Real-time multi-fault diagnosis in wind turbines using a physics-guided hybrid RBF-ANN framework

In this study, a real-time diagnostic system is proposed for fault detection in wind turbine power systems using a hybrid Radial Basis Function–Artificial Neural Network (RBF-ANN) framework. The proposed system analyzes environmental, mechanical, and electrical parameters, including wind speed, temperature, rotor speed, gearbox vibration, torque, voltage, and current, to identify multiple operating conditions. A dataset of more than 22,000 labelled samples was developed using real-time turbine simulator data and MATLAB/Simulink–FAST-based fault simulations. Six diagnostic classes were considered: healthy condition, generator fault, gearbox fault, rotor imbalance, electrical disturbance, and compound fault. The proposed RBF-ANN achieved an overall classification accuracy of 94.8% and a weighted F1-score of 0.942, outperforming SVM, k-NN, and MLP models, which achieved accuracies of 89.1%, 86.3%, and 91.5%, respectively. The model also demonstrated strong class-wise performance, with diagnostic accuracy ranging from 90.8% for compound faults to 98.5% for healthy operating conditions. Real-time implementation on a Raspberry Pi 4 produced an average inference latency of 192 ms per sample, with approximately 62% CPU utilization and 280 MB RAM usage, confirming its suitability for lightweight edge deployment. The false-positive rate remained below 4% across all fault categories, indicating reliable fault discrimination under varying operating conditions. These results demonstrate that the proposed RBF-ANN framework provides an accurate, interpretable, and computationally efficient solution for real-time wind turbine fault diagnosis and predictive maintenance.