Friday November 3rd, 2023 Time: 3:00-4:00 pm
This seminar will be held in ASE 1.126
Professor Kaan Inal, University of Waterloo, ON, Canada
Applications of artificial intelligence/machine learning in computational mechanics is a new and emerging field. This work presents various machine learning frameworks with applications tailored around finite strain plasticity problems. In the first part of the work, a convolutional neural network-based framework (CNNs) is coupled with a rate-dependant crystal plasticity finite element method (CPFEM) formulation. The coupled approach is then used to successfully predict the stress-strain behaviour, texture evolution and localized deformation during proportional and non-proportional strain paths. This new approach presents significant computational efficiency compared to the classical crystal plasticity finite element model (10,000 times faster). In the second part of the work, an advanced, recurrent deep-learning framework capable of predicting the entire simulated stamping response for new geometries, materials, and process parameters is presented. The predictions with the new framework are compared to finite element model simulations, demonstrating a significant reduction in prediction time while maintaining high accuracy in key performance indicators. Finally, the work presented in this talk highlights how these frameworks may be leveraged to improve the formability of structural components further.