Machine Learning- Based Condition Monitoring of Wind Turbines Using SCADA Dataset for Early Fault Detection

Abstract

There has been a steady increase in the amount of wind power installed worldwide as a result of global efforts to replace fossil fuels and lower the average global temperature. The LCO of wind energy includes, among other things, the costs of operating and maintaining wind farms. Using machine learning and SCADA data, this research proposes a framework for early defect identification in wind turbines. The approach is designed for use with wind turbines. The proposed methodology contains data preprocessing methods like data cleaning, data normalization, feature selection, label encoding and SMOTE for the class imbalance issue. Processed data is then classified into fault and non-fault conditions using XGBoost and LightGBM (LGBM) models to effectively classify the data. The experimental outcomes showed that the proposed models have high fault detection performance, with accuracy of 95.2% for XGBoost, and the highest accuracy of 95.6% for LGBM, which has better values of precision, recall, and F1 score. The performance of the proposed LGBM is related to other machine learning models like NB, KNN and SVM by obtaining better classification performance with low misclassification rates. This framework could greatly aid predictive maintenance, keep turbines up and running, cut down on maintenance expenses, and improve the operational stability of wind power systems.