Research And Implementation Of Data-driven Diagnostic And Prognostic Methods For Electronic Systems

According to statistics, more than 80% failures are from the analog circuits in electronic system equipment. Thus, the analog circuit fault diagnosis technology is the key and difficult issue in electronic system health management. In addition, with the development of the electric vehicles, lithium-ion batteries have been widely used. The lithium-ion battery life is a key factor for electric vehicle performance and safety. Therefore, the lithium-ion batteries life prediction is an another important research issue in electronic system health management. Based on the above reasons, major work in this paper are summarized as follows:1. Study on analog circuits fault diagnosis method. Based on the pattern classifier nature of support vector machine(SVM), the statistical features such as mean, variance, standard deviation, entropy, kurtosis, skewness, and centroid, are proposed to constitute the fault feature vector. In current, there exists a shortcoming when the SVM is used for analog circuit fault diagnosis, that uses the same feature vector combination to train all SVM binary classifiers. However, each SVM binary classifier has different classification accuracy for different feature vector combinations. To solve this problem, the particle swarm optimization(PSO) is proposed to select the near-optimal feature vector combination for each SVM binary classifier. The experimental results show that the proposed method improves the fault diagnosis accuracy.2. Study on lithium-ion battery life prediction method. Based on the life prediction principle, the lithium-ion battery life prediction overall framework is proposed. First, the Verhulst model is used as the lithium-ion battery life degradation model. Owing to the traditional Verhulst model prediction accuracy is not high, the PSO algorithm is used to improve the Verhuslt model prediction accuracy. Second, estimate the Verhulst model parameters. The PSO algorithm is also applied to search the Verhulst model parameters in our paper. Finally, in order to reduce the impact of noise, the particle filter(PF) is used to update the Verhulst model parameters timely. The experimental results show that the proposed method could predict the lithium-ion battery life with small error.3. Fault diagnosis and prediction system software design. To meet the demand of fault diagnosis and prediction in electronic systems, fault diagnosis and prediction system software are developed respectively. Fault diagnosis system integrates the SVM algorithm, and can calculate the fault feature for the external data automatically. Moreover, the fault diagnosis results are given rapidly. Fault prediction system integrates many prediction models, such as Verhulst model, GM(1,1) model and AR model, and can select the opposite prediction model for external data sequence automatically. Based on the experiments validation, fault diagnosis and prediction system software could give the accurate results efficiently.