Bridge scour detection method based on Siamese neural networks under bridge-vehicle-wave interaction

Abstract

Bridge scour is one of the predominant factors in the failure of bridges located over waterways. Therefore, a novel bridge scour detection method is proposed in this study. Specifically, a three-dimensional bridge-vehicle-wave interaction model is developed to obtain vibration simulation data under multiple scour damage scenarios. The real bridge vibration signal and the model simulation data are then fed into the Siamese neural network. The Siamese neural network can be utilized to match the scour damage features of both and measure their similarity to capture the structural changes of the real bridge caused by scour damage. The effectiveness of the proposed method can be verified by the field test case, in which the prediction accuracy of the scouring location reaches 81.11%, and the prediction accuracy of the scouring damage level reaches 78.06%. Furthermore, the influence of vehicle speed and transducer arrangement on bridge scour detection accuracy is discussed. The results show that the proposed method has the potential to adequately extract and transfer the knowledge of scour patterns from the numerical model to detect the scour damage of real bridges.