Consistency-Based Fault Isolation for Uncertain Systems with Applications to Quantitative Dynamic Models

This paper presents the Probabilistic General Diagnostic Engine (PGDE), a novel method of offline consistency-based fault isolation. Many existing proposals require qualitative logic models for consistency-based diagnosis due to their ability to speed the search for conflict sets through the use of an ATMS. However, for many applications, quantitative dynamic models are preferred or already available. The key strength of the PGDE is that it allows the use of any modelling language for which an appropriate calculation engine can be written. It also offers graceful degradation in the presence of uncertainty, commonly caused by noise or modelling errors. Finally, given perfect knowledge, it can be shown that the PGDE computes the same result as existing consistency-based diagnosis methods. To demonstrate the performance of the algorithm, we have used a quantitative dynamic model of the fluid power circuit of a singledegree of freedom hydraulic test bench and developed an appropriate calculation engine for computing consistency between measured values and predicted results. Various failures were generated on the physical test bench and the PGDE isolated the faults with approximately 85% accuracy.