Bridge damage detection using machine learning algorithms

The application of accelerated bridge construction (ABC) methods is becoming more widespread owing to their many advantages. In this construction method, prefabricated bridge elements are assembled on-site by establishing in-situ joints to minimize on-site construction time. Despite the improved life-cycle performance and cost benefits of ABC bridges, some concerns exist about the degrading environmental effects on the joints and invisible internal damages. In this study, the long-term performance of an ABC bridge that had been in service for more than 50 years was investigated utilizing machine-learning processes. Observation of reflective cracking on the deck surface and leakage through the joints in this bridge indicated some damage to the bridge longitudinal joints. Damages to the joints are not always visible, nor their extent is known. Therefore, a new damage detection approach is proposed that uses the results of a series of load tests as input in machine-learning techniques with the ultimate aim of detecting the location and severity of joint damages with a high level of certainty. The proposed approach uses the bridge responses obtained from a detailed finite element (FE) model under the assumption of various damage scenarios and predicts the potential damages using the training process of machine-learning algorithms and the actual bridge responses. The results show that the supervised learning algorithm successfully estimated the location and amount of damage in the bridge joints.