Abstract To improve the performance of case based intelligent stamping die design systems, an approach based on Rough Set theory is proposed to mine rules from successful design cases. This approach is based on the mapping relation between stamping features and die designs. The mappings from the attributes of a feature to the related die design constitute the design knowledge. The method extracts the rules from the mapping relationship by applying fuzzy classification and computing attribute reduct. The mined rules can be integrated into traditional RBR systems to provide assistance for the design of a new part. Furthermore, these rules can also be used to speed up the case retrieval process of CBR systems by restricting the search space into a subgroup of cases. Key words: Stamping Die design, Rough Set, Fuzzy Classification, Knowledge Discovery 35824
1. Introduction Metal stampings are used widely in almost any mass-produced product because the metal stamping manufacturing processes are economic and quick means of producing qualified articles in huge quantities[1]. However, its also well known that stamping die design is still a difficult, time-consuming and error-prone process. Recently, with the development in the field of AI (Artificial Intelligence), many efforts have been made to integrate various AI techniques, such as pattern recognition, RBR (Rule Based Reasoning), CBR (Case Based Reasoning) and NN (Neural Networks), into conventional computer aided die design systems. Among many prototype systems constructed to improve the automation and intelligence of design process, CBR and RBR are the two most popular AI approaches. However, both of them have their significant advantages and disadvantages. RBR systems can be easily integrated with CAD systems and give instant advices for designers because the knowledge for die design is expressed by rules[2]. But its difficult to acquire, represent and update design knowledge because stamping die design is more of an art than a science in some sense. CBR systems support designers by recalling similar existing designs and thus get around the tough work of dealing with rules[3]. But the accuracy and speed of case retrieval is still a bottleneck for CBR systems. Although many algorithms are proposed to improve the efficiency of case retrieval, NN(Nearest Neighbor) search is still a main approach because of not having a good indexing method except for in some special fields[4]. A method based on RST (Rough Set Theory) is introduced in this paper to automatically mine rules from design cases and some stamping parts including mainly bending process are used for illustrating the procedure of mining rules. 2. Feature-based case representation In a typical die development process, designers first receive the specifications and drawings of a stamping product, and then establish the stamping process. Finally, the die assembly and die parts will be designed according to the requirement of product. The knowledge of die design will be kept in the drawings and is hard to be found. A stamping die design case can be decomposed into two sections: a stamping part and a corresponding die design. All of the cases congregate and form a case base. From the viewpoint of feature-based design, a stamping part can be pided into features and each feature is represented by a set of attributes. Features and their attributes can be obtained by mainly three approaches: feature recognition, design by features and interactive feature definition[5]. Features with different attributes imply different requirements of stamping parts and can be mapped to special die designs. Usually, a stamping product can be decomposed into features like flat, drawing, hole, bead, flange, slot, bend, etc[6]. This paper mainly deals with the bend features and the other type of features can be treated in the same way. Some factors of a bend feature, such as material, bend shape, bend radius, bend angle and dimension accuracy, are important for a die designer to make his decision of die design. In addition to these major factors, some factors indicating spring back is considered because spring back is a tough problem for bend die design and its hoped that some experiences can be obtained from existing cases and reduce the trial-and-error times. A bend feature is denoted as follows: 冲压模具设计挖掘规则英文文献和中文翻译:http://www.751com.cn/fanyi/lunwen_33960.html