An integrated, deep-shallow expert system for multi-level diagnosis of dynamic systems.

An integrated expert system architecture and strategy for diagnosis of dynamic systems with feedback loops and synchronous or asynchronous state transitions is presented. The dynamic system under diagnosis is modelled using structural and behavioural representations in multiple levels of abstraction. The diagnosis process integrates shallow and deep expertise. It recursively navigates through the structural hierarchy, and at each level tries the shallow expertise first. If it fails it switches to deep, simulation based expertise. A multilevel simulator assists the diagnosis process in verification and elimination of hypothesized suspects. The simulator shifts on demand through several levels, from coarse hypothesized suspects. The simulator shifts on demand through several leves, from coarse qualitative modelling to detailed quantitative modelling. Knowledge of pathological behaviour (failure modes) of lower level components is incorporated in the simulator. Learning is exhibited as deep to shallow expertise transfer, as well as up the abstraction levels of the simulator itself, to improve future efficiency. In addition, knowledge about pathological behaviour can be used for off-line training by artificially generating new cases for diagnosis.