Research on Application of Backpropagation Neural Network in Damage Detection of the Refined Plate Model

Abstract

Artificial intelligence has been widely used in engineering. In this paper, we propose to combine the backpropagation neural network (BPNN) with the refined plate model based on Carrera Unified Formula (CUF) to advance the development of damage detection. The prediction model is built by utilizing the error back propagation function of the neural network. In addition, MATLAB uses Taylor’s interpolation algorithm and lower degrees of freedom yet achieves the same accuracy as ANSYS, and the improved plate model accurately reproduces the mechanical properties of the metal plate. A database is then built based on the mechanical model to detect the location of damaged elements and node displacements. The nodal displacements were used as inputs while the locations of damaged elements were used as training outputs for the neural network. The effectiveness of the proposed method was verified through various damage scenarios. The results show that the method can accurately predict individual damage locations based on node displacements alone. The neural network combined with the plate model achieved a detection accuracy of 91% with a regression coefficient of 0.95.