About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Verification, Calibration, and Validation Approaches in Modeling the Mechanical Performance of Metallic Materials
|
| Presentation Title |
Bridging Machine Learning Constitutive Laws With Classical Model Form to Balance the Variance/Bias Challenge in Constitutive Law Calibration in Computational Homogenization |
| Author(s) |
Kirubel Teferra, Kaushik Bhattacharya |
| On-Site Speaker (Planned) |
Kirubel Teferra |
| Abstract Scope |
Computational homogenization of mechanical response into material constitutive laws relevant at the component scale remains the only viable method to incorporate the underlying micromechanical processes that determine material behavior into the analysis of engineering systems. Recently proposed machine learning constitutive laws increase the generalizability of classical approaches that prescribe a model form but at present, they lack the necessary extrapolation capability due to the lack of prescribed bias in the neural network model form. This work explores formulating rate-independent plasticity using classical model forms, such as Hill’s yield criteria with isotropic and kinematic hardening, while representing model parameters as neural networks. Further generalization can be incorporated through multi-surface plasticity. Training is performed using Ensemble Kalman Inversion, which is a gradient-free optimization method that can easily incorporate linear constraints, thus greatly expanding the use-case from stochastic gradient descent. This presentation formulates the proposed methodology and evaluate its performance on test cases. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
ICME, Machine Learning, Modeling and Simulation |