Segmentation of Dimensionally-Large Rapid Prototyping Objects Y. Tang, H. -T. Loh, J. -Y. -H. Fuh, Y. -S. Wong, S. -H. Lee AbstractAn algorithm was developed to enable efficient segmentation of dimensionally-large objects into smaller components that can be fabricated within the given Rapid Prototyping (RP) machine workspace. The algorithm uses vertical and horizontal flat plane cuts, as well as feature -based volume decomposition. Due considerations were given to the optimisation of the surface accuracy, the build time, the strength and the number of segments generated by the segmentation process. A computer-aided design (CAD) application programme that interfaces with Unigraphics (UG) was also developed to allow import of objects in Standard Triangulated Language (STL) files into UG without loss of accuracy. In addition, the application software provides the functions that facilitate the implementation of the segmentation algorithm in UG. Two case studies were carried out using the algorithm in a Selective Laser Sintering (SLS) RP system. The resulting objects had properties that matched the research objectives with which the proposed algorithm was validated. Index termsComputer-aided Design (CAD), Rapid Prototyping (RP), Selective Laser Sintering (SLS), Segmentation I. INSTRODUCTION C ONVENTIONAL Rapid Prototyping (RP) processes, e.g. Stereolithography (SLA), Selective Laser Sintering (SLS), 3-Dimensional Printing (3DP), and Fused Deposition Modelling (FDM), etc., enable the fabrication of an object through a fully additive process of building thin layers of the object. When the object is too large to fit within the RP machine workspace, the conventional way is to fabricate the object in a bigger RP machine with a larger workspace. Manuscript received October 30, 2002. This work was jointly supported by Singapore-MIT Alliance and Department of Mechanical Engineering, National University of Singapore. Y. Tang is with Singapore-MIT Alliance, National University of Singapore, E4-04-10, 4 Engineering Drive 3, Singapore 117576 (phone: 6568744857; fax: 65-67795922; e-mail: smatyx@nus.edu.sg). H. -T. Loh is with the Mechanical Engineering Department, National University of Singapore, 10 Kent Ridge Crescent, Singapore 119260 (e-mail: mpelht@nus.edu.sg). J. -Y. –H. Fuh is with the Mechanical Engineering Department, National University of Singapore, 10 Kent Ridge Crescent, Singapore 119260 (e-mail: mpefyh@nus.edu.sg). Y. -S. Wong is with the Mechanical Engineering Department, National University of Singapore, 10 Kent Ridge Crescent, Singapore 119260 (e-mail: mpewys@nus.edu.sg). S. -H. Lee was a final year student of 2002 with the Mechanical Engineering Department, National University of Singapore, 10 Kent Ridge Crescent, Singapore 119260. But having a larger workspace is not always feasible, because machine cost escalates exponentially with the increasing workspace dimension, and the workspace can only be increased up to a certain size due to technical limitations. Alternatively, the object can be segmented into smaller components that are then fabricated in the existing RP machine, following which they are reassembled back into a whole object. Personnel experienced in RP area currently performs this type of object segmentation manually and on an ad hoc basis. But such operations are not always reliable and can be very complicated and time -consuming when the object has complex features. Hence it is necessary to develop a more rigorous and algorithmic framework so that the process can be easily carried out and not be so dependent on the skill and knowledge of the experienced operator. Segmentation algorithms for large objects have been developed in certain newer RP systems by Horváth et al [1] and Beascoechea-Gangoiti et al [2]. But such algorithms are not directly applicable to conventional RP systems due to fundamental differences between the conventional and the newer RP systems. For conventional RP systems, Ilinkin et al [3] developed a segmentation algorithm with an optimal decomposition of both convex and non-convex objects using a single horizontal flat cutting plane, with considerations for minimisation of support volume for the SLA process and the ability to control the number of decomposed parts. However, the algorithm does not discuss about larger objects that require more than one cutting plane, or cutting planes in other directions for the purpose of achieving the desired segment size. Qian et al [4,5] developed a feature-based segmentation method such that each feature can be fabricated in its optimal orientation. This allows fabrication of parts with a higher precision, higher quality and in less time. This work also introduces the interaction-free volume decomposition algorithm, which enables elimination of staircase interaction effects between adjacent feature components. Similar work was developed by Wang et al [6], in which a big or complex CAD model is decomposed into several simpler or smaller components according to their geometrical features. The components are then optimised to their individual orientations, and packed optimally within the RP workspace. Although these feature-based decompositions are applicable to decomposing a large object and thus enabling its fabrication, it is observed that pure decomposition by features may generate components with thin walls that are fragile, and corners with very acute angles that are difficult to build. Also, there is no guarantee that the object segment will fit within the RP workspace. Another research work developed by Chang and Prinz et al [7] uses the layer decomposition principles for a new additive/subtractive RP technology called Shape Decomposition Machining (SDM), in which an object is fabricated through the additive deposition of materials for free-form layers of the object with undercuts, and the machining of the freeform layers without undercuts. The object segmentation in this case can generate different freeform layers consisting of entirely undercut surfaces or nonundercut surfaces. Surface geometry, particularly the undercut and non-undercut surfaces, are used to determine the parting surfaces for each layer, hence allowing for free-form parting surfaces instead of only flat planar parting surfaces. Though this freeform object decomposition methodology is specific for SDM only, some relevant insights about object segmentation can still be obtained. Another relevant aspect about object segmentation is the optimisation of object orientation for fabrication in RP systems. This paper develops a segmentation algorithm that combines the advantages of both feature volume decomposition method and flat planar parting surfaces method, while reducing or eliminating their disadvantages. The segmentation will generate a minimum number of object segments, produce an object with the optimal surface and dimensional accuracy, and require minimum build time. The segmentation algorithm should be robust enough for application to different conventional RP systems, with a minimal amount of customization needed. The segmentation should not generate new thin segments or walls that are structurally weak, and should allow for easy location of segments during assembly of segments into the object. The final assembled object should also have good overall strength and dimensional accuracy, especially at the parting surfaces of the segments. The implementation of the algorithm will be carried out using a customized CAD application program, which will allow import, manipulation and export of Standard Triangulated Language (STL) object file. The CAD application should also provide functions to facilitate interactive segmentation of the large object with the proposed algorithm. Two case studies will then be performed to demonstrate and study the feasibility of the proposed algorithm. II. SEGMENTATION ALGORITHM To enable detailed formulation of the algorithm, this novel algorithm was developed based on a particular conventional RP process, namely the SLS process. The algorithm consists of the following eight steps: Feature, Union, Isolation, Orientation, Feature Interaction Volume (FIV), X, Y, Z-cuts, Recombination, and Packing. A. Feature Under a CAD environment, decompose the object model into separate and distinct feature components, using the existing boundary lines between features as the parting loops of the parting surface as much as possible. After feature decomposition, identify critical surfaces that have maximum possible surface accuracy. Examples of critical surfaces are cylindrical and conic surfaces. Other examples of critical surfaces are curved surfaces that require maximal surface accuracy, as part of the surfaces’ functional requirements. Following this, identify the feature components with critical surfaces as critical surface features (Fc), and identify the remaining feature components, which have non-critical surfaces, as non-critical surface features (Fnc). Performance of this feature step will result in a feature decomposition of the object, and the identification of features with maximal surface accuracy requirement. Corresponding parting surfaces will be generated too. B. Union For each parting surface on the critical surface of an Fc, check whether the corresponding parting surface belongs to the Fnc. If it happens, union the Fc and Fnc with this parting surface. On the other hand, if the corresponding parting surface belongs to another Fc, do not union these two Fc but take a note of this parting surface. The rest of the Fnc that are not connected to any Fc will remain as separate feature components. C. Isolation For each Fc that has an adjacent Fnc united to it, isolate and separate this Fc from the adjacent Fnc again using flat cutting planes on the body of each Fnc, ensuring that each Fc shall still remain whole and uncut, as well as the cuts are only performed on the body of each attached Fnc. These cuts will produce modified Fc with portions of the previously attached Fnc, as well as modified Fnc with a new shape. The selection of flat cutting planes shall follow the rules that cut planes are normal to the z, y or x axes and with a minimum portion of Fnc left to attach Fc. Then, unite the adjacent Fnc together at the corresponding parting surface of each pair of adjacent Fnc. All the Fnc that can be united together will form a new and larger Fnc component. All Fnc that cannot be united with other Fnc will remain as a separate Fnc component. All original and modified Fc will remain unchanged during the union of the Fnc components. D. Orientation For each original or modified Fc component, orient it to the maximum build accuracy orientation. For each of the remaining Fnc components, orient it to the best build time or build accuracy orientation. Finally individually rotate each Fc and Fnc components about the z-axis such that the longest horizontal dimension of each component is aligned along the workspace x-axis. E. Feature Interaction Volume (FIV) For each Fc or Fnc, if the surface normal at any point on the parting surface is neither perpendicular nor parallel to the z build direction, staircase interaction will be resulted because of the layered manufacturing mechanism of RP. Hence, a Feature Interaction Volume (FIV) is needed for the corresponding two components. The FIV is a solid connection between features generated from the parting surface. All the planar, uniformly curved or freeform parting surfaces on all the components should be checked for necessity of FIV formation. The method of generating a FIV is projecting the original paring surface to the bottom surface of each feature, then cut the object to the smallest volume. After all required FIV has been formed, orient each FIV to the orientation with the normal of parting surface perpendicular or parallel to the z build direction. F. X, Y, Z-cuts As a result of above feature-based segmentation, a set of components of Fc, Fnc or FIV will be produced. But these components may be larger than the RP working space. Hence all components should be further segmented by X, Y, Z cuts to ensure that all sub-components after the segmentation are fit within the size limitation of the RP machine in X, Y, Z direction. G. Recombination For each pair of corresponding parting surfaces to all sub-components, if the surface normal of all points on both parting surfaces are all perpendicular or parallel to the z build direction, label the pair of parting surfaces as a Possible Recombination Surface (PRS) pair. Then selectively recombine sub-components of the PRS pair such that the resulting object segments can still fit within the RP workspace, and the overall number of object segment is minimised. FIV can also be recombined if above conditions are met. H. Packing Optimally pack all sub components into one RP workspace boundary to prepare for fabrication. and Fnc identification), union, orientation, isolation, XYZcutting, recombination, and packing are performed for this object segmentation. FIV is not necessary here because there is no staircase interaction between any two segments. Optimisations of the object requirements are also considered based on the surface accuracy, build time, part strength and the number of segments throughout the segmentation process. The segmentation process of case study 2 is shown in Fig. 2 (a)∼(f). Unlike case study 1, case study 2 has no Fnc at all. Hence there is no union step for this case. But it is necessary to generate FIV here because for some parting surfaces, the staircases are unavoidable. After segmentation, both the case study 1 and 2 were implemented in a SLS system (shown in Fig. 3), i.e., all segmented and packed components were sintered by the selective laser sintering machine using a plastic material. Then the built components were assembled to the original large objects. It is observed that the final assembled objects have a good overall accuracy, a short build time and a high strength. V. CONCLUTION A novel algorithm of large-object segmentation for the SLS RP system has been successfully developed. The algorithm combines two segmentation methodologies: flat plane segmentation and feature-based segmentation. The algorithm has successfully utilised the advantages of these two methodologies while minimising their inherent disadvantages. With the self-developed CAD application RP-T, the algorithm has been successfully implemented in UG CAD system based on the two case study objects. The two objects were then successfully fabricated using an in-house SLS system. The properties of these two objects were acceptable and met the user’s requirements and expectations, and hence showing that the segmentation algorithm is practically feasible. III. CAD APPLICATION A CAD system, called Rapid Prototyping Toolkit (RPT), has been developed to facilitate object segmentation using the proposed algorithm. RP-T was developed on a HP Visualise 2000 graphical workstation with HP-UX operating system, and interfaces with Unigraphics (UG) version 15.0.3. RP-T consists of the following two modules: STL import module and Interactive Object Segmentation Assistance Module (IOSAM). The STL import module imports a STL file into UG using boundary representation (B-Rep), without loss of accuracy during the import. This allows for manipulation of the STL object in UG environment. The IOSAM is to provide functions to perform the object segmentation process using the algorithm described as above. IV. CASE STUDY Two case studies were conducted to testify the segmentation algorithm. The segmentation process of case study 1 is shown in Fig. 1 (a)∼(k), where feature analysis (Fc REFERENCES [1] Horváth, I., Broek, J. J., Rusák, Z., Kuczogi, G., Vergeest, J. S. M., “Morphological Segmentation of Objects for Thick-layered Manufacturing”, Proceedings of 1999 ASME Design for Manufacturing Conference, Las Vegas, CA, September 1999, pp. 18-24 [2] Beascoechea-Gangoiti, A., 3D packing for solid freeform fabrication: A heuristic approach, Ph.D. thesis, Rensselaer Polytechnic Institute, Troy, NY, 1996 [3] Ilinkin, I., Janardan, R., Majhi, J., Schwerdt, J., Smid, M., Sriram, R., “A Decomposition-Based Approach to Layered Manufacturing”, Proceedings of the Seventh International Workshop on Algorithms and Data Structures, Providence, RI, 2001, pp. 211-216 [4] Qian, X., Dutta, D., “Feature based fabrication in layered manufacturing”, Transactions of the ASME, Journal of Mechanical Design, Vol.123, No. 3, pp. 337-345, 2001 [5] Qian, X., Dutta, D., “Features in the Manufacturing of Heterogeneous objects,” Proceedings of the Solid Freeform Fabrication Symposium, University of Texas at Austin, Austin, Texas, 1999, pp. 689-696 [6] Wang, W. L., Fuh, J. Y. H., Wong, Y. S., Miyazawa, T., “Make-up Fabrication of Big or Complex Parts using the SLA Process”, Proceedings of the Solid Freeform Fabrication Symposium, University of Texas at Austin, Austin, Texas, 1996, pp. 299-306 [7] Chang, Y. C., Pinilla, J. M., Kao, J. H., Dong, J., Ramaswami, K., Prinz, F. B., “Automated Layer Decomposition for Additive/Subtractive Solid Freeform Fabrication”, Proceedings of the Solid Freeform Fabrication Symposium, University of Texas at Austin, Austin, Texas, 1999, pp 111-120 United Fc and Fnc FIGURES United Fc and Fnc (c) Isolated modified Fc Horizontal planar cut (a) Vertical planar cut (d) Critical surface Fc Critical Fnc surface Fnc (b) United adjacent Fnc (e) Oriented Fnc Oriented Fc Build direction X-cut (i) (f) Z-cut Z-cut Recombined components Z-cut (j) (g) No y-cut needed RP machine workspace boundary Build direction Z (k) (h) Fig. 1 Segmentation process of case study 1: (a) Original object, (b) Two Fc identified. Rest of features is Fnc, (c) Fc united with neighboring Fnc, (d) Isolate all Fc from Fnc with planar cuts, (e) Union all Fnc, (f) Orient Fc to required optimal orientation. Orient Fnc for optimal build time/accuracy, (g) Do z-cuts for each component. Three z-cuts are made, (h) Do y-cuts on each sub-component. No y-cut is needed, (i) Do x-cuts on each sub-component. One x-cut is needed, (j) Perform re-combination of sub components where possible, (k) All sub-components reoriented as necessary and packed into RP machine workspace Fc FIV Fc FIV Fc Fc FIV Fc FIV (d) (a) All Fc Fc and adjacent Fc do not unite Z Z-cut Parting surface joined to Fc on two sides Build direction (e) All Fc (b) Z FIV Build direction (f) FIV (c) Fig. 2 Segmentation process for case study 2 (a) Original object, (b) Feature identification. All features are Fc. Skip Fc union with Fnc, since no Fnc exist. Skip isolation of Fc, since all Fc are already isolated, (c) Design FIV. Four pairs of corresponding parting surfaces joins to Fc on both sides; four FIV are required, (d) Orient all Fc and FIV to required build orientation. Skip orientation of Fnc as there is no Fnc, (e) Perform z-cuts. Y and x-cuts are skipped as they are not needed, (f) Recombination not needed. Finally pack and orient sub-components (a) (d) Fig. 3 Implementation of Case Study 1 and 2: (a) Segmented and SLS Built Components of Case Study 1, (b) Segmented and SLS Built Components of Case Study 2, (c) Final Assembled Object of Case Study 1, (d) Final Assembled Object of Case Study 2 (b) (c)