英文翻译1

and die design for manufacturing automotive panels

Abstract

This work improves process planning and die design in automotive panel manufacturing using a novel case-based reasoning (CBR) methodology. An innovative indexing representation and retrieval approach are also addressed. The flat-bend graph, which is utilized to represent a panel model with a B-rep structure, retains geometric and topological data in the Standard for the Exchange of Product model data format. Flat-type faces collected into several groups are represented by graph nodes, and bend-type faces are represented by graph arcs. Based on the topological information between bend-type faces and flat-type faces, a graph is constructed.

Additionally, the holes detected are considered another graph node types. Geometric information and stamping parameters are utilized as graph attributes. To retrieve an appropriate case for a potentially huge search space, independent maximal cliques detection is applied. All independent maximal cliques that represent the maximum number of features shared by models are identified. Based on the retrieval result, previous process plans and die sets can be acquired for use by new cases. Experimental results obtained using the CBR system integrated with the product data management system demonstrate the practicality of reusing previous designs to accelerate stamping process planning and die design.

Keywords ：Automotive panel stamping .Case-based reasoning . Graph-based representation . Similarity assessment . Process planning and die design

1 Introduction

Sheet-metal parts are widely used to replace expensive cast and forged products in mass production in the automotive industries. Sheet-metal stamping to produce parts with freeform geometries (e.g., hood panels and fender panels) has the advantages of little material waste,

reduced costs, high productivity, and high-quality dimension. Process planning and die design are crucial tasks in the sheet- metal manufacturing cycle. At the start of a manufacturing process, appropriate stamping operations are set and organized in an appropriate sequence based on the geometry and features of an automotive panel. Press dies are designed based on panel shape and stamping operations to achieve high-quality complex freeform shapes. However, stamping planning and die design processes are difficult, error-prone, time-consuming, and demand considerable skill and experience from designers [1]. Furthermore, due to the complexity of

freeform shapes, process planning and die design for automotive panels are more complicated than those for common stamped parts.

Experience and knowledge can be applied to a design to improve production of panel stamping efficiency. Ullman [2] concluded that more than 75% of design activities reuse an existing design. When developing a new part with a certain functionality, the new design is typically a modified version of an existing design with similar functionality. Hence,reusing

knowledge embedded in existing components that are similar to the new component can improve production efficiency. In these cases, adapting the process plan and die sets from a similar existing panel to a new panel reduces design effort and increases production efficiency.

Case-based reasoning (CBR) is a methodology for solving problems by using or adapting solutions to old problems [3]. Conceptually, CBR is commonly described by four activities: retrieve, reuse, revise, and retain (the four REs) [4]. First, one retrieves similar cases from a case library for a current problem description, and then reuses a solution suggested by this similar case. If necessary, the solution is revises or adapted to meet new problem requirements. Finally, the new solution with its knowledge is retained in a CBR system once it has been confirmed or validated. The CBR methodology has been applied in the manufacturing domain for the last 10 years. Utilizing the CBR methodology to plan a stamping process and design stamping dies for

automotive panels reuse existed designs to develop a new design. The retrieval of an appropriate case is regarded as a key issue in CBR and has a significant impact on system efficiency. In

automotive panel stamping, parts with similar manufacturing characteristics usually have similar process plans and stamping dies. Therefore, similarity assessment and retrieval of a similar model

1

in the CBR are the major tasks for improving the efficiency of an entire process in automobile panel manufacturing.

Based on the CBR methodology, automotive panel stamping can benefit from retrieving a similar model with documents relevant to the target design. Several methods for analyzing parts in terms of their shapes along with shape-based 3D model retrieval techniques have been widely explored [5, 6]. Unfortunately, most shape-based matching techniques for engineering reuse focus on the 3D solid model rather than the surface model with freeform surfaces. The surface model is widely used in sheet-metal stamping applications because of the thin and flat character-istics of panel models. Similarly, model retrieval using existing methods for retrieving the appropriate surface model in metal stamping applications is unsuitable. Conse-quently, similarity assessment and retrieval architecture for surface models with freeform geometries are desirable in automotive panel production.

This work presents a novel CBR methodology for automotive panel stamping process planning and press die development. This methodology addresses the representa-tion, indexing, and retrieval of cases. The surface model is analyzed to extract geometric and topological

information,which is then converted into graph form as a case representation for indexing. A novel case retrieval archi-tecture based on local feature correspondence is utilized to search for similar models for reuse. The term feature in this work refers to the local geometric and topological characteristics of a sheet-metal part.

The research goals are as follows：

1. Utilize a novel graph-based representation and indexing method with the CBR methodology for automotive panel models.

2. Apply a novel comparison algorithm to assess the similarity between models and embed the retrieved architecture in the CBR methodology for retrieving similar cases.

3. Integrate the surface model retrieval system with CBR methodology to search for similar automotive panels in the product data management (PDM) system.

In the proposed CBR framework, panel models are analyzed to identify three fundamental entities: flat-type faces, bend-type faces, and holes. The topological and geometric data can be converted into features that form a high-level structural representation. In this work, geometric and topological data are combined in an attributed graph for measuring case similarity. The

graph-based indexing scheme is helpful for transforming the case similarity-assessment problem into a graph-matching problem. Graph-matching techniques are powerful tools used in the pattern-recognition field. The proposed approach retrieves a suitable automotive panel with a similar process plan, operational sequence, and press dies. In this work, a CBR system for stamping panel process planning and die design is implemented and integrated into the PDM system [7]. This work demonstrates the advantages of the proposed method over previous techniques for retrieving similar models automatically for reusing sheet-metal components.

The rest of this paper is organized as follows. Section 2 presents an overview of techniques for CBR and retrieving similar models. Section 3 then describes the framework of the proposed CBR system for process planning and die design. Next, Sections 4 and 5 introduce the flat-bend graph as a case representation for stamping parts and describe the novel comparison method for assessing the similarity between two cases. In Section 6, a car fender is used as an example case to demonstrate the implementation process. Conclusions are finally drawn in Section 7, along with recommendations for future research.

2 Related work

2.1 The case-based reasoning methodology

The foundation of CBR lies in the psychological theory of human cognition. Notably, CBR has been successfully applied in the application domains of design and manufacturing. Sycara et al. [8] proposed Cadet, a case-based synthesis tool, for mechanical designs. Sun and Chen [9] used CBR methodology to construct a fixture design system that becomes more intelligent than the traditional computer-aided design system. The intelligent system can learn from every solved case and become experienced.Tiwari et al. [10] utilized a process planning system for machining

prismatic components using CBR methodology.Their CBR system has three modules, i.e., the part input and representation module, the case retrieval module, and the case adaption module. Process

2

planning with CBR is also used in other such areas as the welding process [11],injection molding process [12], and cold forging process[13]. Lee and Luo [14] studied die-casting die design using CBR. They developed a CBR environment for die-casting die design and proposed an effective case representation and adaptation methodology for efficiently reusing previous design resources. Tor et al. [15] presented a CBR approach with an indexing and retrieval strategy for metal stamping die design. By using a hierarchical classification structure, a stamped metal part with four stamping feature categories is represented as a case in the library. Zhang et al. [16]extended the work of Tor et al. by extracting a set of deep drawing features for multi-stage,

non-axisymmetrical sheetmetal deep drawing. Liu et al. [17] presented a representational model for objected-oriented case information relationship. They proposed the Class-Property structure to refine case representations and utilized three hierarchical indexing strategies for case extraction. However, only few studies have focused on process-planning systems for stamping automotive panels using CBR. Leake et al. [18]presented a set of principles and techniques for integrated case-based design support systems. The Stamping Advisor,a system that supports feasibility analysis for sheet-metal automotive parts, was utilized to demonstrate applications of case-based design. Chen et al. [19] recently presented an object-oriented hierarchical case representation scheme for automotive panels in process planning.

2.2 Model similarity assessment and similar model retrieval

Various techniques for assessing the similarity among 3D solid models have been developed [20]. Group Technology(GT) is an effective scheme for model coding and classification using manufacturing information. Iyer and Nagi [21] proposed a GT code-based technique for retrieving and ranking similar solid parts from a database based on answers to a series of questions based on their coding rules. Hermann and Singh [22] developed design similarity measures for process planning; these measures are based on design attributes in GT code. However, the representation of GT code has some limitations, especially when a part is extremely complex.

To describe a shape in detail, statistical techniques are based on sampling points on the surface of 3D models [23–25]. Statistical techniques effectively distinguish between models in broad categories, but perform poorly when discriminating between models with similar gross shapes but different local features. Based on both geometric and topological properties,

graph-based techniques convert models into such graphs as adjacency graphs, Reeb graphs,and skeleton graphs [26–31]. El-Mehalawi and Miller [32]converted mechanical artifacts into attributed graphs for model comparison. Similarity was obtained by approximate graph

comparisons [33]. Hilaga et al. [34] introduced a topology-matching technique for polyhedral models by comparing mutliresolutional Reeb graphs (MRGs). Chen and Ouhyoung [35] extended the work of Hilaga et al. for practical use by applying a pre-processing stage of creating MRGs for 3D models. Bespalov et al. [36], who applied an MRG-based method to a large number of

engineering parts,found that small changes in topology may markedly alter the similarity among models. Iyer et al. [37] used skeletal graphs to describe voxel models converted from solid models. The global feature vectors and skeletal graphs are processed in separate steps to overcome disadvantages associated with skeleton-based approaches [38].

Feature-based schemes are derived from manufacturing feature recognition. Elinson et al. [39] developed a technique based on a graphic representation of 3D model features, in which similarity is obtained from an equivalence hierarchy and classification trees. Ramesh et al. [40] extracted features from a B-rep model to identify similarities between parts by decomposing a model into simple shapes that resemble machining features. Cicirello and Regli [41] defined a model dependency graph (MDG) for each solid model by extracting machining features. The similarity of two solids was measured using the largest common subgraph to the two MDGs. Cardone et al. [42] recently applied algorithms to assess similarity between parts based on

machining features on three-axis machining centers. Tsai and Chang [43] developed a two-stage fuzzy approach for feature-based design retrieval. Though many feature-based techniques for assessing similarity between mechanical parts have been proposed, successful extraction during the feature recognition phase is difficult when using feature-based techniques, especially when complicated feature intersection exists. Additionally, the main focus of these approaches is machining (prismatic and turned) features on solid models rather than stamping features on surface models.

3

2.3 Representation of sheet-metal parts

A lack of consensus exists within the CBR community regarding information that should be included in a case [44].Most case representations are made of features related to manufacturing information. Zhang et al. [45] developed a surface-based approach for recognizing geometric features using surface orientation region subdivision and surface orientation area reconstruction. Sunil and Pande [46]proposed an algorithm composed of region segmentation and region merging to recognize features of sheet-metal components from the surface model in the StereoLithography format. Jargirdar et al. [47] presented a classification system for forming operations and concepts for extracting and recognizing forming features from 3D sheet-metal components created within a wireframe model. Tor et al.[15] developed a feature relational graph representation of astamped metal part to describe topological information. To generate appropriate process plans, Zheng et al. [48] utilized a hierarchical structural model for a panel by extracting stamping features based on feature and stamping technologies.Chen et al. [19] encoded part name, blank material, vehicle type, and name of key stamping processes for retrieving similar parts. Information for the major geometric features of a part is also considered to determine the geometric similarity between parts with same type of properties.

3 Case-based reasoning framework for metal stamping process planning and die design

The CBR approach is based on the principle that a new case can benefit from solutions to similar cases. Figure 1 shows a schematic diagram of the proposed CBR system for metal stamping process planning and die design. Successful cases,process plans, and die designs are stored in their respective libraries. Initially, a new automotive panel to be stamped is inputted into the CBR system. To facilitate case retrieval, the new sheet-metal part is analyzed to indentify fundamental characteristics and represented in a graph-based scheme for indexing. This indexed case is then passed to a case retriever to extract an existing case that most closely resembles the input case. The similarity between the input case and each case in the case library is assessed using the graph-based comparison method. After all similarities are calculated, the case with the highest similarity rank and its solutions are selected. The process plan and die sets relevant to the selected case are evaluated to determine whether the solution is satisfactory. If it is not satisfactory, the solution is revised and then reevaluated. Until a solution is satisfactory for the new design, the new case with its process plan and die sets are retained in the libraries. Whenever a new case is solved, the case library is expanded, and the CBR system is improved. This work focuses on case representation, indexing, and retrieval—the most important functions in the CBR system.

Fig. 1 Schematic diagram of the proposed case-based reasoning system

4

4 Representation and indexing 4.1 Flat-bend graph

Attributed graphs are used in many applications such as computer-aided process planning and feature recognition.Design and manufacturing features employed in computer-aided

applications are embedded in a graph signature. Thus, in this work, local features are compared during the graph comparison process. The case comparison result is the number of local features shared by the cases. Therefore, retrieval of similar cases with local feature correspondence is achieved by comparing attributed graphs.

Surface models with freeform surfaces are widely employed in stamping applications. This work proposes a novel graph-based descriptor, called the flat-bend graph,which represents these surface models. The B-rep scheme of a surface model can be transformed into an attributed graph as a case representation for case comparisons. Three fundamental entities—bend-type faces, flat-type faces, and holes—are identified from the panel model (Fig. 2). The flat-type faces are regarded as graph nodes representing the primary shape of a model. The bend-type faces are regarded as graph arcs for identifying the relationships among flat-type faces. The holes are the other node type, which also describes the model shape. The graph structure is built from

topological data of these fundamental characteristics. Additionally, graph attributes are collected from other geometric data. Figure 3 shows the process for constructing a flat-bend graph. First, all faces from the input surface model are analyzed to determine whether they are bend-type faces or flat-type faces. Once bend-type faces are detected, they are extracted from the model, and the remaining faces are flat-type faces.Connected flat-type faces are then gathered as a group. These groups are nodes in the flat-bend graph describing the primary shape of the surface model. Hole detection in each group is performed to obtain hole features from the model. After these steps, a graph can be built based on the relationships between group nodes and hole nodes connected to group nodes. Finally, some geometric characteristics are set as node attributes and arc attributes.

Fig. 2 Illustration of fundamental entities (flat-type faces, bend-type faces, and holes)

5

Fig. 3 The flat-bend graph construction process Fig. 6 An example of a simplified surface model

and its flat-bend graph

4.2 Graph construction

This section describes the construction of a flat-bend graph from the surface model of a stamping panel. The Spring-Solid system developed by the Solid Model Laboratory, National Taiwan University, is employed to obtain the Brep structure from the surface model in the

Standard for the Exchange of Product model data (STEP) format. Acquiring topological (faces, edges, and vertices) and geometric (surfaces, curves, and points) data from the B-rep structure facilitates construction of a flat-bend graph.

Several portions of a metal part on an automotive panel are bent at certain angles to form the body of the panels. Several stamping operations are needed because of the irregular shape of sheet-metal panels. These operations, such as drawing,flanging, re-striking, and piercing, are utilized to stamp that part into its final shape. Accordingly, the faces on the surface model that represent a sheet-metal part are categorized as either bend-type or flat-type. Most smooth and even areas on a panel are flat-type faces. Bend-type faces are bends occurring on the sheet-metal model and provide a number of clues that distinguish bend-type faces from other faces in the model. For example, surface curvature is an important clue for bend-type faces.

For each face of a model, several points are randomly chosen from the surface. At each point pi on the surface, the principal curvatures are calculated and denoted as κi,max and κi,min. The average curvature, κavg, is then calculated as

4.3 Case library

The case library is a repository of historical design cases as flat-bend graphs. One of the most significant issues in the CBR approach is the efficiency of retrieving the most similar case from a large case library. Cases indexed in an appropriate manner allow the system to determine whether two cases are similar and thereby acquire the most similar case easily. The flat-bend graph represents stamping models and supports rapid indexing of cases. The information of topological and geometric characteristics of a stamping part is condensed into a graph-based

representation with a structural form. These cases are stored in a relational database to form a case library. Each case in the relational library is regarded as an object composed of design

requirements, the original model, process plans, and die sets. This digital information is linked by data pointers stored in the case library. Therefore, process plans and die sets of the most similar case can be obtained from the information in the case library. When the most similar case is

6

retrieved, case solutions, such as process plans and die sets, are also retrieved. The case library initially contains few cases, but continuously expands as new cases are solved using the CBR system. Accordingly, the robustness and effectiveness of the CBR system increase as the number of cases stored in the case library increases.

Fig. 7 The process of detecting independent maximal cliques Fig. 8 The process of similar case assessment

5 Similarity assessment

5.1 Independent maximal clique detection

The quality and robustness of any CBR system depends on its retrieval mechanism. For simplicity, a variation on a classical subgraph isomorphism algorithm is used to assess similarity among attributed graphs. Therefore, the problem of assessing similarity between cases can be simplified as the problem of determining the maximum size of the common subgraph. Many algorithms have been developed for solving the maximum common subgraph (MCS) problem. One strategy is to reduce the MCS problem of two given graphs into a maximum clique (MC) problem [49].

A clique in a graph is a subset of nodes, such that all nodes in the subset are connected. The principal preprocess step in solving the MCS problem by finding the MC is to construct an

association graph from two given graphs. Two nodes with the same attributes in different graphs are mapped together as a mapping node in the association graph. When two nodes in one

attributed graph and their mapped nodes in the other graph have the same connectivity (connected or not connected), these two mapping nodes in the association graph are connected by an arc. Accordingly, the MC in the association graph is equivalent to the MCS between two attributed graphs. Hence, the problem of inding the MCS is transformed into an MC detection problem, i.e., finding the largest clique in the association graph.

To efficiently obtain applicable solutions, an effective scheme called independent maximal cliques (IMC) detection described in [29] is employed. The IMC detection with a simulated

annealing algorithm is used to solve the graphmatching problem. This method measures the local feature correspondence between 3D solid models and outperforms traditional MC methods in terms of both efficiency and efficacy. Hence, applying the IMC detection method in this work is reasonable. The IMC are several maximal cliques in a graph with no vertex belonging to two cliques simultaneously. The IMC detection architecture finds the maximum size of IMC from an association graph built from two attributed graphs. Figure 7 shows the process for detecting IMC. A novel algorithm based on simulated annealing is utilized to identify the maximal clique

effectively and efficiently. When a maximal clique is found, the size of the clique is assessed. If clique size is ≥3, the original attributed graphs are modified by deleting retrieved vertices. The association graph is then reconstructed from the modified attributed graphs for another clique detection iteration. If clique size is <3, the procedure stops. When the detection stage is complete, the final detection result is the sum of graph sizes of IMC.

Both IMC detection and the simulated annealing algorithm are applied to solve the

graph-matching problem. The association graph, which is used for IMC detection, is constructed

7

from two attributed graphs while retaining their attributes. All IMC, which represent the

maximum number of common features between cases, are identified via the simulated annealing algorithm. Therefore, the retrieval framework via IMC detection can achieve the goal of retrieving the most similar case from the case library.

Fig. 9 Illustration of correspondence of local features and their equivalent subgraphs

5.2 Similar case assessment

The IMC detection scheme acquires the maximal clique of the association graph repeatedly in a heuristic manner. Once an assessment is complete, the sizes of IMC are summed to calculate the similarity factor between these two cases. Factor S, which is employed to assess the similarity between two graphs, is calculated as

where N m is the total number of vertices in each independent maximal clique, N1 is the number of nodes in the graph of the query case, and N2 is the number of nodes in the graph of the candidate case for comparison. Figure 8 shows the process of comparing the query case with a candidate case using the proposed heuristic algorithm. First, two cases are uploaded by a user or obtained from the case library. The flat-bend graphs are then obtained from their case representations. The IMC are acquired by the IMC detection procedure combined with the simulated annealing algorithm. Finally, the similarity factor is calculated using Eq. 4.

If two cases have many common features, their similarity factor is high. Conversely, the factor is low when few common features exist. The retrieval operation compares the query case with all cases in the case library and obtains similarity factors for each pair of comparisons. From these comparison results, the case with highest factor is the most similar case. Because the query case and selected case have many shared features, process plans and die designs of the selected case can be reused in the new design. Hence, case solutions for the query case can enhance design and manufacturing efficiencies.

8

Fig. 11 Illustration of the case retrieval system

Fig. 14 Illustration of die set revision

6 Results

Representing a sheet-metal part as a flat-bend graph has practical applications in partial matching. Given two models with attributed graphs, the subpart correspondence can be obtained by finding the subgraph common to the two graphs. Figure 9 shows the local feature

correspondence of two cases with common subgraphs. A portion of the entire model and its

subgraph are extracted in both cases. The bend-type faces are marked red, and the remaining faces are flat-type faces of the model. In the subgraphs, circular nodes represent flat groups, continuous arcs represent concave connections, and dashed arcs represent convex connections. Although the number of flat-type faces and bend-type faces in both cases differ, the expected result indicates the correspondence between local features. The nodes and arcs of these two subgraphs are mapped correctly. Therefore, if two models have a common local feature, a partial graph pattern of this feature must exist in each of the two attributed graphs. The correspondence of local features can be obtained in graphic form by determining whether a common subgraph exists. The same shape may have different features defined in a standard manufacturing library. However, most feature- based methods are not sufficiently robust in extracting manufacturing features and remain a

9

significant research challenge. Comparing parts with local feature correspondence helps overcome this problem schemes based on manufacturing feature recognition.

Figure 10 shows the similarity measurement result obtained through IMC detection, where Ni is the number of nodes in a flat-bend graph, and Sij is the similarity factor between Part-i and Part-j. Models with the number of nodes in the same order are chosen in this example to produce feasible results from the correspondence of local features. Part-1 and Part-2 differ only slightly and have many common features and, thus, have a high similarity factor (Fig. 10). Conversely, the similarity factor between Part-1 and Part-3 is small due to a shortage of common features, as is that for Part-2 and Part-3. This example demonstrates that IMC detection can determine whether two models are similar and can be applied to retrieve the similar models.

The proposed framework is implemented in Java. The CBR system for automotive panels is added to the PDM system with a part library, a process plan library, and a die set library.

Figures 11, 12, 13, and 14 show the development of the process plan and die sets of a new case via the CBR methodology. Initially, the new case, f07.stp, is selected, and the flat-bend graph is built in the CBR system. When the new case is ready for comparison with existing cases, the search dialog is activated. Figure 11 shows the main window for retrieving cases that are similar to the query case. The query model is displayed in the viewer on the left, and retrieved models are shown on the right with their names and similarity factors. The retrieval process compares the query model with all models in the case library. In this example, only the first ten closest models are selected from the comparison result and displayed in comparison result list. The retrieved models are ordered from high to low according to their similarity factors. Figure 12 shows the three most similar models and their similarity factors. Several features exist and are mixed

together in the query model. In the proposed similarity assessment process, the similarity measure for each comparison depends on the local feature correspondence. This experimental result

suggests that numerous local features are common between the best solution (f10.stp) and the new case (f07.stp). After retrieving the most similar case from the case library, the previous solution, process plan, and die design are obtained from the process plan library and die set library, respectively. The retrieval operation sequence and process plans of f10.stp are regarded as

references for the new case (Fig. 13). To find the new solution, several modifications are needed from the references in the process planning system. Notably, the die sets of the new case are also developed by modifying the retrieved die sets to meet new design requirements (Fig. 14). This example demonstrates that the proposed

CBR system with the proposed indexing and retrieval method effectively retrieves similar cases for further development to improve the efficiency of the design and manufacturing processes.

Most manufacturing components, such as process plan and die sets, are examined using model features. Manufacturing information (case solutions) in retrieved cases can be obtained and modified for a new design; this eliminates the need to design a product from scratch. Therefore, the proposed CBR approach enhances design and manufacturing efficiency by retrieving similar cases for reuse.

7 Conclusions

Sheet-metal parts are widely used in mass production in the automotive industries. However, stamping process planning and die design demand considerable skills and experience from

designers. Reusing an existing design enhances product development efficiency. Employing the proposed CBR methodology to retrieve a problemsolving case for reuse/revision of existing process plans and die sets for automotive panels is efficient. Therefore, retrieving an appropriate case is key in the CBR methodology and has a marked impact on system efficiency. Case representation and similarity assessment are the major tasks that improve process efficiency in automotive panel manufacturing.

This work applies a CBR approach with the novel indexing and retrieval strategy for automotive stamping process planning and die design. The most significant contribution of this work is that it converts a surface model of an automotive panel into a flat-bend graph for indexing and retrieval by the CBR architecture. Geometric data of the surface model are categorized into bend-type faces, flat-type faces, and holes. The groups, which are composed of connected flat-type faces, and holes are represented as graph nodes. The bend-type faces that identify the

10

relation-ships between groups are represented as graph arcs. Several attributes are assigned to the nodes and arcs to thoroughly describe the model. A novel heuristic method, IMC detection, is applied to find the IMC for solving the graph-matching problem efficiently. Case similarity is acquired by finding the maximum size of IMC based on local feature correspondence using IMC detection. The CBR system is designed for surface models with a B-rep structure in the neutral STEP format on a PDM system. The system based on the proposed scheme can retrieve cases (models) similar to a query case (model). The case solutions corresponding to the retrieved case are obtained for reuse and further development. Adapting existing process plans and press die sets reduces design effort and increases production efficiency.

Further work should enhance the performance of the proposed scheme in automatic adaption. Adapting a retrieved case to meet the requirements of a new stamping part automatically will further improve development effi- ciency. However, the proposed architecture only allows for manual modification of a retrieved process plan and die sets. For further development, automatic adaption should be considered to achieve the goal of automated design.

References

1. Cheok BT, Nee AYC (1998) Trends and developments in the automation of design and manufacture of tools for metal stampings. J Mater Process Tech 75(1–3):240–252

2. Ullman DG (1997) The mechanical design process, 2nd edn. McGraw-Hill, New York 3. Riesbeck CK, Schank R (1989) Inside case-based reasoning. Erlbaum, Northvale, NJ

4. Watson I (1999) Case-based reasoning is a methodology not a technology. Knowl-Based Syst 12(5–6):303–308. doi:10.1016/S0950-7051(99)00020-9

5. Iyer N, Jayanti S, Lou K, Kalyanaraman Y, Ramani K (2005) Three-dimensional shape searching: state-of-the-art review and future trends. Comput-Aided Des 37(5):509–530. doi:10.1016/j. cad.2004.07.002

6. Cardone A, Gupta SK, Karnik M (2003) A survey of shape similarity assessment algorithms for product design and manufacturing applications. ASME Journal Comput Inf Sci Eng 3(2):109–118. doi:10.1115/1.1577356

7. You CF, Chao SN (2006) Multilayer architecture in collaborative environment. Concurr Eng Res Appl 14(4):273–281. doi:10.1177/1063293X06073128

8. Sycara K, Navin Chandra D, Guttal R, Koning J, Narasimhan S(1991) CADET: a case-based synthesis tool for engineering design. In J Expert Syst 4(2):157–188

9. Sun SH, Chen JHL (1996) A fixture design system using case-based reasoning. Eng Appl Artif Intel 9(5):533–540. doi:10.1016/0952-1976(96)00048-6

10. Tiwari MK, Kotaiah KR, Bhatnagar S (2001) A case-based computer-aided process-planning system for machining prismatic components. Int J Adv Manuf Technol 17(6):400–411. doi:10.1007/s001700170158

11. O’Connor LJ, Partridege LMS, DR LJMF (1992) A case-based reasoning approach to automated weld-process design. Appl Artif Intell 6(3):315–330

12. Kwong CK (2001) A case-based system for process design of injection moulding. Int J Comput Appl T 14(1–3):40–50. doi:10.1504/IJCAT.2001.000259

13. Lei Y, Peng Y, Ruan X (2001) Applying case-based reasoning to cold forging process planning. J Mater Process Technol 112(1):12–16. doi:10.1016/S0924-0136(00)00852-9

14. Lee KS, Luo C (2002) Application of case-based reasoning in die-casting die design. Int J Adv Manuf Tech 20(4):284–295.doi:10.1007/s001700200154

15. Tor SB, Britton GA, Zhang WY (2003) Indexing and retrieval in metal stamping die design using case-based reasoning. J Comput Inf Sci Eng 3(4):353–362. doi:10.1115/1.1630339

16. Zhang WY, Tor SB, Britton GA (2006) Indexing and retrieval in case-based process planning for multi-stage non-axisymmetric deep drawing. Int J Adv Manuf Tech 28(1–2):12–22. doi:10.1007/s00170-004-2333-y

17. Liu W, Yang YY, Xing ZW (2004) Study on process decision methodology for sheet metal stamping by case based technology.Mater Sci Forum 471–472:895–899

18. Leake DB, Birnbaum L, Hammond K, Marlow C, Yang H (1999) Integrating diverse information resources in a case-based design

environment. Eng Appl Artif Intel 12(6):705–716. doi:10.1016/S0952-1976(99)00042-1

19. Chen G, Chen J, Zhao Z, Ruan XY (2005) An object-oriented hierarchical case representation automotive panels in a computer-aided process planning system. Int J Adv Manuf Tech 26(11–12):1323–1330. doi:10.1007/s00170-004-2105-8

20. Tangelder JWH, Veltkamp RC (2008) A survey of content based 3D shape retrieval methods. Multimedia Tools and Applications 39(3):441–471. doi:10.1007/s11042-007-0181-0

21. Iyer S, Nagi R (1997) Automated retrieval and ranking of similar parts in agile manufacturing. IIE Trans 29(10):859–876

22. Hermann JW, Singh G (1997) Design similarity measures for process planning and design evaluation.

Technical Research Report, TR97-74. Institute of Systems Research, University of Maryland, College Park

11

23. Osada R, Funkhouser T, Chazelle B, Dobkin D (2001) Matching 3D models with shape distributions. Shape

Modeling International, Genoa, pp 154–166

24. Ip CY, Lapadat D, Sieger L, Regli WC (2002) Using shape distributions to compare solid models.

Proceedings of the 7th ACM Symposium on Solid Modeling and Applications, Saarbrucken, 17–21 Jun 2002, pp 273–280

25. Hong T, Lee K, Kim S (2006) Similarity comparison of mechanical parts to reuse existing designs.

Comput-Aided Des 38(9):973–984. doi:10.1016/j.cad.2006.05.004

26. McWherter D, Peabody M, Regli WC, Shokoufandeh A (2001) An approach to indexing databases of graphs.

Technical report,Drexel University, Philadelphia

27. Bespalov D, Regli WC, Shokoufandeh A (2006) Local feature extraction and matching partial objects.

Comput-Aided Des 38 (9):1020–1037. doi:10.1016/j.cad.2006.07.005

28. Biasotti S, Marini S, Spagnuolo M, Falcidieno B (2006) Sub-part correspondence by structural descriptors of

3D shapes. ComputAided Des 38(9):1002–1019. doi:10.1016/j.cad.2006.07.003

29. You CF, Tsai YL (2009) 3D solid model retrieval for engineering reuse based on local feature

correspondence. Int J Adv Manuf Technol 46(5–8):649–661. doi:10.1007/s00170-009-2113-9

30. Liu W, He Y (2008) Representation and retrieval of 3D CAD models in parts library. Int J Adv Manuf

Technol 30(9–10):950–958. doi:10.1007/s00170-006-0914-7

31. Sundar H, Silver D, Gagvani N, Dickinson S (2003) Skeleton based shape matching and retrieval.

Proceedings of the Shape Modeling International, Seoul, 12–15 May 2003, pp 130–139

32. El-Mehalawi M, Miller RA (2003) A database system of mechanical components based on geometric and

topological similarity. Part I: representation. Comput-Aided Des 35(1):83–94. doi:10.1016/S0010-4485(01)00177-4

33. El-Mehalawi M, Miller RA (2003) A database system of mechanical components based on geometric and

topological similarity. Part II: indexing, retrieval, matching and similarity assessment. Comput-Aided Des 35(1):95–105. doi:10.1016/ S0010-4485(01)00178-6

34. Hilaga M, Shinagawa Y, Kohmura T, Kunii TL (2001) Topology matching for fully automatic similarity

estimation of 3D shapes. Proceedings of ACM SIGGRAPH, Los Angeles, 12–17 Aug 2001, pp 203–212 35. Chen DY, Ouhyoung M (2002) A 3D object retrieval system based on multi-resolution reeb graph.

Proceeding of Computer Graphics Workshop, Taiwan, June 2002

36. Bespalov D, Regli WC, Shokoufandeh A (2003) Reeb graph-based shape retrieval for CAD. Proceedings of

the ASME DETC’03 Computers and Information in Engineering Conference,Chicago, 2–6 Sept 2003

37. Iyer N, Kalyanaraman Y, Lou K, Jayanti S, Ramani K (2003) A reconfigurable 3D engineering shape search

system Part I: shape representation. Proceedings of the ASME DETC ’03 Computers and Information in Engineering Conference, Chicago, 2–6 Sept 2003

38. Lou K, Jayanti S, Iyer N, Kalyanaraman Y, Ramani K, Prabhakar S (2003) A reconfigurable 3D engineering

shape search system Part II: database indexing, retrieval and clustering. Proceedings of the ASME DETC ’03 Computers and Information in Engineering Conference, Chicago, 2–6 Sept 2003

39. Elinson A, Nau D, Regli WC (1997) Feature-based similarity assessment of solid models. Proceedings of the

fourth ACM SIGGRAPH Symposium on Solid Modeling and Applications,Atlanta, 14–16 May 1997, pp 297–310

40. Ramesh M, Yip-Hoi D, Dutta D (2001) Feature based shape similarity measurement for mechanical parts. J

Comput Inf Sci Eng 1(3):245–256. doi:10.1115/1.1412456

41. Cicirello V, Regli WC (2001) Machining feature-based comparisons of mechanical parts. Proceedings of

Shape Modeling International, Genoa, 7–11 May 2001, pp 176–185

42. Cardone A, Gupta SK, Deshmukh A, Karnik M (2006) Machining feature-based similarity assessment

algorithms for prismatic machined parts. Comput-Aided Des 38(9):954–972. doi:10.1016/j.cad.2006.08.001 43. Tsai CY, Chang CA (2005) A two-stage fuzzy approach to feature-based design retrieval. Comput Ind

56(5):493–505.doi:10.1016/j.compind.2005.02.001

44. Watson I, Marir F (1994) Case-based reasoning: a review. Knowl Eng Rev 9(4):327–354

45. Zhang XQ, Wang J, Yamazaki K, Mori M (2004) A surface based approach to recognition of geometric

features for quality freeform surface machining. Comput-Aided Des 36(8):735–744.doi:10.1016/j.cad.2003.09.002

46. Sunil VB, Pande SS (2008) Automatic recognition of features from freeform surface CAD models.

Comput-Aided Des 40(4):502–517. doi:10.1016/j.cad.2008.01.006

47. Jagirdar R, Jain VK, Batra JL (2001) Characterization and identification of forming features for 3-D sheet

metal components.Int J Mach Tool Manu 41(9):1295–1322. doi:10.1016/S0890-6955(01)00006-2

48. Zheng JQ, Wang YL, Li ZG (2007) KBE-based stamping process paths generated for automobile panels. Int

J Adv Manuf Tech 31(7–8):663–672. doi:10.1007/s00170-005-0239-y

49. Bunke H, Foggia P, Guidobaldi C, Sansone C, Vento M (2002) A comparison of algorithms for maximum

common subgraph on randomly connected graphs. Structural, syntactic, and statistical pattern recognition. Lect Notes Comput Sci 2396:123–132.doi:10.1007/3-540-70659-3_12

12