A reliable framework for predicting wind turbine failures utilising SCADA and Alarm data

Identifying wind turbine failures is vital for assuring the efficiency and reliability of wind farms. SCADA systems supply machine health data that can be used for this purpose. However, manually labelling SCADA data with fault details is tedious and challenging due to the unstructured characteristics of the data. Alarm systems can detect faults but frequently generate more alarms than necessary, resulting in false positives. A novel framework addresses these issues using the turbine alarm system data and a rule-based categorization approach. The framework was tested using data from an operational wind farm, where historical stoppages were precisely tagged using alarm data. The testing used a 10-fold cross-validation approach with a random forest algorithm and bagging, iterating 100 times with a base learner. The investigation showed that the proposed framework obtained a remarkable 95.8% in accurately identifying cases, with only 4.16% misclassified. This study demonstrates machine learning algorithms’ efficacy in forecasting wind turbine failures using SCADA data. Nonetheless, it emphasises the necessity for continued research and fine-tuning to optimise model performance.