Fault Identification Method Based On Multi-level Evidence Reasoning Fusion

For complex industrial system,single sensor-based industrial equipment condition monitoring has rarely satisfied the actual requirements,hence to apply the multi-sensors-based condition monitoring has been becoming an available choice.There are some factors such as the measurement error of the sensor itself,the influence of the external environment and the change of the fault state of the system which all lead to the uncertainty of the monitoring and diagnosis.In addition,the industrial equipment is in a normal working state most of the time and the fault simulation is usually costly,so there are some troubles that the fault samples for modeling the diagnostic model are scarce,the normal samples and the fault samples are not balanced.In order to solve the above problems of "uncertainty" and "unbalance",in the actual background of the irregularity of railway track,this paper proposes a fault identification method based on Evidence Reasoning(ER)fusion.Track irregularity is one of the important indicator of track management,it is of great significance in relation to the traveling comfort and safety of train operation.In this thesis,the multi-level ER fusion method is used to model the nonlinear relationship between the vibration acceleration data and the irregularity amplitudes,and a corresponding track irregularity fault recognition hard/software system is designed.The main research contents are as follows:(1)A review of the background and main methods of industrial system fault diagnosis,to analyze the "uncertainty" and "unbalance" problems in fault diagnosis.This paper introduces the fusion method based on Dempster-Shafer(DS)evidence theory,and ER fusion method,expounds its roots,progress,practical applications.At the same time it explains the mathematical description,reasons principle of information fusion based on evidence reasoning(ER)rule.(2)A method for identifying the track irregularity is proposed based on multi-level ER fusion.The multi-source vibration features collected by the sensor are used as the input of the ER fusion model,and the irregularity amplitudes are taken as the output of the model.The k-NN algorithm is applied to obtain k neighbor samples about the current sample from the historical sample database;the multiple pieces of evidence activated by the current single source sample and its neighbor samples arefused(the first level ER fusion),and the corresponding fusion results coming from different sources are fused again(the second-level ER fusion),and the irregularity amplitude is estimated based on the final fusion result.Since the historical neighbor samples participate in the fusion diagnosis process,the problems of “uncertainty” and“unbalance” can be effectively solved.(3)A track irregularity fault identification system is designed.There are six modules including embedded central processing unit,accelerometer(ACC)acquisition,wireless communication,GPS positioning,storage and display.By loading the multi-level ER fusion model algorithm,the functions of data acquisition,display storage and irregularity amplitude estimation are tested under laboratory conditions.Therefore,the track irregularity state recognition system based on vibration feature fusion analysis is established,which lays a foundation for practical engineering application.