Damage detection in Glass Fiber Reinforced Plastic (GFRP) using neural network based on denoising with different mother wavelets

In this paper, a vibration-based damage detection approach for multi-layered woven glass laminate using time signal processing and Neural Network (NN) is presented. In order to reduce noise in the experimental extracted signals, wavelet-based denoising has been applied. After data mining and feature extraction from processed signals, NN as a classifier is employed to detect the damaged GFRP. Different NN structures were tested in order to enhance the damage detection performance to recognize the most remarkable performance. Also, the performance of the presented method was evaluated when different mother of wavelets at different decomposition levels denoise signals so that the best signal processing method is selected. The results demonstrate the effect of NN structure on the damage detection technique, which in this research the best NN performance was obtained with 75 hidden layers and allocating 80%, 10% and 10% of data to training, evaluation and testing, respectively. Furthermore, denoising using db3 and bior3.7 mother wavelets at 2nd decomposition level leads to the highest accuracy as well as suitable calculation time compared to other mother wavelets. The proposed method based on real data at the data acquisition points detects damage in composite laminate with high accuracy at reasonable calculation time, hence it can be used for condition monitoring of composite laminate either offline or online, provided that adding online data acquisition equipment.