Damage detection for high-speed railway standard box girders based on time–frequency characteristics of train-induced strain

This paper presents a novel approach for detecting damage in high-speed railway standard box girders by leveraging the time–frequency characteristics of train-induced strain. Based on the mechanical and deformation characteristics of high-speed railway box girders, the method involves segmenting the box girder into distinct components based on plate element analysis to identify their damage separately. It utilizes coefficients derived from wavelet transforms as indicators sensitive to damage, and employs the particle swarm optimization algorithm (PSO) to determine the ideal frequency intervals in terms of number and position. Within these optimal frequency intervals, the sum of wavelet coefficients is only sensitive to damage features while remaining unaffected by environmental and operational variations. A convolutional denoising autoencoder is employed to remove noise from the strain time–frequency image, improving the robustness of the proposed method. By utilizing the Gaussian inverse cumulative distribution function to estimate confidence boundary (CB) for damage features (DF), outliers are detected, allowing for precise damage localization and quantification. A case study for the high-speed railway box girder show that the proposed method can effectively identify, locate and quantify damage across all components in the critical key section by using data acquired from just 4 strain sensors. This method holds promise for facilitating the development of effective maintenance strategies for high-speed railway box girders.