Research On Prognostics Strategy Based On Parameter Identification For Power Electronic Circuits

The prognostics and health management(PHM)have very important significance for ensuring the security of systems.The widely use of power electronic equipment is now making the PHM technology for power electronics become more important.PHM can be divided into two parts: fault prediction and health management,in which the former is the foundation of the latter.This thesis focuses on the parametric failure prediction for power electronic circuits based on performance(feature)parameter extraction.A technique based on system identification of feature parameter extraction for power electronics is adopted in this paper.The accuracy of conventional parameter identification of the Buck converter suffers from too many non-ideal factors in practice.In this regard,a more widely used parameter extracting method for the Buck converters is proposed by establishing a linear model containing target components of the circuit.In addiction,the identified values of capacitance and equivalent series resistance of the filter capacitor are deviated from the true values in the original method of parameter identification for the Boost converter.Aiming at this problem,the related reasons are analyzed and a model modification strategy is put forward in this paper.Moreover,power electronic topologies with non-resistive loads are widely adopted in practical applications.Corresponding models are required to be built when the load structure varies.Given the shortcoming stated above,the model establishment and parameter extraction methods proposed in this paper are suitable for converters with different sorts of loads as the output current is monitored directly.Then,the performance including convergence speed and identification precision of the presented methods are verified in Matlab and the experiment platform,which provides effective basis for the future prognostics of power elctronics.Finally,parametric failure prediction for power electronics is preliminary explored with the assist of simulation in this paper due to limited experiment time and cost.The parameter RC is selected as the example and its variation in the simulation circuit is set according to existed emprical formulas.Additionally,time series of RC are obtained by identification strategies listed above.However,prediction error led by variable working condictions is ignored in the primary prognostics for power electronics based on feature parameters.Hence,classification pretreatment of identified parameters according to stress levels experienced by the device is adopted and the improved prediction strategy is confirmed with the LS-SVM toolbox in Matlab.