B. Stein (Germany)
diagnosis, automated modeling, simulation, data mining
This paper is on the automated construction of diagnosis models for complex, continuous-valued systems. Given is the following diagnosis scenario of first principles: (a) The global behavior model of the system in question can be composed from single component models, where (b) context-free models of both the normal and the faulty component behavior are known. This situation enables one to simulate the system with respect to expected inputs along with faults, and to distill a compiled diagnosis model from the huge set of generated data with mining techniques. As well as that, this situation enables us also to define the most informative measurement points for fault isolation purposes. The contributions of this paper are threefold: (1) It outlines the compilation approach and its realization in the domain of hydraulic engineering, (2) it extends the GDE measurement heuristics towards an optimum strategy with respect to an arbitrary observation horizon, and (3) it presents a measure to quantify the diagnosis effort for sys tems whose behavior can be captured by a compiled model.
Important Links:
Go Back
