Fault Diagnosis in a Large Dynamic System: Experiments on a Training Simulator

Fault diagnosis performance of expert trainees was investigated using a low-fidelity marine powerplant simulator. Based on experimental data, the diagnostic process was divided into two stages: 1) hypothesis formation and 2) hypothesis evaluation. Hypotheses were formed based on feasible set of symptom-case pairs. Performance was related to the initial feasible set (IFS) and the transition strategy for stage-shifting. Subjects used either a breadth-depth strategy where a hypothesis was thoroughly checked after a broad search for hypotheses, or a balanced strategy where several hypotheses were formed quickly and maintained at the same time. Good IFS and breadth-depth strategy resulted in better performance that bad IFS and balanced strategy.