Fault Detection Method Based on Multi-scale Convolutional Neural Network for Wind Turbine Gearbox

This paper is concerned with the fault detection problem of planetary gearbox in wind energy conversion systems (WECSs). An effective method based on empirical mode decomposition (EMD) and multi-scale convolutional neural network (MSCNN) is proposed. Specifically, the non-stationary vibration signals of the gearbox are decomposed by using the EMD, so that its signal-to-noise ratio (SNR) and characteristic information are improved. Then, a hierarchical convolutional neural network is applied to adaptively extract multi-scale features from the decomposed signal components. Finally, a binary classifier based on cross entropy is employed to automatically realize the classification of the obtained multi-scale features. The effectiveness and superiority of the proposed method are verified on the experiments with vibration data sets from a true WECS.