Research On DC-DC Converters Prognostics And Health Management Based On Machine Learning

Switching power supply technology can realize high-efficiency power transmission.It has a wide range of applications in military and civilian fields.However,switching power supplies usually work at high frequencies and have strong nonlinearity,which brings stability and safety challenges to the entire circuit.After the switching power supply fails,the whole equipment will be shut down,and the safety of people’s lives and property will be lost.With the advancement of machine learning and artificial intelligence technology,research institutions all over the world are hoping to gradually reduce the losses caused by the traditional "timing guarantee",and actively carry out the "conditional guarantee" based on fault diagnosis and health management of equipment.Therefore,the research on Prediction and Health Management(PHM)of electronic equipment has gradually entered the field of vision of domestic and foreign researchers.However,due to the relatively short research time in this field,there are currently few research results in this field.In the future,PHM technology will be the focus of active exploration by scientific researchers.Prediction and Health Management(PHM)can evaluate the reliability of electronic equipment in practical applications and make relevant recommendations for future use.The application of PHM in DC-DC power supply can monitor the health of the system,predict the Remaining using life(RUL)of the product,and realize fault location.This article analyzes the health degradation process of the switching power supply in detail,selects the circuit transfer function as the monitoring parameter,and uses the Hidden Markov Model(HMM)to establish a complete switching power supply PHM system.By monitoring the key parameters of the circuit transfer function,accurate positioning of faulty components is achieved,with an accuracy rate of over 96%;real-time monitoring of the health status of the power supply and prediction of the remaining using life of the power supply.Compared with the previously commonly used fault location and life prediction methods that use the output voltage as the monitoring quantity,the use of key parameters of the circuit transfer function as the monitoring quantity makes the fault location and life prediction easier and more accurate.This paper mainly focuses on the BOOST circuit of DC-DC switching power supply,and the following research work is carried out:(1)In view of the shortcomings of previous research on power health management that can only deal with a single device failure,this article fully considers the situation of multiple device degradation in practical applications,and designs a Boost power degradation circuit,which can reflect the multiple failures of power.The model breaks through the limitation that power management can only be performed through the equivalent resistance of the filter capacitor;in terms of power health management,most of the research stays on life prediction.This paper adds two models of power failure diagnosis and power health monitoring to improve and developed a power health management system.(2)For the output signal of the switching power supply circuit,the frequency characteristics and mathematical statistical characteristics of the circuit transfer function are selected as the monitoring parameters,and the power supply characteristic vector is constructed based on this.Since the feature vector is extracted based on frequency features and mathematical statistics,the magnitude of the feature is quite different.In order to solve the problem of inconsistent size,the feature vector is normalized,which not only solves the dimension problem,but also speeds up the machine learning model convergence speed,improve the efficiency of failure prediction and health management.In view of the high dimensionality of the feature vector,there is a hidden danger of dimensional explosion,the PCA algorithm is used to reduce the dimensionality of the feature vector,which reduces the consumption of computing resources and reduces the complexity of the model.(3)Aiming at the fault location research of switching power supply with multiple device degradation and multiple failure modes,this paper uses the SVM algorithm to establish a fault location classification model.Aiming at the multi-classification problem in fault location,this paper adopts corresponding multiple SVM two-classifiers to realize the SVM multi-fault classification so as to realize the fault location of the power supply.(4)It is proposed to use Hidden Markov Model(HMM)to monitor the health of the switching power supply.The health of the circuit is regarded as the hidden state of the HMM,and the fault feature vector that can be extracted is used as the observed value of the HMM.First,collect the raw data of the power supply through Monte Carlo simulation in Pspice,and extract the fault feature vector from the raw data to preprocess its features.Then,the Baum-welch algorithm is used to train the Hidden Markov Model(HMM)parameters.Finally,for the health status monitoring of the power supply,the most likely health status of the current power supply is calculated online through the Viterbi algorithm.(5)Aiming at the prediction of the remaining using life(RUL)of the switching power supply,a linear regression algorithm is used to model the degradation process of the power supply.First of all,the fault feature vector in this paper is multi-dimensional,and a theory based on vector similarity is proposed to fuse multiple fault feature vectors to obtain the remaining service life indicator.Then,in order to predict the remaining service life of the power supply,the linear regression algorithm is used to perform curve fitting on the RUL indicator,and the stochastic gradient descent algorithm(SGD)is used to update the model parameters of the linear regression algorithm in real time.Finally,according to the current state of the power supply,the power supply RUL is predicted through the trained regression model.