Fault Diagnosis of Wind Turbine Bearings Using Siamese Networks

Bearing faults pose a significant threat to the reliability and efficiency of rotating machinery systems, particularly in wind turbines. Traditional fault diagnosis algorithms often face limitations, such as processing single signal inputs from each sensor without considering interferences between components. To overcome these shortcomings and leverage the potential of processing multi-dimensional signal inputs, we propose SiamFD-net, a deep learning-based method. Focusing on wind turbine bearing fault diagnosis, our model employs a Siamese network architecture, which distinguishes between similar and dissimilar input pairs. This approach not only addresses the limitations of traditional algorithms but also contributes to mitigating the class imbalance problem commonly encountered in fault diagnosis. Experimental results using the Doosan Wind Turbine bearing dataset demonstrate the comprehensive view and diagnosis capability of our model. Furthermore, we validate the effectiveness of SiamFDnet using the Case Western Reserve University (CWRU) bearing fault diagnosis benchmark dataset. Our code is openly accessible at: https://git.io/SiamFDnet.