Damage detection on rectangular laminated composite plates using wavelet based convolutional neural network technique

Abstract

Listen

Translate article

In this study, a novel method called the Wavelet Transform-based Convolutional Neural Network (WT-CNN) technique is proposed for damage detection of rectangular laminated composite plates (RLCPs). In the proposed method, the convolutional neural networks and two-dimensional wavelet transform are combined to detect the location of damages in RLCPs. The finite element model (FEM) of damaged RLCPs is developed to generate two-dimensional signals for feeding in the wavelet transform. In order to form a dataset, five hundred single-damage scenarios are applied on a typical RLCP and detected using optimal mother wavelet functions and vanishing moments. The correlation between the class of signals and their best (optimal) wavelet function is considered as the criterion for the best (optimal) wavelet selection. WT-CNN is trained to detect the location of damages in RLCPs. Results show that the proposed WT-CNN can predict and detect the location of damages in RLCPs with high accuracy and eliminate problems of trial and error simulations for future input signals of damaged RLCPs.