This study applies statistical process control and machine learning techniques to diagnose wind turbine faults and predict maintenance needs by analyzing 2.8 million sensor data collected from 31 wind turbines from 2015 to 2017 in Taiwan. Unlike previous studies that only relied on historical wind turbine data, this study analyzed the sensor data with practitioners' insight by incorporating maintenance check list items into the data mining processes. We used Pareto analyses, scatter plots, and the cause and effect diagram to cluster and classify the failure types of wind turbines. In addition, control charts were used to establish a monitoring mechanism to track whether operation data are deviated from the controls (i.e., standard deviations) as a mean to detect wind turbine abnormalities. While statistical process control was applied to fault diagnosis, machine learning algorithms were used to predict maintenance needs of wind turbines. First, the density-based spatial clustering of applications with noise algorithm was used to classify abnormal-state wind turbine data from normal-state data. Then, random forest and decision tree algorithms were employed to construct the predictive models for wind turbine anomalies and tested with K-fold cross-validation. The results indicate a high level of accuracy: 92.68% for the decision tree model, and 91.98% for the random forest model. The study demonstrates that, by data mining and modeling, the failures of wind turbines can be detected, and the maintenance needs of parts can be predicted. Model results may provide technicians early warnings, improve equipment efficient, and decrease system downtime of wind turbine operation.
All Science Journal Classification (ASJC) codes