NASA-funded work that ran the method on ADAPT, the electrical power system testbed at NASA Ames. A model of more than 400 nodes, compiled once, returned diagnoses in under a millisecond on average.
ADAPT is a working electrical power system at NASA Ames, built to be representative of the power systems in aircraft and spacecraft, with generation, storage and distribution. The paper set out to diagnose it with probabilistic reasoning and met two problems that come up in every real diagnostic application, how to build the model and how to reason within a hard real-time budget.
For the first, the team wrote a specification language that generates the Bayesian network of a power system from a description of its components. For the second, the network was compiled once into an arithmetic circuit, a small and predictable structure that an avionics computer can evaluate. The circuit answered diagnostic queries on real ADAPT data in under a millisecond on average, on a model of more than 400 nodes covering more than 100 components.
This is the problem Manifest addresses, two decades earlier and in its first form. The specification language is the ancestor of automated model production. The compiled circuit is the reason a diagnosis can run onboard with a known time and memory budget. The work was funded by NASA and published at an AAAI conference.
ADAPT is a ground testbed, not a flight system. This was a research prototype in 2008, not PiLogic's product. The linked copy is the AAAI proceedings version of record; NASA also hosts a preprint, which lists Ann Patterson-Hine as an author. The proceedings version thanks her for her central role in developing ADAPT.