A Performance Metric to Evaluate Frequency-Based Damage Indicators

Abstract

Frequency-based correlation methods have been established as an autonomous tool for extracting key features of structures in Structural Health Monitoring (SHM). Although the literature on the subject is extensive and includes multiple correlation strategies, validation of most of these methods has been performed on simple structures, such as beams or plates. In contrast, their validity for application in more complex structures under more realistic damage scenarios deserves further investigation. In this work, a method called Precision-Recall curve based on the confusion matrix is proposed to objectively evaluate the performance of spectral correlation methods for structural damage detection from vibration data sets. The Precision-Recall curve is then condensed into a scalar value using the Area Under the Curve (AUC) metric. The work is based on extensive experimental lab tests that use three different structures with decreasing modal information, challenging the detection of damage scenarios. In addition, the effects of noise and frequency range are studied as key factors that can reduce or improve the performance of the indicators. The work results also validate the method based on the Complex Frequency Domain Assurance Criterion (CFDAC) previously proposed by the authors in structures with scarce modal information. Finally, experimental evidence allows conclusions to be drawn on the performance of different indicators available in the literature.