Minimum Noise Amplitude Deconvolution And Its Application In Rolling Bearing Fault Diagnosis

Blind deconvolution(BD)algorithms reduce the influence of signal transmission path and noise on fault characteristics and recover the repetitive impact characteristics hidden in the raw signal.It has always been a hot spot for fault diagnosis based on vibration signals.This paper mainly studies the fault diagnosis method of rolling bearing based on BD.The main work and contributions are as follows:(1)Aiming at the problem that the existing BDs are difficult to ensure the solution precision and efficiency simultaneously,a general solution algorithm for BD problem based on backward automatic differentiation is proposed-backward automatic differentiation blind deconvolution(BADBD).BADBD uses the backward automatic differentiation algorithm to calculate the gradient of the maximization criterion on the filter coefficients by itself,which replaces the manual solution and meets the flexibility of the selecting the maximization criterion;BADBD allows multiple cascaded filters to filter signals,which makes up for the lack of performance of a single filter;BADBD iteratively updates the filter coefficients through Adam algorithm,which improves the optimization ability of BD and speeds up the convergence speed of the algorithm.(2)Aiming at the problem that the objective function of the existing BDs cannot effectively characterize the repetitive impact characteristics excited by faults,the BD has a limited effect on the enhancement of fault characteristics of multi-resonance band signals and low signal-to-noise ratio signals.A new objective function,periodic noise amplitude ratio(PNAR),is proposed.PNAR is defined as the ratio of the average amplitude of signal periodic noise to the effective value of the signal.Taking PNAR as the minimization criterion,the PNAR of the filtered signal is minimized with the help of BADBD,and a new BD algorithm called minimum noise amplitude deconvolution(MNAD)is obtained.Different from the existing BD algorithms,MNAD does not need to face the problem of difficult signal fault impact location.MNAD indirectly enhances the signal’s fault impact characteristics by reducing the signal’s periodic noise amplitude.Experiments show that MNAD can adaptively and accurately locate one or more resonance bands excited by rolling bearing faults,which is significantly better than the existing algorithms.MNAD can also considerably enhance the fault characteristics of signals with a low signal-to-noise ratio.(3)Aiming at the problem that the existing rolling bearing health monitoring methods are challenging to monitor the early failure of rolling bearing,a method for monitoring the health status of rolling bearings based on MNAD is proposed.The basic properties of PNAR index are analyzed,and minimum periodic noise amplitude ratio(MPNAR)is defined to characterize each fault stage of rolling bearing.The MPNAR corresponding to the actual fault type is used as the monitoring index of the rolling bearing health state,and the validity of the index is verified through experiments and engineering data.