Introducing Finite Element Method Integrated Networks (FEMIN).

This paper introduces a novel computational framework, Finite Element Method Integrated Networks (FEMIN), designed to accelerate crash simulations significantly. The core innovation is that large regions of the mesh in a Finite Element Method (FEM) crash simulation are replaced by a Neural Network (NN). The NN is directly integrated into the FEM solver and aims to reduce computational time and resources while maintaining high-fidelity results. The introduced NN architectures and training strategies address the inherent challenges posed by crash simulations, such as long time series, dynamic loads, and complex, history-dependent material behaviors. Our work includes the development of load cases tailored to validate the FEMIN approach. Furthermore, we introduce a novel NN architecture, the Temporal gated Multi-Layer Perceptron (TgMLP), both as a stand-alone model and also as an initial state predictor in combination with a conditional Long Short-Term Memory (LSTM) network and compare their effectiveness in FEMIN. The results demonstrate that FEMIN, employing these NN models, can accurately predict the structural response under crash conditions while inferring the loads significantly faster than the replaced FEM mesh. This research paves the way for more efficient vehicle design, offering a promising solution for integrating machine learning into computational mechanics. [ABSTRACT FROM AUTHOR]