The Research Of Model-Based Diagnosis And Its Application In The Dynamic System

Model-based diagnosis is a new type of intelligent reasoning technology, overcoming traditional fault-diagnosis methods'shortcomings. And model-based diagnosis is one of the active branches of Artificial Intelligence, and more and more widely applied. In addition to circuit diagnosis and medical diagnosis, nowadays, MBD has been applied to more practical aspects, such as fault detection and location in VHDL, diagnosis of asynchronization discrete systems, diagnosis of network communications, automotive systems diagnosis, monitoring of gas turbines, and so on.Traditional rule-based diagnosis mainly depends on experts'experiences. And it has strict domain-dependency too. However, model-based diagnosis has strong device-independency. In model-based diagnosis, domains and system models are apart, so that it can be used to diagnosis another system, with the corresponding system model be replaced only. Struss called model-based diagnosis as an important challenge and verify for Artificial Intelligence.Recently, the research of the static diagnosis is very mature, and the hotspot of MBD has turned to dynamic systems. Meantime, it is an open problem. Many experts try to deal with it via different methods. Meanwhile, the real-time requirement is also very important, we usually miss the opportunity to correct fault behavior after identifying it. So the focus of this paper is the most widely used one of the dynamic system of real-time hybrid system fault diagnosis, which are introduced to hybrid dynamic system.Hybrid dynamic system is a kind of complex system that include both discrete speciality and seriate speciality, and the both specialities interact each other. From 90th last century, hybrid dynamic system is becoming the focus to the domain of cybernetics, computer science and mathematics.Introduce a hybrid dynamic system model is helpful to a system with mode switch modeling to a subsection seriate behavior with discrete switch control, in order to predigest the modeling of complex system.As some nonlinear system can be abstracted into subsection seriate behavior system with discrete switch. Along with the producing system increasing the complexity and requirement of the efficiency, the need to the key technique in engineering and technique, faulty diagnose is becoming more and more exigent. Traditional faulty diagnostic method can divide into model-based diagnose and non-modeled diagnostic method. As the progress of DEDS theory researching, they also brought the diagnostic method based on discrete event system(DES).Introducing faulty diagnose of mixed system is based on consideration below: one side, the discrete change of a mixed dynamic system model can be not only used to describe controller's action and autonomic jumping(the switch is aroused by the model itself),but also modeling external behavior(it arouse some unexpected behavior such as system components failure.), introducing failure as discrete event into system, and consider it as a mixed dynamic system. This kind of method can bring a new angle of view to faulty diagnose research, and introduce new methods and new academic fragment to research technique in faulty diagnose of mixed system, accordingly produce transformable influence possibly. On the other hand, at the same time, we pay attention to that some of existent faulty diagnose techniques are suitable to seriate system, and some of them are suitable to discrete system, their connection have stimulative effect to develop faulty diagnose.Just about the infection of this idea, in recent years, many researchers pay interests in this domain, aim at some of problems, they bring forward some methods, and acquire some production.This work is mainly on the basis of theory and MBD hybrid system fault diagnosis and launched, including the behavioral model based on fault diagnosis methods of study; model-based diagnosis of hybrid dynamic systems modeling research based on the mutation, ease of change fault model diagnostic work, as well as fault diagnosis model of the system realization. Specifically, the following aspects:(1)A new method of using failure behavior in model-based diagnosis is proposed for computing all minimal diagnoses, and the computing procedure is formalized by combining revised SE-tree (set enumeration tree) with closed nodes to produce all the resolutions gradually. It can directly compute all the minimal diagnoses, without computing all the conflict sets and therefore the hitting sets of the collection of the corresponding conflict sets like the classical methods. And then the combinatorial explosion caused by calling ATMS (Assumption-based Truth Maintenance System), known as an NP-complete problem, can be avoided as well. As the closed nodes are added into the SE-tree, the non-minimal diagnoses can never be produced, and the true resolutions can not be missed by pruning, either. Results show that the corresponding algorithm is easily implemented, and the efficiency is highly improved.(2) The thesis introduces a modeling method, which applies MBD technology to the diagnosis of the hybrid system, as well as the basic principles of Hybrid Bond Graph. Through the modeling of the hybrid system with the Hybrid Bond Graph, this research provides a basic model for fault-testing, fault tracing and detection, fault isolation and identification.(3) Techniques for diagnosing faults in hybrid systems that combine digital (discrete) supervisory controllers with analog(continuous) plants need to be different from those used fordiscrete or continuous systems. This paper presents a methodology for online tracking and diagnosis of hybrid systems.(4)A semi-qualitative approach to diagnosis of gradually changing faults of a class of hybrid systems is proposed. Under the framework of hybrid bond graph, some improvements are further carried out. The strong tracking filter as a part of hybrid observer is introduced to insure the hybrid behaviors being tracked more exactly between different modes. In the qualitative reasoning, an improved temporal causal graph is proposed by integrating the fuzzy set knowledge. This makes the original method be extended to diagnosis of gradually changing faults of hybrid systems, and the reasoning time is greatly reduced at the same time.