Wind Turbine Gearbox Fault Detection Based on Statistical Learning
Abstract
In recent years, the energy landscape has shifted towards sustainable and renewable power sources, with wind energy playing a significant role in global installations. However, like any other complex machinery, wind turbines are subject to wear and tear, leading to operational inefficiencies, downtime, and increased maintenance costs. To address these challenges, a systematic framework for early warning(s) for wind turbine gearboxes is discussed to identify premature failures. The framework includes an adaptable cutoff point algorithm utilizing single-variable time series variables from the gearbox and different machine-learning algorithms. The front bearing temperature of the gearbox is chosen as the univariate time series to generate the gearbox health status. The results show superiority for fault detection using a tree-based random forest algorithm with the highest accuracy of 86%, followed by 82% using Bayesian classification and Neural network along with a good precision value.