Structural health monitoring of railway truss bridge under moving train load using decision tree models and residual neural networks

Abstract

This paper introduces an efficient machine learning-based structural health monitoring method for railway truss bridges, addressing the time-consuming and error-prone nature of traditional approaches. By utilizing measured vibration responses under train load, the technique employs wavelets, Fourier transforms, and spectrograms to extract damage-induced changes in signals for training machine learning models. Given the slow and impractical data collection from real-world bridges, the paper proposes generating data from a numerical model, onto which a moving train load is applied. The acceleration time history responses from nodes are recorded for various damage cases, forming the dataset. Decision trees and Residual Neural Networks are trained on this data, demonstrating accurate classification of damaged members. Despite the effectiveness, human interpretation remains necessary for structural health monitoring, emphasizing these models as tools to enhance efficiency and reduce human errors in the monitoring process.