About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
Fatigue in Materials: Fundamentals, Multiscale Characterizations and Computational Modeling
|
Presentation Title |
Interpretable Machine Learning for the Prediction of Crack Initiation in Additively Manufactured Inconel 718 |
Author(s) |
Jonas Merrell, Krzysztof Stopka, Michael Sangid, Jacob Hochhalter |
On-Site Speaker (Planned) |
Jonas Merrell |
Abstract Scope |
Additive manufacturing (AM) allows for the creation of highly complex shapes but often produces microstructures and defects not commonly found with traditional manufacturing techniques. These characteristic defects, such as porosity, often become crack initiation sites and reduce the number of cycles to crack initiation and component life. To model this mechanism, we compute the accumulated plastic strain energy density (Wp) as a failure metric using crystal plasticity finite element models (CPFEM), first with defect-free AM Inconel 718 microstructures. Subsequently, an inherently interpretable machine learning model is trained from the CPFEM Wp data using genetic programming based symbolic regression (GPSR). Once trained, the GPSR model can be used to predict crack initiation sites and the interpretable nature provides direct insights into governing microstructure features. Finally, GPSR models trained from microstructures with defects provides interpretable models for the relative effect of defects. |
Proceedings Inclusion? |
Planned: |
Keywords |
Computational Materials Science & Engineering, Machine Learning, Modeling and Simulation |