Research On Localized Fault Diagnosis Method Of Rotating Machinery Based On Sparse Regularization

As the essential transmission components in rotating machinery,gears and rolling bearings are prone to localized faults after long periods of operation under harsh environments,which may lead to disastrous consequences.Therefore,it is vital to monitor the state of rotating machinery and diagnose the existing faults in time.Based on the theory of sparse representation and compressed sensing,the application of the sparse regularization method in localized fault diagnosis of rotating machinery is deeply studied in this paper.In terms of sparse representation,to solve the low accuracy and high computational complexity problems of the existing l₁ regularization method,a novel objective function with convolutional generalized minimax-concave（GMC）penalty is established and its fast solving algorithm is derived for gear localized fault diagnosis based on the analysis of the shift-invariant sparse characteristics of the fault signal.The stationary components with large energy are first removed to improve the signal-to-noise ratio of the localized fault feature,and then the shift-invariant K-SVD is utilized to obtain the impulse pattern from the residual signal.Finally,the sparse coefficient by solving the convolutional GMC problem is obtained,and the fault impulse feature is further extracted.Compared with the commonly used l₁ regularization term-based-solving method,the proposed method effectively alleviates the amplitude underestimation problem with the same regularization parameter setting.The effectiveness of the proposed method is verified through the gear fault simulation signal and experiment signal analysis,and the misdiagnosis of the normal gear signal will not be caused by the proposed method.Aiming at the low computational efficiency problem of the existing convex optimization reconstruction algorithms in compressed sensing,a method for diagnosing rolling bearing compressed fault based on the feature proxy and the convex optimization algorithm is proposed.The shift-invariant K-SVD is employed to learn the impulse pattern.The function of the feature proxy and the noise contained in the proxy are analyzed,and the fast iterative shrinkage threshold algorithm（FISTA）is used to extract the sparse coefficient from the proxy.Finally,the impulse pattern and the sparse coefficient are convolved to reconstruct the fault feature.Compared with the direct impulse feature reconstruction method from the compressed signal using the FISTA,the proposed method reduces the computational complexity while without reducing the solution accuracy.When compared with the commonly used greedy reconstruction algorithm,the proposed method does not require the prior estimation of the signal sparsity and can reconstruct more impulse features.The feasibility of the proposed method is further demonstrated by the localized fault simulation signal and experiment signal analysis of the rolling bearing.For the amplitude underestimation problem of the l₁ regularization term in compressed fault feature reconstruction,a compressed localized fault diagnosis method with non-convex regularization is proposed.The GMC penalty function is used as the regularization term to improve the signal sparsity,and the strategy of signal segmented compression and reconstruction is adopted to improve the compression and reconstruction efficiency.Besides,the regularization parameter selection strategy based on the feature proxy is discussed emphatically.The gear and rolling bearing fault simulation and experiment comparison results show that when the regularization parameters are all set as the recommended values,the GMC penalty function can effectively alleviate the amplitude underestimation problem of the l₁regularization term while suppressing noise more,which can improve the localized fault feature reconstruction accuracy.