Prediction of remaining useful life of wind turbine shaft bearings using machine learning

Abstract

Wind turbines are a major trend in the current green energy market. Wind energy is abundant, and if utilized properly, can result in significant reductions in carbon emissions. Therefore, the development of wind power systems is urgently required. However, wind turbines are mainly built in unmanned areas. Regular inspections require substantial manpower and material resources, and doubts regarding the accuracy of the inspected data may occur. Therefore, it is necessary to establish an automatic diagnostic method for determining the remaining useful life (RUL) of a wind turbine to facilitate predictive maintenance. In this study, a multi-class support vector machine (SVM) and a convolutional neural network (CNN) were employed for fault diagnosis and RUL prediction of the shaft bearings used in wind turbines. During the multi-SVM process, the vibration signal of the shaft bearings was converted into a 15-parameter feature vector input for training and prediction; we achieved a resulting classification accuracy of 95.33%. For the CNN process, the spectrogram of the vibration signal from the wind turbine shaft bearings was used to train a CNN; here, we achieved a resulting classification accuracy of 100%.