MODELING FAULT DIAGNOSIS PERFORMANCE ON A MARINE POWERPLANT SIMULATOR (HUMAN PROBLEM SOLVING, KNOWLEDGE REPRESENTATION)

Human operators of large dynamic systems must possess good fault diagnosis skills. Factors affecting fault diagnosis skills are investigated in this thesis, and a model for fault diagnosis is developed.;A model for the fault diagnosis process which employs a conceptual entity called hypothesis frame to account for the observations from data and protocols has been proposed. Relevant information about a failure is "compiled" in the frames so that subjects could make quick and reliable inferences. Given symptoms of a failure, subjects would first try to match a frame. Once a frame is identified, subjects could use the information contained in the slots to make inferences. When no obvious frames could be processed, subjects could search in the system or symptom knowledge base for relevant information under the guidance of heuristics. This model could form the basis for implementing an on-line coach system for fault diagnosis.;A low fidelity simulator of a marine powerplant, was developed and used to study expert marine engineers' fault diagnosis performance. Based on the data collected, factors affecting the fault diagnosis performance were identified. They are: the initial feasible set (IFS) and the transition strategy. A subject's initial set of actions upon observing the symptoms forms the IFS which reflects the subjects' knowledge about a given failure. Transition strategy is concerned with the schemes subjects used to shift from the hypothesis formation stage to hypothesis evaluation stage. Two types of strategy were identified: breadth-depth strategy and balanced strategy. The data seem to indicate that subjects who had good IFS and used breadth-depth strategy performed better than subjects who had bad IFS and used balanced strategy. This finding seems to imply that both context-specific (IFS) and context-free (transition strategy) factors are important for fault diagnosis training.