This study aims to establish a hybrid method combining the finite element method (FEM), the mechanical–electrical model, and a back-propagation artificial neural network (BP), to simulate the piezoresistive pavement. First, the tire-pavement FEM model with piezoresistive units was established considering the viscoelasticity of the pavement materials. Subsequently, the mechanical responses of the piezoresistive units under various tire and environmental loads were converted into electrical resistance outputs via the mechanical–electrical model. Finally, BP was trained using simulated data to address challenges associated with the back-calculation of tire loads. Results indicate that the electrical resistance of the piezoresistive unit in complete contact with the tire illustrates an overall rising trend as tire load increases, which is attributed to changes in contact stress. However, the adjacent piezoresistive units display an opposite trend, which can be used to determine the lateral position of the tires. Additionally, electrical resistance shows a non-linear decrease with increasing temperature. The single-hidden-layer BP with 13 neurons was validated to demonstrate higher accuracy compared to multi-hidden-layer BP. Moreover, the Genetic algorithm-optimized single-hidden-layer BP (GA-S-BP) shows further improved performance, achieving an MSE of 1.91 and an MAPE of 8.5%, and a low probability of underestimating tire loads. The GA-S-BP designed in this study can effectively predict tire loads within permissible levels to realize the function of piezoresistive pavement.
Read full abstract