基于Weisfeiler-Lehman图核算法的装配体模型比较方法

A Comparison Method of Assembly Models Based on Weisfeiler-Lehman Graph Kernel Algorithm

随着CAD信息技术的在机械领域的广泛应用,在机械产品设计过程中积累了大量包含多源信息的CAD模型,分析利用已有的产品模型可以大大提高新产品的开发效率。对能够比较全面表达产品信息的装配体模型进行比较分析可以很好的支撑通用结构挖掘、模型检索等三维信息重用,提高产品设计效率。提出了基于Weisfeiler-Lehman图核算法的装配体模型比较方法。Weisfeiler-Lehman算法可以有效的解决图匹配问题,核函数不用计算复杂的非线性变换,直接得到非线性变换的内积,应用于图的模型中可以大大简化计算复杂度,实现图之间的相似度计算。首先用图模型对多源装配体模型进行信息的转化表达并进行初步信息归类,形成装配体类码连接图;将W-L图匹配算法和核函数综合应用,对装配体类码图进行相似度计算,实现装配体之间的比较分析。

With the wide application of CAD information technology in the mechanical field,a large number of CAD models containing multi-source information have been accumulated in the process of mechanical product design.Analysis and utilization of existing product models can greatly improve the development efficiency of new products.The similarity analysis of the assembly models which are capable of comprehensive product information can well support common design structure mining,model retrieval and other design reuse,so as to improve product design efficiency.We propose a multi-source assembly CAD model similarityanalysis method based on Weisfeiler-Lehman graph kernel algorithm.Weisfeiler-Lehman algorithm is expert in graph matching problem,and kernel function greatly simplifies calculation of graph information because that directly obtains nonlinearity without calculating complex nonlinear transformation.Thus the combination of WeisfeilerLehman algorithm and kernel function is able to achieve the similarity calculation between graphs.Firstly,the graph model is used to transform and express the information of the multi-source assembly model.Then the nodes and edges are classified according their attributesto generate t class code graphs.Secondly,the W-L matching algorithm and the kernel function are comprehensively applied in graph similarity analysis.The similarity calculation performed on the class code graphes realize the similarity analysis between the assembly models.