Gear Fault Diagnosis Method Based on Deep Transfer Learning

Abstract

Aiming at the problem of insufficient gear fault samples, a fault diagnosis method of transfer learning based on Hilbert-Huang spectrum and pre-trained VGG16 model is proposed. Firstly, the intrinsic mode function (IMF) is obtained by Empirical Mode Decomposition (EMD) of vibration signals, and the time spectrum is obtained by Hilbert transform of IMF with the largest correlation coefficient. Secondly, pre-trained VGG16 is used to extract Hilbert-Huang spectrum image features of gears under varying loads and under various health conditions. Finally, the global average pooling layer is used to replace partial full connection layer of VGG16 model for classification output. Experimental results show that with a small amount of sample data, the accuracy of gear fault diagnosis reaches 98.86%, which is better than the transfer learning methods such as TLCNN and Tran VGG-19. It is proved that the method presented in this paper has some research value in gear fault diagnosis.

Keywords

• Transfer learning;VGG16;Hilbert-Huang spectrum;Gear fault diagnosis;Global average pooling