COMBINING INNOVATIVE PRODUCTION COMPETENCES Fraunhofer-Einrichtung für Gießerei-, Composite- und Verarbeitungstechnik IGCV
Fraunhofer IGCV Facts and figures Es tablis hed: July 01, 2016 Managem ent: Prof. Dr.-Ing. Gunther Reinhart (executive) Prof. Dr.-Ing. Klaus Drechsler Prof. Dr.-Ing. Wolfram Volk Augs burg s ite headquarters Management, administration, fields of research: processing and composite technology Garching s ite: Field of research: casting technology Staff size ~ 140 employees Supported by: Seite 6
Fraunhofer IGCV We use synergies in these fields of research and development: Casting technology Molding materials Sand and gravity die casting processes Simulation and design of mold and cast components Seite 7 Composite technology Hybride Hybrid composite constructions Online process monitoring Materials and test engineering CFRP manufacturing engineering Recycling of composites Efficiency and balancing Processing technology Resource efficiency in factories Intelligent networked production Flexible production Networked modeling and simulation Additive manufacturing
TIA Fraunhofer Workshop Online monitoring and classification of carbon fiber and textile production defects using scalable line scan optics and computer vision Andreas Margraf 05/10/2017
Contents Online process monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 19
Online Process Monitoring Online process monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 20
Online Process Monitoring System solutions Customer-specific concepts Online Dynamic processes Vis ion applications Image processing chains Design & optimization Modularised & scalable Data handling High volume data Classification & analysis Realtime conditions 21
Overview Online Process Monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 22
Overview: carbon fiber and textile defects Carbon fiber production 1 Nitting fault Fiber misalignment Binder application Surface pattern Overlap Automated fiber placement 2 AFP gaps Gap AFP hotspots 1: Source: Bunsell, A. R.: Fibre reinforcements for composite materials. Composite materials series, Bd. 2. Amsterdam, New York: Elsevier 1988 2: Source: Coriolis Composites GmbH, www.coriolis-composites.eu 23
System layouts Online process monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 24
System layouts Preconditions and system features varying conditions along the production process» material state changes significantly» different environments on inspection points Therefore: adaption of software and hardware Material ty pes PAN precurs or ox idized PAN Carbon fiber Fiber color white black black Fiber s urface glossy dull glossy Illumination dark field diffuse incident light diffuse incident light 25
sensor width (px) sensor width (px) TIA Workshop: Carbon fiber and textile defect detection System layouts Vision hardware moving target small defects high resolution line scan sensor scan range extension (factor 3) - patent pending high volume data use of FPGA 1 Scan range extension Camera Illumination Area scan: 8x8 px / t t Line scan: 8 px / t t Line scan sensor over moving target 1: Field Programmable Gate Array Page 26
Image processing Online process monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 27
Image processing Software requirements solid architecture image filtering segment regions of interest (ROI) classification of defects: use of SVM 1 28 PAN PANOX Carbon Fiber Classification with SVM 1: Support Vector Machine
Image processing Image enhancement and segmentation Split regions of interest Merge regions by distance-based algorithms 29
Machine Learning Online process monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 30
Machine Learning Approaches Cartesian Genetic Programming (CGP) Optimizing image processing Simplify configuration ANN 1 /SVM 2 based classification Principles of Organic Computing (OC) 3 Applications Cartesian Genetic Programming 4 Usage: CF, further material defects, imaging technologies 1: Artificial Neural Network 2: Support Vector Machine 3: See expert group for OC: http://fg-oc.gi.de/ 4: A. Margraf, A. Stein, L. Engstler, S. Geinitz, J. Hähner, An Evolutionary Learning Approach to Self-Configuring Image Pipelines in the context of Carbon Fiber Fault Detection (2017). Submitted to: International Conference on Machine Learning and Applications 2017. Page 31
feature 2 TIA Workshop: Carbon fiber and textile defect detection Machine Learning Fiber Monitoring Binder Monitoring Image Processing Data Analysis P r IME Interfaces Data Management Classification 1: Professional Integrated Monitoring Environment feature 1 Seite 32
Integrated Solutions Online process monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 33
Integrated Solutions Dynamic hardware control & defect detection dynamic binder application - patented flexible measurement unit Evaluate binder application homogeneity nitting defect detection optimization of preform permeability acceleration for preform compaction Spool gate Binder application unit Robot based fiber placement Fiber placement with integrated binder application Image acquisition Image enhancement Automated evaluation Inline evaluation of binder application Binder pattern Fiber misalignments and nitting defects Page 34
Integrated Solutions Roving metrology full surface-metrology by InFactory Solutions cooperation with InFactory Soluations and Coriolis Composites automated Tool detection and measurement automated localisation in the FAI process Goals: meet FAI requirements ensure integration in industrial process Overlap Gap Page 35
Integrated Solutions Tem perature control Challenge: Keep element temperature in a critical range solution: embedded control loop for IR heater using an IR camera tasks: collision control hotspot detection realtime conditions» Proof of Feas ability by Fraunhofer IGCV Computed control parameters IR camera view Page 36
Conclusion and outlook Online Process Monitoring Overview System layouts Image processing Machine learning Integrated solutions Conclusion and outlook 37
Conclusion and Outlook S um m ary Nitting defects Detection by image processing Mis aligned Fibers Vision System & Classification Binder application Flexible vision systems EA based approaches Over l a p AFP gaps & overlays Laser line systems G a p Fiber Placem ent Temperature Control System Enhancem ent Page 38
Contact us Andreas Margraf Phone +49 821 90678-424 andreas.margraf@igcv.fraunhofer.de S teffen Geinitz Phone +49 821 90678-222 steffen.geinitz@igcv.fraunhofer.de André Wedel Phone +49 821 90678-223 andre.wedel@igcv.fraunhofer.de Fraunhofer-Einrichtung für Gießerei-, Composite- und Verarbeitungstechnik IGCV Am Technologiezentrum 2 86159 Augsburg Germany Page 39