ANALYTICAL AND KNOWLEDGE-BASED REDUNDANCY FOR FAULT-DIAGNOSIS IN PROCESS PLANTS

The increasing complexity of process plants and their reliability have necessitated the development of more powerful methods for detecting and diagnosing process abnormalities. Among the underlying strategies, analytical redundancy and knowledge-based system techniques offer viable solutions. In this work we consider the adaptive inclusion of analytical redundancy models (state and parameter estimation modules) in the diagnostic reasoning loop of a knowledge-based system. This helps overcome the difficulties associated with each category. The design method is a new layered knowledge base that houses compiled/qualitative knowledge in the high levels and process-general estimation knowledge in the low levels of a hierarchical knowledge structure. The compiled knowledge is used to narrow the diagnostic search space and provide an effective way of employing estimation modules. The estimation-based methods that resort to fundamental analysis provide the rationale for a qualitatively-guided reasoning process. The overall structure of the fault detection and isolation system based on the combined strategy is discussed focusing on the model-based redundancy methods which create the low levels of the hierarchical knowledge base. The system has been implemented using the condensate-feedwater subsystem of a coal-fired power plant. Due to the highly nonlinear and mixed-mode nature of the power plant dynamics, the modified extended Kalman filter is used in designing local detection filters.