Diagnosing continuous dynamic systems using qualitative simulation

Concerns model-based diagnosis of dynamic systems (MBDDS) by fuzzy qualitative simulation (FUSIM). Within a primitive architecture, the system model explicitly describes the structure of the plant to be diagnosed and an algorithm for generating the dynamic behaviour from the plant´s structure. The discrepancy generator compares observations from the plant with predictions from the system model and generates appropriate discrepancies, whilst the candidate generator uses the discrepancies to produce candidates, i.e., the possible faults, to be validated by the system model. The diagnostic supervisor contains the meta-knowledge necessary to control the diagnostic process. This paper discusses how this diagnostic method determines the fault conditions of a physical system. The FUSIM algorithm produces a sequence of system states with related temporal durations, providing a significant advantage over other qualitative simulation techniques for refining diagnosis over time. An implementation of the primitive architecture is described. The mechanism for the comparison between the predictions and the observations is presented; it is the key technique for system-monitoring and fault evaluation, and the basis for discrepancy generation within MBDDS