Best features extraction for ML-based structural health monitoring

Extending the lifespan of structures for sustainability and economic reasons is promoting the development of automated Structural Health Monitoring (SHM) systems. By integrating data-driven strategies with machine learning techniques, SHM can flexibly assess structural conditions using Damage Sensitive Features (DSFs) extracted from structural response data, bypassing the need for complex structural models. This approach, relying on historical sensor data for comparison, is suitable for real-world applications. A key challenge is that DSFs can be affected by Environmental and Operational Variabilities (EOVs), which can obscure damage indicators. This study focuses on using signal processing and clustering analysis for automated health assessment. It introduces a method to extract DSFs, ensuring they are both sensitive to damage and robust against EOVs. The method's effectiveness is demonstrated on a finite element model bridge structure under various damage scenarios and vehicular loads, validating the robustness of the extracted features.