Dynamic behavior prediction of modules in crushing via FEA-DNN technique for durable battery-pack system design

The structural integrity and crashworthiness of the battery-pack system (BPS) in electric vehicles are an emerging concern of engineers. Therefore, corresponding numerical and experimental investigations have to be carried out. Engineers need to select appropriate thicknesses and materials of main components through multiple finite element analysis (FEA), e.g., upper enclosure and bottom shell. This process is laborious and time-consuming. In this paper, a rapid stress prediction method is proposed to help select components' thicknesses and materials under crush scenarios. This method is based on historical FEA data and a deep neural network (DNN) algorithm. First, a nonlinear FE model of a BPS that includes battery modules is developed. The FE model is verified via mesh-sensitivity analysis and modal test results. The crush simulations are performed and the FEA data are collected. Second, a DNN framework with forwarding and backward propagations is used to train the FEA data. Therefore, a DNN model that can describe the relationship between the inputs (thicknesses and materials of related components) and outputs (maximum von Mises stresses of modules) is established The established DNN model can effectively predict the modules' stresses. The accuracy of the DNN model is investigated in terms of error functions. Furthermore, the second-order response surface model, third-order response surface model, and radial basis function neural network model are used to demonstrate the advantages of the DNN model. The proposed crushing behavior prediction method, which can be used in the design of safe and durable BPS, is proven efficient and accurate.

Automated summary

In this paper, a rapid stress prediction method based on a deep neural network algorithm is proposed to select the thicknesses and materials of electric vehicle battery-pack system (BPS) under crush scenarios. By establishing a nonlinear FE model and training historical data, the method accurately predicts the stresses of the modules, demonstrating high efficiency in designing safe and durable BPS.