Classification of Damage on Wind Turbine Blades Using Automatic Machine Learning and Pressure Coefficient

Abstract

Wind turbine blades (WTB) are critical components of wind energy systems. Operating in harsh environments WTBs face significant challenges, since damage to their leading edge caused by erosion or additive surface roughness can reduce performance, and increase maintenance costs and operational downtime. One approach to detect WTB damage is to use machine learning, but properly designing a predictive system is not trivial. Auto machine learning (AutoML) can be used to simplify the design and implementation of machine learning pipelines. This work presents the first comparison of state-of-the-art AutoML methods, Auto-Sklearn, H2O-DAI and TPOT, to detect erosion and additive roughness in WTBs. The Leading-Edge Erosion Study database is used, which provides measurements of the pressure coefficient along the airfoil under different conditions. This is the first work to combine the pressure coefficient and AutoML systems to detect these types of damage. Results show the viability of using AutoML in this task, with H2O-DAI producing the best results, achieving an accuracy above $90\%$ in many cases. However, statistical analysis shows that a standard classifier can achieve similar performance across all problems considered, based on the Friedman test and the Wilcoxon-Holm post hoc analysis with an $\alpha =0.05$ significance level. However, AutoML systems perform better as the complexity and difficulty of the problem increases.