CrashNet: an encoder–decoder architecture to predict crash test outcomes

Abstract Destructive car crash tests are an elaborate, time-consuming, and expensive necessity of the automotive development process. Today, finite element method (FEM) simulations used to reduce costs by simulating crashes computationally. We propose CrashNet, encoder–decoder deep neural network architecture that reduces further models specific outcomes very accurately. achieve this formulating events as time series prediction enriched with a set scalar features. Traditional sequence-to-sequence usually composed convolutional (CNN) CNN transpose layers. concatenate those MLP capable learning how inject given scalars into output series. In addition, we replace 2D layers in order force model process hidden state one The proposed CrashNet can be trained efficiently is able input infer results tests. produces faster at lower cost compared destructive FEM simulations. Moreover, it represents novel approach safety management domain.