Predictive framework for detecting main bearing faults in direct-drive horizontal-axis wind turbines using supervisory control and data acquisition

Main bearing failures are among the most critical and costly faults in wind turbines due to prolonged downtime and complex replacement procedures, particularly in direct-drive configurations where the main bearing and the generator bearing are functionally the same component. Unlike traditional geared turbines, which use a separate, high-speed generator bearing, the direct-drive architecture relies on a single, large, integrated bearing. To reduce reliance on expensive vibration monitoring, this study introduces a SCADA-based predictive framework for early fault detection. After data cleaning, six predictors of generator bearing temperature were identified using Pearson correlation. Regression models—Linear, Stepwise Linear, Fine Tree, and Ensemble Bagged Trees—were tested in MATLAB’s Regression Learner. Healthy turbines showed stable thermal behavior within prediction bounds, while faulty ones displayed persistent deviations beyond ±2 standard deviations, signaling degradation. A condition-based maintenance plan was developed using three weekly indicators combined into a percentile-based monthly health score. This scoring method highlights turbines with sustained abnormal behavior, even when temperature fault records alone do not. The framework demonstrates that SCADA data can reliably detect main bearing faults, enabling turbine health ranking and more effective maintenance prioritization.