Incipient Fault Diagnosis And Prognosis With Application To High-speed Railway Traction Systems

Due to its comfort,convenience,safety and punctuality,high-speed train has become a prior intercity transportation way in China.Traction system,the core of the high-speed trains,is a complex electromechanical coupling system,whose reliability is of critical importance to the safety of the entire train.Along with the growth of running time,some components in the traction system degrade with age and may cause various incipient faults.These incipient faults often have small amplitudes and obscure early-stage symptoms,and the existing FDD(Fault Detection and Diagnosis)strategies have high missing alarms which definitely increase the risk of serious accidents in the trains.Thus,incipient fault diagnosis and prognosis are urgently demanded in high-speed railway traction systems.This thesis has addressed the problems of incipient fault detection,diagnosis(isolation & reconstruction)and prognosis for CRH traction system,including fault diagnosis and prognosis in suspension system,induction motor drive circuit,motor sensors,PWM inverter system and stator/rotor winding.Mathematical system descriptions include linear and time-invariant system,T-S fuzzy sytem,Lipschitz nonlinear system with uncertainties and Lipschitz nonlinear descriptor system.The main work are presented as the following six sections:(1)Several ToMFIR-based fault detection schemes are proposed for linear and time-invariant system including ToMFIR theory in linear and time-invariant system,robust ToMFIR theory,approximate ToMFIR theory and observer residual modification theory.A dynamic model of the CRH suspension system is set up,which contain the track irregularities and several kinds of incipient actuator faults,such as gain faults,bias faults and multiple faults.According to robust ToMFIR theory and approximate ToMFIR theory,a ToMFIR-based incipient fault detection scheme is proposed with an application to a high-speed train suspension system.Simulation results show that the proposed method can detect the common incipient actuator faults effectively,outperforming the observer residual based fault detection methods.(2)The problem of incipient fault detection and diagnosis for Takagi-Sugeno(T-S)fuzzy systems is addressed and studied.Firstly,T-S fuzzy model is used to describe the global dynamics of a nonlinear system and the model of incipient actuator faults is formalized.Secondly,a novel incipient fault detection method is proposed based on fuzzy ToMFIR theory,which can remove the assumptions on system structure in some existing work.Further,an incipient fault isolation scheme is designed by combining sliding-mode observer and ToMFIR-based isolation thresholds.In addition,an adaptive estimator is used for the purpose of fault estimation after the faulty actuator has been isolated.The key novelties of the proposed method are in that,the limitations on system structure are overcomed and a more general FDD(Fault Detection and Diagnosis)framework is given.The proposed incipient FDD framework has been applied to a traction motor control system in CRH2 high-speed train successfully.(3)ToMFIR theory in nonlinear systems is further studied and a solution to incipient fault detection and isolation for a class of uncertain nonlinear systems with multiple sensor faults is presented.The proposed diagnosis scheme can effectively detect and isolate current sensor fault and speed sensor fault which occur at the same time in traction motor systems of CRH2 high-speed train.The results show that,the linearization process can be removed and the design methods based on state observer can be used directly in nonlinear system.(4)Descriptor systems can be used to model a considerable proportion of complex systems in the areas of electrical and electronic engineering,mechanical engineering,aerospace engineering and so on.Thus,this thesis also focus on incipient fault diagnosis in descriptor systems.A novel descriptor observer based fault estimation scheme is presented for nonlinear Lipschitz descriptor systems to estimate the actuator and sensor faults simultaneously.Firstly,by parameterizing the actuator and sensor fault terms in the original descriptor system,an augmented descriptor system model is constructed.Then,a novel descriptor estimator is developed to decouple both the input disturbances and measurement output noises.Thus,the system states,actuator/sensor faults and the output noises can be estimated simultaneously.In order to achieve an optimal performace of signal estimation,an LMI optimization problem is proposed.Finally,application results conducted on the three-phase inverter system of CRH high-speed trains are given to illustrate the effectiveness of the proposed approach.(5)Faults in stator winding or rotor winding are the main kinds of faults in induction motor systems.Incipient winding fault detection and isolation can not only improve the reliability of motors,but also avoid serious failures in the entire traction systems of high-speed trains.Firstly,the state-space mathematical model of induction motor d-q coordinate system is established,based on which,the fault characteristics of induction motor stator/rotor winding are analyzed and incipient winding fault models are also obtained.Then,a robust fault detection and isolation scheme is proposed for the uncertain single output nonlinear system with incipient stator winding fault(an internal winding turn-to turn short of 5% of the entire stator winding).Simulation results show that the incipient fault diagnosis scheme proposed not only has high sensitivity for incipient winding faults but also can accurately isolate faulty winding(stator winding or rotor winding).(6)To investigate the fault propagation mechanism and predict the fault probabilities accurately for a high-speed railway traction system,a fault prognosis approach via Bayesian network and bond graph modeling techniques is proposed.The inherent structure of a railway traction system is represented by bond graph model,based on which,a multilayer Bayesian network is developed for fault propagation analysis and fault prediction.For complete and incomplete data sets,two different parameter learning algorithms,Bayesian estimation and expectation maximization(EM),are adopted to determine the conditional probability table of the Bayesian network.The proposed prognosis approach using Pearl's polytree propagation algorithm for joint probability reasoning can predict the failure probabilities of stator and rotor accurately.