) is no longer applicable. Small cracks can grow even when larger cracks, subjected to the same driving force, K I, would not propagate [90]. To image short crack growth in three dimensions, Birosca et al. [119] has used a synchrotron X-ray beam to collect high resolution XCT data in situ during a fatigue experiment. Absorption and phase contrast tomography was utilised to detect both the surface crack and the grain boundaries respectively. Testing was carried out on a notched fatigue sample of the α + β alloy Ti-6Al-2Sn-4Zr-6Mo, heat treated to produce a Widmanstätten microstructure, with prior β grains of approximately 100 µm and 3 µm to 5 µm α lath widths. By taking an XCT scan each time they observed a difference in the 2D projections, they were able to quantify the short crack growth in 3D. Each XCT scan took approximately one hour where the sample was held at the maximum load of the fatigue cycle. In effect, the test was one of dwell fatigue, which is known to change the fatigue behaviour of some titanium alloys [91]. Figure 2.50 illustrates the crack front recorded by XCT after several different numbers of cycles. Birosca et al. noted that the crack deviated from a perfect ellipse shape, which they referred to as first order undulations. The smaller, local variation in the crack front, giving it its rough appearance, was called second order undulation. With regards to the first order undulation, they saw that initially the lower half of the crack in Figure 2.50 grew faster than the upper half, then the trend was reversed. The phase contrast XCT showed that this was a result of crack bifurcation at prior β grain boundaries, which resulted in local crack shielding. When crossing a β grain boundary the crack would 88

89 2.4. FATIGUE OF TITANIUM Figure 2.50: Crack fronts at various numbers of fatigue cycles measured by XCT. Crack front thresholds (left) and a mirror image showing cracks in different colours. Mirror plane the is sample surface. Reprinted from Birosca et al. [119]. first be diverted before a secondary crack tip would form in the direction predicted by LEFM for mode I crack growth. The secondary crack would then grow more rapidly and replace the initial crack as the primary crack. Crack bifurcation was observed only away from the sample surface, thought to be due to constraint in the bulk material [119]. In contrast, the second order crack undulation was attributed to the α lamellar microstructure, with high angle grain/colony boundaries acting as a barrier to crack growth [119]. In subsection 2.2.4, it was stated that a finer lamellar microstructure results in greater resistance to small crack propagation. In lamellar microstructures, colony boundaries act as obstacles for crack growth and cracks have been observed to follow the edge of α colonies or cross them at 90 [120, 121, 122]. The cracks tend to propagate with slip in the a direction, and are arrested when they encounter a grain favourably orientated for c + a slip, but unfavourably orientated for a slip [123]. It has been found that, in lamellar microstructures, cracks tend to propagate across lamellae orientated for basal slip. Whereas, in a duplex microstructure, cracks tend to propagate through grains preferentially orientated for prismatic slip [122]. The difference in preferred crystallographic planes has been suggested to be due to the fact that cracks need to cross α/β boundaries to travel through a colony of lamellar. As such, the crack follows a plane that is preferential for slip in both the α and β phase. As a result of the Burgers orientation relationship the α basal plane contains a Burgers 89

90 CHAPTER 2. LITERATURE REVIEW Figure 2.51: Schematic diagram of surface crack growth behaviour observed in a Ti-8.6Al alloy with an equiaxed microstructure. Reprinted from Hall [120]. vector parallel to one of the vectors in the(111) plane of theβphase, which is not the case for the prismatic plane [122]. Crack arrest at grain boundaries can result in fatigue crack growth rates being more scattered and less predictable when cracks are short compared to when they are long enough for LEFM to be applied [120]. Figure 2.51 illustrates schematically the crack growth behaviour observed in a Ti-8.6Al alloy with an equiaxed microstructure and the more unpredictable short crack growth regime. In this alloy, the crack was arrested at grain boundaries, but the same effect has been observed in other alloys with a lamellar microstructure at colony boundaries. Once a crack has reached a certain size, similitude is achieved and the crack growth rate can be characterised with LEFM. In addition, as cracks get larger, growth can be slowed by crack closure induced by plasticity, roughness, oxide growth, viscous fluid impregnation or phase transformation. All crack closure mechanisms become more likely to occur when the load ratio (R) is low. Long cracks also become less sensitive to microstructure effects [120]. Plasticity induced closure occurs due to the plastic deformation around the crack tip (Figure 2.52a). The size of the plastic deformation zone increases with crack length (Figure 2.52b) and when the crack advances this results in an envelope of previous plastic zones developing around the crack (Figure 2.52c). The residual tensile plastic strains within this envelop result in the crack closing before the stress is completely removed; any further reductions in the apparent K min do not effect the crack. The stress intensity factor below which the crack will remain closed is denoted K op. If, during the loading cycle, the stress intensity falls below K op, the 90

91 2.4. FATIGUE OF TITANIUM (a) (b) (c) Figure 2.52: Development of plastic zone around advancing crack. Reprinted from Suresh [94]. effective stress intensity range is reduced to K e f f = K max K op. This reduction in stress intensity factor range thus leads to a reduction in crack growth rate due to crack closure effectively shielding the crack tip [94]. Crack closure induced by roughness results from slip preventing alignment of the crack fracture surfaces. A rough fracture surface may then overlap in this new location. If the crack opening displacement is smaller than the overlap between the surfaces they will never fully open; once again leading to a slower crack growth [94]. Roughness induced crack closure is promoted by coarser grains, which themselves result in a rougher more uneven fracture surface. In contrast to short crack growth, long crack propagation is thus hindered more effectively by larger grains or colony sizes [120, 124]. Large grains/colonies can cause cracks to divert from the plane predicted by LEFM, i.e. normal to the direction of load, when they are unfavourably orientated for slip [94]. When the crack is diverted from the normal growth plane, the effective driving force is reduced, and thus, the crack growth is reduced. Near the crack tip the loading conditions can also become mixed mode, even when the far field loading is purely mode I. Additionally, if crack growth is measured in the direction predicted by LEFM, the apparent crack growth will be slower. Stress concentrators, very important in terms of crack initiation, can also affect the rate of crack propagation. The stress and strain concentrations around a crack can be affected by pores in close proximity. In the XCT work by Li et al. [112] mentioned previously, it was found that the Paris equation (Equation (2.8)) could be modified by replacing K with the geometric mean of the stress and strain concentrations (k g ). After calculation of values for C and m, using XCT and FE models to predict k g, it was possible to predict the local crack growth rate. Cracks grew fastest where k g 91

92 CHAPTER 2. LITERATURE REVIEW Figure 2.53: Influence of pores on fatigue crack growth in laser welded Ti-6Al-4V showing a reduction in growth rate following the crack breaching a pore. The inset images show the porosity involved. Reprinted from Tsay et al. [125]. was highest, and thus crack front irregularities, due to non uniform growth, could be predicted [112]. In contrast, when using CT specimens to examine the effect of porosity on longer fatigue crack growth rates in Ti-6Al-4V SLM [103] and laser welds [125], it was found that small pores had little influence on the crack growth rate. Tsay et al. [125] found that in laser welds the instantaneous growth of cracks as they breached pores was compensated for by a reduction in growth rate as pores acted to blunt the crack tip and reduce the stress concentration. The overall result was a growth rate very similar to that observed in material with no porosity present. These opposing effects are illustrated in Figure

93 2.5. POROSITY AND DEFECTS IN POWDER BED AM 2.5 Porosity and Defects in Powder Bed AM It has been shown that the presence of pores within SEBM Ti-6Al-4V samples can significantly influence the high cycle fatigue life, despite comprising only a minor fraction of the volume (0.1 % to 0.5 % [26, 126]). The size, morphology and location of the pores is crucial in determining their influence on crack initiation [7, 10]. To date, the amount of literature on the pores and how they form is somewhat limited. Therefore, a review of the factors that affect porosity in other AM processes is included in this section Typical Appearance and likely Origins of Internal Defects in Powder Bed AM Standard metallographic examination on polished sections from powder bed AM samples has revealed that the most common pores have a circular profile [29, 63, 81, 127, 128] and are relatively small (<100 µm). These defects are thought to be spherical gas pores caused by bubbles becoming trapped in the melt pool during solidification. In SLM, which uses an inert gas atmosphere, some of these pores likely originate from shielding gas becoming trapped during densification of the powder [129]. This is less likely to occur with high vacuum processes like SEBM, but all powder based techniques are still liable to porosity if there is contamination of the powder [63]. Section outlines how argon bubbles can become trapped within powder particles during their manufacture. With SEBM, using standard metallography, generally it has been assumed that the gas pores are randomly distributed [29] and little has been reported on the influence of the process variables on their location. Figure 2.54a and b illustrate two examples of gas pores observed by optical microscopy and SEM. Gaytan et al. [63] state that because of the melt and liquid phase surface tension characteristics of Ti-6Al-4V as well as the low bubble (gas) pressure, it is essentially impossible to eliminate these intrinsic gas bubbles in Ti-6Al-4V manufactured products. When Loeber and Biamino [37] examined titanium alminide components produced by SEBM, the only type of porosity they observed was that resulting from trapped argon gas, despite noting a much higher volume fraction of porosity ( 2 %). A separate study by Ahsan et al. [130] on direct metal deposition of Ti-6Al-4V, has discussed the defects formed in relation to a range of laser powers and mass flow rates in addition to using both gas atomised and PREP powder. Ahsan et al. used X- ray computed tomography (XCT) to quantify the volume fraction of porosity in both the powder feedstock and the consolidated material. They found that although the volume fraction of gas pores in the virgin powder granules affected the final volume 93

94 CHAPTER 2. LITERATURE REVIEW (a) (c) (b) (d) Figure 2.54: Examples of pores seen in SEBM: gas pores observed with (a) optical micrograph and (b) SEM; (c & d) lack of fusion defects. (d) also shows a gas pore at the arrow. Adapted from Gaytan et al. [63]. fraction of pores, other factors were more important [130]. Laser power and the mass flow rate could both alter the volume fraction of porosity more significantly than the prior porosity in the powder type alone. Moreover, all the pores they identified were spherical and thus likely to be due to gas in the powder. Hence, by altering the laser power and mass flow rate, gas bubbles trapped within the powder feedstock were given the opportunity to escape the melt pool. From these results it becomes apparent that the statement by Gaytan et al. [63] is too general; it seems likely that it is possible to alter the melt strategies in SEBM to remove gas bubbles from the melt pool before they form pores. While less common, in SEBM other defect types have also been reported that are associated with undesirable process conditions [29, 63, 128, 77]. Such defects generally result from a lack of fusion between granules of un-melted powder and can be more damaging because of their larger size (e.g.>200 µm [10, 63, 77, 128]) and high aspect ratio. Figure 2.54c & d gives two examples of lack of fusion defects occurring due to regions of powder remaining un-melted [63]. Hrabe and Quinn [128] suggest that they arise from local peaks in the powder thickness due to poor spreading of the powder. The build direction is not given on Figure 2.54, however, other authors have noted that in SEBM, lack of fusion defects tend to be orientated such that they are elongated in the x-y plane (i.e. the plane of the powder layers) rather than the z direction [10, 128]. The angle of lack of fusion defects within Inconel specimens manufactured by SLM has also been found to increase with line offset (distance between melt passes) until the line offset reaches the melt pool width, when the defects become vertical [131]. 94

95 2.5. POROSITY AND DEFECTS IN POWDER BED AM (a) (b) (c) (d) (e) (f) Figure 2.55: Experimental (a, c & d), left, and modelling (b, d & e), right, images of tunnel defects produced with a beam speed of 800 mm s 1 and power of: (a & b) 90 W; (c & d) 120 W; and (e & f) 180 W. Adapted from Bauereiß et al. [76]. Other large defects have been observed that can grow upward, through many layers, parallel to the build direction. These tunnel defects were initially attributed to none optimum settings or the beam being interrupted, leading to them being perpetuated through the build [63]. Figure 2.55a, c, & e illustrates recently produced experimental examples by Bauereiß et al. [76] of this type of defect for varying beam power in SEBM. The lattice Boltzmann modelling shown alongside the experimental results on Figure 2.55 confirmed that these defects are more likely to occur when the beam power is lower. The modelling results also demonstrated that the tunnel defects can be caused when surface tension and wetting effects overcome gravitational forces. Thus, when the melt pool overlap is insufficient, molten particles coalesce with the first solid or liquid surface they contact, which is often nearby particles rather than the previous solid layer [76]. This situation is made more complex by the random packing of particles. However, even if particles fill the cavity it is possible for the lateral surface tension forces to pull them out again [76]. Figure 2.56 illustrates the process by which these tunnel defects form. From Figure 2.55 it is clear that increasing the beam power reduces the probability of these defects occurring. The larger melt pool size associated with increased beam energy results in a more continuous melt pool and less lateral surface tension effects [76]. In parts produced with optimised parameters, fatal fatigue cracks have been mainly reported to initiate at near-surface gas pores rather than at larger, higher aspect ratio lack of fusion pores [7, 10]. This is likely to be due to their higher frequency in the 95

96 CHAPTER 2. LITERATURE REVIEW Figure 2.56: Lattice Boltzmann modelling of the formation of tunnel defects with a beam power of 90 W and translation of 800 mm s 1. For each layer, the first (left) image shows the layer before the powder is applied, the second shows the layer with powder and the third shows the temperature field during melting. In the last picture violet regions are liquid and the beam is light grey. Adapted from Bauereiß et al. [73]. 96

97 2.5. POROSITY AND DEFECTS IN POWDER BED AM deposited material and thus greater chance of being located close to the surface, as their more rounded morphology would produce a lower stress concentration than the more irregular lack of fusion pores [110] Influence of Beam Energy The influence of energy input on defect populations in selective beam melting AM has been investigated by a number of authors [65, 72, 76, 80, 132, 77, 74, 129, 133]. Most of these studies were concerned primarily with the elimination of the lack of fusion defects. It has already been shown that the large tunnel defects are made more probable by insufficient energy input [76]. Two common parameters used to describe the energy input into a build are the energy density and line energy. The line energy is regularly used in welding to describe the heat input and is the ratio of beam power over beam speed, previously defined in Equation (2.2). The applied energy density per unit volume (E v ) is a parameter often used in selective laser melting to compare the effect of different process parameters [80, 132]. It defines the local heat input per unit volume with respect to the beam speed, power and offset between melt tracks, which can be varied simultaneously. Thus, energy density (J mm 3 ) is a useful parameter for comparing different samples and is given by: E v = P v h t (2.13) where: P, v, h and t are the beam power (W), beam velocity (mm s 1 ), line offset (mm) (spacing between melt tracks) and layer thickness (mm), respectively. It should be noted that Equation (2.13) should not be viewed as the net energy input, as it neglects coupling with the work piece, but it is a useful parameter for benchmarking the relative energy input between different process settings. It would be possible to more accurately estimate the heat input per unit volume by including an efficiency factor in Equation 2.13, however, since energy density is generally used to compare samples manufactured with the same equipment, this is rarely applied. With SEBM using the Arcam machine, a constant voltage (60 kv) is maintained during processing so that the power (P = current voltage) is proportional to the beam current only. The nature of lack of fusion defects indicates that when energy input is lower they will become more prevalent because of the corresponding smaller melt pool and less melt track overlap. This was confirmed by Puebla et al. [77], who demonstrated that when using a constant electron beam power, the volume of defects increased with increasing beam speed. The increased speed and, thus, faster cooling rate, also led to finer α plates and an increase in microhardness. Conversely, the macrohardness and Youngs modulus reduced due to the increased prevalence of porosity. Puebla et al. 97

98 CHAPTER 2. LITERATURE REVIEW Figure 2.57: Influence of process setting of pore volume fraction in SEBM Ti-6Al-4V samples. Reprinted from Gong et al. [72]. also noted that when manufacturing cylinders of the same geometry, there tended to be more lack of fusion porosity in cylinders orientated vertically rather than horizontally. They suggest this is due to the increased heat loss from the vertical cylinders. However, since one of the primary aims of the control software used during SEBM is to maintain a constant build temperature this seems unlikely. Other factors, such as the shorter melt track length resulting in more beam path changes, are perhaps more likely to be the reason for the increased number of lack of fusion defects (see discussion in Chapter 5). A more systematic study of the effect of variation of the process parameters on defect populations was carried out by Gong et al. [72], who also compared SEBM to SLM. They reported that energy density has a significant effect on defect population in both SEBM and SLM. The pore volume fraction was measured by employing the Archimedes method in SEBM samples whilst varying the current, focus offset, line offset and speed function. Figure 2.57 summarises the results of their analysis of porosity volume fraction in SEBM samples. Due to the control software of the Arcam machine, which automatically calculates the beam power and speed, they were unable to quantify the exact energy density for the samples, instead, they state the input values used. They found the most influential factor was the speed function, followed by the line offset, focus offset and maximum current [72]. An increased energy density is associated with decreasing either the speed function or line offset, both of which were found to be effective in decreasing the measured pore volume fraction. The relationship 98

99 2.5. POROSITY AND DEFECTS IN POWDER BED AM Figure 2.58: Schematic of influence of focus offset on melt pool geometry. Dotted line indicates the layer thickness. Reprinted from Gong et al. [72]. with the focus offset was less clear, with a minimum level of porosity recorded when the focus offset was 10 ma. With both a smaller focus offset (more intense beam with a smaller diameter), and a larger focus offset (less intense beam with a larger diameter), the volume of porosity increases. They suggest that the increase in porosity with a high focus offset is due to the beam penetration depth no longer being deep enough to fully melt the powder layer, illustrated schematically on Figure Whereas, they do not discuss the reason for the increase in porosity with low focus offset. Furthermore, Gong et al. noted that there appeared to be an increase in the number of small gas pores as the focus offset was increased, although they were unable to detect this using the Archimedes method. Additionally, Gong et al. [72] offered no explanation as to the slight reduction in porosity they observed with increasing current, since the effect of the speed function, which increases beam speed when power is increased, is to keep the energy density approximately constant even when different currents are used [72]. Recently Juechter et al. [65] showed that, in addition to the net energy input, the return time (how long until the beam returns to the same location on the adjacent melt line) can influence defect populations. They showed that when the beam is moving very rapidly, the retained heat from the previous lines can be used to assist melting the current line, leading to bigger melt pools and larger heat affected zones (HAZ). To approximate the hatching speeds required to achieve this, they assumed a critical thermal diffusion depth of approximately three times the layer thickness ( m), based on the assumption that a melt pool size of twice the layer thickness was required for fully dense parts. The thermal diffusion depth (d di f f ), is related to the time available for diffusion, t di f f (s), and the thermal diffusivity, k (m 2 s 1 ), by Equation (2.14). d di f f = 4 k t di f f (2.14) Thermal losses are greatly reduced when the return time is less than the diffusion time for a distance equivalent to three layers [65]. The beam return time, t return (s), is a simply the length of the melt, L (m), divided by the beam speed, v scan (m s 1 ). Thus, the critical speed for hatching (v crit.hatch ), above which thermal losses by diffusion can 99

100 CHAPTER 2. LITERATURE REVIEW Figure 2.59: Processing window of Ti-6Al-4V processed by SEBM proposed by Juechter et al. [65]. The area enclosed by dashed lines constitutes the processing window with dense and smooth parts. be neglected, is given by Equation (2.15). v crit.hatch = 4 L k d 2 di f f (2.15) The critical hatch return speed for a 15 mm length was estimated to be 10 m s 1 [65]. A similar logic was used to calculate the speed required to avoid strong local losses over the time the beam is interacting with the powder, using the interaction time given by the beam diameter (d beam = m) divided by the beam speed. A critical velocity to avoid large local loses was calculated as 0.2 m s 1, using Equation (2.16). v crit.local = 4 d beam k d 2 di f f (2.16) To support these arguments, Juechter et al. produced the processing window for 15 mm square Ti-6Al-4V specimens, based on the beam speed and line energy (Equation 2.2) shown in Figure The upper and lower boundaries of the window, given by the dotted lines, are where the upper surface was judged to become uneven, and the volume fraction of porosity, measured by optical microscopy, exceeded 0.5 %, respectively. Despite the data they present never reaching the critical hatching speed, it indicates that the required line energy for dense parts reduces with increasing beam velocity. In addition, the increased gradient in the window at very low speeds is likely to be due to the increased local heat losses (Equation 2.16). Juechter et al. [65] do not record the type of porosity that was encountered in this 100

101 2.5. POROSITY AND DEFECTS IN POWDER BED AM work. Nonetheless, it is clear that the volume of pores was reduced as the melt pool size increases due to the increased residual heat. However, they also note that drawbacks of increasing the melt pool size include: an uneven top surface; and a loss of aluminium, with larger melt pools and higher surface temperatures leading to increased aluminium evaporation within the vacuum chamber. The reasoning of Juenchter can also be used to explain the lower porosity found with higher beam currents in the work of Gong et al. [72] (mentioned previously). Despite the higher currents being accompanied by higher beam speeds, and thus, similar energy densities, the return time is reduced, leading to more residual heat and larger melt pools. Juenchter s argument can be used to explain the reduction in porosity observed. When Gong et al. [72] compared the effect of energy density in SLM they found that higher energy densities did not necessarily result in more dense parts. In addition, at very high energy densities, the high thermal gradient led to distortion of the part and subsequent failure of the powder deposition system. A number of other authors have confirmed that the relationship between porosity and energy input in SLM is not straightforward, with many individual factors also influencing the volume porosity [80, 132, 133, 134]. For example, at low beam scan speeds, the melt pool can become unstable due to the increased importance of the hydrodynamics [80]. Instability of the melt pool can lead to an effect known as balling [32, 132, 134]. Balling is a complex process with a number of possible causes. However, it is generally attributed to a too high laser power [32, 132, 134]. The low viscosity can lead to the surface tension effects overcoming the wetting of the previous layer and the formation of melt balls [73, 74, 134]. Balling can occur regardless of the beam scan speed, with some authors noting it at high speed [134] and some at slow speeds [32]. Balling has also been reported in work utilising stainless steel powder with SEBM, with a speed either higher or lower than an optimum value [74]. Figure 2.60a shows an example of melt balls formed during SEBM. Balling has been shown to become more probable when there is oxidation of the powder, and thus, less wetting [32], and can also be influenced by the local powder arrangement [73]. Figure 2.60b shows an example of a defect known as delamination, rarely reported in SEBM [74] and more often in SLM [32, 132, 134]. Delamination occurs when the bonding between layers is not strong enough to resist the residual stresses brought about by cooling [32, 74]. In SEBM delamination has been found to occur when high scanning speeds, and thus lower energy densities, were used [74]. Moreover, the melt strategy employed, which influences the residual stresses, can have an important effect (subsection 2.5.3). A search of the literature has revealed little on the influence of energy density on gas pores. However, an early investigation into SLM of stainless steel powder revealed a possible method for the removal of shielding gas bubbles trapped during densifica- 101

102 CHAPTER 2. LITERATURE REVIEW (a) (b) Figure 2.60: Defects encountered during SEBM of stainless steel powder: (a) Melt ball formation; and (b) delamination. Reprinted from Zäh and Lutzmann [74]. tion of the powder. Morgan et al. [129] demonstrated that by remelting a previously deposited layer, without additional material deposition, entrapped gas was given an opportunity to escape. This additional step increased the measured relative density from approximately 98 % to consistently over 99 %. Nonetheless, it is worth noting that some of this reduction in porosity may be due to a smoother upper component surface, leading to a reduction in the initial entrapment of the shielding gas [129] Influence of Melt Strategies In addition to the beam energy, the scanning strategy can influence the defect population within components manufactured by selective beam melting of powder. The standard Arcam SEBM strategy is to first melt the outline of the layer with a strategy known as contouring (this can involve more than a one pass), followed by a snaking infill strategy known as hatching, see subsections and When Karlsson et al. [29] attempted to use non-standard powder sizes, layer thicknesses and beam speeds, they observed that the porosity generated was not randomly distributed. Figure 2.61 shows cross sections, in the x-y plane, from these samples. In Figure 2.61a, the standard build, the distribution of pores appears random. Indeed this is the conclusion that Karlsson et al. drew: as pores are caused by argon gas trapped within the feedstock powder, they are randomly distributed. The same conclusion was reached for a build produced using standard powder, but with thinner layers (Figure 2.61b). Karlsson et al. also suggested that some of the white spots seen are in fact silica particles, a residue from polishing, and that the porosity volume is lower than it initially appears. Furthermore, this polishing residue would likely increase the random appearance of the pores. The remaining two images on Figure 2.61c & d, both show clear evidence of non-random distribution of the pores. In Figure 2.61c a sample produced with standard layer thickness, but smaller powder, the pores appear in straight lines at an approximately constant distance from the edge. Karlsson et al. 102

103 2.5. POROSITY AND DEFECTS IN POWDER BED AM (a) (b) (c) (d) Figure 2.61: Overview of pores (white) in cross sections (x-y plane) from Ti-6Al-4V SEBM samples produced with different powder sizes and layer thicknesses obtained by optical microscopy: (a) 45 µm to 100 µm powder and layer thickness 70 µm (standard settings); (b) 45 µm to 100 µm powder and layer thickness 50 µm; (c) 25 µm to 45 µm powder and layer thickness 70 µm; (d) 45 µm to 100 µm powder and layer thickness 50 µm. Reprinted from Karlsson et al. [29]. 103

104 CHAPTER 2. LITERATURE REVIEW suggest that this indicates they are caused by insufficient melting between the contouring and the hatching strategies. The pores in Figure 2.61d, with both smaller powder size and layer thickness, are far larger and at a similar distance from the surface. In this work they were attributed to variation in thermal conductivity, due to the smaller powder and layer. However, Karlsson et al. did not suggest what the variation would be. From Figure 2.61c & d it is clear there is some influence on porosity populations from the interface between the two melt strategies. In SEBM the melt strategy is defined by the commercial supplier of the equipment, Arcam AB, and most studies use either the prescribed settings or only make minor alterations to beam speed and power, such as in the work of Karlsson or Juenchter. In contrast, Thijs et al. [80] examined three SLM specimens with the same energy density but three separate melt strategies. These consisted of: (i) zigzag hatching with each adjacent melt line rotated by 180 from the preceding line and all layers identical; (ii) unidirectional hatching with identical layers; and (iii) zigzag hatching, but the initial hatch direction rotated by 90 each layer. Relative densities of 99.6 %, % and 99.9 % were recorded for the three strategies, (i), (ii) and (iii), respectively. The strategy that produced the best results, (iii), is the same strategy that is applied during SEBM. Thijs et al. do not describe a reason for the improvement in relative density seen with this scanning strategy. However, it seems likely it is due to a more uniform distribution of heat energy input, i.e any region of a layer that did not receive sufficient heat energy during melting of any given layer would be more likely to receive sufficient energy in the proceeding layer. This is in contrast to the other two samples, where repeating identical melt strategies in each layer would lead to certain regions receiving consistently less energy and increasing the probability of lack of fusion defects Quantification of the Defect Population in SEBM Most studies on defects in SEBM have either only reported the defect size distributions in broad terms (i.e. less than 100 µm), or have only stated the total volume fraction [29, 63, 81, 127, 128]. However, some conference proceedings have included quantification of individual pore sizes in SEBM Ti-6Al-4V samples. Smith et al. [135] utilised XCT to quantify the pore sizes in three dimensions, while Gong et al. [126] used standard two dimensional optical metallography to characterise the porosity with a number of line and focus offsets. Figure 2.62a, presents the histogram produced by Smith et al. for the pore size distributions within a structurally optimised cantilever beam built by SEBM. It is clear that the pores identified are somewhat larger than generally stated in the literature and the modal size of the pores (124 µm to 155 µm) is above the general consensus of <100 µm. A possible explanation for these apparently anomalous results is that the relatively low resolution excluded typical small gas pores from the 104

105 2.5. POROSITY AND DEFECTS IN POWDER BED AM Figure 2.62: Pores in a cantilever beam: (a) Histogram of pore sizes within cantilever beam; (b) visualisation of the XCT data, pores indicated in red. Adapted from Smith et al. [135]. results, with only pores greater than 61 µm identifiable. Moreover, visual analysis of the data (Figure 2.62b) confirmed that the pores identified had the appearance of tunnel defects, rather than spherical gas pores or small lack of fusion defects [135]. Finally, Smith et al. cautioned that some of the regions identified as pores were in fact artefacts introduced as a result of the XCT reconstruction of such a complex part. In contrast, the relatively course bin size histograms of pore areas plotted by Gong et al. [126] showed small pores dominating the distributions. In this work the pore sizes were given in terms of area, which makes comparisons of the histograms challenging. However, it is clear that for all the line and focus offsets examined, pore sizes below µm 2, were found to be the most common. This pore area would correspond to an equivalent circular diameter (subsection 3.4.1) of 70 µm, which is near the resolution limit of Smith et al. [135]. In addition, when the line offset and focus offset were varied from their optimum values, the major influence was an increase in the number of small pores, although some large defects were also recorded. 105

106 CHAPTER 2. LITERATURE REVIEW (a) (b) Figure 2.63: Histogram of pores sizes measured by image analysis (in 10 3 µm 2 ) in Ti-6Al-4V samples manufactured with various line offsets (a) and focus offsets (b). The standard line and focus offset is show on the leftmost plot. R D indicates the pore volume measured by the Archimedes method, while R I indicates the pore volume measured via image analysis. Adapted from Gong et al. [126]. 106

107 2.5. POROSITY AND DEFECTS IN POWDER BED AM Pore Formation in Energy Beam Welding The SEBM process, and indeed most AM processes, can be considered as numerous overlapping consecutive welds, where the filler material and weld bead is used to manufacture the final component. Pores have previously been observed in both laser and electron beam welded Ti-6Al-4V [136, 137, 138, 139]. The melt pool associated with laser and electron beam welding can have a very different geometry to the melt pool associated with SEBM. Very high line energies can lead to the temperature at the centre of the weld becoming higher than the boiling point of the work-piece. The escaping vapour produces a recoil pressure displacing liquid metal from the weld centre and exposing new material to the heat source. The vapour hole that forms in the material is called a keyhole [140]. Figure 2.64 illustrates schematically the keyhole melt geometry and the forces acting. As the beam moves forward, material is melted at the leading edge of the keyhole and moves around to fill the keyhole at the trailing edge. For a detailed description of the process the reader is directed to the literature [140]. During keyhole welding, pores can be generated by instability of the keyhole, entrapment of the shielding gas, chemical reactions in the melt pool, or evaporation of low boiling point elements [140]. For example, if the heat source is removed the upper part of the keyhole can solidify first and prevent the liquid melt flowing down to Figure 2.64: Schematic diagram of keyhole and melt pool formed during electron beam welding. Reprinted from Stone [141]. 107

108 CHAPTER 2. LITERATURE REVIEW Figure 2.65: Calculated growth of bubble, with initial radius of 20 µm, in an electron beam weld melt pool due to hydrogen diffusion. The influence of different levels of hydrogen concentrations in the liquid is also shown. Reprinted from Huang [137]. fill the keyhole [142]. However, most pores in Ti-6Al-4V electron beam welds are attributed to impurities in the melt, particularly hydrogen, but also oxygen and moisture [136, 137]. Spherical cavities observed in the consolidated weld are assumed to be caused by gas bubbles in the melt, very similar in appearance to the gas pores seen in SEBM Ti-6Al-4V. Gas bubbles are thought to be caused by contaminants in the melt pool and are found to be more prevalent when there is a greater roughness or contamination at the surface [136]. Asperities between the touching surfaces of the pieces to be welded can allow micro-bubbles of insoluble gases to exist in the melt. Diffusion of soluble gases into these micro-bubbles result in expansion of the bubbles, and pores in the welded region following solidification of the melt pool. The high internal pressure required to nucleate a bubble in a melt pool suppresses homogeneous nucleation and makes heterogeneous nucleation of bubbles more likely. A model to predict the growth of micro-bubbles in a melt due to the diffusion of hydrogen has been provided by Huang et al. [137]. They derived the equations that governed the diffusion of hydrogen and growth of the gas bubbles, then solved them using standard finite difference methods. They assumed a melt temperature of 2000 K and an ambient pressure of 10 4 mbar, identical to the intended pressure within the vacuum chamber during SEBM. Figure 2.65 illustrates the predicted growth of a bubble, with initial radius 20 µm, due to three different hydrogen concentrations. A higher hydrogen content can be seen to be accompanied by a predicted increase in bubble growth rate. In addition, with the conditions modelled, the bubble did not grow at all with a hydrogen concentration of 100 ppm, indicating there must be a critical level below which there will be no growth. For a 30 µm bubble this was estimated to be 200 ppm, whereas for a 5 µm this increased to 450 ppm. In addition, reducing either the ambient pressure or surface tension meant that a decreased hydrogen content was 108

109 2.5. POROSITY AND DEFECTS IN POWDER BED AM predicted to be required for bubble growth. Since the surface tension of the titanium melt is inversely proportional to the temperature, it appears that bubbles are more likely to grow when the electron beam has a higher line energy (higher temperature). Indeed, when different average powers were tested during laser welding of Ti-6Al-4V, it was found that higher powers were associated with a higher volume fraction of porosity [139]. Alternately, a very good vacuum would also increase the rate of bubble growth Movement of Gas Bubbles within Melt Pools Even if a gas bubble is contained within a melt pool there is a possibility that it may escape prior to solidification. This is known to occur in electron beam welding due to vigorous convection and stirring of the weld melt pool, leading to pores only being observed intermittently [136, 137]. The escape of hydrogen from a melt pool was demonstrated by Huang et al. [137], who observed a much more significant rise in the vacuum chamber pressure when electron beam welding hydrogen charged Ti-6Al-4V plates, as compared to standard plates. Stokes law can be applied to the floatation of bubbles to calculate the upward velocity due to buoyancy forces (v) of a bubble in a melt pool of low Reynolds number, Equation (2.17) [143]. v= 2 9 g r2 ρl ρ g µ (2.17) where: g is the acceleration due to gravity; r is the gas bubble radius; ρ l and ρ g are the densities of the liquid and gas bubble respectively; and µ is the viscosity of the liquid. Equation (2.17) also neglects to include any movement due to convection or Marogoni effects in the melt pool. From Equation (2.17) it is clear that the larger bubbles will rise faster than small bubbles, since the velocity is proportional to the radius squared. Thus, larger bubbles will collide with slower moving smaller bubbles and coalesce forming an even larger bubble. Bubbles will rise until they escape the melt pool or are overtaken by the solidification front Surface Roughness A rough surface can provide ready sites for fatigue crack initiation [94] as valleys in the surface profile act as stress concentrators. The surface roughness of components manufactured by SEBM has been noted as being worse than both those manufactured by casting [62], and SLM [83, 100] using the same component design. Other authors have also expressed concerns about the suitability of SEBM produced components for fatigue loaded applications [14, 29]. Figure 2.66 illustrates the difference in the asbuilt surface finish between SLM and SEBM, with the high surface roughness being 109

110 CHAPTER 2. LITERATURE REVIEW Figure 2.66: External surfaces of cylindrical samples built: (a) vertically by SLM; (b) horizontally by SLM; (c) vertically by SEBM; (d) horizontally by SEBM. In (a) and (c) the build direction is indicated by the arrow, on (b) and (d) the build direction is perpendicular to the image plane. Reprinted from Rafi et al. [83]. primarily caused by the presence of partially melted powder particles attached to the part faces. Despite this, there is little numerical information concerning the surface roughness. Al-Bermani [14] recorded a roughness average (R a ) of 28.1 µm in the z axis, and 31.2 µm perpendicular to the z direction. In contrast, Chan et al. [100] measured the R a in eight thin plates (4 mm) to be a maximum of µm with a standard deviation of µm. However, closer examination of the results reveals that it is likely that Chan et al. [100] mislabelled their data. It appears that the value they quote is the maximum peak to trough distance, R t, not the roughness average, despite their labelling the value R a. This is somewhat disappointing as they have then gone on to demonstrate a dependence of the mean fatigue life on the surface roughness. Chan et al. [100] manufactured Ti-6Al-4V components with a variety of techniques including SEBM, casting and rolling, and showed that those components with the highest surface roughness also had the shortest mean fatigue life. Despite the poor reporting of the surface roughness, it is clear that the surface roughness of as-built components is an area of significant concern to the application of SEBM manufactured components. In addition to the roughness of the vertical surfaces, the upper surface also displays significant variation in height. Bauereiß et al. [76] noted that when melting with a layer thickness of 50 µm the amplitude of the roughness of the final layer was several hun- 110

111 2.5. POROSITY AND DEFECTS IN POWDER BED AM dred µm, much greater than the nominal layer thickness. This would clearly influence the spreading of powder on the top layer and variations in powder thickness is thought to be a contributing factor to the formation of lack of fusion defects [128] Powder Production Techniques Powder bed AM techniques generally use spherical powder particles with a diameter less than 150 µm [17, 29, 133, 144]. Irregular particles, produced by water atomisation, have been used, though the results were poor compared to when using spherical particles produced by gas atomisation (GA). This was attributed to the chemical content as well as the particle morphology [144]. Within the literature, it has been assumed that the gas pores found in SEBM Ti-6Al-4V are a result of gas bubbles trapped within the powder particles. GA powder is produced by forcing molten metal through a nozzle containing a set of jets of high velocity cold inert gas, which breaks the melt stream into small droplets. These droplets freeze in flight to produce relatively spherical powder particles [145]. The particles can have small attached satellites, visible on Figure 2.67a. The average size of the particles can be decreased by increasing the gas velocity, mass flow rate and Figure 2.67: Images of powder particles containing gas pores. (a) and (b) show SEM images of GA and PREP powder respectively. Note the smoother surface of the PREP powder. An optical cross section (c) shows a gas bubble within a particle and (d) a gas bubble bursting through a particle observed by SEM. (a) and (b) reprinted from Ahsan et al. [130]; (c) and (d) from Gaytan et al. [63]. 111

112 CHAPTER 2. LITERATURE REVIEW angle between the gas and molten stream. However, the inert gas (usually argon) can become trapped in some of the powder particles as they solidify [63]. Figures 2.67c & d show examples of gas bubbles trapped within powder particles. Analysis of stainless steel GA powder revealed that the probability of a particle containing a gas bubble increased with size. In this work, small particles were found to be relatively unlikely to contain a gas bubble [146]. Another atomisation process, this time utilising jets of hot ionised inert gas, is known as plasma atomisation (PA). With this method, a wire is lowered into the zone where three plasma torches, angled between 20 to 40, converge [145]. The thermal and kinetic energy of the plasma melts and atomises the titanium. Droplets then cool in flight through the surrounding argon atmosphere, which also ensures low impurities and moisture, to produce spherical powder particles [147]. The plasma rotating electrode Process (PREP) has also been used to produce spherical powder for AM [130, 148, 149]. In this case, a plasma generated by an electric arc is used to melt a rotating electrode from which centrifugal forces eject droplets into the path of a high velocity stream of gas which further refines the droplets size. PREP powder is more expensive than GA, but is more spherical (Figure 2.67b) and free of contamination [14]. Moreover, the volume of gas bubbles trapped within particles was recorded via XCT analysis as being three times less in PREP powder compared to GA powder [130]. Schematics of the three previously mentioned production methods can be found elsewhere [145]. The lowest cost process for producing titanium powder is the hydride-dehydride (HDH) method [150]. With this technique titanium is reacted with hydrogen to produce TiH 2, which, due to its brittle nature, can then be easily crushed and milled. After removing the hydrogen (dehydriding) the resultant material can be broken into single particles [150]. Unfortunately, these chemical process steps lead to the pick-up of oxygen/nitrogen, so this produces a lower quality powder which is unsuitable for critical applications [20]. The particles also have to be spheridised mechanically to improve their flow properties [151] Hot Isostatic Pressing Hot isostatic pressing (HIPing), developed in 1955, involves the simultaneous application of heat and pressure, shown schematically in Figure The major objective of HIPing is the removal of internal porosity by applying a hydrostatic pressure. HIPing has been used on a range of materials, including ceramics, steels, aluminium and titanium alloys [152]. Pressure is applied with a gas, generally inert argon, by means of both a mechanical compressor and by heating the gas in an enclosed volume. The high temperature reduces the material yield stress, and plastic flow and diffusion occurs on a 112

113 2.5. POROSITY AND DEFECTS IN POWDER BED AM Figure 2.68: Schematic diagram of a hot isostatic pressing operation. Reprinted from Atkinson and Davies [152]. microscopic scale. Following collapse of the pores, the interface between the surfaces, which are now in contact, is bonded by diffusion [152]. Standard HIP conditions for Ti-6Al-4V are a pressure of 100 MPa and a temperature from 900 C to 926 C for two hours [26, 27, 85]. HIPing is known to heal internal porosity that acts as crack initiation sites within titanium castings, leading to as much as a twofold increase in the fatigue life [41]. Recently XCT was used to scan cast Ti- 6Al-4V samples before and after HIPing, and it was found that even the largest internal porosity was closed to below the resolution limit of the equipment used (20 µm) [153]. Some pores were found to persist following the HIP cycle but these were located close to the sample surfaces and it was suggested that micro cracks, too small to be detected by XCT, connected them to the exterior. From data available in the literature, it appears that the fatigue life of SEBM Ti-6Al-4V machined from bulk is improved following HIPing. In addition, the detected volume of porosity in SEBM Ti-6Al-4V has been reported as falling from 0.2 % to 0.0 % [26]. However, other authors have noted the presence of retained collapsed pores in polished samples following a standard HIP cycle [63]. Figure 2.69 shows examples of what are thought to be both a collapsed lack of fusion defect and gas pore. When utilising titanium aluminide powder, Loeber and Bianmino [37] reported that the volume fraction of porosity was reduced, from <2 % to <1 % following HIPing, but not eliminated. Furthermore, argon is not soluble in titanium [152]. Therefore, if gas pores are caused by argon entrapped in the powder, then it is likely that following HIPing the argon bubble will persist in the material, but at a significantly smaller size and with a higher internal pressure. 113

114 CHAPTER 2. LITERATURE REVIEW Figure 2.69: SEM images of the remains of pores following HIPing: (a) collapsed void; and (b) symmetrical void collapse. Reprinted from Gaytan et al. [63]. 114

115 2.6. SUMMARY OF THE EXISTING LITERATURE 2.6 Summary of the Existing Literature Additive manufacturing (AM) is a relatively new sector of manufacturing that has generated considerable interest from both academia and industry. Different AM technologies offer a trade off between deposition rate and resolution. Currently the highest deposition rates are achievable through wire and plasma AM, which allows rapid manufacture of large components. Whereas, selective laser melting allows much finer details to be produced, but is currently only able to produce small components at a relatively low build rate. Selective electron beam melting (SEBM) has a slightly lower resolution than SLM and a faster deposition rate. During SEBM, the whole build takes place at a high temperature, which prevents both martensite and residual stresses from being observed in the as built components, and the vacuum chamber ensures a relatively clean build. Despite these advantages, only one company, Arcam AB, currently supplies SEBM equipment. Published information regarding the melting methodology of the process and any effects this may have on the melt pool behaviour, and resulting microstructure and defect population, is consequently limited. Indeed, characterisation of defects size, geometry and frequency within SEBM components in the literature is sparse. Many authors have acknowledged the existence of porosity in built components and stated that the majority are small spherical voids, likely to be caused by gas trapped within the powder. However, only two authors have quantified the size of the defects with sufficient statistics to plot a histogram, and both of these were in conference proceedings, rather than a peer reviewed journal. The overall volume fraction of the defects has been measured to be quite low (<0.2 %), and there is some information regarding the effects of various process parameters on the volume fraction of lack of fusion defects. However, few studies to date have been published that have quantified the residual porosity in sufficient detail to allow reliable statistical relationships to be developed with the main process variables. From the variation in the as-built microstructural scale between different sized builds, and even within individual samples, it is possible to infer that cooling rates vary depending on both geometry and location. Therefore, the melt pool size and geometry may also vary. It is possible that the number and type of pores may change between samples, or at different locations within the same sample, due to these variations. Previously, it has been shown that the lack of fusion defects can be reduced by increasing the melt pool size. This has been achieved by either increasing the power or the residual heat in the sample. However, there is no reported evidence of the effect of melt settings on the probability of gas pores appearing in SEBM. Pores produce a stress concentration which has been shown to severely reduce the fatigue performance of components, manufactured with both conventional methods and AM, by providing a ready site for crack initiation. The location and size of a pore 115

116 CHAPTER 2. LITERATURE REVIEW is critical in determining the probability of a crack initiating from it. Defects near the surface are known to be more detrimental in terms of fatigue life. FE modelling has shown that this is likely to be due to an increased stress concentration around surface defects. However, there has been no systematic study of the effect of defects and their location on the fatigue life of SEBM components. Following HIPing, the pores appear to be mainly removed and the fatigue performance improves. However, some remnants of collapsed pores have been observed in HIPed Ti-6Al-4V. In addition, since argon cannot dissolve in titanium, any argon bubbles trapped in the material are likely to persist, although their size will be significantly reduced. HIPing is generally used to remove gas pores formed by the change in solubility of the gas during melting. Thus it is possible for the gas to diffuse out of the material under the HIP conditions. With AM, the elimination of the pores, without the need for additional costly process steps such as HIPing, would clearly be beneficial to the industrialisation of this new technology for aerospace applications. Therefore, for all AM platforms there is a requirement to develop a better understanding of the relationships between the process parameters and part geometry, in addition to the size, density, and spatial distribution of pores found within a component. Such information is essential in developing strategies for reducing the defect content. Furthermore, it is necessary to gain a better understanding of the effect of defects on the fatigue performance, such that the most detrimental defects can be identified and controlled. For example, being able to determine the probability of pores appearing in critical locations, such as near surfaces, is particularly important. Greater knowledge regarding the formation of porosity may allow melting methodologies to be altered to avoid defects appearing in locations determined to be the most adverse, in terms of fatigue. It is also important to be able to better link the effect of porosity to fatigue life so that their impact on the performance of an engineering component can be predicted. This step has not yet been made in the literature due to a lack of sufficient high quality, spatially resolved data. 116

117 3 Experimental Methodology This chapter describes the operating procedures used to both manufacture and analyse the AM samples studied in this project. The AM system under investigation in this project was an Arcam S12 SEBM machine, which utilises a focused electron beam to preheat and melt layers of titanium powder. The operating procedure for the Arcam and the associated equipment is thus provided, alongside a description of the melt strategies employed. All the samples manufactured during this project are described, including their geometry, alongside any modifications made to the standard Arcam melt strategies. In addition, the techniques employed to characterise the porosity and its effect on the fatigue performance of the Ti-6Al-4V material investigated are described. Alongside traditional optical and electron microscopy, X-ray computed tomography (XCT) was used extensively to quantify the porosity. XCT allows volumes of material to be represented by three dimensional digital images providing accurate information as to how the X-ray attenuation varies within the material. XCT is not yet a commonly utilised technique, therefore a concise review of the principles of its operation is provided alongside descriptions of the experimental settings used. The finite element (FE) methods employed to predict the stress around both idealised spheroid voids and real pore geometries, taken from XCT data, are described in the final section. Also described here is the procedure used to model the stress and strain around fatigue cracks. 117

118 CHAPTER 3. EXPERIMENTAL METHODOLOGY 3.1 Arcam Operating Procedure All samples analysed in this thesis were manufactured using an Arcam S12 SEBM machine at the University of Sheffield with standard Arcam operation methodology, as recommended at one of their in-house training sessions. The training also covered the algorithms used to control the electron beam during melting (subsection 3.1.3). During melting, the electron beam is controlled by the proprietary Arcam EBM Control software; this software is also used to specify the models and settings for each build. For all samples, the feedstock consisted of Ti-6Al-4V pre-alloyed powder with a stated particle diameter range between 45 µm and 100 µm in diameter. As would be the case in a commercially viable process, all the powder used in this study had been recycled following the procedure described in Figure S12 Components An external view of the Arcam S12 SEBM equipment used to produce the samples in this work is given in Figure 3.1a. The left module of the S12 houses the control and power units. The EBM control software is interfaced with via the touch screen. The right module contains the vacuum chamber which is accessed through the door and can be viewed, under vacuum, through the circular leaded glass window. An internal view of the vacuum chamber when the door is open is given in Figure 3.1b, whilst Figure 3.1c shows a schematic diagram of the internal components. The electron gun column (A) is located centrally above the vacuum chamber (B) which is evacuated, then taken to a controlled vacuum (< mbar) by backfilling with helium, before the build begins. Within the vacuum chamber, two hoppers (C) store the powder prior to it being spread with the rake (D). The components being manufactured (E), are built onto a stainless steel baseplate (F), within the build tank (G). Beneath, and in contact with, the baseplate, is a thermocouple to record the temperature during the build process. Following the beam processing each layer, the baseplate is incremented downwards 70 µm by actuators (H) to achieve the correct layer height. The heat shield both protects components in the vacuum chamber and assists maintaining the elevated temperature in the powder bed. For clarity the heat shield is not shown in Figure 3.1c, but can be seen in the centre of the photograph in 3.1b above the build tank. In addition to the S12, secondary hardware, also provided by Arcam, is required to process the powder feedstock. The major components are: two explosion proof vacuum cleaners; a powder recovery system (PRS); and a vibrating sieve, powered by compressed air. One vacuum cleaner was employed to remove clean loose powder (i.e. still within the vacuum chamber), whilst the other was used to remove dirty powder from other 118

119 3.1. ARCAM OPERATING PROCEDURE (a) (b) (c) Figure 3.1: The Arcam S12 SEBM system: (a) photograph of the external cabinet; (b) photograph of the vacuum chamber; and (c) schematic diagram of the internal components (reprinted from Al-Bermani [27]). The coordinate system used throughout this thesis is also given in (c). 119

120 CHAPTER 3. EXPERIMENTAL METHODOLOGY (a) (b) (ii) (ii) (i) Figure 3.2: Photographs of the powder recovery system: (a) external view; and (b) internal view. surfaces contaminated with powder during normal operation of the equipment, e.g. the floor. Fine loose titanium powder poses a number of risks to health and safety such as: explosion or ignition; build up within the lungs if inhaled; and highly reduced friction when on a hard surface. For these reasons any spilt powder was immediately removed using the designated vacuum cleaner and safely disposed of. The clean powder still within the vacuum chamber at the end of a build could be recycled. Powder was first removed with the vacuum cleaner, then deposited into the PRS system shown in Figure 3.2. In addition to recycling the powder, the PRS is used to remove the sintered powder that encases manufactured parts. Sintered powder is removed by means of compressed air mixed with powder particles expelled at high velocity through the gun (i) shown in Figure 3.2b. During the removal of sintered powder the titanium manufactured part is held on the titanium plate (ii). Loose powder then falls through the steel grid (iii), before a sieve is employed to remove any powder particles with a diameter less than 45 µm, in an attempt to mitigate the risk of powder ignition. Powder is then gravity fed through a magnetic grid in an attempt to remove steel contaminate particles before the remainder are collected in a barrel. This completes the PRS step of recycling the powder. Following processing with the PRS, the powder was fed through the air powered vibrating sieve to filter any particles greater than 105 µm in diameter. The sieved powder was then reused in the next build after thorough mixing with any other powder required Build Preparation A standard methodology to prepare the Arcam hardware and software for a build was provided by Arcam at training courses. The steps involved, assuming the previous 120

121 3.1. ARCAM OPERATING PROCEDURE build was a success, are summarised here. Full details are available in the Arcam S12/A2 manual [28]. Prior to running the Arcam SEBM machine, it is necessary to create an Arcam build file, which contains the 2D melt geometries for each model within each slice. The first stage in generating the build file is design of the geometries desired within a CAD program. Samples were designed in either netfabb 5.2 or Autodesk Inventor 2012 and exported in the STL file format. Any STL repairs required (see subsection for possible errors) were carried out using MAGICS software. A virtual build chamber was used to position and orient the CAD models as required, before the proprietary Arcam Build Assembler software was used to slice the STLs into 2D layers of 70 µm thickness, the intended layer thickness during the build. The data concerning melt geometries for each slice was saved as a *.abf file, then imported into the Arcam SEBM machine. Upon completion of any previous build, the vacuum chamber is kept sealed until the thermocouple below the baseplate records a temperature below 100 C. Helium can be let into the vacuum chamber to raise the pressure to 400 mbar so that heat transfer, via convection through the helium, allows a significantly shorter cooling time, approximately 3 to 6 hours, in comparison to if the vacuum chamber pressure was held at < mbar where cooling can take several days. Once the temperature falls to below 100 C, air can be vented into the vacuum chamber and the door unlocked and opened. The heat shield is removed before the build table is moved up to release the previous solidified component from the build tank. The built components and sintered powder that encase them are transferred to the PRS system ready for sintered powder removal and powder recycling. The gun column is opened, examined for contamination and, if necessary, cleaned with high purity alcohol that evaporates without leaving residue. The filament operating history should be checked; if the current use time plus the upcoming build time is greater than the predicted life of the filaments (60 hours), the filament should be replaced. The anode and all sections of the gun column above the vacuum chamber are then checked and cleaned, again with high purity alcohol. The powder within the hoppers is checked to ensure there is sufficient to complete the next build. If additional powder is required, the hoppers must be removed and refilled with either virgin powder or reclaimed powder, before the hopper is returned to the build chamber. Excess loose powder is removed from the build chamber using the designated vacuum cleaner. All surfaces, including the heat shield, must be cleaned to remove metallic residue (evaporated titanium and aluminium) that builds up during use of the system. The build table is lowered such that when the base plate is inserted there is at least 40 mm of powder between it and the table. In addition, the thermocouple must be located so that it is touching the underside of the baseplate. The rake positions 121

122 CHAPTER 3. EXPERIMENTAL METHODOLOGY are calibrated to ensure that they are collecting the correct volume of powder from each of the hoppers during the powder collection and deposition stage. The rakes are also used to level the baseplate before it is cleaned of all powder and the build table is lowered to account for thermal expansion of the baseplate during the preheat stage. Once the heat shield is returned, the rubber seals around the vacuum chamber are cleaned, the door shut, and the vacuum pumps switched on. When the vacuum reaches the required level (< mbar in the chamber and < mbar in the gun), the high voltage power unit is ramped up to 60 kv. A low current ( 0.5 ma) is used to centre and focus the beam on the baseplate. Before starting the new build the EBM controller is used to load the previously created *.abf file and select beam settings for each model in the build file. The standard Arcam settings can be modified at this point and it is possible to select different settings for each model. When all the desired models are defined, the build process can begin. Builds end when the models are complete, although they can be aborted for a number of reasons, including mechanical and electrical faults Electron Beam Control With the SEBM machine, a constant voltage (60 kv) is maintained by the electron gun, but the current, deflection speed and focus is altered to control the energy input. Prior to depositing and melting powder on the baseplate, the baseplate is preheated to 730 C by rapidly scanning a defocused beam across it. The beam focus is adjusted by modification of a parameter termed the focus offset. The focus offset applies a modification current to the electromagnetic lenses and moves the focal point either above the build surface, when a positive focus offset is applied, or below, when a negative focus offset is applied. The first layer of powder is deposited directly onto the preheated baseplate. Prior to melting each layer, the powder is preheated and sintered by rapidly scanning the beam across the powder bed. This includes two consecutive steps with a defocused beam, where Preheat I scans the entire bed and Preheat 2 then pre-scans only the areas to be melted, expanded by 5 mm from their section edges. Figure 3.3a shows a photograph of the powder bed taken during the initial preheat (Preheat I) stage. The very high beam speed makes it appear to the eye as though there are numerous beams. The amount of energy input during the preheating stage is varied by the control software in an attempt to ensure a constant surface temperature prior to melting. This is achieved by modifying the beam speed, current and number of passes. Unfortunately, the exact algorithms used to calculate these beam settings are propriety Arcam IP and unavailable outside the company. However, typical values used during the Preheating stages are given on Table

123 3.1. ARCAM OPERATING PROCEDURE Figure 3.3: Photographs of the powder bed during the SEBM process: (a) preheating; (b) contouring; (c) hatching. Reprinted from Al-Bermani [14], provided by Arcam. Table 3.1: Default Arcam electron beam currents, speeds, focus offsets and line offsets employed during melting step. Strategy Current Speed Focus offset Line offset (ma) (mm s 1 ) (ma) (mm) Preheat I Preheat II Outer contour Inner contour Hatching In the standard build sequence, this is then followed by the melting stage, which uses a more concentrated beam and employs two separate beam rastering strategies (illustrated schematically in Figure 3.4). In the melting stage, firstly, three contour passes are used to melt the outline of each 2D section slice, starting at the section edge and moving inwards. The contour strategy (from here on referred to as contouring) uses a technology known as MultiBeam, which rapidly moves the beam so as to keep several separate melt pools active at one time. As a result of the MultiBeam settings, 50 melt pools are present during the outer contour pass and 10 during the inner contour passes. In the outer contour pass each melt pool is translated more slowly and with a lower power than for the inner two (see Table 3.1 for details). The centre of each section is then filled in by rastering the beam in a snaking melt strategy known as hatching (i.e. with a forwards and backwards beam motion with a continuous path). On completion, the stage is incremented downwards 70 µm to achieve the correct layer height, and the next layer of powder is dispensed from hoppers and spread with a rake. The hatching direction is rotated by 90 between each layer. Whilst the contour strategy uses constant beam settings regardless of model geometry, the beam current during hatching is not directly set by the operator, instead, it is calculated by the EBM control software and, in addition, varied linearly with the length of the hatch line, such that smaller melt lengths have a lower current. Unfortunately, once again, the propriety algorithms used to calculate the power are unknown outside 123

124 CHAPTER 3. EXPERIMENTAL METHODOLOGY Figure 3.4: Schematic of melt strategy employed by Arcam. The first initial contour passes melts the outline of the geometry, this is followed by hatching where a single melt pool snakes across the model. Also indicated is the line offset, which is the distance between the centre line of consecutive hatch lines. Arcam. However, Arcam describes the process by which the power is decided for each layer as follows: The surface temperature controls the amount of power that will be used for melting. All models loaded for the process step are included in the calculation. The goal of the calculation is to maintain a constant surface temperature during the build by adapting the power with regard to the geometries of the models. The hatching current is thus tuned in an attempt to maintain a constant temperature across the powder bed, regardless of the melt cross sections. The calculated hatch current is saved in an automatically generated log file. Following a build cycle it is possible to extract this log file to read the currents used. After calculation of the hatching current, a speed function is used to vary the speed of the beam to try and maintain a constant melt pool depth by maintaining an approximately constant line energy, proportional to I/v. The speed function will determine the beam speed by interpolation from a lookup table containing possible currents alongside a corresponding speed. Higher speed functions associate faster beam speeds with the same beam current. Figure 3.5 illustrates how the beam speed will vary with current for a range of speed functions. The default Arcam methodology when melting 45 µm to 100 µm diameter Ti-6Al-4V powder with a 70 µm layer thickness is to use speed function 36. The hatching beam velocity can be modified further by the application of a turning function and a thickness function. When the hatching reverses direction, the turning 124

125 3.1. ARCAM OPERATING PROCEDURE Speed (m s 1 ) Current (ma) Figure 3.5: Effect of beam current on electron beam speed for a range of speed functions. Table 3.2: Arcam recommended control constants used in Equations (3.1) and (3.2). The constants stated here are used in all models unless indicated otherwise. Function Constant Value Unit Turning Thickness A x 0.75 B x s mm 3 C x s 2 mm 2 A z 1.3 B z 1.4 mm 1 C z 0.25 mm function increases the speed to avoid overheating the already hot recently melted area. The new speed is calculated taking into account the initial speed (v int ) and the distance from the end of the preceding hatch pass (d x ) with Equation (3.1). v new = v int (1+ v) v=a x exp ( {v int [B x d x C x v int ]} 2) (3.1a) (3.1b) The constants: A x, B x and C x, are defined in Table 3.2 and are those recommended by Arcam. Figure 3.6 demonstrates the resulting beam speed on turning when standard settings are applied to a range of initial beam speeds. Further increases to the beam speed are applied when the hatch is melting a layer where the part geometry requires a lower melt depth. For example, for any overhanging sections or negatively angled walls. This function, referred to as the thickness function by Arcam, calculates an increased speed by multiplying the speed by the 125

126 CHAPTER 3. EXPERIMENTAL METHODOLOGY Speed (m s 1 ) m s m s m s m s Distance from end of previous hatch pass (mm) Figure 3.6: Increase in speed when starting a new hatch pass due to the turning points function for a range of initial speeds. 1.8 Speed increase Thickness (mm) Figure 3.7: Increase in speed due to the thickness function when melting part of an overhang of variable depth. speed increase given in Equation (3.2). Speed Increase = 1 + A z exp(b z (d z C z ))+1 (3.2) The speed increase is a function of the required melt depth to un-melted powder (d z ) and the constants are defined prior to the build (default values in Table 3.2). The thickness function is only utilised when the depth of the powder to be melted is less than a constant value (on Arcam recommendation set to 4 mm). Figure 3.7 shows the default beam speed increase for various depths. In this thesis, the standard settings defined above have been used to produce most samples. In cases where the control algorithm has been altered, the changes applied will be clearly stated. 126

127 15 mm 3.2. SAMPLES MANUFACTURED 3.2 Samples Manufactured The geometries and process settings used to produce all the samples examined in this thesis are recorded in this section. During the project, the powder feedstock manufacturing method was changed from gas atomisation (GA) to plasma atomisation (PA). In addition, there were updates to the software which altered the behaviour of the electron beam during the melting process. Therefore, each sample is noted as being GA or PA, and pre or post update, both in this section and when discussing the results. All specimens, except one clearly defined sample, were manufactured directly onto the stainless steel baseplate Geometry Work to characterise the effect of geometry on defect populations involved the manufacture of a number of specimens with GA powder and prior to the software update. Samples with dimensions of 15 mm in the y and z direction were produced while the thickness in the x direction was systematically varied, as shown in Figure 3.8. Further samples for the investigation of the effect of geometry on porosity were manufactured to the dimensions given in Figure 3.9. Three standard geometries were used, consisting of prisms with various base shapes. The bases used, square, circular and equilateral triangle, were chosen to represent possible geometry elements in engineering components. All the samples had the same maximum dimension (15 mm). Two to three build orientations were used for each of the geometries, giving a total of seven geometry samples, all manufactured without any supporting sacrificial structure. Two sets of seven samples were manufactured, and following non-destructive analysis by XCT, 10 mm 15 mm 8 mm 0.5 mm 1 mm 2 mm 4 mm z x y Figure 3.8: Dimensions of samples manufactured to examine effects of wall thickness on defect populations. The axis denotes the orientation of the two orthogonal hatching directions (x & y) and the build direction (z). 127

128 15 mm CHAPTER 3. EXPERIMENTAL METHODOLOGY Gs1 15 mm Gs2 Gc1 Gc2 15 mm 15 mm 15 mm 15 mm Gt1 Gt2 Gt3 z x y Figure 3.9: Dimensions of the prism samples manufactured to examine effects of geometry and build direction. Letters designate the prism base of the sample: Gs, square; Gc, circular; and Gt, triangular. Samples with the same letter designation, but a different number have the same geometry, but were orientated differently in the build chamber, as shown. The axis denotes the orientation of the two orthogonal hatching directions (x & y) and the build direction (z). one set was cut and polished to image pores with optical microscopy, while the other was sent for a HIP cycle (see subsection for details). A subscript H is used to differentiate the HIP samples from those used for optical microscopy. After the introduction of the new powder feedstock and the most recent software updates, further samples were manufactured with slightly varied geometry. These samples (Gp1, Gp2 and Gp3) were also built in specific locations within the build chamber to allow the effect of hatching scan lengths to be investigated. Figure 3.10 specifies the dimensions of the three geometries investigated in the x-y plane and their relative location in the build chamber. All the sample geometries had a height in the z direction of 25 mm and a radius in the x-y plane of 15 mm. Also indicated in Figure 3.10 is the relative orientation of the x and y hatching directions. In addition, one cuboid sample (geometry defined in Figure 3.10) was manufactured with a space of 10 mm between its underside and the baseplate with no supporting structure beneath it, designated sample Gr. 128

129 3.2. SAMPLES MANUFACTURED Gp2 30 mm 30 mm Gp3 15 mm mm mm Gp1 15 mm z x y 15 mm Figure 3.10: Dimensions of samples manufactured to investigate effect of geometry and sample location on defect populations. Samples were arranged in the build chamber as shown. The cuboid geometry, left, was also used to examine the influence of the different process themes. The axis denotes the orientation of the two orthogonal hatching directions (x & y) and the build direction (z) Melt Strategies To examine the effect of the melt strategies on defect populations, cuboid specimens with the geometry defined in Figure 3.10, were manufactured directly onto the baseplate. These samples were built using the latest Arcam themes and PA powder. The samples were orientated in the build chamber such that the orthogonal hatch directions (x & y) were aligned with their outer edges. During melting, the samples were arranged with 30 mm between each model to minimise thermal interaction (significantly further than used in other studies to thermally isolate each part [81, 128]). One build cycle was used to manufacture a set of specimens to examine the effect of varying the implementation of the individual contour and hatching strategies, which are detailed in Table 3.3. These samples were designated C0 C7. Further build cycles were then conducted to systematically examine the influence of the line offset between hatch passes (L0 L2), speed function (S0 S3) and focus offset (F0 F3) on the porosity generated during hatching, whilst all other parameters were kept constant (see Tables 3.2 and 3.1). Following each build, the Arcam software was used to extract the beam speeds and currents from the automatically generated log file. In addition, and with GA powder and prior to the updates to the process, the walls of various widths shown in Figure 3.8 were also manufactured with only the hatching strategy activated. 129

130 CHAPTER 3. EXPERIMENTAL METHODOLOGY Table 3.3: Sample designations for modifications to the standard Arcam settings used with the cuboid shaped samples. Samples numbered 0 are control samples, manufactured with standard Arcam recommended parameters. All settings except those mentioned were kept at Arcam recommended defaults (see Tables 3.1 and 3.2). Sample C0 C1 C2 C3 C4 C5 C6 C7 Modification None (control) Contouring only, hatching disabled Hatching only, contouring disabled Number of contours set to five Contour order changed to inner to outer Hatch first, contour second Turning function disabled Single direction hatching S0 Speed function set to 36 (control) S1 Speed function set to 28 S3 Speed function set to 20 S2 Speed function set to 12 L0 L1 L2 F0 F1 F2 F3 Line offset set to 0.20 mm (control) Line offset set to 0.15 mm Line offset set to 0.10 mm Focus offset set to 19 ma (control) Focus offset set to 12 ma Focus offset set to 6 ma Focus offset set to 0 ma 130

131 3.2. SAMPLES MANUFACTURED Hot Isostatic Pressing Hot isostatic pressing (HIPing) was carried out externally by Bodycote H.I.P. Ltd. (Chesterfield, United Kingdom). Standard HIP conditions for Ti-6Al-4V were used, i.e. a temperature of 920 C combined with a pressure of 100 MPa for 2 hours. Following the HIPing cycle, samples were cooled to room temperature at a rate of 6±2 C per minute Material for Mechanical Testing The samples used for mechanical testing were all manufactured using standard Arcam settings with GA Ti-6Al-4V powder and prior to the update to the control software. The samples were manufactured larger than the required size and machined to specification ahead of testing. In addition to ensuring a higher quality surface finish, machining meant that the material tested was from the bulk hatching region. Material for tensile testing was manufactured as cylinders with 20 mm diameter and 138 mm height vertically orientated in the build chamber. Fatigue samples orientated in the build (z) direction were machined from as-built cylinders of diameter 15 mm and height 60 mm. For testing in the x direction, fatigue specimens were machined from square crosssection samples of width and height 15 mm and length 60 mm aligned in the x direction in the powder bed Alternate Powder Feedstock The effect of powder feedstock on defect population was investigated using different powders, all of pre-alloyed Ti-6Al-4V, as the feedstock to produce solid samples. A total of six powders were tested, two gas atomised (GA), two hydride-dehydride (HDH), a plasma rotating electrode process (PREP) and plasma atomised (PA). Before changing the powder, the S12 machine was thoroughly cleaned to remove any residual powder in the vacuum chamber. The powders were consolidated using standard Arcam settings to produce cuboid samples 70 mm 15 mm 15 mm in the x, y and z directions respectively. Cylindrical samples with diameter 12.5 mm, and length 15 mm, were machined from the central hatching region of each build for XCT analysis. 131

132 10 mm CHAPTER 3. EXPERIMENTAL METHODOLOGY 3.3 Mechanical Testing Mechanical testing was carried out both within the university and by an external testing company. For all mechanical testing, machining was used to produce the samples from the as-built blanks. The material tested that was within the sample gauge length was therefore from the bulk hatching region of each build. Testing methodology and sample dimensions are described here Tensile Testing Following manufacture of the cylinders described in subsection 3.2.4, tensile samples were machined to the dimensions given in Figure Examples of the samples pre and post machining are provided in Figure Tensile testing was conducted using an Instron 5582 machine fitted with a 100 kn load cell. Prior to testing, the average diameter of each machined sample was measured ( 5 individual measurements) and the average cross-sectional area calculated. A 25 mm extensometer was calibrated and attached to the centre of the gauge length. The load and engineering strain (ε e ) were recorded with a frequency of 10 Hz whilst a constant extension rate (2 mm min 1 ) was applied. Engineering stress (σ e ) in the sample was calculated using the recorded load and maximum area calculated. True stress (σ t ) and true strain (ε t ) were calculated 30 mm 15 mm radius 60 mm M16 2 Figure 3.11: Dimensions of tensile samples. Figure 3.12: Example of an as-built cylinder (top) and a machined tensile sample. 132

133 10.6 mm mm 3.3. MECHANICAL TESTING using Equations (3.3) and (3.4) respectively. σ t = σ e (1+ε e ) (3.3) ε t = ln(1+ε e ) (3.4) Following fracture, the reduction in area was recorded using callipers to measure the minimum neck diameter. In addition to testing at room temperature, tests were carried out at elevated temperatures of 150 C and 300 C. Prior to testing, samples were held at temperature for approximately three hours to ensure a uniform temperature Fatigue Testing Fatigue testing was carried out externally by Exova (Lancaster, United Kingdom). Exova was also responsible for the machining and longitudinal polishing of the samples (subsection 3.2.4) to the specifications set out in Figure Axial load controlled testing was carried out in accordance with ASTM E (Standard Practice for Conducting Force Controlled Constant Amplitude Axial Fatigue Tests of Metallic Materials). An R-ratio of 0 and trapezoidal waveform loading with a frequency of 0.25 Hz was applied during testing. This produced a repeating waveform with a one second linearly increasing load, a one second hold at maximum load, a one second linear unload and then one second zero load. Maximum loads ranged between 500 MPa and 760 MPa. Samples were labelled according to their testing direction and maximum stress, e.g. FV575 for a sample tested in the vertical direction with a maximum stress of 575 MPa. Testing ceased after the sample fracture or 100,000 cycles were reached. The fatigue testing was carried out on behalf of an industrial partner, thus, of the samples tested, only some were available for analysis by the author of this thesis mm 0.5'' UNF - 2A 9 mm radius 12 mm Figure 3.13: Dimensions of the fatigue samples. Screw thread was a fine imperial thread. 133

134 CHAPTER 3. EXPERIMENTAL METHODOLOGY 3.4 Metallography Two dimensional imaging of the pores, microstructure and fracture surfaces of the samples was carried out using both optical and electron microscopy. The methodology used to prepare and image surfaces is provided here. In addition, the statistical methods to convert the 2D measurements of pore area to estimations of pore volume that were used are given here Optical Microscopy Samples for microstructure and porosity analysis were first sectioned in either the x-y, x-z or y-z plane with a 6 inch silicon carbide abrasive cutting wheel. A forward speed of 0.02 mm s 1 and blade rotation of 4000 rpm was utilised to cut the material. Following cutting, the samples were hot mounted in Bakelite then systematically ground using silicon carbide abrasive grinding papers. Paper grits were changed sequentially from P180 to P4000 before final polishing with an oxide polishing suspension (OPS). Optical microscopy was carried out using both a Zeiss Axiophot microscope, which was equipped with an Olympus camera, and a Keyence confocal microscope. Software provided by Keyence allowed the stitching together of 99 individual images to produce a large area for examination. All images used for quantification of porosity were collected at 10 magnification. Samples were examined without etching to provide the maximum contrast between pores and the polished surface. The optical micrographs were segmented and quantified in MATLAB. Segmentation was achieved via the Otsu method, which calculates a threshold based on the histogram of grey values (the brightness of each pixel) for each optical micrograph. The Otsu method calculates the intraclass variance for each possible threshold value and selects the value that resulted in the lowest intraclass variance [154]. Otsu showed that minimising the intra-class variance is identical to maximising the inter-class variance, which is far faster to compute and thus was used to select the thresholds. In other words, the Otsu method finds the threshold value where the sum of the spreads of the grey values of both classes is lowest, in this case pores and solid material. This threshold was used to create a binary image where all pixels with a greater grey value were labelled as solid and all pixels with a lesser grey value were labelled as pore. By using this method, the possibility of any operator bias in the choosing of threshold values to segment the pores was removed. In addition, this method should result in more repeatable results and correct for large scale changes of the brightness of images. The cross sectional surface area (A cs ) of each pore was obtained by multiplying the number of segmented pixels by the pixel area. To aid interpretation of the data, the area was converted to an equivalent circular diameter (D 2D ), the diameter of a circle of 134

135 3.4. METALLOGRAPHY equal area, using Equation (3.5). Acs D 2D = 2 π (3.5) The nature of optical microscopy means that all measurements were of a planar section of a volume containing 3D objects (pores) of varying sizes. If all objects are assumed to be spheres, not unreasonable considering the prevalence of gas pores reported in the literature, it is clear that, not only is it unlikely that the imaged plane will lie at the maximum dimension of any sphere, but also that larger spheres are more likely to lie on the imaged plane. The recorded size of each pore via optical microscopy will therefore lie somewhere between 0 and the actual pore size, leading to errors in the size distributions calculated. A statistical method for converting measured 2D circular diameters distributions per area to 3D diameter distributions per volume, is provided by the Schwartz-Saltykov analysis, also known as the Scheil analysis [155]. Schwartz and Saltykov developed a matrix of probability distributions for the number of circles, of size i, intersecting a random plane due to spherical particles, of size j. Inversion of this matrix provides coefficients (α( j,i)) to calculate the number of spheres of size j from the number of circles in size group i. These coefficients are used in Equation (3.6) alongside the experimentally obtained number of circles per area (N A ) in each size group i, where i=1 to k, is the group width and k is the maximum circle size. Thus, an estimate of the number of spheres per unit volume (N V ) in size group j is provided by Equation (3.6). N V ( j)= 1 {α( j, j) N A( j)+α( j, j+ 1) N A ( j+ 1)+ +α( j,k) N A (k)} (3.6) During the derivation of the coefficients (α( j,i)), spheres were assumed to be discrete sizes j, and thus the distribution provided by the Schwartz-Saltykov analysis is shifted below the correct level. Therefore, when plotting the results of Equation (3.6) they must be shifted by /2 to obtain the correct volume size distribution Surface Profile Measurement The Keyence confocal microscope was also used to measure, in high resolution, the variation in surface profiles of as-built surfaces. Lower resolution analysis of the surface was also conducted using a NanoFocus Uscan: Laser profilometer (MMSC), which measured parallel line profiles in the build direction with a step size of 5 µm and spacing of 1 mm. To quantify the surface roughness, the roughness average (R a ) 135

136 CHAPTER 3. EXPERIMENTAL METHODOLOGY was calculated using Equation (3.7). R a = 1 n n i=1 y i y (3.7) where: n is the number of measurements, y i is the elevation of the ith point, and y is the average elevation of all the measurements, after correction for tilt in the sample surface Electron Microscopy Scanning electron microscopy (SEM) was utilised to provide qualitative information regarding pore morphologies in polished cross sections. Additionally, SEM was used to examine the mechanical test samples fracture surfaces. Cross sections were prepared using the same method as described for optical microscopy. Fracture surfaces were prepared by first protecting them with masking tape before removal by a cut normal to the sample length. This allowed the fracture surface to be imaged in the SEM. Pores morphologies were analysed using both secondary and back-scattered electron images acquired with an FEI Magellan High Resolution Scanning Electron Microscope. Fracture surface analysis was undertaken using a CamScan MaXim SEM, FEI Sirion SEM and the depth focus mode with a Zeiss EVO 60 SEM. Typical accelerating voltages used ranged from 8 kv to 20 kv. Electron back scatter diffraction (EBSD) was undertaken to reveal the orientation of grains along fatigue crack growth profiles. Post-mortem analysis of fatigue samples was conducted by cutting a plane normal to the fracture surface and aligned with the crack growth direction. Silicon carbide abrasive papers were used to remove material until the plane was as close to the crack initiation point as practically possible, before final polishing with OPS. Crystal orientation data was acquired using a CamScan MaXim microscope with Aztec HKL software by Oxford instruments and a 0.5 µm step size. Due to the increasing prevalence of EBSD within materials science, and the relatively small role it occupies in this project, a review of the technique is not provided here, however, other sources are available [156] β Grain Reconstruction The initial solidification β grain structure was reconstructed from room temperature EBSD data with software developed at the University of Sheffield by Dr P.S. Davies and Dr B. Wynne [157]. Reconstruction is possible due to the Burgers orientation relationship (Figure 2.2.1) which dictates that for each α grain there are six possible high temperature parent β grains. Each α grain is checked against the neighbouring grains to identify those likely to have the same parentβgrain. By doing this for all the 136

137 3.4. METALLOGRAPHY α grains, the most likely β grain orientations can be obtained. Detailed descriptions of the algorithm, which is based on earlier work by Gey [158] and Humbert [159], can be found elsewhere [157, 10]. 137

138 CHAPTER 3. EXPERIMENTAL METHODOLOGY 3.5 X-ray Computed Tomography X-ray computed tomography (XCT) has been used extensively to characterise the size, geometry and position of porosity in the sample sets described in sections 3.2.1, and 3.2.5, with a range of systems and resolutions used. Furthermore, XCT was also used to characterise crack growth in fatigue test samples. In this work, X-ray absorption tomography was employed and therefore this section only discusses absorption tomography. However other techniques, such as phase or diffraction contrast tomography, can also provide useful data [160]. As XCT is not yet a standard technique, a concise review of the methodology and principals behind the technique is provided here. In addition, the 3D digital image processing techniques employed are described Principles A schematic diagram of the steps involved in collecting a XCT dataset is given in Figure With a laboratory X-ray imaging facility (i.e. not a synchrotron), X-rays are generated by the deceleration of electrons within a beam hitting a target, referred to as the source. The X-ray beam is directed towards a detector, and the sample being examined is placed between the source and detector. On passing through the sample, some of the photons within the X-ray beam are attenuated. The attenuated beam intensity is measured by a detector while the sample is rotated and radiographs/projections are recorded for each rotation step. Rotation steps commonly used range from around 0.1 to 1, and the process of collecting these radiographs while rotating the sample is commonly referred to as scanning. The series of projected images is then used to reconstruct a 3D matrix where each element, referred to as a voxel (3D analogue of a 2D pixel), corresponds to the X-ray attenuation at that point in the sample. This matrix can then be viewed as a series of slices as depicted in Figure Figure 3.14: Schematic diagram of XCT process. Reprinted from Landis and Keane [161] 138

139 3.5. X-RAY COMPUTED TOMOGRAPHY (a) (b) (c) (d) Figure 3.15: X-ray generation and energy spectra. (a) typical X-ray generation tube; (b & c) influence of source current and voltage (schematic); (d) effect of physical filters on energy spectra at 300 kv source voltage. Adapted from Kruth [15]. A typical X-ray tube consists of: an electron beam gun containing an electron emitting filament; an anode to accelerate the electrons; electrodes for control, focusing and deflection of the electron beam; and a target material [15]. The electron beam is directed onto the target material, as illustrated in Figure 3.15a. As described in the discussion of the SEBM system, upon hitting the target, most of the electron energy is converted to heat. However, a small amount (<1 %) is converted to X-rays. Most of the X-ray generation is due Bremsstrahlung (braking) radiation, which occurs when electrons hit the atomic nucleus. Alternatively, when the colliding electron excites an electron within an atomic shell, X-ray photons are produced when the excited electron releases its energy. This second source of X-ray photons is known as characteristic radiation, since it depends on the target material and is characterised by a line spectrum of photon energies and intensities. In contrast, Bremsstrahlung radiation results in a continuous (polychromatic) radiation spectrum with photon energies up to the maximum applied electron voltage. However, even the Bremsstrahlung radiation is influenced by the target material, with some materials producing a greater intensity of high energy photons than others. X-ray tubes can be defined as either reflection or transmission tubes. Within a reflection tube, as shown in Figure 3.15a, the useful X-ray radiation is emitted at approximately 60 to the electron beam direction. The large metal target allows integrated cooling and high beam energies. In contrast, in a transmission tube, a thin plate of target material acts as the vacuum chamber window and the X-ray beam is emitted in the same direction as the electron beam. The thin target allows a smaller electron interaction volume and, thus, smaller X-ray source spot size. The implications of spot 139

140 CHAPTER 3. EXPERIMENTAL METHODOLOGY size are discussed in the following section. To alter the number and energy of the X-ray photons, the electron beam current and voltage can be changed. Increasing the beam current, i.e. more electrons decelerating, results in a greater number of X-ray photons but with the same average and distribution of energy. The effect of electron beam current is illustrated schematically in Figure 3.15b. Conversely, increasing the voltage of the electron beam increases both the number and average energy of the X-ray photons, Figure 3.15c. Additionally, the average energy of the X-ray beam can be altered by the application of a filter between the X-ray source and the sample. The filter material will attenuate the lower energy X- ray photons, whilst allowing the higher energy photons to pass through. Figure 3.15d illustrates the influence of various filter thicknesses on energy spectra of X-rays produced with a 300 kv electron beam. The sharp peaks visible in Figure 3.15d are due to the characteristic radiation of the target. The X-ray beam energy should be chosen after consideration of the sample material, geometry and density, and is not a trivial task [15]. The transmission of X-ray photons through the sample should be high enough to get sufficient information about attenuation for each path. However, the intensity should be not be so high as to saturate the detector. This can be especially difficult to achieve when examining samples with a large aspect ratio, such as plate. In this case, when the beam path is parallel to the plate surface the attenuation can be very high, whereas in the perpendicular direction, the transmission can be very high. Possible solutions for this problem include scanning the sample at more than one energy; however, this is not yet a commonly used solution [15]. The reduction of X-ray intensity on passing through the sample is a function of the attenuation within the sample and the distance travelled through the sample. For a monochromatic and parallel beam, the new intensity is given by Equation (3.8). I = I 0 e µ x (3.8) where I is the intensity (number of photons) of a X-ray beam with initial intensity I 0 after passing a distance of x, through a material with attenuation coefficient µ. Equation (3.8) is known as the Beer-Lambert law, which was originally applied to the absorption of light [15, 162, 163]. Attenuation of an X-ray beam in the typical power range for laboratory XCT (20 kv to 450 kv) is due to both photoelectric absorption and Compton scattering [15]. Photoelectric absorption occurs when the entire energy of the photon is transferred to an inner electron, resulting in the ejection of the electron (Figure 3.16a). Alternately, if a photon interacts with an outer electron, the photon is deflected after transferring some its energy to the electron, known as Compton scattering (Figure 3.16b). The X-ray 140

141 3.5. X-RAY COMPUTED TOMOGRAPHY (a) (b) Figure 3.16: X-ray photon attenuation: (a) Photoelectric absorption; (b) Compton scattering. Reprinted from Kruth [15]. (a) (b) Figure 3.17: Reconstruction of volume from 2D projection data by: (a) simple backprojection; (b) filtered back-projection. Reprinted from Smith [164]. attenuation depends on the type and number of atoms in the beam path. However, in general, attenuation increases with an increased atomic number of the material, and is greater when the energy of the X-ray photons is lower [15, 162, 163]. Several methods to reconstruct a 3D volume from a series of 2D projections have been developed and include both direct and iterative solutions. The most common technique (used for all reconstruction in this thesis) is a direct method called filtered back-projection, which is a modification of the older back-projection technique [15, 162, 163]. Back projection is a relatively simple technique; radiographs are backprojected by setting all the voxels along the path leading to each pixel on the projection to the same recorded value of attenuation. In other words, each projection is smeared back along the direction it was acquired from [164]. The final back-projection image is then the summation of all the smears. Figure 3.17a illustrates the use of the back-projection technique to reconstruct a single slice of the volume. Clearly, the reconstructed image is very blurry, this due to each true point being reconstructed as a circular region that radially decreases in intensity moving from the centre of the true object. To reduce the blurring in the reconstructed volume, the recorded projections are 141

142 CHAPTER 3. EXPERIMENTAL METHODOLOGY Figure 3.18: Image magnification and blurring by moving the object towards a source having a finite X-ray spot. Reprinted from Kruth [15]. filtered prior to back-projecting them; hence this method is referred to as filtered backprojection. Each of the one dimensional horizontal lines of greyscales of the projections (each row of pixels) are convolved with a one dimensional filter kernel to filter the entire projection. The back-projection of the filtered radiographs results in a much sharper image, as shown in Figure 3.17b. Mathematically, assuming both an infinite number of projections and points on the detector, the filtered back-projection algorithm replicates exactly the correct image [164]. In Figure 3.17b two significant changes to the line profiles on the filtered views can be observed. Negative values have been introduced around the large peak in grey value, and the top of the peak has been flattened. The negative values counteract the blur and the flat top gives the circle a uniform intensity. The illustration given in Figure 3.17 is for the reconstruction of a single slice of the 3D volume from a single row of pixels on a radiograph collected with a parallel X-ray beam. However, the same principles can be applied when using a cone shaped X-ray beam, such as that generated by a laboratory X-ray source. The cone shape of the beam also allows magnification of objects by simple geometric effects. Figure 3.18 illustrates how the cone beam leads to the magnification of the projected image when objects are placed between the source and detector. The magnification of the projected object is simply the distance from the focused source to the detector (FDD) divided by the distance from the source to the object (FOD). It is clear how a single cone beam X-ray system can be used to give a range of magnifications by changing the distance from source to the object or detector. In addition to magnifying the object, the cone beam can worsen the effects of a finite, rather than a point, X-ray source when using a cone beam. The resultant blurring is one of a number of potential sources of error and 142

143 3.5. X-RAY COMPUTED TOMOGRAPHY uncertainly discussed in the following section Potential Sources of Error During the acquisition of radiographs and reconstruction of the 3D data, a number of artefacts can be introduced which can lead to incorrect values of X-ray attenuation being calculated at voxels. These errors can negatively effect the results if they are not properly taken account of. Some potential sources of error are listed below. Measures taken to try and minimise the errors during experiments are given in sections and 3.5.4, while the effects of unavoidable noise in the data, and how to account for this when analysing results, is discussed in subsection When reconstructing a 3D volume with projections acquired with a cone beam source, the size of X-ray source can lead to blurring. Reconstruction algorithms assume that the X-rays emanate from a point source. In reality, the source will be a finite sized spot where the electron beam hits the target material. As is clear in Figure 3.18, the blurring this generates is more significant when the magnification is higher. To try and mitigate this effect, the X-ray source size was kept as small as possible. In general, the X-ray source size increases with the beam current. Thus, as low a current as could produce reasonable results was used. As a polychromatic X-ray beam travels through an attenuating sample, the average energy of the photons will increase and the beam will become harder. Beam hardening [162, 15] is caused by the higher attenuation rate of lower energy photons. Beam hardening artefacts in reconstructed volumes result in the outer edges of samples having artificially high attenuation coefficients, due to the attenuation of many low energy photons, whilst the interior of the sample has an erroneously low attenuation coefficient, due to the relative lack of attenuation of high energy photons. A classic approach to the correction of beam hardening is the use of a thin sheet of material (typically copper or aluminium) to filter out the low energy photons before they reach the sample [15]. In addition to a physical filter, most reconstruction software packages contain algorithms for the numerical correction of beam hardening. Scatter of the X-ray beam can also introduce errors. The beam can be deflected within the sample, filter or detector and this can lead to a halo appearing around the sample. Unless proper threshold values are chosen this can result in inaccurate edge detection. The combination of the effects of scatter as well as beam hardening and any beam hardening corrections often leads to the rounding off of sharp edges. Ring artefacts can be caused by faulty pixels on the detector. Ring artefacts appear as rings around the centre of rotation of the sample with a false value of attenuation coefficient. 143

144 CHAPTER 3. EXPERIMENTAL METHODOLOGY The presence of material with the sample of a significantly higher attenuation coefficient can significantly effect the volume around it. Commonly observed around metal parts within material of a lower density, these effects are known as metal artefacts. Metal artefacts often appear as star like high density ghost images in the volume [15]. However, the samples analysed in this thesis were of near uniform density, excluding pores, thus this error source could be neglected. Errors can also be introduced during reconstruction of the 3D data from the collected radiographs. The axis about which the sample rotated must be located either by manual checking of the data or automatic calculation by the reconstruction software. If the axis is not found to a sufficient accuracy, it will lead to a poor quantity reconstructed volume with the subject edges artificially blurry Macro XCT Lower resolution (voxel size >7 µm) XCT was undertaken using the Nikon Metris custom bay with Nikon Metrology proprietary acquisition software (InspectX). An external view of the custom bay is given in Figure 3.19a while an internal view of the bay is available in Figure 3.19b. The system was equipped with a 225 kv static multimetal reflection anode source (Cu, Mo, Ag, and W) with a minimum focal spot size of 3 µm and a Perkin Elmer pixels 16-bit amorphous silicon flat panel detector. The high maximum power of this system (225 kw), combined with the large detector and source to detector distance, allowed the full scanning of macro scale, asbuilt samples. The rotating multi-metal X-ray source allowed the target material to be chosen to provide the most appropriate X-ray spectrum for each sample. As mentioned previously, different materials can result in different X-ray spectra, and for this system, Cu provided the beam with least energy, followed by Mo, Ag and W respectively. Figure 3.19: Nikon Metris custom bay XCT system: (a) external view; and (b) internal view of the custom bay. 144

145 3.5. X-RAY COMPUTED TOMOGRAPHY Table 3.4: Metris custom bay X-ray projection acquisition and 3D volume reconstruction settings settings. Step Settings A B C D Acquisition Reconstruction Source Ag Ag Mo Mo Voltage (kv) Current (ma) Exposure time (s) Cu filter thickness (mm) Source-sample distance (mm) Source-detector distance (mm) Voxel size (µm) Beam hardening Noise reduction Data remapping Table 3.4 lists the experimental settings used to scan and reconstruct all the samples within this thesis. Most samples were scanned using the settings is defined as A in Table 3.4. These settings were used to acquire and reconstruct most of the data within this thesis. The group of settings labelled B was used to compare the effect of different powder feedstock on the porosity found in solid material. XCT scanning of the fatigue samples utilised settings group C for imaging the entire sample and D for higher resolution imaging of the crack. For all samples, the beam settings and exposure time were chosen to provide approximately 10 % to 20 % transmission of the X-ray beam through the sample. In addition, the beam power, source material and filter thickness was chosen such that approximately 60,000 counts were recorded on the detector when the X-ray beam was unattenuated. The copper pre-filter on the beam removed low energy photons, which both avoided saturation of the detector and reduced the effect of beam hardening. The fatigue samples were contained within an aluminium tube during scanning (see subsection 3.5.6), where the tube acted in a similar fashion to a 3.2 mm aluminium filter and thus removed the requirement for the copper pre-filter. Prior to scanning samples, two sets of correction radiographs were acquired. One set of radiographs when the sample was outside the field of view and with the unattenuated X-ray beam, and another set when there was no beam energy. In both cases, 64 separate radiographs were recorded before being averaged in an attempt to avoid noise. These correction radiographs provide both the unattenuated intensity and background radiation intensity necessary for calculating the beam attenuation. In addition, the correction radiographs should correct for any faulty pixels on the detector and minimise ring artefacts in the reconstructed volume. For all samples the number of projection was set to Nikon s recommended optimum (3142), giving a rotation step size 145

146 CHAPTER 3. EXPERIMENTAL METHODOLOGY of rad ( 0.11 ). More projections mean a longer acquisition time and larger file size, but should result in less noise in the final reconstructed image. Reconstruction was performed using Nikon Metrology proprietary software CT- Pro. The voxel sizes given in Table 3.4 indicate the edge length of all the voxels that make up the 3D data. The voxel size can be calculated by dividing the 200 µm pixel pitch on the detector by the magnification. The centres of rotation of the samples were found automatically with the CT-pro software to the best possible standard. The beam hardening and noise reduction settings are numerical corrections for the artefacts discussed in subsection and were decided manually. Both settings had a range from 1 to 6, where 1 has no effect and 6 has the strongest influence on the data. For all samples, the pre-filter (or aluminium tube) removed most of the low energy photons and resulted in no beam hardening correction being required. In contrast, a noise reduction correction of 2 was found to give a slightly higher image quality than when using zero correction. Finally, the data remapping value indicates the range over which the 32 bit histogram was binned to create the 8 bit data High Resolution XCT High resolution XCT scanning was undertaken using two separate systems, the Zeiss Xradia Versa 500 and Zeiss Xradia microct system. The Versa was used to image both solid material produced by the Arcam and powder feedstock, whilst the microct was used only to image solid material. Both systems are equipped with tungsten transmission targets, which results in a smaller X-ray spot size but lower X-ray flux than available on the custom bay instrument. A photograph on the inside of the Versa system chamber is given in Figure 3.20a alongside typical samples imaged with this system shown in Figure 3.20b. Prior to all high resolution XCT scans, samples were secured to a nail with super-glue, as visible in Figure This allowed the three pin sample grips, supplied with both the Versa and microct systems, to easily and securely fix the samples during scanning. The reduced power and field of view of the higher resolution systems necessitated further sample preparation following manufacture in the Arcam AM system. For use with the microct system, small (approximately 1.1 mm diameter) cylinders were machined from the hatching region within the 10 mm wide wall. The Versa system was used with both cylindrical specimens with a diameter of 1.75 mm, machined from the edge and centre of sample Gs1, and cuboid specimens with approximately the same (1.75 mm) horizontal edge length, machined from sample C0. In addition, the six powder types from various production methods were analysed using the Xradia Versa system. Scanning loose powder was not practical, instead polyimide tubes sealed at one end were used to contain the powder. The tubes had an outer and inner diameter of 146

147 3.5. X-RAY COMPUTED TOMOGRAPHY Figure 3.20: Zeiss Xradia Versa 500 XCT system: (a) internal view of chamber; (b) typical samples, left to right, cylinder with 1.75 mm diameter, cuboid with 1.75 mm edge length and powder within polyimide tube. Also indicated in (b) is the approximate field of view ( 1.9 mm) and ( 1.8 mm) respectively. The square detector required that the vertical height of material being imaged was approximately the same as the diameter of the sample. Thus, only a 1.8 mm section height from each sample, shown in Figure 3.20b, was imaged. Both the Versa and microct systems include optical magnification lenses to increase the resolution beyond that which would achieved by the geometric cone beam effects alone. The optical magnification works in a similar fashion to an optical microscope focused on the scintillator in the detector. Unfortunately, the higher resolution this achieves can lead to greater noise in the recorded radiographs. Therefore, for all samples, the pixels in the detector were binned by two, averaging the signal from 4 (2 2 ) pixels. This binning resulted in half the resolution, but significantly less noise in the reconstructed volume. The acquisition and reconstruction settings used for all samples are given in Table 3.5. The source-sample and source-detector distances quoted in Table 3.5 are approximately correct, within 100 µm. Slight variations in the distances used resulted in minor changes to the voxel size in the reconstructed data. For all samples, the X- ray beam energies and exposure times were chosen to provide approximately 20 % to 30 % transmission through the sample as recommended by Zeiss. The filters used are specific to the Zeiss systems and, rather than specify a material and thickness, the associated number indicates filter attenuation, with higher numbers indicating more attenuation. Unlike the Nikon custom bay system, the centre of rotation was decided manually by reconstructing a single slice with multiple different values and choosing the 147

148 CHAPTER 3. EXPERIMENTAL METHODOLOGY Table 3.5: XCT data acquisition settings used for high resolution scans. Step Settings Solid Powder Solid Acquisition Reconstruction System Versa Versa microct Voltage (kv) Power (W) Exposure time (s) Filter LE4 LE3 LE4 Number of projections Binning Optical magnification Source-sample distance (mm) Source-detector distance (mm) Voxel size (µm) 2.05± ± ±0.001 Beam hardening Remapping 10k 50k 10k 45k 10k 50k value that gave the sharpest image. The beam hardening correction was also chosen manually. The manual choice of these values is subjective and does introduce some possibility of error, however, this was reduced by ensuring two operators were present and the values were agreed between them Data Down-sampling The 3D volume data was initially reconstructed as a 32 bit (Custom Bay system) or 16 bit (Versa and MicroCT systems) 3D volume. A full 32 bit volume requires bits = bits = 32 gigabytes of memory, and a full 16 bit volume requires 16 gigabytes. Therefore, to reduce the computational power required, the data was down-sampled to 8 bit using either VGStudio MAX or Aviso 8.0. This reduces the data size to 8 gigabytes for the full volume, but also reduces the possible grey values for each voxel to integers between 0 and 255. The down-sampling step could make small differences in X-ray absorption undetectable, but, if the data range remapping limits are chosen with care, the effect is known to be minimal [161]. The remapping values for all samples are given in Tables 3.4 and Interrupted Fatigue Testing The specimens for the interrupted fatigue testing were machined and polished by Exova to the dimensions described in subsection Testing was conducted in the horizontal direction and samples were labelled IFH1 4. An initial XCT scan was carried out to identify the pores in each test sample using settings group C in Table 3.4. Samples were then subjected to 10,000 fatigue cycles using the conditions described 148

149 3.5. X-RAY COMPUTED TOMOGRAPHY M18 x 1.5 Bolt Washer Upper ring Handle to prevent displacement Upper sample grip Tube Sample Lower sample grip Lower ring M8 bolts Load Cell Figure 3.21: Cross sectional diagram through the centre of the rig developed to hold samples under load during XCT scanning. in subsection at ambient temperature and a maximum stress of 600 MPa, before performing a second XCT scan. The XCT data from the second scan was manually checked for any detectable cracks. If no cracks were visible, the sample was subjected to a further 10,000 fatigue cycles at the same conditions, before performing another XCT scan. These steps were repeated until a crack was detectable in the XCT data. Once a crack was detected, the sample was rescanned at a higher resolution (settings D in Table 3.4) and then was scanned every 1,000 fatigue cycles. This was repeated until sample failure, in order to visualise the crack growth in 3D. To image the fatigue cracks, it was necessary to open them by applying a load. The capacity to do this at high resolution did not exist at the start of the project, so a custom sample holder was developed to hold the fatigue specimen at the maximum load experienced during a fatigue cycle. A scale diagram of this device is shown in Figure

150 CHAPTER 3. EXPERIMENTAL METHODOLOGY Figure 3.22: Steps taken to load fatigue samples prior to XCT scanning: (a) the lower ring and lower sample grip are secured to the load cell; (b) the sample is screwed into both sample grips; and (c) the tube and upper ring are placed around the sample and the bolt tightened to the correct load. When loading the samples to the correct stress (Figure 3.22), at all times care was taken to avoid touching the gauge length of the fatigue bar. A single aluminium alloy (6082-T6) tube was used in the rig to support the load during scanning of the fatigue samples. Turning the bolt at the top of the rig moved the sample grips apart and resulted in tensile stress in the fatigue sample. Rotational displacement of the sample was prevented by the square geometry of the upper sample holder. Following loading, the rig was moved to the custom bay, and both the rig and associated wiring secured to the rotation stage. This was required to avoid the wire from the load cell to the data recorder moving into the field of view. The load cell reading was then recorded at the start and end of each scan. Prior to unloading, the XCT data was checked to ensure that it reconstructed correctly Capability of Tensile Rig to Maintain Consistent Load Prior to using the tensile rig to enable the imaging of fatigue cracks, it was necessary to test its ability to maintain a consistent load on a sample for the length of time required for an XCT scan. The magnitude of the loading, MPa, was taken from the peak load experienced by a sample during a fatigue test with a maximum stress of 600 MPa. Figure 3.23 shows how the load varied with time when a dummy sample was loaded. A small degree of stress relaxation can be seen to have occurred, mostly in the initial 10 minutes following loading. With the XCT settings used to scan the fatigue samples (see subsection 3.5.3) each scan lasted approximately 54 minutes. The reduction of load over 54 minutes was always less than 2 MPa. During the actual experimental procedure, the stress relaxation was reduced further by loading the sample for approximately ten minutes before the start of each XCT scan. The load relaxation, after the first ten minutes, would only result in a <0.001 % reduction is elastic strain. Over the entire 12 mm gauge length this is equal to a displacement of 9.8 µm, slightly less than 150

151 3.5. X-RAY COMPUTED TOMOGRAPHY Load (kn) Time (minutes) Stress (MPa) Figure 3.23: Load and stress relaxation of the tensile rig developed for the project. The three lines indicate three separate tests. one voxel at the lower resolution (10.4 µm). This small displacement is unlikely to significantly effect the results. 151

152 CHAPTER 3. EXPERIMENTAL METHODOLOGY D Image Analysis Visualisation science group s (VSG) Avizo 8.0 was used to analyse the 3D XCT data. Due to noise and partially filled voxels, the average grey values of the pores within the material was much higher than the surrounding air. The nature of the XCT data, with the histogram of grey values dominated by the solid material and surrounding air, meant that standard algorithms for deciding threshold values, such as the Otsu method, would not identify the small amount of porosity in the samples. Most 3D data was therefore segmented using the Otsu method with the threshold constrained to a reduced portion of the grey scale histogram. This is the same method described in subsection 3.4.1, but only certain threshold values were tested for intraclass variance. Thresholds between 110 and 145 was used in the Otsu algorithm, since including more of the histogram resulted in a value that, while adequately separating the solid material and surrounding air, did not detect the pores within the material. Deciding the range over which to use the Otsu method required a user decision, but the same range was used for all samples. All voxels with a grey value lower than the threshold calculated by the Otsu method were labelled as pores and all voxels with a value higher than the threshold were labelled as solid to create a binary volume. In addition, to avoid labelling valleys in the rough, as-built surface as pores, the sample material was segmented with a high threshold of 150 then filled. Any voxels outside this artificially small sample were assumed not to be pores. If this final 150 threshold segmentation step was not applied, then a layer of voxels around the edge of the sample would invariably be labelled as pores by the automatic segmentation techniques used. Following the application of the threshold, a binary volume where each voxel is labelled as either solid or non-solid is created. Any voxels labelled as the same material as another voxel in the immediate neighbouring 26 places were grouped being a part of the same feature in the material. By such a method the volume was divided into solid material, pores and surrounding air. Pores were quantified with the label analysis tool included in the Aviso 8.0 software. However, not all the voxels labelled as non-solid were included in porosity quantification. In X-ray tomography the resolution limit is related to the voxel size and the ability to differentiate features from their absorption difference. The voxel size is controlled by the magnification (geometrical and optical) and is an absolute value determined by the equipment. For the sample sizes and instruments used in this study, the voxel dimensions are given in Tables 3.4 and 3.5. The resolution is more difficult to define, but is the smallest feature perceptible from the reconstructed 3D voxel data [15]. Resolution is affected by a number of factors, including: blurring from a finite rather than point X-ray source; scatter of X-ray photons within the sample; beam hardening; mechanical errors from stage movement; and incorrect determination of the centre of 152

153 3.5. X-RAY COMPUTED TOMOGRAPHY rotation during reconstruction (see subsection 3.5.2). In the measurements performed here, the X-ray attenuation of the void within a pore was much less than that of the solid titanium background. Whilst giving good contrast, this can still be a significant issue when attempting to detect small pores in thicker cross sections. The X-ray source size always remained below the voxel size, the centre of rotation was determined to the best standard possible, and the beam hardening was reduced with a pre-filter. However, irrespective of the quality of the system set up, a single voxel can still not be quantified as it is impossible to disregard the possibility that it is simply noise in the data. In addition, to determine accurate morphological parameters like pore aspect ratio, a greater resolution is required relative to the defect size. This is because when a defect becomes too small, their morphology cannot be accurately represented; i.e. ultimately a pore one voxel in size will appear as a cube. Therefore to assess pore size distributions and volume fractions, a lower pore size cutoff of (8) voxels was used, and for morphological analysis, only objects with a minimum volume of (125) voxels have been analysed, which is in line with resolution limits presented in the literature [33, 165, 166]. Individual pore volumes (V ) were calculated using the number of connected voxels in the pore multiplied by the voxel volume. However, volumetric size is not a convenient measure for interpretation or comparison with other techniques. Therefore, pore sizes in this thesis are quoted with the equivalent spherical diameter (D 3D ). The equivalent spherical diameter of a pore is the diameter of a sphere of equal volume, Equation (3.9). D 3D = 3 6 V π (3.9) In addition, the maximum (L max ) and minimum (L min ) pore length was calculated by checking the pore length in steps of 1 in the x, y & z directions around all possible directions. The aspect ratio was then defined simply as the maximum over the minimum length of each pore. As stated above, morphological parameters such as aspect ratio were only calculated for pores consisting of more than 5 3 = 125 voxels. The caveat being that only the larger pores orientation was quantified, which could influence the results. Pore orientations were quantified by calculating the principal rotational axes of the pores, defined by the eigenvectors of the moment of inertia tensor [167]. The directions of the three eigenvectors give the directions of the three principal rotational axes of the pore. Figure 3.24 illustrates the direction of the three eigenvectors for a spheroid geometry. The angle of the major axis, u 1 in Figure 3.24, was assumed to be the orientation of the pore and used in all quantifications. To better understand the distribution of pores within the samples, distance maps 153

154 CHAPTER 3. EXPERIMENTAL METHODOLOGY Figure 3.24: Principal axis of an ellipsoid. Reprinted from Galvez and Canton [167]. were generated giving the distance from each voxel within the sample to the sample surface. The location of the pores was then combined with the distance map to allow the distance of each pore from the surface to be recorded. In addition, the combined data was binned to record the volume of porosity at each voxel distance from the surface (i.e. the volume of pore voxels 1 voxel from the surface, 2 voxels from the surface, etc.). The volume of solid material at each distance was found in the same fashion. By dividing the volume of porosity by the material volume, the pore volume fraction and its variation with depth was thus calculated. Cracks were detected using the top hat tool within the Aviso label editor. This tool initially calculates the difference between each voxel and all voxels within a cube with edges 5 voxels long and centred on the voxel being tested. Cracks were defined as those voxels with a value greater than 16, i.e. voxels with an average difference of more than 16 in the greyscale between them and the surrounding material were designated as part of a crack. A distance map from the crack initiation location was used to quantify the crack depth. 154

155 Unconstrained plane Constrained plane Constrained plane Mirror plane d 3.6. FINITE ELEMENT MODELLING 3.6 Finite Element Modelling Finite element (FE) modelling was employed to predict the stresses and strains around pore geometries for different idealised conditions as well as real defect geometries measured by XCT. ScanIP was used to generate FE meshes from XCT data, and all other pre and post-processing of the idealised FE models was conducted using the commercial package Abaqus CAE. Abaqus/standard was used to solve all the models described in this thesis Stress Concentrations within Idealised Geometries All idealised geometries described in this section were created in the Abaqus CAE software. Only linear elastic material properties were used to investigate the stress around idealised pore and sample geometries. The gradient of the true stress-strain curves from the tensile testing (subsection 3.3.1) provided the Young s modulus (E = 119 GPa), whilst the Poisson s ratio (ν = 0.32) was taken from the technical guide for Titanium, by Donachie Jr. [20]. To investigate the effect of a free surface on the stress around pores, theoretical modelling of the stress around spherical and spheroid voids was carried out. The depth of the pore was made dimensionless by dividing the distance from the free surface to the maximum pore position by the pore diameter (d/d). Figure 3.25 illustrates the labelling system used. Symmetry of the loading conditions and geometry meant that only a quarter of the pore and surrounding material was required to model the entire stress field around the pore. The mirror planes indicated in Figure 3.25 were given boundary conditions such that there was no displacement either normal to the symmetry plane or rotationally about the about axis parallel to the symmetry plane. The result of these conditions is that the model behaves as if there was another identical geometry mirrored about this plane. For the other constrained planes, boundary conditions were specified such that there was no displacement normal to the plane. The intention was (a) (b) (c) Loaded plane Loaded plane d D Mirror plane Mirror plane D Figure 3.25: Example of the theoretical geometry used to examine the effect of a free surface on stress concentration around an oblate spheroid void: (a) x-z view; (b) y-z view; and (c) isotropic view. In this example d/d=

156 Constrained plane Constrained plane Constrained plane Mirror plane CHAPTER 3. EXPERIMENTAL METHODOLOGY (a) Loaded plane (b) (c) Loaded plane D1 s D2 D 1 s D 2 Mirror plane Mirror plane Figure 3.26: Example of the theoretical geometry used to model the effect of a two voids in close proximity on stress concentration: (a) x-z view; (b) y-z view; and (c) isotropic view. In this example s/d 1 = 0.5 and s/d 2 = 1. to simulate the effect of a relatively large amount of surrounding material, in comparison to the pore size; i.e. plane strain conditions away from the free surface. Perfectly spherical voids were initially modelled before oblate spheroid geometries with aspect ratios of both 2 and 0.5 were tested to examine the effect of the aspect ratio. Figure 3.25 shows an example of an oblate spheroid with aspect ratio 2. In this case, the length used to make the depth dimensionless was the diameter normal to the applied stress, as shown in Figure Spherical voids were also used to model the influence of the proximity of other pores on the local stress concentration. The distance between the two pores was made dimensionless by dividing the separation of the pores (s) by the one of the pore diameters (D), shown schematically in Figure When two voids of identical size were being modelled, due to symmetry, only one eighth needed to be modelled (exactly the same geometry as shown in Figure 3.25) but with additional boundary conditions such that the unconstrained plane became a mirror plane. The effect of different void sizes was modelled using a quarter of the geometry. Boundary conditions were applied as shown in Figure The 3D models mentioned above were meshed using solid hexahedral 3D stress elements and the standard meshing tools available within Abaqus CAE. Quadratic 20 node, fully integrated elements were used in order to improve resolution. An average size element size of 0.05 that of the diameter of the pore under investigation, with further refinement in the proximity of the pore using the curvature control tool, was used. This size of element was chosen to provide a balance between accuracy and time spent both meshing and solving the model. Finer elements could produce different stress concentrations, but the differences were significantly less than the trends being studied. Only the elastic case was considered for these idealised geometries, and since the stress concentration is independent of the global stress, the loading was set to 1 Pa to allow easy calculation of the stress concentration at all points in the model. The stress within a perfect fatigue sample, containing no defects, was modelled 156

157 3.6. FINITE ELEMENT MODELLING using axisymmetric elements. The symmetrical nature of the geometry and loading conditions require that only a quarter of the axisymmetric geometry be modelled to estimate the stress in the entire sample. Dimensions were taken from the engineering drawings used to machine the samples (Figure 3.13). The load on the sample was taken from the maximum load experienced during a fatigue cycle with maximum stress of 600 MPa or kn Modelling of Real Pore Geometries from XCT Data Meshes were created in ScanIP using the +FE Module (Simpleware Ltd, Exeter, UK) from the XCT data to allow direct modelling of the stress and strain around real pore geometries within fatigue samples. The tools provided within ScanIP for image segmentation, analysis and quantification, are less powerful than those available in Avizo. Therefore, the XCT data was segmented into solid and void, and the pores quantified, using Avizo. The binary data was exported as a stack of black and white tiffs for import into ScanIP. The mesh was created using the FE Grid algorithm in ScanIP+FE. A description of the methodology employed by the ScanIP software is given below, however, for more detailed information the reader is directed to the literature [168, 169]. The algorithm first meshes the entire volume, including the surrounding air, with hexahedral elements of a size defined by the sampling rate; each cubic voxel is converted to an element. Figure 3.27a shows how the segmented cubic voxel volume can be converted simply to a cubic hexahedral mesh, with an the example of a cylindrical object. Note that in Figure 3.27a only the voxels of interest (i.e. the solid material) are shown and the surrounding air would also be represented by cuboid voxels or elements. An isosurface denoting the material edges is then used to cut the volume into the 2 phases. The volume of interest is then re-meshed into tetrahedral elements. Figure 3.28a illustrates how an internal hexahedral element can be converted into a set of tetragonal elements, and Figure 3.28b how a voxel or element that intersects the isosurface can be converted. The result of the conversion of the solid volume to a tetragonal mesh, applied to the original cylindrical volume, is shown in Figure 3.27b and illustrates the smoother material edges created using both the isosurface and tetragonal elements. Internal tetragonal elements can then be converted back to hexahedral elements to reduce the total number of elements, as shown in Figure 3.27c. The number of elements can be further reduced by combining 8 (2 3 ) internal hexahedral elements into a single hexahedral element with transitional tetragonal elements to the original finer grid. Figure 3.27d illustrates the application of this decimation of elements. This final step can be repeated as many times as the internal volume without a feature edge can allow. 157

158 CHAPTER 3. EXPERIMENTAL METHODOLOGY (a) (b) (c) (d) Figure 3.27: Creation of finite element mesh from voxel data: (a) original voxel data; (b) tetrahedral mesh; (c) internal hexahedral elements replaced; and (d) mesh coarsening away from features. Reprinted from Young et al. [169]. (a) (b) Figure 3.28: Creation of tetragonal elements from orthogonal voxel data: (a) all vertices within a single material; (b) a single isosurface running through the volume. Reprinted from Young et al. [169]. 158

159 3.6. FINITE ELEMENT MODELLING It should be noted that the description of the meshing algorithm available in the literature [168, 169] and examination of the meshes generated in Figures 3.27 and 3.28 reveals that there is a possibility of distortion of the mesh, especially near feature edges. This could result in erroneous results when calculating the stress distribution around a pore. In addition, since the element size is based on the voxel size, which is constant, the larger pores will have a finer mesh relative to their size. This is likely to result in higher stress concentration around larger pores, even if the geometry is constant, which is not the case in reality. This is clearly an effect of the meshing, and thus, care must be taken when interpreting results. However, due to the difficultly in generating a FE mesh from XCT data, there are few other options and this was chosen as the best method available. An alternative method for generating FE meshes from XCT data was available in the Avizo software, the advancing front method. With this approach, first, a discrete triangulated surface of the feature being investigated must be generated. A set of nodes is then placed behind this surface, inside the feature, to construct an initial layer of tetragonal elements. Another set of nodes is placed further inside the feature to create a second layer of tetragonal element on top of the first layer. This is repeated until there is no more room for nodes to be placed. Clearly, there is an issue where two advancing fronts collide and the elements generated here can be of very poor quality (i.e. elongated with a very high aspect ratio). Furthermore, there is little scope for mesh refinement and the number of elements created is often considerably higher using this method compared to the method used by ScanIP, outlined previously. When the ScanIP software was employed to convert XCT data to FE meshes, default settings were used. This resulted in a mixed hexahedral and tetragonal mesh with most elements having the same length as the voxel size. The elements generated were 3D stress elements. As all the voxels are meshed, not just those labelled solid, at least three voxels of void were left when meshing an edge of the sample. The resulting mesh was exported as an *.inp file for import into Abaqus CAE where the material properties, boundary conditions and loading were defined. Both perfectly elastic (Youngs modulus and Poissons ratio of 119 GPa and 0.32 respectively) and idealised elastic-plastic (with a yield stress of 1 GPa followed by perfectly plastic deformation) material properties were used with these meshes. Any solid model edges representing an internal (x-z or y-z) plane, i.e. not a real edge of the sample but simply the edge of the modelled region, were constrained to zero displacement in the normal direction. Edges that represented the true edge of the model were left unconstrained to simulate the effect of the free surface on the stress and strain fields. A tensile pressure load in the z-direction was applied to the upper x-y plane edge of the model with the magnitude of the loading defined by location of the pore within the sample (see subsection 7.2.3). The lower x-y plane edge was constrained to zero displacement in the z-direction. 159

160 CHAPTER 3. EXPERIMENTAL METHODOLOGY Modelling of Fatigue Cracks Observed by XCT Fatigue cracks were also modelled using XCT data segmented in Avizo. Prior to importing the data into ScanIP the data was binned by to reduce the number of voxels to be meshed by a factor of 8. ScanIP was then used with the same procedure as described in the previous section to generate a mesh from the binary slice data exported by Avizo. The binning resulted in a larger element size than used to model the stress around pores. The elastic/perfectly-plastic material properties described above were assigned during pre-processing of the model in Abaqus CAE. Again, edges of the model representing internal planes (x-z and y-z planes) in the fatigue bar were constrained in the normal direction while the outer surfaces, representing the true edges of the sample, were left unconstrained. A tensile pressure load in the z-direction was applied to one of the x-y plane edges of the model while the opposite edge was constrained to zero displacement in the z-direction. 160

161 4 3D Characterisation of Defects in SEBM and their Origins This chapter reports on the extensive experimental work carried out to characterise the defects generated during SEBM-AM of Ti-6Al-4V using an Arcam powder bed system under standard conditions. The role of the different process parameters is not considered as this is dealt with in the following chapter. Low resolution X-ray computed tomography (XCT) has been used to quantify the size, morphology and frequency of large numbers of pores in whole volumes, alongside high resolution XCT, to image small sections and specific types of defect in more detail. The tomography data is also compared to measurements made by conventional optical microscopy and SEM from 2D sample sections. In addition, the surface roughness has been quantified with a confocal microscope. Firstly results are reported from two identical samples (C0 and F0), with a geometry defined in subsection 3.2.2, manufactured with current Arcam PA powder and the most recent software updates. Sample C0 is used as an example to demonstrate the type and distribution of defects typically observed in Ti-6Al-4V samples manufactured with SEBM and current best practice. Therefore, the defects identified within sample C0 have been scrutinised in detail. Following this analysis, these results are compared to those from sample F0, that was built separately but immediately proceeding C0, to examine the consistency of the defect populations. In addition, the accuracy of the XCT results and effects of XCT data processing on the quantification of porosity results is demonstrated here. The defects seen within sample Gs1, that was made with GA powder and prior to the system update are then characterised. Since much of the literature is relatively old, by examining the differences between these samples, it was possible to check that the previously suggested theories of pore formation were still applicable to the more modern Arcam systems. Finally, the porosity identified within the solid samples is compared to that measured with the two different powder types used by the Arcam machine. The results are then discussed collectively before the chapter concludes with an overall summary of the main findings. 161

162 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM 4.1 Defects within a Standard Sample Manufactured with Current Best Practice In this section, the defects encountered in the standard cuboid sample, C0 (see subsection for full sample description), are characterised with both 2D and 3D imaging techniques. Initial low resolution XCT scanning was performed on the whole volume of C0, photographed in Figure 4.1a. Following this, higher resolution XCT was conducted on the small sections photographed in Figure 4.1b, cut from both the edge (referred to as HR Edge) and centre (HR Centre) of the volume. The results from the XCT analysis are also compared to those produced by more conventional imaging techniques. (a) (b) 10 mm Figure 4.1: Photograph of standard sample C0 used to characterise defects: (a) as built; and (b) after cutting and mounting for high resolution XCT. Also shown in (b) is the powder sample analysed in subsection

163 4.1. DEFECTS WITHIN A STANDARD SAMPLE Overview of XCT Data In Figure 4.2 isometric projections are shown of two datasets from the standard sample C0. The segmented pore surfaces have been highlighted in red, that were obtained from lower resolution XCT scans performed over the whole volume (Figure 4.2a) and high resolution scans carried out on 1.6 mm machined sections from the sample edge and centre (Figure 4.2b). Statistical data from these scans is also summarised in Table 4.1. A range of pore sizes can be seen that at first sight appear randomly distributed. However, it should be noted that because such images are projections to the eye, they (a) (b) 1 mm 5 mm Figure 4.2: Examples of XCT data sets obtained from (a) the standard cuboidal sample with a voxel size of 9.9 µm and (b) from the edge and centre of the same sample with a voxel size of 2.1 µm. The approximate location of the two high resolution scans is shown on (a) by the blue boxes. Table 4.1: Summary of the average pore statistics obtained by XCT from full sample low resolution scans and from high resolution scans of different regions within the standard cube sample, C0. Also shown for comparison purposes is the data obtained by standard metallography and optical microscopy analysis of sample C0. XCT Sample location Whole specimen Voxel Volume/Area Volume Mean equiv. Max equiv. Number size analysed fraction diameter diameter identified (µm) (mm 3 /mm 2 ) (%) (µm) (µm) Hatch centre Edge Powder Optical x-y plane n/a microscopy x-z plane n/a

164 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM tend to over emphasise the volume fraction of pores. Without more careful statistical analysis such images can thus give a false impression of the density and tendency of pores to be spatially clustered. From the high resolution results in Table 4.1, it can be seen that overall the volume fraction of pores measured is low and in the range %. This figure can be assumed to be a lower bound because the micro XCT system could not detect very small pores (less than 5 µm in diameter) and this would be expected to generate only a slight underestimate, as such pores would contribute little to the overall volume Pore Size Distributions In Figure 4.3 the defect size distributions obtained from the 3D datasets shown in Figure 4.2, by XCT scans of the standard cubic sample (C0), are compared to conventional 2D metallographic optical microscopy measurements of the pore sizes. Measurements were performed on the two planes shown in Figure 4.4, which contain 99 individual optical microscope images after being stitched together. Despite the low resolution of Figure 4.4, which is much lower than the 99 individual images used during the quantification of the porosity, large circular pores are clearly visible. However, with this conventional optical approach, only 201 pores were analysed, despite imaging a total area of 195 mm 2. In comparison, with the high resolution XCT, when two scans were added together this resulted in approximately 10 mm 3 of material being analysed, and consequently a larger number of pores was detected (395). In addition, when the whole 1600 mm 3 sample was scanned at a lower resolution with the 225/320 kv Custom Bay machine, 2707 pores were identified. The 3D XCT data directly provides size distributions in terms of the frequency of the equivalent spherical diameter per unit volume, whereas the optical data was measured as the equivalent circular diameter per unit area. Therefore, to allow better comparison, the optical results have been converted into an equivalent volume distribution using the Schwartz-Saltykov (S-S) analysis [155]. This analysis is only applicable to spherical objects but, as by far the majority of the pores were spherical gas pores (see subsection 4.1.3), after correction the three techniques yielded good agreement in in terms of the frequency of pores found per mm 3 with sizes below 150 µm. For larger pore sizes, insufficient pores were detected by both higher resolution XCT and optical microscopy to allow reliable statistical measurement (only 4 pores were detected with an equivalent diameter greater than 120 µm by both high resolution XCT and optical analysis). In contrast, the data from the low resolution scans was statistically more reliable in larger size ranges (329 pores detected >120 µm). For pore sizes smaller than 50 µm, there was also a reduction in the frequency in the macro scans, compared to the optical and the high resolution XCT measurements, as macro 164

165 4.1. DEFECTS WITHIN A STANDARD SAMPLE Frequency µm voxel XCT (mm -3 ) 2.1 µm voxel XCT (mm -3 ) Light microscopy (mm -2 ) S-S analysis (mm -3 ) Equivalent diameter (µm) Figure 4.3: Pore size frequency distributions (11.5 µm bin size) derived from XCT datasets obtained using low and high resolution scans, plotted as an equivalent spherical diameter. The XCT data is also compared to 2D data from conventional optical microscopy, using the equivalent circular diameter, as well as after conversion to an equivalent spherical diameter by the Schwartz-Saltykov method (S-S analysis). (a) (b) 1 mm 1 mm Figure 4.4: Low resolution examples of stitched optical images used to quantify the porosity in sample C0, in the: (a) x-y plane; and (b) x-z plane (build direction vertically upwards in the plane of the page). 165

166 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM XCT could not identify any pores below 25 µm in diameter. Thus, below the resolution limit of the macro XCT there are a large number of small pores that would be missed by this technique. There was also a small disagreement in the frequency of small pores detected between the XCT and corrected optical data. This is likely to be due to the low sample volume, as the two techniques sampled different regions of the build and there was variation in the density of porosity depending on location (discussed further below). Thus, while the lower resolution XCT data cannot detect small pores, by scanning the whole sample it has identified the fewer largest pores that will be most important in terms of fatigue life, which were missed by the other two techniques owing to sampling issues. The limitations in resolution and sampling statistics between the optical microscopy and different resolution XCT techniques thus resulted in differences in measured average pore sizes as can be seen in Table 4.1. The corrected optical analysis and high resolution XCT recorded a mean size of 15 µm while the low resolution XCT gave a result of 82 µm. However, overall comparison of the three distributions shows that the majority of pores are below 100 µm in diameter and few ( 0.02 mm 3 ) exist above 150 µm in size Pore Morphologies Examples of typical pores types seen in SEBM samples when imaged by SEM are provided in Figure 4.5, to enable comparison to the different pore morphologies imaged by XCT shown in Figure 4.6. The pores aspect ratio frequency distribution and their size plotted against their aspect ratio are also provided in Figures 4.7 and 4.8, obtained from high resolution XCT scans of the centre and edge of the standard sample (C0) as well as the lower resolution scans of the whole sample. When observed using 2D polished sections, most pores were circular in cross section (Figure 4.5a) and ranged in size between 5 and 160 µm (Figure 4.3). Such pores are clearly equivalent to the very common spherical pores reconstructed from the XCT data that are designated type i and ii in Figure 4.6. From Figures 4.7 and 4.8 it can also be seen that all the porosity in the centre of the build (HR Centre) had a relatively low aspect ratio; the highest value recorded was 1.3. Visual inspection of the data confirms that virtually all the pores detected in the HR Centre scan were near spherical in morphology (see Figure 4.2b). In addition, when the low resolution data for the entire sample was analysed, it revealed that less than 3 % of the pores had an aspect ratio > 1.5. A rare example of a higher aspect pore, which appears to be two spherical pores joined together, is shown as type iii in Figure 4.6. An example of a small lack of fusion defect observed rarely by SEM is shown in Figure 4.5b and an irregular flaw that is equivalent to this defect is shown in 3D in 166

167 4.1. DEFECTS WITHIN A STANDARD SAMPLE (a) (b) 20 µm 20 µm Figure 4.5: Examples of typical pores seen in SEBM deposits imaged by SEM (backscatter mode) in the x-z plane: (a) two very small circular pores; and (b) a more irregular lack of fusion pore. The build direction is vertically upwards in the plane of the page. Figure 4.6: Examples of pore types observed in the standard sample, C0, at the same scale and translated to fit on a single figure: spherical pores, small (i) and large (ii) (blue); (iii) two near spherical pores joined together (turquoise); irregular pores, small (iv) and large (v) irregular (red). The build direction is z while x and y denote the hatching directions. An enlarged view of the small type (i) pore is provided in the inset. Note that the smaller/thinner pores (types i & iv) were detected using a high resolution scan from the centre of the sample and the coarse pores (ii, iii & v) were imaged using a lower resolution full sample scan. The very small spherical pore (i) has also been enlarged in the inset. 167

168 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Frequency (mm -3 ) Whole sample HR Centre HR Edge Aspect ratio Figure 4.7: Pore aspect ratio distributions obtained from the standard sample (C0) from high resolution data XCT taken at the sample edge and centre and lower resolution data from the whole sample. Aspect ratio Unreliable data (< 125 voxels) HR Centre HR Edge Equivalent diameter (µm) Figure 4.8: Pore aspect ratio against equivalent diameter distributions for all the pores detected by high resolution XCT, including those too small to allow reliable aspect ratio calculations (i.e. less than 125 voxels in size). Note the increased frequency of irregular pores in the sample taken from the cube edge. 168

169 4.1. DEFECTS WITHIN A STANDARD SAMPLE Figure 4.6; designated flaw type iv. Also shown in Figure 4.6 is a larger rare 190 µm irregular pore (type v), observed by the macro XCT full volume scan near the edge of the sample. No corresponding SEM or optical 2D representation was found for this flaw type, because of the low frequency of its occurrence. It can further be observed from the distributions plotted in Figure 4.7 and Figure 4.8 that there was a greater frequency of small high aspect ratio pores near the edge of the sample (HR Edge). In contrast, the sample machined from the centre of the hatched region (HR Centre) contained few high aspect ratio pores, but a number of larger near spherical pores. However, the majority of pores identified in both samples were spherical and relatively small (<75 µm) Pore Alignment In Figure 4.9 the dataset from the standard sample (C0) has been used to plot histograms depicting the orientation distributions of the major axis of elongated pores, relative to the build direction (z) and the beam raster directions (x & y). It can be seen from the graphs in Figure 4.9 that the pores were found to have their largest axis strongly orientated close to the x-y plane, whereas the rotation angle around the build direction showed no preferential alignment with the orthogonal scanning pattern. Hence, irregular pores were found to be elongated in the plane of the deposited layers, but were not strongly orientated relative to the beam raster directions. (a) 1.2 Frequency (mm 3 ) (b) Angle from build (z) direction ( ) Angle from hatching (x) direction ( ) Figure 4.9: Orientation distributions of irregular pores determined from (a) the angle of their major axis to the build direction and (b) the rotation of their major axis around the build direction in the x-y plane, relative to one of the hatching raster directions (lower resolution dataset from the standard cuboid sample C0). Both histograms to the same scale. 169

170 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Spatial Distribution of Pores in the x-y Plane The macro XCT 3D pore datasets for the standard sample have been projected in the build direction in Figure 4.10b, to allow visualisation of the spatial distribution with proximity to the sample edges of all the coarse (>25 µm) pores found. The pore volume fractions have also been averaged with distance from the sample surfaces to the centre of each section and these results are plotted in Figure 4.10a. Figure 4.10c & e show the relative volume fractions of regular (aspect ratio 1.5) and irregular porosity (aspect ratio>1.5). Since it was only possible to measure the aspect ratio of pores greater than 125 voxels in size, the volume fraction of these pores is also shown for comparison in Figure From Figure 4.10a, it is clear that pores >125 voxels make up the majority of the measured volume fraction so this is a reasonable representation. It is also apparent from Figure 4.10 that the standard specimen had a low volume fraction of pores near its edge, within a distance consistent with where the powder was melted by the contouring passes, whereas in the middle hatching region the average pore density was substantially higher. Moving inwards, two peaks in porosity are also observed in the data in Figure 4.10a. The first smaller peak (denoted I in Figure 4.10a) is at a distance of 0.8 mm in from the section surface, which coincides with a position close to the location of the last contour pass and the edge of the hatching region (shown by the line in Figure 4.10). However, there is a much larger second peak (II) in pore volume fraction between approximately 1.5 and 2 mm from the surface (in the x-y plane), near the edge of the hatching region. From Figure 4.10c it is clear that both peaks are almost entirely due to an increase in volume fraction of porosity with a spherical morphology. It is also notable that on moving in from the surface, in the initial 0.7 mm the porosity is almost entirely irregular (Figure 4.10e), whereas further from the surface, irregular porosity only makes up a small fraction of the total pore volume fraction. 170

171 4.1. DEFECTS WITHIN A STANDARD SAMPLE Pore volume fraction (%) Pore volume fraction (%) Pore volume fraction (%) (a) (c) (e) I Contour II (d) (f) > 125 voxels Irregular Hatching Total > 125 voxels > 125 voxels Regular Distance in x-y plane from surface (mm) (b) Lines indicating location of last contour and first hatch 5 mm Figure 4.10: Lower resolution analysis of the standard sample C0 showing the variation in porosity volume fraction in the x-y plane with distance from the sample surface: (a) the variation in the total pore volume fraction; in (b) all the pores detected are also projected in to the x-y plane, to allow visualisation of their spatial distribution; in (c) and (d) quantification and visualisation of only the regular pores is shown; and (e) and (f) provide data for the irregular pores. In all images the approximate width of the regions melted by contouring and hatching are denoted by the background colour. On all images the line at 0.8 mm from the surface indicates the location of the last contour pass. The volume fraction of pores large enough (> 125voxels) to allow accurate shape measurements is also indicated. 171

172 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Surface Roughness The high surface roughness of components manufactured by SEBM is the most obvious defect and can be observed in all the photographs of whole samples presented in this thesis. The cross sections of sample C0 observed by optical microscopy, shown in Figure 4.4, also clearly demonstrate the high roughness of the sample surfaces. When the surface is imaged by SEM, as shown in Figure 4.11, individual, partially melted powder particles are visible in addition to larger undulations. In addition to having a significant impact on the fatigue life of components, the roughness of the as-melted top surface could interfere with the correct spreading of the next powder layer. An image of the top surface of sample C0 is provided in Figure 4.12, which would be very similar to the surface that powder would be spread across during a build. From this low resolution image of the surface, it is clear there is some undulation in the last melted layer. To obtain more information about the surface profile, a colour plot of the relative elevation, measured by a confocal microscope, of the side and top surfaces of the cuboid sample C0, is shown in Figure 4.13a & b, respectively. Both surfaces show (a) (b) 200 µm 100 µm Figure 4.11: Typical component side surfaces as-built by SEBM imaged by SEM, showing partially melted powder particles. Build direction is vertical in the plane of the page. 5 mm Figure 4.12: View of the upper surface of standard sample showing ridges due to melt passes, imaged by a flat bed scanner. 172

173 4.1. DEFECTS WITHIN A STANDARD SAMPLE (a) 1.5 (b) z-direction (mm) y-direction (mm) x-direction (mm) x-direction (mm) Relative elevation (µm) Relative elevation (µm) Figure 4.13: Profile of surfaces of a standard cuboid: (a) vertical surface with the build direction vertically upwards in the plane of the page; and (b) top surface. The final hatching direction was left to right on this image and clear evidence of the melt passes is visible. 173

174 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM significant variation, with a higher peak to trough distance recorded on the side surface, despite imaging a smaller area. A similar trend is observed in the calculated average roughness (R a ) of 31.2 µm and 16.8 µm for the side and top surfaces respectively. The peak to trough distance of the side surface (>290 µm) is significantly bigger than the diameter of the powder (<100 µm) and some regions of high elevation are also larger than the powder size. The peak to trough distance in the top surface (>210 µm) is three times that of the 70 µm layer thickness used to manufacture this sample. Furthermore, the last layer of hatching has left clearly defined valleys and peaks in the surface profile aligned with the left to right hatching (x) direction. The spacing between neighbouring peaks in the y-direction is approximately the same as the 0.2 mm spacing between adjacent hatch melt passes. There is also significant variation in the x-direction, i.e. within a single hatch melt pass. 174

175 4.2. REPEATABILITY OF MEASUREMENTS 4.2 Repeatability of Measurements In this section, an analysis of the repeatability of the results is carried out. Due to the time and expense involved in both manufacturing and analysing samples, repeat tests were not always possible. However, error in XCT measurements can be hard to quantify [166, 170]. An examination of how the data processing techniques used influence the numerical measurements of the porosity is presented here alongside an analysis of the consistency in the measurements between repeated samples Influence of Threshold Value on XCT Image Analysis In 3D image analysis, similar to 2D image analysis, the quality of the numerical quantification depends on the quality of the segmentation of the image into discrete phases. Once down-sampled to 8 bit, the 3D XCT data consists of grey values between 0 and 255 defining the relative X-ray attenuation. The histogram of grey values from the scan of C0 in Figure 4.14 shows two peaks corresponding to the large volume of solid material, with high X-ray attenuation, and surrounding air, with low X-ray attenuation, but no obvious divide between the two phases. Simple segmentation of the data into discrete phases (pores and solid) can be achieved by choosing a threshold to split the voxels into two or more groups. As stated in the methodology chapter, threshold values used in this thesis were chosen by use of the Otsu method [154], which should remove operator bias and make measurements repeatable by minimising the intra-class variance of the two phases, i.e. minimising the spread of each phase. However, other methods are available, ranging from other automatic algorithms to operator decision, that could result in different threshold values [171]. Figure 4.15 shows how the calculated pore volume fraction varies if threshold values between 50 and 170 are used 10 8 Number of voxels Grey value Figure 4.14: Logarithmic plot showing the histogram of grey values (representing relative X-ray attenuation) within the XCT data collected from sample C0. 175

176 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Pore volume fraction (%) Threshold Figure 4.15: Variation in detected volume fraction porosity with threshold value for sample C0. Pore volume fraction (%) Distance in x-y plane from surface (mm) Figure 4.16: Variation in detected volume fraction porosity with depth for various threshold values. to segment the data. Unsurprisingly, a higher threshold results in a greater detected pore volume fraction with a near linear relationship for threshold values between 60 and 140. Greater than 150 and the volume fraction increases sharply. Clearly, the threshold value can strongly influence the final calculated volume fraction of porosity. The effect of four different threshold values on the variation in the volume fraction of porosity plotted against depth is shown in Figure Similar to in Figure 4.15, the threshold value can be seen to strongly influence the volume of porosity detected. However, the trends in the variation of detected volume fraction porosity with depth are very similar, with peaks at 0.8 mm and 1.75 mm. In addition, with high threshold values an initial peak can be seen in the volume fraction very close to the rough surface due to the valleys in the surface profile being mistakenly identified as internal pores. 176

177 4.2. REPEATABILITY OF MEASUREMENTS Comparison of XCT Scans of Identical Samples A summary of the quantification of the porosity detected in two samples with identical geometry and built in separate build cycles, but with all parameters kept identical such as the melt strategies and powder, is given in Table 4.2. In addition to using separate build cycles to manufacture the samples, the XCT scanning and analysis were conducted on two separate occasions. Thus, the results should give an indication of the variation of both the porosity generated during the build and random experimental error when measuring porosity by XCT. Clearly the differences between the two measurements are low, with variation in pore volume fraction detected less than %. The greatest difference is in the number of pores detected, but the statistical measures of individual pore sizes are similar. In addition, when histograms of the pore sizes are compared together in Figure 4.17, the size distributions can again be seen to be very similar. A slightly higher number of pores per unit volume with a equivalent diameter Table 4.2: Quantification of pores detected in two identical samples manufactured in two separate build cycles, C0 and F0, with the low resolution XCT (9.9 µm voxel size). Quantification Sample Difference C0 F0 (%) Volume fraction of detected porosity (%) Number of pores detected Mean pore equivalent diameter (µm) Maximum pore equivalent diameter (µm) Standard deviation pore equivalent diameter (µm) Frequency (mm -3 ) C0 F Equivalent diameter (µm) Figure 4.17: Lower resolution analysis of the two standard identical samples, C0 and F0, showing the histogram of pore sizes detected with the low resolution XCT with a 10 µm bin size. 177

178 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Pore volume fraction (%) C0 F Distance in x-y plane from surface (mm) Figure 4.18: Lower resolution analysis of the two standard identical samples, C0 and F0, showing good agreement in the variation in porosity volume fraction in the x-y plane, with distance from the sample surface. 80 µm to 100 µm can be observed in sample C0 than in sample F0, consistent with the higher number of pores detected overall in C0. However, other than this slight discrepancy, the two distributions are remarkably well aligned. In neither sample was the low resolution XCT able to detect pores below 25 µm, thus, the drop in the histogram at 20 µm to 30 µm is an artefact of the measurement method. When the volume fraction porosity was plotted against depth for sample C0 in Figure 4.10, a clear trend was observed with a non-random spatial distribution of porosity volume fraction. An identical analysis was conducted on sample F0 and a comparison of the two datasets is given in Figure Once again the two datasets are near identical, with both samples showing very similar variation in their pore volume fraction with depth, although there is some, seemingly random, scatter in the datasets at depths greater than 1.5 mm. The peaks in pore volume fraction identified in subsection (i & ii) are clearly also visible in sample F0. 178

179 4.3. SAMPLE MANUFACTURED WITH PREVIOUS CONTROL SOFTWARE 4.3 Sample Manufactured with Previous Control Software During the course of this research project both the software and powder type was updated. To examine whether this influenced the type and distribution of defects, a summary of the defects observed in sample Gs1, that had been manufactured using the older software and GA powder, was carried out. Sample Gs1 had an identical melt cross section to C0, but was only built to a height of 15 mm. In addition, sample Gs1 was manufactured alongside other components, which would result in changes to the electron beam current and speed used during hatching (discussed further in the following chapter). This sample was analysed using the same procedures of macro scale and high resolution XCT, at the same resolutions as samples C0 and F0, as well as by optical microscopy analysis Overview of XCT Data Figure 4.19a shows an isometric view of all the pores observed using macro scale XCT in sample Gs1, highlighted in red. Clearly there are large tunnel defects near one of the vertical surfaces; these defects are unlike any observed in samples C0 or F0. As a result of these tunnel defects the volume fraction of porosity in sample Gs1 is much higher than sample C0 at %. Table 4.3 summarises the pore statistics of all the pores identified with the two XCT systems and by optical microscopy. If defects greater than 400 µm in equivalent diameter, assumed to be tunnel defects, are excluded from the quantification of the porosity, the volume fraction drops to 0.02 %, significantly below that recorded for C0. Furthermore, these tunnel defects increased (a) (b) 1 mm 5 mm Figure 4.19: Example of pores found in sample Gs1 manufactured with GA powder and prior to the software update: (a) all the coarse (>25 µm) pores identified in sample Gs1; and (b) high resolution images of pores found in specimens machined from the edge (left) and centre (right) of sample Gs1. Note the presence of large tunnel defects at the sample edges in (a). 179

180 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Table 4.3: Summary of the average pore statistics obtained by XCT from full sample low resolution scans and from high resolution scans of different regions within the sample Gs1. Also shown for comparison purposes is the data obtained by standard metallography and optical microscopy analysis of sample Gs1. XCT Optical microscopy Sample location Whole specimen Voxel Volume/Area Volume Mean equiv. Max. equiv. Number size analysed fraction diameter diameter identified (µm) (mm 3 /mm 2 ) (%) (µm) (µm) Hatch centre Edge Powder x-y plane n/a the maximum detected pore size to significantly above that observed in either sample C0 or F0. The largest tunnel defect detected had an equivalent diameter of 1.1 mm and a length of 6.7 mm. In contrast, the mean pores sizes are smaller in sample Gs1 than sample C0. In addition, the high resolution XCT, which did not include any tunnel defects, recorded a lower volume fraction of porosity than the high resolution imaging of sample C Pore Size Distributions In Figure 4.20 the pores size distributions per unit volume detected with the macro scale and higher resolution XCT are shown along with optical microscopy measurements per unit area and after correction to volume measurements by S-S analysis. Note that the large tunnel defects have been excluded from the histograms. The three measurements are in agreement as to the distribution of pore sizes, with small pores less than 150 µm dominating the distribution. 180

181 4.3. SAMPLE MANUFACTURED WITH PREVIOUS CONTROL SOFTWARE Frequency µm voxel XCT (mm 3 ) 2.1 µm voxel XCT (mm 3 ) Light microscopy (mm -2 ) S-S analysis (mm -3 ) Equivalent diameter (µm) Figure 4.20: Pore size distributions in sample Gs1, measured by macro scale XCT, high resolution XCT, optical microscopy and after conversion by the Schwartz-Saltykov (S-S) analysis Pore Morphologies The aspect ratios of all the pores observed by high resolution XCT in small specimens machined from the edge and centre of sample Gs1 have been plotted against their equivalent diameter in Figure In common with sample C0, most pores are both small and reasonably spherical (low aspect ratio). Further similarity is observed when comparing the pores observed in the specimen from the edge and centre. The specimen from the centre contained mainly near spherical pores, whereas the specimen from the edge contained many irregular pores. Aspect ratio Unreliable data (<125 voxels) Centre Edge Equivalent diameter (µm) Figure 4.21: Pore aspect ratios against equivalent diameters measured by high resolution XCT of the edge and centre of sample Gs1. 181

182 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Spatial Distribution of Pores in the x-y Plane The variation in pore volume fraction with depth for sample Gs1 is also again plotted in Figure However, in this case the large tunnel defects have significantly altered the appearance of the trend in volume fraction of porosity with depth compared to that seen previously in samples C0 and F0. In these older builds, the distribution is dominated by the presence of the large tunnel defects with a peak in pore volume fraction at approximately 1.05 mm from the sample surface, which is at the edge of the region melted by the hatching. This peak is far greater than any observed in the samples manufactured with the modern software and powder. Pore volume fraction (%) Distance in x-y plane from surface (mm) Figure 4.22: Variation in pore volume fraction with depth in sample Gs1. The trend is dominated by the presence of large tunnel defects. 182

183 4.4. MEASUREMENTS FROM THE POWDER FEEDSTOCK 4.4 Measurements from the Powder Feedstock The results in the previous section confirmed that spherical pores were by far the most commonly encountered in standard samples produced by SEBM. In the literature these spherical pores are attributed to gas bubbles trapped within the powder feedstock. To corroborate this theory, the powder feedstock has been examined by XCT at the same resolution as the high resolution XCT scans of the solid AM builds to allow direct comparisons between porosity in the solidified metal and powder feedstock Current Plasma Atomised Powder In Figure 4.23 a 3D visualisation of a cropped sub-volume of the XCT data of the powder used to manufacture samples C0 and F0 is shown, where the pores detected within the particles have been highlighted with red surfaces. Those powder particles coloured dark grey had a pore detected within them, whereas the other light grey particles where not found to contain a pore. Statistical results from XCT scans of the powder feedstock are depicted in Figure 4.24a. The measured size distribution of the PA powder particles was close to the 45 µm to 100 µm range stated by the manufacturer, with only 8 % of the particles falling outside this range. From Figure 4.24a, it can be seen that a significant fraction of the larger powder particles contained pores. The pores present within the powder particles have also been compared to those in the consolidated material in Table 4.1 with respect to their number density, size and volume fraction. From Table 4.1 it can be noted that the volume fraction in the powder was lower by a factor of 1.5 2, relative to that in the centre of the standard AM 0.5 mm Figure 4.23: Visualisation of a subset of the 3D XCT data of the powder particles used to manufacture samples C0 and F0, showing pores within some powder particles. 183

184 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM (a) (b) Number of powder particles Number of powder particles Particles Porous particles Ratio porous Equivalent diameter (µm) Particles Porous particles Ratio porous Equivalent diameter (µm) Ratio porous Ratio porous Figure 4.24: Powder size distributions of all the powder particles and only the particles containing porosity. Also indicated by the line plot is the fraction of particles found to contain pores for each size bin. Data from the: (a) PA powder; and (b) GA powder; showing the increased prevalence of pores in the PA powder. 184

185 4.4. MEASUREMENTS FROM THE POWDER FEEDSTOCK sample (HR Centre), while the number density (pores per unit solid volume) was an order of magnitude higher within the powder feedstock. Furthermore, the maximum pore size was much greater in the solid material than in the powder, despite the large pores in the solid being spherical in morphology Gas Atomised Powder When the pore volume fraction in the GA powder was measured, it was found to be significantly lower (0.023 %) than in the PA powder (0.090 %). This surprising result suggests that in terms of internal porosity the newer PA powder is actually of a lower quality than the older GA powder. Although relatively few pores were detected in this older powder it still appears that, in general, larger powder particles are more likely to contain porosity. When the porosity in the GA powder is compared to that in the solidified material in Table 4.3, it is apparent that the total volume fraction is again lower in the powder. Furthermore, the number density of pores per unit volume is again higher in the powder than the solid material, and the spherical pores observed in the centre of sample Gs1 were also once more larger than any of the pores observed in the powder. When the histogram of powder sizes are compared for the two types of powder, Figure 4.24b shows that for the GA powder a larger proportion of the particles (23 %) fall outside the stated size range. In addition, the modal powder size is smaller, resulting in the higher number of particles analysed. 185

186 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM 4.5 Discussion In this chapter, 3D XCT datasets have been used to quantify the type, size and spatial distribution of the residual pores found in the SEBM-AM standard test samples in more detail than has previously been reported. XCT analysis was first performed at the scale of whole samples, with a bigger voxel size (9.9 µm), to determine the location of larger (>25 µm) defects. Subsequently, analysis was performed at a finer scale in specific regions of selected samples, with an order of magnitude smaller voxel size (2.1 µm), under conditions that were capable of detecting nearly all the pores present, down to a size limit of 5 µm. The results have revealed that the average volume fraction of porosity is quite low (<0.2 %), and lower than often reported values for SLM [133, 172, 80]. However, the pores present are not randomly distributed; rather there is strong evidence of a link between the pore distributions and the distance from the sample surface. This link is explored further in the following chapter where the effect of the beam conditions and Arcam control systems is investigated. The focus of the discussion in this chapter is on defect formation during the melting process and the accuracy of the measurements of the pore distributions Gas Pore Formation The results in this chapter have confirmed previous observations [63, 128, 127, 29] that small spherical pores less than 100 µm in diameter, such as those depicted in Figures 4.5a and 4.6 (types i and ii ), are by far the most common defect found in SEBM- AM components (97 % of the total). Their smooth spherical morphology confirms their origin as gas bubbles that were unable to escape during solidification [63]. Because the SEBM processing is carried out under vacuum, the main source of gas pores is thought to be argon from the powder feedstock that has been trapped in the powder particles during their manufacture by plasma or gas atomisation. Clear evidence of this source of gas contamination has been found by scanning the virgin powder, which has revealed a significant level of pores within larger powder particles (Figure 4.24). Previously, it has been shown that larger GA powder particles are more prone to gas entrapment than smaller particles [146], but here PA powders have also been shown to show the same trend. The volume fraction of pores measured in the PA powder was 0.09 % compared to 0.18 % in the consolidated PA material, but the average number density of pores was lower by 95 % in the solid samples (Table 4.1). With the GA powder, the pore volume fraction was also lower (0.02 %) in the powder compared to the solid (0.03 %) and had a 40 % lower number density of pores in the powder. The average size of the argon bubbles in both precursor powders was 12 µm. Although the maximum pore size in the powder must be less than the maximum size of the particles 186

187 4.5. DISCUSSION ( 100 µm), gas bubbles can potentially expand in the melt pool as the gravitational hydrostatic pressure from the small melt pool would be low, and there was a reduced pressure of mbar in the build chamber. Indeed, the largest spherical gas pore observed in the centre of the hatched region manufactured with PA powder was 140 µm in diameter, larger than could exist in a powder particle. Gas bubbles swept forward by the solidification front could also coalesce. An example of two gas pores frozen in the process of coalescing is shown in Figure 4.6 (type iii ). The lower volume fraction, but higher number density, of argon filled pores in the powder, can thus easily account for the observed level of gas porosity seen in the consolidated samples, and implies that a substantial proportion of this gas is actually lost from the melt. Furthermore, a greater number of individual gas bubbles are lost when using the PA powder and newer software settings than the older GA powder. Bubbles formed by soluble gases, such as hydrogen, oxygen and nitrogen, can also potentially precipitate at the solidification front due to the large difference in their solubility in the solid and liquid phases [173]. However, contamination in Ti-6Al-4V components has been measured to be very low [174]. Low levels of soluble gas in the melt make homogeneous nucleation of bubbles unlikely due to the high internal pressure required to form the new interface [137, 136]. Pre-existing bubbles or voids may grow due to the diffusion of soluble gases. In electron beam welding the most significant impurity for gas pore formation is considered to be hydrogen [137, 136]. A model of bubble growth has been proposed by Huang et al. [137] which links the growth of bubbles to both the initial bubble/void radius and the hydrogen content in the melt. Upon melting, bubbles of insoluble gasses that pre-exist in the powder, and thus melt pool, could provide volumes for the hydrogen to migrate to. For a bubble of 60 µm diameter to grow, the model proposed by Huang et al. [137] predicts that a hydrogen content of at least 200 ppm would be required. Measured gas concentrations in the powder used to manufacture the samples C0 and F0 were only H: 15 ppm, N: 150 ppm, O: 1210 ppm [174]. Thus, it seems unlikely that bubble growth in the melt pool can be attributed to the diffusion of hydrogen due to the low hydrogen content. Moreover, the high electron beam deflection speed, and corresponding short molten time, would allow little time for hydrogen diffusion to bubbles. Hence, it seems likely that bubbles observed in the Ti-6Al-4V consolidated by SEBM are a result of pre-existing bubbles of insoluble argon in the powder feedstock, rather than from hydrogen gas diffusion. Whether or not a gas bubble can escape before becoming trapped during solidification, by breaking the surface of a moving melt pool, depends on the pool shape as well as the convective and buoyancy forces that control their motion [175]. The high beam travel speed in AM processes typically results in a tear drop shape that is relatively shallow, but very elongated [27, 75]. This melt pool shape would make gas escape easier than, for example, in keyhole laser welding where the pool is typically much 187

188 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM deeper. However, the high speed of beam travel and related rapid solidification rate in AM is generally believed to increase the probability of gas entrapment [63]. However, the distribution of gas bubbles detected by XCT in the samples has been shown to be non-random (Figure 4.10), which is in contrast to the random distribution of gas pores assumed in the literature [29], and this is discussed further in the following chapter. In contrast, examination of the polished planes of the standard sample C0 did not measure the position of enough porosity to reveal locations of greater pore density, highlighting the advantage of XCT in quantifying thousands of pores. From the low resolution XCT results (Figure 4.10), it is clear that the melt strategy can influence the probability of gas bubbles escaping the melt pool prior to solidification. This point is considered further in the following chapter Lack of Fusion Pore Formation The presence of irregular shaped pores in AM builds is potentially more damaging owing to the higher stress concentrations they can generate. Such pores were found to form a minor fraction of the population in sample C0 with only 3 % of pores having an aspect ratio > 1.5 in macro scans of the standard sample. Higher resolution scans revealed no pores with aspect ratio>1.5 in the centre of the hatched region (HR Centre), but 4.2 % by number were irregular at the edge of the sample (HR Edge). These irregular defects had a wide size range, from a minimum of 18 µm up to 190 µm (Figure 4.8). The smaller irregular pores (type iv in Figure 4.6) had a morphology that suggests that they were formed from a lack of fusion and arise from small voids left between partially melted powder particles. It has been demonstrated that the irregular pores showed a strong tendency to be orientated with their major axis (Figure 4.9) lying in the plane of each layer, but that they were not found to have preferential alignment with the orthogonal beam raster directions. This in-plane alignment is to be expected if the irregular pores arise from gaps between partially melted powder particles, owing to the semi-circular transverse section of the melt pool and gravity driven compaction of the semi-solid region during processing. Irregular pores could be generated by random intermittent irregularities in the process, which reduces overlap of the melt tracks. For example, where there has been poor local powder settling during spreading, there was a particularly coarse particle near the edge of the melt pool, or as a result of local variation in coupling of the moving beam and the material in the powder bed. This would result in a lack of fusion between powder particles and a void left in the consolidated material. In Figure 4.13a it was shown that the last layer melted by the hatching had a peak to trough depth of over 210 µm, three times that of the intended layer thickness. Clearly the spread powder is unlikely to have a uniform depth corresponding to the expected 70 µm layer 188

189 4.5. DISCUSSION thickness. Moreover, in Figure 4.13a, it is clear that some of the lowest points on the melted surface lay adjacent to some of the highest points, which could further complicate the spreading of the powder. However, the effect this will have on powder spreading is currently unclear, indeed, to the best of the author s knowledge there is no literature on the spreading of powder in AM, regardless of the surface roughness. The lack of any studies into the effectiveness of powder spreading is surprising, especially when considering the number of AM system that utilise this method. If it assumed that the powder spreading results in a uniform height of powder 70 µm above the highest peak in the melt surface, it is clear that while in some places the beam will be melting only 70 µm powder thickness, in others the powder layer to be melted will be nearly 300 µm. Where deepest, the powder layer would have a depth similar to the melt pool depth predicted by Al-Bermani et al. [27] and greater than the melt pool depth predicted by Antonysamy et al. [75] for a solid Ti-6Al-4V plate. Since irregular pores are observed so rarely, this suggests that either the powder spreading cannot be so simply analysed or the melt pool depth is deeper than predicted by Al-Bermani et al. [27] and Antonysamy et al. [75]. Lattice Boltzmann modelling and experimental testing of the melt pool size when melting powder, rather than solid, was carried out by Körner et al. [73], who found the melt pool was strongly influenced by the packing density and random arrangement of particles. Thus, irregularities in the spread powder layer, possibly caused by the melted surface roughness, is likely to be an important factor in the generation of lack of fusion defects. Although irregular lack of fusion pores were found throughout the samples, in sample C0 manufactured to current best practice, they were dominant in the contour region where, entirely opposite to in the hatching region, they made up the majority of the detected defects. A possible reason for this is the Arcam MultiBeam setting, which keeps multiple melt pools active during contouring. This could lead to a non-steady state response with more chance of material intermittently not fully melting. The images and profile of the vertical surface (Figures 4.11 and 4.13a) also suggest that the melt pool was not uniform near the sample surface. Contouring also employs a more focused beam than the hatching, which Al-Bermani [14] showed resulted in a deeper but narrow melt pool, and this could potentially lead to insufficient overlap between passes. In the older sample Gs1, there was also a much greater density of irregular lack of fusion pores near the surface. However, those caused by the MultiBeam contouring may have been supplemented by the hatching failing to provide sufficient energy at the edge of its pass, discussed further below. 189

190 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM Tunnel Defect Formation Tunnel defects are likely to be even more detrimental to mechanical properties than smaller lack of fusion defects due to their greater size and aspect ratio. In the literature, tunnel defects are not reported as often as either gas pores or lack of fusion defects, but recent modelling work by Bauereiß et al. [76] demonstrated that they can be generated when the line energy of the electron beam is too low, see Figure They develop during a layer-wise deposition process when capillary and wetting effects overcome gravitational forces. This leads to the melted powder tracks separating by beading up, rather than filling in large voids present in the preceding layer (see subsection 2.5.1, page 93). The appearance of the tunnel defects at the edge of the hatching region (Figure 4.22), near the overlap between the hatching and the contouring, therefore suggests that with the old control software there was insufficient energy input in this region. Two possible reasons for this are most obvious: that there was insufficient overlap between the two themes; or that the turning function increased the beam speed by too great an extent when the beam turns around at the edge of a hatch melt track. Either reason could lead to there being insufficient energy input at the edge of the hatching and the formation of the tunnel defects by the process described by Bauereiß et al. [76]. It will be shown in the proceeding chapter that the turning function is likely to be the reason behind the reduced energy, rather than insufficient overlap between the strategies. It should be noted that the update to the software prior to the manufacture of samples C0 and F0 appeared to correct this problem and no tunnel defects were observed in the most recently produced samples Variation Between Modern and Older Arcam SEBM Methodologies It is immediately obvious from examination of the 3D visualisation of pores in Figures 4.2 and 4.19 that there are large differences between the samples manufactured with the most recent powder supply and machine control software compared to with the older procedures first used at the start of this project. In particular, the presence of large tunnel defects indicates that the older methodology was not supplying sufficient energy to samples at the edge of the edge of the hatching region where it overlaps with the contour strategy. However, the size distribution and morphology of the smaller gas pores and lack of fusion defects, were similar to that in sample C0. In contrast, when the tunnel defects were excluded from the pore quantification, the older sample actually contained a smaller volume fraction of porosity. A possible reason for the lower volume fraction of porosity in sample Gs1 (when the tunnel defects are excluded) is the change in the powder feedstock. When the porosity was measured in the older 190

191 4.5. DISCUSSION GA powder, it was found to contain only 26 % of the volume fraction and 30 % of the number density that was observed in the newer PA powder used to manufacture sample C0. It will be shown in chapter 6 that the volume fraction of pores in the powder is approximately linearly related to the volume fraction of pores in the solid material. In addition, sample Gs1 was manufactured with a number of other components where standard practise was to apply the hatching step to multiple models in the build chamber, i.e. the hatching will melt all aligned models in a single pass before turning back on itself. This did not occur in samples C0 and F0 as they were the only models being melted with standard settings in the build chamber. The influence of sample geometry and location in the build chamber, as well as electron beam current and speed is discussed further in the following chapter Accuracy and Consistency of XCT Results From the results presented in subsection 4.2.1, it is clear that the data processing techniques used to segment the data, and in particular the threshold value below which to count voxels as part of a pore, can significantly alter the final volume of porosity detected. For void-solid segmentation, such as that performed here, a global threshold is generally regarded as reasonably accurate [161, 166]. The unsurprising rise in detected volume fraction porosity with increasing threshold values (Figure 4.15), demonstrates the importance of using a consistent method to segment the data and has been reported previously by Maire and Withers [166]. If methods were varied between samples, serious systematic errors between measurements could be introduced. Therefore, all pores in this thesis were segmented using the Otsu technique, as outlined in the methodology section. It should be noted that the change in volume fraction is not directly proportional to the grey value histogram (Figure 4.14) since the positive identification of pores required a cluster of 8 voxels with a grey value less than the threshold value. However, given the wide, shallow trough in the histogram (Figure 4.14), it is unsurprising that small changes to the threshold when within this trough resulted in small near linear changes to the calculated volume fraction [161]. The pore volume fraction calculated optical microscopy was higher than that calculated by the low resolution XCT. Since the optical microscopy can detect small pores (<25 µm), that the low resolution XCT misses, it is unsurprising that a higher volume fraction should be derived from its measurements. The higher resolution XCT, could also identified smaller pores and reported a higher volume fraction. In addition to missing small pores, a too-high threshold value may underestimate the size of the larger pores when measuring porosity by low resolution XCT. Voxels overlapping the edge of pores will contain some percentage of solid material, therefore, the calculated X-ray attenuation will be somewhere between that of void and solid. Hence, a too-high 191

192 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM threshold value may detect the presence of a pore, but only include the voxels away from the pore surface, and thus calculate a too low pore size. However, from the comparison of pore sizes quantified with the two XCT resolutions and optical microscopy provided in Figure 4.3, it seems that pore size measurements are in good agreement for the larger pores. Ergo, the low resolution XCT provides reasonable accuracy in terms of the larger pores. The volume fraction porosity calculated by the low resolution XCT should thus perhaps be considered the volume fraction of large pores, rather than the absolute pore volume fraction. Moreover, the calculated volume fraction with depth into the sample (Figure 4.16) showed that the threshold value was not important in terms of defining the shape of the trend line, only the height of the peaks. Therefore, it can be safely assumed that this non-random distribution of pores is a true feature of the Ti-6Al-4V parts produced by SEBM. When the same Otsu segmentation technique was used on two separate samples, the variation in volume fraction porosity with depth (Figure 4.18) showed that both sample C0 and F0 had a near identical trend. This similarity was also reflected in the overall calculated pore volume fraction which showed less than 1 % difference. Not only this, but the number of pores detected per mm 3 were very closely aligned (Figure 4.17). This indicates that the SEBM machine produces a predictable total volume and density of porosity in Ti-6Al-4V components and some locations are more probable to contain porosity. Secondly, that any error in the measurement of porosity by XCT when using a standardised set up and data processing procedure is systematic rather than random. Of course, it is possible that random errors in the XCT measurement cancelled out random variation in the volume of porosity in the two samples. However, the chances of this seem remote and it appears far more likely that if there are errors in the XCT measurement they are systematic, e.g. consistently underestimating the volume fraction by a certain percent. Thus, overall XCT has been shown to be an appropriate and reliable tool to compare any variations between samples with confidence Surface Roughness The roughness of vertical surfaces has been measured to be high (R a >30 µm), similar to that found after poor quality metal sawing or sand casting. Components manufactured by SEBM are unlikely to be suitable for fatigue loading applications without further finishing operations, which negates the freedom of design that is often quoted as a major advantage of AM. In comparison to the results from other authors, the measured R a for vertical surfaces were very similar to those of Al-Bermani [14] (28.1 µm to 31.2 µm), but much lower than that of Chan et al. [100] (131.4 µm). However, as stated previously it seems likely that Chan et al. misreported the R a value. Al-Bermani 192

193 4.5. DISCUSSION does not record the size of the sample used to measure the surface roughness, but Chan et al. measured 60 mm 60 mm 4 mm walls, so it is also possible that the sample geometry had an influence on the surface roughness. Some of the pits and crevices visible in Figure 4.4 are overhung and would not be detectable by the confocal microscope. Thus, the true surface roughness is likely to be higher than that measured. The SEM images and height measurement of the vertical surface (Figure 4.11 and Figure 4.13a) show that the roughness is due to both partially melted powder particles attached to the surface and larger undulation in the surface profile. The longer wavelength undulations imply that the extent of the melt pool into the powder bed varies with height and position. In turn, this suggests a variable melt pool size or beam position when melting the edge of a sample. Both of these could be due to the Arcam MultiBeam setting of the contour strategy, which keeps 50 melt pools active when melting the very edge of a sample cross section. Poor overlap between these separate melt pools could lead to the longer wavelength surface undulation observed in the x and y directions. Further, since the melt pool has been shown to have an approximately semicircular cross section [27, 73], it is inevitable that a vertical wall will contain undulations with a wavelength corresponding to the layer thickness. Undulations were also observed on the upper surface of sample C0 (Figure 4.13b). Again, both shorter and longer wavelengths in the undulations were visible. It has been shown that a single pass of the electron beam results in convex weld bead upper surface [27]. Hence the linear undulations aligned with the hatching direction (x) observed in Figure 4.13b can be attributed to the electron beam passes. The reason for the larger scale undulation is less apparent. It is possible that the powder spreading by the rake failed to provide a uniform layer thickness, and thus some regions had more material added than others. Surface tension could also lead to thicker regions of powder forming convex melt pools rather than spreading out and assuming the shape that gravitational effects alone would suggest [76]. The high undulation of the top surface in comparison to the layer thickness also suggests why the default Arcam setting is to have a melt pool depth much greater than the layer thickness. 193

194 CHAPTER 4. DEFECTS AND THEIR ORIGINS IN SEBM 4.6 Summary Overall, the results from high resolution XCT scans gave good agreement with more conventional 2D measurements by optical microscopy, down to a size limit of 5 µm in diameter, giving confidence in the results. However, coarser scale scans with a resolution limit of 25 µm were found to be very useful for locating all the larger scale flaws within an entire build and the spatial distribution of defects. Analysis of the XCT data has shown that the average volume fraction of the pores was quite low (<0.2 %) and below that usually found in other AM processes like SLM [80, 133, 172]. Unfortunately, the measured surface roughness of the SEBM sample is high, and this must be considered before fatigue loading SEBM components. With the standard samples manufactured to current best practices analysed here, it has been found that the vast majority of voids were small spherical gas pores. These pores are thought to predominately originate from argon contamination in the powder feedstock, with the smaller gas bubbles trapped in the powder granules expanding and coalescing in the melt pool, owing to the reduced pressure in the build chamber. Rarer irregular shaped pores were found to be related to a lack of fusion between layers. Some of these flaws were quite large (up to 190 µm) and they were predominantly concentrated in the contour region. The older sample built with a previous generation of Arcam control software contained very large tunnel defects at the edge of the hatching region, but when these were excluded from the data set, the volume fraction of gas pores was lower and this was related to the lower pore volume fraction in the GA powder. The measurements of porosity volume fraction, size and spacial distributions made by XCT have been shown to be highly repeatable, with any errors in the quantification being systematic rather than random. The XCT technique has therefore been demonstrated to be a suitable method for quantifying the variation in porosity between samples. In addition, the trends in pore volume with depth have been shown to be repeatable when using the same powder feedstock. From these trends, it is clear that the melt strategies (contouring the outer edge followed by hatching the inner area) have a great effect on the porosity distribution and the effect of different melt strategies on defect population is explored further in the following chapter. 194

195 5 Influence of Melt Strategies and Sample Geometries on Defect Population The results presented in the previous chapter showed that most defects found within Ti-6Al-4V parts built by the Arcam AM machine are caused by gas bubbles unable to escape the melt pool. However, certain regions of each sample were found to be more prone to gas bubble entrapment. In this chapter, the melt strategies and electron beam settings have been altered to elucidate more information as to how such peaks in gas pore volume fraction are generated and effected by the Arcam machine s dynamic control system. In addition, the electron beam settings used for the hatching strategy have been changed to try and identify ways to avoid the appearance of porosity by giving the argon gas trapped in the powder feedstock more opportunity to escape the melt pool. The samples used in this study were all designed to have a constant melt cross section to give consistent average energy density with build height. The electron beam settings were read from the log file generated during the build to calculate the energy input provided by the Arcam machine s dynamic control software. The effect of sample geometry and location in the build chamber on defect populations has also been analysed by examining a number of different standard geometries. Unfortunately, when melting these samples, the electron beam settings could not be easily determined as the log file only records the speed of the very first hatch pass on each layer. Since the hatch melt length varies, the current and speed, which are dependent upon the hatch length used, also varied within each layer. However, knowledge of the algorithms used to control the electron beam allowed reasonable assumptions to be used to determine how the energy density would have varied in each build. As a result of this work it has been shown that some shapes and geometries are likely to contain a higher volume fraction of porosity. The large differences observed in the previous chapter between samples manufactured with the older software and GA powder and the modern software and PA powder, mean that it is necessary to differentiate between the two. Therefore, the software and powder type will be referenced in the text and figure captions will be labelled to indicated whether samples were manufactured with GA or PA powder. 195

196 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES 5.1 Modifications to Standard Melt Strategies The two peaks in pore volume fraction seen on moving in from the surface of the standard cuboid sample C0, shown previously in Figure 4.10, demonstrate clearly that the Arcam melt strategies can significantly influence the defect population detected by XCT imaging. In this study, selective activation, deactivation, or altering the behaviour of the individual melt strategies (hatching and contouring), has thus been used to gain insight into why these peaks form and whether they can be avoided. It should be noted that the samples examined here were manufactured with the most recent software updates and PA powder. All the samples used for this analysis had a standard cuboid geometry with 10.6 mm square cross sections, and a height of 25 mm (see subsection 3.2.2) Electron Beam Melting Settings Before discussing the defects detected with the samples manufactured using modifications made to the standard melt strategy, it is important to understand the effect of changing the Arcam machine process settings on the beam behaviour and energy density delivered to each sample. The sample codes and exact process modifications are listed in Table 3.3. Samples C0, S0, F0 and L0 were all melted with the standard melt strategy as described in subsection Figure 5.1 shows the melting pattern for all the samples analysed in this section. It should be noted that, while sample sets C0 C7 and F0 F3 were manufactured with the same powder batch analysed in subsection 4.4.1, separate powder batches were used for both sets of S0 S3 and L0 L2 samples. When using the Arcam system, it is not possible to directly alter the beam current and speed as this is calculated dynamically by the proprietary control software, based on the process settings. For all samples, the contouring beam current and speed is not adjusted by the control system and did not deviate from the standard values (see Table 5.1). In contrast, some of the process modifications used in samples C1 C7 resulted in changes to the energy density used for hatching. At the start of each hatch melt strategy in each layer, the control system first calculates the hatch begin speeds, and these are shown in Table 5.1 alongside the corresponding beam current for the speed function. The values in Table 5.1 were taken from the Arcam log file, which is automatically generated during the build, and records the calculated system beam parameters. It can be seen from Table 5.1 that there was a slight increase and decrease in both beam current and speed between the production of the standard sample, C0, and samples C2 and C3 respectively. This is caused by the change in hatch length that occurred due to the removal or addition of contour passes and occurs because the 196

197 5.1. MODIFICATIONS TO STANDARD MELT STRATEGIES (a) (b) (c) (d) (d) Figure 5.1: The effect of process changes (Table 3.3) on the melt pattern: (a) C0: Default settings, as described in subsection (b) C1: Contour only, the number of contours is increased and each subsequent contour moved inwards until the entire cross-section is melted. (c) C2: Hatch only, no contouring is used and the hatching area is increased to cover the whole cross-section. (d) C3: Number of contour passes set to 5, giving a smaller hatched region. (e) C7: Single direction hatching, all hatch melt lines are in the same direction and identical layers. Melt strategies for the other samples not mentioned were modified by changing the order in which the strategies were applied, the beam speed or focus or the line offset between hatch passes as described in Table

198 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES Table 5.1: The electron beam settings used by the Arcam machine for the different samples with the standard settings and with parameter modification (see Table 3.3). The corresponding applied energy density in each case E a has been calculated by Equation (5.1), based on the begin speed and the average speed. Process Sample Melt length v begin I h E a:begin E a:average numbers (mm) mm s 1 (ma) (mm) (J mm 3 ) (J mm 3 ) Outer contour Inner contour Hatch All All 9.5 & 9 (8.5 & 8 for C3 only) C0, C4, C5, F0 F3, S0 & L0 C C C6, C S S S L L control software assigns longer hatch passes higher currents, which, through the speed function, increases the beam speed to maintain an approximately constant ratio of I/v. More significant changes in speed and energy density were recorded when the speed function was deliberately decreased in samples S1 S3. The resultant applied energy density (E a ) for each melt condition was calculated using Equation (5.1) and is shown in Table 5.1. E a = P v h t (5.1) Where: P, v, h and t are the beam power (W), beam velocity (mm s 1 ), line offset (mm) (spacing between melt tracks) and layer thickness (mm), respectively. For all samples the layer thickness was 70 µm. This was calculated initially based on the begin speed which suggests that hatching for the standard sample (C0) had a higher energy density than for both contours. However, although the initial hatch pass of each sample started at this speed, subsequent hatch lines are affected by a turning function, which increases their speed without increasing the beam power and thus reduces the energy density. This function is designed to increase the beam speed at the start of each new reverse hatch track to avoid over heating the already hot area recently melted in the forward pass. On turning, the speed increase is controlled by an exponential function of the initial speed and the distance from the previously melted area, while 198

199 5.1. MODIFICATIONS TO STANDARD MELT STRATEGIES (a) Hatching beam velocity (mm s 1 ) C3 C0 C2 C2: Begin C6 = C3: Begin C0: Begin Distance from end of preceeding hatch line (mm) (b) Hatching energy density (J mm 3 ) C3 C3: Begin C2: Begin C0 C6 = C0: Begin C Distance from end of preceeding hatch line (mm) Figure 5.2: The effect of the turning function on: (a) the beam velocity; and (b) the resultant energy density for samples (C0 C7). Note; the lines for C0 and C6 also represent the speeds of the other samples indicated in Table 5.1. the beam power is kept constant as described in subsection The variation in beam speed and energy density with distance from the end of the previous hatch line is plotted in Figure 5.2, which shows the values for all the hatch lines. With standard settings, the turning function can impart a maximum increase of speed of 75 % at the start of a new hatch pass. Since energy density is proportional to 1 / v, this results in a 43 % reduction in energy density at the start of the hatch pass. The effect of this speed increase can be observed in Figure 5.2, by comparing samples C0 and C6, where the turning function was enabled and disabled, with all other settings kept constant. It can be seen that the turning function adjusts the speed over a considerable distance from the edge of the hatched area and in the small samples studied the speed is higher over the entire section width, never returning to its initial speed or energy density (equal to the C6 line). Thus, the actual average energy density of the hatched area is far lower than the values calculated in Table 5.1, based on the hatch 199

200 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES begin speed. The estimated actual average hatch energy has thus also been included in Table Effect of Process Modification on Defect Distribution To aid visualisation of the pore densities, in Figure 5.3 all the pores detected in the macro scale XCT scans of samples C0-C7 that were produced with different process modifications have been projected into the x-y plane. In Figure 5.4 the total pore volume fractions have again been plotted with distance from the sample surfaces, and the overall volume fractions of the porosity in all the samples are indicated in each legend. It is immediately obvious from comparing the images in Figure 5.3 and the graphs in Figure 5.4, that changing the machine settings can radically affect the distribution and density of pores. For example, build C1 (Figure 5.3b), produced by only using the contour settings across the entire section, had an average pore volume fraction much lower than that seen for the standard build. In addition, this build did not show any strong peaks in void density (Figure 5.4a). In contrast, sample C2 (Figure 5.3c), which was produced by hatching across the entire section, had a higher pore density than the original standard sample and a peak in pore volume fraction at a distance of mm from its edge, similar to the second peak (II) seen in the standard build (Figure 4.10 in the previous chapter). It is notable that this peak had moved outwards by the same distance (0.8 mm) as the extra hatching length required when contouring was turned off. This larger peak must therefore be a feature of only hatching and not due to the interaction between the hatching and contouring regions. Furthermore, when the number of contour passes was increased from 3 to 5 (sample C3), the surface layer with low porosity became correspondingly wider and the first smaller peak in pore density (I in the standard sample) moved inwards by a distance equivalent to that of the offset caused by the additional two contour passes ( mm) to mm (Figure 5.4b). In addition, when contouring was changed so that the first contour pass started at 0.8 mm in from the surface, and the subsequent passes moved out to the edge of the specimen (sample C4, Figure 5.3e), the variation in volume fraction with depth was very similar to that recorded with standard settings. However, when the contouring was performed after the hatching (sample C5), the first peak (I) was no longer observable, but the larger second peak (II) was again present in the same location as in the standard build (Figure Figure 5.4c). In contrast, when the snake function was turned off (sample C6) so that the hatching was performed by sweeping in a single direction (right to left in Figure Figure 5.3g), the first peak became far more pronounced, while the second peak disappeared. From Figure 5.3g (sample C6), it also appears that there is a tendency for pores to be left closer to the end (left) of 200

201 5.1. MODIFICATIONS TO STANDARD MELT STRATEGIES (a) C0: Control (b) C1: Contour only (c) C2: Hatch only (d) C3: Contour 5 (e) C4: Contour in-out (f) C5: Hatch first (g) C6: Snake function off (h) C7:Turning function off y 5 mm x Hatching directions Contour Hatching Build direction (z) Figure 5.3: Lower resolution 3D XCT full sample scans showing visualisation of all the pores (red) detected in each sample projected in to the x-y plane to allow visualisation of their spatial distribution, showing the effect of the process modifications defined in Table 3.3. In all images the approximate width of the regions melted by contouring and hatching are denoted by the background colour. (PA samples.) 201

202 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES a hatch track than the start (right). Therefore it can be concluded that peak I is probably generated by the end of the hatching lines. Single direction hatching would also negate the effect of the turning function as the beam does not turn back on itself and as a result there is no peak II in the hatching region. To confirm this relationship, in sample C7 (Figure 5.3h) normal hatching was resumed, but the turning function was disabled and the overall density of pores can again be seen to be much lower in the hatching region than in the standard sample. 202

203 5.1. MODIFICATIONS TO STANDARD MELT STRATEGIES (a) (b) (c) (d) Pore volume fraction (%) Pore volume fraction (%) Pore volume fraction (%) Pore volume fraction (%) mm 0.5 mm max % C0: % C1: % C2: % C0: % C3: % C4: % C0: % C5: % C6: % C0: % C7: % Distance in x-y plane from surface (mm) Figure 5.4: Effect of process modification, compared to the standard recommended settings (sample C0), on the pore volume fraction distribution with distance from the build edge, from lower resolution XCT scan data. The parameters changed are described in Table 3.3 for each sample type (C1- C7). The average volume fraction for the whole sample is given in the figure legend. (PA samples.) 203

204 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES Influence of Beam Speed, Offset and Focus on Pore Volume Fractions The effects of changes to the beam speed, line offset, and focus on the measured pore volume fractions obtained by the lower resolution XCT full sample scans is summarised in Figure 5.5. For these experiments, all the other beam parameters were kept constant (see Table 3.3). It is clear from Figure 5.5 that decreasing both the beam speed and line offset reduces the level of porosity (>25 µm diameter). However, the relationship between focus offset and pore volume fraction was found to be more complex. A reduction in focus offset, from its standard value (F0: Focus offset = 19 ma) first reduced the detected pore volume fraction by an order of magnitude, before increasing it slightly again when the beam was at its most focused (F4: Focus offset = 0 ma). Pore volume fraction (%) S0 S3: Speed function L0 L2: Line offset F0 F3: Focus offset Sample number Figure 5.5: Influence of the speed function, line offset and focus offset on detectable pore (>25 µm diameter) volume fraction. The sample codes are described in Table 3.3. (PA samples.) 204

205 5.2. EFFECT OF SAMPLE GEOMETRY AND LOCATION IN BUILD CHAMBER 5.2 Effect of Sample Geometry and Location in Build Chamber To analyse the effect of sample geometry on defect population, generic geometries were built to represent possible engineering components. Early work, utilising GA powder and prior to the Arcam system updates, was conducted to examine the influence of the sample geometry and build direction on pores in samples. Later work, with PA powder and after the Arcam system updates, has been included to show the effect of sample location in the build chamber. Unfortunately, unlike the samples analysed in the previous section, the non-constant cross section of these geometries made it much harder to determine the electron beam settings used to melt each sample as the settings are varied when the hatch length changes Influence of Wall Thickness All the data presented in this section is taken from samples produced using the older Arcam methodologies (i.e. GA powder and earlier software). A photograph of the samples of identical length and height, but various widths manufactured to examine the effect on pore volume fraction, is provided in Figure 5.6. The variation in pore volume fraction detected by macro scale XCT (10 µm voxel size) is shown in Figure 5.7. There is a clear increase in pore volume fraction as the wall width gets larger. In addition, there is a smaller peak in the volume fraction of porosity in the 2 mm wide wall. Visual inspection of the 3D XCT data revealed that the 10 mm wide wall contained some large tunnel defects. From the 3D visualisation of the pores detected in the 10 mm wall (see Figure 5.8), it is clear that the tunnel defects are concentrated near the edge of the samples and, in particular, near vertices. Similar to those observed in sample Gs1, the tunnel defects tended to be located at a distance of 1 mm from the sample edge, where the hatching strategy ends. Despite the presence of these tunnel defects, the total pore volume fraction of the 10 mm wall is of a similar magnitude to the samples manufactured with the modern methodology. Since the pores in the 10 mm Figure 5.6: Photograph of samples manufactured to examine effect of wall thickness. (GA samples.) 205

206 Pore volume fraction (%) CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES Wall width (mm) 8 10 Figure 5.7: Variation in pore volume fraction with sample width detected by macroscale (10 µm voxel size) XCT. (GA samples.) (a) (b) 5 mm Figure 5.8: Lower resolution 3D XCT full sample scans showing visualisation of all the pores (red) detected in the 10 mm width wall sample including large tunnel defects, in the: (a) x-z plane; and (b) y-z plane. (GA samples.) 206

207 5.2. EFFECT OF SAMPLE GEOMETRY AND LOCATION IN BUILD CHAMBER (a) (b) Relative height (mm) Pore volume fraction (%) Figure 5.9: The spatial distribution of pores with height in the hatch region obtained by micro XCT of specimen machined from 10 mm wide wall: (a) 3D visualisation in the x-z plane; and (b) the variation in volume fraction of porosity with build height. Build direction is vertically upwards in the plane of the page. Images (a) and (b) to same distance scale. (GA sample.) modern samples consisted primarily of gas pores, this indicates that the level of gas porosity is much lower in these older samples. A small section was machined from the 10 mm wide wall and subjected to high resolution XCT (1.2 µm voxel size) using the microct system. Inspection of the XCT data revealed that, all the pores were spherical gas pores and the pore volume fraction appeared to exhibit regular periodicity with build height, as shown in Figure 5.9. A Fourier transform was applied to the pore volume fraction with height to quantify the spacing between peaks. The dataset was first given a zero mean volume fraction, by subtracting the average value, and then multiplied by a Hann window to reduce the spectral leakage associated with finite observation intervals [176]. The fast Fourier transform MATLAB function was then applied and the maximum peak in the periodogram identified, as shown in Figure The first strong peak in Figure 5.10 is the true wavelength, whereas the secondary peaks in Figure 5.10 are harmonics of the first and can be ignored. Since with this older control setting the Arcam machine deposited layers 70 µm thick, this wavelength closely matches every fourth layer in the build cycle. The effect of wall thickness on surface roughness was also measured with multiple (>10) line measurements in the z-direction with a laser profilometer for all the wall samples. Figure 5.11 shows the variation in roughness values with wall width. Clearly, the average roughness measured on the 8 mm wide wall is separated from the general trend of slightly increasing roughness with decreases in wall width. The 10 mm wall shows a very similar average roughness to sample C0 previously measured with the 207

208 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES Frequency = 3.21 mm 1 Period = 0.31 mm Power Frequency (mm 1 ) Figure 5.10: Periodogram resulting from the application of an fast Fourier transform algorithm to the pore volume fraction with build height in the high resolution XCT data of the 10 mm wide wall shown in Figure 5.9b Ra (µm) Wall width (mm) Figure 5.11: Effect of wall thickness on surface roughness in the z direction across the samples. Error bars indicate the standard deviation between the line measurements. (GA samples.) 208

209 5.2. EFFECT OF SAMPLE GEOMETRY AND LOCATION IN BUILD CHAMBER 5 mm Figure 5.12: Visualisation of the defects observed in the 10 mm wide wall manufactured with only the hatching theme activated. (GA sample.) confocal microscope. However, the change in surface roughness is small and it appears that factors other than wall thickness are important in determining the roughness of vertically built surfaces. In addition to walls built with standard Arcam settings (prior to the update), samples with the same geometry were manufactured but with only the hatching strategy applied. Only the largest (10 mm wide) wall was analysed by XCT and a visualisation of the defects observed is shown in Figure Clearly, the tunnel defects are present in this sample and, once again, are concentrated at one edge. They do however propagate further in to the sample, i.e. greater than 1 mm from the surface, than previously observed Influence of Sample Geometry and Orientation When the standard geometries defined in Figure 3.9 were examined using XCT, further large tunnel defects were observed. Figure 5.13 shows a visualisation of the defects identified in all the samples. The different sample geometries and orientations within the build chamber clearly influenced the likelihood of tunnel defects appearing. The pore volume fractions detected by both XCT and optical microscopy are shown in Figure 5.14, while statistical results from the quantification of individual pores within each of the samples is shown in Table 5.2. Comparison of the XCT results in Figures 5.13 and 5.14 reveals that those samples found to contain tunnel defects were also found to have a larger volume fraction of porosity. The pore volume fraction detected by optical microscopy depended on whether a tunnel defect was present in the examined images. In all cases, where the pore volume fraction was measured to be lower with the optical microscopy, compared to the XCT, it was because tunnel defects were present in the sample that did not intersect the imaged plane. In contrast, the 2D images of sample Gs1 contained multiple tunnel defects and resulted in the higher volume fraction recorded by microscopy than XCT. The tunnel defects have a tendency to be 209

210 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES Gs1 Gs2 Gc1 Gc2 Gt1 Gt2 Gt2 10 mm Figure 5.13: Lower resolution 3D XCT full sample scans showing visualisation of all the pores (red) detected in samples with various geometries and build directions. The build direction is vertically upwards in the plane of the page. Note the presence of large tunnel defects at the edges of samples. (GA samples.) Pore volume fraction (%) XCT Optical microscopy Gs1 Gs2 Gc1 Gc2 Gt1 Gt2 Gt3 Sample Figure 5.14: Variation in pore volume fraction due to sample geometry and build direction, measured by low resolution XCT and optical microscopy. (GA samples.) 210

211 5.2. EFFECT OF SAMPLE GEOMETRY AND LOCATION IN BUILD CHAMBER Table 5.2: Individual pore quantification by XCT and optical microscopy in samples manufactured to analyse the effect of geometry and build direction on defect population. (GA samples.) Sample XCT Optical microscopy Mean equiv. Max. equiv. Mean equiv. Max. equiv. diameter (µm) diameter (µm) diameter (µm) diameter (µm) Gs Gs Gc Gc Gt Gt Gt (a) (b) 5 mm Figure 5.15: Example slice of XCT data showing tunnel defects in proximity to surface in sample: (a) Gs2; and (b) Gc2. The grey-scale colouring indicates relative X-ray attenuation at each point. Note how the defect stays approximately 1 mm from the edge even when this involves non-vertical growth. (GA samples.) located near sample edges, and in particular, near the corners. This tendency was even observed when the sample edge was non-vertical. Furthermore, none of the tunnel defects propagate through the contouring region and only penetrate the sample surface when hatching is used at the surface. Example slices from the XCT data in Figure 5.15 clearly illustrate both these effects for samples Gs2 and Gc2. Finally, in samples Gs2 and Gt3 there is a significant concentration of porosity in the lower vertex. 211

212 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES Influence of Sample Geometry and Location in the Build Chamber Further investigation of the influence of sample geometry was carried out with the software update and PA powder. The location of these samples (Gp1 3) within the build chamber was also carefully recorded, as shown in Figure All the pores detected by macro-scale XCT (9.9 µm voxel size) in a 16 mm height for the three different shaped samples have been compressed in the build direction in to the x-y plane in Figure Figure 5.16 also shows the approximate location in the build chamber of the three specimens after the 25 mm spacing between samples in the x direction has been reduced. The pore volume fractions have also been averaged with distance from the sample surfaces to the centre of each section in the x-y plane, and these results are plotted in Figure From Figures 5.16 and 5.17 it is apparent that all the specimens had a low volume fraction of pores near their edge, within a distance consistent with the powder melted by the contouring section outline passes, whereas in the middle hatching region, the average pore density was substantially higher. Moving inwards, two peaks in porosity are also observed in the data in Figure 5.17, similar to peaks I and II in Figure Once again, the first smaller peak is at a distance of 0.8 mm from the section surface, which coincides with a position at the end of the hatch pass (shown by the light lines in Figure 5.16a c). In particular, a large volume of pores were found to cluster in the vertex region marked (i) of the triangular section sample (Figure 5.16c). The Figure 5.16: Lower resolution 3D XCT full scans of samples: (a) Gp1; (b) Gp2; and (c) Gp3; showing visualisation of all the pores (red) detected in samples manufactured to examine the effect of location in build chamber projected in to the x-y plane to allow visualisation of their spatial distribution, showing the effect of sample geometry and location in the build chamber. Note that during the build, the samples where arranged with 30 mm between them and no other samples were present. In all images the thin blue line indicates the approximate location of the centre line of the final contour pass and first hatch. (PA samples.) 212

213 5.2. EFFECT OF SAMPLE GEOMETRY AND LOCATION IN BUILD CHAMBER Pore volume fraction (%) Square Circular Triangular Distance in x-y plane from surface (mm) Figure 5.17: Variation in pore volume fraction moving inwards from the sample surface, for the square circular, and triangular cross section samples, obtained by low resolution XCT. (PA samples.) circular and triangular samples also showed a high second peak (similar to peak II in Figure 4.10) in pore volume fraction between 1.6 and 2 mm from their surfaces (in the x-y plane). In averaging the data with distance from the sample faces this peak has been affected by the clustering of defects in specific locations in the circular and triangular sections. Specifically, towards the top edge of the hatching region in the circular sample (Figure 5.16b), with this sample, because of its shape and relative position in the build chamber, the hatching in the x-direction only included the exposed chords of the cross section not aligned with the other samples. Pores were also concentrated within the apex marked (ii) of the triangular sample (Figure 5.16c), where the hatch length in y becomes progressively smaller. While this peak in pore volume was not as apparent with the square cross section (Figures 5.16a and 5.17), the cuboid sample equally showed a peak in pore number density at the same distance from its edge. It should be noted that in all cases the defects had a similar distribution to that recorded for sample C0, i.e. irregular pores in the contouring and mainly spherical in the hatching regions. A single cuboid was also manufactured raised up 10 mm from the baseplate with no attached supporting structure and no other surrounding samples. The measured volume fraction of porosity is plotted against the approximate height in the build chamber in Figure 5.18 alongside an example slice from the XCT data used to measure the pore volume fraction. A very large increase in pore volume fraction was observed in the lower 1 mm of this sample that was built directly onto the powder, 10 mm to 11 mm in Figure 5.18a. From Figure 5.18b it is clear that this peak is caused by the presence of tunnel defects, originating from the first layer melted onto powder. The region affected by tunnel defects is however fairly small, with all the tunnel defects failing to 213

214 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES Approximate height (mm) (a) Pore volume fraction (%) (b) Figure 5.18: Effect of an overhang on pore volume fraction: (a) measured pore volume fraction against approximate height in the build chamber; and (b) example slice of XCT data showing tunnel defects which were not detected by automated segmentation techniques. Graph and image to same distance scale. (PA sample.) grow further than 1.1 mm into the sample. The average volume fraction of porosity is % in the lower 1 mm of the sample, whereas, above 1 mm from the lower surface the volume fraction drops to %, identical to that in sample C0. Finally, the melted area was also lower in the region closest to the powder, and the lower surface was not flat as designed in the CAD model. Instead, the bottom surface curves upward near the sample edges Relationship between Defect Population and Microstructure Variation 1 To examine whether any relationship could be observed between microstructure variation and the volume of porosity, an automated procedure was employed to map the titanium α phase plate width in the transformation microstructure across the triangular sample (Gp3) cross section. The result from this approach is shown in Figure 5.19 and involved image processing a regularly spaced array (200 µm pitch) of 1124 high magnification (10,000 ) backscatter SEM images (this work was not conducted by the author of this thesis and full details of the procedure are available elsewhere [177]). From the resultant α plate width map it is apparent that there is some coincidence between the locations of coarse pore clusters and a smaller α lamellar width. For example, in the section of the sample shown it can be observed that there is a region of 1 The results in this section (5.2.4) were collected by Hao Zhao and should not be regarded as the work of the author of this thesis. However, the experiment was conducted on a sample manufactured by the author and provides insight into possible reasons for pore formation. 214

215 5.2. EFFECT OF SAMPLE GEOMETRY AND LOCATION IN BUILD CHAMBER 2 mm Mean plate spacing (µm) Figure 5.19: Results from the automated image analysis mapping of the α phase lamellar spacing in the triangular section sample (mid plane) conducted by Hao Zhao. (PA sample.) reduced α lamellar width that matches closely with the location of increased porosity in the vertex region seen in Figure 5.16c. 215

216 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES 5.3 Discussion The results presented in this chapter have demonstrated that the defect population is strongly correlated to both the melt strategies and sample geometries. Furthermore, it has been shown that the volume fraction of defects can be strongly reduced by minor changes to the standard settings. By discussing the influence of the electron beam melt strategy on defect population first, the reasoning behind some geometries being more prone to porosity will become more apparent Influence of Energy Density To better understand the differences between the samples manufactured with modifications to the melt strategy, the overall average energy density used to melt each of their cross sections has been calculated by integrating the beam velocity, taken from the respective sample build file, over the entire hatch length (i.e. including the effect of the turning function) and summing the proportions from the contouring and hatching regions relative to their respective areas. Figure 5.20 shows the calculated average energy density plotted against the observed volume fraction of porosity for the individual changes to the process (C0 to C7), as well as the effect of more systematic changes to the speed function and line offset. From Figure 5.20 it appears that the energy density and porosity volume fraction have an inverse relationship. Therefore, it seems likely that as an increased energy density results in larger and deeper melt pools, with greater overlap and re-melting in repeated passes, this gives more opportunity for gas bubbles to escape. Although the energy density provides a convenient way for broadly comparing the effect of line offset and beam speed, it appears, from the scatter in Figure 5.20, that other individual factors can be more important. For example, lowering the beam speed (samples S0 to S2,), which will increase both the melt pool width and depth, appears to be more beneficial than reducing the overlap (samples L0 to L2). In addition, it appears that a lower focus offset has a very substantial effect on the measured volume fraction of pores (samples F1, F2 and F3). The beneficial effect of focus and other aspects of the beam control, such as the turning function, are discussed further below. Unfortunately, there is currently little published data from SEBM-AM samples with which to corroborate these observations. With similar laser based (e.g. SLM) processes, results reported on the effect of energy density on residual porosity levels are very scattered, as individual variables such as speed can have separate effects on melt pool stability. Hence, higher energy densities do not always result in more dense parts [80, 132, 133, 172, 178]. Previous SEBM work has shown that when the beam speed or line offset is reduced, which would increase the energy density, the volume 216

217 5.3. DISCUSSION Pore volume fraction (%) 0.1 L0 S0 Standard focus 0.08 C2 C0 Increased focus C4 C C3 S1 C L1 L F3 C7 F1 0 F2 C1 S2 S Mean energy density (J mm 3 ) Figure 5.20: Sample average pore volume fractions, measured by low resolution XCT, plotted against the average energy density for a whole layer in each sample, produced with different process parameters (see Table 5.1 and Table 3.3). The background colour indicates whether the focus offset was different for the majority of the build. fraction of lack of fusion defects can be reduced [77, 172]. The presence of very large tunnel defects (none of which were observed in the samples manufactured with modifications to their melt strategy using the most recent Arcam methodologies) has equally been predicted to be caused by an insufficient energy input [76]. However, the results here suggest that the volume fraction of gas pores, as well as that of lack of fusion defects, is reduced when the energy density of the electron beam is increased above the Arcam recommended standard settings. In direct laser melting of stainless steel powders, increasing the energy input to aid removal of gas pores has equally been shown to lead to lower porosity; in that work it was suggested that remelting of previously solidified layers without more material deposition was effective in helping gas bubbles to escape [129]. Increasing energy density is not without drawbacks; in addition to elevated operating costs, it is associated with an increased level of aluminium evaporation [65], which may alter both the microstructure and mechanical properties Effect of the Different Melt Strategies The observed trends in porosity density as a function of distance from the surface measured in the samples (Figures 4.10, 5.3 and 5.4), suggest that the pore distribution is influenced, in particular, by differences between the contouring and hatching beam parameters. A substantially lower average porosity was repeatedly found in the contouring region than in the hatching area, which had a far larger pore volume fraction (Table 4.1), although in the hatch region they were mainly spherical gas pores (Figures 4.7 and 4.8). Thus, it is apparent that the smaller melt pool, higher beam speed, and 217

218 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES reduced overlap between layers associated with hatching has given less opportunity for any gas bubbles to escape. A further important observation is that the pores were not uniformly distributed across the standard sample section, even if the beneficial effect of the contour passes was not considered. Peaks in porosity were repeatedly found at two specific locations; at the absolute edge of the hatching area and further inwards where the pore volume fraction rises towards the edge of the hatch region. These peaks were previously identified as I and II in Figure In the sample sections built with standard parameters, a smaller narrow peak (I) in pore volume fraction was seen at 0.9 mm in from the edge, where relatively large spherical pores dominated (Figure 4.10). In contrast the larger peak (II), builds up sharply at depth of 1.6 mm before decaying towards the build centre to the higher level seen in the hatched region. A similar trend for pores to be located near the edge of the hatch region was observed in metallographic images of non-optimised builds by Karlsson et al. [29]. In the work here these two positions of pore concentration have been shown to be clearly related to the edge of the hatching lines. For example, when a sample was produced just by using contouring, which spiralled inwards from the edge, very few pores and no peaks in the pore population were found. In addition, when the hatching strategy was changed to rastering in one direction only, pores were found at the end of the hatching lines at the position of peak I. This implies that pores pushed forward by the solidification front tend to be dumped at the end of the hatching track and it is likely that this also happens when the beam abruptly reverses direction when the normal weaving raster pattern is used. Whatever the origin of these pores, it is apparent that they were removed by re-melting the edge of the hatching area, when contouring was performed after hatching (sample C5, Figure 5.4c). Hence, although the higher heat input and larger overlap of the contour melt tracks tended to reduce gas porosity in general, by allowing more opportunity for gas bubbles to escape from the larger melt pool, there also appears to be a negative effect of a sudden change of beam direction associated with hatching, which does not happen in contouring. The presence of peak II within the hatching area can be explained by considering the effect of the turning function which increases the beam speed, reducing both the energy density and the time available for gas bubbles to escape towards the start of each raster line (see Figure 5.2). A moving melt pool will also become narrower and less deep, in response to a higher travel speed, thereby reducing the melt track overlap, if this is not adequately compensated for by residual heat from the previous pass. In fact, the rise in porosity towards the edge of the hatching region shown in Figure 5.4 mirrors the change in energy density plotted in Figure 5.2b. Hence, the turning function appears to increase the probability of gas pores remaining trapped in the melt pool close 218

219 5.3. DISCUSSION to the edge of the hatching area. This implies that the energy input near the edge of the hatch region is reduced too far by this control function, and there is not sufficient re-melt depth to give gas adequate opportunity to escape from the melt pool. This conclusion has been confirmed by the observation that the peak in porosity near the edge of the hatched area disappeared when the turning function was disabled (Figures 5.3h & 5.4d) which would increase the energy density at the edge of the hatching area. In addition, in the sample produced when hatching with a constant beam speed (C7), the overall volume fraction of porosity was lower across the entire section, and an inverse behaviour was found, in that the pore density decreased towards the edge of the sample (Figures 5.3h & 5.4d). This occurs because when the turning function is disabled, the melt tracks will actually become wider and deeper than in the centre near the edge of the hatch region, because of the residual heat from the forward beam travel. Although powder bed AM is most suited to producing complex, relatively small components, it should also be noted that when melting parts with thicker sections or longer hatch lengths, the turning function will have a smaller effect. All the samples studied that had modifications to the melt strategy were both small and melted individually, i.e. the hatching turned back on itself within the small sample and thus the turning function was critical in determining the beam speed. Conversely, in the samples manufactured to analyse the effect of geometry on defect population, the hatching was conducted across all the sample in the bed, i.e. only the samples at the edge of the bed would be effected by the turning function. As the control software modifies parameters based on the line length, the average energy density will be higher and the average volume fraction of porosity is lower, as shown in section 5.2; this effect is discussed further in subsection The trough in pore volume fraction between peaks I and II is not as easily explainable without more information concerning the transient melt pool behaviour, as the beam path reverses direction through a tight arc as it moves to the next hatch line (Figure 5.1a). This relatively low porosity region was still observed when only hatching was used to melt the sample (C2, Figure 5.4a). When the electron beam reverses direction it will first have to accelerate in the reverse direction to the high speed requested by the turning function. This region, which is close to the turning point, therefore, probably does benefit from the additional residual heat from the previous track. In addition, if gas bubbles that build up by being at the rear melt pool surface in the forward motion of the prior beam track are dumped out during turning, there will be a brief period before the number of bubbles build up again to a steady state level. It is notable that the porosity recorded within the contour only sample (C1) was significantly lower than for the hatched sample S1 which was produced with the same energy density via modifications to the speed function. This could be related to the lack 219

220 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES of beam turning in the contour patterns, but with samples F0 F3 it has been shown that this is could also be due to the more focused beam used by the Arcam machine when contouring is used, rather than hatching. The reduction in pore volume factions in samples F0 F3 suggests that when the focus of the beam is increased, the probability of gas bubbles escaping the melt dramatically rises. Al-Bermani [14] has shown that in the SEBM process, the focus offset had a significant effect on the melt pool geometry. In particular, a lower focus offset (closer to 0 ma) was found to result in a narrower but deeper melt pool. Thus, increasing the amount of remelting of the previous layer seems to be an important factor in reducing the level of porosity. This could also be the main benefit from increasing the energy density by reducing the travel speed in the results described above. Indeed, in Al-Bermanis work, the deepest melt pool was measured with a focus offset of 10 ma that produced a focal point just above the surface [14]. This would imply that samples F1 and F2 would have also have had greater melt pool depths, which could assist in the escape of gas bubbles, by a greater level of re-melting of the previously consolidated layer Influence of Sample Geometry and Location As discussed above, the melt pool geometry, and in particular melt depth, is crucial in determining the local volume fraction of gas bubbles that are trapped in the melt pool. When the geometry is altered from a constant cross section, the electron beam behaviour is altered in an attempt to compensate for changes to the thermal conditions, with the intention of keeping a constant melt pool size. It has already been shown that the control algorithms do not provide a constant heat input in a constant cross section due to the application of the turning function. The situation is more complex when melting multiple samples in a single build, where the default hatching strategy is to include all the models in the melt pattern. Multiple models, aligned in the current hatching direction (x or y), will be melted in a single hatching line before repeating an adjacent hatch line. A simple graphical example of this process is provided in Figure Between the models, the beam moves across the powder bed near instantaneously and thus avoids melting. Therefore, when analysing the effect of sample geometry on defect population, it is important to consider the behaviour of the electron beam in addition to any changes to the local thermal properties. In fact, the corrections applied to the beam speed may be more important in determining melt pool depth than sample geometry alone. Due to the significant changes ot the SEBM process implemented during this project, it is appropriate to separate the samples into those built prior to the updates (denoted GA) and those built following the installation of the latest software and new powder. First those built prior to the updates will be discussed. When melting the walls of var- 220

221 5.3. DISCUSSION Figure 5.21: Hatching pattern when melting multiple models. The control software considers all models and the total line length when calculating the hatching beam current and speed. ious thicknesses the highest volume fraction of porosity was recorded in the walls of 2 mm and 10 mm width (Figure 5.7). Importantly, the wall thickness samples were built as a single model and these two samples were at the end of the x-direction hatching, and thus would be most affected by any transient effects caused by the beam movement reversal, whereas the beam would not reverse in x within the other samples. The high resolution XCT analysis of the 10 mm wide wall revealed that the spherical gas porosity had a significant periodicity in height corresponding to four layer thicknesses (Figure 5.10). This wavelength is not surprising since the hatching theme scanning pattern rotates by 90 every layer, which leads to every fourth layer having the same direction of beam travel. This observation suggests that this causes a difference in melt pool size every fourth layer, although without the exact beam settings it is unclear what this would be. However, the change is likely to be related to the change in hatch length which had different distances of 82 mm in x and only 15 mm in y when the samples were built in a single model, resulting in changes to the beam current, speed and application of the turning function. The gradual increase in pore volume fraction with wall width is consistent with the increasing fraction of the sample melted with the hatching strategy, and in particular the edge of the hatching near the overlap with the contouring, which has been shown to contain more pores. In contrast, only contouring was used to manufacture the two smallest walls and a very low pore volume fraction was recorded. When the samples with varying geometry and build direction (Gs, Gc & Gt) were examined, it was clear that both factors had significant influence on the on defect popu- 221

222 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES lation. Unfortunately the position of these samples in the build chamber was not accurately recorded and they were built alongside other components which would also alter the hatching strategy. However, it is clear that the tunnel defects were more frequent in certain locations, such as at one of the vertical walls of sample Gs2 (Figure 5.13). Tunnel defects are caused when the beam line energy is too low [76] and it is likely that their appearance is a function of both sample location in the build chamber and geometry. If they were due to either sample geometry or poor overlap between the hatching and contouring, the tunnel defects would likely be more symmetrically distributed than observed in Figure It seems likely that they occur at the end of a hatch pass where the beam reverses direction. This theory is supported by the presence of tunnel defects in the wall manufactured with just the hatching theme activated which showed tunnel defects only at one edge of the sample where the beam reversed direction (Figure 5.12). In short, a likely origin for the tunnel defects is the turning function, reducing the line energy of the beam too far in all the GA samples. With regards to the surface roughness of the various walls, it appears that reducing the wall width increases the R a. However, the maximum surface roughness measured here is still far below that recorded by Chan et al. [100] who used a larger layer thickness than used for these samples (100 µm compared to 70 µm). The thicker layers used by Chan et al. [100] may have changed the surface tension induced melt bead profile and increased the roughness. In addition, the roughness recorded for the 8 mm wall was significantly higher than all other samples. Since the powder size was constant, increased roughness must be due to increased undulations in the surface, i.e. the permeation of the melt pool into the surrounding powder bed is more variable. This implies poorer electron beam control during the contouring strategy. The reason for this is unclear without more detailed knowledge of the process algorithms, but could be due to surrounding samples altering the behaviour of the contouring. For example, increasing the distance between samples means the beam has to move further between samples and the very high speed movement of the beam between samples may result in reduced spacial accuracy in its new location. Fortunately, following the application of the software update, tunnel defects were largely removed from the process, but the geometry and location of samples was still found to influence the distribution of other smaller defects (Figure 5.16). From Figure 5.16 it appears that sample location in the build chamber is more important in terms of defect population than sample geometry alone. For example, on one side of the circle sample (Figure 5.16b) there is a significant increase in the number of pores. This position is also where the hatching length in the x-direction is shortest, thus the beam would be reversing direction most often and the turning function most influential. Similarly, in the position marked (ii) in the triangular sample (Figure 5.16c), shorter hatch melt lengths in the y-direction led to a greater effect of the turning function, thus lower 222

223 5.3. DISCUSSION heat input and more porosity. This local reduction in net heat input near the apex of the triangular sample is confirmed by the α plate thickness map in Figure 5.19, which shows a narrower average plate width in the same region. A smaller, shallower melt pool in this location is likely to be the reason behind the greater concentration of gas porosity in this location. From these results it appears that ensuring gas porosity is kept low in any arbitrary geometry is dependent on maintaining a melt pool depth sufficient to give the gas bubbles adequate opportunity to escape. Unfortunately, it appears that the corrections for geometry applied by the Arcam control software are too great, and locations where it may be expected the heat would build up from the beam repeatedly turning back on itself, such as in the apex of the triangle, actually receive less average energy than other regions. Therefore, to reduce concentrations of the porosity, some refinement of these functions is required to maintain a constant melt pool depth. If this problem is not addressed, it could be particularly problematic when trying to utilise the SEBM process for industrial applications, where it would mean that the arrangement in the build chamber for particular components may have to be fixed, as well as individual component geometry and process settings. The peak in pore volume at position (i) in sample Gp3 (Figure 5.16c) may be less avoidable by altering the effect of the turning function. As discussed in subsection 5.3.2, this peak in porosity is thought to be caused by gas bubbles being deposited at the end of the hatch line. At this apex of the triangle, the following hatch passes are less likely to result in sufficient remelting to allow this gas to escape or move it forward in the new hatch pass. Finally, the increased volume fraction of porosity at the bottom of a sample raised up from the baseplate (Figure 5.18), is further indication of the overcompensation applied by the control software when there is potential for the heat to build up. The thickness function increases the beam speed without increasing the beam energy when melting sections with a depth less than 4 mm. This reduction in line energy resulted in tunnel defects forming and growing upwards through the sample. However, as the effect of the thickness function reduced, and the beam energy increased, the tunnel defects were prevented from growing any further into the sample and the pore volume fraction became very similar to that observed in a bulk sample of the same geometry. 223

224 CHAPTER 5. MELT STRATEGIES AND SAMPLE GEOMETRIES 5.4 Summary The pores/defects have been conclusively shown not to be randomly distributed, and a strong correlation was found with the process parameters and strategies used to outline (contouring) and infill (hatching) a part section, making their impact on fatigue life potentially more significant. Few gas pores were found within the surface layer melted by the contour passes, with the majority being concentrated in the infill hatched area. This behaviour has been attributed to the higher energy density used in the contour step. This produces a larger, and more importantly, deeper melt pool, giving more opportunity for gas bubbles to escape by encouraging a greater level of re-melting of the previous layer. The irregular shaped pores found within the contour region are assumed to be an artefact of the MultiBeam setting not giving sufficient melt pool stability, but more work is needed to definitively prove this connection. Overall, the lower average energy density and shallower melt pool depth in the hatching region clearly correlated to a higher average gas pore density. Simple changes to the process parameters to increase the energy density in this region produced significant reductions in the pore populations. In addition, the use of a more focused beam offered the opportunity to reduce the gas porosity without increasing the energy input, owing to the deeper melt pool this generates. Under standard build conditions, moving in from a section edge, two peaks in porosity were seen at depths of 0.9 mm and 1.6 mm to 2 mm. Both peaks were related to the edge of the hatching region. The first peak was formed by gas bubbles being moved to the edge of each hatch pass and subsequently deposited at the end of a hatch line, when the melt pool changes direction. The second peak is thought to be due to the effect of the turning function within the hatching region. It appears that this function overcompensates for residual heat left by the forward beam pass, by over accelerating the beam when it reverses its trajectory. This results in too low an energy density, which leads to more gas pores being seen near the edge of the hatched area. A reduction in energy density can also be used to explain the different levels of gas porosity observed in different geometries. When the hatch length is shorter, and thus the turning function is more influential, there both appears to be a lower energy input, reflected in the microstructural size, and there is higher pore volume fraction. In the older samples, tunnel defects were observed and these are thought to be caused by the hatching strategy failing to provide sufficient line energy, especially at the end of hatch passes, which results in the melt track beading and not wetting the voids in the previous layers. 224

225 6 Influence of External Changes to SEBM-AM Methodology on Porosity To reduce the level of porosity in SEBM components, without altering the control software, changes can be made to the powder feedstock or by adding an extra processing step post manufacture. The results presented in the previous chapter suggested that a recent change in Arcam supplied powder, from GA to PA, may have actually resulted in an increase in the volume fraction of gas porosity, due to the increased level of porosity found in the PA powder. To study the influence of powder in more detail, in this chapter other powder types have therefore been used in the SEBM process to attempt to correlate the volume fraction of porosity seen in the consolidated martial to that found in the powder feedstock. In addition, hot isostatic pressing (HIPing) is now being widely implemented by industry to remove porosity before putting SEBM components into service. Thus, the effectiveness of HIPing in removing all the defect types discussed in chapter 4 has also been analysed. XCT was used to quantify the effects of both the powder feedstock type and HIPing on the pore volume fraction in built samples. By conducting high resolution XCT examination of the different powder feedstock, quantifying the volume fraction of porosity, and then manufacturing and analysing samples with different powders, it was possible to explore any relationship between the porosity in the built parts and the feedstock. Likewise, XCT was used to examine the same SEBM samples in an as-built condition and after undergoing a HIP cycle, so that the non-destructive nature of the technique could be used to allow the effect of the HIP cycle on individual pores to be quantified. 225

226 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY 6.1 Influence of Powder Feedstock on Porosity in Consolidated Samples A total of six different powder types, including the two already analysed in section 4.4, were quantified in terms of the size distribution and volume fraction of porosity they contained and were then used to manufacture components by SEBM. These powders consisted of two manufactured by gas atomisation (GA), one by plasma atomisation (PA), one produced by the plasma rotating electrode process (PREP) and two by the hydride-dehydride (HDH) and milling process. Each of the virgin powders was subjected to high resolution XCT analysis using a 2.0 µm voxel size, which was able to detect pores down to an equivalent diameter of 5 µm when using a small volume of powder contained in a 1.7 mm diameter polymer tube (see subsection 3.5.4). Following this analysis, 12 mm diameter samples were machined from the centre of standard cuboid samples (15 mm 15 mm 70 mm) manufactured from each powder type using similar arrangements in the build chamber. The machined samples were used to allow higher resolution and reduce any transient effects on the porosity observed near sample edges as described in subsection These samples were examined at a lower resolution than the powder using a 7.5 µm voxel size, which was able to detect pores down to an equivalent diameter of 19 µm. All of these AM samples, except the one manufactured with original Arcam GA powder, were built after the recent software updates Powder Analysis A summary of the measurements made from all six powder types is shown in Table 6.1. Where more than one powder with the same manufacturing process was examined (i.e. the same process but from different commercial suppliers), they were numbered with the different suffixes as shown. The different powder types have a wide range of pore Table 6.1: Summary of powder size statistics and detected pore volume fraction contained within particles in different powder feedstocks. Superscripts indicate the current Arcam supplied PA powder (a) and the original GA powder (b). Powder Number of Ratio Mean equiv. Standard Max. equiv. Pore volume type particles porous(%) diameter (µm) deviation (µm) diameter (µm) fraction (%) PA a GA-1 b GA PREP HDH HDH

227 6.1. INFLUENCE OF POWDER FEEDSTOCK ON POROSITY volume fractions and percentage of particles that contain pores. The size distribution of powder types PA and GA-1 (the two powder types supplied by Arcam) have already been represented by histograms in Figure 4.24 (section 4.4), which showed an increase in the ratio of porous particles as the particle size increased. These histograms have been reprinted in Figure 6.1, to allow easier comparison to the histograms for the remaining four powder types shown alongside. When interpreting these graphs, it should be noted that the y-axes have been scaled to highlight trends in the data. A general increase in the ratio of porous particles with particle size is evident in Figure 6.1. Some powder types also contained more pores than others, with the current PA powder supplied by Arcam being the most porous when measured by the ratio of porous particles, and the GA-2 powder being the most porous when measured by pore volume fraction. The particle size histograms of all the powder types presented in Figure 6.1 also shows a significant variation in their size distributions. For example, powder GA-2 had a much lower modal and mean particle size, which also resulted in a far higher number of particles being quantified in the same field of view (Table 6.1). 227

228 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY (a) (b) (c) Number of powder particles Number of powder particles Number of powder particles Particles Porous particles Ratio porous Equivalent diameter (µm) Particles Porous particles Ratio porous Equivalent diameter (µm) Particles Porous particles Ratio porous Equivalent diameter (µm) Ratio porous Ratio porous Ratio porous Figure 6.1: Statistical analysis of the size distributions of the different powder feedstock tested, showing both all the particles and only the particles containing porosity. Also indicated is the fraction of particles found to contain porosity. Powder types: (a) PA (current Arcam); and (b) GA-1 (previous Arcam); (c) GA-2. Figure continues on following page. 228

229 6.1. INFLUENCE OF POWDER FEEDSTOCK ON POROSITY (d) (e) (f) Number of powder particles Number of powder particles Number of powder particles Particles Porous particles Ratio porous Equivalent diameter (µm) Particles Porous particles Ratio porous Equivalent diameter (µm) Particles Porous particles Ratio porous Equivalent diameter (µm) Figure 6.1 continued: (d) PREP; (e) HDH-1; and (f) HDH-2. 1 Ratio porous Ratio porous Ratio porous 229

230 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY Comparison of Porosity Observed in Built Samples and Powder Feedstock In Figure 6.2 the volume fraction of porosity measured by lower resolution XCT in the consolidated AM material has been plotted against the pore volume fraction in the powder measured by high resolution XCT. Despite the different resolutions used, it appears that the two variables are correlated and a linear trend line has been fitted by the least-squares method to illustrate this. In other words, the total volume fraction of detected pores (>19 µm in diameter) in the solid material can be related to the volume fraction of measured pores (>5 µm in diameter) in the powder. However, there is some scatter in the data. For example, both GA powders had a volume fraction of porosity in the solid material below the trend line. It has been demonstrated previously that the porosity volume fraction can be affected by the sample position in the build chamber (subsection 5.3.3) and this may account for some of the scatter in the data. However, for these sample their positions were unfortunately not all recorded. Pore volume fraction in consolidated material (%) PA GA HDH PREP Pore volume fraction in powder (%) Figure 6.2: Comparison between the pore volume fraction detected by XCT in the powder feedstock measured by high resolution XCT to that seen in consolidated material measured by lower resolution XCT. Linear trend line fitted by the least-squares method. 230

231 6.2. EFFECT OF POST MANUFACTURE HIPING ON POROSITY 6.2 Effect of Post Manufacture Hot Isostatic Pressing on Porosity To examine the effect of post manufacture HIPing on porosity in SEBM parts, standard as-built samples were subjected to full volume, low resolution XCT (9.9 µm voxel size) as-built to identify all the coarse porosity they contained, before being HIPed and rescanned to allow direct correlation between individual pores before and after HIPing. The image registration tool in the Avizo software was used to align the datasets using a least squares method to match their grey levels. In addition, the small 1.7 mm diameter cylindrical specimens, machined from sample Gs1, (described in subsection 3.2.1, analysed in section 4.3), were also HIPed to allow higher resolution (2.0 µm voxel size) imaging of the regions known to previously contain porosity Coarse Pores in Whole As-Built Samples An example of a 3D visualisation of the pores detected in the standard samples Gc1 H and Gt3 H as-built is given in Figure 6.3a while the pores detected following HIPing are shown in Figure 6.3b. In sample Gc1 H, it can be seen by comparison of the images in Figure 6.3 and the net pore volume fraction data presented in Figure 6.5, that all the detectable internal porosity present after AM were removed by the HIPing treatment. In contrast, while all the smaller gas porosity in sample Gt3 H was removed by HIPing, some tunnel defects still persisted in the HIPed samples. Prior to analysis of the post-hip XCT results, the data sets were exactly aligned with the data collected from the samples in the as-built condition. Thus, the example slices compared in Figure 6.4 from geometries Gc2 H and Gt3 H show the same internal slice of each sample before and after HIPing. In Figure 6.4a it can be seen that even the coarse tunnel defects have all been removed from the cylindrical sample, whereas in Figure 6.4b, with the inverted prism geometry, some remain. Detailed examination of individual slices revealed that all the tunnel defects remaining after HIPing in sample Gt3 H were connected by ligaments to the surface, even if the ligaments were too small for automatic segmentation to detect them. In contrast, porosity completely enclosed by solid material was healed below the detection limit following HIPing; i.e. it had shrunk in size to below the resolution of the equipment for the full sample scans and was not detectable by either automatic segmentation (as shown in Figure 6.3) or manual examination of slices (Figure 6.4). The average volume fraction of porosity detected by automatic segmentation of the as-built and HIPed samples is shown in Figure 6.5. In all cases the detected pore volume fractions had reduced following the HIP cycle. In sample geometries Gs2 H, Gc1 H, Gc2 H, and Gt2 H this was to below the detectable limit of the coarse scan. All the pores that persisted after the HIP cycle were tunnel defects that were found to have 231

232 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY (a) Gc1 H Gt3 H (b) 5 mm Figure 6.3: Isometric 3D visualisation of defects (red) in samples Gc1 H and Gt3 H : (a) as-built; and (b) after HIPing. All detectable internal porosity was removed in sample Gc1 H, whereas in sample Gt3 H tunnel defects connected to the surface persisted in the material. 232

233 6.2. EFFECT OF POST MANUFACTURE HIPING ON POROSITY (a) (b) 5 mm Figure 6.4: Examples of aligned slices of the XCT data collected from samples asbuilt (left) and after HIPing (right): (a) Gc2 H ; and (b) Gt3 H. In (b) note the tunnel defects that breach the surface (arrow) that are still present after HIPing. Pore volume fraction (%) As-built Post HIP cycle Gs1 H Gs2 H Gc1 H Gc2 H Gt1 H Gt2 H Gt3 H Sample Figure 6.5: Pore volume fraction measured by low resolution (9.9 µm voxel size) XCT in SEBM Ti-6Al-4V samples as-built and after a HIP cycle. 233

234 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY breached the top surface of the samples. Only samples with a top surface melted by the hatching strategy had this open porosity present after HIPing (i.e. Gs1 H, Gt1 H and Gt3 H ). In contrast, tunnel defects present in samples built where the upper surface had been melted with the contour strategy (such as Gs2 H, Gc2 H and Gt2 H ) were not connected to the surface and were closed by the HIPing process, despite the similar coarse nature of the original flaws. All of these samples showed no detectable porosity following HIPing at the 9.9 µm resolution limit of the full sample scan Fine Porosity within Machined Cylinders High resolution XCT examination of the two small cylindrical specimens machined from the edge and centre of sample Gs1 again revealed no small pores could be detected following HIPing, either by automatic segmentation, or manual examination of the data. High resolution example slices from both cylinders are provided in Figure 6.6, again showing the same regions before and after HIPing. It can be seen that the two pores types initially present, a spherical gas pore (Figure 6.6a) and a lack of fusion defect (Figure 6.6b), are both undetectable in the XCT data acquired following HIPing. The aligned XCT data also allowed targeted manual examination of volumes where pores where known to exist prior to HIPing. However, even if the 8 voxel minimum size requirement was removed, no pores with a size great enough to significantly affect the voxel grey value could be detected, indicating that these pores had shrunk to a size below 1 voxel or 2 µm in diameter. 234

235 6.2. EFFECT OF POST MANUFACTURE HIPING ON POROSITY (a) (b) 500 µm Figure 6.6: Examples of aligned slices from high resolution XCT data collected from the cylindrical samples machined from sample Gs1 prior to (left) and after the HIPing (right). Showing: (a) a gas pore within the cylinder machined from the centre; and (b) a lack of fusion defect from the cylinder machined from near the sample edge. 235

236 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY 6.3 Discussion The results in this chapter have shown that it is possible to reduce the level of porosity in SEBM components without altering the algorithms used to control the electron beam. Two important methods being pursued by industry have been identified and these are discussed in turn below Influence of Powder Feedstock on Porosity It has been shown that, in general, a lower fraction of porosity in the powder used to build an SEBM component results in a lower fraction of porosity in the consolidated solid material (Figure 6.2). This is unsurprising when considering the domination of gas pores, over other types of defect in the solid material. As has been discussed in the previous chapter, the main source of gas that produces these defects can only originate from argon trapped in the powder as components are built under vacuum conditions and the powder has very low levels of soluble gas contamination. The increase in pore volume fraction found in both the solid and powder in the current Arcam supplied PA powder, compared to the old Arcam supplied GA powder, makes the decision to change the feedstock somewhat surprising and was likely driven by economic cost rather than powder quality. However, the size distribution of the different powder types (Figure 4.24 and Figure 6.1) could have also affected the porosity by influencing the melting process and melt pool size. Karlsson et al. [29] have demonstrated that using smaller sized powder, but similar electron beam settings, can result in more porosity observed in Ti-6Al-4V samples manufactured by SEBM. They suggested this was caused by changes to the thermal properties, but had not examined the porosity in the virgin powder to confirm that both powder types had similar levels of porosity prior to melting. In addition, the different powders may have other properties that influence likelihood of defects appearing in the SEBM process, such as the rheology, which would alter the powder spreading behaviour. Therefore, there are many variables to be accounted for and caution should be applied before drawing conclusions between powder properties and porosity in the built samples. For all the powder types analysed, it appears that larger particles were more likely to contain porosity. When powder is atomised by flowing gas (GA, PA and PREP), small droplets will spheroidise and freeze in flight without trapping any gas bubbles break up rather than solidifying around a gas bubble [146]. Whereas, with larger particles it is possible for the high viscosity liquid metal to form a bag around the gas, and, upon solidification, entrap a gas bubble. The same argument cannot be applied to HDH powder, which is produced by mechanical milling [150], but a defect, such 236

237 6.3. DISCUSSION as a crack, would require a certain amount of material around it to prevent fracture of the powder particle. Defects may simply not fit inside the smaller particles. However, some increase in porosity with particle size may be due to larger particles making up a larger volume of material and thus having a higher probability of containing a pore. Since the ratio of particles that are porous has been normalised by the number of particles, not volume, in this case this would result in a greater ratio of porous particles. However, an investigation into the exact distribution of porosity in powder is beyond the scope of this project. HDH powder is generally considered a low quality powder unsuitable for critical applications due to chemical impurities introduced during the HDH process [20], and the low porosity observed in both the HDH powder and solid could therefore be considered surprising. However, this is consistent with the assumption that most porosity in Ti-6Al-4V components manufactured by SEBM is caused by inert argon gas bubbles in the powder. In addition, it suggests that the diffusion of contaminate soluble gases is unimportant for porosity formation during SEBM of Ti-6Al-4V. Nonetheless, the solid material produced by the HDH process may still be unsuitable for mechanically loaded applications due to high levels of chemical contamination. In contrast, PREP is an expensive powder production process that has been shown to produce powder with low levels of porosity, and a corresponding low volume fraction of porosity in the solid material Removal of Porosity by Hot Isostatic Pressing The results presented have thus shown HIPing to be very effective in closing porosity contained within Ti-6Al-4V components, manufactured by SEBM-AM. This in itself is unsurprising, given the success HIPing has enjoyed when applied to Ti-6Al-4V castings [41, 153]. However, large tunnel defects, which could appear as separated segments in the 3D visualisations of the XCT data due to the small size of the ligaments connecting them, were found to persist in some samples following the HIPing process. When this occurred, more careful manual analysis revealed that in all cases these flaws were connected to the top surface of the build and this would allow the infiltration of the pressurised argon gas into the tunnel cavity during HIPing, preventing it closing. The potential of these large, high aspect ratio defects to have a detrimental effect on the mechanical properties of SEBM samples is aggravated by the failure of HIPing to close those that were open to the surface. However, surface connected pores in traditionally manufactured components can be removed by the application of a coating prior to HIPing, which effectively makes the porosity internal [152]. Although in general the presence of tunnel defects has been reduced by the Arcam software updates, surface connected tunnel defects were detected in sample Gr (Figure 5.18) built 237

238 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY with current best practise onto the powder bed rather than the base plate. Thus, even with samples built with the current methodology there is still the possibility of tunnel defects remaining after a HIPing process. Furthermore, argon cannot diffuse readily through titanium. Indeed, this is the gas used to exert pressure during HIPing [152]. Hence, if the gas pores in SEBM components are caused by argon bubbles, they must also persist in the sample after HIPing, all be it at a significantly smaller size and a much higher internal pressure. Since no remnants of gas pores could be observed by high resolution XCT (Figure 6.6a), any bubbles remaining must be below approximately 2 µm in diameter. During HIPing, the internal gas pressure of the pores will oppose the shrinkage and will increase as the bubbles shrink [152]. However, the initial pressure in the gas pores could be quite low, due to the solidification of the melt pool taking place in a vacuum chamber, and is clearly insufficient to prevent the collapse of the pores. When HIPing gas porosity in castings, which is typically caused by hydrogen, it is assumed that the gas is soluble and can thus diffuse out of the casting [152], whereas here the argon is likely to remain in the material with a higher internal pressure. However, the initial pressure in the gas pores could be quite low, due to the solidification of the melt pool taking place in a vacuum chamber, and is clearly insufficient to prevent the collapse of the pores. When HIPing gas porosity in castings, which is typically caused by hydrogen, it is assumed that the gas is soluble and can thus diffuse out of the casting [152], whereas here the argon is likely to remain in the material with a higher internal pressure. Nonetheless, the high cycle fatigue life of Ti-6Al-4V samples manufactured by SEBM has been shown to be improved following a HIP cycle [26, 85]. High cycle fatigue is generally dominated by the number of cycles required to initiate a crack [94], so the removal of ready crack initiation sites is likely to be the primary reason behind the increase in fatigue life. The size of a spherical pore does not in fact change its stress concentration factor [110], but of importance in determining fatigue initiation is the reduction in the size of the plastic zone near a defect relative to the scale of microstructural barriers to early stage crack growth [94]. It has been shown here that any remaining argon filled pores would be less than 2 µm in diameter, and this would increase the stress concentration in only a very small volume of material above the elastic limit and such tiny pores are hence unlikely to initiate a crack able to break through microstructural barriers such as α colony boundaries [94]. The microstructural coarsening that occurs in high temperature HIPing cycles, and associated increase in crack propagation resistance, has also been suggested as a reason for the increase in fatigue strength seen in Ti-6Al-4V SEBM components following HIPing [85]. While this may increase the crack propagation resistance and reduce the smaller crack propagation contribution to high cycle may contribute to high cycle fatigue life, it seems more likely that the reduction in pore size caused by HIPing is the major reason for the 238

239 6.3. DISCUSSION improvement observed, given the reasoning and results outlined above. 239

240 CHAPTER 6. EXTERNAL CHANGES TO SEBM-AM METHODOLOGY 6.4 Summary The porosity seen in AM components has been shown to be reduced by making changes to two key factors outside of the SEBM machine itself. Namely, changing the powder feedstock or adding a post manufacture hot isostatic pressing (HIPing) step. By studying samples built with different powder feedstocks, an approximately linear trend was observed between the volume fraction of small pores detected in the powder and of larger pores found in the solidified material. Other factors, such as differences in the powder size distributions and rheology, may also be important, but this would need further investigation than was possible here. HIPing has been confirmed to be an effective method to remove even large scale internal porosity from worst-case AM components built under non-optimised conditions, but surface connected porosity was seen to be retained. Comparison of samples built with different geometries revealed that this was predominantly caused by large tunnel defects present in builds which had a top surface formed by the hatching strategy. Even at high scan resolutions, no evidence of internal porosity could be observed in the HIPed samples by XCT, therefore any residual voids can be assumed to have collapsed to below the resolution limit of the equipment, which was approximately 2 µm. It is proposed that this small residual pore size is responsible for the large increase in fatigue life seen in HIPed AM components. 240

241 7 Effect of Porosity on Mechanical Properties of Ti-6Al-4V SEBM Samples This final results chapter documents the experimental and modelling work carried out to quantify the effect of the porosity characterised in the previous chapters on the mechanical properties of Ti-6Al-4V SEBM samples. Presented first are the results from standard tensile and fatigue testing and fracture surface analysis of samples machined from bulk SEBM Ti-6Al-4V. Following this, a more detailed analysis of the effect of the porosity on fatigue crack initiation is provided, which the fracture surface analysis revealed was strongly influenced by the presence of porosity and particularly pores near the sample surface. By conducting XCT scans on fatigue samples prior to testing it was possible to identify pores that could lead to crack initiation. A methodology was then developed to rank the pores in the order of those most likely to initiate the crack. The ranking system was based on Finite Element (FE) modelling which was conducted to first quantify the stress concentration around idealised pore geometries. The stress around real pore geometries was then estimated using the trends observed for generic shapes. XCT models of the pores identified as being most dangerous were then exported for further FE analysis of the stress field around real pore geometries. Finally, the crack growth behaviour was characterised in 3D using XCT on interrupted fatigue tests. The influence of the local grain orientation on the crack profile was also analysed using EBSD maps of the orientations of grains along the crack profile. 241

242 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES 7.1 Mechanical Properties of the SEBM Samples Tensile testing on samples machined from SEBM deposited material was carried out parallel to the build direction at both ambient and elevated temperatures (see subsection 3.3.1). Fatigue testing was carried out externally by Exova (Lancaster, United Kingdom). Fatigue samples were tested in the horizontal and vertical (build) directions, but only the vertical samples were available for fracture surface analysis. The numerical results from these tests are presented here alongside fractures surface analysis Tensile Testing Engineering stress (σ e ) and strain (ε e ) curves for samples machined from the bulk hatching region are shown in Figure 7.1. The three samples tested at room temperature (20 C) show very similar stress strain curves and very closely aligned yield and ultimate stresses; likewise, for the two samples tested at each of the elevated temperatures investigated (150 C and 300 C). The significant variation in maximum strain recorded for the two samples tested at 300 C shown in Figure 7.1 is a result of the testing conditions rather than significant variation between the samples. In particular, due to the large sample gauge length (60 mm) in comparison to the 25 mm extensometer used to record the engineering strain. During the testing of one of the samples, marked (i) in Figure 7.1, the extensometer was positioned around where the neck formed in the sample, and thus recorded the high strain in the necking region. When testing the other sample (ii) the extensometer was positioned away from where the neck formed and a lower maximum 1,200 1,000 σ e (MPa) (ii) ε e (i) 20 C 150 C 300 C Figure 7.1: Engineering stress-strain data from tensile testing out carried using samples machined from the bulk hatching region at ambient and elevated temperatures. 242

243 7.1. MECHANICAL PROPERTIES OF THE SEBM SAMPLES 1, σe (MPa) 1, UTS σ 0.2y RA RA (%) Test temperature ( C) Figure 7.2: Variation of tensile properties with test temperature, where σ 0.2y, UTS and RA are the 0.2 % offset yield stress, ultimate tensile stress and reduction in area follwing fracture respectively. value of strain was recorded at fracture. The test marked (i) was the only test where the extensometer recorded the strain in the neck. The stress-strain curves shown in Figure 7.1 were used to calculate the 0.2 % offset yield stress (σ 0.2y ) and ultimate tensile stress (UTS), which have been plotted alongside the reduction in area (RA) measured post fracture in Figure 7.2. An increase in testing temperature is accompanied by a decrease in both σ 0.2y and UTS, with the σ 0.2y showing more sensitivity to the testing temperature than the UTS. Conversely, the RA increased with temperature, but the increase from 150 C to 300 C was much less significant than from 20 C to 150 C. Due to the small number of test samples the error bars in Figure 7.2 denote the maximum and minimum values recorded, while the solid lines pass through the mean value. The low level of scatter between the yield and ultimate tensile stresses recorded is demonstrated by the small range of the error bars visible in Figure 7.2. For all temperatures, both the yield and ultimate stresses exhibited a maximum variation from the mean value of less than 1 %. In contrast, the maximum variation in the RA was larger, being 12.6 %, 2.0 % and 0.9 % from the mean, for room temperature, 150 C and 300 C respectively. The consistency between separate tests at the same temperature is further illustrated in Figure 7.3, where the engineering stress-strain curves have been converted to true stress strain curves. The data has been cropped at the UTS because necking of the sample prevented accurate determination of the true stress at higher strains. From Figure 7.3 it can be seen that the average strain to UTS increases with temperature, whereas the strain to yield decreases. 243

244 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES 1,250 1,000 σ t (MPa) ε t 20 C 150 C 300 C Figure 7.3: True stress-strain data from tensile testing carried out at ambient and elevated temperatures. Data only shown prior to UTS. 244

245 7.1. MECHANICAL PROPERTIES OF THE SEBM SAMPLES Tensile Fracture Surface Analysis In Figure 7.4a c low magnification examples of images of the tensile sample fracture surfaces are shown for each temperature. Higher magnification images of the tensile fracture surface are given in Figure 7.4d f. The low magnification images show cup and cone fracture surfaces for all the testing temperatures. The higher magnification images show dimples on the fracture surface, which show some increase in size with higher testing temperatures. In Figure 7.4a, a pore can be observed at the fracture surface. However, the reduction in cross sectional area due to the pore would be negligible, <0.01 %, and would have only a very small influence on the stress distribution in the rest of the sample. (a) (b) (c) 2 mm (d) (e) (f) 20 µm Figure 7.4: Tensile fractures surfaces at two different magnifications. Low magnification images of fracture surface following tensile testing at (a) room temperature, (b) 150 C and (c) 300 C. Higher magnification images of the centre of the fracture surfaces at are shown in (d), (e) and (f) for room temperature, 150 C and 300 C respectively. 245

246 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES Fatigue S-N Data In Figure 7.5 the number of fatigue cycles to failure is plotted against the peak applied load for samples tested at room temperature in the z (build) and x directions. There is large scatter visible in the data, particularly in the results for samples tested in the vertical (build) direction. For example, one sample tested at a maximum stress of 600 MPa ran out to the end of the test (survived 100,000 cycles), whereas another fractured after only 33,771 cycles. In general, the horizontal samples showed a longer fatigue life than the vertical samples tested at the same stress. The two vertical samples that remained unbroken at the end of the test (after 100,000 cycles) were examined by XCT under loading with the rig described in subsection Analysis of the XCT data revealed that both samples contained cracks that had initiated from porosity. The sample loaded to 500 MPa (FV500) showed a crack that initiated from a pores at the surface, whereas the sample loaded to 600 MPa (FV600 3 ) showed a crack emanating from an internal pore. These samples were then subjected to interrupted fatigue testing and XCT scanning until failure. The observed crack growth is described in subsection Maximum stress (MPa) Vertical Horizontal Cycles to failure Figure 7.5: Fatigue cycles to failure against maximum stress. Results are shown for samples tested in the vertical and horizontal direction according to ASTM E Note that testing ceased after 100,000 cycles. 246

247 7.1. MECHANICAL PROPERTIES OF THE SEBM SAMPLES Fatigue Fracture Surface Analysis Examples of fatigue fracture surfaces are provided in Figure 7.7 and Figure 7.6 for two of the samples tested in the vertical direction. Sample FV575, shown in Figure 7.6, survived the greatest number of cycles until failure (73,068) excluding those that reached run out, having been tested with a maximum stress of 575 MPa. Figure 7.6a shows the entire fracture surface observed by SEM and also indicates the approximate locations where the higher resolution images shown in Figure 7.6b d were acquired. Figure 7.6b shows a magnified image of the crack initiation point where there is clear evidence of a pore at the sample surface. The crack which propagated from the pore appears to grow in a plane normal to the direction of loading with only small levels of deviation observable and striations visible at high magnification (Figure 7.6c). Once the crack reached a critical size, fast fracture occurred with the final fracture region showing dimpled overload fracture features (Figure 7.6d). Sample FV760, shown in Figure 7.7, survived the fewest cycles prior to fracture (5,540), having been tested at the highest maximum stress (760 MPa). In Figure 7.7a two separate cracks can be observed that contributed to the failure. Higher resolution images of the two crack initiation points are provided in Figure 7.7b e. The larger of the two cracks initiated at a smooth facet at the sample surface (Figure 7.7d). In contrast, the smaller crack initiated from an irregular pore near to the sample surface that appears to be two gas pores frozen during coalescence (Figure 7.7e). Further examples of crack initiation locations imaged by SEM are shown in Figure 7.8. Figure 7.8a c show pores at the crack initiation. The remaining image, Figure 7.8d, shows a facet feature which is distinct from the porosity visible at the other crack initiation sites. In Table 7.1 the features of the fatigue crack initiation points have been summarised for all the vertical fatigue samples. Specifically, the location of the crack initiation has been noted alongside a measurement of the maximum cross section pore area using image analysis and corresponding equivalent circular diameter (if applicable) the crack emanated from. It should be noted that there was significant uncertainty when measuring the area of each pore. The similarity of the image intensity between the pores and the surrounding fracture surface meant that an automatic threshold was unable to segment the pore. Instead, manual segmentation was required, meaning the results were, to some extent, dependent on the user. To reduce the uncertainty, each pore was measured on three separate occasions and the standard deviation is indicated in Table 7.1. For internal porosity the diameter of a circle of equivalent area can be used to estimate the size of the pore. However, in the case of open porosity, some of the as-solidified pore was machined off during manufacture of the fatigue sample, and thus the quoted diameter does not represent the full as-built pore size. The number of cycles to failure has been updated to include the extra cy- 247

248 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES Figure 7.6: Fatigue fracture surface of sample FV575 tested at 575 MPa showing crack initiation from a pore. (a) shows the whole surface while (b) (d) provide higher resolution images of the indicated areas. (b) illustrates the crack initiation from a pore. (c) shows striations in the crack growth region and (d) shows a magnified image of the fast fracture region. 248

249 7.1. MECHANICAL PROPERTIES OF THE SEBM SAMPLES Figure 7.7: Fatigue fracture surface of sample FV760 tested at 760 MPa showing crack initiation from both a pore and a featureless facet. (a) shows the whole surface while (b) (e) provided higher resolution images of the indicated areas. (d) illustrates the crack initiation from a smooth facet, whereas (e) shows a crack initiation point at two conjoined gas pores. 249

250 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES (a) (b) 200 µm 200 µm (c) (d) 200 µm 200 µm Figure 7.8: Examples of fatigue crack initiation in samples: (a) FV600 2 ; (b) FV550; (c) FV525; and (d) FV660. A pore can be seen at each of the crack initiation sites in (a) (c), whereas a facet at the surface is visible in (d). Table 7.1: Quantification of the fatigue fracture initiation sites, and testing conditions for samples tested in the vertical direction. The area of the pore at the crack initiation site, measured from SEM images, is given alongside the corresponding equivalent circular diameter rounded to the nearest µm. A pore size of N/A indicates that the crack initiation was not from a pore. Sample Max. stress Cycles to Pore area Pore dia. Initiation (MPa) failure (µm 2 ) (µm) location FV , ±0.4 Surface FV , ±1.6 Surface FV , ±2.2 Surface FV , ±2.2 Surface FV , ±1.0 Surface FV , ±1.0 Surface FV , ±2.0 Sub-surface FV , ±2.0 Surface FV , ±1.2 Surface FV ,063 N/A N/A Surface 2 FV , ±1.9 Surface FV , ±1.5 Surface FV ,540 N/A N/A Surface 2 250

251 7.1. MECHANICAL PROPERTIES OF THE SEBM SAMPLES cles (>100,000) required to fracture the two vertical samples that remained unbroken at the end of the standard test (FV500 and FV600 3 ). The application of extra cycles and the crack growth prior to their failure is described in subsection Analysis of the 13 fatigue fracture surfaces revealed that fatal fatigue cracks in 11 samples had initiated from pores and, of these, 10 had initiated from pores very close to the surface. Only two samples were found to have cracks initiating at features other than pores, but also near the sample surface. Furthermore, both these two samples also showed a secondary crack emanating from a pore near the surface, as shown in the example given in Figure 7.7. All but one of the fatal fatigue cracks initiated near the surface of the samples and propagated inwards. Moreover, the single sample FV600 3 that failed from a crack initiating at a sub-surface location was also one of the two samples that remained unbroken after 100,000 cycles, although it eventually failed after 101,868 cycles at a maximum stress of 600 MPa Curve Fitting to Observed Fatigue Life In an attempt to fit a trend line to the S-N data provided in Figure 7.5, Basquin s expression relating the stress amplitude to the number of stress reversals to failure to failure, Equation (7.1), was used. σ 2 = σ a = σ f (2 N f ) b (7.1) where: σ, σ a, σ f, N f and b are the true stress change, stress amplitude, fatigue strength coefficient, number of cycles to failure and fatigue strength exponent. Thus 2 N f is the number of load reversals to failure. Basquin noted that when the logarithm of the true stress amplitude was plotted against the logarithm of the number of load reversals to failure a linear relationship was often observed [94]. Fitting a linear trend line to the logarithms of the fatigue data allows the constants σ f and b in Equation (7.1), to be calculated. Figure 7.9 shows the logarithmic true stress amplitude plotted against the logarithmic number of stress reversals to failure for the vertical fatigue samples. Only samples found to have a pore at the initiation site of the fatal crack have been included in Figure 7.9; samples FV660 and FV760 with facets at the initiation location have been excluded. Despite the similar initiation sites, the large scatter in the data makes the linear trend line, fitted by the least squares method, a fairly poor model for the results. The coefficient of determination was calculated to be 0.53, which is fairly low and indicates that only around half of the variance on the data is explained by the model. Hence, Equation (7.1) in its conventional form is not useful in predicting the fatigue life. The number of fatigue cycles required to initiate a crack is a crucial factor in determining the number of cycles to failure [94]. Therefore, to try and better fit an equation 251

252 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES True stress amplitude (MPa) Stress reversals to failure Figure 7.9: Logarithmic plot of fatigue results for only those samples which failed from cracks initiating from pores. Linear trend line fitted by the least squares method. to the measured fatigue life, Basquin s equation was modified to include the effect of the porosity on the crack initiation. This was achieved by estimating the stress intensity factor generated by each of the pores at the crack initiation sites. Murakami [114, 115] showed that defects such as pores or inclusions are equivalent to cracks from the viewpoint of fatigue. The maximum stress intensity factor (K max ) generated by an internal defect is given by Equation (7.2). K max = 0.5 σ π A n (7.2) Where σ and A n are the global applied stress and the pore normal area respectively. When a defect is identified as being at a surface, Murakami [114, 115] has suggested the constant (0.5) in Equation (7.2) is changed to 0.65 to reflect the increased stress intensity factor of surface cracks. Thus, an approximation of the stress intensity factor range ( K I ) for the pore at each crack initiation location could be calculated based on the pore area measured from SEM images of the fracture surface (Table 7.1) and Equation (7.2). In this testing R=0, and hence K I = K max. It was assumed that cracks initiated at the widest part of the pore and accordingly the measured area was equal to the normal area of the pore. Basquin s equation was then modified by exchanging the true stress range ( σ) for the stress intensity factor range ( K I ). The resultant equation is shown in (7.3). K I = K I (2 N f ) b Where K I and b are new constants, analogous to σ f and b in Basquin s original equation. (7.3) 252

253 7.1. MECHANICAL PROPERTIES OF THE SEBM SAMPLES 300 KI (MPa mm) , ,000 1,000,000 Stress reversals to failure Figure 7.10: Logarithmic plot of fatigue cycles to failure against predicted stress intensity factor using Equation (7.2) at the pore that initiated the fatal crack. Linear trend line fitted by the least squares method. Figure 7.10 shows the logarithmic plot of the relationship between the calculated stress intensity factor range at the pore at the crack initiation site and the number of stress reversals to failure. While there is still deviation from the linear trend line fitted with the least squares method, the scatter is greatly reduced in comparison with Figure 7.9. Indeed, the coefficient of determination of the trend line was calculated to be 0.84, indicating a much better fit to the data. The gradient of the trend lines shown in Figures 7.9 and 7.10 were used to give b and b respectively, while the intercepts are equal to σ f and K I. In Figure 7.11a the fatigue results for the vertical fatigue samples that had a fatal fatigue crack initiating from a pore has been plotted alongside the trend line fitted with Basquin s original equation (7.1). For comparison, in Figure 7.11b the stress intensity factor range for the same samples has been plotted against the number of cycles to failure alongside the S-N curve predicted by the modified Basquin equation (7.3). It is clear from comparison of the data shown in Figure 7.11a & b that the modified Basquin equation (7.3) gives a much better fit to the data than the original equation. 253

254 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES (a) (b) Maximum true stress (MPa) Cycles to failure KI (MPa mm) Cycles to failure Figure 7.11: Fatigue data with Basquin s equation fitted: (a) true stress against cycles to failure; and (b) stress intensity factor generated by pore at crack initiation site against cycles to failure. 254

255 7.2. PREDICTION OF FATIGUE CRACK INITIATION 7.2 Prediction and Characterisation of Fatigue Crack Initiation It has been shown that the stress intensity factor at the crack initiation site strongly influenced the number of fatigue cycles to failure. From this result it was hypothesised that fatigue cracks initiated at the pore with the greatest stress intensity factor. Therefore it was attempted to predict locations where fatigue cracks are likely to initiate in SEBM samples by using X-ray computed tomography (XCT) data to identify the worst case pores prior to testing and to further reduce scatter in fatigue test data. This section details the methods used to develop simple search rules to select the pores most likely to lead to fatigue cracks directly from XCT data. This was achieved by using FE analysis of idealised geometries to find general rules that could be applied to all porosity. More accurate FE modelling of a short list of the worst case real pore geometries was then conducted in an attempt to further refine the prediction based on the local stress/strain conditions. The better understanding gained by this work of why certain pores are more likely to initiate fatigue cracks can be used to explain some of the results presented in the previous section. Interrupted fatigue testing was carried out using the same testing standards (ASTM E466-07) as for the conventional tests, but testing was paused following every 10,000 cycles and the samples were subjected to periodic XCT scans. By such a method the approximate number of cycles required to initiate a crack detectable by XCT was identified Stress Concentrations within Idealised Geometries It has been shown previously that approximately 97 % of defects within Ti-6Al-4V samples manufactured via SEBM are near spherical gas pores (chapter 4). Thus spheres and ellipsoids of revolution were used to quantify the approximate effect of the pores found in the fatigue samples on the local stress concentration. Idealised geometries were modelled using Abaqus standard, with 3D voids used to estimate the stress around pores in 3D, and an axis-symmetric model used to predict the variation in stress in a homogeneous fatigue sample free from defects. Examples of visualisations of the stress distributions within models used to predict the stress concentration in a pore near a surface are shown in Figure The maximum tensile stress (σ z ) calculated was then divided by the globally applied stress to calculate a single linear elastic stress concentration (K t = max(σ z )/σ ) for each model. The effect of proximity to a free surface on the tensile stress concentration (K t (sur f ace)) around voids with three different aspect ratios has been plotted in Figure To allow these results to be applied to any void size, the pore location has been made dimensionless by dividing the depth from the surface (d) by the pore diameter (D). Thus, numbers less than 1 indicate a pore that is open at the surface (such 255

256 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES (a) (b) σ 1 σz/σ 0.5 max. σ z max. σ z 0 Figure 7.12: Visualisation of tensile stress around two pores with different distances to a free surface. Only a quarter of the pores is required to accurately model the geometry due to the symmetry. These models correspond to: (a) d/d=0.6; and (b) d/d=1.2, for AR=1 in Figure Kt(sur f ace) AR = 2.0 AR = 1.0 AR = d/d Figure 7.13: Tensile stress concentration generated by a void at different depths from the surface. Pore depth (d) has been made dimensionless by dividing by the pore diameter (D). When d/d < 1, the pore is open to the surface and when d/d>1 the pore is completely enclosed. The trend with voids of three different aspect ratios (AR) is shown. 256

257 7.2. PREDICTION OF FATIGUE CRACK INITIATION Kt(proximity) s/d 1 D 1 = D 2 2 D 1 = D 2 Figure 7.14: Tensile stress concentration due to the interaction of two voids arranged normal to the loading direction. Results are shown for the interaction of two identical voids (D 1 = D 2 ) and where one void was twice as large as the other (D 1 = 2 D 2 ). as the void modelled in Figure 7.12a), whereas numbers greater than 1 signify that the pore is completely enclosed (such as in Figure 7.12b). All the models showed that the greatest stress concentration was at the surface of the pore, at the widest point relative to the loading direction. In addition, for any model where d/d > 0.4, the greatest stress was at the edge of the pore closest to the free surface, as shown in Figure From Figure 7.13 it can be noted that there is a considerable increase in the stress concentration generated when the pore is close to touching the surface, regardless of the aspect ratio of the void. Where d/d = 1, the pore would be just touching the surface and the infinitely small volume of the surface ligament would lead to an infinite stress concentration in a purely elastic condition. Curves for all pore aspect ratios therefore asymptotically approach infinity at d/d = 1. Of course, in experimental conditions, plastic deformation would prevent this. At the other extreme, when a pore is much further than 1 diameter from the free surface (i.e. d/d > 2) the effect on the stress concentration becomes negligible. In fact, the rise in stress concentration is only significant at approximately half a diameter from the surface (d/d < 1.5). Furthermore, when the majority of the pore is outside of the sample (d/d < 0.5), the stress concentration reduces below that experienced for a pore in a bulk material (d/d>2). Consequently, the near surface positions leading to significant increases in the stress concentration around a pore are fairly limited to a surface layer of approximately 0.5 pore diameters. The stress concentration generated by the interaction of two spherical pores arranged normal to the direction of loading (K t (proximity)) is shown in Figure Again, the position has been made dimensionless, this time by dividing the separation between the pores (s) by the one of the pore diameters (D 1 ). Results are shown for two 257

258 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES σ 1 σz/σ 0.5 max. σ z 0 Figure 7.15: Visualisation of tensile stress around two pores of different sizes. Only a quarter of the pores is required to accurately model the geometry due to the symmetry. This model corresponds to D 1 = 2 D 2 and s/d 1 = 0.5 in Figure identical pores interacting with each other (D 1 = D 2 ) and for one pore with twice the diameter of the other (2 D 1 = D 2 ). It can be seen that in this scenario the increase in stress concentration is only noticeable when the pores are separated by less than a diameter of the smallest pore, and only increases significantly when the distance is less than half a diameter. The results are slightly different when the pores have different sizes, but the overall trends remain similar. An example of the stress distribution around two pores of different diameters is shown in Figure The greatest tensile stress is always at the maximum width of the pore, relative to the loading direction, and closest to the other pore. When the two pores are of identical diameters, both pores experience the same stress concentration. However, when the pores have different diameters, the greatest tensile stress is found at the surface of the smaller pore, as shown in Figure Finally, the tensile stress (σ z ) distribution in a homogeneous fatigue test sample, without any internal defects, loaded to 600 MPa is shown in Figure 7.16a, while Figure 7.16b shows the variation of stress with position in the z-direction at specific radial locations. This figure illustrates how the stress distribution in a standard fatigue test samples is actually slightly non-uniform with a location of highest stress being predicted at the surface of the sample near the gauge end. The tensile stress at the centre of the sample gradually decreases moving towards the neck (Figure 7.16b), whereas at the surface of the sample the stress increases near the neck before reducing. Less than 4 mm from the centre of the gauge length, in z, all the radial positions show the same stress, but closer to the sample neck radial locations further from the centre show a higher stress. Additionally, the stress within the gauge length (613 MPa) was higher than the intended stress (600 MPa), due to the allowable ±0.05 mm tolerances in the gauge length diameter during machining. A sample at the maximum permissible size would have a stress in the gauge length equal to 600 MPa. Hence, the sample geometry 258

259 7.2. PREDICTION OF FATIGUE CRACK INITIATION (a) σz (MPa) max. σ z (b) 800 σz (MPa) Centre 1 mm from centre 4.5 mm from centre Sample surface z-position (mm) Figure 7.16: Tensile stress distribution within the fatigue test sample: (a) overview of the longitudinal stress distribution in cylindrical coordinates; and (b) variation in stress with distance at specific radial locations. (a) and (b) to same length scale. 259

260 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES alone results in a stress concentration of at the end of the gauge section compared to the average stress in the gauge length Initial Sorting Algorithm The XCT scans of the machined, but untested, fatigue samples revealed multiple pores within the gauge length of each specimen. A histogram of the pore sizes detected in the 4 samples is shown in Figure The 10.4 µm voxel size used for these initial scans would be able to detect pores down to a size limit of 26 µm and a total of 1298 internal pores were detected by standard automatic threshold image analysis in Avizo Fire. To try and predict the defect location where fatigue crack initiation would most probably occur, a number of methods could be used, the simplest being to rank the pores in terms of their size. However, a number of authors have shown that the area of a defect projected onto a plane perpendicular to the applied stress (A n ) is a more useful way of estimating its effects on the fatigue properties of samples than using the entire defect volume [112, 114, 115]. In addition, a number of other authors have shown the dependence of defect location on the likelihood of fatigue crack initiation [113, 10, 102, 105]. Finally, pores open to the surface are not detected by the standard segmentation tools available in Avizo, as they are regarded as being part of the surrounding air rather than an individual feature. Open porosity has been shown to be important in terms of both increased stress concentrations (Figure 7.13) and crack initiation (Figure 7.8), and thus must be included in any prediction of the crack initiation site. Number of pores detected Sample 1 Sample 2 Sample 3 Sample Equivalent diameter (µm) Figure 7.17: Number of pores of detected in each fatigue sample binned by their equivalent diameter. Markers are only shown when pores were detected in a size bin. 260

261 7.2. PREDICTION OF FATIGUE CRACK INITIATION Therefore, to attempt to capture all of these effects and, in particular, to include the effect of defect location in the predictions, an attempt was made to estimate a stress intensity factor for each pore. The pore with the highest stress intensity factor was then assumed to be the one from which fatigue cracks would first initiate. A first estimation of the stress intensity factor was made using the equations provided by Murakami [115] shown previously, Equation (7.2). These equations include a multiple that is varied based on the defect location. Thus, Equation (7.2) provides a very simple method of sorting the pores into those most likely to initiate a crack that includes the effect of surface pores. An attempt to further improve the capability of Equation (7.2) to predict the pores most likely to initiate a fatigue crack was made by replacing the variable based on the pore location and the global stress (0.5 σ) by the maximum local elastic stress (σ max ) around a pore giving Equation (7.4). K max σ max π A n (7.4) In this relationship σ max can then be estimated in the first instance by interpolation from the output generated by the systematic approach to FE modelling generic ellipsoidal pores of equivalent aspect ratio and proximity to a surface, or another pore, as exemplified in Figures 7.13, 7.14 and With 3D XCT data it is also possible to generate FE meshes of real defect geometries and thus more accurately predict the local stress and strain distributions. However, the number of pores identified in a typical sample means that it is not feasible to model each pore individually and the computation required to model the entire fatigue sample with a fine enough mesh size to maintain sufficient accuracy was prohibitive. Hence, to reduce calculation time, an initial sorting algorithm was written in MAT- LAB to pick out the top 2.5 % of pores deemed most likely to initiate a fatigue crack for FE analysis. The full script can be found in the Appendix, and a summary of the approach used is provided here to highlight the methodology used to sort the pores into those most likely to initiate a crack. All segmentation was first conducted in Avizo, as detailed in the experimental methodology section, and a stack of binary images was imported into MATLAB. An example of a slice of the 3D reconstructed data from sample IFH2 is shown in Figure 7.18a, alongside the same slice after segmentation by the Otsu method (Figure 7.18b). The first step of the script was to recreate the intended fatigue bar geometry and identify the open porosity that was missed by segmentation in Avizo. This was achieved by analysing each slice individually and fitting a convex hull to the data. An example of a slice and the corresponding convex hull is shown in Figure 7.18b and c, respectively. A convex hull is the smallest convex geometry that still contains all the solid mate- 261

262 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES (a) (b) (c) (d) Figure 7.18: Example of the method used to identify surface porosity from the XCT data: (a) slice of original data; (b) the segmented data; (c) a convex hull used to approximate the ideal geometry; and (d) by combining (b) and (c) it is possible to identify all the porosity including that which is open to the surface (highlighted). rial. In practice, this will draw a straight line across the open porosity (highlighted) as shown in Figure 7.18c. By masking the ideal shape with the original data it was then possible to identify all the porosity, including pores open to the surface which were missed by the standard techniques in Avizo. The pores identified in Figure 7.18d include the open pore (highlighted) that actually led to the fatal fatigue crack in sample IFH2. The linear segments used to construct the convex hull meant that the size of the open porosity was slightly underestimated, but this effect is very small. The idealised hull shape was also used to define a radius and centre of the solid material for each slice. The individual pores were then separated and their centroids calculated using inbuilt MATLAB functions. It was assumed that the fatigue samples were vertical during XCT scanning, and thus the area of the pore normal to the loading direction (A n ) was simply the area of the pore in the x-y plane. This normal area was also used to provide the equivalent diameter of a spherical void of the same normal area (D n ). The aspect ratio of all the internal pores was calculated using the maximum dimension in the x-y plane divided by the height in z. It was assumed all the pores were 262

263 7.2. PREDICTION OF FATIGUE CRACK INITIATION ellipsoids of revolution and this was used to assign a stress concentration due to the pore aspect ratio (K t (AR)), based on the FE modelling results in Figure The aspect ratio of pores open to the surface was calculated based on the aspect ratio of the 2D opening in the sample surface. The idealised hull geometry (Figure 7.18c) was also used to create a distance map for each slice, which denoted the distance of each internal voxel from the ideal shape. In this way, the depth of the open porosity, in addition to the distance of all the internal porosity from the surface, could be calculated. The maximum and minimum depth of each pore was then calculated by checking the distance map value at all the voxels around each pore perimeter. Those pores with a minimum depth of 1 voxel (i.e. touching the idealised geometry edge) were denoted as open porosity. Those pores that were open to the surface had their depths normalized with their height in the z- direction. The depths of all the internal pores were normalized by D n and those closer than 1.5 diameters from the surface, whose stress field would be affected by its proximity (Figure 7.13), were identified. Two subsets of pores, those open to the surface and those close enough to the surface to cause a greater local stress were thus identified (i.e. < 1.5 D n ). Linear interpolation was then used to fit the normalised depth to the stress concentrations predicted by FE modelling (K t (sur f ace) in Figure 7.13). Next, the distance between each pore centroid was calculated using an in-built MATLAB function and then normalized by D n. Those pores with less than 3.5 diameters between their centroids were identified. Analysis of all the perimeter voxels in the two pores allowed the length of material between their edges to be calculated. Again, linear interpolation was used to combine the theoretical FE analysis data regarding the effect of a pores proximity to another pore, with respect to the normalised distance between the pores, to approximate the increase in stress concentration (K t (proximity)) in Figure 7.14). The final step was to estimate the background stress at each pore location due to the sample geometry (σ z (position)). To do this it was first necessary to convert the pore locations to the cylindrical coordinate system shown in Figure This was achieved using the centre and radii of the convex hull of each slice. From the diagram of the sample (Figure 3.13) it is clear that in the 12 mm gauge length the radius was constant (2.25 mm), whereas at the neck the radius increases with a 9 mm curvature. The deviation from the constant radius was found at either end of the fatigue sample (R dev = R 2.25). This was used to construct a right angled triangle with a hypotenuse (c r ) of 9 mm and sides of a r and b r, where a r was the distance from the constant radius and b r = 9 R dev. The Pythagorean theorem was then used to calculate the distance from the constant radius using Equation (7.5). a r = c 2 r b 2 r = 9 2 (9 R dev ) 2 = R dev (18 R dev ) (7.5) 263

264 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES The stress at the pore location (σ z (position)) was then multiplied with any stress concentrations from the aspect ratio (K t (AR)), surface (K t (sur f ace)) or other pore proximity (K t (proximity)) to estimate the local stress around each pore. This was combined with the square root of D n to provide an estimation of the relative stress intensity factor (K I ) for each pore, as shown in Equation (7.6). K I σ z (position) K t (AR) K t (sur f ace) K t (proximity) D n (7.6) The result of Equation (7.6) should not be regarded as the absolute stress intensity factor; it is only a relative measure for each of pores. After sorting the pores using Equation (7.6) the top 2.5 % of pores with the highest relative stress intensity factors were identified. The binary slices defining each of the most detrimental pores and their surrounding environment were then exported for further analysis. These slices included any neighbouring pores and the sample surface, as originally defined by the XCT data. The exported volume was centred on the pore and expanded by 5 times the pore size in each direction; hence larger pores resulted in a larger exported volume. From the modelling results shown in Figures 7.13 and 7.14, it is clear that any other features, such as a free surface or other porosity, will have negligible influence if they are more than 5 diameters away from the pore of interest. Each pore location was also defined so that it could be located and checked against the fatigue cracks observed later, following testing FE Analysis of Pores most Likely to Initiate Cracks The image slices in each pore s local volume were then imported into ScanIP to generate an FE mesh based on the real pore geometry, which was then imported into Abaqus. The loading conditions applied to each model were defined by the stress due to the position of the pore within the fatigue sample. Purely elastic material properties were used to rank the pores using the maximum elastic principal stress predicted by FE (σ max:elastic ) and the normal area. This provided another approximate relative stress intensity factor, Equation (7.7). K I σ max:elastic D n (7.7) In an aluminium alloy fatigue samples, cracks have been shown to initiate at the location of the greatest stress/strain concentration when plastic deformation was accounted for [112]. Hence, the model was also rerun with elastic/perfectly-plastic material properties and then the maximum stress/strain concentration (k g ) calculated with 264

265 7.2. PREDICTION OF FATIGUE CRACK INITIATION Equation (7.8). k g = K σ K ε (7.8) Where K σ and K ε are the respective local plastic principal stress and strain concentrations calculated by FE analysis. Thus, a total of six ranking systems were tested to sort the porosity into those most likely to initiate fatigue cracks ranging in sophistication from: the greatest normal area, the greatest relative linear elastic stress intensity factor, calculated with Equations (7.2), (7.6), and (7.7) and the greatest stress/strain concentration under elastic-plastic conditions calculated by Equation (7.8) Comparison Between Predicted and Observed Crack Initiation Cracks were observed by XCT after 120,000, 100,000, and 70,000 cycles in samples IFH2, IFH3 and IFH4 respectively. No cracks were observed in sample IFH1 at 130,000 cycles, but it fractured prior to reaching 140,000 cycles when the next XCT examination was to take place. In addition, a secondary crack was observed initiating in a separate location in sample IFH3 after 102,000 cycles. Comparison of the number of cycles required before a crack could be identified, and the total number of cycles to failure, showed the great majority of the fatigue cycles were spent initiating a crack large enough to be detected by XCT. The smallest crack detected by XCT had a maximum diameter of approximately 0.9 mm in sample IFH4. Once a crack had been detected, the XCT data collected in the previous scan (less 10,000 cycles) was inspected in more detail with knowledge of where the crack should lie, but in all cases there was no visible evidence of the crack. All the fatigue cracks initiated from pores, and three of the cracks initiated from pores at the sample surface. The pore size was also shown to be important in sample IFH4, which cracked at a large internal pore, despite the presence of smaller porosity near the sample surface, whereas clustering of pores did not appear to be an important factor, with no cracks initiating from pores in close proximity. Quantification of the pores at the crack initiation site for each sample is shown in Table 7.2 alongside the relative ranking of each pore using the six systems outlined in the previous section, where 1 indicates the pore deemed most likely to initiate a crack. Table 7.2 also shows the quantification and ranking of the pore in sample IFH3 that lead to the secondary crack, although this has not been included in the calculation of the mean used to judge the effectiveness of the different ranking systems. From Table 7.2 it is clear that using the pore size (normal area) alone is not appropriate for predicting the most detrimental pores in terms of fatigue crack initiation. In all samples the largest pore did not lead to the fatal fatigue crack. A significant 265

266 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES Table 7.2: Quantification of the pores that led to the fatal fatigue crack in each sample studied by XCT. Pore size was measured from both the XCT data and SEM images of the fracture surface. The pore location is indicated alongside the maximum pore size detected by XCT and the number of cycles to failure. Following this is ranking of the pore that lead to the fatal fatigue crack using the systems outlined in subsections and Sample IFH4 remained unbroken at the end of the test but XCT revealed a large crack and failure was imminent. Also shown is the quantification and ranking of the pore that lead to a secondary crack in sample IFH3, although this was not included in the calculation of the mean. continued Sample Cycles to Total number D n (µm) failure of pores XCT SEM Location IFH1 139, Surf. IFH2 122, Surf. IFH3 102, Surf. IFH3 S Surf. IFH4 77, Sub-surf. Sample Ranking of pore at crack initiation using: D n Eq. (7.2) Eq. (7.6) Eq. (7.7) Eq. (7.8) IFH IFH IFH IFH3 S IFH Mean

267 7.2. PREDICTION OF FATIGUE CRACK INITIATION (a) (b) 5 mm 5 mm (c) 1,000 (d) σz (MPa) 0 σyz (MPa) Figure 7.19: Fatigue crack initiation location in sample IFH4. Slices of the XCT data in: (a) y-z plane and (b) x-z plane. Result of the FE modelling with half the model hiden to enable the: (c) tensile stress (σ z ) and (d) shear stress (σ yz ) at the internal pore surfaces to be visualised. improvement in the prediction is gained by taking account of whether the pore is at the surface, using Equation (7.2), provided by Murakami [114]. However, for pores close to, but not touching, the surface it can be unclear as to how to categorize them using this approach. Making the transition between surface pores and subsurface pores more gradual by using Equation (7.6) and interpolating the stress concentration variation from the simple FE models depicted in Figure 7.13, resulted in more accurate prediction of crack initiation. With this approach all the observed cracks initiated at one of the top 2.5 % ranked pores output by the MATLAB routine for further FE analysis. In addition, modelling the stress distribution around the real pore geometry (Equations (7.7) and (7.8)), did not result in any improvement in the prediction. Closer examination of the XCT data of crack initiation in sample IFH4 revealed that the crack (arrowed in Figure 7.19a) did not begin at the widest part of the pore. Instead the crack initiated at one end of the pore, where it had a significantly smaller cross section. Comparing this result to the FE model of the pore prior to testing, suggests that crack initiation occurred where the shear stress (Figure 7.19d) was greatest, not the tensile stress (Figure 7.19c). However, the resolution of the XCT data was not high enough to confirm whether this was the case in the other samples, where the pores at the crack initiation site were much smaller. 267

268 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES Effect of Local Stress/Strain Distribution on Total Fatigue Life By using Equations (7.2), (7.7), and (7.8) to predict the local maximum stress at the defect at the crack initiation site in each sample, an S-N curve has been plotted in Figure 7.20 depicting all the fatigue test data. The standard fatigue life data is also included, but as all the tests were conducted at the same applied level of stress this data is only useful for showing the level of scatter seen between samples. It can be noted that Equations (7.2), (7.7), and (7.8) show the same trend; in general a higher stress intensity factor or stress/strain concentration was associated with a reduced number of cycles to failure. However, there is a minor increase in the stress intensity factor calculated for sample IFH1 compared to sample IFH2, despite sample IFH1 actually surviving more cycles prior to failure. σ (MPa) / KI (MPa mm) σ Eq. (7.2) Eq. (7.6) Eq. (7.7) , , ,000 Cycles to failure 2 kg Figure 7.20: Results of fatigue testing. In addition to the standard approach of using the global sample stress, the number of cycles to failure has been plotted against the factors identified. 268

269 7.3. CHARACTERISATION OF FATIGUE CRACK GROWTH 7.3 Characterisation of Fatigue Crack Growth The interrupted fatigue testing combined with XCT scanning, used in the previous section to identify crack initiation, was also used to characterise crack growth in 3D. XCT scanning was carried out using a tensile rig developed during the project. Two samples from the interrupted fatigue testing (IFH3 and IFH4) and the two samples that remained unbroken at the end of the standard fatigue test (FV500 and FV600 3 ) were used to characterise the fatigue crack growth. To allow easier comparison between samples, all were loaded to a maximum stress of 600 MPa during both fatigue testing and XCT scanning. This meant the loading conditions were changed for sample FV500, which had experienced 100,000 cycles at 500 MPa prior to the interrupted fatigue testing. The influence of grain orientation on fatigue crack propagation was analysed by EBSD of sectioned failed samples D Characterisation of Crack Morphology An example of a growing fatigue crack observed by XCT is shown in Figure 7.21, with the segmented crack shown in red after various numbers of fatigue cycles. It is clear that as the number of fatigue cycles is increased, both the crack size and the crack growth rate increased. Also shown in Figure 7.21 is the residual porosity within the sample. However, from Figure 7.21 alone it is hard to differentiate the effect of the porosity on crack growth rate. To allow better visualisation of the crack front detected after different numbers of fatigue cycles, the cracks observed by XCT are shown projected on to the x-y plane in Figure The planes of the cracks observed in 3D were closely aligned with the x-y plane (<4 angle between the two planes) normal to the loading direction (see Figure 7.23). Thus, the difference in crack length projected onto the x-y plane, in Figure 7.22, and the 3D crack length will be less than 0.2 %. Figure 7.22a & b show (a) (b) (c) 5 mm Figure 7.21: Fatigue crack (red) and pores (blue) imaged by XCT in sample FV500, after: (a) 100,000 cycles; (b) 103,000 cycles; and (c) 104,000 cycles. 269

270 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES (a) (b) (c) (d) 1 mm Figure 7.22: Crack growth measured by XCT projected onto the x-y plane, for different numbers of fatigue cycles. Pores within ±0.5 mm of the crack plane are shown in red. The sample designation and number of fatigue cycles (in 1000 s) were: (a) IFH3 after 100, 101 & 102 cycles; (b) IFH4 after 70, 71, 72, 73, 74, 75, 76 & 77 cycles; (c) FV500 after 100, 103 & 104 cycles; and (d) FV600 3 after 100 & 101 cycles. 270

271 7.3. CHARACTERISATION OF FATIGUE CRACK GROWTH (a) (b) (c) (d) 1 mm Figure 7.23: Single slices of XCT data in a plane normal to the fatigue crack planes, showing the crack profile. The sample designation and number of cycles for each image are given by: (a) IFH3, 102,000; (b) IFH4, 77,000; (c) FV500, 104,000; and (d) FV600 3, 102,000. All samples were tested at a maximum stress of 600 MPa and all images are to the same scale. the growth of fatigue cracks in samples IFH3 and IFH4, respectively, which were tested in the horizontal direction, whereas Figure 7.22c & d show the crack growth in samples FV500 and FV600 3, which were tested in the vertical direction. In Figure 7.22a, b & c the crack front is shown as measured after steps of 1000 cycles, whereas in Figure 7.22c an initial step of 3000 cycles was used. Pores within ±0.5 mm of the crack plane are shown in red. The extent of the sample at the crack plane is shown in light grey. All the cracks in Figure 7.22 show an approximately elliptical morphology in the x-y plane. The cracks in the horizontal samples (Figure 7.22a & b) appear to show more deviation from a perfect ellipse than the cracks in the vertical samples (Figure 7.22c & d). No noticeable increase in crack growth in the x-y plane due to porosity can be observed; the effect of pores on the stress concentration around the crack is explored further in subsection In Figure D slices from the x-z plane of the XCT data used to generate Figure 7.22 are shown. The fatigue cracks in samples IFH3 and IFH4, tested in the horizontal direction, are shown in Figure 7.23a & b respectively, and the cracks in samples 271

272 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES FV500 and FV600 3, tested in the vertical direction, are shown in Figure 7.23c & d, respectively. It can be seen that there is some diversion from the direction normal to the loading direction. This is most significant near the crack initiation location. However, the amplitude of the crack diversion was small in comparison to the crack growth in the normal direction. In addition, the diversion of the fatigue crack was more significant close to the sample surface, when the crack initiated from a surface pore (Figure 7.23a & c) rather than an internal pore. No difference in crack profile could be observed between samples tested in the horizontal and vertical directions. Figure 7.23 also highlights the difficulty in identifying the full extent of the crack from the XCT data. While the crack can have significant size in the x and y directions, when closed it is essentially a 2D feature and therefore very difficult to identify with 3D voxels. This problem was alleviated to some extent by the application of a load to open the crack Measured Crack Growth Rates To allow comparison of the all the crack growth rates observed by XCT, the change in measured crack area per cycle (da/dn) has been plotted on a logarithmic scale against the stress intensity factor range ( K) in Figure In this case, crack growth was quantified by the change in the crack area divided by the number of cycles between XCT inspections. The stress intensity factor range was estimated using Equation (7.2), provided by Murakami [115], to calculate the maximum stress intensity factor K max experienced by a crack during the fatigue cycle immediately prior to the XCT examination. In the testing performed here R = 0, and therefore K = K max. The crack growth between each XCT scan has been plotted against the mean K calculated for the two scans. If the crack growth for the three lowest stress intensity factor ranges is neglected, the crack growth shows an approximately linear trend. The trend line shown in Figure 7.24 was fitted by the least squares methods, excluding the points indicated with unfilled markers. In general a higher stress intensity factor range is associated with an increased crack growth rate. However, the data shows some scatter around the trend line. There is no noticeable difference between the crack growth rates measured for samples tested in the horizontal direction and those tested in the vertical direction. In addition, the growth of surface (samples IFH3 and FV500) and subsurface (samples IFH4 and FV600 3 ) cracks fit the same trend line after using the appropriate equation to calculate the stress intensity factor range. 272

273 7.3. CHARACTERISATION OF FATIGUE CRACK GROWTH IFH3 IFH4 IFH4 - excluded FV500 FV600 Linear trend line da/dn (mm 2 ) K (MPa m) Figure 7.24: Crack growth rate measured by XCT against change in stress intensity factor for SEBM Ti-6Al-4V samples with a linear trend line fitted by the least squares method, but excluding three points from sample IFH4. 273

274 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES Effect of Porosity on Stress-Strain Concentration around Cracks To estimate the influence of pores near the crack tip on the local stress-strain concentration, whole fatigue cracks, as imaged by XCT, were converted into an FE mesh. In order to generate a model that could be run without requiring excessive computation power, it was necessary to down-sample the data by binning by 2 2 2, such that 8 voxels were combined to form a single larger voxel. This reduced the number of elements in a model of the crack generated by ScanIP from > 10 million to around 2 million. Figure 7.25 shows the result of FE analysis of the stress around the crack observed in sample FV600 3 after 100,000 cycles, using an elastic-perfectly plastic material model. In Figure 7.25a the entire model is shown illustrating the von Mises stress at the sample surface, while in Figure 7.25b the model has been sliced to allow the internal stresses to be observed. Further 2D slices of the 3D model in Figure 7.25c e and f h show the stress and equivalent plastic strain respectively around the crack. It is immediately apparent that the crack generates a large stress concentration, (a) (b) (c) (d) (e) (f) (g) (h) ,000 σ v (MPa) ε pe 10 2 Figure 7.25: Results of FE analysis of the stress and strain distribution around the fatigue crack observed in sample FV500 after 100,000 cycles. (a) visualisation of the von Mises stress (σ v ) in the whole model; (b) half of the model removed to allow visualisation of internal stress around the crack; (c), (d) & (e) slice of the mesh showing the von Mises stress around the crack with no pores in close proximity, a pore just ahead of the crack tip and a pore perpendicular to the crack tip respectively; (f), (g) & (h) same slices as (c e) but showing the equivalent plastic strain (ε pe ) around the crack tip. 274

275 7.3. CHARACTERISATION OF FATIGUE CRACK GROWTH (a) (b) 1 mm Figure 7.26: XCT data showing interaction of fatigue crack with residual porosity in sample FV500 after: (a) 103,000 cycles; and (b) 104,000 cycles. which is greatest at the sample surface. In comparison, the stress concentration experienced at the deepest point of the crack in somewhat less (Figure 7.25a & b). When residual porosity was close to the crack tip (Figure 7.25d & f) the change in the stress distribution around the crack was slight compared to when there were no pore present (Figure 7.25c). The plastic strain shows some increase near the pore in Figure 7.25g but again the influence is small. Some difference between stress and strain concentration calculated in the planes shown in Figure 7.25c h is due to the different depths of the planes from the sample surface, with Figure 7.25c being further from the surface than d and f and thus having a lower stress concentration. Following the application of another 3000 fatigue cycles no diversion of the crack towards the pores show in Figure 7.25 was observed. However another pore, not visible in Figure 7.25, was found to exist in a plane aligned with, and close to, the advanced crack tip. Figure 7.26a & b show x-y planes of the 3D voxel data of the crack after 103,000 and 103,000 cycles respectively. Only part of the crack is visible due to a slight misalignment between the x-y plane of voxels and the crack plane. However, no diversion of the crack towards the pore can be detected, with the observable crack front remaining smooth and elliptical both prior to reaching, and after passing through the pore. The segmented voxel data regarding the fatigue crack after 103,000 cycles (Figure 7.26a) was then converted to an FE mesh after being down-sampled by In addition, the pore shown in Figure 7.26 was artificially filled in the segmented data to generate a secondary model, identical in all respects to the original except that there was no pore near the crack. In Figure 7.27a & b the calculated von Mises stress and equivalent plastic strain for whole model are shown. A slice of the model is shown to visualise the stress and strain around the crack near the pore in Figure 7.27c & d 275

276 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES (a) (b) (c) (d) (e) (f) ,000 σ v (MPa) ε pe 10 2 Figure 7.27: Results of a finite element model of the stress and strain around the fatigue crack detected in sample FV500 after 103,000 cycles. The von Mises stress (σ v ) and equivalent plastic strain (ε pe ) in the entire model are show in (a) & (b) respectively. In (c) & (d) the von Mises stress and equivalent plastic strain around a slice of the model with a pore near a crack tip is shown. In (e) & (f) the same slice is shown, but with the pore artificially removed. respectively. In Figure 7.27e & f the same location is shown but with the pore removed from the model. Comparison of Figure 7.27c & e, Figure 7.27d & f, reveals that the pore had no noticeable impact on the stress and strain distribution at the crack tip. The maximum equivalent plastic strain at the crack tip was with the pore present and with the pore removed, a difference of less than 0.3 %. Due to the low number density of pores in the fatigue samples this was the only sample where the crack passed through a pore. In no specimens was diversion towards a pore observed Influence of Grain Orientation on Crack Growth Two fatigue samples were further analysed by polishing a plane perpendicular to the fracture surface and using EBSD to extract the grain orientations along the crack growth direction. Figure 7.28a shows the results of the EBSD analysis of the crack profile for sample FV760, shown in Figure 7.7, tested at the highest maximum stress 276

277 7.3. CHARACTERISATION OF FATIGUE CRACK GROWTH of 760 MPa. Also shown in Figure 7.28b is the initial β grain structure that formed on solidification, which was reconstructed using the algorithm developed by Davies [157] (see subsection 3.4.4). For comparison the room temperature α grains and reconstructed β grains along the crack of sample FV550 tested at 550 MPa are shown in Figure 7.28c & d respectively. All figures are presented in inverse pole figure (IPF) colouring approximately aligned with the build direction. Both fatigue cracks show greater diversion from normal to the loading direction closer to their initiation points (left). In Figure 7.28a, of the higher stress sample, this effect is more marked, with the initial 150 µm of crack growth occurring closer to 45 from the loading direction than normal. In contrast, the crack in Figure 7.28c (lower stress sample), shows less pronounced deviation from the normal direction, but the crack diversion is more significant further (>150 µm) from the crack initiation location. From the reconstructed β grains it can be seen that there is a correlation between crack diversion and the prior β grain boundaries. 277

278 CHAPTER 7. EFFECT OF POROSITY ON MECHANICAL PROPERTIES (a) (b) (c) Crack propagation (d) 200 µm Figure 7.28: Grain orientations measured by EBSD along sections through fatigue crack surfaces shown in IPF colouring for samples (a b) FV760 and (c d) FV550: (a) & (c) room temperature α grain structure quantified by EBSD; and (b) & (d) reconstructed β grain structure. In each case the crack propagation direction is indicated by an arrow, and the build and loading direction is vertical in the plane of the page. All images are to the same scale. 278

279 7.4. DISCUSSION 7.4 Discussion It has been shown in this chapter that pores can strongly influence the fatigue properties of Ti-6Al-4V samples manufactured by SEBM. In contrast, the effect on the static mechanical properties appeared to be fairly weak. The mechanical properties recorded in this chapter are compared to those available in the literature for Ti-6Al-4V components manufactured by more conventional routes. Following this, the influence of pores on the fatigue behaviour of SEBM Ti-6Al-4V samples and the use of XCT coupled with stress intensity modelling to predict fatigue crack initiation is discussed Comparison to Conventionally Manufactured Ti-6Al-4V Samples The relatively few fatigue tests reported in this thesis, combined with the statistical nature of fatigue [94], make it hard to draw definitive conclusions as to how the fatigue performance of SEBM Ti-6Al-4V compares to that manufactured by conventional processing routes. However, to get an overview of how the fatigue results presented in this chapter compare to what is available in the literature for Ti-6Al-4V, they have been overlaid onto the fatigue limits presented in Titanium: A Technical Guide by Donachie Jr. [20] in Figure From Figure 7.29 it appears that the fatigue lives of SEBM material is generally marginally better than cast parts and is comparable to cast parts following HIPing. As a near-net-shape process, SEBM may be most suitable as a re- 1,250 1,000 Maximum stress (MPa) Cast parts SEBM Wrought anneal Cast plus HIP Cycles to failure Figure 7.29: Comparison of smooth axial fatigue life in cast, cast plus HIP, wrought, and SEBM Ti-6Al-4V. Data for cast, cast plus HIP and wrought anneal taken from Donachie Jr. [20]. 279