Fault Diagnosis Method Based on MTF-ResDSCNN Two-dimensional Image

Abstract

In order to effectively capture the fault features contained in the vibration signals of the rotating machinery and complete the fault diagnosis task efficiently, a fault diagnosis model combining two-dimensional image features and lightweight neural network is designed. Firstly, the collected one-dimensional vibration signals are decomposed by modified ensemble empirical mode decomposition (MEEMD) to obtain the intrinsic mode function (IMF) components, and the corresponding IMF components are selected for sum reconstruction to enhance the amplitude fluctuation of vibration signals. Then, Markov transition field (MTF) could be used to more effectively characterize the fault features in the reconstructed signals. Secondly, the 2D feature map generated by MTF is input into residual depth separable convolutional neural network (ResDSCNN) for feature extraction and fault diagnosis. The planetary gearbox fault data set is used to verify the performance of the model, and the results show that the diagnosis accuracy of the model for all kinds of gear faults can reach more than 98%.

Keywords

• Markov transition field;Depth separable convolution;Fault diagnosis;MEEMD