The Sdudy On Candidate Diagnosis Solving Methods And Monotonicity Of Candidate Daignosis Space In Model-Based Diagnosis

Model-based diagnosis (MBD) is based on the related technologies to support the development of artificial intelligence. As a high flexibility of reasoning technology, it is replacing the method based on expert diagnosis of faults; and greatly promotes the developing of artificial intelligence. In recent years, with system integration, automation degree is increasing, and the demanded reliability and maintainability of the system is also increasing. The applications of model-based diagnosis are also widely spread. At present, the use of MBD troubleshooting and circuit system can be ruled out, including medical diagnosis system research, the network communication system fault diagnosis, fault diagnosis large cars, ships, etc.Early experts’ experience provides the basic for the traditional fault diagnosis methods, but the shortcoming is that the update of experts’experience is quite slow and also not easy to obtain. Moreover, once the system changes, experts’ experience is hard to use. While MBD doesn’t rely on the system structure, and thus the established model is independent with the whole system.System modeling is to use appropriate language to represent the system structure. Conflict identification is to find out the system of normal work conflicts widget set at the same time. The candidate to produce parts from the conflict focus to find possible fault components set, is really identify the fault components by increasing the measurement point out parts, to determine the fault components. Model-based diagnosis is composed of the above four processes.For system modeling, it is necessary to select the appropriate method, so that it can correctly represent actual system. Currently the first-order logic language is the best method.Candidate diagnosis generation, which is a procedure of generating all minimal hitting sets (MHSs), is proved as an NP-complete problem. Many researchers have studied the problem. However, some solutions may be lost by pruning in some methods, or a tree or a graph needs to be constructed with the corresponding data structure and algorithm being quite complex, or computing MHSs by recursion is rather inefficient in some methods, etc. In order to overcome the shortcomings mentioned above, a method of CHS-tree (Cardinality-based Hitting Set-tree) is put forward, to compute all MHSs (Minimal Hitting-Sets) based on all conflicts. The set with the smallest cardinality is selected from the current conflict set cluster. In some cases, the efficiency CHS-tree is higher than the current most efficient Boolean method.The identification of diagnoses is one of very important steps in model-based diagnosis. Its basic idea is to find the real fault component set in the candidate diagnostic space by gradually adding new more measurement points. We studies the optimization problem of choosing suitable measurement points, and analyses the monotonicity of the size of a candidate diagnostic space on the new measurement point, by the way of logic reasoning and checking. Then, we give some characteristics of measurement points, and based on which, the number of candidate diagnoses can be monotonic increase or monotonic decrease on the increasing relevant measurement points. Finally, we prove that with increasing the relevant measurement points, the corresponding candidate diagnostic space will be monotonic decrease, which provides a basic theoretical foundation for the optimization of practical measurement point selection.