Browse Theses

The need for customization in expert systems is evident in domains where no single interpretation of the domain concepts satisfies all end users or reflects their individual expertise. In expert systems dealing with such domains, there should be room for the customization of portions the knowledge base in order to allow groups of users with different opinions to share the same application without conflicts. However, specifically in engineering, not all the tasks to be performed by the expert system will be subject to customization, since there is a large pool of commonly agreed knowledge. This coexistence of individual and common knowledge in an application should be identified and taken into account in the design of expert systems.

This thesis proposes an approach to enable customization of expert systems that consists in splitting the knowledge base of an expert system into two components: (1) a core knowledge base containing the commonly agreed domain knowledge; and (2) a custom knowledge base, containing the individual heuristics and definitions of subjective concepts in the domain of interest. The development of an expert system with the proposed approach starts with the definition of the contents of the core and custom knowledge bases by the knowledge engineer. The definition of the contents of each of the two knowledge bases is made, identifying the processes in which there is no general consensus of how they should be executed or interpreted. This identification of the contents of each knowledge base is made through consultation sessions with the different experts participating in the development of the expert system.

A proper knowledge acquisition facility must be provided to allow individual users to create and maintain their individual knowledge bases. The knowledge acquisition facility is the component responsible of building, modifying and reviewing the custom knowledge base. The design of the knowledge acquisition facility of the expert system depends on the representation scheme selected for custom knowledge. A representation scheme for custom knowledge and a facility for maintaining custom knowledge are presented in this thesis.

This thesis illustrates the proposed framework with EVAL, a knowledge-based system for evaluating the seismic resistance of existing buildings. The domain of seismic evaluation of existing buildings was a good candidate for the proposed framework since it incorporates individual judgment and heuristic knowledge while the same time relying on a body of common knowledge, such as structural analysis and structural mechanics.