January_2021_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | J A N U A R Y 2 0 2 1 2 1 i.e., high confidence data points and why others are low confidence data points in the fuzzy overlap between two clusters or assigned to the wrong cluster. Additionally, it can be used as “pre-labeling” for supervised ML. CONCLUSIONS Combining data science know- how with our materials science knowl- edge enables us to forge new paths in microstructure research and reach high- er qualities in microstructure segmen- tation and classification. This marks a further step toward comprehensive and integral processing-structure-prop- erties-performance correlations and understanding. Regarding ML-based segmentation and classification, a core aspect is the ground truth assignment, which must be well-founded and as ob- jective as possible. For that we must look at the entire process of building a ML-based segmentation and classifi- cation pipeline in a holistic approach, which not only focuses on images and/ or data and the corresponding ground truth, but starts by selecting suitable samples, establishing reproducible sample contrasting, and finding the optimum imaging technique. Metada- ta and a uniform ontology also play an increasingly important role in this con- text. Therefore, our domain knowledge in materials science and metallography is still indispensable. When applying machine learn- ing to complex microstructures such as bainite, the microstructure’s visual appearance under a microscope might no longer be sufficient for an objec- tive ground truth assignment and sup- porting methods must be found. The approaches presented in this bainite case study, i.e., correlative characteri- zation using EBSD, specific production of reference samples, and use of unsu- pervised learning techniques, are also transferable to other microstructure characterization tasks. ~AM&P Acknowledgments The authors wish to acknowledge the EFRE Funds of the European Com- mission and the State Chancellery of the Saarland for support of activities within the ZuMat project. The authors would also like to thank steel manu- facturer, AG der Dillinger Hüttenwerke, for providing the sample material and TA Instruments for producing samples with the quenching dilatometer. For more information: Martin Muel- ler, doctoral student Saarland Univer- sity and Material Engineering Center Saarland, martin.mueller1@uni-saa- rland.de. Dominik Britz, group lead- er Saarland University, deputy director Material Engineering Center Saarland, director on the IMS board of directors, d.britz@mec-s.de . Campus D3.3, 66123 Saarbruecken, Germany, www.fuwe. uni-saarland.de, www.mec-s.de . References 1. D. Sculley, et al., Hidden Technical Debt in Machine Learning Systems, Advances in Neural Information Pro- cessing Systems, January, 2503–2511, 2015. 2. S.M. Azimi, et al., Advanced Steel Microstructural Classification by Deep Learning Methods, Scientific Reports, 8 (1), 1–14, 2018, https://doi.org/10.1038/ s41598-018-20037-5. 3. B.L. DeCost, et al., High Throughput QuantitativeMetallography for Complex Microstructures using Deep Learning: A Case Study in Ultrahigh Carbon Steel, Microscopy and Microanalysis, 25 (1), 21–29, 2019, https://doi.org/10.1017/ S1431927618015635. 4. M. Müller, et al., Classification of Bainitic Structures using Textural Parameters and Machine Learning Techniques, Metals, 630 (10), 1–19, 2020, https://doi.org/10.3390/met10050630. 5. Z.L. Wang, and Y. Adachi, Property Prediction and Properties- to-Microstructure Inverse Analysis of Steels by a Machine-Learning Approach, Materials Science and Engineering A, 744 (June), 661–670, 2019, https://doi. org/10.1016/j.msea.2018.12.049. 6. Y. Liu, et al., Machine Learning in Materials Genome Initiative: A Review, Journal of Materials Science and Tech- nology, 57, 113–122, 2020, https://doi. org/10.1016/j.jmst.2020.01.067. 7. E.A. Holm, et al., Overview: Com- puter Vision and Machine Learning for Microstructural Characterization and Analysis, Metallurgical and Materials Transactions A: Physical Metallurgy and Materials Science, 51 (12), 5985–5999, 2020, https://doi.org/10.1007/s11661- 020-06008-4. 8. M. Ge, et al., Deep Learning Analysis on Microscopic Imaging in Materials Science, Materials Today Nano, 100087, 2020, https://doi.org/https ://doi.org/ 10.1016/j.mtnano.2020.100087. 9. H.K.D.H. Bhadeshia, Bainite in Steels (3rd ed.), Maney Publishing, 2015. 10. L. Fielding, The Bainite Controversy, Materials Science and Technology, 29 (4), 383–399, 2013, https://doi.org/10.1179/ 1743284712Y.0000000157. 11. D. Britz, et al., Identifying and Quantifying Microstructures in Low- alloyed Steels: A Correlative Approach, Metallurgia Italiana, 109 (3), 5–10, 2017. 12. D. Britz, J. Webel, and J. Gola, A Correlative Approach to Capture and Quantify Substructures by Means of Image Registration, Practical Metal- lography, 54 (10), 685–696, 2017. 13. J. Gola, et al., Objective Micro- structure Classification by Support Vector Machine (SVM) using a Com- bination of Morphological Parameters and Textural Features for Low Carbon Steels, Computational Materials Science, 160 (January),186–196,2019,https://doi. org/10.1016/j.commatsci.2019.01.006. 14. S. Zajac, V. Schwinn, and K.H. Tacke, Characterisation and Quantification of Complex Bainitic Microstructures in High and Ultra-High Strength Linepipe Steels, Materials Science Forum, 500–501, 387–394, 2005, https://doi. org /10.4028 /www.scientific.net / MSF.500-501.387. 15. A.R. Kitahara, and E.A. Holm, Microstructure Cluster Analysis with Transfer Learning and Unsupervised Learning, Integrating Materials and ManufacturingInnovation,7 (3),148–156, 2018, https://doi.org/10.1007/s40192- 018-0116-9.

RkJQdWJsaXNoZXIy MjA4MTAy