AN APPROACH TO THE COMPILATION OF OPERATIONAL KNOWLEDGE FROM CAUSAL-MODELS

The interest for model-based approaches to diagnosis has grown very rapidly over the last decade. However, the computational complexity of such approaches has stimulated the search for mechanisms able to enhance the performance of model-based diagnostic systems. Knowledge compilation (a process for automatically synthesizing "operational knowledge" from "deep knowledge") has been proposed as a solution to the problem of increasing computational efficiency by means of heuristic knowledge, without requiring an explicit phase of acquisition of experiential knowledge from human experts. An approach to the synthesis and use of operational knowledge in diagnostic problem solving is proposed. The approach presented significantly departs from previous approaches to knowledge compilation in the sense that the authors do not aim at compiling an autonomous heuristic problem solver from a deep one but they only aim at deriving a set of conditions whose evaluation can focus and speed-up diagnostic reasoning on causal models. In particular, it is argued that operational necessary conditions can be used to focus causal diagnostic reasoning by pruning significant portions of the search space to be considered. The process for synthesizing operational knowledge is performed by running a case-independent simulation on a causal model using constraint propagation techniques. This is a major difference with respect to the approaches based on the use of examples. Many of the problems arising in the other approaches to the synthesis of heuristics from deep knowledge are solved in our system, as it will be discussed in detail in the paper.