An approach for integrating multi-modal, model-based diagnostic components

We describe an approach for modeling and diagnosing complex multi-modal sys tems. A novel feature of this approach is the proposed method for integrati ng multiple specification modalities. We apply this approach to a real-worl d rocket engine turbopump. The example model consists of two submodels, one describing liquid flows, and one describing mechanical vibrations. Typical methods for modeling such systems will treat these two modalities as being completely disjoint, even though they have significant interactions. We de scribe a method for integrating these sub-models using variables that play different roles in the different sub-models. We focus on how the integrated causal network model allows us to diagnose a more complete and physically plausible set of faults than by using the independent sub-models.