- Email: [email protected]
Multi-material virtual prototyping for product development and biomedical engineering
Computers in Industry 58 (2007) 438–452 www.elsevier.com/locate/compind
Multi-material virtual prototyping for product development and biomedical engineering S.H. Choi *, H.H. Cheung Department of Industrial and Manufacturing Systems Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong, China Received 27 January 2005; accepted 8 September 2006 Available online 18 October 2006
Abstract This paper describes the development of a multi-material virtual prototyping (MMVP) system, which integrates virtual reality (VR) and layered manufacturing (LM) for digital fabrication of multi-material prototypes to facilitate advanced product development and biomedical engineering. The MMVP system incorporates a topological hierarchy-sorting algorithm, two algorithms for sequential and concurrent multi-toolpath planning, a build-time estimation algorithm, and a virtual prototyping system. Based on the topological hierarchy of slice contours, the sequential multitoolpath planning algorithm generates sequential toolpaths that avoid redundant tool movements, while the concurrent multi-toolpath planning algorithm generates collision-free concurrent toolpaths to further reduce the build-time. To verify the resulting toolpaths and to facilitate estimation of fabrication cost, a mathematical model is incorporated for build-time estimation of multi-material fused deposition processes. The MMVP system is useful for planning, stereoscopic simulation, build-time estimation and validation of multi-toolpaths, and subsequent digital fabrication and analysis of multi-material prototypes. # 2006 Elsevier B.V. All rights reserved. Keywords: Multi-material prototyping; toolpath planning; Build-time
1. Introduction Multi-material prototypes may be made of materials that change gradually from one type to another, or of a collection of heterogeneous materials. In contrast to single-material prototypes, multi-material prototypes can differentiate clearly one part from others, or tissues from blood vessels of a human organ; and they perform better in rigorous environments [1]. As high value-added products now often involve complex designs while surgical operations become more complicated, there are strong demands for multi-material prototypes, especially for advanced product development and biomedical applications. It is therefore desirable and timely to develop multi-material layered manufacturing (MMLM) technology for fabrication of multi-material prototypes [2]. MMLM technology provides a relatively convenient way to fabricate multi-material parts, such as electronic products, advanced communication components, drug delivery devices,
* Corresponding author. Tel.: +852 2859 7054; fax: +852 2858 6535. E-mail address: [email protected] (S.H. Choi). 0166-3615/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.compind.2006.09.002
and innovative cellular and cell-containing tissue scaffolds [3,4]. Although its benefits have been widely recognised, practical MMLM machines have yet to be developed, particularly for fabrication of relatively complex objects. In general, an MMLM machine may consist mainly of two sub-systems namely: (i) a hardware material-depositing mechanism and (ii) a computer software system for processing complex slice contours to plan multi-toolpaths that control the material deposition mechanism. Some researchers have developed experimental MMLM machines for relatively simple objects. Lappo et al. [5] developed a discrete multiple material selective laser sintering (M2SLS) machine for fabrication of prototypes with discrete multiple materials. Yang and Evans [2] proposed a multicomponent powder-dispensing device for selective laser sintering (SLS) to make parts of 3D functional gradients. Jafari et al. [6] extended the fused deposition modelling (FDM) hardware mechanism to form a fused deposition of multiple ceramics (FDMC) machine that could deposit up to four different materials in a layer of a multi-material ceramic part. However, an intelligent control software system would be required to deposit more materials. Cho et al. [7] reported a 3D
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
printing (3DP) machine developed in the Massachusetts Institute of Technology (MIT) that contained a multiple nozzle print-head mechanism for fabrication of objects with local composition control (LCC), such as drug delivery devices and gradient index lenses. For biomedical products, a multi-nozzle deposition manufacturing (MDM) system was proposed for direct fabrication of tissue-engineering scaffold with gradient 3D structure [8]. Some companies have also attempted to produce commercial MMLM machines. Examples include FDMM and MP [9] for electroceramic components, and DMDTM [10] and LENSTM [11] for tooling parts and mould cavities. Although these systems were more suitable for relatively simple parts of a limited variety of materials, they were among the pioneering work and could be adapted for further development of the hardware material-depositing mechanism. However, a viable computer software system for MMLM has yet to be developed. To address this software issue, some researchers have proposed CAD representation schemes for heterogeneous objects to provide geometric and material information needed for fabrication of multi-material prototypes [12,13]. Qiu et al. [14] developed a CAD system to plan multitoolpaths for MMLM of relatively simple objects. Nevertheless, it is crucial to develop an integrated software system to process complex objects for practical applications of MMLM. The system should be able to process complex multi-material slice contours, which are subsequently sorted and intraconnected to relate specific slice contours to a particular tool, such that the area is appropriately deposited with the desired material. This facilitates toolpath planning for effective control of tools to deposit the selected materials at the appropriate contours, and to avoid redundant movements and possible collisions of the tools that otherwise move independently to increase efficiency. For multi-axis NC machining which generates a part by removing material from a workpiece secured at a fixture, tool collisions may occur between the cutter edge and the workpiece in the vicinity of the cutter contact points, namely local gouging, or between the cutter holder and the fixture, namely global collision [15]. Many advanced toolpath planning strategies have been developed to help avoid tool collisions in NC machining [15–17]. However, such strategies cannot be used for layered manufacturing (LM), which forms a part by adding materials layer by layer, instead of by removing material from a workpiece as in NC machining. Hence, tool collisions in NC machining are different from those in MMLM. Indeed, tool collisions in MMLM often occur between multiple tool (nozzle) holders when they are controlled to deposit different materials on the related contours concurrently. Therefore, it is necessary to develop a novel toolpath planning strategy for MMLM to detect and avoid possible tool collisions. In the following sections, some related works on toolpath planning for LM are reviewed; the complexity of toolpath planning for MMLM, particularly for avoidance of redundant tool movements and detection of tool collisions, are explained in detail. Subsequently, a practical multi-toolpath planning
439
approach for MMLM and a build-time estimation algorithm for verification of the resulting toolpaths are proposed. Toolpath planning for LM is mainly concerned with (i) contour filling strategy and (ii) tool sequencing strategy. Two common approaches to contour filling are the zigzag toolpaths parallel to a specific direction, and the spiral toolpaths by inward offsetting of contours. Contour filling is similar to pocket machining in several respects, and it has been well studied [18–20]. However, few researchers have attempted to develop tool sequencing strategies. An obstacle is how to sort and group complex slice contours into an appropriate sequence. Tang and Pang [21] introduced a maximum linear intersection algorithm to link a set of contours for minimising non-cutting laser motion of the laminated object modelling (LOM) system. This approach enhanced the efficiency of single-material LOM. However, it is not necessarily suitable for toolpath planning for layered manufacturing of multi-material prototypes. In comparison, toolpath planning for MMLM is much more complicated, in that it is necessary to process material information for sequencing an array of tools to deposit specific materials on the appropriate contours. Without efficient multitoolpaths, redundant back-and-forth movements and possible collisions of tools may occur. Toolpath planning for MMLM is based on 2D slice contours, which are mostly generated by slicing the STL file of an object converted from either a CAD design or MRI/CT/laser scan data. For each tool to deposit a specific material on the related contours, it is necessary to identify and relate specific contours of a slice to a particular tool, such that the area is properly deposited with the desired material, and at the same time to detect and avoid possible collisions of tools. This is a complicated task because high value-added products and biomedical models, such as a human spine in Fig. 1, tend to be complex and their slice contours are random in nature with no explicit topological hierarchy relationship. 1.1. Topological hierarchy of slice contours As shown in Fig. 2a, a slice of an object may contain groups of contours, each of which may have multiple external (solid) and internal (void or cavity) contours. Medical imaging, in particular, often involves complex human structures surrounded by tissues and blood vessels, or test specimens with lots of internal cavities. Due to the complexity and a lack of topological hierarchy information, the slice contours to be made of a particular material cannot be easily identified and linked up accordingly. Consequently, it is extremely difficult to generate multi-toolpaths to control an array of nozzles or tools for efficient fabrication of such layers, because it is necessary to identify and relate contours to a particular nozzle one by one. This is particularly tedious if potential tool collisions are to be detected and avoided accordingly, when the tools deposit materials concurrently to reduce build-time. Assuming that contours do not intersect and that two touching contours are considered as two separate contours, the relationship of any two contours of a slice can be either (1) external-and-external or (2) external-and-internal or (3)
440
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 1. A slice of a human spine.
Fig. 2. (a) A slice of a multi-material machine part with external-and-internal contours. (b) Topological hierarchy based on a parent-and-child relationship. (c) Slice contours may be treated as five contour entities containing seven families.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
internal-and-internal. On the one hand, it can be seen in Fig. 2a that contours C3 and C12 illustrate an external-and-external case as they are independent of each other, while contours C3 and C1 give an external-and-internal case with C1 being an internal contour of C3. On the other hand, internal-and-internal contours should be isolated by at least one external contour in the slice. For example, two internal contours C11 and C2 are isolated by an external contour C10. In Fig. 2b, C3, C6, C7, C9, and C12 are the five outermost contours of the slice. These five contours are at Level 0, meaning that they have no parent and are therefore not contained by any contours. Similarly, a Level 1 contour, such as C1, C5, C8, and C11, has one and only one parent, although a parent may have multiple children. Further, C4 and C10 are Level 2 contours, for they are contained by C5 and C6, and by C11 and C12, respectively. In addition, C2 is a Level 3 contour for it is contained by C10, C11, and C12. This means it is the fourth generation. Hence, according to the parent-and-child relationship that defines the contour containment, the contours can be grouped into five contour entities, namely entity 1 (C3 ! C1), entity 2 (C12 ! C11 ! C10 ! C2), entity 3 (C7), entity 4 (C9 ! C8), and entity 5 (C6 ! C5 ! C4). Isolating the contours of one generation of parent-and-child relationship, the contour entities can be regarded as containing seven contour families, namely family 1 (C3 ! C1), family 2 (C12 ! C11), family 3 (C10 ! C2), family 4 (C7), family 5 (C9 ! C8), family 6 (C6 ! C5), and family 7 (C4). It can be seen that a contour family may contain (1) one or more than one pair of external-internal contours and/or (2) one or more than one external contour. A contour family is indeed a solid area of a specific material property; and contour families of the same material property can be linked together and treated as a single solid area. Based on such a topological hierarchy, groups of related external-and-internal contours can be linked to form contour entities or families, as shown in Fig. 2c, to simplify the complexity of the slice contours. Subsequently, toolpath planning for MMLM becomes straightforward because it is no longer necessary to identify and relate individual contours one by one to a particular tool. In conclusion, an explicit topological hierarchy relationship of slice contours is of vital importance to toolpath planning for MMLM. For this purpose, the author has developed a topological hierarchy-sorting algorithm [22], based on which an intelligent approach to toolpath planning for MMLM of heterogeneous multi-material is developed [23,24]. It includes a sequential multi-toolpath planning algorithm that generates sequential multi-toolpaths without redundant tool movements, and a concurrent multi-toolpath planning algorithm that generates collision-free multi-toolpaths for concurrent deposition of materials. To verify the resulting multi-toolpaths, a build-time estimation algorithm for MMLM processes has been developed. It can estimate the build-time required for fabrication of a multi-material prototype using either sequential or concurrent multi-toolpaths. This is useful for subsequent scheduling and cost estimation of fabrication of multi-material prototypes.
441
2. The MMVP system The above algorithms are integrated with virtual reality (VR) technology to form a multi-material virtual prototyping (MMVP) system for digital fabrication of heterogeneous multimaterial prototypes for advanced product development and biomedical engineering. It consists mainly of a topological hierarchy-sorting algorithm for processing slice contours, a sequential multi-toolpath planning algorithm, a concurrent multi-toolpath planning algorithm, a build-time estimation algorithm, and a virtual reality (VR) simulation system for visualisation and optimisation of MMLM processes. The topological hierarchy-sorting algorithm processes the hierarchy relationship of complex slice contours. It builds a parent-andchild list that defines the containment relationship of the slice contours, and subsequently arranges the contours in an appropriate sequence that facilitates toolpath planning for MMLM by avoiding redundant movements and potential tool collisions. The resulting multi-toolpaths may be verified quantitatively using the build-time estimation algorithm to estimate the required build-time for fabricating a multi-material prototype. Subsequently, a VR simulation system simulates an MMLM process for digital fabrication of multi-material prototypes, with stereoscopic visualisation of the resulting digital multi-material prototypes for quality analysis and process optimisation. Using the MMVP system, the designer can improve the prototype quality before physical fabrication by optimising the process parameters, such as layer thickness, hatch space, and build direction. In this way, the development cycle and cost of a product can be reduced substantially. In the following sections, the machine part in Fig. 2a is used to illustrate the MMVP system. Firstly, planning of multitoolpaths based on topological hierarchy is described; secondly, build-time is estimated to verify the resulting multi-toolpaths and facilitate cost estimation; finally, the machine part is digitally fabricated for stereoscopic visualisation and analysis. 2.1. Topological hierarchy-based multi-toolpath planning for MMLM The layer of the machine part in Fig. 2a is to be made of five materials denoted by m1, m2, m3, m4, and m5, respectively. Based on unsorted contours, there will be redundant tool movements, as shown in Fig. 3. In order to reduce build-time, five nozzles, namely N1, N2, N3, N4, and N5, should be appropriately controlled to deposit m1, m2, m3, m4, and m5 materials, respectively, on the related contours without redundant movements and collisions. To this end, the topological hierarchy-sorting algorithm is used to establish the explicit topological hierarchy relationship that represents the connectivity of the external contours and the internal cavities, as shown in Fig. 4. A parent-and-child list is built to define the containment relationship of the contours. Hence, five entities, namely C3 ! C1, C12 ! C11 ! C10 ! C2, C7, C9 ! C8, and C6 ! C5 ! C4, are formed accordingly. Based on the material properties, these entities are further sorted into seven contour
442
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 3. Redundant tool movements of multi-toolpaths based on unsorted contours in Fig. 2a.
families, namely C3 ! C1 (family 1), C12 ! C11 (family 2), C10 ! C2 (family 3), C7 (family 4), C9 ! C8 (family 5), C6 ! C5 (family 6), and C4 (family 7). Subsequently, a nozzle is assigned to deposit a specific material on each contour family. Hence, multi-toolpaths may be planned simply by treating each contour family as a solid without having to handle all the contours one by one. 2.1.1. Sequential multi-toolpath planning Using the topological hierarchy information above, the seven contour families of the machine part layer are treated as seven independent inputs to generate hatch vectors that define the multi-toolpaths. Thus, seven toolpaths (PC3,1, PC12,11, PC10,2, PC7, PC9,8, PC6,5, and PC4) for the nozzles to deposit the respective materials are generated. In order to avoid redundant tool movements, the seven toolpaths should be further arranged based on the specific geometric and material information of each contour. For instance, PC3,1 and PC7 are grouped into a toolpath-set {S1(PC3,1 ! PC7)}, which is associated with nozzle N1 because contour families 1 and 4 are of material m1. In the same way, PC12,11, PC10,2, PC9,8, and PC6,5 and PC4 are grouped into {S2(PC12,11)}, {S3(PC10,2)}, {S4(PC9,8)}, and {S1(PC6,5 ! PC4)}, which are associated with N2, N3, N4,
and N5, respectively. After associating the toolpath-sets with the related nozzles, the toolpath-sets are arranged in a sequence of S1 ! S2 ! S3 ! S4 ! S5 according to the hierarchy levels of the contours. As a result, practical sequential multi-toolpaths can be conveniently generated, and the movement sequence of the nozzles is given as N1 ! N2 ! N3 ! N4 ! N5. In comparison with six deposition steps of the unsorted toolpaths in Fig. 3, the sequential multi-toolpaths reduce the total number of deposition steps to five by avoiding redundant tool movements, as shown in Fig. 5, and thus the build-time is reduced accordingly. 2.1.2. Concurrent multi-toolpath planning Build-time can be reduced further, particularly for relatively large and complex prototypes, by depositing specific materials on the related contours of different materials simultaneously. However, planning of concurrent multi-toolpaths requires detection of possible tool collisions, which may occur when the tools move at the same time. A concurrent multi-toolpath planning algorithm has been developed to generate collisionfree multi-toolpaths to control the tools that deposit materials concurrently. It uses parametric polygons to construct tool envelopes for contour families of the same material property to
Fig. 4. Topological hierarchy relationship of the machine part slice in Fig. 2a.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
443
Fig. 5. Sequential multi-toolpaths based on topological hierarchy of contours.
simplify detection of tool collisions. The tightness of polygons can be controlled to suit the processing speed and optimality of the resulting concurrent multi-toolpaths. The algorithm is briefly explained below, while its detail has been presented in Ref. [24]. The concurrent multi-toolpath planning algorithm mainly consists of the following steps: (i) Parametric polygons are used to approximate the envelopes in which specific tools can fabricate contour families of the same material properties. (ii) Overlap tests are performed among all pairs of such envelopes in each layer. (iii) Based on the result of the overlap tests, collision-free concurrent multi-toolpaths are planned accordingly. With the parametric polygonal approach, a polygon is constructed around each contour family by selecting and linking a limited number of dominant points along the exterior of the contour family. The polygon is then offset outward a distance of the tool-holder radius; and polygons of the same material property are linked to form an envelope for the tool. A tool envelope is an area in which only the related tool or nozzle is allowed. The number of dominant points defining a tool envelope polygon is proportional to the value of the parameter t, in the range of 0 and 1. To suit the requirements of the processing speed and the optimality of the resulting concurrent multitoolpaths, the value of t can be varied to control the accuracy or tightness of the polygon. On the one extreme when t = 0, the polygon is reduced to only four dominant points, forming an axis-aligned bounding box which is the simplest possible for detection of tool collisions but is less optimal for the toolpaths. On the other extreme for t = 1, the polygon is an exact offset of the exterior of a contour family, resulting in relatively slow collision detections but good optimality of the toolpaths. In general, a value of t = 0.3 is deemed a good compromise.
Having constructed all tool envelopes, potential tool collisions may now be detected by overlap tests, which are achieved by calculating the minimum separation distance between all pairs of envelopes. If the minimum separation distance between two envelopes is larger than zero, the two tools can move concurrently in their respective envelopes without collisions. Hence, the number of overlap queries is given by O((1/2)n(n 1)), where n is the number of toolpathsets in a layer. In comparison, the number of queries without topological hierarchy relationship is O((1/2)m(m 1)), where m is the total number of individual contours of a layer. Since n is generally much smaller than m, the concurrent multi-toolpath planning algorithm simplifies collision detection considerably. The layer of the machine part in Fig. 2a is again used to illustrate this algorithm. Referring to Fig. 6, five envelopes, namely E1, E2, E3, E4, and E5, for five nozzles, namely N1, N2, N3, N4, and N5, are firstly constructed with t = 1, respectively. Secondly, overlap tests are performed among the five tool envelopes. Table 1 shows the result of overlap tests, OT(Ei, Ej), for the pairwise combinations of the five envelopes in Fig. 6. The value of OT(Ei, Ej) indicates whether or not concurrent movements can be planned for nozzles Ni and Nj to deposit materials in the contours bounded by envelopes Ei and Ej, respectively. If OT(Ei, Ej) = non-overlap, the envelope pair
Fig. 6. Tool envelopes of nozzles constructed with parametric polygons in which t = 1.
444
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Table 1 Result of overlap tests for tool envelopes in Fig. 6 Column Envelope, Ei E2
E3
E4
E5
Row E1 Non-overlap Non-overlap Non-overlap Overlap Envelope, Ei E2 Overlap Non-overlap Non-overlap E3 Non-overlap Non-overlap Non-overlap E4 Result of Overlap Tests, OT(Ei, Ej)
(Ei, Ej) do not overlap and concurrent movements can be planned for nozzles Ni and Nj; otherwise the envelope pair overlap and concurrent tool movements cannot be planned. Subsequently, concurrent multi-toolpaths are generated based on the result of envelope overlap tests. The sequence of concurrent nozzle movements is given in Table 2. The tool sequence of the resulting concurrent multi-toolpaths involves two steps:
build-time and some process parameters, including layer thickness, hatch space, table speed, and nozzle speed. Such works laid a good foundation for build-time estimation of FDM, though they could not be applied directly to MMLM processes. It is therefore desirable to develop a build-time estimation algorithm for the multi-material FDM processes. 2.2.1. A mathematical model for build-time estimation of multi-material FDM If a multi-material prototype is to be made of N kinds of materials, N nozzles are used to fabricate it. The time required for a nozzle to complete depositing the related contours of a layer is given by T nozzle ¼ T idle þ T deposition
(1)
where Tidle is the nozzle idle time and Tdeposition is the nozzle deposition time. The nozzle idle time includes nozzle positioning time and cleaning time: T idle ¼ T positioning þ T cleaning
(2)
For the sequential multi-toolpath planning described in Section 2.1.1, there are five deposition steps in the tool sequence:
where Tcleaning is the time for cleaning and purging the nozzle and Tpositioning is the time for positioning a nozzle, which
From the two sets of tool sequences above, it is apparent that concurrent multi-toolpath planning can shorten the tool sequence and hence the build-time significantly. The reduction in buildtime will be more substantial for relatively large and complex parts. This can be quantitatively verified using the build-time estimation algorithm described below to estimate and compare the build-times of sequential and concurrent multi-toolpaths.
includes (i) the time for the nozzle to move from the datum point to the entry point of the first toolpath; (ii) the time for the nozzle to return to the datum point from the exit point of the last toolpath; (iii) the time for the nozzle to move from the exit point of a toolpath to the entry point of the subsequent toolpath. Tpositioning is given by T positioning ¼
2.2. Build-time estimation of multi-material FDM process Fused deposition modelling (FDM) is a common type of LM machines that fabricates a part by depositing semi-solid material onto successive layers according to predefined toolpaths. For single-material FDM, estimation and minimisation of build-time have been well studied. Alexander et al. [25] developed a method to estimate build-time by adjusting and optimizing part geometry orientation. Thrimurthulu et al. [26] developed an approach that determines the optimal part deposition orientation for a good compromise of build-time and part quality. Han et al. [27] analysed the relationship between
Tool movement for envelope, Ei (nozzle, Ni) E1 (N1)
E2 (N2)
E3 (N3)
E4 (N4)
E5 (N5)
Sequence Yes No
Yes No
No Yes
Yes No
No Yes
(3)
where Dp is the total positioning distance of a nozzle and Vidle is the average nozzle speed for idle motion. The nozzle deposition time is given by T deposition ¼ T predelay m þ T scanning
(4)
where Tpredelay is the time to start up the servo-pump before a nozzle begins to deposit material, m the number of connected toolpaths to be deposited by a nozzle and Tscanning is the time for a nozzle to fill up the solid part of the connected toolpaths with the desired material. It is given by T scanning ¼
Table 2 Sequence of concurrent nozzle movements
Dp V idle
Ds Vs
(5)
where Ds is the total scanning distance that depends on the hatch space; the smaller the hatch space, the longer the total scanning distance. Vs is the nozzle deposition speed. With Eqs. (1)–(5), the time taken for a nozzle to complete depositing the related contours of a layer, Tnozzle, can be found for subsequent calculation of the build-time of a complete layer based on the results of multi-toolpath planning, as follows.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
The build-time of a layer is the sum of the platform setup time and the layer fabrication time: T layer ¼ T setup þ T fab
(6)
where Tsetup is the platform setup time and Tfab is the time required for complete deposition by all the related nozzles of the layer
Therefore, the total build-time for fabricating a multimaterial prototype is given by T total ¼
L X i¼1
S X T step
(7)
j¼1
where Tstep is the time for completing a deposition step and S is the total number of deposition steps of the layer. For sequential multi-toolpath planning, only one nozzle is allowed in each deposition step to deposit a material on the related contours. This means that the total number of deposition steps of a layer, S, is equal to the total number of nozzles, n, currently used for the layer. Both n and S should be less than or equal to the total number of nozzles, N, needed for complete fabrication of the prototype. Hence, for sequential multitoolpath planning, the time required for a deposition step is given by T step ¼ T nozzle
(8)
For concurrent multi-toolpath planning, one or more nozzles may be depositing materials on the related sets of contours in a deposition step concurrently; the total number of deposition steps of a layer, S, is generally smaller than or at most equal to the total number of nozzles, n, currently used for the layer. Hence, the time for a deposition step is the longest nozzle time among all the nozzles performing deposition in the step, i.e., T step ¼ maxðT nozzle Þ
(9)
T layer ¼
L X
ðT setup þ T fab Þ
i¼1
L S X X ¼ T step T setup þ i¼1
T fab ¼
445
(10)
j¼1
where L is the total number of layers. Hence for sequential multi-toolpath planning, the total build-time is L S X X T total ¼ T nozzle : j 2 ½1; S ¼ n (11) T setup þ i¼1
j¼1
while for concurrent multi-toolpath planning, the total buildtime is L S X X T total ¼ maxðT nozzle Þ : j 2 ½1; S n (12) T setup þ i¼1
j¼1
where Tnozzle is obtained using Eqs. (1)–(5). 2.2.2. Verification of estimated build-time The mathematical model above has been implemented in the proposed sequential and concurrent multi-toolpath planning algorithms. As practicable MMLM machines are not yet commercially available, the accuracy of build-time estimated above is instead verified using a Stratasys FDM3000 machine, which is a single-material fused deposition modelling machine. It is assumed that the material-depositing mechanism of an MMLM machine consists of an array of nozzles of the same operation characteristics, and each nozzle deposits a specific
Fig. 7. Fabrication of a machine part prototype on FDM3000 machine.
446
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
material similar to that of FDM3000. With this assumption, an experiment of fabricating the machine part in Fig. 2a on FDM3000 was conducted to verify the build-time of sequential multi-toolpath planning. Although the result may not be entirely applicable to concurrent multi-toolpath planning, it does verify the accuracy of the parameters of the mathematical model for estimating concurrent build-time. Fig. 7 shows the machine part prototype being fabricated on FDM3000, together with the base support and the resulting prototype. The operation parameters of FDM3000, such as the platform setup time per layer (Tsetup) and the time required for cleaning and purging the nozzle (Tcleaning) were measured. The machine part in Fig. 2a was of dimensions 100 mm 60 mm 3.5 mm; it was fabricated and the build-time recorded, as follows. InsightTM, the proprietary control software of FDM3000, was used to slice the machine part and generate the base structure. Using the default setting of the layer thickness t and the road width w as 0.178 mm and 0.356 mm, respectively, 28 layers were sliced, with the first 8 layers for the base structure and the remaining 20 layers for the part. The layer thickness and the road width are two major factors that affect the build-time and the quality. In order to obtain a good compromise between the buildtime and the quality, the distance between centre points of two adjacent roads (toolpaths), called hatch space h, should be appropriately controlled. If the distance is too large (small), underfills (overfills) between the two adjacent roads will occur [27,28]. To overcome such problems, a road geometry model [27,28] was adopted to find the necessary negative offset f between the two adjacent roads as shown in Fig. 8. Referring to Fig. 8, the hatch space is expressed as h¼wþtþ f
(13)
where f is expressed as f ¼
ðp 4Þt 4
(14)
Hence, the proposed build-time algorithm adopts Eqs. (13) and (14) to obtain the required hatch space for generation of toolpaths. Based on the default values of the layer thickness t = 0.178 mm and the road width w ¼ 0:356 mm, the hatch space h = 0.496 mm was obtained to generate toolpaths. ABS P400 and P400SR were used, respectively, for the part material and the base material, and two T12 nozzle tips were used as the part and the base. The nozzle speed Vs depends on the nozzle tip size and the type of material used; and Vs is assumed constant in a deposition process [27]. From the specifications of FDM3000 [29], the maximum nozzle speed for T12 was 20.320 mm s1 (0.800 in. s1) and the time to start
Fig. 8. Relation between road width and hatch space.
Table 3 Actual build-time and estimated build-times of the machine part Total build-time of the machine part (min) Actual build-time on FDM3000 Estimated build-time by FDM3000’s control software, InsightTM Estimated sequential build-time by the proposed algorithm Estimated concurrent build-time by the proposed algorithm
Difference (%)
136.27 151.00
– 10.809
136.01
0.191
114.46
16.005
up the servo-pump before a nozzle began to deposit material Tpredelay was approximately 0.1 s. For idle motion, the average nozzle speed Vidle was 20.320 mm s1. The platform setup time per layer Tsetup and the time for cleaning and purging the nozzle Tcleaning were measured to be 1.0 s and 5.0 s, respectively. The machined part was physically fabricated on FDM3000. The total build-time was 178.77 min, of which the base structure took up 42.50 min while the part 136.27 min. The estimated build-time by FDM3000’s control software, InsightTM, was 197.55 min, with 46.55 min of which for the base and 151.00 min for the part. The proposed build-time estimation algorithm was also used to estimate the build-time of the machine part using the operation parameters of FDM3000. For simplicity, the base structure was ignored because its design and the way of fabrication depend largely on the proprietary control software of LM machines as well as on the geometry and orientation of the part. Table 3 shows the actual and estimated build-times of the machine part fabricated on FDM3000 and their percentage differences. It can be seen that while FDM3000’s control software overestimated the build-time of the part by 10.809%, the sequential build-time and the concurrent build-time estimated by the proposed algorithm are 136.01 min and 114.46 min, respectively. The sequential build-time is literally the same as the actual build-time, while the concurrent build-time represents a reduction of 16.005%. This result shows that the proposed build-time estimation algorithm is accurate and reliable; it also confirms the expectation that concurrent multi-toolpath planning helps reduce build-time. 2.3. Digital fabrication of multi-material prototypes The algorithms above are integrated with a VR simulation system to form an MMVP system for toolpath planning, buildtime estimation, stereoscopic simulation, validation of multitoolpaths, and visualisation of MMLM processes. Fig. 9 shows a snapshot of the simulation of concurrent deposition of materials on the layer shown in Fig. 2a. To complement buildtime estimation, VR simulation facilitates verification of multitoolpaths through visualisation. Subsequently, when digital fabrication of a prototype is completed, it can be superimposed on its STL model to identify the resulting dimensional deviations. The overall accuracy of the prototype can be highlighted by showing the maximum and the average cusp
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
447
Fig. 9. Simulation of concurrent nozzle movements for the layer in Fig. 2a.
heights. Hence, the designer can analyse prototype quality iteratively without incurring much cost until a good compromise between quality and build-time is struck with an appropriate set of process parameters, such as build direction, layer thickness, and hatch space. 3. Case studies 3.1. A jewellery pendant Hong Kong is renowned for its jewellery products. To survive competition in global markets, most manufacturers
have adopted advanced technologies, including CAD/CAM and LM technologies. Hence, the MMVP system will be useful for the jewellery industry to help reduce product development time and cost. Therefore, a pendant, as shown in Fig. 10, is used to demonstrate digital fabrication, using concurrent multitoolpaths, of multi-material prototypes for development of jewellery products. The pendant STL model is sliced into 25 layers with layer thickness 0.178 mm. The first layer of 25 contours is to be made of 5 kinds of materials. Its topological hierarchy relationship in Fig. 11 is established, in which the contours are grouped into 18 contour families, and 18 toolpaths (PC15,8,9,10,11,12,13,14, PC7,
Fig. 10. A layer of contours of a multi-material pendant.
448
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 11. Topological hierarchy relationship of contours of the pendant layer.
PC1, PC2, PC3, PC4, PC5, PC6, PC25,20,21,22,23,24, PC16, PC17, PC18, PC19, PC35,30,31,32,33,34, PC26, PC27, PC28, and PC29) are generated for these contour families accordingly with a hatch space of 0.496 mm. Based on the material information, the toolpaths with the same material property are grouped into five toolpath-sets, namely from S1 to S5, which are associated with five nozzles from N1 to N5, respectively. Subsequently, five work envelopes from E1 to E5 for each of these nozzles are constructed to facilitate planning of concurrent multi-toolpaths. Potential collisions of nozzles can be detected by overlap tests of the envelopes E1, E2, E3, E4, and E5. According to the overlap test result, the system can efficiently generate concurrent multitoolpaths. Three steps are generated and the sequences of concurrent nozzle movements are as follows:
multi-toolpath planning becomes substantial when the part is complex, in comparison with the machine part in Fig. 2a. The MMVP system can digitally fabricate the pendant prototype for quality analysis through visualisation. Fig. 12 shows the digital fabrication process of a few pendant layers, while Fig. 13 highlights the excessive materials of the resulting prototype when it is superimposed on the STL model. Its average and maximum cusp heights, which are 0.085 mm and 0.178 mm, respectively, are calculated and highlighted. Therefore, using the MMVP system, the designer can conveniently perform design iterations and quality analysis of the resulting prototypes. Thus, an optimal combination of process parameters, such as layer thickness, build direction, and hatch space can be obtained for cost-effective fabrication of physical prototypes.
With this, the MMVP system can estimate the build-time to verify the resulting multi-toolpaths. Table 4 shows the estimated build-times for the sequential multi-toolpaths and the concurrent multi-toolpaths, which would be useful for subsequent process planning and cost estimation. It can be seen that the estimated build-time by concurrent multi-toolpath planning is 31.342% shorter than by sequential multi-toolpath planning. This shows that build-time reduction by concurrent
3.2. A pelvis model with two inserted pins
Table 4 Estimated sequential and concurrent build-times for the pendant
Sequential multi-toolpaths Concurrent multi-toolpaths
Estimated build-time (min)
Difference (%)
36.66 25.17
– 31.342
LM technology has now been widely adopted to produce medical prototypes for studying human bone or organ structures and surgical planning [30,31]. As multi-material prototypes are more revealing for identifying complex organ structures and thus more helpful for managing delicate surgical conditions, the MMVP system is deemed useful for biomedical applications. Surgeons can study and plan a surgical operation conveniently in the VR environment provided by the system; they can also design a prosthetic component of appropriate size, shape, and orientation that would fit the patient best before physical fabrication takes place. A pelvis model with two simple pins inserted for fixation, as shown in Fig. 14, is therefore used to illustrate the usefulness of the MMVP system for biomedical engineering.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 12. Digital fabrication process of a multi-material prototype of the pendant.
The pelvis is sliced into 721 layers with a layer thickness of 0.178 mm. Fig. 14 shows a layer containing 21 contours to be made of five kinds of materials, namely m1, m2, m3, m4, and m5. Five nozzles, N1, N2, N3, N4, and N5, will deposit specific materials, m1, m2, m3, m4, and m5, respectively. Fig. 15 shows the topological hierarchy relationship of the layer contours. According to the contour families, eight toolpaths namely, PC2,21,15,16,17,3,5, PC1, PC6,7,14, PC4, PC11, PC8,18,19, PC9,10,12,20, and PC13 are generated by hatching the contours with hatch space 0.496 mm. Subsequently, the
449
Fig. 13. Superimposition of pendant prototype on its STL model to highlight dimensional deviations.
toolpaths with the same material properties are grouped together to form five toolpath-sets, namely S1, S2, S3, S4, and S5, respectively. Five work envelopes from E1 to E5 for nozzles from N1 to N5 are then constructed to facilitate planning of concurrent multi-toolpaths. Potential nozzle collisions can be detected by overlap test of the envelopes E1, E2, E3, E4, and E5. According to the overlap test result, the system generates concurrent multi-toolpaths, which are arranged in four steps with the sequence of nozzle movements as follows:
Fig. 14. A layer of contours of a multi-material pelvis with two pins.
450
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 15. Topological hierarchy relationship of contours.
The build-time of the pelvis prototype based on concurrent multi-toolpath planning is estimated to be 1763.87 min. In comparison with the estimated sequential build-time of 2475.72 min, as shown in Table 5, a reduction of 28.753% is achieved by concurrent multi-toolpath planning. This shows that build-time reduction by concurrent multi-toolpath planning for relatively large and complex parts is substantial. The estimated build-time facilitates cost estimation of customised prosthetic components as well as scheduling of preoperative tasks. Furthermore, the MMVP system provides
stereoscopic visualisation of the resulting prototypes to help surgeons design prostheses with best fit possible for the patient. The pelvis prototype can be superimposed on its STL model, as shown in Fig. 16, to highlight dimensional deviations. In addition, the system can calculate the average and maximum cusp heights of the prototype to facilitate evaluation of the overall accuracy of the prototype. Thus, the surgeons can repeat digital fabrication of the pelvis prototype until an optimal tradeoff between the quality and the production cost of the pelvis prototype can be achieved.
Fig. 16. Superimposition of pelvis prototype on its STL model to highlight dimensional deviations.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452 Table 5 Estimated sequential and concurrent build-times for the pelvis prototype
Sequential multi-toolpaths Concurrent multi-toolpaths
Estimated build-time (min)
Difference (%)
2475.72 1763.87
– 28.753
3.3. Limitation of the proposed MMVP system Case studies show that the MMVP system can generate sequential and concurrent multi-toolpaths for effective control of MMLM processes; reduction in build-time by concurrent multi-toolpath planning is more substantial for relatively large and complex prototypes; the estimated build-time is useful for validation of multi-toolpaths, as well as for scheduling and cost estimation of prototype fabrication. Nevertheless, the resulting concurrent multi-toolpaths may not be of absolute optimality, because the tools in overlapping envelopes are not allowed to move concurrently. In the next step of our development work, the concurrent multi-toolpath planning algorithm would be enhanced to decompose overlapping envelopes into smaller ones for detailed detection of possible tool collisions in order to increase the likelihood of concurrent nozzle movements. Indeed, the optimality of concurrent multi-toolpaths may be further enhanced by incorporating more vigorous optimisation rules, such as critical path, to find the shortest nozzle movement distance possible while allowing as many nozzles to deposit concurrently as possible. Obviously, this requires much computing effort and is not an easy task, which would be undertaken in our further work. 4. Conclusion To address the key software issue of MMLM, an MMVP system has been developed for digital fabrication of heterogeneous multi-material prototypes. The MMVP system integrates VR and MMLM technologies for advanced product development and biomedical applications. It consists of a suite of algorithms for multi-toolpath planning, build-time estimation, simulation, optimisation, and visualisation of MMLM processes. Based on the topological hierarchy relationship of slice contours, the sequential multi-toolpath planning algorithm can arrange the contours in an appropriate sequence to generate sequential multi-toolpaths without redundant movements. To save build-time further, the concurrent multi-toolpath planning algorithm can generate concurrent multi-toolpaths. To validate the resulting sequential and concurrent toolpaths, a build-time estimation algorithm for MMLM process is developed. Designers can use the MMVP system to iterate digital fabrication of prototypes conveniently until an optimal trade-off between the quality and the cost is achieved. The MMVP system may be adapted as an intelligent control software system for practical and viable MMLM machines.
451
Acknowledgements The authors would like to acknowledge the Research Grant Council of the Hong Kong SAR Government and the CRCG of the University of Hong Kong for their financial support for this project. Thanks are also due to Dr. Y.H. Chen of the Department of Mechanical Engineering at the University of Hong Kong for facilitating the fabrication of the machine part on the Stratasys FDM 3000 machine. References [1] K. Samanta, B. Koc, Feature-based and material blending for free-form heterogeneous object modeling, Computer-Aided Design 37 (3) (2005) 287–305. [2] S.F. Yang, G.J.R. Evans, A multi-component powder dispensing system for three dimensional functional gradients, Materials Science & Engineering A 379 (1–2) (2004) 351–359. [3] A. Safari III, J. Cesasrano, P.G. Clem, B. Bender, Fabrication of advanced functional electroceramic components by layered manufacturing (LM) methods, in: Proceedings of the 13th IEEE International Symposium on Applications of Ferroelectronics, Nara, Japan, (2002), pp. 1–6. [4] E. Sachlos, N. Reis, C. Ainsley, B. Derby, J.T. Czernuszka, Novel collagen scaffolds with predefined internal morphology made by solid freeform fabrication, Biomaterials 24 (8) (2003) 1487–1497. [5] K. Lappo, B. Jackson, K. Wood, D. Bourell, J.J. Beaman, Discrete multiple material selective laser sintering (M2SLS): experimental study of part processing, in: D.L. Bourell, et al. (Eds.), Solid Freeform Fabrication Symposium, Austin, Texas, The University of Texas, Austin, TX, 2003, pp. 109–119. [6] M.A. Jafari, W. Han, F. Mohammadi, A. Safari, S.C. Danforth, N. Langrana, A novel system for fused deposition of advanced multiple ceramics, Rapid Prototyping Journal 6 (3) (2000) 161–175. [7] W.J. Cho, E.M. Sachs, N.M. Patrikalakis, D.E. Troxel, A dithering algorithm for local composition control with three-dimensional printing, Computer-Aided Design 35 (9) (2003) 851–867. [8] Y.N. Yan, Z. Xiong, Y.Y. Hu, S.G. Wang, R.J. Zhang, C. Zhang, Layered manufacturing of tissue engineering scaffolds via multi-nozzle deposition, Materials Letters 57 (18) (2003) 2623–2628. [9] R.E. Bremnan, S. Turcu, A. Hall, N.M. Hagh, A. Safari, Fabrication of electroceramic components by layered manufacturing (LM), Ferroelectrics 293 (1) (2003) 3–17. [10] DMDTM, http://www.pom.net/, 2004. [11] LENSTM, http://mfgshop.sandia.gov/1400_ext/1400_ext_LENS.htm, 2004. [12] V. Kumar, S. Rajagopalan, M. Cutkosky, D. Dutta, Representation and processing heterogeneous objects for solid freeform fabrication, in: Sixth IFIP WG 5.2 International Workshop on Geometric Modelling: Fundamentals and Applications, Tokyo, Japan, The University of Tokyo, 1998. [13] S.M. Morvan, G.M. Fadel, Heterogeneous solids: possible representation schemes, in: D.L. Bourell, et al. (Eds.), Solid Freeform Fabrication Symposium, Austin, Texas, The University of Texa, Austin, Texas, 1999, pp. 187–198. [14] D. Qiu, N.A. Langrana, S.C. Danforth, A. Safari, M. Fafari, Intelligent toolpath for extrusion-based LM process, Rapid Prototyping Journal 7 (1) (2001) 18–24. [15] S. Ding, M.A. Mannan, A.N. Poo, Oriented bounding box and octree based global interference detection in 5-axis machining of free-form surfaces, Computer-Aided Design 36 (13) (2004) 1281–1294. [16] O. Ilushin, G. Elber, D. Halperin, R. Wein, M.-S. Kin, Precise global collision detection in multi-axis NC-machining, Computer-Aided Design 37 (9) (2005) 909–920. [17] E.L.J. Bohez, N.T.H. Minh, B. Kiatsrithanakorn, P. Natasukon, R.-Y. Huang, T. Son, The stencil buffer sweep plane algorithm for 5-axis CNC tool path verification, Computer-Aided Design 35 (12) (2003) 1129– 1142.
452
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
[18] Y. Yang, J.Y.H. Fuh, H.T. Loh, An efficient scanning pattern for layered manufacturing processes, in: Proceedings of the 2001 IEEE International Conference on Robotic & Automation, Seoul, Korea, (2001), pp. 1340– 1345. [19] J.H. Kao, F.B. Prinz, Optimal motion planning for deposition in layered manufacturing, in: Proceedings of ASME Design Engineering Technical Conferences, Atlanta, GA, (1998), pp. 1–10. [20] J.S. Liang, A.C. Lin, System development of adaptive spraying paths for rapid prototyping, Rapid Prototyping Journal 9 (5) (2003) 289–300. [21] K. Tang, A. Pang, Optimal connection of loops in laminated object manufacturing, Computer-Aided Design 35 (11) (2003) 1011–1022. [22] S.H. Choi, K.T. Kwok, A topological hierarchy-sorting algorithm for slice contours, Rapid Prototyping Journal 10 (2) (2004) 98–113. [23] S.H. Choi, H.H. Cheung, A multi-material virtual prototyping system, Computer-Aided Design 37 (1) (2005) 123–136. [24] S.H. Choi, H.H. Cheung, A topological hierarchy-based approach to toolpath planning for multi-material layered manufacturing, ComputerAided Design 38 (2) (2006) 143–156. [25] P. Alexander, S. Allen, D. Dutta, Part orientation and build cost determination in layered manufacturing, Computer-Aided Design 30 (5) (1998) 343–356. [26] K. Thrimurthulu, P.M. Pandey, N.V. Reddy, Optimum part deposition orientation in fused modeling, Machine Tools & Manufacture 44 (6) (2003) 585–594. [27] W.B. Han, M.A. Jafari, K. Seyed, Process speeding up via deposition planning in fused deposition-based layered manufacturing processes, Rapid Prototyping Journal 9 (4) (2003) 212–218.
[28] W.B. Han, M.A. Jafari, S.C. Danforth, A. Safari, Tool path-based deposition planning in fused deposition processes, Transactions of the ASME, Journal of Manufacturing Science and Engineering 124 (2) (2002) 462–472. [29] FDM, FDM1 System Documentation, Stratasys Inc., 2001. [30] P. Peters, F. Langlotz, L.-P. Nolte, Computer assisted screw insertion into real 3D rapid prototyping pelvis models, Clinical Biomechanics 17 (5) (2002) 376–382. [31] E.D. Momi, E. Pavan, B. Motyl, C. Bandera, C. Frigo, Hip joint anatomy virtual and stereolithographic reconstruction for preoperative planning of total hip replacement, International Congress Series 1281 (2005) 708–712. Dr. S.H. Choi is associate professor in the IMSE Department at the University of Hong Kong. He obtained both his B.Sc. and Ph.D. degrees at the University of Birmingham. He worked in computer industry as CADCAM consultant before joining the University of Hong Kong. His current research interests include CADCAM, advanced manufacturing systems and virtual prototyping technology.
Mr. H.H. Cheung gained his B.Eng. degree from the IMSE Department at the University of Hong Kong. He continued his postgraduate research study in the Department, and his research interest is in virtual prototyping technology.
Multi-material virtual prototyping for product development and biomedical engineering S.H. Choi *, H.H. Cheung Department of Industrial and Manufacturing Systems Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong, China Received 27 January 2005; accepted 8 September 2006 Available online 18 October 2006
Abstract This paper describes the development of a multi-material virtual prototyping (MMVP) system, which integrates virtual reality (VR) and layered manufacturing (LM) for digital fabrication of multi-material prototypes to facilitate advanced product development and biomedical engineering. The MMVP system incorporates a topological hierarchy-sorting algorithm, two algorithms for sequential and concurrent multi-toolpath planning, a build-time estimation algorithm, and a virtual prototyping system. Based on the topological hierarchy of slice contours, the sequential multitoolpath planning algorithm generates sequential toolpaths that avoid redundant tool movements, while the concurrent multi-toolpath planning algorithm generates collision-free concurrent toolpaths to further reduce the build-time. To verify the resulting toolpaths and to facilitate estimation of fabrication cost, a mathematical model is incorporated for build-time estimation of multi-material fused deposition processes. The MMVP system is useful for planning, stereoscopic simulation, build-time estimation and validation of multi-toolpaths, and subsequent digital fabrication and analysis of multi-material prototypes. # 2006 Elsevier B.V. All rights reserved. Keywords: Multi-material prototyping; toolpath planning; Build-time
1. Introduction Multi-material prototypes may be made of materials that change gradually from one type to another, or of a collection of heterogeneous materials. In contrast to single-material prototypes, multi-material prototypes can differentiate clearly one part from others, or tissues from blood vessels of a human organ; and they perform better in rigorous environments [1]. As high value-added products now often involve complex designs while surgical operations become more complicated, there are strong demands for multi-material prototypes, especially for advanced product development and biomedical applications. It is therefore desirable and timely to develop multi-material layered manufacturing (MMLM) technology for fabrication of multi-material prototypes [2]. MMLM technology provides a relatively convenient way to fabricate multi-material parts, such as electronic products, advanced communication components, drug delivery devices,
* Corresponding author. Tel.: +852 2859 7054; fax: +852 2858 6535. E-mail address: [email protected] (S.H. Choi). 0166-3615/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.compind.2006.09.002
and innovative cellular and cell-containing tissue scaffolds [3,4]. Although its benefits have been widely recognised, practical MMLM machines have yet to be developed, particularly for fabrication of relatively complex objects. In general, an MMLM machine may consist mainly of two sub-systems namely: (i) a hardware material-depositing mechanism and (ii) a computer software system for processing complex slice contours to plan multi-toolpaths that control the material deposition mechanism. Some researchers have developed experimental MMLM machines for relatively simple objects. Lappo et al. [5] developed a discrete multiple material selective laser sintering (M2SLS) machine for fabrication of prototypes with discrete multiple materials. Yang and Evans [2] proposed a multicomponent powder-dispensing device for selective laser sintering (SLS) to make parts of 3D functional gradients. Jafari et al. [6] extended the fused deposition modelling (FDM) hardware mechanism to form a fused deposition of multiple ceramics (FDMC) machine that could deposit up to four different materials in a layer of a multi-material ceramic part. However, an intelligent control software system would be required to deposit more materials. Cho et al. [7] reported a 3D
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
printing (3DP) machine developed in the Massachusetts Institute of Technology (MIT) that contained a multiple nozzle print-head mechanism for fabrication of objects with local composition control (LCC), such as drug delivery devices and gradient index lenses. For biomedical products, a multi-nozzle deposition manufacturing (MDM) system was proposed for direct fabrication of tissue-engineering scaffold with gradient 3D structure [8]. Some companies have also attempted to produce commercial MMLM machines. Examples include FDMM and MP [9] for electroceramic components, and DMDTM [10] and LENSTM [11] for tooling parts and mould cavities. Although these systems were more suitable for relatively simple parts of a limited variety of materials, they were among the pioneering work and could be adapted for further development of the hardware material-depositing mechanism. However, a viable computer software system for MMLM has yet to be developed. To address this software issue, some researchers have proposed CAD representation schemes for heterogeneous objects to provide geometric and material information needed for fabrication of multi-material prototypes [12,13]. Qiu et al. [14] developed a CAD system to plan multitoolpaths for MMLM of relatively simple objects. Nevertheless, it is crucial to develop an integrated software system to process complex objects for practical applications of MMLM. The system should be able to process complex multi-material slice contours, which are subsequently sorted and intraconnected to relate specific slice contours to a particular tool, such that the area is appropriately deposited with the desired material. This facilitates toolpath planning for effective control of tools to deposit the selected materials at the appropriate contours, and to avoid redundant movements and possible collisions of the tools that otherwise move independently to increase efficiency. For multi-axis NC machining which generates a part by removing material from a workpiece secured at a fixture, tool collisions may occur between the cutter edge and the workpiece in the vicinity of the cutter contact points, namely local gouging, or between the cutter holder and the fixture, namely global collision [15]. Many advanced toolpath planning strategies have been developed to help avoid tool collisions in NC machining [15–17]. However, such strategies cannot be used for layered manufacturing (LM), which forms a part by adding materials layer by layer, instead of by removing material from a workpiece as in NC machining. Hence, tool collisions in NC machining are different from those in MMLM. Indeed, tool collisions in MMLM often occur between multiple tool (nozzle) holders when they are controlled to deposit different materials on the related contours concurrently. Therefore, it is necessary to develop a novel toolpath planning strategy for MMLM to detect and avoid possible tool collisions. In the following sections, some related works on toolpath planning for LM are reviewed; the complexity of toolpath planning for MMLM, particularly for avoidance of redundant tool movements and detection of tool collisions, are explained in detail. Subsequently, a practical multi-toolpath planning
439
approach for MMLM and a build-time estimation algorithm for verification of the resulting toolpaths are proposed. Toolpath planning for LM is mainly concerned with (i) contour filling strategy and (ii) tool sequencing strategy. Two common approaches to contour filling are the zigzag toolpaths parallel to a specific direction, and the spiral toolpaths by inward offsetting of contours. Contour filling is similar to pocket machining in several respects, and it has been well studied [18–20]. However, few researchers have attempted to develop tool sequencing strategies. An obstacle is how to sort and group complex slice contours into an appropriate sequence. Tang and Pang [21] introduced a maximum linear intersection algorithm to link a set of contours for minimising non-cutting laser motion of the laminated object modelling (LOM) system. This approach enhanced the efficiency of single-material LOM. However, it is not necessarily suitable for toolpath planning for layered manufacturing of multi-material prototypes. In comparison, toolpath planning for MMLM is much more complicated, in that it is necessary to process material information for sequencing an array of tools to deposit specific materials on the appropriate contours. Without efficient multitoolpaths, redundant back-and-forth movements and possible collisions of tools may occur. Toolpath planning for MMLM is based on 2D slice contours, which are mostly generated by slicing the STL file of an object converted from either a CAD design or MRI/CT/laser scan data. For each tool to deposit a specific material on the related contours, it is necessary to identify and relate specific contours of a slice to a particular tool, such that the area is properly deposited with the desired material, and at the same time to detect and avoid possible collisions of tools. This is a complicated task because high value-added products and biomedical models, such as a human spine in Fig. 1, tend to be complex and their slice contours are random in nature with no explicit topological hierarchy relationship. 1.1. Topological hierarchy of slice contours As shown in Fig. 2a, a slice of an object may contain groups of contours, each of which may have multiple external (solid) and internal (void or cavity) contours. Medical imaging, in particular, often involves complex human structures surrounded by tissues and blood vessels, or test specimens with lots of internal cavities. Due to the complexity and a lack of topological hierarchy information, the slice contours to be made of a particular material cannot be easily identified and linked up accordingly. Consequently, it is extremely difficult to generate multi-toolpaths to control an array of nozzles or tools for efficient fabrication of such layers, because it is necessary to identify and relate contours to a particular nozzle one by one. This is particularly tedious if potential tool collisions are to be detected and avoided accordingly, when the tools deposit materials concurrently to reduce build-time. Assuming that contours do not intersect and that two touching contours are considered as two separate contours, the relationship of any two contours of a slice can be either (1) external-and-external or (2) external-and-internal or (3)
440
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 1. A slice of a human spine.
Fig. 2. (a) A slice of a multi-material machine part with external-and-internal contours. (b) Topological hierarchy based on a parent-and-child relationship. (c) Slice contours may be treated as five contour entities containing seven families.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
internal-and-internal. On the one hand, it can be seen in Fig. 2a that contours C3 and C12 illustrate an external-and-external case as they are independent of each other, while contours C3 and C1 give an external-and-internal case with C1 being an internal contour of C3. On the other hand, internal-and-internal contours should be isolated by at least one external contour in the slice. For example, two internal contours C11 and C2 are isolated by an external contour C10. In Fig. 2b, C3, C6, C7, C9, and C12 are the five outermost contours of the slice. These five contours are at Level 0, meaning that they have no parent and are therefore not contained by any contours. Similarly, a Level 1 contour, such as C1, C5, C8, and C11, has one and only one parent, although a parent may have multiple children. Further, C4 and C10 are Level 2 contours, for they are contained by C5 and C6, and by C11 and C12, respectively. In addition, C2 is a Level 3 contour for it is contained by C10, C11, and C12. This means it is the fourth generation. Hence, according to the parent-and-child relationship that defines the contour containment, the contours can be grouped into five contour entities, namely entity 1 (C3 ! C1), entity 2 (C12 ! C11 ! C10 ! C2), entity 3 (C7), entity 4 (C9 ! C8), and entity 5 (C6 ! C5 ! C4). Isolating the contours of one generation of parent-and-child relationship, the contour entities can be regarded as containing seven contour families, namely family 1 (C3 ! C1), family 2 (C12 ! C11), family 3 (C10 ! C2), family 4 (C7), family 5 (C9 ! C8), family 6 (C6 ! C5), and family 7 (C4). It can be seen that a contour family may contain (1) one or more than one pair of external-internal contours and/or (2) one or more than one external contour. A contour family is indeed a solid area of a specific material property; and contour families of the same material property can be linked together and treated as a single solid area. Based on such a topological hierarchy, groups of related external-and-internal contours can be linked to form contour entities or families, as shown in Fig. 2c, to simplify the complexity of the slice contours. Subsequently, toolpath planning for MMLM becomes straightforward because it is no longer necessary to identify and relate individual contours one by one to a particular tool. In conclusion, an explicit topological hierarchy relationship of slice contours is of vital importance to toolpath planning for MMLM. For this purpose, the author has developed a topological hierarchy-sorting algorithm [22], based on which an intelligent approach to toolpath planning for MMLM of heterogeneous multi-material is developed [23,24]. It includes a sequential multi-toolpath planning algorithm that generates sequential multi-toolpaths without redundant tool movements, and a concurrent multi-toolpath planning algorithm that generates collision-free multi-toolpaths for concurrent deposition of materials. To verify the resulting multi-toolpaths, a build-time estimation algorithm for MMLM processes has been developed. It can estimate the build-time required for fabrication of a multi-material prototype using either sequential or concurrent multi-toolpaths. This is useful for subsequent scheduling and cost estimation of fabrication of multi-material prototypes.
441
2. The MMVP system The above algorithms are integrated with virtual reality (VR) technology to form a multi-material virtual prototyping (MMVP) system for digital fabrication of heterogeneous multimaterial prototypes for advanced product development and biomedical engineering. It consists mainly of a topological hierarchy-sorting algorithm for processing slice contours, a sequential multi-toolpath planning algorithm, a concurrent multi-toolpath planning algorithm, a build-time estimation algorithm, and a virtual reality (VR) simulation system for visualisation and optimisation of MMLM processes. The topological hierarchy-sorting algorithm processes the hierarchy relationship of complex slice contours. It builds a parent-andchild list that defines the containment relationship of the slice contours, and subsequently arranges the contours in an appropriate sequence that facilitates toolpath planning for MMLM by avoiding redundant movements and potential tool collisions. The resulting multi-toolpaths may be verified quantitatively using the build-time estimation algorithm to estimate the required build-time for fabricating a multi-material prototype. Subsequently, a VR simulation system simulates an MMLM process for digital fabrication of multi-material prototypes, with stereoscopic visualisation of the resulting digital multi-material prototypes for quality analysis and process optimisation. Using the MMVP system, the designer can improve the prototype quality before physical fabrication by optimising the process parameters, such as layer thickness, hatch space, and build direction. In this way, the development cycle and cost of a product can be reduced substantially. In the following sections, the machine part in Fig. 2a is used to illustrate the MMVP system. Firstly, planning of multitoolpaths based on topological hierarchy is described; secondly, build-time is estimated to verify the resulting multi-toolpaths and facilitate cost estimation; finally, the machine part is digitally fabricated for stereoscopic visualisation and analysis. 2.1. Topological hierarchy-based multi-toolpath planning for MMLM The layer of the machine part in Fig. 2a is to be made of five materials denoted by m1, m2, m3, m4, and m5, respectively. Based on unsorted contours, there will be redundant tool movements, as shown in Fig. 3. In order to reduce build-time, five nozzles, namely N1, N2, N3, N4, and N5, should be appropriately controlled to deposit m1, m2, m3, m4, and m5 materials, respectively, on the related contours without redundant movements and collisions. To this end, the topological hierarchy-sorting algorithm is used to establish the explicit topological hierarchy relationship that represents the connectivity of the external contours and the internal cavities, as shown in Fig. 4. A parent-and-child list is built to define the containment relationship of the contours. Hence, five entities, namely C3 ! C1, C12 ! C11 ! C10 ! C2, C7, C9 ! C8, and C6 ! C5 ! C4, are formed accordingly. Based on the material properties, these entities are further sorted into seven contour
442
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 3. Redundant tool movements of multi-toolpaths based on unsorted contours in Fig. 2a.
families, namely C3 ! C1 (family 1), C12 ! C11 (family 2), C10 ! C2 (family 3), C7 (family 4), C9 ! C8 (family 5), C6 ! C5 (family 6), and C4 (family 7). Subsequently, a nozzle is assigned to deposit a specific material on each contour family. Hence, multi-toolpaths may be planned simply by treating each contour family as a solid without having to handle all the contours one by one. 2.1.1. Sequential multi-toolpath planning Using the topological hierarchy information above, the seven contour families of the machine part layer are treated as seven independent inputs to generate hatch vectors that define the multi-toolpaths. Thus, seven toolpaths (PC3,1, PC12,11, PC10,2, PC7, PC9,8, PC6,5, and PC4) for the nozzles to deposit the respective materials are generated. In order to avoid redundant tool movements, the seven toolpaths should be further arranged based on the specific geometric and material information of each contour. For instance, PC3,1 and PC7 are grouped into a toolpath-set {S1(PC3,1 ! PC7)}, which is associated with nozzle N1 because contour families 1 and 4 are of material m1. In the same way, PC12,11, PC10,2, PC9,8, and PC6,5 and PC4 are grouped into {S2(PC12,11)}, {S3(PC10,2)}, {S4(PC9,8)}, and {S1(PC6,5 ! PC4)}, which are associated with N2, N3, N4,
and N5, respectively. After associating the toolpath-sets with the related nozzles, the toolpath-sets are arranged in a sequence of S1 ! S2 ! S3 ! S4 ! S5 according to the hierarchy levels of the contours. As a result, practical sequential multi-toolpaths can be conveniently generated, and the movement sequence of the nozzles is given as N1 ! N2 ! N3 ! N4 ! N5. In comparison with six deposition steps of the unsorted toolpaths in Fig. 3, the sequential multi-toolpaths reduce the total number of deposition steps to five by avoiding redundant tool movements, as shown in Fig. 5, and thus the build-time is reduced accordingly. 2.1.2. Concurrent multi-toolpath planning Build-time can be reduced further, particularly for relatively large and complex prototypes, by depositing specific materials on the related contours of different materials simultaneously. However, planning of concurrent multi-toolpaths requires detection of possible tool collisions, which may occur when the tools move at the same time. A concurrent multi-toolpath planning algorithm has been developed to generate collisionfree multi-toolpaths to control the tools that deposit materials concurrently. It uses parametric polygons to construct tool envelopes for contour families of the same material property to
Fig. 4. Topological hierarchy relationship of the machine part slice in Fig. 2a.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
443
Fig. 5. Sequential multi-toolpaths based on topological hierarchy of contours.
simplify detection of tool collisions. The tightness of polygons can be controlled to suit the processing speed and optimality of the resulting concurrent multi-toolpaths. The algorithm is briefly explained below, while its detail has been presented in Ref. [24]. The concurrent multi-toolpath planning algorithm mainly consists of the following steps: (i) Parametric polygons are used to approximate the envelopes in which specific tools can fabricate contour families of the same material properties. (ii) Overlap tests are performed among all pairs of such envelopes in each layer. (iii) Based on the result of the overlap tests, collision-free concurrent multi-toolpaths are planned accordingly. With the parametric polygonal approach, a polygon is constructed around each contour family by selecting and linking a limited number of dominant points along the exterior of the contour family. The polygon is then offset outward a distance of the tool-holder radius; and polygons of the same material property are linked to form an envelope for the tool. A tool envelope is an area in which only the related tool or nozzle is allowed. The number of dominant points defining a tool envelope polygon is proportional to the value of the parameter t, in the range of 0 and 1. To suit the requirements of the processing speed and the optimality of the resulting concurrent multitoolpaths, the value of t can be varied to control the accuracy or tightness of the polygon. On the one extreme when t = 0, the polygon is reduced to only four dominant points, forming an axis-aligned bounding box which is the simplest possible for detection of tool collisions but is less optimal for the toolpaths. On the other extreme for t = 1, the polygon is an exact offset of the exterior of a contour family, resulting in relatively slow collision detections but good optimality of the toolpaths. In general, a value of t = 0.3 is deemed a good compromise.
Having constructed all tool envelopes, potential tool collisions may now be detected by overlap tests, which are achieved by calculating the minimum separation distance between all pairs of envelopes. If the minimum separation distance between two envelopes is larger than zero, the two tools can move concurrently in their respective envelopes without collisions. Hence, the number of overlap queries is given by O((1/2)n(n 1)), where n is the number of toolpathsets in a layer. In comparison, the number of queries without topological hierarchy relationship is O((1/2)m(m 1)), where m is the total number of individual contours of a layer. Since n is generally much smaller than m, the concurrent multi-toolpath planning algorithm simplifies collision detection considerably. The layer of the machine part in Fig. 2a is again used to illustrate this algorithm. Referring to Fig. 6, five envelopes, namely E1, E2, E3, E4, and E5, for five nozzles, namely N1, N2, N3, N4, and N5, are firstly constructed with t = 1, respectively. Secondly, overlap tests are performed among the five tool envelopes. Table 1 shows the result of overlap tests, OT(Ei, Ej), for the pairwise combinations of the five envelopes in Fig. 6. The value of OT(Ei, Ej) indicates whether or not concurrent movements can be planned for nozzles Ni and Nj to deposit materials in the contours bounded by envelopes Ei and Ej, respectively. If OT(Ei, Ej) = non-overlap, the envelope pair
Fig. 6. Tool envelopes of nozzles constructed with parametric polygons in which t = 1.
444
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Table 1 Result of overlap tests for tool envelopes in Fig. 6 Column Envelope, Ei E2
E3
E4
E5
Row E1 Non-overlap Non-overlap Non-overlap Overlap Envelope, Ei E2 Overlap Non-overlap Non-overlap E3 Non-overlap Non-overlap Non-overlap E4 Result of Overlap Tests, OT(Ei, Ej)
(Ei, Ej) do not overlap and concurrent movements can be planned for nozzles Ni and Nj; otherwise the envelope pair overlap and concurrent tool movements cannot be planned. Subsequently, concurrent multi-toolpaths are generated based on the result of envelope overlap tests. The sequence of concurrent nozzle movements is given in Table 2. The tool sequence of the resulting concurrent multi-toolpaths involves two steps:
build-time and some process parameters, including layer thickness, hatch space, table speed, and nozzle speed. Such works laid a good foundation for build-time estimation of FDM, though they could not be applied directly to MMLM processes. It is therefore desirable to develop a build-time estimation algorithm for the multi-material FDM processes. 2.2.1. A mathematical model for build-time estimation of multi-material FDM If a multi-material prototype is to be made of N kinds of materials, N nozzles are used to fabricate it. The time required for a nozzle to complete depositing the related contours of a layer is given by T nozzle ¼ T idle þ T deposition
(1)
where Tidle is the nozzle idle time and Tdeposition is the nozzle deposition time. The nozzle idle time includes nozzle positioning time and cleaning time: T idle ¼ T positioning þ T cleaning
(2)
For the sequential multi-toolpath planning described in Section 2.1.1, there are five deposition steps in the tool sequence:
where Tcleaning is the time for cleaning and purging the nozzle and Tpositioning is the time for positioning a nozzle, which
From the two sets of tool sequences above, it is apparent that concurrent multi-toolpath planning can shorten the tool sequence and hence the build-time significantly. The reduction in buildtime will be more substantial for relatively large and complex parts. This can be quantitatively verified using the build-time estimation algorithm described below to estimate and compare the build-times of sequential and concurrent multi-toolpaths.
includes (i) the time for the nozzle to move from the datum point to the entry point of the first toolpath; (ii) the time for the nozzle to return to the datum point from the exit point of the last toolpath; (iii) the time for the nozzle to move from the exit point of a toolpath to the entry point of the subsequent toolpath. Tpositioning is given by T positioning ¼
2.2. Build-time estimation of multi-material FDM process Fused deposition modelling (FDM) is a common type of LM machines that fabricates a part by depositing semi-solid material onto successive layers according to predefined toolpaths. For single-material FDM, estimation and minimisation of build-time have been well studied. Alexander et al. [25] developed a method to estimate build-time by adjusting and optimizing part geometry orientation. Thrimurthulu et al. [26] developed an approach that determines the optimal part deposition orientation for a good compromise of build-time and part quality. Han et al. [27] analysed the relationship between
Tool movement for envelope, Ei (nozzle, Ni) E1 (N1)
E2 (N2)
E3 (N3)
E4 (N4)
E5 (N5)
Sequence Yes No
Yes No
No Yes
Yes No
No Yes
(3)
where Dp is the total positioning distance of a nozzle and Vidle is the average nozzle speed for idle motion. The nozzle deposition time is given by T deposition ¼ T predelay m þ T scanning
(4)
where Tpredelay is the time to start up the servo-pump before a nozzle begins to deposit material, m the number of connected toolpaths to be deposited by a nozzle and Tscanning is the time for a nozzle to fill up the solid part of the connected toolpaths with the desired material. It is given by T scanning ¼
Table 2 Sequence of concurrent nozzle movements
Dp V idle
Ds Vs
(5)
where Ds is the total scanning distance that depends on the hatch space; the smaller the hatch space, the longer the total scanning distance. Vs is the nozzle deposition speed. With Eqs. (1)–(5), the time taken for a nozzle to complete depositing the related contours of a layer, Tnozzle, can be found for subsequent calculation of the build-time of a complete layer based on the results of multi-toolpath planning, as follows.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
The build-time of a layer is the sum of the platform setup time and the layer fabrication time: T layer ¼ T setup þ T fab
(6)
where Tsetup is the platform setup time and Tfab is the time required for complete deposition by all the related nozzles of the layer
Therefore, the total build-time for fabricating a multimaterial prototype is given by T total ¼
L X i¼1
S X T step
(7)
j¼1
where Tstep is the time for completing a deposition step and S is the total number of deposition steps of the layer. For sequential multi-toolpath planning, only one nozzle is allowed in each deposition step to deposit a material on the related contours. This means that the total number of deposition steps of a layer, S, is equal to the total number of nozzles, n, currently used for the layer. Both n and S should be less than or equal to the total number of nozzles, N, needed for complete fabrication of the prototype. Hence, for sequential multitoolpath planning, the time required for a deposition step is given by T step ¼ T nozzle
(8)
For concurrent multi-toolpath planning, one or more nozzles may be depositing materials on the related sets of contours in a deposition step concurrently; the total number of deposition steps of a layer, S, is generally smaller than or at most equal to the total number of nozzles, n, currently used for the layer. Hence, the time for a deposition step is the longest nozzle time among all the nozzles performing deposition in the step, i.e., T step ¼ maxðT nozzle Þ
(9)
T layer ¼
L X
ðT setup þ T fab Þ
i¼1
L S X X ¼ T step T setup þ i¼1
T fab ¼
445
(10)
j¼1
where L is the total number of layers. Hence for sequential multi-toolpath planning, the total build-time is L S X X T total ¼ T nozzle : j 2 ½1; S ¼ n (11) T setup þ i¼1
j¼1
while for concurrent multi-toolpath planning, the total buildtime is L S X X T total ¼ maxðT nozzle Þ : j 2 ½1; S n (12) T setup þ i¼1
j¼1
where Tnozzle is obtained using Eqs. (1)–(5). 2.2.2. Verification of estimated build-time The mathematical model above has been implemented in the proposed sequential and concurrent multi-toolpath planning algorithms. As practicable MMLM machines are not yet commercially available, the accuracy of build-time estimated above is instead verified using a Stratasys FDM3000 machine, which is a single-material fused deposition modelling machine. It is assumed that the material-depositing mechanism of an MMLM machine consists of an array of nozzles of the same operation characteristics, and each nozzle deposits a specific
Fig. 7. Fabrication of a machine part prototype on FDM3000 machine.
446
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
material similar to that of FDM3000. With this assumption, an experiment of fabricating the machine part in Fig. 2a on FDM3000 was conducted to verify the build-time of sequential multi-toolpath planning. Although the result may not be entirely applicable to concurrent multi-toolpath planning, it does verify the accuracy of the parameters of the mathematical model for estimating concurrent build-time. Fig. 7 shows the machine part prototype being fabricated on FDM3000, together with the base support and the resulting prototype. The operation parameters of FDM3000, such as the platform setup time per layer (Tsetup) and the time required for cleaning and purging the nozzle (Tcleaning) were measured. The machine part in Fig. 2a was of dimensions 100 mm 60 mm 3.5 mm; it was fabricated and the build-time recorded, as follows. InsightTM, the proprietary control software of FDM3000, was used to slice the machine part and generate the base structure. Using the default setting of the layer thickness t and the road width w as 0.178 mm and 0.356 mm, respectively, 28 layers were sliced, with the first 8 layers for the base structure and the remaining 20 layers for the part. The layer thickness and the road width are two major factors that affect the build-time and the quality. In order to obtain a good compromise between the buildtime and the quality, the distance between centre points of two adjacent roads (toolpaths), called hatch space h, should be appropriately controlled. If the distance is too large (small), underfills (overfills) between the two adjacent roads will occur [27,28]. To overcome such problems, a road geometry model [27,28] was adopted to find the necessary negative offset f between the two adjacent roads as shown in Fig. 8. Referring to Fig. 8, the hatch space is expressed as h¼wþtþ f
(13)
where f is expressed as f ¼
ðp 4Þt 4
(14)
Hence, the proposed build-time algorithm adopts Eqs. (13) and (14) to obtain the required hatch space for generation of toolpaths. Based on the default values of the layer thickness t = 0.178 mm and the road width w ¼ 0:356 mm, the hatch space h = 0.496 mm was obtained to generate toolpaths. ABS P400 and P400SR were used, respectively, for the part material and the base material, and two T12 nozzle tips were used as the part and the base. The nozzle speed Vs depends on the nozzle tip size and the type of material used; and Vs is assumed constant in a deposition process [27]. From the specifications of FDM3000 [29], the maximum nozzle speed for T12 was 20.320 mm s1 (0.800 in. s1) and the time to start
Fig. 8. Relation between road width and hatch space.
Table 3 Actual build-time and estimated build-times of the machine part Total build-time of the machine part (min) Actual build-time on FDM3000 Estimated build-time by FDM3000’s control software, InsightTM Estimated sequential build-time by the proposed algorithm Estimated concurrent build-time by the proposed algorithm
Difference (%)
136.27 151.00
– 10.809
136.01
0.191
114.46
16.005
up the servo-pump before a nozzle began to deposit material Tpredelay was approximately 0.1 s. For idle motion, the average nozzle speed Vidle was 20.320 mm s1. The platform setup time per layer Tsetup and the time for cleaning and purging the nozzle Tcleaning were measured to be 1.0 s and 5.0 s, respectively. The machined part was physically fabricated on FDM3000. The total build-time was 178.77 min, of which the base structure took up 42.50 min while the part 136.27 min. The estimated build-time by FDM3000’s control software, InsightTM, was 197.55 min, with 46.55 min of which for the base and 151.00 min for the part. The proposed build-time estimation algorithm was also used to estimate the build-time of the machine part using the operation parameters of FDM3000. For simplicity, the base structure was ignored because its design and the way of fabrication depend largely on the proprietary control software of LM machines as well as on the geometry and orientation of the part. Table 3 shows the actual and estimated build-times of the machine part fabricated on FDM3000 and their percentage differences. It can be seen that while FDM3000’s control software overestimated the build-time of the part by 10.809%, the sequential build-time and the concurrent build-time estimated by the proposed algorithm are 136.01 min and 114.46 min, respectively. The sequential build-time is literally the same as the actual build-time, while the concurrent build-time represents a reduction of 16.005%. This result shows that the proposed build-time estimation algorithm is accurate and reliable; it also confirms the expectation that concurrent multi-toolpath planning helps reduce build-time. 2.3. Digital fabrication of multi-material prototypes The algorithms above are integrated with a VR simulation system to form an MMVP system for toolpath planning, buildtime estimation, stereoscopic simulation, validation of multitoolpaths, and visualisation of MMLM processes. Fig. 9 shows a snapshot of the simulation of concurrent deposition of materials on the layer shown in Fig. 2a. To complement buildtime estimation, VR simulation facilitates verification of multitoolpaths through visualisation. Subsequently, when digital fabrication of a prototype is completed, it can be superimposed on its STL model to identify the resulting dimensional deviations. The overall accuracy of the prototype can be highlighted by showing the maximum and the average cusp
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
447
Fig. 9. Simulation of concurrent nozzle movements for the layer in Fig. 2a.
heights. Hence, the designer can analyse prototype quality iteratively without incurring much cost until a good compromise between quality and build-time is struck with an appropriate set of process parameters, such as build direction, layer thickness, and hatch space. 3. Case studies 3.1. A jewellery pendant Hong Kong is renowned for its jewellery products. To survive competition in global markets, most manufacturers
have adopted advanced technologies, including CAD/CAM and LM technologies. Hence, the MMVP system will be useful for the jewellery industry to help reduce product development time and cost. Therefore, a pendant, as shown in Fig. 10, is used to demonstrate digital fabrication, using concurrent multitoolpaths, of multi-material prototypes for development of jewellery products. The pendant STL model is sliced into 25 layers with layer thickness 0.178 mm. The first layer of 25 contours is to be made of 5 kinds of materials. Its topological hierarchy relationship in Fig. 11 is established, in which the contours are grouped into 18 contour families, and 18 toolpaths (PC15,8,9,10,11,12,13,14, PC7,
Fig. 10. A layer of contours of a multi-material pendant.
448
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 11. Topological hierarchy relationship of contours of the pendant layer.
PC1, PC2, PC3, PC4, PC5, PC6, PC25,20,21,22,23,24, PC16, PC17, PC18, PC19, PC35,30,31,32,33,34, PC26, PC27, PC28, and PC29) are generated for these contour families accordingly with a hatch space of 0.496 mm. Based on the material information, the toolpaths with the same material property are grouped into five toolpath-sets, namely from S1 to S5, which are associated with five nozzles from N1 to N5, respectively. Subsequently, five work envelopes from E1 to E5 for each of these nozzles are constructed to facilitate planning of concurrent multi-toolpaths. Potential collisions of nozzles can be detected by overlap tests of the envelopes E1, E2, E3, E4, and E5. According to the overlap test result, the system can efficiently generate concurrent multitoolpaths. Three steps are generated and the sequences of concurrent nozzle movements are as follows:
multi-toolpath planning becomes substantial when the part is complex, in comparison with the machine part in Fig. 2a. The MMVP system can digitally fabricate the pendant prototype for quality analysis through visualisation. Fig. 12 shows the digital fabrication process of a few pendant layers, while Fig. 13 highlights the excessive materials of the resulting prototype when it is superimposed on the STL model. Its average and maximum cusp heights, which are 0.085 mm and 0.178 mm, respectively, are calculated and highlighted. Therefore, using the MMVP system, the designer can conveniently perform design iterations and quality analysis of the resulting prototypes. Thus, an optimal combination of process parameters, such as layer thickness, build direction, and hatch space can be obtained for cost-effective fabrication of physical prototypes.
With this, the MMVP system can estimate the build-time to verify the resulting multi-toolpaths. Table 4 shows the estimated build-times for the sequential multi-toolpaths and the concurrent multi-toolpaths, which would be useful for subsequent process planning and cost estimation. It can be seen that the estimated build-time by concurrent multi-toolpath planning is 31.342% shorter than by sequential multi-toolpath planning. This shows that build-time reduction by concurrent
3.2. A pelvis model with two inserted pins
Table 4 Estimated sequential and concurrent build-times for the pendant
Sequential multi-toolpaths Concurrent multi-toolpaths
Estimated build-time (min)
Difference (%)
36.66 25.17
– 31.342
LM technology has now been widely adopted to produce medical prototypes for studying human bone or organ structures and surgical planning [30,31]. As multi-material prototypes are more revealing for identifying complex organ structures and thus more helpful for managing delicate surgical conditions, the MMVP system is deemed useful for biomedical applications. Surgeons can study and plan a surgical operation conveniently in the VR environment provided by the system; they can also design a prosthetic component of appropriate size, shape, and orientation that would fit the patient best before physical fabrication takes place. A pelvis model with two simple pins inserted for fixation, as shown in Fig. 14, is therefore used to illustrate the usefulness of the MMVP system for biomedical engineering.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 12. Digital fabrication process of a multi-material prototype of the pendant.
The pelvis is sliced into 721 layers with a layer thickness of 0.178 mm. Fig. 14 shows a layer containing 21 contours to be made of five kinds of materials, namely m1, m2, m3, m4, and m5. Five nozzles, N1, N2, N3, N4, and N5, will deposit specific materials, m1, m2, m3, m4, and m5, respectively. Fig. 15 shows the topological hierarchy relationship of the layer contours. According to the contour families, eight toolpaths namely, PC2,21,15,16,17,3,5, PC1, PC6,7,14, PC4, PC11, PC8,18,19, PC9,10,12,20, and PC13 are generated by hatching the contours with hatch space 0.496 mm. Subsequently, the
449
Fig. 13. Superimposition of pendant prototype on its STL model to highlight dimensional deviations.
toolpaths with the same material properties are grouped together to form five toolpath-sets, namely S1, S2, S3, S4, and S5, respectively. Five work envelopes from E1 to E5 for nozzles from N1 to N5 are then constructed to facilitate planning of concurrent multi-toolpaths. Potential nozzle collisions can be detected by overlap test of the envelopes E1, E2, E3, E4, and E5. According to the overlap test result, the system generates concurrent multi-toolpaths, which are arranged in four steps with the sequence of nozzle movements as follows:
Fig. 14. A layer of contours of a multi-material pelvis with two pins.
450
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
Fig. 15. Topological hierarchy relationship of contours.
The build-time of the pelvis prototype based on concurrent multi-toolpath planning is estimated to be 1763.87 min. In comparison with the estimated sequential build-time of 2475.72 min, as shown in Table 5, a reduction of 28.753% is achieved by concurrent multi-toolpath planning. This shows that build-time reduction by concurrent multi-toolpath planning for relatively large and complex parts is substantial. The estimated build-time facilitates cost estimation of customised prosthetic components as well as scheduling of preoperative tasks. Furthermore, the MMVP system provides
stereoscopic visualisation of the resulting prototypes to help surgeons design prostheses with best fit possible for the patient. The pelvis prototype can be superimposed on its STL model, as shown in Fig. 16, to highlight dimensional deviations. In addition, the system can calculate the average and maximum cusp heights of the prototype to facilitate evaluation of the overall accuracy of the prototype. Thus, the surgeons can repeat digital fabrication of the pelvis prototype until an optimal tradeoff between the quality and the production cost of the pelvis prototype can be achieved.
Fig. 16. Superimposition of pelvis prototype on its STL model to highlight dimensional deviations.
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452 Table 5 Estimated sequential and concurrent build-times for the pelvis prototype
Sequential multi-toolpaths Concurrent multi-toolpaths
Estimated build-time (min)
Difference (%)
2475.72 1763.87
– 28.753
3.3. Limitation of the proposed MMVP system Case studies show that the MMVP system can generate sequential and concurrent multi-toolpaths for effective control of MMLM processes; reduction in build-time by concurrent multi-toolpath planning is more substantial for relatively large and complex prototypes; the estimated build-time is useful for validation of multi-toolpaths, as well as for scheduling and cost estimation of prototype fabrication. Nevertheless, the resulting concurrent multi-toolpaths may not be of absolute optimality, because the tools in overlapping envelopes are not allowed to move concurrently. In the next step of our development work, the concurrent multi-toolpath planning algorithm would be enhanced to decompose overlapping envelopes into smaller ones for detailed detection of possible tool collisions in order to increase the likelihood of concurrent nozzle movements. Indeed, the optimality of concurrent multi-toolpaths may be further enhanced by incorporating more vigorous optimisation rules, such as critical path, to find the shortest nozzle movement distance possible while allowing as many nozzles to deposit concurrently as possible. Obviously, this requires much computing effort and is not an easy task, which would be undertaken in our further work. 4. Conclusion To address the key software issue of MMLM, an MMVP system has been developed for digital fabrication of heterogeneous multi-material prototypes. The MMVP system integrates VR and MMLM technologies for advanced product development and biomedical applications. It consists of a suite of algorithms for multi-toolpath planning, build-time estimation, simulation, optimisation, and visualisation of MMLM processes. Based on the topological hierarchy relationship of slice contours, the sequential multi-toolpath planning algorithm can arrange the contours in an appropriate sequence to generate sequential multi-toolpaths without redundant movements. To save build-time further, the concurrent multi-toolpath planning algorithm can generate concurrent multi-toolpaths. To validate the resulting sequential and concurrent toolpaths, a build-time estimation algorithm for MMLM process is developed. Designers can use the MMVP system to iterate digital fabrication of prototypes conveniently until an optimal trade-off between the quality and the cost is achieved. The MMVP system may be adapted as an intelligent control software system for practical and viable MMLM machines.
451
Acknowledgements The authors would like to acknowledge the Research Grant Council of the Hong Kong SAR Government and the CRCG of the University of Hong Kong for their financial support for this project. Thanks are also due to Dr. Y.H. Chen of the Department of Mechanical Engineering at the University of Hong Kong for facilitating the fabrication of the machine part on the Stratasys FDM 3000 machine. References [1] K. Samanta, B. Koc, Feature-based and material blending for free-form heterogeneous object modeling, Computer-Aided Design 37 (3) (2005) 287–305. [2] S.F. Yang, G.J.R. Evans, A multi-component powder dispensing system for three dimensional functional gradients, Materials Science & Engineering A 379 (1–2) (2004) 351–359. [3] A. Safari III, J. Cesasrano, P.G. Clem, B. Bender, Fabrication of advanced functional electroceramic components by layered manufacturing (LM) methods, in: Proceedings of the 13th IEEE International Symposium on Applications of Ferroelectronics, Nara, Japan, (2002), pp. 1–6. [4] E. Sachlos, N. Reis, C. Ainsley, B. Derby, J.T. Czernuszka, Novel collagen scaffolds with predefined internal morphology made by solid freeform fabrication, Biomaterials 24 (8) (2003) 1487–1497. [5] K. Lappo, B. Jackson, K. Wood, D. Bourell, J.J. Beaman, Discrete multiple material selective laser sintering (M2SLS): experimental study of part processing, in: D.L. Bourell, et al. (Eds.), Solid Freeform Fabrication Symposium, Austin, Texas, The University of Texas, Austin, TX, 2003, pp. 109–119. [6] M.A. Jafari, W. Han, F. Mohammadi, A. Safari, S.C. Danforth, N. Langrana, A novel system for fused deposition of advanced multiple ceramics, Rapid Prototyping Journal 6 (3) (2000) 161–175. [7] W.J. Cho, E.M. Sachs, N.M. Patrikalakis, D.E. Troxel, A dithering algorithm for local composition control with three-dimensional printing, Computer-Aided Design 35 (9) (2003) 851–867. [8] Y.N. Yan, Z. Xiong, Y.Y. Hu, S.G. Wang, R.J. Zhang, C. Zhang, Layered manufacturing of tissue engineering scaffolds via multi-nozzle deposition, Materials Letters 57 (18) (2003) 2623–2628. [9] R.E. Bremnan, S. Turcu, A. Hall, N.M. Hagh, A. Safari, Fabrication of electroceramic components by layered manufacturing (LM), Ferroelectrics 293 (1) (2003) 3–17. [10] DMDTM, http://www.pom.net/, 2004. [11] LENSTM, http://mfgshop.sandia.gov/1400_ext/1400_ext_LENS.htm, 2004. [12] V. Kumar, S. Rajagopalan, M. Cutkosky, D. Dutta, Representation and processing heterogeneous objects for solid freeform fabrication, in: Sixth IFIP WG 5.2 International Workshop on Geometric Modelling: Fundamentals and Applications, Tokyo, Japan, The University of Tokyo, 1998. [13] S.M. Morvan, G.M. Fadel, Heterogeneous solids: possible representation schemes, in: D.L. Bourell, et al. (Eds.), Solid Freeform Fabrication Symposium, Austin, Texas, The University of Texa, Austin, Texas, 1999, pp. 187–198. [14] D. Qiu, N.A. Langrana, S.C. Danforth, A. Safari, M. Fafari, Intelligent toolpath for extrusion-based LM process, Rapid Prototyping Journal 7 (1) (2001) 18–24. [15] S. Ding, M.A. Mannan, A.N. Poo, Oriented bounding box and octree based global interference detection in 5-axis machining of free-form surfaces, Computer-Aided Design 36 (13) (2004) 1281–1294. [16] O. Ilushin, G. Elber, D. Halperin, R. Wein, M.-S. Kin, Precise global collision detection in multi-axis NC-machining, Computer-Aided Design 37 (9) (2005) 909–920. [17] E.L.J. Bohez, N.T.H. Minh, B. Kiatsrithanakorn, P. Natasukon, R.-Y. Huang, T. Son, The stencil buffer sweep plane algorithm for 5-axis CNC tool path verification, Computer-Aided Design 35 (12) (2003) 1129– 1142.
452
S.H. Choi, H.H. Cheung / Computers in Industry 58 (2007) 438–452
[18] Y. Yang, J.Y.H. Fuh, H.T. Loh, An efficient scanning pattern for layered manufacturing processes, in: Proceedings of the 2001 IEEE International Conference on Robotic & Automation, Seoul, Korea, (2001), pp. 1340– 1345. [19] J.H. Kao, F.B. Prinz, Optimal motion planning for deposition in layered manufacturing, in: Proceedings of ASME Design Engineering Technical Conferences, Atlanta, GA, (1998), pp. 1–10. [20] J.S. Liang, A.C. Lin, System development of adaptive spraying paths for rapid prototyping, Rapid Prototyping Journal 9 (5) (2003) 289–300. [21] K. Tang, A. Pang, Optimal connection of loops in laminated object manufacturing, Computer-Aided Design 35 (11) (2003) 1011–1022. [22] S.H. Choi, K.T. Kwok, A topological hierarchy-sorting algorithm for slice contours, Rapid Prototyping Journal 10 (2) (2004) 98–113. [23] S.H. Choi, H.H. Cheung, A multi-material virtual prototyping system, Computer-Aided Design 37 (1) (2005) 123–136. [24] S.H. Choi, H.H. Cheung, A topological hierarchy-based approach to toolpath planning for multi-material layered manufacturing, ComputerAided Design 38 (2) (2006) 143–156. [25] P. Alexander, S. Allen, D. Dutta, Part orientation and build cost determination in layered manufacturing, Computer-Aided Design 30 (5) (1998) 343–356. [26] K. Thrimurthulu, P.M. Pandey, N.V. Reddy, Optimum part deposition orientation in fused modeling, Machine Tools & Manufacture 44 (6) (2003) 585–594. [27] W.B. Han, M.A. Jafari, K. Seyed, Process speeding up via deposition planning in fused deposition-based layered manufacturing processes, Rapid Prototyping Journal 9 (4) (2003) 212–218.
[28] W.B. Han, M.A. Jafari, S.C. Danforth, A. Safari, Tool path-based deposition planning in fused deposition processes, Transactions of the ASME, Journal of Manufacturing Science and Engineering 124 (2) (2002) 462–472. [29] FDM, FDM1 System Documentation, Stratasys Inc., 2001. [30] P. Peters, F. Langlotz, L.-P. Nolte, Computer assisted screw insertion into real 3D rapid prototyping pelvis models, Clinical Biomechanics 17 (5) (2002) 376–382. [31] E.D. Momi, E. Pavan, B. Motyl, C. Bandera, C. Frigo, Hip joint anatomy virtual and stereolithographic reconstruction for preoperative planning of total hip replacement, International Congress Series 1281 (2005) 708–712. Dr. S.H. Choi is associate professor in the IMSE Department at the University of Hong Kong. He obtained both his B.Sc. and Ph.D. degrees at the University of Birmingham. He worked in computer industry as CADCAM consultant before joining the University of Hong Kong. His current research interests include CADCAM, advanced manufacturing systems and virtual prototyping technology.
Mr. H.H. Cheung gained his B.Eng. degree from the IMSE Department at the University of Hong Kong. He continued his postgraduate research study in the Department, and his research interest is in virtual prototyping technology.