Fusion of Variational Modal Decomposition and ResNet Network for Intelligent Fault Diagnosis of the Bearing under Time-varying Speed Condition

Given that the bearing often works under time-varying speed condition, vibration signal is modulated by speed, making difficulty in extracting fault features, resulting in low fault diagnose accuracy by the traditional intelligent diagnostic methods. To address above challenge, this paper proposes a novel method for intelligent fault diagnosis of the bearing under time-varying speed condition, which combines the advantage of joint feature extraction of one-dimensional convolution and two-dimensional convolution, as well as the training advantage of ResNet network. Firstly, Variational modal decomposition (VMD) is used for decomposing the original signal into multiple intrinsic mode functions (IMFs). Secondly, one-dimensional convolution is used for reconstructing all IMFs into a two-dimensional feature map called VMD feature map. Finally, the abstract features in VMD feature map are extracted by ResNet network. The experimental result indicates that the proposed method can weaken the influence of speed modulation, and improve the quality of the bearing abstract features under time-varying speed condition, thus raising the recognition accuracy of fault diagnosis. Compared with current first-rate methods, the proposed method shows better fault diagnosis performance.