Wind turbines are one of the renewable sources of energy production, where the stability of their operation plays an important role in the efficiency of power generation systems. This thesis presents a data-driven framework for the early detection of faults in wind turbines, which consists of three main stages: data preprocessing, statistical process control, an‎d machine learning-based modeling. In the statistical analysis section, repeated exceedances of defined thresholds by temperature-related variables, such as hydraulic oil temperature, were detected, indicating potential early warning signs of failure. In the machine learning phase, after labeling the pre-failure periods, feature selec‎tion was performed using correlation coefficients an‎d one-way ANOVA to extract the most discriminative parameters between normal an‎d pre-failure states. The ran‎dom forest model outperformed the decision tree in both accuracy an‎d stability, demonstrating a stronger ability to identify critical conditions. The results show that combining statistical analysis with machine learning methods provides an effective framework for predictive maintenance an‎d fault prediction in wind turbines.

ساره گلي فروشاني

مريم كلكين نما , زهرا صابري