Optimization of spring fatigue life prediction model for vehicle ride using hybrid multi-layer perceptron artificial neural networks

摘要

In this study, hybrid multi-layer perceptron artificial neural network (HMLP ANN) models were developed to predict the fatigue life of automotive coil springs with high accuracy based on the vertical vibrations of the vehicle and natural frequencies of the vehicle suspension system. The design and development of vehicle suspension systems involve numerous steps from conceptual design to prototyping and testing, including fatigue life evaluation and vehicle ride analysis. Optimizing HMLP ANN models will significantly simplify the design and development process, which forms the motivation of this study. Simulations were conducted on a quarter car model to extract the loading signals using the measured acceleration signals and artificial road profiles as inputs. The fatigue life was predicted based on the Coffin-Manson, Morrow, and Smith-Watson-Topper strain-life models whereas the comfort ride index was assessed according to the ISO 2631-1:1997 standard. Various HMLP ANN models were trained using the Levenberg-Marquardt backpropagation algorithm to determine the optimum architectures. The lowest mean square error (0.0117) is obtained for the Morrow HMLP ANN model with three hidden layers. The coefficient of determination values are more than 0.9559, indicating that there is good fit between the training/testing datasets and the data predicted by the optimum HMLP ANN models. These models were validated using the conservative correlation approach and there is good agreement between the targeted and predicted fatigue life values. It can be concluded that the optimum HMLP ANN models are capable of predicting the fatigue life of automotive coil springs with acceptable accuracy. (C) 2018 Elsevier Ltd. All rights reserved.