Research On Improvement Method Of Model Based Diagnoses In Discrete Event System

In modern industry and aviation, the behaviors of equipments always shown overwhelmingly complexity, the behaviors in the running of telecommunication net and the detecting task of lunar rover in the unknown surrounding could be vividly examples. The same characters were appeared in the interaction, testing and debugging of large software components. These software and hardware systems were typically in large scale and behave in a very complex interaction way. It was impossible to make sure all the components to behave correctly, in fact. Once the systems access fault or even paralysis down by the fault, the consequent damage could be enormous. The fault diagnosis method was proposed for this problem. The early fault diagnosis method just could diagnosis for one system. The method built system description according the domain experts'knowledge couldn't diagnosis other systems'failures, and it is just one of the shortcomings of the method. This disadvantage was an obstacle for the diagnosis within the swift developing in industry and aviation application. Under this motivation, method of model based diagnosis was proposed.Model based diagnosis method build system's model according the inner structure and behavior of the subject system, and represent the model by a first-order propositions set. After the building of model, we could simulate the behavior of the system, such as the process of input, output and the action of observation. The diagnose analysis those behavior and compute the problems in the model, to get the logic result from the reasoning procedure. The diagnoser could approach the actual state of system, such as abnormal or not, and if the state is abnormal the diagnoser could detect the fault component and try to reason about how the failure happened.According the state is dynamic or not, when the diagnosis process undergoing, the diagnosis methods could be divided into static diagnosis and dynamic diagnosis. The static diagnosis means that the method just could handle the system diagnosis problem under the static assumption, and the diagnosis method could be named as dynamic diagnosis if the static assumption relaxed. Obviously the dynamic is more suitable for real problems.Discrete event system diagnosis is a dynamic diagnosis. The discrete event system is a dynamic model, but not dynamic overall, especially the event are assumed as ideal event, it does not need time span for execution. The state in the discrete event system is"jump"from one to another. The model are represented as automaton, and the system's behavior with a time span could be represent as state; and those without time span were represented as events. All of the messages transported by the sensors were represented as observations, which could be synchronized with trajectory in model, to monitor the running of system.Discrete event system could capture some characters of dynamic system. The object system could changing when the diagnosis procedure running on. And the running system could be damaged by some failure, the diagnosis process of diagnosis must be very efficiently therefore. Modern applications are always enormous and typically need thousands of states to represent, and the searching space is very large. And the critical challenge is to find the correct diagnosis in the shortest time.We proposed some method to deal with the time shorting diagnosis in huge searching space. Those method are mainly separated as pre-process for the model, and space pruning and automation distributing.The pre-processing process are typically focused upon the components emerge, to propose the emerge method for different component's automations and build the isomorphism automation. Those automation have the same state and with the same translation set, and only has different in observation, could be defined as isomorphism automation. Those isomorphism process could simplify the distribute automation. In the real diagnosis process the isomorphism process take those different automation model and observation in to account only. And the diagnosis could just compute those automation with the same structure in linear-time, avoid the redundancy.We proposed a state coding algorithm to pre-process the distribute automation, since all the state in the path could provide some information about the system conditions, we could find a way to remark those path in a uniquely way. The state coding is to store information in the model, and when diagnosis we could analysis the state code to find the most possible diagnosis path. The advantage of the method is easy to find the diagnosis path, and doesn't need too much observation results to diagnosis. But the method need exponential coding process, and will be worse with the number of states growth.To overcome the above disadvantage of exponential length coding for distribute automation, we propose a conflict-based diagnosis method to deal with the big scale automation in diagnosis. In the diagnosis, the state connects with the events by prior probability and posterior probability. Under the assumption of correct observation sequence, every observation could be used as a fact. Based upon the independent fact in the probability theory, we could compute the conflict between the states and events. If some observation independence with some state in the automation or events, we could detect a conflict, and those states could be the prune point of the searching algorithm. But, when there is not any conflict between the observation and states and events, we could re-compute the posterior probability and find the most possible path to isomorphism. This method could be advantaged in the automation diagnosis with lots of states; every state could be induced by several events. And the disadvantage of the method is too heavy computation work upon the probability information.All the methods proposed above were based on synchronization process. And synchronization was the center part of computation in discrete event system diagnosis. When we couldn't deduce the fault just based upon the distribute automaton, synchronization between automaton based on communication events mainly, and keep all path from the initial states to termination states. This kinds of synchronization was suitable for automaton with intimate relations and embodiment causal relationship between those automatons. The process could be a good attachment part for the minimum hitting set diagnosis method, when it get into incomplete causal relationship solve space.After the building of automaton model, the process of synchronization could approach the diagnosis results. And the result could be named as a deep diagnosis, to emphasis the character of the result not only provide which component get in fault state but also provide how it become bad event by event, and why the current state could be fault of attached by a fault token. This kind of diagnosis could provide much more information for the fault recover, the logic next step of diagnosis.The diagnosis is not just the purely synchronization process. We need to build the model and accomplish a diagnosis able verification procedure, or judge whether the method could approach a meaningful result within the time limitation. The whole problem also includes how to arrange correct observation sequence, and how to isolate faults and recover them. Those problems could be research in future.