Vibration-based SHM of a laboratory-scale wind turbine blade

Abstract

The necessity of the green transition is today more extensive than ever before. Consequently, wind turbines along with wind farms are compelled to grow rapidly to encounter the obligation of a sustainable source of energy. Facing extensive expenses associated with maintenance, the exploration of a well-functional Structural Health Monitoring (SHM) system is crucial. Hence, by utilizing data from a laboratory-scale wind turbine blade in a controlled environment, numerous Damage Sensitive Features (DSFs) along with several techniques to mitigate the effect of Environmental and Operational Variability (EOV) are investigated. In this work, we consider DSFs originating from Vector Auto-Regressive (VAR) models, including natural frequencies and damping ratios as a physical quantity, and the parameters of the models as a non-physical quantity. The inevitable obscuring of the DSFs caused by EOV is mitigated by utilizing two distinct approaches. Bayesian non-linear regression models and the Principal Component Analysis (PCA) methodology, designated as an explicit and implicit approach respectively. In addition, a novel approach combining the implicit and explicit methodology is proposed. Through a comparative analysis, various considerations of interest were derived. The applicability of the various DSFs was measured in terms of correctly detected damages utilizing the multivariate squared Mahalanobis distance as a one-class classifier. The derivation was that the usage of the non-physical quantity did show prominent applicability. Generally, the combined implicit-explicit methodology showed superior efficiency. Despite the acknowledgement regarding complicating aspects entering an uncontrolled environment, the conviction of efficiency is preserved yet reduced to some extent.