Modified Stockwell Transform Based Generative Adversarial Network Model for Estimating Temporal Frequency Variation of Vehicle-Bridge Interaction System

Abstract

Massive and directional railroad vehicles causes temporal frequency variations in the fundamental frequencies of the vehicle-bridge interaction (VBI) system. As efforts to investigate complex VBI systems, conventional structural inspection methods are struggling since the achieved solutions are mainly focused on the simple VBI system. Thus, this thesis proposes a generative adversarial network based temporal frequency estimation framework to address those limitations. As the conventional methods, analytical eigenvalue approach, numerical substitution approach, and wave propagation approach which is based on the Fourier series approximations are also studied. Moving on to the dynamic analysis, modified stockwell transform (MST), one of the time-frequency analysis method to comprehend non-stationary signals is introduced and optimal parameters are selected. Then, framework with paired dataset of MST data and numerical substitution data is proposed. Bridge acceleration data achieved from the dynamic analysis are utilized as an input to the proposed framework, and framework is trained with different hyperparameters to achieve best results. The framework’s validation involves quantitative assessments and loboratory scale experiments, illustrating its robust performance across diverse scenarios and demonstrating potential applicability for evaluating bridge conditions under the operational settings.