Visual-Based Deep Convolutional Neural Network Method for Detecting Damage in Bridge Plate Rubber Bearings

Abstract

With the widespread use of plate rubber bearings in bridge construction, there is an urgent need for an objective, efficient, and labor-saving disease detection method to improve the quality and efficiency of bridge maintenance. In this study, three common bridge plate-rubber-bearing diseases are the research objects, along with normal bearings. A bridge plate-rubber-bearing disease classification model based on the VGG16 network was constructed using transfer learning methods. The study compared the performance of four deep convolutional neural network models obtained by training from scratch and transfer learning. To further enhance model performance, data augmentation was applied during transfer learning on the VGG16 model constructed in this study. The results showed that the data set enhanced with image augmentation presented an improvement in overall recognition accuracy of 18% in the fine-tuning transfer learning algorithm of the VGG16 model compared with training from scratch. This validates the effectiveness and feasibility of the proposed model and provides strong support that it improves the efficiency of identifying and classifying diseases in bridge rubber bearings and facilitates post-bridge maintenance.