AI-driven automated and integrated structural health monitoring under environmental and operational variations

Abstract

An automated framework for structural health monitoring is presented in this paper, encompassing modal identification, health monitoring, and damage localization while accounting for environmental and operational variations. The proposed framework automates the modal identification process using covariance-driven stochastic subspace identification, coupled with a Gaussian mixture model clustering approach for automatic pole selection. It further integrates autoencoder neural network and proposed thresholding process for ongoing health monitoring. For the automated damage localization step, a pattern recognition–based method is proposed that integrates the decomposition capabilities of advanced signal processing techniques, such as discrete wavelet transforms, with the learning capabilities of long short-term memory models, designed to minimize false positives and enable precise identification of stiffness loss zones. Experimental validation on a laboratory bridge structure subjected to simulated damage scenarios demonstrates the framework’s effectiveness. Designed with a user-friendly interface, the system eliminates the need for manual intervention and facilitates infrastructure health monitoring.