Constructing a Design Knowledge Base Using Natural Language Processing

Abstract

Mechanical engineering, like other engineering disciplines, has witnessed maturation of various aspects of its domain, obsolescence of some areas and a resurgence of others. With a history of over 200 years of continuous research and development, both in academia and industry, the community has generated enormous amounts of design knowledge in the form of texts, articles and design drawings. With the advent of electronics and computer science, several of the classical mechanisms faced obsolescence, but with the emergence of MEMS and nanotechnology, the same designs are facing a resurrection. Research and development in mechanical engineering would derive enormous benefit from a structured knowledge-base of designs and mechanisms. This paper describes a prototype system that synthesizes a knowledge-base of mechanical designs by the processing of the text in engineering descriptions. The goal is to construct a system that stores and catalogs engineering designs, their sub-assemblies and their super-assemblies for the purposes of archiving, retrieval for launching new designs and for education of engineering design. Engineering texts have a relatively clear discourse structure with fewer ambiguities, less stylistic variations and less use of complex figures of speech. The text is first passed through a part-of-speech tagger. The concept of thematic roles is used to link different parts of the sentence. The discourse structure is then taken into account by anaphora resolution. The knowledge is gradually built up through progressive scanning and analysis of text. References, interconnections and attributes are added or deleted based upon the nature, reliability and strength of the new information. Examples of analysis and resulting knowledge structures are presented.