Date of Award

8-2008

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Legacy Department

Mechanical Engineering

Advisor

Summers, Joshua D

Committee Member

Fadel , Georges

Committee Member

Ziegert , John

Committee Member

Malloy , Brian

Committee Member

Kurz , Mary Beth

Abstract

Model reuse is typically facilitated by search and retrieval tools, matching the sought model with models in a database. This research aims at providing similar assistance to users authoring design exemplars, a data structure to represent parametric and geometric design problems. The design exemplar represents design problems in the form of a bi-partite graph consisting of entities and relations. Authoring design exemplars for relatively complex design problems can be time consuming and error prone. This forms the motivation of developing a search and retrieval tool, capable of retrieving exemplars that are similar to the exemplar that a user is trying to author, from a database of previously authored exemplars.
In order to develop such a tool, similarity measures have been developed to evaluate the similarity between the exemplar that a user is trying to author and target exemplars in the database. Two exemplars can be considered similar based on the number and types of entities and relations shared by them. However, exemplars meant for the same purpose can be authored using different entities and relations. Hence, the two main challenges in developing a search and retrieval tool are to evaluate the similarity between exemplars based on structure and semantics.
In this research, four distinct similarity metrics are developed to evaluate the structural similarity between exemplars for exemplar retrieval: entity similarity, relation similarity, attribute similarity, and graph matching similarity. As well, a thorough understanding of semantics in engineering design has been developed. Different types of semantic information found in engineering design have been identified and classified. Design intent and rationale have been proposed as the two main types of semantic information necessary to evaluate the semantic similarity between exemplars. The semantic and structural similarity measures have been implemented as separate modules in an interactive modeling environment. Several experiments have been conducted in order to evaluate the accuracy and effectiveness of the proposed similarity measures. It is found that for most queries, the semantic retrieval module retrieves exemplars that are not retrieved by structural retrieval module and vice versa.

Share

COinS