Research On Design Oriented CAD Model Retrieval

In the field of mechanical engineering, different products often have the same orsimilar parts. The successful product designs with strong experience and weak theorydepend heavily on the designer’s relevant design experience. And the existing design casesthat have been proved to be correct in practice can be modified to meet the requirement fornew products. Consequently, reuse of existing design in new product design is a pragmaticand inevitable choice. Some surveys show that about40%of product designs can directlyuse the existing design resources, about40%of product designs can modify the previousdesign cases to get the new designs, and only20%of product designs need to start fromscratch. In order to locate the reference cases, the designers used their60%of the time tosearch for desired design resources. Even so, there are widespread repeat design andmanufacturing phenomenon in the manufacturing industry, which cause huge economiclosses for enterprise. The reason is the lack of effective design resources management andsearch tools.The popularity of three-dimensional (3D) CAD systems in the product design ofmanufacturing industry brings about the emergence of a large number of3D CAD models.Since CAD models are geometry expression the text-based management and searchtechnologies have serious limitations. First, the descriptive texts for CAD model are ofgreat uncertainty as the change of CAD systems, designers and design dates. Second, theaccuracy of the shape description based on text is poor. Thus, in the last decade,content-based CAD model search has received extensive attention in the academiccommunity.This dissertation mainly studies the CAD model retrieval methods for mechanicalproduct design, which includes the following aspects:(1) A CAD model retrieval method based on inexact graph matching is presented inorder to resolve the problem that the exact graph matching is unable to support the similarmodel retrieval. First, a representation of face attributed relational graph (ARG) for eachCAD model is extracted from its B-Rep model. Then, the vertex compatibility matrix andedge compatibility matrix between the ARGs of the target and searched model arecalculated, and the measure of the two model’s similarity is created from the compatibilitymatrices, which serves as the objective function for optimally selecting vertex mappingmatrix M between the two models. At last, the optimal vertex mapping matrix M is foundusing Lagrange multiplier method for the optimization problem after relaxing itsinequality constraints to be equal.(2) Based on the gradient flows in Lie group, a partial retrieval approach for CAD models is presented. First, a representation of the face ARG for a CAD model is createdfrom its B-rep model and thus partial retrieval is converted to a subgraph matchingproblem. Then, an optimization method is adopted to solve the matching problem, wherethe optimization variable is the vertex mapping and the objective function is themeasurement of compatibility between the mapped vertices and between the mappededges. Different from most previously proposed methods, a homogeneous transformationmatrix is introduced to represent the vertex mapping in subgraph matching, whosetranslational sub-matrix gives the vertex selection in the larger graph and whoseorthogonal sub-matrix presents the vertex permutation for the same-sized mapping fromthe selected vertices to the smaller graph’s vertices. Finally, a gradient flow method isdeveloped to search for the optimal matching matrix in Special Euclidean group SE(n).(3) A CAD model retrieval method based on B-rep decomposition is presented. First,according to the salient geometric features of the mechanical part, the surface boundary ofa solid model is divided into local convex, concave and planar regions with the minimalnumber. Then, we give a kind of region codes that describe the surface region and theirlinks in CAD model. At last, the similarity between two models is measured by thecomparison of their region codes. The retrieval for CAD model can be applied to theglobal and partial search.(4) A similarity evaluation for assembly model based on component attributedrelational graph (CARG) is presented. First, assembly models are created from CADsystems. The information of constitute parts, excluding for connection and transmissionparts, and the complex relationships between them in an assembly are extractedautomatically or manually; meanwhile, assembly model is described as CARG. Then, thepart compatibility matrix between two assemblies are calculated, which serve as themeasure of their similarity. Finally, the optimal matching under the measures is calculatedusing Kuhn-Munkres algorithm.Based on the methodologies mentioned above, a3D retrieval system is developedusing Visual C++6.0, geometry engine ACIS and display engine HOOPS. The systemsupports the inexact retrieval, the exact partial retrieval, globe and local retrieval for CADmodel. Meanwhile, some experimental results show their correctness and effectiveness.