Presentations

Abstract

Anticipating failures in major wind turbine components is essential for reducing maintenance costs and ensuring wind farm safety. It enables optimized maintenance scheduling, timely ordering of spare parts, and minimizes turbine downtime and production losses. Proactively stopping a turbine can also prevent severe consequences from component failures during operation. However, identifying problematic assets within a fleet of hundreds or thousands of turbines presents a significant challenge. Leveraging readily available SCADA data, turbine component monitoring has proven effective over the past decade in predicting mechanical failures.  In this study, we introduce models for monitoring two critical failure modes of generator components: thermal fatigue of the windings and wear of the slip ring brushes. Both approaches rely on physics-based models to estimate damage accumulation during turbine operation. The inputs to the models are readily available 10-minute statistics from SCADA systems, ensuring wide applicability. A key advantage of these models is their lightweight, "out-of-the-box" nature, allowing seamless integration into existing monitoring frameworks. Model constants are initialized using standard values from the literature but can be adjusted for specific turbine models or wind farms when sufficient validation data are available.  The thermal fatigue model assesses damage to the windings' insulation caused by thermal cycles. It uses generator temperature, the endurance curve of the insulation class, and Miner's rule to compute cumulative damage. Applied to data from two wind farms, the model's predictions were validated against historical failure records. Results show that turbines with failures exhibited higher fatigue accumulation than others. Notably, a few high-temperature events in the months preceding failure accounted for the majority of the damage, while bulk temperature distributions remained unaffected. This highlights the model’s ability to detect early failure indicators that statistical methods might overlook.  The slip ring model applies tribology principles to estimate brush wear resulting from contact with the rotating shaft. The primary input is generator speed, which determines the total sliding distance, with slip ring temperature used to adjust the wear coefficient. Applied to the same wind farm data, the model showed excellent agreement with maintenance records, with 90\% of turbines predicted to have higher wear aligning with reported cases of abnormal degradation. Further SCADA analysis linked increased wear rates to elevated slip ring operating temperatures, suggesting potential contact defects between the brush and shaft.  These models provide aggregated metrics for assessing component health, enabling the automatic flagging of problematic assets through threshold-based alarms. Their lightweight implementation, reliance on readily available data, and strong correlation with failure data make them practical and accurate near-real-time monitoring solutions.