Fault Diagnosis Of On-orbit Spacecraft Based On Evidence Theory

Fault diagnosis of on-orbit spacecraft is an important technology in space missions. With the flourishing development of China's space industry, newer and higher requirements in the process of on-orbit spacecraft fault diagnosis are needed. Fault diagnosis of on-orbit spacecraft is a research topic with great meanings and challenges. Evidence theory, as a method of uncertainty reasoning, has flexible and convenient characteristic in expression and reasoning of uncertain information. The treatment process to uncertain information is similar to the thinking habit of human experts. Therefore, evidence theory is usually used in information fusion as the theory basis and can be used for fault diagnosis according to multisource information. Faults in on-orbit spacecrafts have common features, such as multisource information representation, uncertainty, sudden and timely treatment. These features make evidence theory suitable for fusion diagnosis.Combining the requirements of on-orbit spacecraft fault diagnosis and the advantages of evidence theory in information fusion, fault diagnosis of on-orbit spacecraft based on evidence theory is studied. Common features of multisensor data measured in the on-orbit spacecraft faults and common requirements of fusion diagnosis are considered. General generation of evidence from fault feature is designed and general model of fault information extraction in the diagnosis frame based on evidence theory is proposed. The general methods are specially realized for two kinds of typical faults in on-orbit spacecraft. Fusion diagnosis experiments were carried. Exploratory research of fusion diagnosis of unrecorded fault based on evidence theory is introduced finally.Effective generation of fusion evidences and accurate expression of uncertain information are essential during information fusion using evidence theory. Choosing of combination rule reflects fusion thought. In Chapter 3, general fusion diagnosis frame based on evidence theory is designed, including automatic generation of evidence according to multisensor data information, decision for diagnosis, and comparison of combination rules. Firstly, an existing method of generation evidence from sensor data is introduced and its disadvantages are analyzed. According to the thinking of the existing method, a more reasonable and general method for evidence generation is designed. Decision principles of fusion diagnosis are proposed then. Three common combination rules of evidence theory are compared and evaluated, while information entropy is chosen as the measurement of uncertain. Applicable conclusions are got and these can be the theoretical basis of using Dempster's rule in the following fusion diagnosis experiments.Fault diagnosis based on evidence theory belongs to decision level information fusion. The decisions before fusion directly affect the fusion result. The evidence generation in this thesis directly uses multisensor data information. Extraction of fault information from multisensor data is the key to the question. General fault information extraction model is designed in Chapter 4. Characteristics of on-orbit spacecraft faults are studied. Concept of diagnosis window is proposed then. General expression of fault information, using statistic of data segment, is devised. The fault information extraction model for fusion diagnosis of on-orbit spacecraft fault is designed.Availability of general method is tested by two kinds of typical faults in on-orbit spacecraft. Cooperative working fault and vibration divergence fault are chosen for the fusion diagnosis experiments in Chapter 5. Cooperative working fault diagnosis experiments use the measured data in the space mission while vibration divergence fault diagnosis experiments use representative vibration isolation system vibration divergence faults data. Diagnosis processes are realized in terms of fault representation, evidence generation and fusion diagnosis. Fault-tolerance improvement is made for the abnormal transmissions of sensor data. Evidence redistribution is designed for the fault-tolerance problem.Diagnosis of unrecorded fault has important meaning. However, the existing fusion diagnosis frame based on evidence theory can not realize unrecorded faults. Exploratory research of fusion diagnosis of unrecorded fault based on evidence theory is taken in Chapter 6. Evidence expressions of unrecorded fault are analyzed and the"open world"thought and evidence discount method are chosen as the mechanism of fusion diagnosis. Fault focal elements with no intersection and with transitive intersections are considered separately. Theoretical derivations of discount fusion process in open frame are taken. Simulation experiments are designed to verify the effectiveness and rationality of the fusion diagnosis of unrecorded faults.