Wind Turbine Gearbox Anomaly Detection Using Signal-to-Image Processing Algorithms and Convolutional Autoencoder

Abstract In this study, signal-to-image conversion techniques coupled with a convolutional auto encoder (CAE) are used for the detection of anomalies in the wind turbine (WT) gearbox system. Firstly, the time series data is converted to images using six different algorithms. Thereafter, these images are stacked into the multi-dimensional structures known as “data cubes”, which is finally fed into CAE for the anomaly identification. The results of this study demonstrate the enhanced efficacy of the method specially in the Gramian Angular Field model in detecting anomalies accurately, suggesting a viable path towards the implementation of dependable and affordable WT monitoring systems. This will open the door for the renewable energy industry’s condition monitoring procedures to become more automated and digitalized.