Unsupervised Ambient Vibration-Based Feature Extraction for Structural Damage Detection

Abstract

Structural damage detection (SDD) is one of the major components of the Structural health monitoring (SHM) which essentially indicates the sustainability of industries, buildings, and bridges. Manual field surveys have traditionally been used to evaluate gaps. However, this approach is typically unreliable, time-consuming, and dangerous to surveyors. To overcome these challenges, researchers have been developing automated structural damage methods which include data collection, processing, structural damage identification and diagnosis. Deep learning (DL), a subcategory of ML approaches based on multiple neural networks, has advanced rapidly in recent years. Unlike standard ML approaches, DL does not necessitate a preset feature-based stage and, by providing additional data, may train more broad and robust models. As a consequence of the success of ML approaches, several specific goals in mind in ML-based fracture detection have been undertaken. This study aims to provide a solution using artificial intelligence-based damage detection research in order to guide the SHM. The proposed model achieved 98% accuracy with better testing results on unseen datasets.