Detecting Structural Deterioration: Investigating Changes in Power Spectral Density Using Deep Learning on Damaged Steel Beam Structures

The article focuses on detecting structural deterioration in damaged steel beam structures by investigating changes in power spectral density (PSD) using deep learning. To simulate damage, cracks are introduced to alter the stiffness of the steel beams. The study aims to replicate a realistic traffic scenario over bridges by measuring vibration signals obtained from acceleration sensors distributed along the steel beams. The article proposes a new parameter that tracks the deterioration of structures by analyzing the PSD when a moving load is applied to the steel beams with defects. Features generated from modified forms of the PSD are used to identify structural deterioration via steel beam damage and deep learning in a training dataset. The study found that differences in PSD shape caused by damage are more effective in detecting damage in various beam structures than those in the value of the fundamental beam frequency. Although the PSD method has been utilized in earlier research to identify steel beam defects, the use of deep learning in this study offers numerous novel and advantageous benefits.