Hybrid reasoning methods for intelligent operation support systems

Process operations are time critical situations. Faced with vast amount of process data, human operators may not be able to contribute a timely and effective solution. The industry requires new technologies to reduce the cognitive load placed on operators in nonroutine operations. This thesis aims at developing artificial intelligence solutions to the operation support problems.; A multi-dimensional problem solving model is developed following the analysis of the recognition behaviour of human operators and the characteristics of the process operations. A multilayer modularity architecture is proposed to achieve the decomposition and coordination of system functions, production processes and the reasoning methods.; The thesis focuses on the integration of various knowledge representations and reasoning methods. A hybrid reasoning method, which employs the case-based reasoning (CBR) as the principal reasoning paradigm and the other methods as the supplemental paradigms, is developed. A dynamic case-based reasoning (DCBR) method is developed to extend the CBR to dynamic problems. The DCBR utilizes new indexing mechanisms to incorporate system dynamics, and to improve problem solving coverage and flexibility. It allows easy integration of various reasoning paradigms into one environment. The model-based reasoning method is integrated with the DCBR for case adaptations and novel problem solving. A principal component analysis (PCA) method and a Kalman filter-based algorithm are developed and integrated with the DCBR for problem feature interpretation and abnormal condition identification. By compensating the limitations of the individual reasoning methods, the best system performance can be achieved. To enhance the intelligent operation support system (IOSS) with the capability of handling multimedia information, an intelligent hypermedia on-line manual is developed as the external knowledge base and multimedia operator interface of the IOSS.; An actuator and sensor design algorithm is developed for operation support systems based on fault distances and objective trees that represent the instrumentation requirement of the IOSS. This algorithm selects an optimal set of actuators and sensors that ensures good system performance and the least hardware cost.; The methods developed in this thesis improve the problem coverage, problem solving efficiency, knowledge representation and operator acceptability of the IOSS. A prototype IOSS has been developed and implemented on a bleached chemi-thermo-mechanical pulp plant with satisfactory results.