BY SARA MCMAINS Layered Manufactu Technologies They are transforming one-off prototyping and mass customization of complex 3D parts directly from computer-aided design models. F rom the holodeck in Star Trek to the slice-by-slice reconstruction of the alien Leeloo played by actress Milla Jovovich in the movie The Fifth Element, science fiction has long predicted the possibility of automated one-off construction of the most complex objects from a stream of digital information. While we are still a long way from being able to build living beings in this manner, digitally controlled layered manufacturing (LM) technologies are transforming prototyping and mass customization of mechanical and biomedical parts of extraordinary geometric complexity. 50 June 2005/Vol. 48, No. 6 COMMUNICATIONS OF THE ACM
SPRINKLER PROTOTYPE (ABS PLASTIC, THE KIND U S E D I N L E G O B R I C K S ) C R E AT E D THROUGH FUSED DEPOSITION MODELING. ( S T R ATA S Y S, I N C., EDEN PRAIRIE, MN) ring COMMUNICATIONS OF THE ACM June 2005/Vol. 48, No. 6 51
Traditional mass-production manufacturing methods involve very high initial start-up costs and long lead times before the first part is produced. For example, a plastic injection-molded part requires a metal mold that typically costs from tens of thousands to hundreds of thousands of dollars and takes weeks to plan and machine. Alternatively, to produce a metal part using numerically controlled (NC) machining, each part feature must be programmed and machined sequentially, and each operation might require custom jigs and fixtures to position, immobilize, and support the work in process. This initial investment is made up for by low marginal costs and short cycle times to produce each subsequent part, because the molds, jigs, fixtures, and/or NC code are reusable many times over. Making prototypes or custom parts is traditionally a time- and labor-intensive process, without the production volumes that allow amortizing the start-up costs over large production runs. LM (also known as solid freeform fabrication or rapid prototyping) has revolutionized one-off production by automating the planning aspects of the manufacturing process, thereby making possible quick and economic prototyping of complex 3D parts directly from computeraided design (CAD) solid models. In all LM processes, the 3D CAD model is sliced into horizontal layers of uniform thickness. Each cross-sectional layer is successively deposited, hardened, fused, or cut, depending on the particular process, and attached to the layer beneath it. The stacked layers form the final part. Here, I describe three popular LM technologies, comparing their cost, speed, accuracy, and capabilities. STEREOLITHOGRAPHY 3D Systems StereoLithography Apparatus (SLA) was the first commercially available LM technology, introduced in 1988, and is still probably the best known (www.3dsystems.com). It employs an ultraviolet laser to harden a light-sensitive liquid resin photopolymer; the original brittle acrylate polymers are now largely replaced by somewhat tougher epoxies [4]. The first step in stereolithography, as in all LM processes, is to generate a 3D CAD solid model of the part. The CAD file must then be translated into a triangulated boundary representation format (known as STL, for stereolithography [6]) understood by the SLA machine and transferred to the SLA s computer. There, support structures must be designed to connect the part to the support platform during all phases of building, so it won t float away and to prevent cantilevered features from sagging downward or curling upward. The SLA software then slices the STL files for the part and its supports, generating cross sections describing the contours of the horizontal layers in the SLC (slice) format. This slice contour format is also used to import slice geometry directly into the system from third-party software applications (such as for slicing curved surfaces without first converting them to a triangulated approximation). Figure 1. Schematic of the main components of a stereolithography apparatus from 3D Systems, showing a dark layer of hardened resin at the top level of the liquid and additional hardened geometry and its supports in the vat. The main components of the SLA machine are a vat of liquid photopolymer, an elevator-support platform, and a mirror-deflected UV laser (see Figure 1). The part is built from the bottom layer up, starting with the elevator one layer-thickness below the surface of the liquid resin. The laser beam, its position controlled by the deflecting mirrors, scans the outline of the new layer s cross section. The laser initiates a chemical reaction in the photopolymer, causing it to gel in the exposed area; the laser-scanning speed is calibrated so the polymer cures just deeply enough to adhere to the previous layer. The elevator is then lowered one layer-thickness, and the process is repeated to build up the entire part, which is then removed from the vat, cleaned in a solvent bath, and post-cured in a UV curing oven. Manual post-processing is required to remove the supports and sand and polish the part. SLA parts are also widely used in molding and casting processes for limited production runs [2]. 52 June 2005/Vol. 48, No. 6 COMMUNICATIONS OF THE ACM
POWDER-BASED 3D PRINTING With 3D printing (3DP) technology, originally developed at MIT under the direction of Emanuel M. Sachs in the late 1980s, a powdered material is distributed one thin layer at a time and selectively hardened and joined together by depositing drops of binder from an inkjet print head [8]. For each layer, a powder hopper and roller system distributes a thin layer of powder over the top of the work tray. The inkjet nozzles then apply binder in parallel during a backand-forth scan of the entire work area, selectively hardening the part s cross section. A piston then lowers the part so the next layer of powder can be applied. The loose powder that isn t hardened remains, acting as a Figure 2. Finite element analysis visualization printed on a color 3D printer. (Z Corporation) support for subsequent layers. After the part is built up, the entire tray can be dried in an oven or fired in a kiln; loose powder is removed with brushes and compressed air. Finished parts can be impregnated with various materials to make them smoother, stronger, or more flexible. The 3DP process is more flexible in terms of its choice of materials than stereolithography, since any combination of a powdered material with a binder with viscosity low enough to form droplets is usable. A potential disadvantage is that the parts are always porous due to density limitations on the distribution of dry powder. 3DP technology was first commercialized in 1993 by Soligen Technologies, Inc. (www.soligen.com) using ceramics. The main focus with ceramics is on producing sacrificial ceramic molds for casting metal. Diverse metals, including copper, bronze, aluminum, cobalt chrome, stainless steel, and tooling steel, have all been cast in the ceramic shells produced through this process. Z Corporation (www.zcorp.com) sells a popular line of 3DP machines that build parts from starch or plaster. For these systems, a full-color option using a color-inkjet print head produces a dazzling variety of colors impossible to achieve with any other system (see Figure 2). THERMOPLASTIC FUSED DEPOSITION MODELING In fused deposition modeling (FDM), developed by Stratasys, Inc. (www.stratasys.com) and introduced in 1993, the part is built up in layers formed by extruding a melted plastic available in many colors. Inside the FDM head, the plastic is heated to just above its melting temperature, then extruded through a nozzle. Controlled by an NC tool path, the head moves to trace out the cross section of the layer. Faster build techniques trade off strength for speed [5, 7]. The melted material adheres to the build platform for the first layer or to the previous layer, hardening in less than a second. After each layer is deposited, an elevator adjusts the distance between the platform and the FDM head so the next layer can be deposited on top of the previous layers. Support structures are important for thermoplastic systems, since neither liquid photopolymer nor a powder bed supports the top layer being deposited; supports are required wherever the top layer extends more than minimally over the profile of the previous layer. The FDM software generates dense, corrugated support structures for these areas, balancing strength and ease of removal. To improve ease of removal, FDM systems use a second extrusion head to deposit a different support material. Depending on the system, this second material is either a plastic that bonds weakly with the primary material or a water-soluble support that can be dissolved away in an ultrasonic cleaning bath. In either case, removing the supports is time- and/or labor-intensive for complex geometries (such as those in Figure 3). FDM systems require no post-curing, throw no messy powders into the air, are fairly safe because the materials are nontoxic, and require no lasers. As a result, FDM is one of the few technologies that can be used in an office environment for desktop manufacturing. The plastic produces sturdy prototypes, compared to brittle SLA parts and fragile 3DP parts. APPROACHES COMPARED In addition to the materials they process, these technologies can be compared in terms of cost, speed, and range and precision of the geometries they produce. COMMUNICATIONS OF THE ACM June 2005/Vol. 48, No. 6 53
a Figure 3. Two views of a fused deposition model of a 3D object after artist M.C. Escher s Belvedere etching: (a) impossible view; (b) alternate view. (Gershon Elber, Technion Israel Institute of Technology [3]) Cost. Operating LM equipment is expensive, requiring a minimum of about $30,000 for an entrylevel machine from Z Corporation or Stratasys, plus a computer, maintenance, materials, and in some cases, ventilation. High-end SLA systems sell for hundreds of thousands of dollars. In general, thermoplastic systems are cheaper to maintain because lasers require frequent and costly replacement (typically a new $10,000 laser once or twice per year), and powder dust creates a hostile environment for mechanical and electrical components. Thermoplastic systems also tend to be cheaper to install, since SLA should be installed in a ventilated lab, and 3DP generates excessive dust without a dust-collection system. Consumable costs are on the order of a dollar per cubic inch of consumable material, while recyclability varies across systems depending on the materials being used. With 3DP, unhardened powder is carefully brushed off the part, then strained and reused when mixed with 50% new powder. Entire FDM parts and supports could theoretically be ground up and reused as a portion of the raw material to create partially recycled filament. In SLA systems, uncured resin left over in the vat can be reused. Per-part material costs for Z Corporation 3D printers vary enormously, depending how tightly the build volume is packed, since the surrounding powder is also degraded, and a significant amount of binder might be used up flushing the print heads after each layer. For SLA and thermoplastic approaches, though, the cost of the material used in support structures must also be considered. However, if you aren t ready to buy an LM machine, a number of service bureaus will build parts from your CAD files for prices ranging from a hundred to thousands of dollars, depending on part size and complexity. Speed. 3DP tends to be the fastest of the three technologies covered here, since its print heads contain hundreds of tiny nozzles that deposit binder in parallel, though large parts may still take hours to print. For small parts, FDM can achieve speeds of two vertical inches/hour using thick layers, but larger parts with thinner layers might take days to complete. With SLA, increasing the layer thickness eventually decreases speed, because the scan rate must be reduced significantly for the laser to penetrate deeply enough to cure the liquid to a sufficient depth; laser b 54 June 2005/Vol. 48, No. 6 COMMUNICATIONS OF THE ACM
power is the most important factor in build speed for these systems. In addition to machine time during the build, pre- and post-processing operations can add significantly to the total time to produce parts. Preprocessing can involve creating support structures, processing the input files (including merging and slicing multiple part files), and setting up the machine. Post-processing the finished part might involve removing excess material in the form of unhardened powder or uncured resin, post-curing, removing support structures, and sanding the part. When comparing quotes from service bureaus, determine how much optional post-processing (to improve the surface finish) is included in the price. Resolution, accuracy, surface finish. A process s resolution the minimum addressable distance between part features should be interpreted only as a lower bound on actual geometric error (tolerance). Resolution is usually coarsest along the Z axis, where it is the layer height; stereolithography is capable of.001-inch to.004-inch or thicker layers; 3DP.003- inch to.01-inch; and FDM.005-inch to.01-inch. If you are buying parts through a service bureau, you may get a better price if you are willing to accept thicker slices. Accuracy in reproducing the geometry of the CAD model depends on the size and shape of the part, the axis of measurement, and how recently the machine was calibrated. Axis-aligned rectilinear parts tend to be the most accurate. With 3D printing, the weight of the binder fusing large upper layers can deform smaller lower layers or even cause them to collapse, making for poor accuracy in such cases. SLA systems have a reputation for the best accuracy overall, due to their thin layers and years of related research into controlling shrinkage and warping [4]. The best accuracy and surface finish are often along different axes, so these two considerations must be traded off when choosing a build orientation. An experienced operator often produces a far more accurate part than a novice by adjusting build parameters based on a better understanding of the idiosyncrasies of the process. Resolution is but one indicator of surface roughness. Any process requiring attached external support structures has the potential for surface flaws where the structures are broken off. For processes using powdered materials, surface finish can be worse than resolution alone would suggest, depending on the extent to which the powders retain their shape in the process. One result is the widespread use of dipped coatings One advantage of many LM processes over traditional prototyping is the ease of making complicated shapes with relatively inaccessible interior chambers (such as ship-in-a-bottle-type geometries). COMMUNICATIONS OF THE ACM June 2005/Vol. 48, No. 6 55
during post-processing of parts made from powder. Dipped coatings improve the surface finish but potentially at the expense of accuracy if the thickness and behavior of the coating process are not modeled and compensated for when building the original part. Part geometry. Part geometry capabilities are limited by minimum feature size (typically about twice the layer height), the build area of the system, support requirements, and how readily support material can be removed. Entry-level LM systems typically have a build area slightly less than one cubic foot; more expensive SLA systems provide larger build areas up to about two feet per side. Software is available for segmenting parts too large to be built in one piece into smaller components to be assembled after building. One advantage of many LM processes over traditional prototyping is the ease of making complicated shapes with relatively inaccessible interior chambers (such as ship-in-a-bottletype geometries). Completely enclosed voids cannot be manufactured; a small outlet is always needed to remove supports. For SLA and FDM with breakaway supports, a larger access hole is needed, unlike the relatively small outlet required for removing powder or water-soluble supports. Easy-to-remove supports even allow the building of fully assembled mechanisms (such as interlocking gears). CONCLUSION If you have a CAD file you are ready to turn into an LM part, consider the following issues when choosing an LM technology. Stereolithography machines typically produce some of the smoothest and most accurate parts. Thermoplastic parts are sturdy, and the systems that build them are relatively easy to use and cost less to install and maintain. Parts made from powder involve the easiest support removal. And 3DP offers the only full-color models. Many other LM technologies, beside the ones described here, are also available. For example, a more expensive powder-based technology selective laser sintering (SLS) uses a laser instead of a binder to fuse the powder [1]. Commercial SLS systems can make parts directly (without a mold) from a greater variety of materials, including nylon, investment casting wax, polycarbonate, metal composites, or a flexible, rubber-like elastomer, than other LM systems. Yet other LM technologies can produce parts with material properties ranging from plywood (laminated object manufacturing, or LOM) to metal (laser-engineered net shaping, or LENS). A number of companies have developed LM machines that use many jets in parallel to deposit wax, thermoplastics, UV plastics, and photopolymers. For links to virtually all LM equipment manufacturers, go to: www.wohlersassociates.com and www.cc.utah.edu/asn8200/rapid.html. Most LM systems are used for creating prototypes and conceptual models, but, as the technology matures, more and more functional parts are likely to be manufactured with LM. Someday, instead of waiting to get a replacement for a broken part to be shipped to a remote location, technicians will just download its CAD model and 3D print it, just like we download and print instruction manuals today. Combined with 3D body-scanning technology, LM is already being used to automatically custom-manufacture products (such as orthodontia and hearing aids) fit to the shape of an individual customer or patient. The ability to make customized products with fast turnaround times might even reverse the current trend throughout U.S. industry toward offshore manufacturing. c References 1. Beaman, J., Marcus, H., Bourell, D., Barlow, J., Crawford, R., and McAlea, K. Solid Freeform Fabrication: A New Direction in Manufacturing. Kluwer Academic Publishers, Dordrecht, Germany, 1997. 2. Chen, Y. and Rosen, D. A reverse glue approach to automated construction of multipiece molds. J. Comput. Inform. Sci. in Engineer. 3, 3 (Sept. 2003), 219 230. 3. Elber, G. Escher for Real. Technion Israel Institute of Technology, Haifa, Israel, 2005; www.cs.technion.ac.il/~gershon/escherforreal/. 4. Jacobs, P. Rapid Prototyping and Manufacturing: Fundamentals of Stereolithography. Society of Manufacturing Engineers, Dearborn, MI, 1992. 5. McMains, S., Smith, J., and Séquin, C. Thin-wall calculation for layered manufacturing. ASME J. Comput. Inform. Sci. Engineer. 3, 3 (Sept. 2003), 210 218. 6. McMains, S., Smith, J., and Séquin, C. The evolution of a layered manufacturing interchange format. In Proceedings of the 28th Design Automation Conference (Montréal, Sept. 29 Oct. 2). ASME Design Engineering Technical Conferences, New York, 2002, 945 953. 7. McMains, S. Double-sided layered manufacturing. In Proceedings of the 2002 Japan-USA Symposium on Flexible Automation (Hiroshima, July 14 19). Institute of Systems, Control, and Information Engineers, Kyoto, 2002, 269 272. 8. Sachs, E., Williams, P., Brancazio, D., Cima, M., and Kremmin, K. Three-dimensional printing: Rapid tooling and prototypes directly from a CAD model. In Proceedings of Manufacturing International 1990 (Atlanta, GA, Mar. 25 28). ASME, New York, 1990, 131 136. Sara McMains (mcmains@me.berkeley.edu) is an assistant professor in the Department of Mechanical Engineering at the University of California, Berkeley. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. 2005 ACM 0001-0782/05/0600 $5.00 56 June 2005/Vol. 48, No. 6 COMMUNICATIONS OF THE ACM