Efficient Multiscale Composite Modeling via an Embedded Long Short Term Memory Surrogate Microscale Model
Artificial Intelligence (AI) is becoming more and more a part of the activities traditionally covered by the engineering analysis and simulation community. Recent advances in the application of AI, machine learning (deep learning) and predictive analytics, have brought these technologies to the fore in every area of industry.
This seminar hosted by the NAFEMS Americas Steering Committee brought together speakers from the end-user, consultancy, and academic industries to discuss where we are and how these technologies are being used to advance significantly the engineering analysis and simulation capabilities and approaches over the next 10 years.
A neural network was trained as a computationally efficient surrogate for a physics-based micromechanics model and embedded within classical lamination theory. The microscale model used generalized method of cells to predict the transversely isotropic homogenized stiffness of composite plies during loading as damage progressed. The surrogate model was able to learn a latent material damage representation to capture loading history and path dependence using long short-term memory layers in the neural network architecture. When the surrogate was embedded within a macroscale classical lamination theory model, the efficient model was able to make predictions with a 29 times speed increase compared to generalized method of cells and 145 times faster than high fidelity generalized method of cells. This speedup came without almost no loss in accuracy. The surrogate predicted almost identical laminate response as the physics-based multiscale model with a mean absolute error of 10.5 MPa and and R-squared of 0.98.