A novel approach is presented in this thesis that exploits uncertain information on the behavioural description of system components to identify possible fault behaviours in physical systems. The result is a diagnostic system that utilises all available evidence at each stage. The approach utilises the standard conflict recognition technique developed in the well-known General Diagnostic Engine framework to support diagnostic inference, through the production of both rewarding and penalising evidence. The penalising evidence is derived from the conflict sets and the rewarding evidence is derived, in a similar way, from two sets of components both combining to predict the same value of a given variable within the system model. The rewarding evidence can be used to increase the possibility of a given component being in the actual fault model, whilst penalising evidence is used to reduce the possibility. Markov matrices are derived from given evidence, thereby enabling the use of Markov Chains in the diagnostic process. Markov Chains are used to determine possible next states of a system based only upon the current state. This idea is adapted so that instead of moving from one state to another the movement is between different behavioural modes of individual components. The transition probability between states then becomes the possibility of each behaviour being in the next model. Markov matrices are therefore used to revise the beliefs in each of the behaviours of each component. This research has resulted in a technique for identifying candidates for multiple faults that is shown to be very effective. To illustrate the process, electrical circuits consisting of approximately five hundred components are used to show how the technique works on a large scale. The electrical circuits used are drawn from a standard test set.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:662071 |
| Date | January 2002 |
| Creators | Smith, Finlay S. |
| Publisher | University of Edinburgh |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/1842/11417 |
Page generated in 0.0026 seconds