Similarity Metrics Applied to Graph Based Design Model Authoring

Abstract

Model reuse is typically facilitated by search and retrieval tools, matching the sought model with models in the case base or database. This paper looks at providing assistance to users authoring design exemplars, a new data structure to represent parametric and geometric design problems, by providing alternative configurations. Four distinct similarity metrics are proposed for model retrieval in an interactive modeling environment: entity similarity, relationship similarity, attribute similarity, and structural similarity. A brief review of graph matching literature has been provided, followed by a technical session showing the application of a set of derived similarity metrics.Keywords: design exemplar, CAD query language, retrieval, interrogation INTRODUCTIONIn engineering design, as with many other disciplines, knowledge or model reuse is sought to reduce workload and development effort. Model reuse is typically facilitated by search and retrieval tools, often matching sought models based upon "similarity metrics". For example, a large tire manufacturing firm in North America designs mold inserts for its tire treads. The tooling cost for each mold insert is approximately 2000$. Hence, in order to reduce the manufacturing cost, the designers retrieve inserts that are similar to the desired shape geometry in that, which fall within a specified tolerance envelope. Different strategies have been employed to retrieve similar models [13,19]. For example, shape-based search methods are used to retrieve 3-D models and sketches in which spherical harmonics are used in creating discriminating similarity measures [13]. An Engineering Advisory System (EAS) has been developed, which involves the retrieval of knowledge associated with 3-D models [19]. The system takes a 3-D shape as a query and converts it into two simpler "fingerprint representations": feature vectors and skeletal graph. Feature vectors represent the global shape or geometry of the object, whereas the skeletal graphs minimally represent the topology and the local geometry of the object. Examples of feature vectors include the Geometric Moment Invariants and Geometric Ratios. Search for similar models can be performed through a combination of these two representations. This paper examines similarity in engineering model authoring contexts, offers a set of metrics that might be employed in graph based model authoring, and demonstrates these metrics by an illustrative case study. Specifically, this approach is applied to the authoring of the design exemplar, a standard representation of mechanical engineering design problem knowledge [39]. These metrics are used in a system that offers the model developer, in this case the exemplar developer, options of existing models that are matched to what has been partially authored. As mentioned above, the design exemplar is a new data structure that has been developed to represent parametric and geometric design problems [7]. Most queries associated with mechanical engineering design include parametric and geometric v...

Design for X-Abilities of a Mechatronic System-A Concurrent Engineering and Graph Theory Based Approach

“…Also, there are some contributions on graph base design. For example, analyzing the fundamental challenges in applying graph transformations in design context [20], graph-based design and analysis of dynamic software architecture [21], development of interactive similarity metrics in graph-based design model [22], etc. A combined graph theory and DFX-based approach has not been applied in concurrent design of a mechatronic system.…”