Vibration-based Damage Detection in Bridges via Machine Learning

Environmental corrosion and external loads degrade the performance of a bridge over the course of its service life. Although dynamic fingerprints are damage-sensitive, they are rarely applied to bridges in-situ due to environmental noise. Machine learning techniques can facilitate effective structural damage detection. This paper proposes a detection method based on dynamic fingerprints and machine learning techniques for multi-damage problems in bridges. Vibration analysis is conducted to acquire the dynamic fingerprints, then the Bayesian fusion is used to integrate these features and preliminarily locate the damage. The RSNB method, which combines Rough Set theory and the Naive-Bayes classifier, is introduced as a robust classification tool for damage qualification. A continuous bridge is numerically simulated to validate the effectiveness of the proposed method. The RSNB method is compared with back propagation neural network, support vector machine, and decision tree techniques, it is found that the RSNB outperforms other three methods in terms of transparency, accuracy, efficiency, noise robustness, and stability.