SUMMARY
The cables are extremely important and vulnerable components in the cable-stayed bridges. Because cable tension is one of the most crucial structural health indicators, therefore, assessing the cable condition based on the cable tension is a major interest in the structural health monitoring (SHM) of the cable-stayed bridges. This paper aims to develop a deep convolutional neural network (DCNN)-based transfer learning method that is integrated with a continuous wavelet transform (CWT) for the health condition identification of the cables in a cable-stayed bridge using the one-dimensional time series cable tension data. For this purpose, the CWT is adopted to convert the cable tension to the images of a time-frequency representation. The last three new layers emerged in the pre-trained DCNN model, which is called AlexNet, as a new learning framework to use for the identification of the cable condition. The performance of the proposed DCNN model is examined using several statistical measures that include accuracy, sensitivity, specificity, precision, recall, and the F-measure. The results show that the proposed DCNN model gives superior accuracy (100%) for the identification of the undamaged cables and the damaged cables based on the cable tension data.