Expert systems in on-line process control and fault diagnosis
In this research expert systems for on-line process control and fault diagnosis have been investigated and the majority of the research is on using expert systems in on-line process fault diagnosis. Several on-line expert systems, including a rule based controller and several fault diagnosis systems, have been developed in this research and are reported in this thesis. The research results obtained demonstrate that rule based controllers can be used in situations where mathematical models for the controlled process cannot be obtained or are very difficult to obtain. The research on on-line fault diagnosis emphasises deep knowledge based approaches. Two avenues in deep knowledge based approaches, namely causal search and qualitative modelling based diagnosis, have been investigated. In the approach of causal search the research results reveal that diagnostic efficiency can be achieved through structural decomposition. A systematic approach for developing diagnostic rules based on structural decomposition is presented in this thesis. Much of the research has been done on qualitative model based fault diagnosis. A qualitative reasoning method which utilizes knowledge on the quantitative relations among variables to reduce ambiguity and can cope with a wider range of situations than Raiman's Order of Magnitude Reasoning is investigated. In the qualitative model based diagnosis the function of the qualitative model is to predict the behaviour of the process under various hypotheses and, therefore, to verify these hypotheses. Further research concerning self-reasoning has been done for the qualitative model based diagnosis approach. Self-reasoning is achieved by backward tracing through the model of the diagnosis system and makes this diagnosis system more intelligent. Self-learning of heuristic rules based on qualitative modelling is investigated and heuristic rules can add efficiency to model based diagnosis. During investigating self-learning of heuristic rules, the good learning property of neural networks is recognised and, neural networks based on-line fault diagnoses are also investigated. The research results reveal that neural networks based diagnosis systems are easy to develop and perform robustly provided that the training data are available.