The Diagnostic Algorithm And Diagnostic Strategy Of Model-Based Diagnosis

Model-based diagnosis(MBD) is one of the central study area of artificialintelligence. It is a intelligent reason method which can overcome theshortcomings of the initial diagnostic expert system. And it is applied indomains as different as networks, traffic control, telecommunication,networks, electronic device, space crafts, etc. MBD has many advantages. Themodel can be get when the device is first manufactured. And it is easy tomaintain ,expand and compute. Otherwise, it is machine-independent.The basic thinking is: express the system to be diagnosed as a abstractmodel. Then predict the system behaviour with the model and compare it withthe practical behaviour to find the fault components. MBD has two genres:consistency-based diagnosis(CBD) and abductive diagnosis(ABD). CBDrequire the system description(SD), observations(OBS), and abnormalityassumptions(△) are consistent. ABD, in addition, require that OBS can getfrom SD and △. So, ABD is a strong notion diagnosis. A ABD is a CBD, buta CBD is not always a ABD.Most computation of diagnosis is based on conflict recognition.Diagnosing every component is impossible because it needs much time, andmost work is unnecessary since most components are normal. So, only partsof the components need diagnosis, i.e. conflicts.Model and test are important too in diagnosis. A good model canimprove efficiency of a diagnosis and the diagnosis can get right result moreprobably. The result needs to be tested to ensure that the components are fault.In the area of diagnosis, many excellent theories and MBD systems weredeveloped. GDE is one of the easiest and most influential diagnosis system. Ituse the assumption-based truth maintenance system。When different valuesappear at the same point, take the assumption as its conflict. REVISE is adiagnosis system based on extended logic programming. It can revise theacknowledge base in diagnosis. With the priority, it provides a two-valuedrevision of assumptions to remove contradictions. Because REVISE is basedon extended logic programming, its result is reliable. DRUM embed diagnosisin a general logical frame and diagnose with the acknowledge of the systemmodel. It uses static pre-compiled information, so can solve large complicatedproblems with attractive time and space complexity. LIVINGSTONE2 is apart of the software of spacecraft "Deep Space 1". And it can detect, diagnose,repair automatically. Satisfy the demand of safety, maintainability, minimumand portability. With the development of MBD, the initial logic programming can notsatisfy the new demand any longer. So, it must be extended. It is morepowerful to express after the introduction of negation and integrity constraints.There are two algorithms computing revision. One is bottom-up algorithm. Itcomputes the conflicts with the support set of ⊥. It must begin to computefrom the computation of the model WFSX(P) and all conflicts will becomputed. So, its efficiency is low. The other is top-down algorithm. We canget a conflict by collecting default literals from the proof of a successfulT-tree of ⊥. Then, extend it to a hitting-set tree. And the revision can be getfrom the tree. So, only parts of conflicts need to be computed. In order to work as plan, the process of diagnosis need control. Thechoice of diagnostic assumptions and models is decided by working...