Structural health monitoring of bridge spans using Moment Cumulative Functions of Power Spectral Density (MCF-PSD) and deep learning

Abstract

This article proposes a new parameter in evaluating mechanical behaviors of defected bridge spans. It is Moment Cumulative Function of Power Spectral Density (MCF-PSD) based on changes in shape of power spectrum and trained via cumulative function of spectral moment value by deep learning model. This new parameter allows evaluating stiffness attenuation along time, thereby helps to forecast the workability of bridge span. It can identify risky positions in not only a bridge span but also various spans of the same bridge, which proves its sensitivity to the structure’s behavior change over time. This study reveals that training MCF-PSD using cumulative function algorithm has gained outstanding results in comparison with previous studies in structural quality assessment. Therefore, it fulfills criteria of evaluating the damage level in a structure and also fosters new development of defect diagnosis and forecast. Conclusions from this study show that the change of this function is the basis to evaluate difference among measurement positions in the same span or among different spans of the same bridge and behaviors at different positions in the same span. Therefore, MCF-PSD is more sensitive than other parameters in evaluating the structure’s stiffness attenuation.