Research On Blind Deconvolution Algorithm For Repetitive Transient Impact Detection Of Rotating Machinery

Rotating machinery system is an important part of rail transit train.Its health state directly affects the operation safety of rail transit train.With the continuous development of automatic driving technology and smart train technology,the traditional temperature-based pre-warning technology of rotating machinery has been unable to meet the new requirements of intelligent development.It is urgent to develop tracking monitoring,self-diagnosis,and real-time prewarning technology for the health status of rotating machinery systems.In the meanwhile,with the continuous increase of the mileage,speed and density of the rail transit,the traditional maintenance strategy based on “period maintenance and planned maintenance” has been unable to adapt to the increasingly heavy maintenance tasks.Condition-based predictive maintenance is the basic direction of the reform and development for the maintenance process and maintenance system of rail transit equipment.Health condition monitoring technology of rotating machinery is the key to realize condition-based maintenance of the critical systems.Because of the higher sensitivity of vibration response to rotating machinery faults and the increasingly mature real-time acquisition technology of vibration data,the health monitoring technology based on vibration detection has been widely used.However,it is still a difficult task to detect rotating machinery faults from vibration data.Therefore,it is of great significance to strengthen the research of fault vibration detection technology in complex operating environment for making maintenance plans,preventing equipment damage and ensuring equipment operation safety.The purpose of this study is to improve the existing fault diagnosis method of rotating machinery based on vibration signals to detect the repetitive transient impact caused by rotating machinery faults(repetitive transient impact is the main symptom of rotating machinery fault).The repetitive transient impact caused by bearing faults is taken as the key research object in this study.Aiming at this research task,this thesis summarizes the framework of repetitive transient impact detection,and on this basis,three blind deconvolution techniques based on this framework are proposed to realize repetitive transient impact detection.Firstly,based on the mathematical generation model of repetitive transient impact,the theoretical framework of repetitive transient impact detection is summarized and the three key techniques(characterization index,filter selection and optimization method)of the theoretical framework are analyzed.Based on the summarized theoretical framework,the advantages and disadvantages of the existing methods of repetitive transient impact detection(filtering-based detection method)are analyzed and discussed.Additionally,the conditions that need to be satisfied to design a good repetitive transient impact detection method are given.Secondly,aiming at the problem that the traditional blind deconvolution algorithm is easy to fall into the local optimal solution(corresponding to the optimization method in the three elements of theoretical framework for repetitive transient impact detection),the traditional blind deconvolution algorithm is extended to the multi-node structure blind deconvolution algorithm.The blind deconvolution algorithm with multi-node structure is a multi-node neural network,which transforms the blind deconvolution process into an unsupervised feature learning problem.In the meanwhile,a new multi-node blind deconvolution algorithm with a special structure is proposed.The new algorithm can obtain better results with fewer nodes,thus alleviating the problem that the traditional blind deconvolution algorithm is easy to fall into the local optimal solution.Thirdly,in view of the problem that the traditional time-domain blind deconvolution algorithm is sensitive to outliers(corresponding to the “observation” in the three elements of theoretical framework for repetitive transient impact detection),a nonparametric envelope spectrum blind deconvolution algorithm is proposed.On the basis of mathematically proving that the traditional time-domain characterization index is sensitive to outliers and the envelope spectrum characterization index is robust to outliers,a blind deconvolution algorithm based on the nonparametric envelope spectrum characterization index is proposed.In the proposed method,the cyclostationary tool(envelope spectrum)is applied to the blind deconvolution algorithm,which makes the proposed method robust to time-domain outliers and sensitive to repetitive transient impacts.The performance of the proposed method is significantly better than that of the traditional time-domain blind deconvolution method.Fourthly,since the proposed nonparametric envelope spectrum blind deconvolution method is unable to realize concurrent fault detection,a parameterized envelope spectrum blind deconvolution method based on the harmonic noise ratio index of the envelope spectrum is proposed.The proposed method is robust to time-domain outliers and introduces the capability of concurrent fault detection.The performance of the new method is significantly better than the existing blind deconvolution methods with concurrent fault detection capability.The performances of all the methods proposed in this thesis are verified on simulation signals,open data sets,experimental data of high-speed train axle box bearing and experimental data of high-speed train gearbox bearing.The experimental results show that all the proposed methods have excellent fault detection performance and engineering application value.