Research On Fault Characteristic Parameters Extraction And Health Forecast Methods Of Power Electronic Circuits

As the power electronic technology rapidly develops, kinds of power electronic products arepopping up everywhere, such as electric traction devices, electric actuators and inverters. Powerelectronic circuits are essential subsystems of engineering systems whose failure leads to theunplanned standstill of the systems. Therefore, with the purpose of improving the reliability andreducing the breakdown loss of the engineering systems, prognostic and health management of thepower electronic circuits are essential.This thesis mainly focuses on the research of fault characteristic parameters extraction andprognostic methods of power electronic circuits based on performance degradation. The researchcontent contains following aspects:(1) The selection of key components’ fault characteristic parameters is studied. Throughanalyzing the failure mechanism and failure mode of key components, the fault characteristicparameters which can reflect the degradation condition of the components are decided and the failurethresholds of the parameters are confirmed.(2) The parametric fault diagnosis and identification method are researched for power electroniccircuits. The method is developed based on hybrid bond graph and genetic algorithm. Case studyshows the developed method is applicable to estimation both single fault and multiple faults ofparametric nature, and can gain accurate diagnostic results.(3) The extraction methods of system-level fault characteristic parameters of power electroniccircuits are presented in this thesis. The system-level fault characteristic parameters of the typicalDC/DC converters are obtained based on Least Squares Support Vector Machines. A BOOST circuit isconsidered as the illustrative example, the experiment result shows that the proposed method caneliminate the influence of working condition and reflect the degradation degree of the circuit.(4) On the basis of all the work done before, the component-level and system-level prognostictechniques are implemented for the power electronic circuits. A model-based prognosticsmethodology using Kalman filter is employed to track the state of health and predict the remaininguseful life of the electrolytic capacitor and MOSFET. Two data-driven approaches based on LeastSquares Support Vector Machines and Gaussian Process are introduced to failure prediction of theBOOST circuit.