A comparison of rapid prototyping technologies

International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

A comparison of rapid prototyping technologies D.T. Pham*, R.S. Gault Cardiff Rapid Prototyping Centre, Systems Division, School of Engineering, University of Wales Cardiff, PO Box 688, Cardiff CF2 3TE, UK Received 16 October 1997

Abstract Until recently, prototypes had to be constructed by skilled model makers from 2D engineering drawings. This is a time-consuming and expensive process. With the advent of new layer manufacturing and CAD/CAM technologies, prototypes may now be rapidly produced from 3D computer models. There are many different rapid prototyping (RP) technologies available. This paper presents an overview of the current technologies and comments on their strengths and weaknesses. Data are given for common process parameters such as layer thickness, system accuracy and speed of operation. A taxonomy is also suggested, along with a preliminary guide to process selection based on the end use of the prototype.  1998 Elsevier Science Ltd. All rights reserved. Keywords: Rapid prototyping; Stereolithography; Selective laser sintering; LOM; 3D printing; Fused deposition modelling

1. Introduction Prototyping is an essential part of the product development and manufacturing cycle required for assessing the form, fit and functionality of a design before a significant investment in tooling is made. Until recently, prototypes were still largely handmade by skilled craftsmen, adding weeks or months to the product development time. Because of this, only a few design iterations could be made before tooling went into production, resulting in parts which at best were seldom optimised and at worst did not function properly. Rapid prototyping (RP) is a term which embraces a range of new technologies for producing

* Corresponding author. 0890-6955/98/$19.00  1998 Elsevier Science Ltd. All rights reserved. PII: S 0 8 9 0 - 6 9 5 5 ( 9 7 ) 0 0 1 3 7 - 5

1258

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

accurate parts directly from CAD models in a few hours, with little need for human intervention. This means that designers have the freedom to produce physical models of their drawings more frequently, allowing them to check the assembly and function of the design as well as discussing downstream manufacturing issues with an easy-to-interpret, unambiguous prototype. Consequently, errors are minimised and product development costs and lead times substantially reduced. It has been claimed that RP can cut new product costs by up to 70% and the time to market by 90% [1]. RP technologies may be divided broadly into those involving the addition of material and those involving its removal. According to Kruth [2], the material accretion technologies may be divided by the state of the prototype material before part formation. The liquid-based technologies may entail the solidification of a resin on contact with a laser, the solidification of an electrosetting fluid, or the melting and subsequent solidification of the prototype material. The processes using powders compound them either with a laser or by the selective application of binding agents. Those processes which use solid sheets may be classified according to whether the sheets are bonded with a laser or with an adhesive. Figure 1 shows Kruth’s classification, which has been adapted to include new technologies. In the following, RP technologies are presented according to the arrangement shown in this figure.

Fig. 1. Classification of rapid prototyping methods (adapted from [2]).

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1259

2. Material addition technologies All of the processes reviewed require input from a 3D solid CAD model, usually as slices. The designer therefore first uses a CAD package to design the product which he wishes to manufacture. This model is then tessellated and exported as an STL file, which is the current industry standard for facetted models, although it may be possible, in future, to slice models directly from the CAD system without first facetting them [3]. If supports are necessary to brace any overhangs, proprietary software may now add these to the model. It is then sliced and the slices sent to the RP machine for the production of the final physical part. By convention, the data slices are said to be in the X–Y plane and the part is built in the Z direction. An important problem is automatic support generation and part orientation. This is because part orientation will influence the final prototype build time and the surface finish of critical areas. The number and position of the supports depend to some extent upon the build direction chosen and may also adversely affect the build time and surface finish of the prototype [3–5]. 2.1. Processes involving a liquid 2.1.1. Solidification of a liquid polymer Of the five processes in this category, which all involve the solidification of a resin via electromagnetic radiation, three construct the part using points to build up the layers whilst the other two solidify entire layers or surfaces at once. 2.1.1.1. Stereolithography (SL) The most popular among currently available RP technologies is perhaps stereolithography. This relies on a photosensitive monomer resin which forms a polymer and solidifies when exposed to ultraviolet (UV) light. Due to the absorption and scattering of the beam this reaction only takes place near the surface. This produces parabolically cylindrical voxels (three-dimensional pixels) as shown in Fig. 2 which are characterised by their horizontal line width and vertical cure depth [6]. An SL machine consists of a build platform (substrate) which is mounted in a vat of resin and a UV helium–cadmium or argon ion laser (Fig. 3). The first layer of the part is imaged on the resin surface by the laser using information obtained from the 3D solid CAD model. Once the contour of the layer has been scanned and the interior either hatched or solidly filled, the platform is next lowered to the base of the vat in order to coat the part thoroughly. It is then raised such that the top of the solidified part is level with the surface and a blade wipes the resin leaving exactly one layer of resin above the part. The part is then lowered to one layer below the surface and left until the liquid has settled [7]. This is done to ensure a flat, even surface and to inhibit bubble formation. The next layer may then be scanned. All new SL machines now employ a method to apply the resin that is superior to the deep-dip process described above. Because of the high resin viscosity, after the deep dip and recoating, either too little or too much resin is left by the recoating blade, which affects part accuracy. The new method involves spreading resin on the part as the blade traverses the vat. Because the blade applies only the required amount of resin, good accuracy is achieved. This method also provides a smoother surface finish and reduces non-productive recoat time. Another important advantage is the elimination of ‘trapped volume’ problems. A trapped volume is a volume of resin that

1260

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 2. Single cured line of photopolymer (adapted from [6]).

cannot drain through the base of the part (Fig. 4). The presence of a trapped volume in the deepdip process affects part accuracy and may lead to delamination or collision of the blade and part because of a build up of unwanted polymerised resin at the surface. Once the part is completed, it is removed from the vat and the excess resin drained. Due to the resin viscosity, this stage may take several hours. The supports are removed and the ‘green’ part is then placed in a UV oven to be postcured. This ensures that no liquid or partially cured resin remains. Solid or partially solid parts are made with either acrylic or epoxy resins in one of several build styles, the three most common being ACES, STARWEAVE and QuickCast [8]. Completely hollow parts are not normally constructed as these are very fragile in the green state and deform on handling. When adopting ACES, the interior of the part is almost wholly cured by the laser (Fig. 5). This is achieved by using a hatch-spacing which is equivalent to half the line width. This spacing

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 3.

Stereolithography.

Fig. 4. Trapped volume in stereolithography.

1261

1262

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 5. ACES build style: repeated, even laser exposure produces a flat base.

is chosen such that all the solidified resin receives the same cumulative UV exposure and hence the downward facing surfaces are flat. This style may only be used with epoxy resins that do not shrink much when polymerised otherwise the connected lines would cause warping in the prototype. It is the most accurate of the three build styles for low-distortion resins and is employed when making high precision parts although the drawing time is the longest of the three styles [9]. STARWEAVE provides stability to a solid part by hatching the interior with a series of grids which are offset by half of the hatch spacing every other layer (Fig. 6). The grids are drawn such that the ends are not attached to the part border to reduce the overall distortion. Also, to keep the distortion low, the gridlines do not touch one another. However, they are located as closely together as possible to improve the green strength of the part [8,9]. This build style should be employed with acrylic resins which have a higher shrinkage when polymerised. It is sometimes used with epoxy resins in preference to ACES because the draw time is lower. QuickCast is usually adopted when the prototype is to be employed as a pattern for investment casting as it produces almost hollow parts. The outline of the layer is drawn before the interior is hatched. Either squares (QuickCast version 1.1) or equilateral triangles (QuickCast version 1.0) are used to fill the part and these are offset after a specified vertical build distance to facilitate resin drainage. The triangles are offset such that the vertices of one section are above the centroids of the triangles in the previous section (Fig. 7). The squares are offset by half of the hatch spacing. Since squares have larger interior angles than triangles, the meniscus of resin will be smaller so better drainage is achieved [9]. Horizontal sections that form the outer surface of the part are completely solidified and are referred to as skinfill areas. Three layers are drawn with skinfill areas corresponding to the part surface to avoid the formation of ‘pinholes’ when the supports are removed and to prevent the upwards-facing horizontal surfaces from sagging [9,10]. These

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1263

Fig. 6. STARWEAVE build style. (a) One layer of STARWEAVE. This is composed of a cross-hatched grid which is detached from the part border; (b) alternate layers of STARWEAVE are offset by half the hatch-spacing.

skinfills support the part surface, which means that the hatch spacing may be larger and a smaller percentage of the prototype is solid [9]. Vents and drains must be designed into these areas to allow the excess resin to bleed from the part. These parts will collapse quickly upon firing so that little stress is developed on the ceramic investment shell, preventing it from being damaged. Because QuickCast parts have a large surface area and the resin is hygroscopic, they should be used as quickly as possible and stored in an area with controlled humidity to prevent later distortion due to water absorption. Hatch spacing must be determined so that the voxels are situated sufficiently near to each other to allow the layers to be connected, but not so closely that the laser scan time is unacceptable or residual stresses are developed through overcure. The layer thickness will obviously affect the closeness of the voxels in the vertical direction — if the layers are too thick, surfaces will not connect [7]. Voxels on sloped surfaces must be nearer to avoid gaps through which resin may drain or through which slurry may invade in later processes such as investment casting. The advantages of stereolithography are that it produces a surface finish that is comparable to that of NC milling, it is a well proven system with over 500 machines in use worldwide and it is reasonably fast and accurate [11,12]. To utilize the resin vat fully, several parts may be built at once. The disadvantages are that the material is expensive, smelly and toxic and must be shielded from light to avoid premature polymerization; there is also a limited choice of resins. The parts may be brittle and translucent and they need supports which may adversely affect the surface finish when removed. The system has an accuracy of ± 100 ␮m and can achieve layers 50 ␮m thick [13]. A machine with a build chamber of 250 × 250 × 250 mm, the most common size, will cost approximately

1264

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 7. QuickCast build style: parts are hatched with offset triangles.

£150 000. The largest build chamber commercially available measures 500 × 500 × 584 mm [14]. The recoat time is 35 s for the new method and more than 50 s for the deep dip method. The draw time is proportional to the cross-sectional area of the part; a layer with a cross-section of 50 × 50 mm2 takes about 78 s to solidify, according to the laser power and curing parameters. Further research is being actively conducted into materials and into the accuracy, warping and shrinkage of the parts.

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1265

2.1.1.2. Liquid thermal polymerization (LTP) This process is similar to SL except that the resin is thermosetting and an infrared laser is used to create the voxels. This difference means that the size of the voxels may be affected through heat dissipation, which may also cause unwanted distortion and shrinkage in the part. However, the problems are apparently no worse than those caused by SL and are controllable [2]. This system is still being researched. 2.1.1.3. Beam interference solidification (BIS) This process uses two laser beams mounted at right angles which emit light at different frequencies to polymerise resin in a transparent vat (Fig. 8). The first laser excites the liquid to a reversible metastable state and then the incidence of the second beam polymerises the excited resin. To date, there are no commercial applications of this technology because there are still technical difficulties to be solved: 쐌 Shadows are cast from previously solidified sections. 쐌 There is a problem with light absorption because the intensity of the lasers drops with depth. 쐌 It is hard to intersect the laser beams due to diffraction variations in the resin caused by temperature gradients or solid sections [2].

Fig. 8. Beam interference solidification (adapted from [2]).

1266

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

2.1.1.4. Solid ground curing (SGC) This system again utilizes photopolymerising resins and light. Data from the CAD model is used to produce a mask which is placed above the resin surface. The entire layer can then be illuminated with a UV lamp (Fig. 9). Once the layer has been cured, the excess resin is wiped away and any spaces are filled with wax. The wax is cooled with a chill plate, milled flat and any chips removed. A new layer of resin is applied and the process is repeated. The mask itself is a sheet of glass which is prepared whilst the current layer is being waxed, cooled and milled. The negative image of each subsequent layer is produced electrostatically on the glass and developed using a toner in a similar manner to laser printing. Because wax is used to fill the gaps in the cured resin, no further supports need to be added by the interface software. The wax supports any overhangs in the design and anchors any discrete protrusions which may be drawn on a layer. It also theoretically reduces distortion due to warping and curl since the part is surrounded and means that the machines do not need to be vibration proofed as the part cannot move in the vat [2,15]. Builds may also be paused to allow other, more urgent parts to be made [16]. An advantage of this system is that the entire layer is solidified at once, reducing the part creation time, especially for multi-part builds. Parts may also be nested to utilise the build volume fully. All the resin within a layer is completely cured by this method, and so no postcuring is required, parts may be more durable than the hatched prototypes created

Fig. 9. Solid ground curing.

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1267

using other processes and operators need not handle partially cured, toxic resin [16]. The wax may be removed automatically in a special machine. The disadvantages of this system are that it is noisy, large and heavy and needs to be constantly manned. It wastes a large amount of wax which cannot be recycled and is also prone to breakdowns [1,9]. The mask is produced by raster scanning the image [16] which may cause steps in the X–Y plane, affecting accuracy. The resin models produced using SGC are solid and so cannot be used for later investment casting since the coefficient of thermal expansion of the resin is an order of magnitude greater than that of the ceramic system so the ceramic moulds will crack when the sacrificial part is burnt out [1]. The resolution is 100 ␮m in the horizontal X–Y plane and 100 ␮m in the z direction. The least expensive SGC machine costs around £180 000 and weighs about 5000 kg. The largest build chamber available is 500 × 350 × 500 mm. Typically, a layer can be built in 65–120 s, depending on the machine used. Of this building time, 3 s are for exposing the layer to a 2000 W UV lamp, the remaining time being needed to clear the part of resin and to add, chill and mill the wax [16,17]. 2.1.1.5. Holographic interference solidification (HIS) A holographic image is projected into the resin causing an entire surface to solidify. Data is still obtained from the CAD model, although not as slices. The build space is 300 × 300 × 300 mm [2]. There are no commercial systems available yet. 2.1.2. Solidification of an electroset fluid: electrosetting (ES) Electrodes are printed onto a conductive material such as aluminium. Once all the layers have been printed, they are stacked, immersed in a bath of electrosetting fluid and energised. The fluid which is between the electrodes then solidifies to form the part. Once the composite has been removed and drained, the unwanted aluminium may be trimmed from the part. Advantages of this technology are that the part density, compressibility, hardness and adhesion may be controlled by controlling the voltage and current applied to the aluminium. Parts may be made from silicon rubber, polyester, polyurethane or epoxy. The hardware for such a system may be bought off the shelf and costs about £5000. The software for the system is still being developed [18]. 2.1.3. Solidification of molten material There are four technologies which involve the melting and subsequent solidification of the part material. Of these, the first three deposit the material at discrete points whilst the fourth manufactures the whole layer at once. 2.1.3.1. Ballistic particle manufacture (BPM) A stream of molten material is ejected from a nozzle. It separates into droplets which hit the substrate and immediately cold weld to form the part (Fig. 10). If the substrate is rough, thermal contact between it and the part is increased which will reduce stresses within the part [19]. The stream may be a drop-on-demand system or a continuous jet. When a continuous jet is adopted, it is ejected through the nozzle which is being excited by a piezoelectric transducer at a frequency of about 60 Hz [20]. To avoid melting the transducer, it is located at a distance from

1268

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 10. Ballistic particle manufacture.

the nozzle. Although a capillary stream will naturally decompose into droplets [21], the disturbance at the nozzle forces the production of a stream of small, regular droplets with uniform spacing and distance. Using a low-frequency carrier wave modulated by a higher frequency disturbance, tailor made streams have been produced where the user is able to specify larger droplet separations than would otherwise be obtainable with just a single frequency. Regular streams have also been produced consisting of a few small, close droplets followed by larger, more distant droplets [22]. This should allow more time for the nozzle to move to a new position or for the droplets to solidify if necessary. Parameters that will affect the eventual part characteristics are the temperature and velocity of the droplets and the charge that they carry. The charge is acquired electrostatically when the stream is ejected and can be used for the accurate placement of the material. Since the maximum charge which may be held by a drop is limited, the maximum deflection of such a drop is also limited and the substrate or the jet must therefore be movable in order to produce a large enough build area. The temperature will control the speed at which the molten material solidifies. If the droplets are too cold they will solidify midflight and will therefore not weld to the part. If they are too hot, the part will lose shape. The deformation and placement accuracy of the droplet depend on its velocity. If it is moving too slowly, placement accuracy will be poor; if it moves too quickly the droplet will be highly deformed on impact [19]. The resolution of the prototypes is related to the droplet diameter which is typically 50–100 ␮m. Droplets may be released in nitrogen or in vacuo to avoid their oxidation and dispersion. The deposition rate is up to 15 000 droplets per second using a single nozzle and a continuous jet

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1269

[19]. In this process, the supports are usually made from a different material which facilitates their subsequent removal from the part. Advantages of BPM are that it is cheap and environmentally safe and that metal parts made using this technology have a finer grain structure than the equivalent cast parts. This is because the splat cooling of the droplets means that they retain an amorphous structure instead of crystallising, giving the prototype good mechanical properties. Materials which may currently be employed for part construction are tin, zinc, lead, other low (⬍ 420°C) melting point alloys and thermoplastics. Systems are being developed to deposit copper which melts at 1100°C [19]. A disadvantage is the small range of commercial materials available to construct the prototypes. Of the systems available, either speed or accuracy is possible, but not both attributes. There are several commercial dual material systems available which can deposit either thermoplastic or wax. One of the most accurate, BPM1, uses a drop-on-demand jet to eject the molten material. The droplets are spheres, 76 ␮m in diameter, which flatten on impact to give discs which have a diameter of 101 ␮m and are 63 ␮m thick. After each layer is deposited, the part is milled to achieve accurate dimensions in the z direction. In order to maintain the tolerances in the horizontal plane, the layer contours are drawn using linear interpolation (not raster scanning) before the interior of the part is filled. The system is able to vary the layer thickness in order to provide speed in areas where the geometry remains unchanged from layer to layer without losing accuracy in critical areas. A future improvement is the use of a larger nozzle to deposit material within the boundary of the part. This should significantly reduce the build time. The system is claimed to have an exceptionally good accuracy of ± 25 ␮m, layer thicknesses of 13–130 ␮m and resolution of 101 ␮m in the X–Y plane. It operates at 18–24°C and can build at a linear speed of 310 mm s−1 [23]. The cost of a machine with a build chamber of 300 × 150 × 220 mm is about £60 000. It is intended to produce parts for downstream manufacturing and so offers a very high accuracy and low layer thickness. A similar system, BPM2, employs a head with 5 d.f. to deposit the material. This ensures that the direction of the jet is perpendicular to the normal of the surface and should eliminate steps in the build direction. The system uses a proprietary thermopolymer material to build models with a maximum size of 250 × 203 × 150 mm. A BPM2 machine will cost approximately £25 000. It has a resolution of 558 ␮m and an accuracy of ± 17 ␮m [24]. Another implementation of this technology, known as Multi Jet Modelling (MJM), employs 96 jets which scan each layer in a raster fashion. Parts are constructed from a thermopolymer material within a 250 × 200 × 200 mm build envelope. The parts have a layer thickness of 33 ␮m, an X–Y resolution of 85 ␮m and a droplet placement accuracy of ± 100 ␮m [25]. The cost of an MJM machine is around £50 000. The machine offers a high part creation speed and is intended primarily for model visualisation. 2.1.3.2. Fused deposition modelling (FDM) The FDM machine consists of a movable head which deposits a thread of molten material onto a substrate. The build material is heated to 0.5°C above its melting point so that it solidifies about 0.1 s after extrusion and cold welds to the previous layers (Fig. 11). Factors to be taken into consideration are the necessity for a steady nozzle speed and material extrusion rate, the addition of a support structure for overhanging parts, and the speed of the head which affects the overall layer thickness [15,26]. The latest FDM system includes two nozzles, one for the part material and one for the support

1270

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 11. Fused deposition modelling.

material. The latter is cheaper and breaks away from the prototype without impairing its surface. It is also possible to create horizontal supports to minimise material usage and build time [26,27]. An advantage of this system is that it may be viewed as a desktop prototyping facility in a design office since the materials it uses are cheap, non-toxic, non-smelly and environmentally safe. There is also a large range of colours and materials available, such as investment casting wax, ABS plastic, medical grade ABS (MABS) and elastomers. Parts made by this method have a high stability since they are not hygroscopic [26]. A disadvantage is that the surface finish of the parts is inferior to that produced using SL due to the resolution of the process which is dictated by the filament thickness [28]. It has not yet been demonstrated whether the material extrusion may be stopped quickly enough to produce small holes in vertical sections [9]. A typical commercially available machine is a stand alone system measuring 660 × 914 × 1067 mm which weighs 160 kg and operates at about 80°C. The build chamber in such a system measures 254 × 254 × 254 mm. The system costs around £100 000, deposits approximately 380 mm of material a second, produces layer thicknesses of 50–762 ␮m and has an accuracy of ± 127 ␮ [27]. 2.1.3.3. Three dimensional welding (3DW) This experimental system uses an arc-welding robot to deposit weld material on a platform as simple shapes which may then be built into more complex structures. Unlike most RP technologies, therefore, the prototypes are not built using sliced CAD files. Parts with a resolution of a few millimeters have been made which may be used for sandcasting or directly as tooling. Several problems still remain to be solved. Since there is no feedback, heat buildup during manufacture can cause the prototypes to melt and because the layers do not form a smooth surface

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1271

the torch may hit the part [11,29]. It is also not known whether complex structures can be built. Some method needs to be found to generate the robot program directly from the CAD file. The orientation of each section to be built should be generated as well as the order in which the sections are to be assembled. Another system which is being researched deposits the weld material in layers. Feedback control is established by the use of thermocouples which monitor the temperature and operate an on-line water cooling system. There is a grit blasting nozzle to minimise the oxidisation of the part and a suction pump and vacuum nozzle to remove excess water vapours and grit [18]. 2.1.3.4. Shape deposition manufacturing (SDM) This still experimental layer-by-layer process involves spraying molten metal in near net shape onto a substrate, then removing unwanted material via NC operations. Support material is added in the same way either before or after the prototype material depending on whether the layer contains undercut features (Fig. 12). The added material bolsters subsequent layers. If the layer is complex, support material may need to be

Fig. 12.

Shaped deposition manufacturing (adapted from [30]).

1272

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

added both before and after the prototype material. Each layer is then shot-peened to remove residual stresses. The prototype is transferred from station to station using a robotised pallet system which can position the workpiece to within an accuracy of ± 5 ␮m. Droplets of 1–3 mm diameter are deposited at a rate of 1–5 droplets per second. To date, stainless steel parts supported with copper have been produced. The copper may then be removed by immersion in nitric acid. These prototypes have the same structure as cast or welded parts with the accuracy of NC milling. Multiple materials may be employed and components can be embedded in the structure. As yet, no temperature control system for the substrate has been implemented, and the temperature, size and trajectory of the droplets are also not controlled [30]. 2.2. Processes involving discrete particles These processes build the part by joining powder grains together using either a laser or a separate binding material. 2.2.1. Fusing of particles by laser Selective Laser Sintering (SLS) is the main process in this category. With Gas Phase Deposition (GPD), the discrete grains are the result of the interaction between a reactive gas and a laser. However, the laser is also used to fix the grains with respect to the part. 2.2.1.1. Selective laser sintering (SLS) SLS uses a fine powder which is heated with a CO2 laser of power in the range of 25–50 W such that the surface tensions of the grains are overcome and they fuse together. Before the powder is sintered, the entire bed is heated to just below the melting point of the material in order to minimize thermal distortion and facilitate fusion to the previous layer [31]. Each layer is drawn on the powder bed using the laser to sinter the material. Then the bed is lowered and a powder-feed chamber raised. A new covering of powder is next spread by a counter-rotating roller. The sintered material forms the part whilst the unsintered powder remains in place to support the structure and may be cleaned away and recycled once the build is complete (Fig. 13). There is a large range of materials available for this process — basically any material which can be pulverised may be employed. At present, nylon, nylon composites, sand, wax, metals and polycarbonates are in use, and it is claimed that these materials have engineering grade properties [32]. They are cheaper than the resins used for SL, are non-toxic and safe and may be sintered with relatively low-powered lasers. However, parts need a long cooling cycle on the machine before they can be removed. For example, wax parts require 12 h to cool down. The materials employed by the system are sensitive to the different heating and laser parameters and each material requires distinct settings. These can be difficult and time-consuming to obtain. Parts may be finished by infiltration with molten metal to achieve 100% density. A drawback is that the recycled powders require sieving to ensure that no globules are present that would interfere with the smooth application of the next powder layer. The system also requires an inert nitrogen atmosphere in which to sinter the materials [32]. The least expensive machine which sinters thermoplastics costs around £250 000. The

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 13.

1273

Selective laser sintering (adapted from [2]).

maximum build chamber size is 330 × 380 × 425 mm. The layer thickness is 76 ␮m with an accuracy in the horizontal plane of ± 51 ␮m. The build speed is 12–25 mm h−1 [32]. A similar system, still under development, involves feeding powder through a nozzle onto the part bed whilst simultaneously fusing it with a laser. The powder nozzle may be on one side of the bed, or coaxial with the laser beam. If it is to the side, a constant orientation to the part creation direction must be maintained to prevent solidified sections from shadowing areas to be built. If the powder feeder is coaxial, there may be inaccuracies in the geometry of the part and the layer thickness if the beam and the powder feeder move out of alignment. The heating of the powder can lead to thermal distortion of the prototype. It is necessary to cool the part when it becomes too hot in order to prevent distortions in the final piece. An alternative would be to add a temperature control system. The minimum wall thickness depends on the feed rate and the width of the particle stream and the laser spot size, speed and power. Walls of 0.5–0.7 mm have been achieved [31]. 2.2.1.2. Gas phase deposition (GPD) In this process, the molecules of a reactive gas are decomposed using either light or heat to leave a solid. The solid result of the decomposition then adheres to the substrate to form the part (Fig. 14). Three slightly different methods of constructing the part are currently being researched. In the first, called SALD (Selective Area Laser Deposition), the solid component of the decom-

1274

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 14. Gas phase deposition.

posed gas is all that is used to form the part. It is possible to construct parts made from carbon, silicon, carbides and silicon nitrides in this way. The second method, SALDVI (Selective Area Laser Deposition Vapour Infiltration), spreads a thin covering of powder for each layer. Then the decomposed solids fill in the spaces between the grains. In the third method, SLRS (Selective Laser Reactive Sintering), the laser initiates a reaction between the gas and the layer of powder to form a solid part of silicon carbide or silicon nitride. A resolution of 1 ␮m is hoped for [11,33]. 2.2.2. Joining of particles with a binder 2.2.2.1. Three dimensional printing (3DP) Layers of powder are applied to a substrate then selectively joined using a binder sprayed through a nozzle (Fig. 15). In order to avoid excessive disturbance of the powder when it is hit by the binder, it is necessary to stabilise it first by misting with water droplets [34]. Once the part is completed, it is heated to set the binder then the excess powder, which was supporting the part, is removed by immersion in a water bath [35]. The part is next subjected to a final firing at 900°C for 2 h in order to sinter it [15]. It is possible to press the green part isostatically before this final firing to increase its density to over 99% of that of a solid part [36]. After firing, the part may be dipped in binder and refired so that its strength is improved. Since there is no state change involved in this process, distortion is reduced [28]. The resolution is dependent on the size of the binder droplets and the powder grains, the placement accuracy of the nozzle and the way that the binder diffuses through the powder due to capillary action. Neighbouring grains which have been wetted by a binder droplet are pulled together into a voxel of approximately spherical shape due to the surface tension. The entire voxel then shrinks as it dries [34]. The layer thickness is affected by the compression of the powder due to the weight of subsequent layers. This compression is most noticeable in the center of the part. At the base, there is no room to compact the powder. At the top of the part, there are fewer layers to cause the compaction. However this effect is mitigated when using more densely packed powders [34].

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1275

Fig. 15. Three dimensional printing.

Parts made using this process do not require supports to brace overhanging features. They do however need to include a hole so that excess powder can be removed [20]. Disadvantages of this technology are that the final parts may be fragile and porous, and it can be hard to remove the excess powder from any cavities. A further drawback is that the layers are raster-scanned by the printhead which leads to a stair-stepping effect in the X–Y plane as well as in the build direction [9]. The materials employed by 3DP are metal or ceramic powders, or metal–ceramic composites with colloidal silica or polymeric binders [20]. At present, this technology is available through a service bureau only and is used to create cast metal parts. A 3DP machine has a build chamber measuring 355 × 457 × 355 mm, a layer thickness of 177 ␮m, a resolution of 508 ␮m and an accuracy of ± 127 ␮m. The build speed is 18–25 mm h−1 [37]. A similar technology, known as Topographic Shape Formation (TSF) is used primarily for rapid production of moulds, which may then be used to create the prototype. The system prints paraffin wax about a centimeter below the surface of a silica powder. Once each layer has been completed, more powder is applied and the process is repeated. The wax binds the powder to form the part and also partially melts the previous layer to ensure good adhesion. Once the part is completed, it is sanded, coated in wax and then employed as a mould for the customer’s part. Materials in use include concrete, fibreglass and expanding foam.

1276

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

An advantage of this technology is that it can build very large parts quickly and cheaply, which may be expensive and time-consuming if constructed by other RP methods. A disadvantage is that the moulds have a ‘gritty’ surface finish and may need to be finished by an operator. At present, TSF is in use in a service bureau only. The machine has a maximum build envelope of 3353 × 1829 × 1219 mm, a layer thickness of 1270–3810 ␮m, resolution of 12 700 ␮m and an accuracy of ± 1270 ␮m [38]. 2.2.2.2. Spatial forming (SF) This technology is being developed for prototyping specialised medical equipment with metal. It is designed to produce high precision parts within a small build envelope of 2 × 2 × 300 mm. A negative of each layer is printed onto a ceramic substrate with a ceramic pigmented organic ‘ink’. The layer is then cured with UV light and the process repeated. After approximately 30 layers, the positive space left by the printing, which corresponds to the part cross section, is filled using another ‘ink’ which contains metal particles. This is then cured and milled flat. The process continues until the whole part is finished. Once the prototype is complete, it is heated in a nitrogen atmosphere to remove the binders in both the positive and negative ‘inks’ and to sinter the metal particles. The ceramic negative can then be removed in an ultrasonic bath to reveal the final piece, which is infiltrated with liquid metal to produce the metal prototype. The sintering process causes shrinkage of up to 20% in all directions which needs to be taken into account when designing the part. Further research includes optimizing the binder removal process and automating the addition of the positive material and the later milling [39]. A prototype of this system is currently being employed to construct preassembled microstructures for medical purposes. To date, no commercial system is available and only extruded parts with a constant cross-section can be produced. In theory, however, completely arbitrary geometries should be feasible. 2.3. Technologies which use a solid There are two different technologies which use solid foils to form the part. Laminated Object Manufacture (LOM) bonds the different sheets with an adhesive and then cuts the part contour using a laser. The second, Solid Foil Polymerisation (SFP), bonds sheets of foil by curing them with UV light. 2.3.1. Sheets bonded with adhesive: laminated object manufacture (LOM) The build material is applied to the part from a roll, then bonded to the previous layers using a hot roller which activates a heat-sensitive adhesive. The contour of each layer is cut with a laser that is carefully modulated to penetrate to a depth of exactly one layer thickness. Unwanted material is trimmed into rectangles to facilitate its later removal, but remains in place during the build to act as supports (Fig. 16). The sheet of material used is wider than the build area so that, once the part cross-section has been cut, the edges of the sheet remain intact. This means that, after the layer has been completed and the build platform lowered, the roll of material can be advanced by winding this excess onto a second roller until a fresh area of the sheet lies over the part. The whole process can then be repeated. The system employs a 25 or 50 W CO2 laser to cut the material. Smaller hatches must be used

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1277

Fig. 16. Laminated object manufacture (LOM1).

on up- and down-facing surfaces to facilitate the removal of waste material which has bonded to the part. It may also be necessary to stop the build to excavate paper from otherwise hard-toaccess places. Once the parts have been completed, they should be sealed with a urethane, silicon or epoxy spray if made of paper to prevent later distortion of the prototype due to water absorption. The height is measured and the cross-sections are calculated in real time to correct for any errors in the build direction [9]. Advantages of LOM include the wide range of relatively cheap materials available — parts may be made using paper for example, or from more expensive materials such as plastic or fiber reinforced glass ceramic. The parts may be quite large compared to those produced by other RP methods. Since they have the appearance of wooden pieces when finished, they are popular with model makers. Speed is another strong point of LOM. As only the outlines of the parts need to be traced, this method is about 5–10 times faster than other processes [40]. A drawback is the need to prise the finished parts off the table which adversely affects their

1278

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

surface finish. It is also hard to make hollow parts due to the difficulty in removing the core and there are serious problems with undercuts and re-entrant features. Other disadvantages of this technology are that there is a large amount of scrap, the machine must be constantly manned, parts need to be hand finished and the shear strength of the part is adversely affected by the layering of adhesive and foil [1,41]. Because the laser cuts through the material, there is a fire hazard which means that the machines need to be fitted with inert gas extinguishers. The drops of molten material (dross) which form during the cutting process also need to be removed [2]. The cost of a LOM machine is between £120 000 and £235 000 depending on the size of the build chamber. Available machines have a maximum build chamber of 813 × 559 × 508 mm. The minimum layer thickness that they can handle is 76–203 ␮m and their maximum accuracy is ± 127 ␮m. The maximum cutting speed achievable is 508 mm s−1 [40]. A similar process, LOM2, includes the ability to bond the sheets selectively to the part crosssection. Here, the cross-section of the part is printed onto a sheet of paper which is applied to the work-in-progress and bonded using a hot roller. A knife is then used to cut the outline of the part and cross-hatch the waste material. This process is repeated until the part is finished, when the excess material may be peeled away from the model. This can then be sealed with epoxy. Since a knife is used to cut the paper, this system should be less hazardous and cheaper than LOM1. The waste material is also easier to remove and so finer features may be built. A LOM2 machine costing approximately £130 000 has a build chamber of 400 × 280 × 300 mm. The system has a throughput of 1 sheet per minute. The parts have a layer thickness of 100 ␮m, X– Y resolution of 25 ␮m and an accuracy of ± 200 ␮m [42]. Another development which yields a low-cost machine involves using layers cut from adhesive material on backing paper or from foam laminating material. These layers are then assembled by hand using special positioning marks and the backing is removed. Once the prototype is completed, it may be coated to protect and strengthen it. This RP technology (LOM3) is perhaps one of the most inexpensive available, with machines costing approximately £8500 [43], although the finished parts are somewhat ‘tacky’ and the assembly process has to be performed manually. 2.3.2. Sheets bonded with UV light: solid foil polymerisation (SFP) In SFP, the part is built up using semi-polymerised foils. On exposure to UV light, the foil solidifies and bonds to the previous layer. It also becomes insoluble. Once the cross-section has been illuminated, a new foil can be applied. The areas of foil which do not constitute the eventual part are used to support it during the build process, but remain soluble and so are easy to remove. Once the part is complete, the non-bonded pieces can be dissolved to leave the finished part [2,44]. No commercial systems are available yet. 2.4. Material removal technology: desktop milling (DM) This is a process which removes material from the workpiece as in traditional machining processes instead of creating the part by gradual material buildup. The prototypes can be made with a high degree of accuracy because they do not deform after they have been completed. If NC machining is to be employed to manufacture the finished design, features which are difficult to create will also be detected at this stage. Any CNC machine may be employed to make prototypes from an inexpensive material such

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1279

as wax. There is a commercial entry-level desktop milling system available which is capable of dealing with STL files, even those which contain gaps and self-intersecting surfaces. This means that the designer does not need to spend time verifying and correcting the files, as is the case with the material accretion technologies. The machine is inexpensive and can handle a wide variety of materials. It generates the NC tool path automatically and may be operated with no NC training. The cost of the basic model is approximately £4500. Cutting speeds of 0.06–3.6 m s−1 can be obtained depending on the model purchased. The resolution can be as high as 10 ␮m for an inexpensive entry level machine, with an accuracy of ± 10 ␮m [45,46].

3. Applications There are many uses for RP. Unlike conventional prototypes which may take a skilled artisan weeks or months to produce, RP parts may be made cheaply by a machine in a few days or less, with little human intervention. Therefore the designer may prototype the part as often as necessary to check for appearance and function. Changes may then be easily incorporated into the model and another prototype generated. This facilitates the optimisation of the design and saves timeconsuming and expensive alterations at a later production date. There are many other applications for the prototyped parts which would have been impractical with conventional models. Some of these applications are listed below. 3.1. Visualisation 쐌 Parts may be employed to facilitate communication of ideas in a concurrent engineering environment. 쐌 Some companies now routinely include a prototype made from the CAD file with their sales proposal to allow the customer to see and assess the part [27]. 쐌 Complex models may be produced for teaching purposes [47]. 3.2. Working models/functional parts 쐌 Small batches of plastic parts can be commercially manufactured. Because patterns for injection moulding are expensive to produce, the break-even point for a production run is a few thousand parts [2]. RP technologies can be used on their own or in conjunction with other more conventional technologies to manufacture parts in quantities as low as one. 쐌 Parts may be produced with intricate internal shapes that could not be manufactured using traditional technologies. Examples include medical equipment such as the interlocking tip assembly for a catheter system to investigate arteries [39] and monolithic ceramic filters [48]. 쐌 One-of-a-kind parts such as bone replacements may be made accurately from a scanned model of the original. The bone may be imaged using X-ray tomography and the data translated to a CAD file which is then used to drive the RP process [47].

1280

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

쐌 Parts can be produced with well-defined microstructures by using technologies which can deposit different materials [47]. 3.3. Tooling/manufacturing pattern

쐌 It is possible to employ RP parts directly as tooling. SLS or ceramic 3DP parts may be infiltrated with liquid metal to produce a dense tool with a well-defined distribution of ceramic or metal particles [19,37]. RP models may be sprayed with metal to produce EDM electrodes which may be used to manufacture up to 1000 parts [9,11]. TSF parts may be used as moulds for concrete, fibreglass or expanding foams [38]. 쐌 Parts made by RP may be used to produce tools indirectly. Tooling lead-times may be reduced from 12–26 weeks to 1–6 weeks. Parts made of wax or other low melting point materials may be sprayed with metal and the wax subsequently removed by melting. The metal shells may then be employed for plastic injection moulding [20]. 쐌 Parts made with a low-melting point material may be used for investment casting purposes. The parts are coated with a ceramic slurry and then burnt out. As mentioned previously, SL parts should be built using a draw-style such as QuickCast to avoid cracking the ceramic moulds. The FDM, BPM and SLS investment casting waxes burn out, leaving little to no ash content (⬍ 0.002%), and therefore are ideal for investment casting. LOM parts made of paper may be burnt out at 760°C leaving approximately 3% in ash [40]. When adopting 3DP, the ceramic moulds may be made directly, which has the effect of tightening tolerances as there are fewer shape transfers. It is also possible to produce moulds with integral cores. This means that they do not have to be manually located and again tolerances are tightened. Another possibility is to print the cores in a different material so that they are easy to remove at a later date [20]. An advantage of these RP technologies is that the expensive conventional tooling used to produce the mould which makes the sacrificial wax patterns is not needed to create the prototype, allowing multiple trials before the design is finalised [1,49]. 쐌 SL, SLS and LOM prototypes may be used in the sand casting process for short runs of cast parts [1].

4. Selection of RP processes Tables 1 and 2 contrast the main features of the different RP systems. The technologies are split into those which are commercially available and those which are still being researched. There are alternative systems listed under each of the categories of BPM and LOM and data for these alternatives have been included in the table. LOM1 is the fully automatic LOM process, employing a laser, LOM2 is the selective bonding process which uses a knife and LOM3 is the manual assembly process. As described previously, the BPM processes are the dual-jet BPM1, the 5axis BPM2 machine and the multi-jet MJM. The figures for DM refer to the entry level system mentioned earlier.

Table 1 Features of rapid prototyping processes (commercial) SGC

BPM1

Postcuring Yes required

No

No

Supports required Material used

No

Yes

Yes

Epoxy or Resin acrylic resin

Laser Yes used Layer 50 thickness (␮m) X–Y 200–250 Resolution (␮m) Accuracy ± 100 (␮m) Scan N/A speed (mm s⫺1) Time to 113 (50 × complete 50 mm) a layer (s) Maximum 500 × part 500 × dimensions584 (mm3) Cost 150–390 (£1000)

1

BPM2

No

Thermo- Thermoplastic or polymer wax

MJM

FDM

SLS

No

LOM1

No (firing Yes may be required) No No

No

No

No

No

No

No

Sand and Paper, Paper wax plastic or ceramic

Paper or foam

Various

No

Nylon, Ceramic metals, or metal wax, or poly carbonate Yes No

No

Yes

No

No

No N/A

Yes

Thermopolymer

ABS, MABS, wax or elastomers

LOM3

No

No

100–200

13–130

Not available

33

50–762

76

177

1270– 3810

76–203

100

110–140

100

101

558

85

254

Not available

508

12 700

203–254

25

Not 10 available

± 500

± 25

N/A

310

65

± 17

± 100

± 127

± 51

± 127

6200

380

0.001– 0.008

0.007

N/A

12 000 particles per second N/A

N/A

N/A

N/A

N/A

500 × 350 × 500

300 × 150 × 220

250 × 203 × 150

250 × 200 × 200

254 × 254 × 254

330 × 380 × 425

180–300

60

25

50

100

250–365

Not available Not available

± 127

± 200

Not ± 10 available Not 60 available

508 (cutting speed)

N/A

Not available

N/A

60

355 × 457 × 355

3353 × 1829 × 1219

813 × 559 400 × 280 610 × × 508 × 300 6101

120 × 100 × 120

Bureau service only

Bureau service only

120–235

4.5

Since prototypes made with the LOM3 system are assembled manually there is no height constraint. These figures refer to an entry-level system only.

2

LOM2

DM2

TSF

Yes

3DP

130

N/A

8.5

N/A

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

SL

1281

1282

Table 2 Features of rapid prototyping processes (non-commercial)

Postcuring Yes required Supports Yes required Material used Resin Laser used Layer thickness (␮m) X–Y Resolution (␮m Accuracy (␮m) Scan speed (mm s ⫺ 1) Time to complete a layer (s) Maximum part dimensions (mm3) Cost (£1000)

Yes 100

100

BIS

HIS

3DW

SDM

GPD

Yes

No

No

No

Yes

No

No

Yes

Yes

No

Resin

Resin

Weld beads Metal

Yes

Yes N/A

No 1450

300–600 ± 500

No

SF

SFP No

No

No

No

Reactive gas Metal

Resin foils

Yes

Yes

Electro-set fluid No

No 0.5

10 ± 25

8

300 × 300 × 300 × 300 × 300 300

ES

2 × 2 × 300

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

LTP

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1283

The accuracy data in the tables was obtained from technical publications and from company literature. In the main, these represent the best accuracies achievable with finely tuned equipment when operated by a skilled technician and not the average accuracies and resolutions achieved by the users. The layer thickness shown in the table is taken to be equivalent to the z resolution of the part. Depending on whether the part is built point by point or layer by layer, either the linear build rate is quoted, or the time to complete a layer is given. Of the systems listed, the most accurate is the dual-jet BPM1 machine. However, the build chamber and therefore the maximum part size is small. The cheapest systems are the LOM3 machine and the entry-level DM system. As already mentioned, the drawback of the LOM3 system is that the parts produced are ‘tacky’ and need to be assembled manually. Disadvantages of the low-cost DM machine are that its work envelope is small and it cannot manufacture shapes as complex as those created using the material accretion technologies. There was less information available for the non-commercial processes and for some technologies no accurate figures could be obtained. Figure 17 is a quick guide to selecting RP processes. The selection is based on the end use of the part, part size, whether or not all features may be freely accessed, whether or not the part is hollow, part accuracy and part strength. For completeness, approximate capital and running cost information is provided on each process and this is then used to rank the different alternatives. Only commercially available processes are represented.

5. Conclusion Rapid prototyping is an enabling technology for concurrent engineering. Its goal is to reduce product development and manufacturing costs and lead times, thereby increasing competitiveness. Impressive steps towards that goal have been made. However, the field of RP is still new, with much effort to be expended on improving the speed, accuracy and reliability of RP systems and widen the range of materials for prototype construction. Another area of improvement will be costing, as most RP systems are currently too expensive to be affordable by any but the larger firms. Although RP technology will continue to be available to all companies via bureaux which, often in partnership with traditional model makers, can provide a comprehensive service from design through to short-run production, the future is likely to see more user-owned RP machines as their costs are reduced. There will also be two different types of RP systems for two distinct markets: the design-office ‘3D-plotter’ for rapidly generating parts for design verification and the workshop/model-making shop machine for producing accurate functional parts and tooling.

Acknowledgement This work was supported by the European Regional Development Fund which is administered by the Welsh Office for the European Commission.

1284

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

Fig. 17. RP process selection guide.

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1285

Fig. 17. Continued

References [1] N.A. Waterman, P. Dickens, Rapid product development in the USA, Europe and Japan, World Class Design To Manufacture 1 (3) (1994) 27–36. [2] J.P. Kruth, Material incress manufacturing by rapid prototyping technologies, CIRP Annals 40 (2) (1991) 603–614. [3] A. Dolenc, I. Ma¨kela¨, Slicing procedures for layer manufacturing techniques, Computer-Aided Design 26 (2) (1994) 119–126.

1286

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

[4] D. Frank, G. Fadel, Preferred direction of build for rapid prototyping processes, 5th International Conference on Rapid Prototyping, Dayton, Ohio, 12–15 June, 1994, pp. 191–201. [5] W. Cheng, J.Y.H. Fuh, A.Y.C. Nee, Y.S. Wong, H.T. Loh, T. Miyazawa, Multi-objective optimization of partbuilding orientation in stereolithography, Rapid Prototyping Journal 1 (4) (1995) 12–23. [6] P.F. Jacobs, Fundamentals of stereolithography, First European Conference on Rapid Prototyping, Nottingham, July 1992, pp. 1–17. [7] K. Renap, J.P. Kruth, Recoating issues in stereolithography, Rapid Prototyping Journal 1 (3) (1995) 4–16. [8] 3D Systems, Maestro Workstation User Guide, 3D Systems, Worldwide Corporation HQ, 26081 Avenue Hall, Valencia, CA, 1996. [9] P.F. Jacobs, Stereolithography and Other RP and M Techniques, ASME Press, New York, 1996. [10] P.F. Jacobs, QuickCast 1.1 and rapid tooling, 4th European Conference on Rapid Prototyping and Manufacturing, Nottingham, 13–15 June 1995, pp. 1–27. [11] P.M. Dickens, Research developments in rapid prototyping, Proceedings IMechE, Journal of Engineering Manufacture, Part B, 1995, vol. 209, pp. 261–266. [12] R. Ippolito, L. Iuliano, A. Gatto, Benchmarking of rapid prototyping techniques in terms of dimensional accuracy and surface finish, CIRP Annals 44 (1) (1995) 157–160. [13] 3D Systems, The Edge, vol. V, 1, 3D Systems, Worldwide Corporation HQ, 26081 Avenue Hall, Valencia, CA, 1996. [14] 3D Systems, SLA-500 Series, 3D Systems, Worldwide Corporation HQ, 26081 Avenue Hall, Valencia, CA, 1996. [15] S. Au, P.K. Wright, A comparative study of rapid prototyping technology, Proceedings ASME Winter Conference, New Orleans, November 1993, vol. 66, pp. 73–82. [16] Cubital Ltd, Advantages of the Solider System, Cubital Ltd., 13 Hasadna St., PO Box 2375, Industrial Zone North, Raanana, 43650 Israel, 1996. [17] Cubital Ltd, Solider 5600, Cubital Ltd., 13 Hasadna St., PO Box 2375, Industrial Zone North, Raanana, 43650 Israel, 1996. [18] Anon, State of the Art Review-93-01, MTIAC, 10 West 35 Street, Chicago, IL 60616, U.S.A., 1993. [19] Anon, Manufacturing parts drop by drop, Compressed Air, March 100 (2) (1995) 38–44. [20] E. Sachs, M. Cima, P. Williams, D. Brancazio, J. Cornie, Three dimensional printing: rapid tooling and prototyping directly from a CAD model, Transactions of ASME: Journal of Engineering for Industry 114 (1992) 481–488. [21] Rayleigh, On the instability of jets, Proceedings of the London Mathematical Society 10 (4) (1878) 4–13. [22] M. Orme, K. Willis, J. Courter, The development of rapid prototyping of metallic components via ultra uniform droplet deposition, Proceedings of the 5th International Conference on Rapid Prototyping, Dayton, Ohio, 12–15 June 1994, pp. 27–37. [23] Sanders Prototype Inc., ModelMakerII, Sanders Prototype Inc., PO Box 540, Wilton, New Hampshire 03086, U.S.A., 1996. [24] BPM Technology Inc., The Personal Modeler 2100 Printer, BPM Technology Inc., 1200 Woodruff Road, A19 Greenville, SC 29607, 1997. [25] 3D Systems, Actua 2100, 3D Systems, Worldwide Corporation HQ, 26081 Avenue Hall, Valencia, CA, 1996. [26] Stratasys Inc., Fused deposition modelling for fast, safe plastic models, 12th Annual Conference on Computer Graphics, Chicago, April 1991, pp. 326–332. [27] Stratasys Inc., FDM-1650, Stratasys Inc., 14950 Martin Drive, Eden Prairie, Minneapolis 55344-2020, U.S.A., 1996. [28] B. Bidanda, V. Narayanan, R. Billo, Reverse engineering and rapid prototyping, in: R.C. Dorf, A. Kusiak (Eds.), Handbook of Design, Manufacture and Automation, Wiley, NY, 1991, pp. 977–991. [29] P.M. Dickens, M.S. Pridham, R.C. Cobb, I. Gibson, Rapid prototyping using 3-D welding, Proceedings of the 3rd Symposium on Solid Freeform Fabrication, Austin, Texas, September 1992, pp. 280–290. [30] R. Merz, F.B. Prinz, K. Ramaswami, M. Terk, L.F. Weiss, Shape deposition manufacturing, Proceedings of the 5th Symposium on Solid Freeform Fabrication, Austin, Texas, 8–10 August 1994, pp. 1–8. [31] F. Klocke, T. Celiker, Y.-A. Song, Rapid metal tooling, Rapid Prototyping Journal 1 (3) (1995) 32–42. [32] DTM Corporation, A platform that supports the entire design process, DTM Corporation, 1611 Headway Circle, Building 2, Austin, Texas 78754, 1996. [33] Laboratory for Freeform Fabrication, Web pages of University of Texas at Austin, Texas, U.S.A., 1996.

D.T. Pham, R.S. Gault/ International Journal of Machine Tools & Manufacture 38 (1998) 1257–1287

1287

[34] E. Sachs, J. Cornie, D. Brancazio, J. Bredt, A. Curodeau, T. Fan, S. Khanuja, A. Lauder, J. Lee, S. Michaels, Three dimensional printing: the physics and implications of additive manufacturing, CIRP Annals 42 (1) (1993) 257–260. [35] E. Sachs, E. Wylonis, M. Cima, S. Allen, S. Michaels, E. Sun, H. Tang, H. Guo, Injection moulding tooling by three dimensional printing, a desktop manufacturing process, Proceedings 53rd Annual Conference ANTEC, Boston, May 1995, vol. 1(1), pp. 997–1003. [36] J. Yoo, M. Cima, E. Sachs, S. Suresh, Fabrication and microstructural control of advanced ceramic components by three dimensional printing, Proceedings of the American Ceramic Society, Jan Cocoa Beach, FL, 1995. [37] Soligen Technologies, Private communication, Soligen Technologies, 19408 Londelius Street, Northridge, CA 91324, 1995. [38] Formus Ltd, Topographic shape fabrication, Formus Ltd, 185 Lewis Road, Suite 31, San Jose, CA 95111, 1997. [39] C.S. Taylor, P. Cherkas, H. Hampton, J.J. Frantzen, B.O. Shah, W.B. Tiffany, L. Nanis, P. Booker, A. Sabhieh, R. Hansen, Spatial forming, a three dimensional printing process, Proceedings IEEE Micro Electro Mechanical Systems Conference, Amsterdam, January 1995, pp. 203–208. [40] Helisys Inc., 2030H System, Helisys Inc., 24015 Garnier Street, Torrance, CA 90505, 1997. [41] S.S. Crump, Fast precise, safe prototypes with FDM, ASME Annual Winter Conference, Atlanta, December 1991, vol. 50, pp. 53–60. [42] YUASA Warwick Machinery Ltd, Private communication, YUASA Warwick Machinery Ltd., Rothwell Road, Wedgenock Ind. Est., Warwick CV24 5PY, U.K., 1997. [43] Schroff Development Corporation, JP System 5, Schroff Development Corporation, PO Box 1334, Mission, Kansas 66205, USA, 1996. [44] S. Corbel, A.L. Allanic, P. Schaeffer, J.C. Andre, Computer-aided manufacture of three-dimensional objects by laser space-resolved photopolymerization, Journal of Intelligent and Robotic Systems 9 (1994) 310–312. [45] Delft Spline Systems, Desk proto, offering in-house rapid prototyping, Delft Spline Systems, PO Box 2171, 3500 GB Utrecht, The Netherlands, 1996. [46] Delft Spline Systems, CNC-500, Delft Spline Systems, PO Box 2171, 3500 GB Utrecht, The Netherlands, 1996. [47] A.P. Nyaluke, D. An, H.R. Leep, H.R. Parasaei, Rapid prototyping in academic institutions and industry, Computer and Industrial Engineering 29 (1995) 345–349. [48] M. Parish, A.B. Jettery, Ceramic filter elements with tailored macro and microstructures, Filtration and Separation 32 (1) (1995) 31–36. [49] C.C. Kai, 3D rapid prototyping technologies and key development areas, Computing and Control Engineering Journal August (1994) 200–206.