| Any knowledge-based system that is to be placed in the cockpit must deal with a wealth of time-varying data, and do so in real time. An avionics knowledge-based system thus requires a conventional signal processing front end to identify relevant signal characteristics, a knowledge representation scheme that lends itself naturally to the representation of uncertain numeric data, and an efficient, interruptible inferencing method.; A hardware architecture for knowledge-based avionics systems is proposed which consists of a signal processor and one or more symbol processors. The signal processor ascertains whether new input data is different enough from previous data to warrant interruption of the inference process; if so, it passes the new data, along with any computed attributes of the data, to the symbol processors. If a symbol processor requires new data, it queries the signal processor, which passes the data and attributes to the symbol processor.; The blackboard paradigm is proposed as the basis for the software architecture, as it allows asynchronous events to alter the course of inference.; Dempster-Shafer theory is used to express uncertain numerical data. It is shown that the Dempster-Shafer formalism and the four-valued logic of Belnap are complementary, allowing numeric and non-numeric expressions of uncertainty to co-exist.; It is also demonstrated that the Kalman filter, a common software component of avionics systems, can be used to provide uncertainty information about numeric data to a Dempster-Shafer evidential reasoning system.; An initial software implementation developed in the Ada programming language is discussed. |
