Research On Multiple Fault Diagnosis For Grid-side Cascaded H-bridge Converter In Electric Railway Traction

In recent years,a new electrical traction system with the power electronic traction transformer(PETT)has attracted much attention and has become an important development trend for high-speed trains due to the efficiency and lightweight requirement.As the conversion port of the medium ac voltage to medium dc voltage in the PETT,the grid-side multilevel converter represented by the cascaded H-bridges is responsible for voltage conversion and the stable operation of the system.Therefore,ensuring its safe and reliable operation becomes especially important.The external environment is very destructive to the system and is prone to failure.The internal structure of the electrical traction system is complex.Affected by the closed-loop control of the system and component degradation,the fault is prone to spread and generate multiple faults.In addition,the cascaded multilevel system has more components with high failure rates such as power devices and sensors.Meanwhile,higher voltage and power levels,and more complex closed-loop control pose higher challenges to the safe operation.To this end,this paper takes the grid-side cascaded H-bridge converter(CHBC)in electric railway traction as the research object and conducts online diagnosis algorithm research on open-circuit faults of power devices and sensor faults,which provides an important foundation for timely maintenance and fault-tolerant operation of the system,reducing downtime cost and maintenance burden.Firstly,for the power devices with the highest failure rate,an open-circuit fault diagnosis method based on the comprehensive fault characteristics of grid-side current and dc-side voltage is proposed.Considering the continuous and discrete features in the converter,a mixed logic dynamic(MLD)model of the single-phase CHBC is established.Through the design idea of feedback gain in the sliding-mode observer(SMO),the state estimator is established to obtain the estimated values of the grid-side current and dc-link voltages.Comparing the estimated values with the measured values,a normalized voltage detection threshold is designed to judge the faulty module.Furtherly,the specific driving signals are injected into the fault module to identify the specific faulty switch by judging the changing trend of the current residual.The proposed diagnosis method can realize the rapid diagnosis of a single open-circuit fault without the additional hardware and has good adaptability for different operating conditions.Then,considering the problem that the faults are easy to spread,causing multiple faults of power devices,a residual-changing-rate-based diagnosis method is studied.The grid-side current estimation model is established.Comparing the estimated values with the measured ones,the theoretical expression of the current residual is obtained through exponential function simplification.Analyzing the corresponding relationship between the residual change rate and the driving signals under single and double open-circuit faults,multiple faults can be equivalent to the superposition of different single faults.Considering the current direction,driving signal,and signal isolation of similar faulty switches,independent diagnostic variables for different switches are designed to achieve accurate localization of different faulty switches.All monitoring signals used for diagnosis come from the existing control system,and the effective diagnosis of single or multiple open-circuit faults can be realized within one current fundamental cycle under different operating conditions.Furthermore,considering the serious coupling of fault features in the grid-side current and the separating difficulty of similar faults under multi-cascaded modules,the output currents of different modules are selected as monitoring signals.The output current estimation model and residual expression are established to decouple the fault information of different modules.The independent diagnostic variables of different switches are designed and normalized.The nonlinear filtering,parameter estimation,and enhanced detection algorithm are introduced to deal with the measurement noise and parameter time-varying problems of the actual system,ensuring that the algorithm can maintain high diagnostic accuracy and robustness in the long-term high-power operation.The algorithm only needs to use two normalized thresholds to locate the multiple switch faults of different modules in less than one fundamental cycle.What’s more,the proposed method is not affected by the number of faulty modules and fault coupling,with good scalability.Moreover,in order to achieve a good balance between fastness,execution complexity,robustness,and diagnostic ability of the multiple fault diagnosis algorithms,a voltage residual-based diagnosis algorithm is designed.The actual value of the input-side multilevel voltage is calculated through the system circuit parameters and the measurable signals,and the estimated value is obtained by using the switching function and the dc-link voltages.The voltage residual is further normalized to analyze the relationship between the residual features and switching states under different faults.Then,the count comparators are introduced to obtain the maximum count value,determining the faulty module.Further,the zero-switching state of the faulty module is compared with the residual to determine the faulty switch.After the identification of a faulty switch is completed,the fault information of the detected switch is cleared and the counting comparators are reset to re-detect and locate more faulty switches.Thanks to the simple design of fault diagnosis,interference elimination,and parameter estimation,the proposed algorithm can use lower computing resources to realize single and multiple fault diagnosis regardless of the number and position of the faulty modules and faulty switches.Finally,considering the problem of numerous sensors and their frequent failures in the CHBC,a multiple fault diagnosis method for sensors is designed from the perspective of fault estimation,and the fault-tolerant operation of the control system is completed.The augmented descriptor system of the CHBC is established to unify the fault disturbance information and the state variables of the original system into augmented state variables.And then,a linear reduced-order observer is designed through the equivalent matrix transformation and Lyapunov stability criterion,where the fault disturbance information is decoupled and the coefficient matrix of the observer is calculated to realize the asymptotic estimation of augmented state variables.The fault disturbance information is compared with the normalized threshold to detect the sensor fault.In addition,the estimated state variables of the original system are used to finish the control reconfiguration,so that the controller can continue to work after a single or multiple sensors fail,maintaining the normal operation of the system.In the design of the fault estimation,considering the influence of measurement noise,sampling error,delay,the voltage drops of power devices,etc.,voltage drop compensation and state feedback adjustment are introduced to obtain better estimation results,ensuring the diagnosis accuracy and the good fault-tolerant control performance.