Research And Implementation Of Health Management Technology In The Space Application Of Electronic System

With the constant development of systems and facilities to miniaturization, complication and intelligence, the problem about the reliability, security, maintainability and indemnificatory expense becomes more and more seriously. Then, Prognostics and Health Management, named PHM, whose major content is monitoring a system's state, diagnosing and predicting its fault and making decision about maintainability, will be the important basis and key research on future automatic logistic.The object of the research in this paper is an ordinary electronic system that running in space environment, based on which some research and applications to PHM's architecture and some key technologies will be done. It includes a design of hardware platform and applications to algorithms and the specific work are shown as follows:(1) Structuring the architecture of Health Management system, named HM, based on the independent minimized electronic system. The architecture is designed according to the thought of hierarchical logic functions and also based on analyzing and summarizing two typical architecture of PHM. Then, some key technologies are simply listed in this paper.(2) Establishing the healthy index system of the electronic system in space. As the first step of researching HM system, the establishment is completed by summarizing the principles and methods about it. Then, a hardware platform that is used to simulate the fault and verify the index system was designed and implemented on the base of 1-Wire sensors, which is integrated with the methods of collecting and transferring data. Finally,several indices was selected as a group, whose data from the platform will be the input of all the following research.(3) Building the healthy state models of the electronic system via clustering algorithms. Because of the autonomy of HM, the system's state models of entire data should be built by unsupervised clustering process with clustering algorithms and accumulated historical data. The analysis and verification shows that Fuzzy C Means algorithm and an anthropopathic clustering algorithm proposed in this paper can complete the whole cursory clustering and partial refined clustering respectively.(4) Evaluating the object system's state at some point based on its standard state models. The problem about evaluation of states can be turned to that about pattern recognition between sample set and standard set according to their data form. Three methods of pattern recognition has been used to verify the capability of evaluating states during the research, and the fuzzy pattern recognition based on statistics is proved to be the most reasonable and viable one.(5) Utilizing the trend prediction of system's indices to realize the trend prediction of system's state. The time series of system's index was analyzed and processed firstly, and then its prediction was accomplished with data-driven methods. Grey prediction model and BP neural network have been adopted in this paper and the comparison of their verification shows that the latter does well in the predictive performance of volatility time series.