Diagnosis Based-on The Relationship Between Terminal Output And Components

Model-based diagnosis rises from the mid-70s with the technology of intelligent reasoning, is a very active field of artificial intelligence, an important branch of artificial intelligence research in the field which plays a significant role. The main idea of MBD is the right system modeling, while there are differences between predicted values and observe values, we diagnosis the system and find the problem. The researches of MBD includes of the theory and the application of model-based diagnosis, experienced from the static diagnosis to the dynamic diagnosis and then to model-based reconstruction system, through the establishment of the system model, use the internal structure of the system as well as its component behavior knowledge for diagnosis.The traditional way of diagnosis are computing the Minimal Conflict Sets and then generating the Minimal Hitting Sets. It is well known that the number of diagnoses can be exponential in the size of system even when the preferred diagnoses are restricted(e.g. minimal diagnoses). Many researchers made great efforts to decrease complexity of diagnosis in space and time, such as the use of the system logical structure and components behavior to obtain diagnostic solutions, the introduction of binding rules to strike the problem of diagnosis.The paper investigates the diagnostic problems from the aspect of relationship between components and terminal outputs of system models, takes the relationship into consideration and abstracts the whole system through Area we proposed in this paper, which can decrease the space complexity of diagnosis, then compute the sets of diagnoses by analyzing the abstract diagnoses generated. In particular, the paper proposes the method for pre-process encoding the domain theory of the system model in terms of areas which are partitioned by relationship between components and terminal outputs. Moreover, the paper describes how to solve diagnostic problems through the information of areas, then gets the diagnoses of the problem while avoiding the procedure of traditional way for computing the Minimal Conflict Sets and then generating the Minimal Hitting Sets. The paper gives an approach to extracting diagnoses with the minimum number of faults from the explanation of areas which represent the entire space of diagnoses. To explain the idea of ours clearly, this paper uses the ISCAS85-benchmark C17 Circuit as an example and shows the procedure of our algorithm in detail.Second, we combine the Area we abstract from the relationship between the terminal outputs and components into the replacement test techniques. Replacement test is a kind of test, which replace the suspicious components and observe the results of its outputs again, according to its outputs observations to determine whether the replace component is failure, until the outputs values are consistent with the expected value, all the components be replaced can lead changes in the terminal outputs are faulty components.In real life, when the system when the system abnormal, the single component failure resulting exception may be far greater than the multi-component joint failure, the more multi-component joint failure, the less likely, therefore, in the replacement test , how to select the appropriate replacement of components is an important issue of concern to us, from the probability point of view we prefer to choose the diagnosis with the less component numbers to be first replaced to enhancing the efficiency of replacement test. However, the diagnoses we get are usually disorder unless we reorder the diagnoses, so we take the components in Area be replaced. For Area are abstracted from the relationship between the terminal outputs and components, which procedure is finished before diagnosis, therefore the abstract of Area does not affect the efficiency of replacement test, when choose the components to be replaced we give priority to the components in the Area which affect terminal outputs are consistent with the abnormal terminal outputs. If observation of terminal outputs doesn't change, once again choose component in the Area to be replaced, unless the observation changed, then compare the observation to the predicted values and continue the replacement test likely until the observation is the same to he predicted values. Then the set of all the components be replaced that can lead changes in the terminal outputs is diagnosis.Finally, we take the ISCAS85-benchmark C432 Circuit as an example to demonstrate the effectiveness of our algorithm with results of some diagnostic problems, and the results show the method in this paper is efficient enough for on-line diagnosis. The method in this paper is especially useful to the model with components'behavior may be not completely given and terminal output abnormality are known.