Energy dissipation-based fatigue damage detection in beam-like structures under moving load using FNN and 1D-CNN

This study proposes a method for localizing and quantifying fatigue damage in beam-like structures by utilizing energy dissipation as a sensitive feature. A scheme was developed to generate complete responses of the beam under random moving loads by combining individual responses from different moving load velocities, simulating real-world conditions and augmenting the training data set. The input load is assumed to be an approximate form of white noise. The responses were collected from three different beams using accelerometers evenly distributed along the length of beams. The Dissipation Spectrum (DS) was then calculated from the generated random response. The proposed approach is a two-step process: first, a Feedforward Neural Network (FNN) analyzes the correlation coefficients between DS at multiple locations on the beam, achieving 96-100% accuracy in damage localization; second, a 1D Convolutional Neural Network (1D-CNN) quantifies the damage level using the DS specifically from the localized damage, yielding an 80-100% accuracy in severity estimation. This two-step process, using correlations of DS for localization and local DS for quantification, has established a dependable method for structural health monitoring, offering a promising approach for engineering applications, particularly those requiring high sensitivity to early-stage fatigue damage.