Fuzzy reliability evaluation and machine learning-based fault prediction of wind turbines

Abstract

The swift growth of the wind power industry necessitates comprehensive evaluation and efficient fault prediction of wind turbines. Given the challenges of integration and optimization of reliability evaluation and fault prediction models, a systematic method of reliability fuzzy evaluation and fault prediction based on the Supervisory Control and Data Acquisition (SCADA) data is proposed. A mid-to-long-term reliability fuzzy evaluation model is constructed using Fuzzy Comprehensive Evaluation (FCE). The mid-term evaluation results in ten failure modes reveal that the model's hazard ranking results match the situation better than the RPN method. And the long-term evaluation results of 5 years in the operating mode show that the model effectively gathers the evaluation information each year and provides a clear and accurate reflection of reliability. Meanwhile, fault prediction is studied using alarm logs because they are better at expressing the status of wind turbines than monitoring data. And the tree-based algorithms and unsupervised statistical learning methods are used to mine the mapping relationship between input variables and predefined tags. The fault prediction achieves both accuracy and recall of 0.784 and saves over 163k Euros based on local wind turbine maintenance expenditures. Overall, the reliability evaluation and fault prediction complement each other, which may either affect future wind farm management or prevent unnecessary maintenance costs.