An automatic case based reasoning system using similarity measures between 3D shapes to assist in the design of metal castings

In this paper, we present current research using the ShapeCBR system that automates the process of creation and selection of cases to populate a CBR system for retrieval of 3D shapes to assist with the design of metal castings. The special feature of this system is that similarity is derived primarily from graph matching algorithms. The particular problem of such a system is that it does not operate on simple search indices that may be derived from single cases and then used for visualisation and principal component analyses. Rather, the system is built on a similarity metric defined directly over pairs of cases and is primarily structural. An overview of previous research in this area is presented. This demonstrates the feasibility of a CBR approach to the design of metal castings. This paper describes further research into the use of the traditional componentisation as used in method engineering to provide a shape representation suitable for efficient retrieval of design knowledge. The paper presents current work aiming mainly at enhancing the efficiency and accuracy of the similarity metrics used in the ShapeCBR system. The architecture of the ShapeCBR system is presented. Finally, performance measures for the CBR system and the metrics used are given, and the results of trials of the system are presented and compared to results obtained from previous research.