面向组件信息回路的机电产品模块划分方法  

Modular Partition Method of Mechatronic Products Based on Component Information Loop

作　　者：李中凯[1] 邹光宇 Li Zhongkai;Zou Guangyu(School of Mechatronic Engineering,China University of Mining and Technology,Xuzhou 221116)

机构地区：[1]中国矿业大学机电工程学院,徐州221116

出　　处：《计算机辅助设计与图形学学报》2022年第8期1186-1192,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(51475459);江苏省优势学科建设工程资助项目。

摘　　要：现有的模块化设计方法多注重对模块化指数的优化迭代,为提升模块划分时组件间信息交互性能,提出一种基于集成学习算法的产品模块划分方法.首先构建产品的连接和功能设计结构矩阵;然后依据组件间信息的传递和反馈构建弱分类器;最后组合大量弱分类器构建强分类器模型,并由强分类器给出产品的模块划分结果.对采矿液压支架ZY3200型的模块划分结果表明,与直接聚类和遗传算法模块结果相比,集成学习模块规划策略的模块化指数I1值为0.8438,具有更高的对角线聚集度;敏感性分析显示,当回路组件数大于3时Q值稳定在0.5562,进一步验证了所提方法具备机器学习的智能性和鲁棒性特征.Most of the existing modular design methods focus on the optimization iteration of modular index.In order to improve the information interaction between components,a product modular partition method based on ensemble learning algorithm is proposed.Firstly,the product connection and function design structure matrix are constructed.Secondly,the information transfer and information loop between components are taken as the construction criteria of weak classifiers.In the last,the strong classifier model is constructed by combining a large number of weak classifier models,and the product module partition results are given by the strong classifier.A ZY3200 hydraulic support case is used to demonstrate the proposed module partition method.Compared with direct clustering and genetic algorithm,the modularity index I1 of ensemble learning algorithm is 0.8438,which is closer to the design structure matrix diagonal.Sensitivity analysis shows that the Q value is stable at 0.5562when the number of loop components is greater than 3,which further verifies that the proposed method has the intelligence and robustness of machine learning.

关 键 词：设计结构矩阵 模块划分 集成学习算法 信息传递 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]