Research On Fault Diagnosis And Fault-tolerant Control Of Complex Systems

This paper studies the method based on multi-agent system for fault diagnosis and tolerance of complex systems. First, the approach based on multi-agent system for analysis, Gaia modeling and fault diagnosis of complex system is put forward. And in the research of fault diagnosis, an advanced algorithm of binary decision diagram (BDD) is proposed to speed up the search of fault source of fault trees. Then the method of establishment and reasoning of fault diagnosis knowledge using ontology language is also proposed to improve the understanding among agents during their interaction, and this method can be used to build the knowledge base of fault diagnosis of aircraft control surface. Finally, the method of multiple-fault diagnosis based on multi-agent system combining neural network and fuzzy control is studied. And all the methods are verified by simulation using examples of combination fault diagnosis and fault-tolerant control of F-16 aircraft control surface. The main content of this paper is as follows:In the research of modeling and fault diagnosis of complex systems based on multi-agent system, the Gaia method is used to modeling complex system. Then the roll models and interaction models of fault diagnosis agent are analyzed and built according to their functionalities. At the same time, the behavior and cooperation algorithm are designed in detail for this agent. And the function and characteristics of fault diagnosis for complex systems based on multi-agent system are also analyzed.In the research of specified method of fault diagnosis, in order to overcome the limitation of the requirement of a predetermined sequence of basic events, two new rules are added to the connection rules of component connection approach for fault tree conversion to Binary Decision Diagram (BDD) to ensure the unique structure of the final integrated BDD. Then through comparing the probabilities of cut sets with the same length of Huffman code, the ordering of checking the fault source can be determined. This method can be applied to digital systems of which the multiple-fault mechanism can be expressed by fault trees with independent basic events. Because this method has no need to simplify the integrated BDD and determine the minimal cut sets, it's more suitable for computer execution and has higher efficiency than traditional methods of fault tree diagnosis.In the research of interactions among multi agents, for the problems of acquisition and reasoning of fault diagnosis knowledge, the complexity of aircraft system was analyzed from the engineering point of view and the concept of aircraft family was introduced to the ontology modeling of aircraft. At the same time, the fault diagnosis of aircraft control surface was also studied. First, the domain knowledge model of aircraft ontology was built by Protégé. Then the knowledge of single fault and combined faults diagnosis were listed as SWRL rules. Finally, the diagnosis result could be concluded through JESS reasoning to select the right plan of self-repairing. This method can be used to realize the modeling of complex systems and accurate selection of fault diagnosis plan. And the knowledge base of fault diagnosis can also be improved by adding new rules.In the research of fault tolerance of multi-agent system, the basic concepts of multiple-agent systems (MAS) are expanded to build the related concepts of fault tolerant multi-agent system (FATMAS). And based on all these concepts, the definitions of various functions are also achieved to realize multiple fault tolerance in FATMAS. The multiple faults can be detected by calling different functions of the system, and the system can recover from multiple faults to run by copying the tasks of non-critical agents. The relevant flows are listed and an example is given to illustrate the application of this method in the aircraft multiple fault diagnosis and self-repairing in JACK platform. This method can be used in complex system for multiple fault diagnosis and tolerance. The clear inter-agent message passing mechanism is also provided and the number of replicated agents is effectively reduced. So this method is relatively close to the process of dealing with failures in actual system.In the end, through the analysis and selection of existing multi-agent development platforms, and the modeling of fault diagnosis agent, the online fault diagnosis and monitoring systems of aircraft control surface are realized based on JADE and Simulink. During this process, MACSim is used to realize the information exchange between the two platforms. The system maintenance, real-time fault diagnosis and monitoring capacity of the whole system has been improved to some degree, and the coordinated fault diagnosis of combined faults can be realized to achieve the required performance.