Hybrid weighted sequential learning technique for structural health monitoring using learning approaches

Abstract

Structural Health Monitoring (SHM) plays a vital role in damage detection, offering significant maintenance and failure prevention benefits. Establishing effective SHM systems for damage identification (DI) traditionally requires extensive experimental datasets collected under varied operating and environmental conditions, which can be resource-intensive. This study introduces a novel approach to SHM by leveraging a Hybrid Weighted Sequential Learning Technique (HWSLT) classifier, which uses Finite Element (FE) computed responses to simulate structural behaviors under both healthy and damaged states. Initially, an optimal FE model representing a healthy, benchmark linear beam structure is developed and updated using experimental validation data. The HWSLT classifier is trained on SHM vibration data generated from this model under simulated load cases with uncertainty. This allows for minimal real-world experimental intervention while ensuring robust damage detection. Results demonstrate that the HWSLT classifier, trained with optimal FE model data, achieves high accuracy in predicting damage states in the benchmark structure, even when mixed with random disturbances. Conversely, data from non-ideal FE models produced unreliable classifications, underscoring the importance of model accuracy. These findings suggest that the integration of ideal FE models and deep learning offers a promising pathway for future SHM applications, with potential for reduced experimental costs and enhanced damage localization capabilities.