A multiaxial fatigue life analysis method for automotive components based on LSTM-CNN

Abstract

The structural components of a moving vehicle are subjected to non-stationary multi-directional excitations from irregular road profiles. Multiaxial fatigue analysis under non-stationary dynamic excitations plays a crucial role in accurately predicting the fatigue life of automotive components. A time-domain method for multiaxial fatigue analysis of automotive components under non-stationary loads is proposed based on deep neural networks. Firstly, a time–frequency domain data augmentation technique is employed to construct long-term multiaxial loads from measured load histories for eight typical roads. Subsequently, a hybrid model integrating Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) is developed to represent the nonlinear mapping from excitation to response, serving as a high-dimensional time-series prediction surrogate model. Using the surrogate model trained only on short-term load and response data, the stress and strain components at multiple critical points of the automotive structural component can be predicted quickly and accurately. Furthermore, the influences of different fatigue criteria, various road types, and varying durations of dynamic response calculations on the fatigue life prediction results are thoroughly discussed. Numerical simulation analysis of an automotive control arm demonstrates that, compared to fatigue life analysis based on transient response calculations using the finite element method, the proposed method achieves a computational efficiency improvement of 2–3 orders of magnitude, and the discrepancy in multiaxial fatigue life calculations is less than 9%.