Research On Some Problems About Model-Based Diagnosis

With the rapid development and application of the electronic technology in modernindustry, the integration level and the complexity grade in the mixed-signal circuit alsodramatically increase. In the last two decades, possible faults appear in a number of keylarge-scale modern equipments, resulting in significant losses. In order to avoid theseriousness of the disaster from occurring, diagnose problems attract more and more attention.A recently developed theory of model-based diagnosis provided a uniform method for test ofthe digital and analog signal in the mixed-signal circuits. Model-based diagnosis is adiagnostic method which uses internal system constructions and behavior knowledge todiagnose possible faulty functions of a system.In this paper, to achieve the diagnostic goal, we propose some methods about associationinformation between outputs and components, computing the minimal test set, simplifiedmodel, diagnosability of system, incomplete model and so on. Details as follows:(1) Basedon Assumption-based truth maintenance system (ATMS), the paper proposes a method to getthe reasonable multiple faults diagnosis solutions based on the relationship between outputsand components in systems. In this paper, we set up a bilayer model which is a sort ofLayered Structure of Abstraction based on the relationship between outputs and components.This model contains all the components information of system outputs and allows us to obtainminimal diagnoses directly by obtaining the related components based on the informationassociated when the faults occur. It avoids deriving the minimal conflict set and the minimalhitting set based on the conventional diagnosing methods, and which can remarkably improveefficiency. Furthermore, we apply this method to MFMC and it shows that the method hasgreat flexibility and application.(2) In the mixed-signal circuit, test and diagnosis of thecircuit under test (CUT) have been studied widely. But the computing of the minimal test setof circuits is still a key and difficulty in this field. A recently developed theory of discreteevent system (DES)[13] provided a uniform method for test of the digital and analog signalin the mixed-signal circuits. In an off-line model, the fault was described by state not event,and the phenomena of the fault were defined by events. So the process of diagnosing was touse the phenomena to divide the fault into partitions until the partition was fine enough. Asfor finding the minimal test set of circuits in the testability study based on DES, We also propose a new algorithm for computing finer partitions. Furthermore, we give a new methodfor computing the minimal test set, which is efficient for computing the optimal partitioningand the minimal test set.(3) As the trend of exponential growth of system’s scale, we give amethod to reduce the redundant information of the model and propose some rules of mergingdifferent states. In this research, we give some reducing rules of DES model. These rulescould reduce the unobservable events and merge the states connected by unobservable events.What is more, we reconstruct the paths leading to these states based on new rules, so that, theobservation could be compatible with the model.(4) In the same time, we give two differentmethods of diagnosability testing and optimizing: Matrix-based testing algorithm andreverse-twin-plant-based testing algorithm and there is no need to generate a complete copy ofthe DES model to determine the diagnosability in all the two methods. Our methods onlyconsider the paths on sub-model instead of all paths on global model. So the execution time islargely reduced, and the methods are feasible in actual application.