一种基于特征的冲模CBD系统的学习方法

Feature-based learning method of CBD design system

尽管CBD曾被认为是很有前途的智能设计支撑技术,其近年来的发展并不令人满意,其中一个主要原因是因为缺乏对于学习功能的深入研究。事例的存储和索引在冲压模具CBD的学习功能中起着重要的作用。本文提出了一种基于特征的CBD系统的学习方法。该方法首先建立冲压特征和对应的模具设计之间的映射关系;然后对冲压特征进行模糊聚类,模糊聚类将所有相似的事例归入同一个簇中;最后利用粗糙集理论,去除冗余属性并抽取产生式规则。在本方法的基础上,开发了一个原型系统。经运行表明本方法是可行的,尤其适合于从模具设计中发掘知识。

Although case based design （CBD） had been considered an appealing supporting technology for intelligent design, its development was not satisfying. One of the main hinders is the lack of deep research on learning. The way of storing and indexing cases plays a key role on the learning function of a CBD system. A feature-based learning approach for CBD system was presented in this paper. The method firstly created mappings from stamping features and corresponding die designs. Secondly, fuzzy clustering was applied to put similar cases into one cluster. Finally, redundant attributes were removed and production rules were extracted from the clustered decision table by applying Rough Set Theory. Running of a prototype system based on this method shows that it is feasible and suitable for learning knowledge from die designs.