Research Of Failure Prediction Of BUCK Circuit Based On Mahalanobis Distance

With the Prognostics and Health Management (PHM) technology continues to evolve, hardwaresand softwares with PHM technology have been used in a variety of electronic devices. The PHMtechnology increases the reliability and safety performance of electronic equipment substantially, andso does to economic benefits. And the reliability of power pack is so important to the performance ofelectronic equipments that ensuring the reliability of power electronic circuits is especially importantin the equipments’ security.The failure prediction of power electronics circuit is of great significance.This paper analyzes the research status of power electronic circuit failure prediction, and the hotpoint, difficulty and trends of it. And it also discusses the thoery of least squares Support VectorMachine and BP neural network and Genetic Algorithm. It analyzes the degradation of components inthe Buck circuit; And builds the control limits chart with the characteristic parameters of theMahalanobis distance (MD), to estimates the health degree of Buck circuit by probability method, soto monitor the health status of Buck circuit. On this basis, this paper does some failure predictionresearch of Buck circuit with algorithms of least squares support vector machine (LS-SVM) and BPneural network, while the prediction process was optimized by arithmetic of grid search and GeneticAlgorithm. Then this paper estimates the remaining life of the circuit. In addition, we analyze theBuck circuit output waveform with wavelet decomposition, and get the control limits chart, healththreshold and correlative fault feature parameter. After that we predict the status of the BUCK circuit,and discuss the pros and cons of this method. The above research work has carried on the simulationexperiment and verification in MATLAB, to verify the feasibility and validity of the method whichproposed in this paper. Finally, it presents a summary of the research and some follow-uprecommendations.