Abstract
Due to the non-linearity and non-stationarity of sound signals from complex transmission structure and frequent start-stop of industrial equipment, such as sewing machine, its incipient faults cannot be accurately identified with classical fault diagnosis methods. Therefore, a new fault diagnosis method with an adaptive variational nonlinear chirp mode decomposition (AVNCMD) is proposed. Firstly, a pre-processing step of convex optimization with generalized minimax-concave penalty function and spline-kernelled chirplet transform is applied to obtain the time–frequency distribution (TFD) with energy concentration, and a peak frequency extension method is proposed to adaptively search initial instantaneous frequencies of AVNCMD on the TFD; then the Pearson correlation coefficient is introduced as the termination condition of the adaptive signal decomposition. Secondly, multi-domain features are extracted from the decomposed nonlinear chirp modes, and an optimal features set is constructed by the feature preference method of an improved ReliefF. Finally, the fault classification is realized by the extreme learning machine, whose input weights and biases are optimized via the salp swarm algorithm. The AVNCMD is testified to be accurate in simulation with nonlinear amplitude modulation–frequency modulation signals and be effective in actual sound signals of industrial sewing machines on four different states. Through the detailed analysis of the measured sound signals in factory, the proposed fault diagnosis method can identify four states of the industrial sewing machines with an overall identification accuracy of 95.0%.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call