Machine-learning based prediction of crash response of tubular structures

This paper proposes a machine learning based methodology for predicting the buckling response of tubular structures. An extensive dataset force-time curves is generated using calibrated finite element model within parametric space where highly non-linear. Based on fully connected neural network template, hyper-parameters are determined and resulting evaluated separate test set, with regard to maximum average load energy absorption errors. evaluation shows non-random error distribution which can be correlated physical properties structural collapse. To validate this assumption, similar analysis conducted between simulations varying geometric imperfections. Evaluation imperfection sensitivity reveals comparison individual that errors made by have interpretation. These results indicate proposed approach capable crushing level accuracy comparable would caused minor change in imperfection.