A Hybrid CNN-LSTM Approach for Early Gear Fault Detection in Wind Turbines Using Vibration and SCADA Data

Detecting gear faults at an early stage is important to keep wind turbines running smoothly and to reduce expensive repairs in renewable energy systems. In this study, we test a hybrid method that combines Convolutional Neural Networks (CNNs) with Long Short-Term Memory (LSTM) models to diagnose gear faults using both vibration sensor data and wind turbine SCADA signals. The vibration data, collected under multiple load conditions, are processed using time-domain normalization, frequency-domain transformations (FFT), and time-frequency spectrograms, while SCADA data are analyzed for operational anomalies via statistical z-score methods. The CNN-LSTM model learns spatial and temporal patterns, showing strong and consistent results across different fault scenarios. Our results suggest that this model performs better than standard methods, offering more precise fault detection and better adaptability. This combined approach can help improve maintenance planning and make turbine monitoring more effective.