Research And Implementation Of Several Methods For Fault Diagnosis Based On Model

Traditional methods of fault diagnosis mainly rely on experts’ experience and knowledge. Unfortunately, experts’ experience is not easy to obtain, and is slow to update. Once the system changes, experts’ experience will be insufficient for fault diagnosis.Model-Based Diagnosis (MBD) can overcome the shortcomings of traditional diagnostic methods. It does not rely on experts’experience. The model of the system built by the MBD is independent. MBD has the advantage of portability and high recycling rate.Initially, MBD is only used for static system, and recently model-based diagnosis of Discrete-Event Systems (DESs) has received more and more attention. The basic idea of diagnosis of DESs:Firstly, construct the system model by using such as finite-state machines, process algebra, Petri nets, and others, to describe discrete dynamic systems. Secondly, monitor system operation to obtain observation information. Finally, synchronize observation and the system model to infer the track of the system, which contains possible fault information.Because of the incompleteness of the system model or the obtained observation, we usually have to consider the diagnosability of the system model. That is, given enough observation information, whether one can determine a failure has definitely occurred or not.In this paper, based on finite-state machines for modeling discrete-event dynamic systems, some related diagnostic methods and diagnosability were deeply analyzed. The main contents are listed as follows.1) Classical diagnostic approaches for discrete-event systems are studied. That is, a diagnoser is constructed to locate the fault, and the diagnoser can also be used for determining the diagnosability of a system.2) The detailed algorithm for classical diagnoser is given and implemented.3) An improved model twin-plant for diagnosing DESs is studied, which has a lower computational complexity than the classical diagnoser method.4) The detailed algorithm for twin-plant method is given and implemented.5) A diagnostic platform for DESs is implemented. The platform integrates classical diagnoser method and the improved twin-plant method. In future, it can also provide possible extension for other diagnostic methods.6) Online diagnosis method based on two successive temporal windows is studied. The detailed algorithm is implemented as a prototype system, to infer fault diagnosis in real-time.