Research On The Performance-degradation With Singular Signal For Trend Prediction Of Electronic Systems

In recent years, for fully improving the diagnostic performance and intelligent decision-making level of electronic system, the trend prediction techniques based on performance-degradation data has became one of the hot topics of the current academia and electronics engineering. Many previous researches have focused on regular performance-degradation data. However, due to some transient or intermittent faults occurring in the degradation process, there always exist some tiny singular signals in these degradation data, which will greatly reduce the forecasting accuracy of time series, and the prediction result may also have a serious deviation from the actual curve. It is a difficult problem, yet to be adequately resolved. Therefore, research on the performance-degradation with singular signals for trend prediction of electronic systems is an important research issue. Based on the above reasons, major works in this paper are summarized as follows:1. The studies on the analysis of singular signal in the degradation process: This part is the foundation of the full text behind the work. To this, the paper first introduces the mathematical model of the singularity, then focuses on the the generation of singular signal and its influence on prediction in the degradation process. By the analysis, the singularity seriously reduces the life prediction accuracy, and has a negligible impact on the data prediction after it. What’s more, the difference between the singular signal and the noise, and the effect of the noise on the singularity detection are also analyzed. The results have shown that the existence of noise not only affects the signal singularity detection, but also seriously affect the quantitative characterization of singular degree. It further points out that the current researches on the performance degradation with singular signals for trend prediction of electronic systems—first identify the singular data from a time series, then remove them and carry out the prediction to the processed time series which no longer contains the singular data—have a lot of deficiencies. Thus, it provides a basis for the following research.2. Two methods about the trend prediction of electronic systems for performance degradation with singular signal are proposed in this paper: Firstly, an approach based on spline hidden Markov time series modeling is put forward. The HMM takes the spline cell as the observation sequence, determinates the optimal spline cell and outputs the optimal spline parameters. As the output prediction cell of an HMM, the cubic non-polynomial spline is an effective model to generate the good forecasting data without first identifying and eliminating the singular points in a time series. Secondly, taking into account the optimally-pruned extreme learning machine(OPELM) model and the Volterra series are inherently close contact and similarity, in this paper, we propose to utilize the Volterra series accurate modeling and the OPELM algorithm within a very good compromise between the computational speed, accuracy and robustness to establish VKOPP model, which effectively guarantee the accuracy of time series forecasting for electronic equipment. The major advantage of these two proposed methods are that they don’t need to identify the singularity in a time series, and this is very important in some cases of no effective method to identify singular points. By analyzing some simulation experiment results, it can draw the conclusions that our models are potentially very useful for singluar degradation time series data analysis, and it can be adapted to a broad range of applications for time series forecasting.3. Predictive analytics software design and implementation: This predictive software integrates 9 different prediction models, including some classic prediction algorithm and the methods proposed in this paper, and can select the opposite prediction model for external data sequence automatically according to each model prediction accuracy and time-consuming analysis. In addition, when the degradation data with singular signals, the users can obtain the singularity locations and values by calculate the wavelet transform modulus maxima curve and Lipschitz Exponent through this software. Based on the validation for the Li-Ion batteries performance-degradation data set which was provided by the Prognostics Center of Excellence at NASA Ames, predictive analytics software could give the accurate results efficiently and can provide real-time and reliable theory basis for the health management of the electronic system.