A combined implicit-explicit vibration-based SHM method for damage detection of wind turbine blades

Today, the green transition is being pursued more than ever, leading to a significant increase in the installation of wind turbines and wind farms. Maintaining and servicing the wind turbines is accompanied by extensive expenses. Therefore, predictive maintenance techniques are in increasing demand, including the implementation of a well-functioning Structural Health Monitoring (SHM) system. In this work, physical and non-physical vibration features along with various techniques to mitigate the effect of Environmental and Operational Variability (EOV) are investigated. Vibration features originating from Vector Auto-Regressive (VAR) models are considered. Physics-domain DSFs comprise the natural frequency, while the non-physical DSFs comprise the VAR model coefficients. A challenging and inevitable aspect in SHM is the effect of EOV that can alter the DSFs so as damage does. Two mitigation approaches are considered, the first based on Bayesian non-linear regression -explicit- and the second based on Principal Component Analysis (PCA) -implicit-. Additionally, a novel combined approach is proposed, featuring regression of Principal Components (PCs). As some of the PCs will be EOV-insensitive, the specific corrected PCs are automatically selected using the F-statistic of individual regressions. A comparative analysis is carried out in a controlled experiment, featuring the vibration response of a lab-scale wind turbine blade under various temperature and damage conditions. The methods are assessed in terms of correctly detected damages from a one-class classifier based on the squared Mahalanobis distance on the EOV-corrected features, and best performing method combinations are thereby identified.