Structural Damage Detection with Modal Strain Energy Using Convolution Neural Network

Abstract

Based on the excellent feature extraction and non-linear learning ability of a convolutional neural network (CNN), a structural damage detection method is proposed in this paper. When the structure is damaged, the changes of its modal parameters reflect the damage information of the structure. A simply supported beam was used and structural damage was introduced at different locations. The finite element method was used to simulate the free vibration of the beam and obtain the first-order modal strain energy for various damage scenarios. The obtained modal parameters and the damage information were used as the training samples of the neural network. A CNN was designed to detect damage (both location and level), which detected damage location with 100% accuracy and damage level with 5% relative error. Compared with a traditional Back Propagation (BP) neural network, the CNN had more advantages than the BP neural network in detecting damage location, and it was more economical in computational costs, the uptime of the CNN was about 5%–40% that of the BP neural network. It is found the CNN has excellent performance in detection of both damage locations and levels, the detection effect exceeds BP neural network, and it is more economical in computational cost than a BP neural network as it uses convolutional operation.KeywordsStructural damage detectionConvolutional neural networksModal strain energyFeature extractionSimply supported beam