Architecture, algorithms and benchmarks for three dimensional shape searching
Abstract
Estimates suggest that more than 75% of engineering design activity comprises reuse of previous design knowledge to address a new design problem. Three-dimensional shape searching is a problem of current interest in several domains. This dissertation proposes a three-tier architecture, various shape representations, a benchmark database and a new multi-step search approach for 3D shape searching in the mechanical engineering domain. Hierarchical skeletal graphs are proposed as a shape representation to search for similar 3D models from a database. The principal advantages of hierarchical skeletal graphs are: they preserves geometry and topology of the query model, they are considerably smaller than B-Rep graphs, and they are insensitive to minor perturbations in shape, but sensitive enough to capture the major features of a shape. We found that skeletal graphs provide over 58% better search performance than 3D shape distributions for prismatic parts. Additional work done in this dissertation includes developing the first benchmark database of engineering shapes. Precision recall curves were generated for twelve different shape representations across the benchmark database. It was found that different shape representations capture different aspects of shape and hence work differently for various part categories. It was found that search methods based on 2D views outperform other methods in literature. The later part of this work includes proposing and quantifying a user controlled multi-step search strategy for improving search performance. From our tests conducted for 801 models from the benchmark database, it was found that multi-step search increases the number of relevant models as compared to human perception by about 24%. Multi-step search also increases the ranking correlation as compared to human perception by about 14%. The multi-step methods that work best for a particular class of parts are also identified and it is hoped that this will yield better search systems in the future.
Degree
Ph.D.
Advisors
Ramani, Purdue University.
Subject Area
Mechanical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.