LIFETIME PREDICTION MODELING OF AIRFOILS FOR ADVANCED POWER GENERATION. Ventzislav Gueorguiev Karaivanov

Transcription

1 LIFETIME PREDICTION MODELING OF AIRFOILS FOR ADVANCED POWER GENERATION by Ventzislav Gueorguiev Karaivanov Mehanial Engineer, Tehnial University, Sofia, Bulgaria, 1995 Submitted to the Graduate Faulty of Swanson Shool of Engineering in partial fulfillment of the requirements for the degree of Dotor of Philosophy University of Pittsburgh 2009

2 UNIVERSITY OF PITTSBURGH SWANSON SCHOOL OF ENGINEERING This dissertation was presented by Ventzislav Gueorguiev Karaivanov It was defended on June 16 th, 2009 and approved by Dr. Minking K. Chyu, Professor, Department of Mehanial Engineering and Materials Siene Dr. Brian Gleeson, Professor, Department of Mehanial Engineering and Materials Siene Dr. Patrik Smolinski, Professor, Department of Mehanial Engineering and Materials Siene Dissertation Diretor: Dr. William S. Slaughter, Professor, Department of Mehanial Engineering and Materials Siene ii

3 Copyright by Ventzislav G. Karaivanov 2009 iii

4 LIFETIME PREDICTION MODELING OF AIRFOILS FOR ADVANCED POWER GENERATION Ventzislav Karaivanov, PhD University of Pittsburgh, 2009 The use of gases produed from oal as a turbine fuel offers an attrative means for effiiently generating eletri power from our Nation's most abundant fossil fuel resoure. The oxy-fuel and hydrogen-fired turbine onepts promise inreased effiieny and low emissions on the expense of inreased turbine inlet temperature (TIT) and different working fluid. Developing the turbine tehnology and materials is ritial to the reation of these near-zero emission power generation tehnologies. A omputational methodology, based on three-dimensional finite element analysis (FEA) and damage mehanis is presented for prediting the evolution of reep and fatigue in airfoils. We took a first look at airfoil thermal distributions in these advaned turbine systems based on CFD analysis. The damage mehanis-based reep and fatigue models were implemented as user modified routine in ommerial pakage ANSYS. This routine was used to visualize the reep and fatigue damage evolution over airfoils for hydrogen-fired and oxy-fuel turbines onepts, and regions most suseptible to failure were indentified. Model allows for interation between reep and fatigue damage thus damage due to fatigue and reep proesses ating separately in one yle will affet both the fatigue and reep damage rates in the next yle. Simulation results were presented for various thermal ondutivity of the top oat. Surfae maps were reated on the airfoil showing the development of the TGO sale and the Al depletion of the bond oat. iv

5 In onjuntion with model development, laboratory-sale experimental validation was exeuted to evaluate the influene of operational ompressive stress levels on the performane of the TBC system. TBC oated single rystal oupons were exposed isothermally in air at 900, 1000, 1100 o C with and without ompressive load. Exposed samples were ross-setioned and evaluated with sanning eletron mirosope (SEM). Performane data was olleted based on image analysis. Energy-dispersive x-ray (EDX) was employed to study the elemental distribution in TBC system after exposure. Nanoindentation was used to study the mehanial properties (Young s modulus and hardness) of the omponents in the TBC system and their evolution with temperature and time. v

6 TABLE OF CONTENTS PREFACE...xiii 1.0 INTRODUCTION NICKEL BASED SUPERALLOYS SUPERALLOY MICROSTRUCRURE CREEP IN NICKEL SUPERALLOYS Type I slip Type II slip FATIGUE IN SINGLE CRISTAL SUPERALLOYS DAMAGE MECHANICS Stress Rupture Modeling Creep Modeling Fatigue Modeling Damage Equivalent Stress FINITE ELEMENT IMPLEMENTATION Finite element model, boundary onditions and loading steps User reep-fatigue routine Finite Element Modeling Results Hydrogen-fired turbine vi

7 3.0 AIR PLASMA SPRAYED YSZ TOP COAT APS YTTRIA STABILIZED ZIRCONIA Phase hange Miro-struture hanges Thermal ondutivity MODELING CHALLENGES OXY-FUEL TURBINE MODELLING THERMAL BARRIER COATINGS THERMALLY GROWN OXIDE KINETICS OF THE THERMALLY GROWN OXIDE External TGO growth and modeling Internal TGO growth and modeling β-phase depletion and modeling MATERIAL TESTING HYPOTHESIS CONCEPT SEM AND EDX EXAMINATION INDENTATION TESTING RESULTS AND DISCUSSION Exposure to 900 o C Exposure to 1000 o C Exposure to 1100 o C vii

8 6.0 CONCLUSION MODELING Thermal solutions Strutural, reep and fatigue modeling MATERIAL TESTING SEM and EDX observations Nanoindentation Material Testing APPENDIX A APPENDIX B BIBLIOGRAPHY viii

9 LIST OF TABLES Table 1. Chemial omposition of superalloys... 6 Table 2. Projeted Gas omposition (vol%) Table 3. TBC properties, Hydrogen Turbine Table 4. TBC properties, Oxy-fuel turbine Table 5. Test matrix Table 6. Clearanes and stress level for different testing temperatures Table 7. Summary of bond oat depletion behavior in all 1100 o C, 600 h. exposed samples Table 8. Mean vertial thikness of the oxide sale and bond oat at 1100 o C Table 9. Measured sample and oatings dimensions in all 1100 o C, 600 h. exposed samples Table 10. Chemial omposition sample #21, 1100 o C, 600 hours Table 11. Material properties evaluated by nanoindentation, all samples Table 12. Nominal dimensions of TBC system omponents and substrate, all samples ix

10 LIST OF FIGURES Figure 1. Crystal struture of and '... 7 Figure 2. TEM of CMSX-4 / ' mirostruture... 7 Figure 3. Experimental Ultimate Tensile Stress Figure 4. Experimental reep urves from CMSX-4 at 950 o C Figure 5. Experimental reep urves from CMSX-4 at 750 o C Figure 6. Damage parameter versus time Figure 7. Creep strain versus time Figure 8. Temperature distribution for Hydrogen turbine, h = 1000W/m 2 K Figure 9. Temperature distribution for Hydrogen turbine, h = 3000W/m 2 K Figure 10. Stress distribution for Hydrogen turbine in MPa, for h = 3000W/m 2 K Figure 11. Creep damage evolution for Hydrogen turbine at hours Figure 12. Fatigue damage evolution for Hydrogen turbine at hours Figure 13. Creep-fatigue interation Figure 14. Y O ZrO phase diagram Figure 15. X-ray diffration pattern of YSZ Figure 16. APS YSZ shemati mirostruture Figure 17. YSZ thermal ondutivity versus temperature x

11 Figure 18. Temperature distribution, K, Hydrogen fired turbine, k=1.5, 2.1, 2.4 W/mK Figure 19. Temperature distribution, K, Hydrogen fired turbine, k=0.9, 2.4 W/mK Figure 20. Substrate temperature distribution for Oxy-fuel turbine, h = 3000W/m 2 K Figure 21. YSZ surfae temperature distribution for Oxy-fuel turbine Figure 22. Minimum ooling effetiveness for Oxy-fuel turbine Figure 23. Substrate temperature distribution for Oxy-fuel turbine with external film ooling.. 48 Figure 24. Creep damage evolution for oxy-fuel turbine with external film ooling at 800 h Figure 25. Fatigue damage for Oxy-fuel turbine, R= Figure 26. TBC system on Rene N5 superalloy Figure 27. SEM of Compat TGO sale in 300μm APS YSZ system Figure 28. Temperature dependant growth rate onstant for ompat oxide sale Figure 29. Compat TGO thikness for Hydrogen turbine, 16000h: Figure 30. Inward TGO growth thikness for Hydrogen turbine, 16000h Figure 31. SEM of β phase depletion in 300μm APS YSZ system Figure 32. Normalized β phase depletion for Hydrogen turbine, 16000h, Figure 33. Test fixture (Hexoloy SA SiC by Saint-Gobain Ceramis) Figure 34. Test sample (ReneN5, MCrAlY, APS YSZ by NETL) Figure 35. Third prinipal stress, MPa, Test fixture with 1273K Figure 36. SEM Image of ross-setioned sample prior to thermal exposure Figure 37. Typial load-displaement urves Figure 38. Indentation loations shown on SEM image Figure 39. Indentation pattern Figure 40. SEM images of ross setion samples exposed to 900 o C for 100, 300 and 1000 h xi

12 Figure 41. Spetrum mapping of bond oat / oxide sale, 900 o C, 1000 hours Figure 42. Compat TGO thikness at 900 o C Figure 43. Young s modulus, GPa, at 900 o C, all samples Figure 44. Hardness, GPa, at 900 o C, all samples Figure 45. SEM images of ross setion samples exposed to 1000 o C for 100 and 300 h Figure 46. Young s modulus, GPa, at 1000 o C, all exposed samples Figure 47. Hardness, GPa, at 1000 o C, all exposed samples Figure 48. SEM images of ross setion samples exposed to 1100 o C for 100, 300 and 600 h Figure 49. Compat TGO thikness at 1100 o C Figure 50. Coalesent raks within the oxide sale, 600 hours, 1100 o C Figure 51. SEM image of the mapped area and spetrum maps - fully depleted bond oat Figure 52. SEM image of the mapped area and spetrum maps - not fully depleted bond oat.. 89 Figure 53. Oxide map of the ternary system Ni-Cr-Al at 1000 o C Figure 54. Young's modulus, GPa, at 1100 o C, all samples Figure 55. Hardness, GPa, at 1100 o C, all samples xii

13 PREFACE This tehnial effort was performed in support of the National Energy Tehnology Laboratory s on-going researh in aero-thermal-material integrated thermal protetion under the RDS ontrat DE-AC26-04NT The author would like to speifially aknowledge Mary Anne Alvin (NETL) for her ontinued support and help. I would like to express my deepest gratitude to the Department of Mehanial Engineering and Materials siene for the support and the opportunity to be part of and ontribute to this projet. More speifially to my advisor - Dr. William S. Slaughter for his invaluable guidane and providing an exellent atmosphere for researh, and to the ommittee members - Dr. Minking K. Chyu, Dr. Brian Gleeson, and Dr. Patrik Smolinski for their ontribution and guidane throughout. I would like to aknowledge the work of Dr. Minking K. Chyu and his students - Danny Mazzotta, Sean Siw and Pavin Ganmol, on aero-thermal modeling of the hydrogen-fired and oxy-fuel turbine onepts without whih our researh ould not have started. I would like to thank to my friend Dr. Serge Sidorov and Dr. Ilya Avdeev (Ansys In.) for their guidane and help in beoming an Ansys user. The author aknowledge the failities, sientifi and tehnial assistane of the Materials Miro-Charaterization Laboratory of the Department of Mehanial Engineering and Materials xiii

14 Siene, Swanson Shool of Engineering, University of Pittsburgh, speifially to the Diretor of MMCL Dr. Jorg M.K. Wiezorek and Mr. Albert Stewart. Speial thanks to Dr. Meltem Yanar for her support throughout the experimental part of our projet. My deepest gratitude to all faulty and students at the Department of Mehanial Engineering and Materials Siene. Last but not least to my wife Mina Pantheva M.D. and our families who made all this possible. xiv

15 1.0 INTRODUCTION The use of gases produed from oal as a turbine fuel offers an attrative means for effiiently generating eletri power from our Nation's most abundant fossil fuel resoure. The Department of Energy is developing key tehnologies that will enable advaned turbines to operate leanly and effiiently when fueled with oal derived synthesis gas and hydrogen fuels [1]. Developing the turbine tehnology is ritial to the reation of these near-zero emission power generation tehnologies. The onept of ultra low or near-zero emission power generation has foused on an oxyfuel ombustion proess that burns pulverized oal in a mix of oxygen and re-irulates waste gases to reate a high onentration of arbon dioxide (CO 2 ) in the gases exiting the boiler. The exhaust stream of the oxy-fuel proess is separated into onentrated CO 2 and water. The power generation turbine inlet temperature (TIT) is expeted to reah 1760 o C [1]. Alternatively, hydrogen-fired turbine systems are oneptually based on a omplex yle omposed of a losed Brayton yle and a Rankine yle. Hydrogen and oxygen are supplied as the fuel and oxidant, respetively, to a ompressor, and burned in steam. The pressurized steam feed enters the turbine at about 1400 C [2]. Creep, fatigue, oxidation and hot orrosion of industrial land-based power generation gas turbines are the prinipal onerns, partiularly in view of long-term operation (30,000 hrs) and operation with advaned ombustion yles that potentially target turbine inlet temperatures of more than 1400C [3]. Single rystal Ni-based superalloys have been used for manufaturing turbine blades beause they an withstand high temperatures and resist reep [4]. However, 1

16 when operating in advaned ombustion yles, exposed to turbine inlet temperature of more than 1400 o C, they will need appropriate thermal protetion. Current tehnology employs different tehniques simultaneously external, internal ooling and thermal barrier oatings (TBCs) [5, 6]. Turbine blades are ooled by air extrated from the turbine ompressor and direted through ooling passages designed into the omponent. The TBC system's role is thermal and, to some extent, oxidation protetion. This system usually onsists of an intermetalli bond oat, a thermally grown oxide layer and a erami top oat. It is designed to have very low thermal ondutivity as well as thermal expansion very lose to that of the underlying superalloy substrate. Every omponent is subjeted to diverse potential failure proesses, whih require good understanding in order to properly model the system behavior under omplex thermo-mehanial loading [5, 6]. Modeling of those proesses is needed to guide the development and appliation of stable TBCs with better insulation properties, as well as implementation of aeptable internal omponent ooling tehniques, in order to ahieve extended operational life of oxy-fuel and hydrogen fired land-based turbines. Previous models [7] for thermal-mehanial life predition have typially assumed that the loading state of the turbine airfoil along its staking line does not deviate signifiantly from that at its midsetion. Therefore, in order to lower omputational ost, a two-dimensional thermo-mehanial analysis is performed on a representative ross setion. Inreasing the work fluid temperature and diversifying the hemial ontent hanges the aerodynami and thermal loading. This demands better understanding of the development and evolution of ritial zones over the entire blade struture. 2

17 There are several models apable of good life predition of the individual elements of the system but hallenges are arising when ombining the failure modes and effets of one onstituent (i.e. TBC spalling) over the rest of the system. A good life predition methodology ould be an essential tool for improving design and material development. Durability designs and omponent life assessment for advaned turbine airfoils subjeted to high rotational speeds and gas temperatures require in-depth understanding of thermalmehanial oupling on the fatigue and reep damage mehanisms in the superalloy substrate. Therefore, a modeling effort should start with a thorough study of the material struture, harateristis, reep, and fatigue apabilities. We used a damage mehanis approah for the failure models development. Starting with one-dimensional models for reep and fatigue, fitted well to experimental data, we extended them for three-dimensional use and implemented them as a user-modified routine for the ommerial software pakage ANSYS. Thus, the models an readily be used for other similar appliations. The models developed required information for the three-dimensional temperature, stress and strain fields. Thus, as a first step, the field alulations were performed, followed by determination of the loal reep and fatigue damage in time and their interation over the entire airfoil domain. Aomplishments in this area are disussed in Chapters 2 and 3. Miro-strutural and phase hanges in the erami top oat part of the TBC system, and their influene on all other damage mehanisms are disussed in Chapter 3. In general, a TBC system redues thermal load to the airfoil and oxidation and orrosion rate into the airfoil substrate [5, 6]. Typial thermal barrier oatings are made of a top layer of yttria stabilized zironia (YSZ) and inter-metalli plasma sprayed bond oat (BC). During high temperature exposure the bond oat forms an Al-rih thermally grown oxide (TGO) sale and it 3

18 has a ritial role in the failure of the TBC system [5, 6, 8, 9]. The goal is to ahieve longer replaement intervals through a stable system and ontinuous protetion. Maps are reated on the surfae of the airfoil showing the development of the TGO sale and the Al depletion of the bond oat. The modeling approah is based on the kinetis of the protetive sale and uses performane data from the literature [10] and also from our own experimental work. In onjuntion with the model development, laboratory-sale experimental validation was exeuted to evaluate the influene of operational ompressive stress levels on the performane of the TBC system. TBC oated single rystal oupons were exposed isothermally in air at 900, 1000, 1100 o C with and without ompressive load. Exposed samples were ross-setioned and evaluated using a sanning eletron mirosope (SEM). Performane data were olleted based on image analysis. Eletron dispersive spetrometry (EDS) was employed to study the elemental distribution in the TBC system after exposure. Nanoindentation was used to study the mehanial properties (Young s modulus and hardness) of the omponents in the TBC system and their evolution with temperature and time. 4

19 2.0 NICKEL BASED SUPERALLOYS Single rystal nikel based superalloys are ommonly employed as a high temperature reep and oxidation resistant blade alloys at the first stages of modern gas turbines. A superalloy is a metalli alloy whih an be used at high temperatures, often in exess of 70% of their absolute melting temperature. Creep and oxidation resistane are usually the prime design riteria. 2.1 SUPERALLOY MICROSTRUCRURE Superalloys an be based on iron, obalt or nikel, the latter being best suited for aeroengine and advaned land based turbine appliations. The essential solutes in nikel based superalloys are aluminum and/or titanium, with a total onentration whih is typially less than 10 atomi perent. This generates a two-phase equilibrium mirostruture, onsisting of gamma (γ) and gamma-prime (γ'). It is the γ' whih is largely responsible for the elevated-temperature strength of the material and its inredible resistane to reep deformation. The γ phase forms the matrix in whih the γ' preipitates. The amount of γ' depends on the hemial omposition and temperature [4, 11]. The single-rystal superalloys are often lassified into first, seond and third generation alloys. The seond and third generations ontain about 3 wt% and 6wt% of rhenium respetively. Rhenium is a very expensive addition but leads to an improvement in the reep 5

20 strength. Typial hemial omposition of several first, seond and third generation proprietary alloys are listed in table 1: Table 1. Chemial omposition of superalloys [4, 11]. Alloy Cr Co Mo W Ta Nb Al Ti Re Hf CMSX CMSX CMSX CSMX Rene N Rene N RR SRR Of partiular modeling interest here are the CMSX-4 and Rene N5 alloys. Both of them are seond generation (3% Re) and have omparable harateristis. The former has been extensively studied and experimental results were published in number of journal artiles. The later is used by National Energy Tehnology Laboratory (NETL) as a substrate material. For a given hemial omposition, the fration of γ' dereases as the temperature is inreased. This phenomenon is used in order to dissolve the γ' at a suffiiently high temperature (a solution treatment) followed by ageing at a lower temperature in order to generate a uniform and fine dispersion of strengthening preipitates [11]. The γ-phase is a solid solution with a ubi-f (fae entered) lattie and a random distribution of the different speies of atoms. By ontrast, γ' has a ubi-p (primitive ubi) lattie in whih the nikel atoms are at the fae-enters and the aluminum or titanium atoms at the ube orners. This atomi arrangement has the hemial formula Ni 3 Al, Ni 3 Ti or Ni 3 (Al,Ti). In addition to aluminium and titanium, niobium, hafnium and tantalum an partition preferentially into γ'. The γ phase has a random distribution of Ni, Al and Ti atoms, whereas the γ' has its nikel atoms loated at the fae-enters, as illustrated on Figure 1 a) and b). 6

21 Figure 1. Crystal struture of and a) b) '. Lattie vetors along <110> diretions, lying on the {111} planes [11]. The γ phase forms the matrix in whih the γ' preipitates. Sine both the phases have a ubi lattie with similar lattie parameters, the γ' preipitates in a ube-ube orientation relationship with the γ. This means that its ell edges are exatly parallel to orresponding edges of the γ phase. Furthermore, beause their lattie parameters are similar, the γ' is oherent with the γ when the preipitate size is small. Disloations in the γ nevertheless find it diffiult to penetrate γ', partly beause the γ' is an atomially ordered phase. The order interferes with disloation motion and hene strengthens the alloy. Figure 2. TEM of CMSX-4 / ' mirostruture, [12]. In all nikel based superalloys, the presene of the ' phase hinders the motion of disloations even though it is oherent with the matrix. This is beause it has an ordered rystal struture. In 7

22 a the Burgers vetor of a disloation is 110 2, whih is, a lattie vetor (fig. 1a) so that slip a does not alter the rystal struture. However, for ' is not a lattie vetor, atually a 110 (fig. 1b) is the lattie vetor for a primitive ubi lattie. It follows that the motion of an a disloation into the ' will disrupt the order, leaving behind an anti-phase domain boundary. However, the passage of a seond suh disloation through ' on the same slip plane restores the order [4, 11]. As a onsequene, the penetration of ' has to our by pairs of disloations, or `super disloations'. The requirement for pairing makes it more diffiult for disloations to penetrate the ' and hene improves the resistane to reep deformation. The small misfit between the and ' latties is important for two reasons. Firstly, when ombined with the ube-ube orientation relationship, it ensures a low / ' interfaial energy. The ordinary mehanism of preipitate oarsening is driven entirely by the minimization of total interfaial energy. A oherent or semi-oherent interfae therefore makes the mirostruture stable, a property whih is useful for elevated temperature appliations. The magnitude and sign of the misfit also influenes the development of mirostruture under the influene of a stress at elevated temperatures. The misfit is said to be positive when the ' has a larger lattie parameter than. The misfit an be ontrolled by altering the hemial omposition, partiularly the aluminum to titanium ratio. A negative misfit stimulates the formation of rafts of ', essentially layers of the phase in a diretion normal to the applied stress. This an help redue the reep rate if the mehanism involves the limb of disloations aross the preipitate rafts. 8

23 The strength of most metals dereases as the temperature is inreased, simply beause assistane from thermal ativation makes it easier for disloations to surmount obstales. However, nikel based superalloys ontaining ', whih essentially is an inter-metalli ompound based on the formula Ni 3 (Al,Ti), are partiularly resistant to temperature [11]. Ordinary slip in both and ' ours on the {111}<110>. If slip was onfined to these planes at all temperatures then the strength would derease as the temperature is raised. However, there is a tendeny for disloations in ' to ross-slip on to the {100} planes where they have a lower anti-phase domain boundary energy. This is beause the energy dereases with temperature. Situations arise where the extended disloation is then partly on the losepaked plane and partly on the ube plane. Suh a disloation beomes loked, leading to an inrease in strength. The strength, however, dereases beyond about 800 o C where the thermal ativation is suffiiently high to allow the disloations to overome the obstales. In summary, the presene of ' is responsible for the fat that the strength of nikel based superalloys is relatively insensitive to temperature. Figure 3 shows a plot of the ultimate tensile strength (UTS) versus temperature for four superalloys inluding the seond generation CMSX-4 [12]. Figure 3. Experimental Ultimate Tensile Stress (UTS) values for four superalloys inluding single rystal CMSX-4. [12]. 9

24 2.2 CREEP IN NICKEL SUPERALLOYS The reep deformation behavior of single rystal superalloys is strongly temperature, stress, orientation and alloy dependant. They also exhibit the usual primary, seondary and tertiary reep behavior. There are two mehanisms of slip deformation dominating reep and plastiity of these superalloys. They our under different operating onditions and are usually referred as Type I and Type II slip. The next two setions look into more details of both mehanisms and why they are important from modeling point of view Type I slip This deformation mehanism ours on <101>{111} slip system and is found under ondition set of temperatures over 800 o C, lower stresses, and orientations lose to <001> rystallographi axis [13]. It is responsible for the seondary or steady-state reep behavior. It results in stable homogeneous deformation, where all matrix hannels are populated with disloations, forming a three dimensional network whih is then maintained throughout the steady-state reep. Preipitates are partiularly resistant to Type I deformation exept at very high stresses where a passage of disloations results in anti phase boundary as explained in setion 2.1. Figure 4a shows the effet of the magnitude of the applied stress on the reep strain evolution at 950 o C for samples within 10 o from <001>. There is strong dependene upon the stress level, however, there is a weak dependene upon disorientation from <001> axis as seen in figure 4b). In this regime reep strain rate inreases monotonially with reep strain in a linear fashion. 10

25 a) b) Figure 4. a) Experimental reep urves from CMSX-4 tested at 950 o C and various stress levels for orientation within 10 o from <001>. b) at 185 MPa and varying angle away from<001>. [13] Type II slip This deformation mehanism ours on <112>{111} slip system and is found under ondition set of lower temperatures espeially around 750 o C and high stress. Type II slip has the ability to shear both the matrix ( ) and the preipitate ( ' ) phases, and it is responsible for the primary reep behavior. It is also favored by orientations more that 25 degrees away from <001> and for stress states with shear omponents. Figure 5 a) illustrates the effet of magnitude of the applied stress on reep strain evolution at 750 o C. The reep strain is relatively insensitive to the magnitude of the stress one the stress level exeeds some border value between 600 and 750MPa. Figure 5 b) illustrates the effet of disorientation from <001>. Primary reep varies signifiantly for the same applied stress and different angle of orientation. Here K, L, M, and N samples have orientation respetively 5.27, 7.98, 17.47, and degrees away from <001> [13]. 11

26 a) b) Figure 5. Experimental reep urves from CMSX-4 tested at 750 o C: a) at various stress levels for orientation within 10 o from <001>. b) at 750 MPa and varying angle away from<001> [13]. Airfoils in advaned land based Hydrogen and Oxy-fuel turbine appliations are projeted to experiene metal surfae temperatures of o C (as shown in setion and 3.2) on the outside metal interfae and o C at the internal ooling hannels ore. Beause of this wide range of temperatures within different parts of the omponent it is expeted that both deformation mehanism desribed above will be ative. The very high reep resistane of the modern single rystal superalloy for temperatures from 700 to 800 o C is enabling the onept of internally ooled airfoils. Stress analysis of turbine blades shows that beause of signifiant thermal stresses and stress redistribution from high temperature areas an inrease the stresses experiened from the internal hannels to MPa and may be possible to exeed the border value and ativate Type II deformation. MaLahlan et al. [12, 14-15] found that rupture life of CMSX-4 redues from 5000h for 600MPa to 400h for 750 MPa at temperature of 750 o C and attributed this hange to orientation and the onset of deleterious type II slip mehanism. 12

27 2.3 FATIGUE IN SINGLE CRISTAL SUPERALLOYS In general Fatigue damage is divided in two types: High-yle fatigue (HF) is a miro-strutural damage mehanism that results from small stress amplitude yli loading, suh as vibrations. Failure is expeted after a relatively large number of yles yles is the dividing number between both types fatigue. Low-yle fatigue (LF) is a miro-strutural damage mehanism that results form large stress amplitude yli loading. Failure is expeted after a relatively short number of yles. Very abrupt thermal hanges, suh as start and stop yles are the usual driving fore for this failure mode. Land-based turbine engines have very long running times in between very few start stop yles. It is expeted that low-yle fatigue may not be a life limiting fator for most of these appliations. Sine the amplitude of the yli loading has a major effet on the fatigue performane, the Stress-Number of yles (S-N) relationship is usually determined for one speifi loading amplitude. The amplitude is express as the R ratio value, whih is the minimum peak stress divided by the maximum peak stress or R min, (1) max where R=0 for on-off stress appliation, R=-1 for fully reversible loading and R=1 for no hange in the applied stress. MaLahlan et al. [16] performed series of interrupted and uninterrupted fatigue tests on CMSX-4. Here are some of the harateristi findings for HF and LF at different temperatures. Failure during HF yling at 750 and 850 o C is entirely rystallographi. Initiation usually ours at a asting pore at intersetion of three rystallographi planes. For HF, from this point on, 13

28 failure is distinguished by whether the rystallographi surfaes are formed by yli rak growth or by atastrophi shear failure. At these lower temperatures, intense shear band are found on rystallographi planes and it is evident that rak growth ours along these bands on the weakened planes. The rak usually forms vary late in life and it takes only one or two thousand yles to propagate to suffiient length for failure. For HF yling at 950 o C failure is different from the lower temperatures. Initiation ours at a asting pore in the enter of a irular flat region. After a brief growth on rystallographi planes it adopts a planar growth (perpendiular to primary stress axis). When the rak reahes ertain length is shears on different rystallographi planes during final shear and there are no indiations for yli growth [16]. Low-yle fatigue test usually show similar failure mehanisms at all temperatures. The rak initiates from pore or oxide spike, planar rak propagation follows, and final shear ours on rystallographi planes. The differene in failure mehanism with inreasing temperature are restrited to the method of initiation, where at 750 o C rak initiates internally, and at higher temperatures rak usually begins at oxide spike on the surfae. Internal rak would grow radially, whether surfae one would join few other smaller raks and shear together. 14

29 2.4 DAMAGE MECHANICS Conepts from ontinuum damage mehanis will be used in the development of the life predition methodology. A brief overview of some of the relevant previous work involving damage mehanis follows. Damage mehanis theory was developed over the last few deades by Kahanov, Lemaitre, and Chabohe [17-21], and has been applied to a wide variety of problems, as well as reep and stress-rupture of a range of materials. For phenomenologial desriptions of mirostrutural hanges it is usual to introdue some internal variables. A damage parameter, ω, is a state variable and is a measure of hange in the material from its fully heat-treated ondition, whih affets its ability to resist monotoni or yli deformation. It has a value of zero for undamaged material and a value of one at failure. However, failure usually ours at values less than one and most models orret for this by use of an additional parameter. Beause there is no reovery, ω is a monotonially inreasing quantity. Turbine blades are omplex layered strutures, operating in high temperature and stress fields, whih are subjet to different damage mehanisms. In ases where several mehanisms of damage our simultaneously, eah of them may be represented by a state variable ω i. The hoie of damage parameter is made by either a physial mirostrutural analysis or by generalization of experimental data. Kahanov onsidered the damage variable as a surfae density of intersetions and avities, and in the simplest ase it is a salar funtion [19]. In general, different mehanisms of damage are disriminated both by the funtion used to desribe how they evolve with time/yles, and by the funtion used to desribe how they affet material properties. 15

30 In order to desribe the influene of damage on reep deformation, Kahanov and Rabotnov introdued a damage parameter in the reep equation, wherein the damage rate is a funtion of the damage state and operating onditions [19], dr g, T, r. (2) dt Rabotnov introdued the onept of effetive (atual) stress ating on a representative volume element as the nominal stress modified by the damage, eff 1r. (3) The stress-strain behavior of the material an be represented by the onstitutive equation of the as-manufatured material with the stress replaed by the effetive stress. It is also assumed that the rate of damage growth is ontrolled primarily by the level of atual stress [19]. If it is also assumed that the temperature loally is onstant, a popular hoie for the funtion desribing damage evolution is [12, 14, 15, 19] dr dt v C, (4) 1r where C and v are material parameters. This equation an easily be integrated analytially for r from 0 to 1, and for t between 0 and time to rupture T r. The substrate, or bulk material of a turbine blade for advaned turbines is usually a single-rystal superalloy, whih in their seond (René N5, CMSX-4) and third generation have exellent high-temperatures reep and fatigue resistane. MaLahlan, et al. [12, 14, 15], have applied ontinuum damage mehanis in life assessment models for stress rupture, reep and fatigue of first, seond and third generation superalloys. Their formulation is based on modified Kahanov-Rabotnov equations and shows a very good ability to desribe the evolution of 16

31 damage and strains as a funtion of time, temperature, and applied stress in uniaxial reep tests. Their version of equation (4) orrets for the fat that failure would our as stress goes to infinity rather than ultimate tensile stress, and provides a orrelation of short-term stress-rupture and tensile data and longer term reep data. MaLahlan et al. [12, 14, 15], have developed a damage mehanis based model for analyzing uniaxial behavior of superalloy single rystals, implemented on a slip system level and using single rystal plastiity theory. In order to aount for the effets of rotation and bending, they modeled the entire reep speimen using finite element (FE) software and employing a user material subroutine. In this model they based damage aumulation on strain and strain rate [15]. Chabohe and Gallerneau applied damage mehanis models to evaluate durability at elevated temperatures. The ontinuum damage model is stress-based and assumes that the progressive deterioration proesses an be desribed by salar damage variables. Their formulation inludes the interation effet between reep, fatigue and oxidation for oated and unoated superalloys through four separate damage variables [17] Stress Rupture Modeling The reep damage model uses onepts from damage mehanis as explained above. The damage parameter is a state variable, r, and it is a measure of hange in the material from its fully heat-treated ondition whih affets its ability to resist monotoni or yli deformation. In general, r varies from 0 to 1. However failure would atually our before r reahes one. 17

32 As explained above, MaLahlan et al. [12, 14, 15] modified equation (4) as: dr dt v eff C UTS eff. (5) Thus, the ultimate tensile strength (UTS) is used as an upper bound on rupture strength and the rate of damage aumulation. It is also onsistent with the onept of effetive stress and the priniple of strain equivalene that the material should fail when the effetive stress or the initial stress ating upon undamaged material reahes the UTS [12, 14, 15]. At reep failure the damage parameterr r f, thus eff UTS, r f 1 r f UTS, (6) UTS provides a orrelation of short-term stress-rupture and tensile data and longer-term reep data. By integrating equation (5) for 0r r f and 0t T r, we have an expression for the total time to rupture in terms of applied stress and ultimate tensile strength: v1 UTS T r C v1 UTS v. (7) The material parameter C is found empirially to follow a Boltzmann funtion of temperature: 1 C A exp H RT B H RT 1 2 exp. (8) It reflets the role of reep softening with temperature of the γ matrix on damaging the overall struture. The first term in the denominator of (8) aounts for this effet. Rafting, whih is a diffusion ontrolled proess, has the opposite effet on damage or redued flow ourring in the vertial matrix hannels, and thus an attenuation parameter is used. The seond term in (8) desribes the effet of rafting and the rate of attenuation again assumes a Boltzman temperature 18

33 funtion [12, 14]. R is the universal gas onstant, T is the absolute temperature, A and B are material parameters. H is similar to the ativation energy for self diffusion of nikel, H 1 2 is the ativation energy for attenuation of matrix flow. In general, these two energies are different as the diffusion proesses involved in rafting are unlikely to be the same as those in reep. For seond generation superalloys (CMSX-4) they have the same value of 278 (kj/mol), whih is similar for self diffusion of nikel. The onstants A and B are found to be 3.1x10-10 and 2.63x10 12, respetively. The material parameter v is independent of temperature and is 3.2 for CMSX-4. All of the above values are found to give good results for both CMSX-4 and third generation CMSX-10 for a temperature range of 1023 to 1423 K [12,14, 15]. The damage profile for onstant initial stress an be obtained by integrating equation (5) up to time t and damage to derive an expression for the damage state: r 1 v 1 t r f 1 1 T r, (9) where r f is given by (6). As previously indiated, the damage parameter r will vary from 0, at t 0, to r f at t =T r, where r f assumes values between 0 and 1, but depends on the initial stress state. Simplifying, we define damage D suh that r D, (10) r f 19

34 And now D assumes value D 0 at t 0, and D 1 at r r f or t. Figure 6 shows damage parameter evolution for several loading onditions at onstant stress and temperature. This damage parameter will be used in presenting reep damage for Hydrogen and Oxy-fuel turbine appliations. T r Figure 6. Damage parameter versus time at onstant stress and temperature Creep Modeling If it is assumed that the parameter ontrolling the evolution of damage during steady state and tertiary reep is the same as that ontrolling rupture life, the damage rate equation an also be oupled with a strain rate of similar form, d r dt eff E 0 UTS eff u, (11) E 0 and u are another two material parameters in addition to C and v. C and v are usually determined first by the stress-rupture behavior, then E 0 and u are obtained from the shape and position of the reep urve [12,14, 15]. This equation is apable of modeling the seondary and 20

35 tertiary reep behavior, whih is made possible beause of the use of effetive stress (3). Although UTS is temperature dependant, in order to model reep strain over wide range of temperatures, E 0 has to take several different values. To overome this issue E 0 an take a form of ativation funtion exp H E 1 0 C V 2 RT. (12) Equation (12) annot model primary reep, and having in mind the omplex behavior shown in figures 4 and 5, it will obviously be very diffiult to find a single equation that an represent all three reep regimes over the entire range of temperatures and stresses. On first approximation Cunha et al. [22] used modified time hardening onstitutive equation in a similar form to the one Chen et al. [23] used to model reep behavior: m n H2 m' H 1 r pr ss C1 t exp C2 texp V V RT RT, (13) where C 1, C 2, m, n, m are material parameters, and VH are ativation energies, and is VH 1 2 the applied stress. This equation has two parts: the first one models primary reep as non-linear funtion of time and the seond one models steady-state reep as a linear funtion in time. In general ativation energies are not the same in order to have ontributions from both parts pr and over different temperatures. In order to desribe the steady-state over wider range ss of stresses we propose to modify ss by hanging the stress dependene as follows: u H C2 texp V UTS RT 1 ss. (14) This model has the advantages of simpliity and onveniene for omputation. Considering that large-sale plastiity or tertiary reep will not appear during servie periods of turbine blades, the 21

36 first two reep stages are usually involved. However, if it is desired to model tertiary reep the applied stress an be substituted with the effetive stress eff (3) to give: m n H 2 eff H 1 r pr ss ter C1eff t exp V C2 texp V RT UTS R eff T. (15) Effetive stress is used not only in the seondary but in the primary formulation beause there are loadings ombinations (low temperature / high stress, or Type II slip) where reep behavior hanges from primary diretly to tertiary, without any steady state region. u Figure 7. Creep strain versus time, modeled by equation (15), at different onstant stresses and temperatures Fatigue Modeling There are several different ways to approah fatigue modeling. For example, the ommerial pakage ANSYS [24] has a fatigue module that an perform two type of analysis based on different loading type, mean stress effets and orretion, only at predetermined points of interest. Claudio et al. [25] used finite element methods to obtain the J integral and stress intensity fator solutions for a dis with two types of raks. Chen et al. [23] used FEA 22

37 simulation results to determine assessment positions and predited the life of turbine blades using five different models for reep-fatigue interation. Chabohe and Gallerneau [17] showed use of damage mehanis in reep-fatigue interation. As oppose to reep, where during tertiary reep multiple rak initiation our and the reep failure is a result of linking of these raks, at LF and HF yle fatigue failure generally results from a single dominant rak. Despite the fat that the final failure is aused by a single marosopi rak, fatigue damage aumulation in single rystals is ontrolled by initiation and an be regarded as a ontinuous proess. The simplest form of linear damage aumulation is Miner s law, assuming that fatigue damage evolves linearly as a funtion of the life fration, or: N i 1, (16) i N fi where N i is the number of yles at ertain stress level and N f is the number of yles to failure at that same stress level. In rate form this equation beomes: d fat 1, (17) dn N where fat is the fatigue damage state variable and it still evolves linearly. Where it is found that damage does not aumulate in liner fashion, as the ase it is here, damage mehanis offers different formulation, expressing the yli life as a power funtion of the stress [18-21]: f d dn fat b A 1 fat v, (18) where A, b, and v are material parameters. In order to aount for the stress level and allow for the effet of the mean stress to be inluded last equation an be written as [18-21]: 23

38 b fat A max, (19) d dn v max, 1 fat where is the mean stress and the damage variable is raised to a power whih is funtion of the stress level. MaLahlan et al. evaluated different tehniques to inlude the effet of the load ratio and found that using the Walker fator (w) is the best fit for relate the ontributions of the stress range and the maximum stress to fatigue life. The Walker fator was first proposed to relate stress ration to rak propagation but is also used in fatigue lifing: max 1 1 R w w w max, (20) d dn fat w v max 1 R A. (21) 1 fat In single rystal super alloy, as strain rate is inreased in tensile testing, the value of the maximum tensile strength inreases as well as the temperature at whih it ours. For LF, with typial low strain rates, yield strength inreases slightly up to 700 o C and then dereases. At HF, with typial high strain rates, both maximum yield stress and peak temperature are inreased. To quantify this effet MaLahlan et al. [16] normalized the yli stress by the yield stress (0.1%) as a measure of the resistane of the material to miro-plastiity and rak initiation. And in order to quantify the effet of strain rate on yield strength they fit the yield strength as a funtion of strain rate and temperature as: y B Q & Q &, T Aexp 1 exp. (22) RT & 0 RT,0.1% 1 2 Here AB&,, 0 are material parameters, Q1, Q2 are ativation energies, R andt are the gas onstant and temperature respetively. The modified fatigue damage rate equation beomes [16]: 24

39 d dn fat v w max 1 R A &, T 1 fat y,0,1%. (23) For ases when reep and fatigue proesses our simultaneously (ex. LF), there is a need to ombine the effets of both on the overall damage of the material. The simplest way is to use linear aumulation of fatigue and reep through the use of ombination of Miner's law for fatigue and Robinson's rule for reep. The ombination of both leads to the linear life fration rule: 1 1 1, (24) N N N f r where N f,and N are the number of pure fatigue yles and number of yle in pure reep r respetively. Although been very popular, this equation does not predit any signifiant reepfatigue interation. However, if interation does our, eq. (24) an be modified as ombination of the rates for fatigue and reep damage: d f, r r r fat d f, fat fat r fat,, d d d f f r fat r fat (25) where and are reep and fatigue aumulated damage. This way damage due to fatigue r fat and reep proesses ating separately in one yle affets both the fatigue and reep damage rates in the next yle and a synergism is reated. The amount of reep-fatigue interation depends on the individual aumulation laws hosen for both proesses. MaLahlan et al. [16] have found that the amount of life redution predited by eq. (25) is not signifiant ompared to the one from eq. (24), whih ould mean that there is no substantial interation, and failure ours either by reep of fatigue only. 25

40 2.4.4 Damage Equivalent Stress Our onept is to apply this model at every material point (node) and thus obtain damage evolution over the entire airfoil domain. In order to be able to do that, we need to relate the three-dimensional state of stress experiened by every node to the uniaxial stress used to desribe the above relations. Starting from the strain energy density release rate, Lemaitre and Desmorat [20, 21] defined a triaxiality funtion in order to find a damage equivalent stress. This damage equivalent stress is defined as the one-dimensional stress σ* whih, for the same value of the damage yields the same value of elasti strain energy density as that of a three-dimensional state. For the ase of linear isotropi elastiity they defined: Where H is hydrostati stress and eq * eq R v, (26) 2 2 R H v, (27) 3 eq is von Mises equivalent stress. However, sine our material has ubi symmetry, we need to define a different funtion in order to relate both stress states. Starting again from the strain energy density release rate for anisotropi material, we then define strain energy density release rate for a material with ubi symmetry and for the degenerate ase of uniaxial loading. Energy density release rate: (J.L. Chabohe 1976): Y * * e e, (28) D 1 D * where e is the elasti strain energy density, and for the ase of anisotropi elastiity without any damage: 26

41 * 1 1 D D 1 2 e Sijkl ij kl Sijklij kl SijklH ij kl, (29) H 1 kk, (30) 3. (31) D ij ij H ij * Next step is to define the damage equivalent stress, or the uniaxial stress whih would give the same amount of elasti strain energy density as multiaxial state of stress following the approah of J. Lemaitre (isotropi): From equation (28) and using the results: (A 1), (A 2): Y * 1 Sijkl 1 S e D D ijkl 2 ij kl Hijkl (1 D) 2 (1 D) 2 (1 D) D D D D D D D D D D D D S 2 S 4 1 2(1 D) kk 3S1111 6S1122 2(1 D) D D D kk D D D D D D kk S S (1 D) D D D S2323 (32) S For the ase of uniaxial stress, using the result of equation (A 8 ): Y 1 S 2(1 D) 1111 * 2 (33) * After equating (32) and (33) and solving for equivalent stress : * D D D D D D D D D 1122 kk kk S D 2 D 2 D S S S (34) 27

42 where D kk ij ij 3 ij is the deviatori stress, and S 1111, S 1122, and S 2323 are the three independent onstants of the ompliane tensor haraterizing ubi material. Elasti onstants are also temperature dependant. This would be the stress value used in the alulations for reep strain and damage evolution. 2.5 FINITE ELEMENT IMPLEMENTATION Implementation of the models desribed above is performed by using ommerial software pakage ANSYS v.11 through the means of ustomizable user routine. Suh approah allows us to use ANSYS s geometry integration, meshing algorithms, standard elements, robust solvers, visualizing apabilities, and post proessing environment along side ustomized reep-fatigue routine that makes use of state variables, suh as the reep and fatigue damage parameters. The next three subhapters are devoted to FE modeling, reation of the user modified reep-fatigue routine and results obtained by using this approah Finite element model, boundary onditions and loading steps In order to redue omputational time, previous efforts for thermal-mehanial life predition have typially assumed that the loading state of the turbine blade along its staking line does not deviate signifiantly from that at its mid-setion. Therefore, two-dimensional analysis was performed on this representative ross-setion [7, 22]. Inreasing the working fluid temperature and altering its hemial omposition hanges the aerodynami and thermal loading, thus 28

43 demanding better understanding of the development and evolution of ritial zones over the entire blade struture in order to evaluate the apability of proposed designs. Thus, threedimensional analyses are required. For simulation purposes, a solid model of the NASA E 3 blade was reated using Pro/Engineer. Based on available data for the root, mean, and tip ross-setions and the orresponding diameters and staking line a blend protrusion was reated that formed the airfoil body [26]. With the geometry data available, the airfoil was reated without its base and tip setions and with a serpentine geometry for internal ooling hannels using the originally speified wall thikness. The serpentine ooling geometry stands for hannels running alongside from the bottom to the top of the blade, and the ooling air an irulate the hannels in many different ways depending on overall design. In general, this implies that the ooling air temperature and heat transfer oeffiient are not onstant through the length of the hannel. The methodology ould, however, be applied to other designs. Table 2. Projeted Gas omposition (vol%), Turbine Inlet Temperature (K) and pressure (MPa) for Hydrogen and Oxy-fuel turbines [1]. Hydrogen-fired Turbine Oxy-fuel Turbine Turbine Inlet Temperature, o C (K) 1425 (1698) 1760 (2033) Turbine Inlet Pressure, MPa Combustor H 2 O (17.3) Exhaust CO 2 (1.4) Composition, (%) O2 (8.2) N2 (72.2) Ar (0.9) H 2 O (75-90) CO 2 (25-10) O 2 N 2 Ar (1.7) To begin the thermal-mehanial analysis of the solid substrate, the level of thermal load or heat transfer oeffiient and pressure distributions on the airfoil s external surfae were previously 29

44 determined via three-dimensional CFD simulation modeling using ommerial pakage FLUENT [2, 7, 27]. Boundary onditions were made available for both Hydrogen-fired and Oxy-fuel turbines with projeted gas ompositions, turbine inlet temperature (TIT), and turbine inlet pressure as shown in Table 2. Steady state thermo-mehanial analysis was performed using ANSYS. The airfoil geometry was imported into ANSYS and the entire volume meshed with tetrahedral elements. Two types of solid elements were used - strutural Solid 187 and its thermal ounterpart Solid 87. In order to apply the boundary onditions obtained from the CFD simulation (surfae heat transfer oeffiient and pressure), the entire surfae was meshed with surfae effets elements, namely Surf 154 and Surf 152. Approximately solid tetrahedral elements were used to mesh the blade volume and ~ D surfae effets elements were used to mesh all surfae areas. In order to simulate the effet of the oating mass on the substrate under entrifugal loading from rotation the outside surfae elements have mass assigned to them. As the airfoil was modeled without the base and the tip setion, in order to apply the mehanial onstraints properly an extension was modeled suh that the airfoil extended from the root ross-setion. The length and form of this extension was reated suh that the effet of the boundary would be negligible on the blade volume (root to tip). No layers have been reated on the geometri model in order to model the presene and effet of the TBC. Rather, an overall effetive heat transfer oeffiient, related to the thermal resistane of the TBC, is imposed as a boundary ondition on the external airfoil surfae. One the surfae heat transfer oeffiient on the airfoil is obtained from the FLUENT flow simulation, the effetive heat transfer oeffiient ( h ) an be evaluated as follows: 30

45 1 1 l l a b l, (35) h h ka k k b where h is the surfae loal heat transfer oeffiient, and l and k are the thikness and thermal ondutivity of eah TBC layer, respetively. The properties of the yttria-stabilized zironia (YSZ) based thermal barrier oating used in this effort are given in Table 3. This effetive heat transfer oeffiient and the loal distribution of pressure determined from the CFD analysis were inluded as boundary onditions to the finite element model. Table 3. TBC properties Thermal Condutivity, k [W/mK] Thikness, l [μm] Top Coat (a) Thermally grown oxide (b) Bond Coat () The internal heat transfer oeffiient depends strongly on the atual surfae enhanement and the internal oolant passage design. Based on modern-day power turbine designs, the internal heat transfer oeffiient (h ) typially varies between 1000 and 3000 W/m 2 K, and the oolant temperature (T ) ranges between 800 and 1000 K. Here, as an approximation, a onstant surfae heat transfer oeffiient is imposed on the internal surfae of the airfoil. We onsidered two ases: A - h = 1000 W/m 2 K and T = 800 K, and B - h = 3000 W/m 2 K and T = 800 K for Hydrogen turbine, and only B for Oxy-fuel turbine. On the tip ross-setion of the airfoil and the external areas of the extension no heat transfer is assumed. 31

46 An angular veloity of rad/s (3600 rpm) is applied with respet to an axis of rotation, where the root radius is 0.6m. This ompletes the boundary ondition statement. The airfoil's substrate is modeled as a single-rystal superalloy whih has ubi symmetry and the <001> rystallographi axis is oinident with the pull diretion of the entrifugal fore experiened by the blade. Material parameters used are those of the seondgeneration Ni-based superalloy CMSX-4 [28]. Sine material parameters are temperature dependant, are is taken to speify this relation properly for elasti onstants, density, thermal expansion, and thermal ondutivity. Finite element analysis is performed in three steps. First, a thermal solution is generated. After solving for the thermal distribution, it is later imposed as a body fore loading. Together with the pressure load determined form CFD simulation and the angular veloity, they omplete the loading for the subsequent step - a stati, elasti strutural analysis. Additionally, we assume and also verify that the stress is below the yield stress orresponding to the temperature thus this initial instantaneous response is entirely elasti. Third step is time dependant reep-fatigue analysis desribed in the next setion User reep-fatigue routine The reep damage model, desribed in setions , an be implemented as a post proessor based on results from FEA simulation in either FEA pakage itself or other program, MatLab for example. However, this approah leads to over estimation of reep damage beause the effet of stress relaxation is not inluded. In order to inlude this effet, we developed a reep-fatigue user routine for ANSYS whih simulates the reep strain based on the model 32

47 desribed above, and alulates the damage equivalent stress, the damage parameter, and its evolution simultaneously. The fatigue damage method based on equations (22) and (23) was implemented as a part of the time dependant user routine for ANSYS, where both damage aumulations an be omputed simultaneously with or without interation between them. This approah allows us to monitor reep and fatigue damage evolution at every time step over the whole blade domain. Creep and fatigue damage parameters are monotonially inreasing and assigned to user state variables reorded at the end of eah time step. Subsequent time steps use this end values as an initial ondition to build up upon, thus preserving the loading history. Thus we have the opportunity to hange the loading (or boundary onditions) at subsequent time steps and not lose the damage history. This routine is alled at every element integration point. Creep and fatigue damage an be allow to interat at the required level with respet to different assumptions for material behavior. This way damage due to fatigue and reep proesses ating separately in one yle affets both the fatigue and reep damage rates in the next yle and a synergism is reated. Implementation based on this approah allows us to alulate fatigue damage aused from one loading ratio at a single simulation run. If it is assumed that there is no signifiant reep-fatigue interation then two separate aumulation rules an be used as: d f (, ), d f (, ). r r r fat fat fat In this ase the fatigue model an be implemented in suh way that in a single simulation several load ratios / frequenies an be evaluated. In this effort we implemented the fatigue damage aumulation for four loading ases where the loading ratios were assumed to be uniform over the entire airfoil, and the loal was the same for all simulations. max 33

48 These ratios and frequenies were: R = 0 at 1 yle / hour R = 0.33 at 1 yle / 6 hours R = 0.95 at 100Hz R = 0.85 at 100Hz, thus representing two low yle and two high yle fatigue ases. Loading ratios were ahieved by imposing different variations on the maximum stress. User routines for ANSYS v.11 are written in Fortran, then by using Intel Fortran ompiler a ustom ANSYS exeutable is reated, whih is later used to run the program. As with all ustomizable features rigorous verifiation is left to the user. Verifiation at varying temperatures and stress loads was done by simulating one diretional loading on three dimensional bar meshed with the same element used for meshing the volume of the airfoil. Results mathed analytial solution for time steps up to 0.25 hours. The one exeption is found to be equi-axed hydrostati ompressive pressure. This is beause the reep models in ANSYS are plasti models and therefore assumed inompressible. The third load step time dependant reep-fatigue analysis uses the results obtained from the first two load steps and depending on the tehnologial yle an simulate different loading time lengths. This routine is alled at every element integration point. The modified userreep.f file is inluded in Appendix B. The version listed uses equation (14) to alulate strain inrement. Sample input illustrating the use of state variables for the above routine is listed at the end of Appendix B. 34

49 2.5.3 Finite Element Modeling Results Hydrogen-fired turbine Results from the FEA first step thermal solutions are shown in figures 8 and 9 for Hydrogen turbine projeted operating onditions and internal ooling heat transfer oeffiient h respetively 1000 and 3000W/m 2 K. The later h value represents the urrent state of the art apabilities to enhane surfae heat transfer. External film ooling is not onsidered in these two ases. Plot shows the pressure and sution sides of the airfoil, as well as ross setion perpendiular to the staking line at 30 mm from the root (approximately in the middle). It is found that in the seond ase the maximum temperature does not exeed 904 o C (1177K). For the materials of interest (CMSX-4, Rene N5) this would allow longer running times if all boundary onditions do not hange in time. However, as disussed in the next hapter this may not be the ase beause the thermal ondutivity of the YSZ erami oating an hange with its exposure to elevated temperature. Figure 8. Temperature distribution for Hydrogen turbine, h = 1000W/m 2 K. Highest temperatures are found on the leading edge and first part of the sution side. This is also where it is usual to onsider for external film ooling. Through thikness gradient is found to be 35

50 Figure 9. Temperature distribution for Hydrogen turbine, h = 3000W/m 2 K. highest at the leading edge. The lowest temperatures are found in the internal ribs where onditions are reated for type II slip and onsequent reep damage. Based on the temperature distribution, pressure loading and angular veloity a stress distribution is obtained in the seond loading step. Von Mises stress is shown for this initial loading in figure 10 under 0 hours. As desribed earlier at this point the time dependant analysis begins. Results for stress distribution following relaxation from reep strain at three different running times an be seen in the plots to the right in figure 10. Figure 10. Stress distribution for Hydrogen turbine in MPa, for h = 3000W/m 2 K. 36

51 Simultaneously reep damage evolution has been alulated be the user reep-fatigue routine and is results are presented on figure 11 for hydrogen turbine. Simulation was extended to hours or approximately two years. Results are presented in olor plot and the sale is logarithmi. At hours the highest reep damage developed at the surfae of the sution side where we find high stresses ombined with high temperature. However, there is no signifiant damage being developed through thikness. Although this surfae area indiates high reep aumulation, it is very important to look at the atual damage values or replot on different sale. The highest reep damage in the area of interest was found to be D=0.23, whih indiates that the material response is still in the steady state region. Reahing value of D=~0.5 would indiate a transition to the highly non-linear tertiary reep state. Figure 11. Creep damage evolution for Hydrogen turbine at hours, for h = 3000W/m 2 K. Fatigue damage was alulated simultaneously over the whole blade for four loading ratios at three frequenies. By far the highest fatigue damage was found for loading ratio of zero or load unload variation, simulation a shut down and restart proess. As expeted, highest values are found at the greatly stressed parts of the blade. 37

52 Figure 12. Fatigue damage evolution for Hydrogen turbine at hours, for h = 3000W/m 2 K. The highest fatigue damage was found to be orders of magnitude lower than damage from reep. There was no interation between reep and fatigue in this ase. This was onsidered after tests were run on a small model to evaluate the possible interation for the temperatures and stresses found in our model. In figure 13 a) we see the damage evolution for reep and fatigue allowed to develop without interation. In fig. 13 b), damage from both reep and fatigue affets the damage evolutions in the next yle. Until the reep damage enters a tertiary phase very small influene is found on the fatigue damage evolution. On other hand, fatigue damage values are orders of magnitude lower than reep thus they have very little influene on reep damage evolution. Beause of this we did not allow for interation in this simulation and thus were able to evaluate four (possibly more) loading ratios at one. Otherwise four runs would be needed. 38

53 Figure 13. Creep-fatigue interation: a) no interation; b) Damage from reep and fatigue in one yle affets the overall damage in the next yle. Calulations for 250MPa maximum stress at 950 o C and loading ratio R = 0. 39

54 3.0 AIR PLASMA SPRAYED YSZ TOP COAT YSZ erami top oat is applied over the bond oat by either Eletron-Beam Physial Vapor Deposition (EB-PVD) or Air Plasma Spray (APS). The final result from the former method has a tooth like struture with raks oriented vertially, and the later method produes struture with raks embedded in it (splat boundaries) and parallel to the surfae. The top oat is designed to have a low thermal ondutivity, high oxygen permeability, high oeffiient of thermal expansion, and strain tolerant. These goals are ahieved by depositing a struture that ontains a large number of raks and pathways. However, both miro-struture and phase omposition are dynami during servie. Both methods have their advantages, with EB-PVD being favored for aeroengine appliations beause of its better surfae roughness and higher adhesion strength. APS applied YSZ have a lower thermal ondutivity, are less expensive, and preferred for land-based appliations [5, 6]. 40

55 3.1 APS YTTRIA STABILIZED ZIRCONIA Phase hange During deposition the molten YSZ undergoes rapid ooling upon impat with the bond oat and forms yttria-rih non-equilibrium phase. This phase ontains 7 wt% yttria, whether the YO 2 3 ZrO2 phase diagram (fig. 14) would predit tetragonal t ZrO2 to ontain 4wt% yttria. Figure 14. Y O ZrO phase diagram [29] Upon heating, the exess yttria starts diffusing out until it reahes an equilibrium onentration. That onentration is dependant on the heat treatment temperature. The yttria that diffuses out forms a new ZrO 2 phase with approximately 14 wt% YO 2 3. Aording to Trie et al.[30], this onversion happens at ~50 hours exposure to 1200 o C for the system they tested. After exposure to 1400 o C final transformation would take plae with the tetragonal phase ompletely transformed to monolini my2o3 and ubi ZrO2. 41

56 Figure 15. X-ray diffration pattern of YSZ, as proessed and after heat treatment [30] Miro-struture hanges In as sprayed ondition, the YSZ has lamellar morphology with eah lamella omprised of olumnar grains, submiron porosity between adjaent lamellae - intra-lamellar porosity, as seen on figure 16, and intra-lamellar raks. Trie et al. have found that after exposure to 1000 o C for 50 hours, intra-lamellar raks are losed but the long olumnar grains and lamellar Figure 16. APS YSZ shemati mirostruture [23]. 42

57 morphology are still present. After exposure to 1200 o C for 50 hours, the lamellar morphology is largely eliminated by hanges in the grain shape with some elongated grains still present ( t ZrO 2 ) but mostly replaed by equiaxed grains ( ZrO2 grains are rounded with porosity at triple points of grain lusters. ). After 50 hours at 1400 o C, the Thermal ondutivity In thermal insulators as ZrO 2, heat transported by lattie vibrations and radiation. This means that thermal ondutivity an be redued by inreasing phonon or photon sattering in the lattie struture of the erami [5]. Thermal ondutivity of the sprayed YSZ is redued ompared to that of the sintered bulk YO ZrO by the various defets in the mirostruture introdued by the appliation proess. Most signifiant ontribution in lowering the thermal ondutivity omes from porosity and miroraks oriented perpendiular to the diretion of heat flow through the oating. Changes in the shape, distribution and surfae area of the defets that our during upon exposure to high temperature alter the thermal ondutivity. Figure 17 shows plots of thermal ondutivity versus temperature for YSZ as proessed and heat treated. Figure 17. YSZ thermal ondutivity versus temperature [30]. 43

58 The inrease of thermal ondutivity at 1000 o C is attributed to the intra-lamellar miroraks losure. Sine they are oriented mostly normal to the oating surfae their relative ontribution is not expeted to be big, but rak satter around the normal inreases it. Further inrease of ondutivity at 1200 o C is attributed to the redution of porosity as a result of sintering between the adjaent lamellae. At 1400 o C, the inrease of ondutivity is attributed to the phase hange from t ZrO to 2 m ZrO 2, as explained in the two previous sub-hapters, beause signifiant volume fration of the oating is replaed by a phase with higher ondutivity. The thermal ondutivity of 98% dense m ZrO 2 at 800 o C is 3.6 W/mK ompared to 2.8 W/mK for similar density t ZrO MODELING CHALLENGES As shown, both YSZ's miro-struture and phase omposition are dynami during servie. Taking into aount YSZ s role in the TBC system, these hanges mean that the thermal load on the airfoil will evolve in time. Changed stiffness of the YSZ top oat should also be expeted as a diret result of densifiation, whih may affet the TGO/YSZ interfae and adherene. So far, for all modeling results presented, it was assumed that the thermal ondutivity of the top oat is onstant in time and has a value of k=0.9 W/mK, presenting the best possible value. Obviously, this is simplifiation whih an be justified in some ases but ultimately will lead to over estimation of the omponent apability. To illustrate the signifiane of this fator we ompleted simulations where the thermal ondutivity of the top oat was assumed to be k=1.5, 2.1, and 2.4 W/mK, orresponding to 50 44

59 hours of exposure at 1000 o C, 1200 o C, and 1400 o C respetively [30]. Results from the thermal solution over the airfoil for Hydrogen turbine projeted onditions are presented in figures 18 and 19. Figure 18. Temperature distribution, K. Hydrogen turbine, with thermal ondutivity of the top oat k = 1.5, 2.1, and 2.4 W/mK. Figure 19. Temperature distribution, K. Hydrogen turbine, with thermal ondutivity of the top oat k = 0.9 and 2.4 W/mK. It is learly shown that maximum temperature on the leading edge has inreased by 100 degrees for k=2.4 W/mK. Thermal distribution is also different over the airfoil. Sine all failure mehanisms disussed so far are, indeed, temperature dependant, the effets of YSZ property hanges should be inluded in the modeling proess based on availability of performane data 45

60 for the YSZ in the expeted working temperature window. Some newly developed high purity APS oatings sintering and phase transition were found not signifiant if the YSZ temperature does not exeed 1300 o C [31]. For hydrogen-fired turbine that might not be an issue, however, the oxy-fuel s turbine inlet temperature is signifiantly higher and more attention is needed to ensure this requirement is satisfied. 3.3 OXY-FUEL TURBINE MODELLING In this hapter we look in more detail at the oxy-fuel turbine. TBC material parameters used in this effort are listed in table 4. Thermal ondutivity of the bond oat is taken at it maximum reported value. Condutivity for the YSZ is assumed at 1.0 W/mK as for advaned oating being ontinuously kept under 1300 o C [31]. Table 4. TBC properties Thermal Condutivity, k [W/mK] Thikness, l [μm] Top Coat (a) Thermally grown oxide (b) Bond Coat () Figure 20 shows substrate temperature distribution for Oxy-fuel turbine with h = 3000W/m 2 K. For lower values of h the maximum metal surfae temperature will far exeed urrent material apabilities. Even at the maximum temperature of 1115 o C the omponent will have short life due to signifiant reep damage. It is obvious that with the metal surfae temperature at this level the top oat temperature will also exeed the allowed maximum of 1300 o C. Figure 21 46

61 shows the temperature distribution on the surfae of the YSZ assuming that its thermal ondutivity is still k=1.0 W/mK. The maximum surfae temperature is 1606 o C. Figure 20. Substrate temperature distribution for Oxy-fuel turbine, h = 3000W/m 2 K. Figure 21. YSZ surfae temperature distribution for Oxy-fuel turbine. Figure 22. Minimum ooling effetiveness required to maintain YSZ temperature under 1300 o C. 47

62 A ombination of two possible solutions was onsidered: a low ondutivity overlay oating and external film ooling. It was determined that the overlay oating ould not, by itself, redue the thermal load to an aeptable level. However, by introdution of external film ooling on the airfoil surfae, it was shown that film ooling ould, by itself, redue the thermal load to aeptable level. On the other hand, one needs to onsider the turbine loses assoiated with the ooling. External film ooling is ahieved by plaing ooling holes in the path of the hot gas ahead of areas with high temperature. It is possible to alulate a minimum external film ooling effetiveness that is required to maintain YSZ surfae temperature as presribed. Results of suh alulation are shown in figure 22. With this level of ooling the maximum YSZ temperature is 1292 o C. If we realulate the substrate temperature distribution, as shown in figure 23, we find a maximum temperature of 953 o C or 163 o C lower than before. It is also believed that the YSZ thermal ondutivity an be maintain at the urrently assumed k=1.0 W/mK. Figure 23. Substrate temperature distribution for Oxy-fuel turbine with external film ooling and h = 3000W/m 2 K. With this new thermal loading we an run the reep-fatigue user routine and evaluate reep and fatigue damage for the oxy-fuel turbine. This simulation was extended to 800 hours where the first signifiant reep damage aumulation was found, similarly to the hydrogen turbine at hours. Compared to the hydrogen-fired turbine here the maximum temperature is 50 o C 48

63 higher and we find omparable level of damage aumulation 20 times faster. Here we find the maximum damage on the pressure side surfae in areas that do not have muh higher temperature but signifiant stresses initially. These stresses also drive the fatigue damage aumulation at those same spots. These stresses an be alleviated by hange in the design as the internal ooling hannels in our model blade are more or less arbitrary. Creep damage is D=~0.2 at these spots whih is still in the steady state part of the reep urve. Figure 24. Creep damage evolution for oxy-fuel turbine with external film ooling at 800 hours. Fatigue damage aumulation is again found biggest for R=0, where now its rate is faster but still orders of magnitude lower than this of reep damage. This means that failure would either be reep or fatigue related and no signifiant interation is expeted. Figure 25. Fatigue damage for Oxy-fuel turbine, R=0. 49

64 4.0 THERMAL BARRIER COATINGS Typial thermal barrier oatings are made of a top layer of yttria stabilized zironia (YSZ) and a inter-metalli plasma sprayed bond oat (BC). Funtionally the TBC system should provide low thermal ondutivity and long life on a super alloy substrate. Figure 26. TBC system on Rene N5 superalloy: MCrAlY bond oat and YSZ top oat in as proessed ondition. (TGO is not present) During high temperature exposure, the inter-metalli bond oat forms a thermally grown oxide sale (TGO) that onsists predominantly of alumina. The formation of the TGO is found to have a ritial role in the failure of the TBC system. The usual mehanism of failure is assoiated with spallation at or lose to the TGO interfaes with YSZ or the bond oat. In the next two setions fous at the strutural and funtional hanges in both: the YSZ and bond oat / TGO as well as modeling approahes to predit failure of TBC system. 50

65 4.1 THERMALLY GROWN OXIDE Primary funtion of the bond oat is to protet the underlying substrate material from oxidative attak. The reliability of TBC systems is ritially dependant on the oxidation behavior of the bond oat. In perfet onditions, the bond oat is expeted to form a protetive, slow-growing, non-porous, free of defets, and adherent Al2O3 sale layer on its interfae with the YSZ. Establishment of an aluminum rih sale depends on availability of ritial ontent of aluminum, the struture of the alloy, and the oxidation onditions, in partiular temperature and oxygen partial pressure. The growth of the TGO during isothermal oxidation an also be haraterized by three stages [5, 10]: - Transient during this initial high temperature exposure stage a transient, heterogeneous sale is formed, onstituting of several transitional Al2O 3 strutures suh as,, and. These metastable strutures grow at a faster rate than Al2O3 and an onsiderably influene the overall amount of oxidation and the subsequent growth of stable Al2O3 phase. Later transformation of metastable Al2O3 to the stable Al2O3 produes 8-13% volume redution, whih an result in tensile stresses and raking in the sale. In general this stage ould last from seonds to hundreds of hours. - Steady state this stage begins with establishment of slow growing, long-term stable phase. - Breakaway stage this stage is haraterized by the failure of the steady state phase and formation of less protetive sales. 51

66 When the effets of property mismath are ombined with volumetri hanges assoiated with the oxide formation, espeially when the later are aompanied by a onstrained expansion of the TGO, large stresses an develop upon ooling, whih an lead to the nuleation of raks at or near the interfaes of the oxide [32]. The magnitude of these stresses depends on the interfae morphology, TGO thikness, temperature and on aumulated exposure time, and material properties. However, it is diffiult to determine the relative effets of these fators, sine all mehanisms are strongly oupled. Bond oat oxidation is believed to be one of the major degradation mehanisms ontrolling the lifetime of the TBC systems. The influene of oxidation has been desribed by the TGO thikness in most referenes, but in general there is no agreement on a single thikness value. Some publiations are reporting thikness values as low as 3μm and as high as 10μm to have detrimental effet, whether others report even higher thiknesses without failure. It is obvious that TGO thikness should be evaluated as a system dependant and that adherene of the BC/TGO/YSZ interfaes should play an important role to desribe the system failure. However, there is still a need to model the oxidation proess properly in order to evaluate its ontribution to the overall failure. 4.2 KINETICS OF THE THERMALLY GROWN OXIDE The exposure of the bond oat YSZ interfae to elevated temperature and oxygen leads to formation of thermally grown oxide. The TGO grows in two diretions internally and externally. Next two sub-setions disuss the kinetis of both and the modeling approah. 52

67 4.2.1 External TGO growth and modeling Externally, the TGO grows from the TGO/YSZ interfae towards the YSZ. This is a ompat oxide layer and its thikness is only slightly inreasing with time, and usually onsists of Al O. Depending on the system, it will be the only oxide visible for the first few hundred hours of exposure. 2 3 Figure 27. SEM of Compat TGO sale in 300μm APS YSZ system exposed to 1000 o C for 100h [10]. It is possible after observing oxidation behavior for one partiular system under several different temperatures and time exposures to obtain an average values for the ompat sale thikness. This thikness an be fitted to equation of the type: n Q h texp RT. (36) Here h is the sale thikness, t is exposure time, Q is ativation energy, R is universal gas onstant, T is the temperature, and n is an exponent approximately equal to This type equation was used by Meyer et al. [33] and later by other authors in slightly different form. We deided to use the following form (after [10]), whih is dimensionally orret and straightforward: 53

68 n t hox(, t T) Kn( T). (37) t0 Here t 0 is for normalizing time, n has the same value, and Kn(T) is temperature dependant growth rate onstant as: K ( T) K exp( E / RT), (38) n 0 where K 0 is a onstant [m], E A is ativation energy [J mol -1 ]. Ehsler et al. observed the behavior of several TBC systems and were able to get an average values for the growth rate onstant, as shown in Figure 28: A Figure 28. Temperature dependant growth rate onstant for ompat oxide sale [10]. The diagram learly shows different rate of growth below 1000 o C suggesting the formation of metastable alumina, and over 1050 o C formation of stable Al2O3. We have implemented the approah based on equations (37) and (38) as a post-proessor (in ANSYS) to the thermal solution for Hydrogen fired and Oxy-fuel turbines. Figure 29 a) shows the ompat TGO thikness for Hydrogen turbine projeted ondition after h of exposure based on the above published data. A maximum thikness of 12.6 μm is attained at the areas exposed to the higher thermal load at the leading edge and part of the sution side. We 54

69 have assumed that the thermal protetion ability of the YSZ has not hanged for the entire length of exposure. However, as shown in hapter 3.2, this may be underestimation beause of inreased thermal ondutivity of the YSZ. It is also assumed that the oxidation behavior of the system will be the same with different working fluid in the turbine. Figure 29 b) shows the ompat TGO thikness for Hydrogen turbine projeted ondition after h of exposure based on our experimental results for a TBC system provided by NETL as desribed in the next hapter. Figure 29. Compat TGO thikness for Hydrogen turbine projeted onditions, 16000h: a) published data [10], b) experimental results Internal TGO growth and modeling Internally, the TGO grows from the TGO/BC interfae towards the BC. This sale starts forming after a few hundred hours into exposure yle, and has a auliflower-like morphology when a ross setion is examined. It has faster rate of growth than the ompat layer and an be modeled by a similar equation: 55

70 t hinward (, t T) An ( T) t m, (39) 0 where A n (T) is temperature dependant rate onstant in meters, m is time exponent found to be m=0.7 [26]. Using this approah and again implemented as a post-proessor for the thermal solution one an obtain the expeted TGO internal growth over the surfae of the airfoil for different exposure to elevated temperature. Figure 30 shows the results from simulated hydrogen-fired turbine operating onditions at 16000h. Signifiant internal sale growth is expeted at the leading edge and part of the sution side wall of the omponent. This growth is of ourse possible only if aluminum is still available in the bond oat. Perhaps this is a very good indiator to implement external ooling in those areas, everywhere else the sale formation is onsiderably lower. Figure 30. Inward TGO growth thikness for Hydrogen turbine projeted onditions, 16000h β-phase depletion and modeling As the oxidation proess ontinues and Al is being removed from the bond oat, a β-nial phase depleted zone is reated with generally no more aluminum available to form a protetive oxide. Indiation for this is the dissolution of the Al rih β-phase. Aluminum form the bond oat an 56

71 also diffuse into the substrate and ontributes to the total depletion of the bond oat. Observing and measuring the depleted zone thikness an provide data for the kinetis of the reation. Ehsler et al. fitted the urves of averaged values for the depleted zone thikness to equation: t h phase (, t T) Bn ( T) t0 p, (40) where B n (T) is temperature dependant depletion rate onstant and p is the time exponent found to be between 0.6 and 0.85 for different system between 950 and 1100 o C. Figure 31. SEM of β phase depletion in 300μm APS YSZ system exposed to 1000 o C for 3000h [10]. Based on the above kinetis, there is no obvious relation of how the ompat layer and inward growth thikness would affet the reliability of the system. However, the depletion of the Al resoure in the bond oat provides an opportunity to establish a damage state variable based on the TBC s ability to protet the underlying substrate from oxidation. One the promotion of stable oxide phase is no longer ative other more brittle oxides are formed and this should not happen within the useful time of the system. For this purpose we define a damage state variable over the surfae of the airfoil that would indiate of how far this proess has reahed. Our new variable will again starts at zero for an as proessed material, and attains value of one as the depleted zone reahes the bond oat thikness. 57

72 D oat h phase. (41) h bond oat Implementing this approah, similarly to the ones for ompat sale and inward growth, we an visualize the oxidation protetion ability of the TBC system over the surfae of the air foil as seen on figure 32. Figure 32. Normalized β phase Depletion for Hydrogen turbine projeted onditions, 16000h, bond oat thikness is assumed at 200x10-6 m. 58

73 5.0 MATERIAL TESTING In onjuntion with model development, laboratory-sale experimental validation was exeuted to evaluate the influene of operational ompressive stress levels on the performane of the TBC system. TBC oated single rystal oupons were exposed isothermally in air at 900, 1000, 1100 o C with and without ompressive load. Exposed samples were ross-setioned and evaluated with sanning eletron mirosope (SEM). Performane data was olleted based on image analysis. Energy-dispersive x-ray (EDX) was employed to study the elemental distribution in TBC system after exposure. Nanoindentation was used to study the mehanial properties (Young s modulus and hardness) of the omponents in the TBC system and their evolution with temperature and time. For this purpose NETL has provided 35 oupons with simple geometries (2.54 m x 2.54 m x m) made from representative nikel-based single-rystal René N5 matries, oated with TBC systems (MCrAlY bond oat and 7% YSZ Air Plasma Sprayed (APS) top oat). Bond oat thikness is in the range of µm and top oat has thikness of µm. Turbine airfoils have highly irregular geometries and under onditions found in servie they are subjeted to omplex loadings. In most publiations results are reported from either uniaxial isothermal tensile reep or thermal yli tests. The former usually providing data for the reep of the superalloy and the overhaul performane, and the later - on TBC spallation as a funtion of thermally-grown oxide (TGO) growth, surfae roughness, temperature and thermal 59

74 mismath between the layers. As both tests provide very valuable data there is little known to the effet of ompressive stress on the performane of the superalloy / protetive system. 5.1 HYPOTHESIS A typial low pressure plasma sprayed MCrAlY bond oat for an APS YSZ TBC system has a high porosity and a highly irregular interfae with the top oat, as seen in figure 36. It is hypothesized that a uniaxial in-plane stress will affet the rate of diffusion and densifiation in the bond oat and thus the rate of growth and raking behavior of the thermally-grown oxide (TGO) layer. As a onsequene, the time to oating failure will also be effeted due to the altered rate of TGO growth and other mirostrutural hanges being a diret onsequene of applied stress. To better understand the loading and failure mehanisms of the oating system under loading onditions, it will be equally important to haraterize the evolution of the material s mehanial properties with temperature and exposure time, as it is expeted that signifiant hanges will be observed due to the mirostrutural and phase hanges. 5.2 CONCEPT The onept here is to plae the sample in a slotted test fixture and both in laboratory furnae where they are exposed to elevated temperature without yling. The differene in the oeffiients of thermal expansion for the nikel superalloy substrate of the samples and for the 60

75 test fixture an be utilized to introdue thermally indued ompressive stress in the samples. The fixture must be made of a material that an withstand the temperatures up to 1100ºC and has a oeffiient of thermal expansion that is less than that of the samples. The hoie was Silion Carbide (SiC), in partiular Hexoloy SA by Saint-Gobain Advaned Ceramis. It has a maximum operating temperature of 1900 o C, oeffiient of thermal expansion of 4.02x10-6 mm/mmk, Young's modulus of 410 GPa, and Poisson ratio of Saint-Gobain Ceramis produed three test fixtures by speial order. Piture of the test fixture an be seen in figure 33. Test sample is as proessed ondition is shown on figure 34. Figure 33. Test fixture (Hexoloy SA SiC by Saint-Gobain Ceramis) Figure 34. Test sample (ReneN5, MCrAlY, APS YSZ by NETL). Prior to thermal exposure the oupons are initially subjeted to non-destrutive examination (NDE) tehniques at West Virginia University (as a separate part of NETL program to evaluate the sample before exposure). The are subsequently subjeted to uniaxial ompressive stress at 61

76 elevated temperatures (i.e., C) for a maximum of 3,000 hours (as per test matrix provided in Table 5) in order to observe the effet on the TBC system. Table 5. Test matrix 100 h 300 h 1000 h 3000 h 900 o C (1173K) 1000 o C (1273K) 1100 o C (1373K) Stressed unstressed stressed unstressed stressed unstressed At all temperatures and running times, two samples were tested - one in the fixture's hannel and one on the fixture s side top surfae. The later one serves as a ontrol in order to evaluate the role of the applied stress and strain in the sample. The stress in the samples depends on the temperature, the effetive ross setional area of the fixture, and the learane between the sides of the samples and the fixture. Both a basi onedimensional analysis and three-dimensional finite element analysis of the thermal stress have been performed. The stress in the samples is highly sensitive to the aforementioned learane. Therefore two sides of the samples were further modified to ahieve the desired learane value for eah temperature. Table 6 lists the learane required in order to ahieve the stress levels desired at eah testing temperature, as well as stress relaxation as a funtion of time. Table 6. Clearanes and stress level for different testing temperatures. 900 o C (1173K) 1000 o C (1273K) 1100 o C (1373K) Clearane 0.007in (1.778x10-4 m) 0.008in (2.032x10-4 m) 0.009in (2.286x10-4 m) Stress (initial) 235 MPa 245 MPa 250 MPa Stress (1hour) 50 MPa 73 MPa 88 MPa Figure 35 shows a ontour plot of the expeted initial third prinipal stress and total displaement for the ase of exposure to 1000 o C with learane of 0.008in (2.032x10-4 m). 62

77 Figure 35. Third prinipal stress, MPa. Test fixture with 1273K. The samples were plaed in the test fixture and then together in laboratory benh furnae made by Carbolite, model CWF 13/13 apable of maintaining 1200 o C for extended periods of time, and maximum temperature of 1300 o C. After exposure to elevated temperature and stress all samples plaed in test fixture exhibited a strong reation between the SiC fixture and the Ni based René N5 quikly forming four interdiffusional layers. Removal of the samples was therefore impossible without damaging the side of the fixture. In order to prevent this unwanted reation Alumina strips were utilized between the fixture and the sample. These strips were ut to size from rigid Alumina paper and subsequently ompressed between two flat plates to ahieve the almost final thikness. Based on this thikness, the sides of the samples were modified to aommodate the strips and maintain the required gap with the fixture. All tests were uninterrupted, non-yled for the time length stated, exept for the 1000 and 3000 hours test whih ran simultaneously with the 1000 hour group in order to save time, thus they were thermally yled one beause of that and seond time beause of power outage. 63

78 5.3 SEM AND EDX EXAMINATION After ompleting the test, samples were sent for NDT evaluation one more time and subsequently mounted in resin and ross-setioned using standard metallographi preparation tehniques. In order to avoid harging in the Sanning Eletron Mirosope (SEM), after polishing the sample is sputter oated with Palladium. Cross-setioned samples are investigated using SEM Philips XL-30: After exposure and ross setioning all samples were subjet to SEM examination inluding two unexposed samples dubbed as NETL#1 and NETL#2: NETL# 1. One ontrol sample in as proessed ondition whih serves as referene point. Figure 36 shows SEM image of the top oat / bond oat / substrate system. The TGO sale is not present at this time; NETL# 2. Sample whih underwent ND testing - in order to evaluate destrutiveness of this method when applied to our system. While examining the ross-setion with the SEM, indents were diffiult to identify, sine surfae roughness exeeds indentation depth. However, at regions earlier identified under the optial mirosope ontaining indentations, no visible damage is found ompared to regions where no indentations were performed, thus deeming the method non-destrutive and safe to be used. 64

79 Figure 36. SEM Image of ross-setioned sample prior to thermal exposure. By using image analysis software on the SEM images several usually reported harateristis of the protetive system were quantified as a funtion of exposure time and temperature: Nominal TGO thikness; Nominal YSZ thikness; Nominal bond oat thikness; Nominal interdiffusion layer thikness; Nominal oupon thikness. When measuring the nominal thikness of the bond oat, we measured the length from the original bond oat / substrate interfae to the bond oat / TGO interfae after exposure. Compat TGO thikness was evaluated by measuring the length from the bond oat / sale interfae to the oxide / YSZ interfae in diretion normal to the bond oat surfae. This measurement was done in areas with well formed ompat sale. These two measurements do not quantify neither the real amount of the oxide sale (both on the surfae and internal) nor the amount of bond oat loss 65

80 due to oxidation around internal porosity. In order to assess the relation between the bond oat thikness, depletion, and the amount and quality of the resulting oxide sale we measured the mean vertial thikness of the bond oat and the oxide sale, inluding ompat and internal. Mean vertial thikness is defined as the average thikness per unit length and is found from image analysis. YSZ porosity was evaluated by point ounting utilizing a grid of 16 point imposed over the image and ounting the points falling over pores. This method has the advantage of being very quik and although the ount needs to repeat many times it still obtains results muh faster that area or line fration tehniques. Some of the relevant results are shown in the following disussion setion. 5.4 INDENTATION TESTING L. S. Cook et al. [34], measured the dynami modulus and internal frition in low pressure plasma sprayed NiCrAlY bond oatings at different temperatures using a piezoeletri ultrasoni omposite osillator tehnique. They investigated oatings with different ompositions: Ni- 15.6Cr-5.2Al-0.20Y, Ni-17.2Cr-11.6Al-0.98Y, and Ni-33.0Cr-6.2Al-0.95Y. Reported values of Yong s modulus at room temperature were respetively: 205.0, 199.8, and GPa. With inrease of temperature the modulus in all ompositions was found to redue in a linear fashion to 137 (900 o C), (900 o C), and GPa (1000 o C) respetively. The authors onluded that, sine the measured Young modulus for LPPS oatings at all temperatures was higher than the one for APS top oating (usually 20-60GPa) differenes in the oating modulus would not aount for TBC durability differenes. 66

81 Franke et al. [35] investigated a single and two-phase platinum modified aluminade oating on nikel-based superalloy in as oated and after thermo-mehanial fatigue loading by SEM and nanoindentation. They followed the mirostrutural evolution in the β-nial grains as the PtAl2 phase dissolves and its influene on the assoiated mehanial properties. They found that both the Young s modulus and the hardness dereased after exposure. The hange in the loal mehanial properties was related to the mirostrutural and hemial hanges in the oating. For their partiular system they found a redution of Young s modulus from 225 GPa to 150GPa and redution of the hardness from 12 to 6 GPa. A strong influene of the Al onentration is also noted on the modulus of the NiAl phase, where a derease of the Al ontent was found to lead to redued modulus and inreased hardness. Li et al. [36] investigated the mirostrutural evolution of the NiCrAlY/CrON system after heat treatment. They measured the elasti modulus and residual stress of the oating system by nanoindentation and photo-stimulated luminesene spetrum methods. They onluded that the phase transformation β γ γ results in lower modulus for the heat treated sample, from 204.2GPa in annealed ondition to GPa at 100 hours exposure. Although this result oinides with the finding of Franke et al. [30] the redution of the modulus for this partiular system is not as great. Measured values of the Young s modulus are influened from both the underlying rystal struture and the hemial omposition of the oating. It should be expeted that systems with different omposition may have a dissimilar evolution of their mehanial properties. With respet to our testing program our goal is to obtain the evolution of the Young s modulus and hardness as a funtion of the exposure temperature and time, and operational level of stress in the substrate / protetive layers system. 67

82 Mehanial properties of the substrate and oating layers of all exposed and two as proessed samples were investigated by nanoindentation. Tribo Indenter by Hysitron was used in this investigation. It was fitted with 2D, three-plate apaitive fore/displaement transduer and a three-faed pyramid Berkovih indenter. The Berkovih tip is the standard nanoindentation tip. It has a total inluded angle (angle from one edge to the opposite side) of degrees. A typial radius of urvature for a standard Berkovih tip would be approximately 150nm. Nanoindentation is used to determine two key material parameters - Young's modulus and hardness. With typial indentation depth between nm measured properties are loal and several indents need to be performed to obtain an average value. Several alulations are performed based on the measured load-displaement urve [37]. A typial loading urves is shown of figure 37. Figure 37. Typial load-displaement urves obtained by indenting with the same load within the substrate material. Exluding both upper and lower ends, a selet portion of the unload urve is used to fit the following power law: 68

83 P Ah h m f (42) where P is the applied fore, A is the ontat area, and h is the indentation depth. The first derivative of the fore with respet of the indentation depth evaluated at the maximum fore is the ontat stiffness: The ontat depth is alulated as: S dp dh. (43) h P hmax 0.75 S max (44) where 0.75 is geometri onstant to aount for edge defets. From here the hardness is alulated as: H P Ah ( ) max (45) and the redued modulus is: E r S. (46) 2 Ah The ontat area is determined from the probe area funtion Ah where is found by equation (44). This area funtion is experimentally obtained and aounts for imperfetions and wear of the indentor tip. The obtained redued modulus depends on both - the indentor and sample materials. The modulus of elastiity is related to the redued modulus through the following equation: h E E E r sample indentor. (47) 69

84 Properties for standard diamond indentors are known and listed as: Eindentor 1140GPa indentor In order to evaluate the modulus of all tested layers within the sample we need to assume a Poasson's ratio for every material. The representative values used in our alulations are: Substrate: 0.3 Bond oat: 0.3 YSZ: 0.1. There are several mehanism by whih the high temperature oatings degrade during servie. These inlude oxidation, orrosion and interdiffusion with the underlying alloy substrate. Although MCrAlY oatings have good orrosion and oxidation resistane, after extended exposure to elevated temperature, element interdiffusion an greatly impart the stability of the oating and its ability to protet the substrate. During oxidation, aluminum from the bond oat diffuses to the oating interfae with the YSZ to form a protetive alumina sale. With ontinuous exposure this sale grows onsiderably and most predition models link its thikness to the oating failure. This proess ontinues until the aluminum onentration beomes insuffiient to form a ontinuous protetive sale. At this point less protetive oxides of other elements (spinel) begin to form. Elasti modulus is a key material parameter of interest and through nanoindentation we an asses its evolution with exposure time and temperature at different regions of the oating layers. Measurements were performed on already ross-setioned and polished, exposed and "as proessed" samples. Partiular interest is plaed on five regions: substrate bulk (SB), substrate interfae diffusion zone with the bond oat (SI), bond oat bulk (BCB), bond oat interfae 70

85 diffusion zone with the TGO (BCI), and the bulk of the top oat (TC). Figure 38 shows the measurements loations on a SEM image of the substrate / TBC system. Figure 38. Indentation loations shown on SEM image of a sample exposed to 1100 o C for 300 hours. In order to determine the indentation load we initially performed indents with fores varying form 500µN to 7000µN at all loations. As reported in many publiations, it was found that at small loads the indentation size effet played signifiant role and harness and modulus values were signifiantly higher. For indentation depth over 100nm both hardness and modulus beome onstant with a small deviation. Thus for indentations within the substrate and the bond oat we used a loading funtion 10 seonds load and unload, maximum fore of 5000µN and no holding period. For indentations within the top oat we used a loading funtion 10 seonds load and unload, maximum fore of 7000µN and no holding period. Indentations were performed in patterns of 4 x 3 indents, where the last two were removed to make the total of 10 indents in a pattern. By exeuting this kind of pattern we were able to orrelate indentations results with an SEM image. For the bulk of the substrate and bond oat indentations were 10µm apart where in 71

86 the rest of the loations they were spaed at 5µm in order to fit the whole pattern in a smaller region. At eah indentation loation a minimum of three patterns per sample were exeuted to obtain an average value. Figure 39. Indentation pattern - 10 indents, 5µm apart within the β-phase depleted zone of the bond oat (BCI). 5.5 RESULTS AND DISCUSSION Exposure to 900 o C As per the test protool in table 5, six samples in three pairs were exposed in air to 900 o C for 100, 300, and 1000 hours respetively. Before and after the exposure they were tasted in a related study using non-destrutive methods whih were deemed to be unaffeting the samples in any signifiant way. Failure by loss of the protetive oating did not our in any of the 900 o C samples. Figure 40 is omposed of SEM images of ross setions of the above samples A to F. 72

87 Several observations an be made: Bond oat / top oat interfae is highly irregular shape and roughness. This is typial for APS top oat sine it promotes better adhesion. As a onsequene the oxide sale has the same irregular shape as oppose to the sale found in EBPVD top oats whih shape an be approximated as a sine urve. The bond oat porosity is found to be progressively dereasing from A to F. The stressed 100h sample (B) had lower porosity ompared to the unstressed (A). The same trend was observed at the higher temperatures. At the 300h stressed sample (D) porosity was almost non-existent exept for pores with oxygen availability where oxide sale was formed. The APS YSZ has progressively densified from A to F sample, where the 1000h stressed sample (F) has lower porosity than the unstressed (E). With exposure to elevated temperature and oxygen the bond oat onsistently forms a ompat oxide sale whih thikness is listed in table 12 as a funtion of time at onstant temperature. Exept for sale growth around existing pores there is very little internal oxidation. 73

88 Figure 40. SEM images of ross setion samples exposed to 900 o C for 100, 300 and 1000 hours. Unstressed samples are shown at left olumn, stressed at the right. There is small interdiffusion zone between the bond oat and the substrate inreasing with exposure time and temperature (in the following ases). At 900 o C there is little or no influene from the applied stress on the thikness of this zone. 74

89 At 100 and 300 hours exposure there were no raks assoiated with the oxide sale. At 1000 hours, there are raks found within the oxide sale propagating from / into imperfetions on the interfae between the sale and the top oat or splat boundaries. These raks were loal and did not propagate into the next undulation of the sale. Initially the bond oat is a two phase material onsisting of the ordered intermetalli β-nial and the fae entered ubi γ-ni. In presene of oxygen and elevated temperature the β-nial is apable of forming a protetive α-al 2 O 3 sale by diffusion of Al towards the interfae with the oxygen permeable top oat. However, there is also an inward Al diffusion towards the Ni-based superalloy whih an lead to a loss of Al to the superalloy Figure 41. Spetrum mapping of bond oat / oxide sale, 900 o C, 1000 hours and promote unwanted topologially lose paked (TCP) phase formation. In our 900oC exposed samples this TCP formation an in found as a thin uniform layer below the bond oat interfae with the substrate. Aording to Gleeson et al. [38] and [39] a bond oat 75

90 with Al ontent over 10 at.% (4.8 wt.%) has the potential to form exlusively α-al 2 O 3. In partiular a Ni-Al alloy needs at least ~34 at.% Al in order to form an exlusive α- Al2O3 and only ~10 at.% Al if it has 6 at.% Cr. EDX analysis of our bond oat s β and γ phases revealed an Al ontent of ~(25 38) and ~(13-16) at.% respetively, where Cr was ~ (4 6) at.% in β- and ~(13-16) at.% in γ-phase. Taking into aount the error of the EDX system (2-3%), we are still over the ritial Al level in both the β- and γ-phase. Spetrum mapping of a bond oat/tgo/top oat region on 900 o C /1000 h. sample, as seen in Figure 41 (region A), reveals the elemental distribution and exlusively Al and O in the TGO part of this segment. Further testing on points within the TGO showed a 100% α-al 2 O 3. Regions B and C show sales formed on the top of the alumina where some Cr inlusions are visible suggesting the presene of other oxides. Compat TGO growth 900 o C, m -6 Thikness, m Exposure time, h 900C stressed 900C w/o stress Literature Figure 42. Compat TGO thikness at 900 o C, measured and data from literature. After 100 hours exposure to 900 o C both samples (stressed and non-stressed) ahieved basially the same ompat TGO thikness as seen in figure 42. After taking into aount the deviation of the thikness results, the same is true for 300 and

91 hours exposure. There is no evidene that the stress applied to the sample had any effet on the TGO thikness in all of the 900 o C ases. The yellow line in figure 42 shows TGO thikness predition based on published average data by Eshler et al. [10] on APS YSZ / MCrAlY systems. Predition and measurement from our system agree at 100 hours. However, in our speimens we find a more exponential growth in the next 200 hours where date from other systems indiates almost linear thikening. From 300 to 1000 hours the rate is the same between both systems. Therefore we an expet signifiantly different oxide sale thikness than the one modeled by the average data. If we assume that the ativation energy in equation 38 is still the same, the best fit to our experimental data is found by hanging the pre-exponent oeffiient to 3.0x10-5. So far, when measuring the nominal thikness of the bond oat, we measured the length from the original bond oat / substrate interfae to the bond oat / TGO interfae after exposure. Compat TGO thikness was evaluated by measuring the length from the bond oat / sale interfae to the oxide / YSZ interfae in diretion normal to the bond oat surfae. This measurement was done in areas with well formed ompat sale. These two measurements do not quantify neither the real amount of the oxide sale (both on the surfae and internal) nor the amount of bond oat loss due to oxidation around internal porosity. In order to assess the relation between the bond oat thikness, depletion, and the amount and quality of the resulting oxide sale we now measure the mean vertial thikness (MVT) of the bond oat and the oxide sale, inluding ompat and internal. Mean vertial thikness is defined as the average thikness per unit length and is found from image analysis. For the ase of 1000 hours exposure to 900 o C the mean vertial sale thikness was found to be and µm for the stressed and nonstressed samples respetively. The mean vertial bond oat thikness was and μm 77

92 respetively. These results illustrate the amount of internal oxidation not aptures by the ompat TGO thikness. In this partiular ase there is ~10 µm internal oxide sale ompared to ~ 6 µm for the ompat external growth showing signifiant amount of Al being lost internally. It also onfirms that there is no signifiant differene between stressed and non-stressed samples as both samples had the same MVT of oxide sale and bond oat. Material properties were evaluated by nanoindentation on all 900 o C samples. The omplete data set is listed in table 11. Plotted data for Young s modulus and Hardness in all measured regions as desribed in setion 5.4 are in figure 43 and 44 respetively. Figure 43. Young s modulus of substrate bulk (SB), substrate interfae (SI), bond oat bulk (BCB), bond oat interfae (BCI), and top oat bulk (TC) at 900 o C, all samples. During the preparation of our samples the ontrolled diretion of the single rystal (substrate) was <001> through the sample s thikness. Subsequently the other diretions were not 78

93 ontrolled. When the samples were ross setioned we didn t know the exat orientation of the indentation surfae with respet to the rystal lattie. Thus, as expeted, the results for the Young s modulus from indentations in the substrate varied randomly between the samples from ~235 GPa to ~270 GPa. As the interdiffusion zone at 900 o C is very small values for the substrate interfae were strongly influened form the bulk lattie orientation. Measured values from the bond oat bulk have a large deviation whih is to be expeted beause of the two phase omposition of the oating. At 900 o C there is no signifiant evolution in the modulus with exposure time inrease, however, a larger deviation is notieable at 1000 hours as the relative amount of β- and γ-phase started to hange. Measured values for the interfae are the same with the bulk at 0 hours (for both substrate and bond oat) beause there are no signifiant diffusion proesses started. At 100 and 300 hours it is very diffiult to evaluate the bond oat interfae values as the γ-phase zone under the TGO is very small, thus values for BCB and BCI are essentially the same. At 1000 hours, where the γ-phase zone is easily identifiable under the optial mirosope of the nanoindentor there is signifiant deviation of the modulus of BCI ompared to the bulk espeially for the stressed sample. This inrease of the modulus an be attributed to the higher onentration of Cr and Co found in the γ-phase lose to the interfae (Ni-50, Al-14, Cr-18, Co-16 at.%) ompared to those in the β-phase (Ni-49, Al-38, Cr-4, Co-8 at.%) and γ-phase (Ni-51, Al-16, Cr-15, Co-18 at.%) in the bulk of the oating. Measured modulus values from the bulk of top oat within one sample also showed a large deviation (+/- 10%) and signifiant drop in the modulus is observed with exposure time. Data from the one of the 1000h samples was not usable and therefore missing from the plot. 79

94 Figure 44. Hardness, GPa, of substrate bulk (SB), substrate interfae (SI), bond oat bulk (BCB), bond oat interfae (BCI), and top oat bulk (TC) at 900 o C, all samples. There is general inrease in the hardness in all loations exept the YSZ, most signifiantly at the substrate interfae. The bulk of the top oat, however, showed a signifiant derease in the modulus from ~257GPa in as proessed ondition to ~200GPa after 1000 hours exposure in air to 900 o C. 80

95 5.5.2 Exposure to 1000 o C Two pairs of samples were exposed to 1000 o C for 100 and 300 hours. There was no failure in all samples. These samples were fully investigated with SEM and nanoindentation. Currently two more pairs are being exposed to 1000 and 3000 hours. Sine these tests are not ompeted at this time the 1000 o C exposure will not be disussed in detail in this work. However, all measured data is found in tables 11 and 12. SEM images form the 100 and 300 hours samples are found in figure 45 and material property data is plotted in figures 46 and 47. Figure 45. SEM images of ross setion samples exposed to 1000 o C for 100 and 300 hours. Unstressed samples are shown at left olumn, stressed at the right. 81

96 Figure 46. Young s modulus of substrate bulk (SB), substrate interfae (SI), bond oat bulk (BCB), bond oat interfae (BCI), and top oat bulk (TC) at 1000 o C, all exposed samples. Figure 47. Hardness, GPa, of substrate bulk (SB), substrate interfae (SI), bond oat bulk (BCB), bond oat interfae (BCI), and top oat bulk (TC) at 1000 o C, all exposed samples. 82

97 5.5.3 Exposure to 1100 o C There were two pairs of samples exposed for 100 and 300 hours, and 6 samples in three pairs exposed to 600 hours. Figure 48 is omposed of SEM images of the above samples. At 100 hours a ompat oxide layer is visible, its thikness reahing 5.9 (+/- 1.91) μm in the stressed sample and 5.11 (+/- 1.66) μm in the unstressed one. There is an oxide sale formed around enlosed pores in the bond oat (with aess to oxygen) and depleted β-phase surrounding them. Between 100 and 300 hours the ompat oxide sale thikened as well as the internal oxide. Although being μm thik there is no signifiant raking in the sale. However, raks are found (even at 100 hours) near or at existing imperfetions on the bond oat / YSZ interfae as seen in figure 48 B). Between 300 and 600 hours the internal oxidation developed more aggressively and the distintion of the ompat sale beame more diffiult, and was measured only in the segments where it was well defined. The ahieved thikness is plotted in fig. 49 along with data from literature [10]. There is again deviation between our experimental and the published data where it predits μm less then measured even at 100 hours. However, both the predited and measured have approximately the same urvature and by modifying k o to 4.0x10-5 in eq. 38 we an get a good fit to the experimental data. Thus now we an estimate the potential ompat sale growth in the models shown in hapter 4 with respet to our urrent material apabilities. The ompat sale was examined with EDX and was found to be 100% Al 2 O 3 in the dark regions (on the SEM images) and ~95% Al 2 O 3 and ~5% Y 2 O 3 around some of the lighter spots within the TGO. Compat oxide sale growth does not show any dependene on the applied stress in the samples. However, porosity in the bong oat is being eliminated faster in the stressed samples, best seen in images C and D of fig

98 Figure 48. SEM images of ross setion samples exposed to 1100 o C for 100, 300 and 600 hours. Unstressed samples are shown at left olumn, stressed at the right. Of all 600 hours exposed samples one failed after exposure and sitting on the desktop for 3 days. Failure ourred by rak propagation within the TGO sale. Propagation started from the edge of the sample and extended towards the middle. The top oat did not ompletely ome off. This sample was not stressed during exposure. For the rest of the samples failure did not our. 84

99 Compat TGO growth 1100 o C, m -6 Thikness, m Exposure time, h 1100C stressed 1100C w /o stress Literature Figure 49. Compat TGO thikness at 1100 o C, measured and data from literature. However, in all of the 600 hours samples a severe raking and oalesent raks extending for more than 100µm an be seen at the samples' edges, as seen in figure 50. Figure 50. Coalesent raks within the oxide sale at the edge of the sample at 600 hours, 1100 o C. Several observations an be made for the samples with 600 hours exposure: All samples featured vertial raks within the YSZ either starting from the surfae or embedded. 85

100 All samples featured a signifiant oxide sale formation in the form of: - Compat TGO. - Internal growth around porosity with oxygen availability. - "Cauliflower" like internal growth. - Spinnel. Craking within and around the TGO was observed in all samples. Craking around and in the oxide sale an be haraterized as: - Craks ontained within the sale. - Separation between the sale and the bond oat possibly propagating from / into the sale. - Separation between the sale and the top oat. - Craks starting in the TGO and propagating within the YSZ possibly onneting with other neighboring raks. - There ould be more than two parallel raks through the thikness of the sale. While all samples featured raks within the oxide sale, an important distintion an be made that oxide sales over well depleted β-nial phase zones of the bond oat (fig.48 F)) had fewer and smaller raks than sales over less depleted zones (fig.48 E)). On other hand, β-phase depletion in the bond oat was uniform neither between the samples nor within the sample. No diret orrelation ould be made between the depletion and test onditions (indued stress in the sample). The following table summarizes these results. 86

101 Table 7. Summary of bond oat depletion behavior in all 1100 o C, 600 hours exposed samples C Bond oat β-phase depletion Full Partial Uniform 100 str 3 x x x x 300 str 16 x x x x 600 str 15 x x x x 600 str 8 x x x x 600 str 26 x x x x There was one sample with uniformly and fully depleted bond oat, two with uniformly partially depleted bond oat and three samples featuring zones with full and zones with partial depletion. As explained earlier, in order to assess the relation between the bond oat thikness, depletion, and the amount and quality of the resulting oxide sale we now measure the mean vertial thikness (MVT) of the bond oat and the oxide sale, inluding ompat and internal. MVT was evaluated for all 600h samples. Results are listed below, in table 8. Table 8. Mean vertial thikness of the oxide sale and bond oat in all 1100 o C, 600 hours exposed samples. Mean vertial thikness, µm Sale BC BC not depl. 600 str str str str It an be seen that the samples (and regions within the samples) with thinner bond oat ( C 173.8μm) had formed thiker sale ( μm), but their bond oats were fully β-phase 87

102 depleted. The samples (or the regions within one sample) with thiker bond oats ( μm) had formed less sale ( μm), and featured a non-depleted β-phase zone. Table 9 lists measurement for all 600h samples made in the traditional way. In all ases the thikness of the sale (from table 8) far exeeded the thikness of the ompat sale beause it inludes the internal oxidation. Table 9. Measured sample and oatings dimensions in all 1100 o C, 600 hours exposed samples. Exp. Hours str Sample # Comp. TGO Bond Coat Depleted zone Interdiffusion zone µm std. dev. µm std. dev. µm std. dev. µm std. dev. 600 str str str As desribed above, regions with full bond oat depletion featured a more ompat oxide sale ontaining less raking and separation than those regions without full depletion. Several suh regions were investigated with EDX spetral mapping to determine the elemental distribution within and around the sale. Two of these are disussed below. Fully depleted bond oat and ompat sale: Figure 51 shows a region with full depletion of the bond oat and thik but ompat oxide sale formed on the bond oat / YSZ interfae (dark grey areas), other oxides formed on the top of the ompat sale (light grey areas) as well as some internal oxidation. From the spetrum mapping an be seen that the ompat sale onsists of aluminum and oxygen, whih was onfirmed by spot omposition analysis to be Al 2 O 3 and some Y 2 O 3 (~5%). Traes form Cr, Ni and Co whih other oxides an be seen on the top of the sale (or the light gray areas). 88

103 Figure 51. SEM image of the mapped area and spetrum maps - fully depleted bond oat. Not fully depleted bond oat and non-ompat sale: Figure 52 shows a region with not fully depleted bond oat within the same sample as above. Although the example is given for one sample the same is found true for samples that are Figure 52. SEM image of the mapped area and spetrum maps - not fully depleted bond oat. uniformly, fully or not fully, depleted. In this ase a signifiant oxide growth an be seen on the bond oat interfae with the YSZ. A ompat sale annot be distinguished beause of severe 89