A comparison of Feature Extraction Methods for Crack and Ice Monitoring in Wind Turbine Blades: System Identification and Matrix Decomposition

The global trend of increasing the relevance of renewable energy in the power grid is likely to remain in the following years. As wind power is a relevant renewable energy source, wind capacity generation has increased considerably. Consequently, the costs of maintenance of wind turbines increase as well. Therefore, the development of Structural Health Monitoring (SHM) systems is important since they can detect defects as early as possible, reduce the wind turbine’s downtime, and maintain the efficiency of the generation sites, which reduces losses. In this work, we use a public wind turbine blades benchmarking dataset to build a model that predicts and classifies fault scenarios. The dataset presents cases considering various fault events under different climate conditions. First, it is analyzed which of the sensors best captures the dynamics difference when there is a fail presence on the wind turbine blade. Then, we extract features from the sensor signal through Principal Component Analysis (PCA) and system identification, an Auto-Regressive Moving-Average (ARMAX) model. After, the addition of a fault classification module that uses Machine Learning classification algorithms completes the development of the SHM systems.