Component ontological representation of function for candidate discrimination in model-based diagnosis

This dissertation proposes representation of function based on component ontology. Further, it proposes the use of function for suspect ordering in model based diagnosis.; In component ontological representation, function of a component is expressed in terms of its ports. This representation has many advantages over the currently used state and process ontological representations: (i) Function of a component can be represented in isolation of its environment. Therefore, libraries of function models can be built. These models are re-usable. (ii) Function model of a complex device can be built by composing the function models of its components. This ensures the fidelity of representation. Further, this process can be automated, and takes time linear in the number of components in the device. (iii) The representation is linear in space complexity: Function model of an atomic component is linear in the number of its ports, and that of a complex device is linear in the number of its components. In comparison, state and process based representations can be typically exponential in space complexity, and hence, intractable.; In the dissertation, the principles behind component-ontological representation of function are enunciated. Classes are presented as a computational model for function, based on these principles. Algorithms are developed to automatically compose the class model of a complex device from those of its components.; In addition, function is proposed to be used for candidate discrimination in model based diagnosis. It is as effective as the traditionally used fault probabilities because both prompt for similar sequences of measurements to verify diagnosis. The advantages of using function instead of fault probabilities are: (i) It facilitates generation of explanations during diagnosis, based on causality. (ii) It is readily available from device design whereas fault probabilities are hard to obtain.; In the dissertation, two techniques are proposed for function based candidate discrimination. They are device-independent. An algorithm is developed, which uses these techniques to diagnose the class model of a device. Classes is a computational model for function based diagnosis. Its scalability and domain-independence are demonstrated by applying it to an electronic printer buffer and an automobile carburetor.