Diagnosing Strong-fault Models with a Two-step A* Search Method

Abstract

Many model-based diagnosis researches, such as conflict directed A* search, have been made on weak-fault models, which have no fault mode behavior. However, in the real world, behaviors of common fault modes are usually known. In this situation, strong-fault models are built. Compared with weak-fault models, it is difficult to diagnose strong-fault models because their mode space is greater and non-monotonic. To diagnose strong-fault models efficiently, this paper proposes a two-step A* search, based on the conflict directed A* search. In our method, the consistency over modes, observations and models, is tested by assumption-based truth maintenance system, which generates multiple conflict sets if a fault occurs. Then fault isolation and identification are accomplished separately: firstly, possibly faulty components are isolated based on the conflict sets from the truth maintenance system; then different mode combinations of the faulty components are tested to obtain the specific fault modes. A* search is employed in both steps, as indicated by the name. By separating isolation and identification in two stages, memory and time requirements are reduced significantly. In the case study, a heat control system is utilized to demonstrate the proposed approach.