Peter Struss, Martin Sachenbacher, Florian Dummert
Diagnosing a Dynamic System with (almost) no Observations.
A case study in off-board diagnosis of the hydraulic circuit of an anti-lock braking system
Workshop Notes of the 11th International Workshop on Qualitative Reasoning QR 97, Cortona, Italy, pages 193-201, 1997.
We present requirements and results of a case study in model-based (off-board-)diagnosis of the hydraulic circuit of an anti-lock braking system. The primary problems to be addressed quite fundamental: it is impossible to predict the behavior of the system and, additionally, there are no measurements of the actual behavior available. Both might be considered fatal for model-based diagnosis. We tackle these problems by applying models that capture qualitative deviations of variables and parameters from nominal behavior. They allow to exploit vaguely described symptoms such as "brake pedal too soft". The models are used in a state-based diagnosis framework, i.e. only the observed states are checked for consistency with the model, and no simulation of the dynamic behavior is required. The crucial step in making the approach work is to exploit basic constraints on continuity and for complementing the directly obtained observations by information about derivatives. Experimental results are compared to expert knowledge represented in existing failure mode and effects analysis (FMEA) documents and prove to be adequate.