Gearbox fault diagnosis method based on optimized VMD-NLM with 1DDRSN

ObjectiveAiming at the problem of poor accuracy of gearbox fault diagnosis under noise interference, a new fault diagnosis method for gearboxes based on the denoising methods of optimized variational modal decomposition (VMD）and non-local means (NLM) was constructed, combined with a one-dimensional deep residual shrinkage network (1DDRSN).MethodsFirstly, the parameters in the VMD were automatically optimized using the subtractive average-based optimization (SABO); secondly, each intrinsic mode function (IMF） after the decomposition of the VMD was filtered using sample entropy, and the noise-containing components were subjected to the NLM denoising and reconstruction; then, a residual network that combines the attention mechanism with soft thresholding was introduced to model 1DDRSN; finally, the denoised and reconstructed signals were inputted into the 1DDRSN for fault diagnosis and identification. And the validation was carried out through the DDS test bench.ResultsThe results show that the denoised signal improves the fault accuracy by 3.16% compared with the original signal, which indicates that the optimized VMD-NLM has a better noise reduction effect. The diagnostic accuracy of the 1DDRSN reaches 99.33%, and compared with the CNN and ResNet, the accuracy improves by 5.97% and 1.17%, respectively, which verifies the method’s feasibility and the efficiency of the diagnosis effect.