摘要
分析了ART1神经网络用于制造单元设计的天生缺陷,提出了两条改进途径:以模糊C均值算法对机床-零件矩阵分类问题进行预处理,以提高分类精度;通过修改模式向量的计算方法来克服保存在网络中的模式向量比较稀疏的情况。改进的ART1算法克服了标准ART1算法的不足,成为一种实用有效的制造单元设计方法。设计了新的算法流程并基于相似系数比较尺度在MATLAB软件平台上进行了算法仿真,与前人的研究结果相比,新算法产生了较好的分组效率。
The drawbacks that keep the standard ART1 paradigm from being a truly effective technique for optimizing the machine part matrix were analyzed, and two changes to the standard ART1 paradigm were proposed. The first change involved pre-processing by fuzzy C-MEANS so as to promote classification precision; the second change was to modify the vector memory pattern to a void too sparse representation vectors. The modified solution above-mentioned overcomes the shortcomings and makes the standard ART1 paradigm become a new effective method which can be used in real manufacturing cells design. A new algorithm chart was described. Simulation based on the standard of similarity coefficient was done in the platform of MATLAB and asserted its better results compared with the former researches. Finally, an engineering application was given in this way.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2006年第10期1040-1043,共4页
China Mechanical Engineering
基金
国防基础科研项目(K1800020502)
关键词
制造单元
ART1神经网络
零件族
相似系数
分组效率
manufacturing cell
ART1 neural network
part family
similarity coefficient
grouping efficiency
