Development and evaluation of a laboratory-scale test rig prototype for full characterization of wind turbine main bearing dynamics and condition monitoring

Effective condition monitoring of the main bearing is essential for preventing unexpected wind turbine failures. However, the massive size of real-life main bearings poses challenges for laboratory monitoring. To address this, a laboratory-scale test rig prototype was developed for the full characterization of wind turbine main bearing dynamics and condition monitoring. Turbine scaling laws were employed to estimate the parameters of a full-scale wind turbine rotor on a smaller scale, enabling the accurate simulation of real-world conditions in a controlled environment. The scaled wind turbine parameters yielded a test rig with a main shaft diameter of 30 mm, a shaft speed range of 100–300 RPM, a maximum aerodynamic thrust of 1500 N. Three spherical roller bearings were used as test specimens, each artificially induced with 1 mm-deep and 1 mm-wide cracks on the inner race, outer race, and rolling elements. Vibration data were then collected from the rig for analysis. Preliminary evaluation using one-way ANOVA (F = 35.8, p = 0.00001) confirmed significant variations in key statistical characteristics of the vibration signals across bearing health states, indicating that the data collected from the rig were discriminative. A Bayesian-optimized LSTM model was then developed for main bearing fault diagnosis. Results from this investigation demonstrated that Bayesian optimization improved main bearing fault classification accuracy by an average of 3.49 %, while also reducing potential false alarms by 50 %. Additional evaluation metrics indicated a precision of 97.6 %, a recall of 97.8 %, an F1-score of 97.7 %, and an area under the curve (AUC) of 0.98. These findings underscore the effectiveness of the proposed test rig as a valuable research tool for advancing the understanding of wind turbine main bearing behaviour and condition monitoring. Moreover, insights from this research methodology may be extended to other wind turbine bearings, thereby contributing to advancements in wind turbine predictive maintenance.