基于最大均值差异测度的装配体相似性研究

Assembly similarity stady based on the maximum mean discrepancy measure

基于距离的装配体相似性度量方法由于忽略了对距离分布的分析,在其案例初筛过程中,采用该方法时易导致丢失部分相似的案例。针对这一问题,提出了一种基于最大均值差异(MMD)的装配体相似性度量方法。首先,利用装配体中零部件数量及零部件类型数量、连接数量及连接类型数量,共4个参数,将装配体模型化为一维数据集合;然后,使用最大均值差异(MMD)算法,将表示装配体模型的一维数组映射到再生核希尔伯特空间(RKHS),在该空间内计算出装配体间的距离,并利用离散系数对距离进行了统计学分析;最后,通过基于实例的实验和基于装配体参数生成规则的仿真比较实验对其进行了验证。实验及研究结果表明:在准确度上,MMD算法与欧氏距离(ED)和加权距离(WD)算法一致;在鲁棒性上,无论进行相似性分析的两装配体零部件数量是否一致,该方法的距离分布在零部件数量超过6个后即可达到基本稳定,最高离散系数约为WD算法的23%,距离分布的鲁棒性有了较大程度的增强。

To address the problem that the distance-based assembly similarity metric ignores the analysis of the distribution of distance,which can easily lead to the loss of some similar cases during the initial screening of cases,an assembly similarity metric based on the maximum mean discrepancy(MMD)was proposed.Firstly,the assembly was modeled as a one-dimensional data set using four parameters,such as the number of parts and the number of part types,the number of couplings and the number of coupling types in the assembly.Then,the one-dimensional array representing the assembly model was mapped to the reproducing kernel Hilbert space(RKHS)using the maximum mean discrepancy(MMD)algorithm,in which the distances between assemblies were calculated and the distance was statistically analyzed using discrete coefficients.Finally,it was verified by case-based experiments and simulation comparison experiments based on assembly parameter generation rules.The experimental and research results show that MMD algorithm is consistent with Euclidean distance(ED)and weighted distance(WD)algorithms in accuracy;from the perspective of robustness,regardless of the number of parts of the two assemblies for similarity analysis,the distance distribution of the method is basically stable after the number of parts exceeding six,and the highest dispersion coefficient is about 23%of the WD algorithm,which improves the robustness of the distance distribution is greatly enhanced.