Enhanced Fault Detection of Wind Turbine Using eXtreme Gradient Boosting Technique Based on Nonstationary Vibration Analysis

Wind turbines serve a vital role in renewable energy generation but operate in harsh environments and endure variable loading. Monitoring wind turbine blade conditions is therefore critical to prevent unscheduled downtime and revenue losses. This research investigates the application of machine learning techniques for detection, monitoring, and diagnosis of wind turbine blade faults, specifically blade twist, surface erosion, and tip damage, using vibration data. A discrete wavelet transform was implemented to analyze the non-stationary vibration signals acquired from instrumented turbine blades. Salient features were extracted from the transformed signals, and the ReliefF algorithm utilized for selection of the most discriminative features for effective classification. These features were input into K-nearest neighbor, support vector machine, and eXtreme gradient boosting (XGBoost) classifiers to categorize turbine blade state into four conditions: healthy, cracked, eroded, and twisted. Experimental results revealed the XGBoost model in conjunction with ReliefF feature selection achieved 99.4% accuracy, outperforming the other techniques. These outcomes highlight the efficacy of the proposed approach for accurate wind turbine blade fault classification based on vibration signatures. This research contributes novel insights into the application of machine learning for enhanced monitoring and maintenance of wind turbines. The demonstrated methodology shows significant potential for improving wind energy system reliability through early fault detection and predictive maintenance strategies.