Intelligent damage classification for tensile membrane structure based on continuous wavelet transform and improved ResNet50

Abstract

Traditional structural dynamic detection methods are difficult to accurately identify damage features in vibration signals from tensile membrane structures, thus a damage classification method for membrane materials based on continuous wavelet transform and an improved ResNet50 is proposed. This damage classification method takes ResNet50 as the backbone, and ResNet50 is improved by embedding the convolutional block attention modules and parameter-transfer learning. Firstly, the planar tensile membrane structure vibration test bench is built, and vibration acceleration signals of four damaged membrane materials under three vibration excitations are collected. Secondly, continuous wavelet transform is applied to perform time–frequency conversion on the raw signals. Finally, the proposed method is used for time–frequency image classifications, and compares with other mainstream networks. The feature mappings are discussed based on Grad-CAM. The results show that the proposed method can achieve accurate damage classification, and the precisions on three custom datasets are respectively 99.67%, 99.74%, and 97.20%.