Distributed and uncertain reasoning in a knowledge-based system for preventive diagnosis of power transformers

This paper presents a knowledge-based system called ASTRA, dedicated to the task of preventive diagnosis of power transformers. ASTRA features a novel problem-solving framework, that merges distributed reasoning and uncertainty management. This framework has been designed to meet the specific requirements of the preventive diagnosis task, which involves both the use of a variety of different knowledge sources and the presence of uncertainty affecting both input data and domain knowledge. The framework includes a proposal of a general architecture for distributed reasoning, called specialist net, and an approach to uncertain reasoning for the different types of specialists that may be present in a specialist net, namely procedural specialists, rule-based specialists, causal-evidential specialists. The application of the framework to the case of ASTRA is discussed and a simplified example of ASTRA reasoning is given.