Nonconvex Regularized Multi-source Sparse Decomposition Method For Gearbox Fault Diagnosis

With the continuous improvement of the level of modernization,vehicles are developing in the direction of intelligence.Research on condition monitoring and fault diagnosis of vehicle key components is an important way and inherent requirement to realize vehicle intelligence.Gearbox is a key component of vehicle system to transmit power and an important object of vehicle operation and maintenance.While transmitting power,the gearbox also bears strong load impact,which is prone to failure.The failure of the gearbox will cause serious consequences to the entire vehicle system and even casualties.Therefore,it is of great practical significance to study the condition monitoring and fault diagnosis of the vehicle gearbox to ensure its safe and reliable operation.Because of the complex working environment of the gearbox system,the collected vibration signals often contain strong background noise.The aliasing of multiple response components and the interference of background noise greatly increase the difficulty of fault feature extraction.Therefore,it is necessary to study effective signal analysis techniques in order to extract the characteristic components related to the characteristic information of the gearbox state from complex signals.This paper aims to achieve accurate fault feature extraction of multi-source signals of gearboxes.Based on the application of sparse decomposition method in gearbox fault diagnosis,a fault feature extraction method based on non-convex regularized multi-source sparse decomposition method is studied.The main research contents of this article are as follows:(1)Research on the generalized smooth log-regularized sparse decomposition method and its application in bearing fault feature extraction.The characteristics of several common non-convex penalty functions are analyzed,and based on the construction method of GMC penalty function,a generalized smooth logarithmic penalty function is designed based on the traditional logarithmic penalty function,which is applied to the sparse decomposition model.In order to improve the precision and accuracy of bearing fault feature extraction,a generalized smooth log-regularized sparse decomposition algorithm based on generalized smooth log-regularization is studied to solve the problems of underestimating signal amplitude by traditional convex penalty functions such as L1 norm and insufficient extraction accuracy of GMC penalty function under strong background noise.Simulation and engineering experiment analysis verify the superiority of the proposed method in accurate extraction of fault features and the reliability of bearing fault diagnosis.(2)Research on the nonconvex multi-source sparse decomposition method under tight frame and its application in gearbox fault diagnosis.The sparse representation dictionary is constructed by studying the Tunable-Q wavelet transform that satisfies the tight frame conditions.It does not involve high-dimensional matrix inversion operation,which can be solved efficiently and quickly.By adjusting the Q-switched wavelet parameters,a dictionary matching different components in the fault signal is constructed respectively.The multi-source sparse representation model of the gearbox is constructed,and then the generalized smooth logarithmic non-convex penalty function proposed in this paper is applied to the multi-source sparse decomposition model.Finally,the nonconvex multi-source sparse decomposition algorithm is obtained by using the optimization algorithm.Simulation and experimental signal analysis verify the applicability and superiority of the proposed method in the gearbox fault diagnosis.(3)Research on the sparse decomposition method of nonconvex penalty function under balance model and its application in gearbox fault diagnosis.Most of the vibration signal sparse decomposition models are based on synthesis sparse decomposition model,while the sparse decomposition models based on analysis and balance are less studied.In order to verify the performance of the balance model in fault feature extraction and to improve the accuracy of gearbox fault feature extraction,a balance,synthesis and analysis sparse decomposition algorithms based on arctangent non-convex penalty function are proposed and applied to gearbox fault feature extraction.Finally,the simulation and engineering experiments verify that the proposed non-convex penalty function sparse decomposition method under the balance model has excellent performance in gearbox fault feature extraction.Through the research in this paper,the key feature information is accurately extracted from the gearbox fault signal with strong background noise,which provides a feasible method for further gearbox fault severity assessment.And it has important theoretical and practical significance for improving the intelligent operation and maintenance level of the vehicle system.