A data-driven approach for the real-time modeling of spot-welded patches under impact using Deep Learning

Solving large structural problems with complex localized behaviour is extremely challenging. To address this difficulty, both intrusive as well as non-intrusive multiscale methods have been developed in the past. To reduce the computational time further, we propose the use of Model Order Reduction (MOR) of local scale. This paper presents the MOR technique based on a novel physics-guided architecture (PGA) of neural networks, incorporating physical variables into the architecture. The proposed approach is illustrated in the case of spot-welded plates undergoing large deformation.