UC Berkeley UC Berkeley Electronic Theses and Dissertations
Title Multiphysics Modeling of Selective Laser Sintering/Melting
Permalink https://escholarship.org/uc/item/6gt2q327
Author Ganeriwala, Rishi
Publication Date 2015
Peer reviewed|Thesis/dissertation
eScholarship.org Powered by the California Digital Library University of California Multiphysics Modeling of Selective Laser Sintering/Melting
by
Rishi Kumar Ganeriwala
A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy
in
Engineering - Mechanical Engineering
and the Designated Emphasis
in
Energy Science and Technology
in the
Graduate Division of the University of California, Berkeley
Committee in charge: Professor Tarek I. Zohdi, Chair Professor Daniel Kammen Professor Hayden Taylor
Fall 2015 Multiphysics Modeling of Selective Laser Sintering/Melting
Copyright 2015 by Rishi Kumar Ganeriwala 1
Abstract
Multiphysics Modeling of Selective Laser Sintering/Melting by Rishi Kumar Ganeriwala Doctor of Philosophy in Engineering - Mechanical Engineering and the Designated Emphasis in Energy Science and Technology University of California, Berkeley Professor Tarek I. Zohdi, Chair
A significant percentage of total global employment is due to the manufacturing industry. However, manufacturing also accounts for nearly 20% of total energy usage in the United States according to the EIA. In fact, manufacturing accounted for 90% of industrial energy consumption and 84% of industry carbon dioxide emissions in 2002. Clearly, advances in manufacturing technology and efficiency are necessary to curb emissions and help society as a whole. Additive manufacturing (AM) refers to a relatively recent group of manufacturing technologies whereby one can 3D print parts, which has the potential to significantly reduce waste, reconfigure the supply chain, and generally disrupt the whole manufacturing industry. Selective laser sintering/melting (SLS/SLM) is one type of AM technology with the distinct advantage of being able to 3D print metals and rapidly produce net shape parts with complicated geometries. In SLS/SLM parts are built up layer-by-layer out of powder particles, which are selectively sintered/melted via a laser. However, in order to produce defect-free parts of sufficient strength, the process parameters (laser power, scan speed, layer thickness, powder size, etc.) must be carefully optimized. Obviously, these process parameters will vary depending on material, part geometry, and desired final part characteristics. Running experiments to optimize these parameters is costly, energy intensive, and extremely material specific. Thus a computational model of this process would be highly valuable. In this work a three dimensional, reduced order, coupled discrete element - finite dif- ference model is presented for simulating the deposition and subsequent laser heating of a layer of metal powder particles sitting on top of a substrate. Validation is provided and parameter studies are conducted showing the ability of this model to help determine appropriate process parameters and an optimal powder size distribution for a given mate- rial. Next, thermal stresses upon cooling are calculated using the finite difference method. Different case studies are performed and general trends can be seen. This work concludes by discussing future extensions of this model and the need for a multi-scale approach to achieve comprehensive part-level models of the SLS/SLM process. i
To my parents, Manju and Suri ii
Contents
List of Figures v
List of Tables ix
1 Introduction 1 1.1 Additive Manufacturing Techniques ...... 1 1.2 Applications of AM ...... 5 1.3 Outline of this Work ...... 6
2 Energy and Societal Impacts of Additive Manufacturing 9 2.1 Manufacturing Energy Use and CO2 Emissions ...... 9 2.2 Life Cycle Impacts of AM Technologies ...... 11 2.3 Societal Impacts of AM ...... 13
3 Selective Laser Sintering Process and Considerations 18 3.1 SLS Process Description ...... 19 3.2 Process Parameters ...... 22 3.2.1 Laser power ...... 22 3.2.2 Scan speed ...... 22 3.2.3 Spot size ...... 23 3.2.4 Hatch spacing ...... 23 3.2.5 Scan strategy ...... 23 3.2.6 Powder material and manufacturing ...... 24 3.2.7 Powder size distribution ...... 25 3.2.8 Layer thickness ...... 26 3.2.9 Surrounding gas atmosphere ...... 26 3.3 Material Properties of Parts Fabricated via SLS/SLM ...... 26 3.3.1 Density ...... 26 3.3.2 Microstructure ...... 27 3.3.3 Strength ...... 27 3.3.4 Hardness and surface roughness ...... 28 3.4 Other Considerations during SLS ...... 28 iii
3.5 Previous Modeling Attempts ...... 30
4 Powder Deposition and Laser Heating Model Description 33 4.1 Particle Dynamics ...... 33 4.2 Particle Thermal Effects ...... 37 4.2.1 Particle-to-particle heat transfer ...... 38 4.2.2 Laser beam modeling ...... 39 4.2.3 Phase change ...... 40 4.2.4 Thermal softening and melting of particles ...... 42 4.3 Finite Difference Modeling of Substrate ...... 43 4.4 Numerical Solution Scheme ...... 45 4.4.1 Time-stepping ...... 45 4.4.2 Binning and OpenMP Parallelization ...... 47 4.5 Programming Algorithm ...... 48
5 Powder Deposition and Laser Heating Model Validation and Results 51 5.1 Material Properties and Parameter Values ...... 51 5.2 Model Validation ...... 51 5.3 Numerical Examples ...... 54
6 Residual Stress Modeling 60 6.1 Background ...... 60 6.2 Modeling Framework ...... 62 6.2.1 Mechanical effects - balance of linear momentum ...... 62 6.2.2 Thermal effects - balance of energy ...... 65 6.2.3 Numerical solution scheme ...... 66 6.3 Numerical Examples ...... 67 6.3.1 Cooling of a solid block ...... 67 6.3.2 Cooling of a porous block ...... 75 6.3.3 Cooling of a single laser scan ...... 81
7 Conclusions and Future Extensions 90 7.1 Summary of this Work ...... 90 7.2 Model Limitations and Future Extensions ...... 93
Bibliography 97
A Economics and Projected Growth of Additive Manufacturing 110
B Numerical Derivatives 114 B.1 Spatial Derivatives using Finite Differences ...... 114 B.2 Time Marching Schemes ...... 117 B.2.1 Euler Methods ...... 117 iv
B.2.2 Runge-Kutta Schemes ...... 118 B.2.3 Other Schemes ...... 120
C Basic Parallelization Techniques 121 C.1 Binning Algorithm ...... 121 C.2 OpenMP ...... 122 v
List of Figures
1.1 List of common industrial AM processes, adapted from [66] ...... 3 1.2 Devices produced using AM techniques. SLS produced fuel nozzle for GE LEAP jet engines (left) [5] and acetabular cup for a hip implant (right) [3] 7
2.1 Global energy related CO2 emissions by scenario. 450 scenario is considered necessary to limit global temperature rise to 2 ◦C [50] ...... 10 2.2 US greenhouse gas emissions by sector [133] ...... 10 2.3 LCAs showing largest impact sources for the SLS of PA2200 nylon (left) and SLM of stainless steel (right) [57] ...... 13 2.4 IEA projections of energy related CO2 emissions per manufacturing sector by 2050 (left) and employment per manufacturing sector by 2050 (right). Two business as usual (BAU) scenarios and the efficient G2 scenario are depicted [102] ...... 16
3.1 Different stages depicting the solid state sintering of metal powders [69] . 19 3.2 Schematic of a typical SLS/ SLM set-up ...... 20 3.3 EOS P 800 SLS machine [1] ...... 20 3.4 Different scanning strategies used during SLS. Typical parallel zig-zag pat- tern (left) and island scanning strategy (right). Dashed lines refer to laser path...... 24 3.5 Bimodal packing distribution ...... 25 3.6 Microstructure of SLM-processed Inconel 718 [51] ...... 27 3.7 Comparison against standard bulk material properties of yield strength, tensile strength, and breaking at elongation for stainless steel parts pro- duced via SLM. Bar color indicates direction tested [65] ...... 28 3.8 SEM images depicting balling behavior of a single laser scan at different scan speeds [74] ...... 29 3.9 Warping of parts due to uneven thermal expansion/contraction [89] . . . . 30
4.1 Discrete element representation of pre-sintered powder particles (SEM im- age from [151]) ...... 34 4.2 Particle-to-particle overlap ...... 35 4.3 Heat transfer to an individual particle ...... 38 vi
4.4 Contact area of two intersecting particles ...... 39 4.5 Gaussian laser beam depiction ...... 40 4.6 Laser beam illustration as a series of light rays (length of each ray corre- sponds to intensity). Note that this approach is not used in the current work...... 41 4.7 Illustration of how heat capacity is updated (left) and ensuing phase change diagram (right) ...... 42 4.8 Finite difference stencil and coordinate system used for an arbitrary mate- rial property A. Indices i, j, and k are used to represent the x, y, and z coordinates, respectively...... 44 4.9 Depiction of boundary conditions used for the finite difference mesh: all B.C.s (left), Neumann B.C. for top face (right) ...... 45 4.10 Simulation flow chart ...... 47 4.11 Illustration of binning algorithm. Only particles in the shaded boxes will be checked for contact with the red particle in the center box...... 48
5.1 Comparison of experimental melt pool size by Khairallah and Anderson (left) [58] and simulation melt pool size depicted in red (right). Note that the figure on the left is Figure 5(a) from [58]. The experimental melt height, width, and depth are 26 µm, 75 µm, and 30 µm, respectively. The simulation melt pool dimensions are 30 µm, 85 µm, and 20 µm...... 54 5.2 Screenshots depicting the deposition of a layer of 316L SS particles (600 particles total) ...... 55 5.3 Screenshots showing the temperature evolution of a layer of 316L SS par- ticles and the underlying 316L SS substrate as a laser is passed over (tem- perature in Kelvin). Top view on upper row. Cross-sectional view from the side on bottom row...... 56 5.4 Screenshots depicting the melt pool (red) of a layer of 316L SS particles and the undelying 316L SS substrate as a laser is passed over. Top view on upper row. Cross-sectional view from the side on bottom row...... 56 5.5 Melt pool size as laser power is varied from 40 - 400 W , scan speed is constant at 2.0 m/s ...... 57 5.6 Melt pool size as scan speed is varied from 0.4 to 4.0 m/s, laser power is constant at 200 W ...... 57 5.7 Melt pool from a single laser pass over a mono-modal (left) and bimodal (right) powder distribution. Note that solid particles are blue, molten ones are green, and gaseous ones are red. Notice how some small particles are vaporized in the bimodal distribution. Laser power is 200 W and scan speed is 2.0 m/s. All parameters between the two simulations are identical other than particle size distribution...... 59
6.1 Boundary conditions for residual stress calculations ...... 69 vii
6.2 Initial temperature distribution (temperature in K) ...... 69 6.3 Average temperature of the fully dense 1 mm x 1 mm x 1 mm block of 316L SS ...... 72 6.4 Average Von Mises stress of the fully dense 1 mm x 1 mm x 1 mm block of 316L SS ...... 72 6.5 Average equivalent strain of the fully dense 1 mm x 1 mm x 1 mm block of 316L SS ...... 73 6.6 Temperature distribution (in K) of the fully dense 1 mm x 1 mm x 1 mm block of 316L SS at t = 0 (left), t = 0.5 ms (middle), and t = 1 ms (right) 73 6.7 Von Mises stress distribution (in MP a) of the fully dense 1 mm x 1 mm x 1 mm block of 316L SS at t = 0 (left), t = 0.5 ms (middle), and t = 1 ms (right) ...... 74 6.8 Equivalent strain distribution of the fully dense 1 mm x 1 mm x 1 mm block of 316L SS at t = 0 (left), t = 0.5 ms (middle), and t = 1 ms (right) 74 6.9 Phase distribution of 70% dense 1 mm x 1 mm x 1 mm block of material. Red points are phase 1 (316L SS) and blue are phase 2 (argon gas) . . . . 76 6.10 Average temperature of the 70% dense 1 mm x 1 mm x 1 mm block of 316L SS ...... 77 6.11 Average Von Mises stress of the 70% dense 1 mm x 1 mm x 1 mm block of 316L SS ...... 77 6.12 Average equivalent strain of the 70% dense 1 mm x 1 mm x 1 mm block of 316L SS ...... 78 6.13 Temperature distribution (in K) of the 70% dense 1 mm x 1 mm x 1 mm block of 316L SS at t = 0 (left), t = 0.5 ms (middle), and t = 1 ms (right) 78 6.14 Von Mises stress distribution (in MP a) of the 70% dense 1 mm x 1 mm x 1 mm block of 316L SS at t = 0 (left), t = 0.5 ms (middle), and t = 1 ms (right) ...... 79 6.15 Equivalent strain distribution of the 70% dense 1 mm x 1 mm x 1 mm block of 316L SS at t = 0 (left), t = 0.5 ms (middle), and t = 1 ms (right) 79 6.16 Finite difference mesh of a layer of powder particles ...... 80 6.17 Phase distribution of the SLS simulation with using pre-deposited particles, sampled from a 100 µm x 100 µm x 100 µm domain. Red points are phase 1 (316L SS) and blue are phase 2 (argon gas) ...... 81 6.18 Average temperature of the 100 µm x 100 µm x 100 µm domain ...... 82 6.19 Average Von Mises stress of the 100 µm x 100 µm x 100 µm domain . . . 82 6.20 Average equivalent strain of the 100 µm x 100 µm x 100 µm domain . . . 83 6.21 Temperature distribution (in K) of the 100 µm x 100 µm x 100 µm domain at t = 0 (left), t = 0.5 µs (middle), and t = 1 µs (right) ...... 83 6.22 Von Mises stress distribution (in MP a) of the 100 µm x 100 µm x 100 µm domain at t = 0 (left), t = 0.5 µs (middle), and t = 1 µs (right) ...... 84 6.23 Equivalent strain distribution of the 100 µm x 100 µm x 100 µm domain at t = 0 (left), t = 0.5 µs (middle), and t = 1 µs (right) ...... 84 viii
6.24 Initial temperature distribution of the 300 µm x 500 µm x 300 µm powder- steel mixture after a single laser scan. All nodes over 1100 K are assumed to have molten and be solid 316L SS...... 86 6.25 Phase distribution of the 300 µm x 500 µm x 300 µm powder-steel mixture after a single laser scan. Red points represent solid 316L SS and blue represents unmolten powder...... 86 6.26 Average temperature of the 300 µm x 500 µm x 300 µm powder-steel mixture ...... 87 6.27 Average Von Mises stress of the 300 µm x 500 µm x 300 µm powder-steel mixture ...... 87 6.28 Average equivalent strain of the 300 µm x 500 µm x 300 µm powder-steel mixture ...... 88 6.29 Temperature distribution (in K) of the 300 µm x 500 µm x 300 µm powder- steel mixture at t = 0 (left), t = 5 µs (middle), and t = 10 µs (right) . . . 88 6.30 Von Mises stress distribution (in MP a) of the 300 µm x 500 µm x 300 µm powder-steel mixture at t = 0 (left), t = 5 µs (middle), and t = 10 µs (right) ...... 89 6.31 Equivalent strain distribution of the 300 µm x 500 µm x 300 µm powder- steel mixture at t = 0 (left), t = 5 µs (middle), and t = 10 µs (right) ...... 89
7.1 Fluid flow modeled by the Lattice Boltzmann method around a Discrete Element sphere, demonstrating couple feasibility (color represents fluid ve- locity) [28] ...... 94 7.2 DMG Mori LASERTEC 65 laser deposition welding and milling machine. Outer case (left) and close-up of a part being built (right) [6] ...... 95
A.1 Worldwide sales of industrial AM systems since 1988 [143] ...... 111 A.2 AM Revenue forecasts through 2020 [82] ...... 111 A.3 Metal AM powder demand forecasts by industry [30] ...... 112
B.1 Illustrative mesh of a domain used in the finite difference method . . . . . 115
C.1 Illustration of binning algorithm. Only particles in the shaded boxes will be checked for contact with the red particle in the center box...... 122 C.2 Linked list binning for a 2D example problem ...... 123 ix
List of Tables
5.1 Material properties for 316L stainless steel as a function of temperature and phase [48, 124] ...... 52 5.2 Remaining material properties and simulation parameters used [48, 2] . . 52 5.3 Affect of powder size distribution on loose bed density, last row represents a bimodal distribution ...... 58
6.1 0.2% Yield strength, ultimate tensile strength, and instantaneous coeffi- cient of thermal expansion for 316L SS [124, 48] ...... 70 6.2 Additional material properties used in this simulation [101, 124] ...... 70 6.3 Argon material properties [77] ...... 75 6.4 Powder material properties [77, 116, 86, 124] ...... 85 x
Acknowledgments
I would like to thank many people for their inspiration and help in completing this manuscript. First, I would like to thank Professor Tarek Zohdi for his years of guidance and advice. I have learned more during the past 4+ years in the CMRL than at possibly any other time of my life. Thanks for all you’ve taught me and for always pushing me to keep working. I would also like thank my other commitee members, Professor Daniel Kammen and Professor Hayden Taylor. Thank you Professor Kammen for teaching me so much about energy and really fostering my love for sustainable technologies that can benefit society. Thank you Professor Taylor for your inquisitive nature and offering me useful feedback on my research. Additionally, I would like to thank all the members of the CMRL lab at Berkeley. Our countless discussions have been really useful in helping me formulate this thesis and solve problems I encountered along the way. Thanks to the rest of my family and friends too, for helping me along this process. Most importantly, I would like to thank my mom and dad, Manju and Suri Ganeriwala. You guys have always served as the best example for me and instilled in me the value of hard work and doing things the right way. Thanks for your continued love and support. This work was primarily funded by the Siemens corporation. Many thanks go to Ramesh Subramanian, Marco Brunelli, Nicolas Vortmeyer and Vinod Philip for their support over the last 4 years. Thanks to all other Siemens employees who attended my monthly research updates and offered useful feedback along the way. This work was also funded in part by the Army Research Laboratory through the Army High Performance Computing Research Center (cooperative agreement W911NF-07-2-0027). Many thanks go to Raju Namburu for his support. 1
Chapter 1
Introduction
3D printing is a buzz phrase thrown around by so many in our society today. But what is it really? And how does it work? Well, turns out that 3D printing is actually just a phrase given to a variety of different techniques that fall under the general category of additive manufacturing (AM). AM is defined as the process of building 3D objects by adding layer-upon-layer of material. This is in contrast to traditional machining which typically removes material from a larger block, also called subtractive manufacturing. AM, or 3D printing, has seen its use rapidly grow in recent years, with a recent Economist article even describing the digitization of manufacturing as the third industrial revolution due to its potential to radically change the manufacturing industry [85], a thought that has since been echoed by many others [90]. In this chapter the main commercialized AM techniques will be discussed, followed by a look at just a few of the potential applications. This chapter will conclude with a section outlining the remainder of this manuscript.
1.1 Additive Manufacturing Techniques
Within the broad category of additive manufacturing, or 3D printing, there exist many different techniques that can be used depending on the material and desired part characteristics. However, they generally follow the same basic steps [31]:
1. A 3D model of the part to be produced is digitally produced, typically using CAD software. That file is converted into a type as specified by the AM technology (often a STL file).
2. The file is sliced into layers by software on the computer and sent to the AM machine.
3. The AM machine builds the part up layer-by-layer. The fabrication path and method is controlled by computer software.
4. The final part is built and removed from the machine. Any support structures or adhering material is removed. Minimal post-processing operations may be necessary 2
depending on the technique used and desired part characteristics.
The most commonly used AM technologies are stereolithography (SLA or SL), fused de- position modeling (FDM), inkjet printing (IJP), laminated object manufacturing (LOM), and selective laser sintering (SLS) [66]. These processes and more are summarized in Figure 1.1. Stereolithography was patented in 1986 by Charles Hull [46] and commercialized by 3D Systems Inc. This technique uses a photo-sensitive polymer resin and a UV laser to build parts layer-by-layer. A wiper blade deposits a thin layer of resin onto the build platform. A UV laser scans over the resin in the shape of the desired object, curing the resin it contacts. The build platform is lowered and a new layer of resin is deposited on top. Once the resin has settled the UV laser scans over it again joining it to the underneath surface. The part is built up in this manner. Support structures are required to attach the part to the build platform. Upon completion the part is taken out of the resin bath and support structures are removed. Often the part may continue to be cured in a separate chamber under UV exposure. This process can offer high resolution parts but typically only for small part sizes. Additionally, the photo-polymer resins can be quite expensive, can be toxic, and are limited in material. The entire process typically takes from a few hours up to a couple days depending on part size and resin characteristics. Fused deposition modeling was developed by S. Crump in the late 1980s [15] and commercialized in 1990 by Stratasys. This technique extrudes molten material from a movable nozzle. This nozzle traverses over the build platform depositing thin layers of the molten material. After each layer, the build platform lowers and the nozzle deposits a new layer on top of the previous layers. The molten material solidifies and bonds with the previous layer almost immediately after extrusion from the nozzle. Support structures are needed to support the weight of the part. While initially only used with polymer plastics, many new materials (wax, metals with binder, and ceramics with binder) have also been incorporated [66]. FDM is generally the cheapest AM technology and is commercially available in desktop sizes for personal use. The main limitations are surface finish and accuracy due to the seam line between layers and delamination of the layers if not done correctly. Build time is typically on the order of hours to days depending on part size. Computer software can automatically make a solid model either partially or completely hollow if desired by the user (often hollow parts are sufficient for prototyping purposes). This greatly speeds up build time and decreases material use. Inkjet printing has origins tracing back to the late 19th century and was first com- mercialized by Siemens in 1951 [70]. It was originally developed for printing 2D images by ejecting ink droplets onto a substrate. The ink contains solutes dissolved in a solvent, which evaporates upon deposition on the substrate, leaving only the original solute. Ex- tensions to 3D printing have come about through the use of pre-patterned substrates at multiple layers of processing. Many materials have been introduced in IJP to additively produce a variety of products including solar cells, sensors, and thin-film transistors [45]. The main drawbacks are expensive ink cartridges and fragile print heads. 3
Figure 1.1: List of common industrial AM processes, adapted from [66] 4
Laminated object manufacturing was patented in 1988 by Michael Feygin [25]. In this process, layers of adhesive coated sheet materials are glued together. 3D parts are created by laminating and adhering 2D cross-sections on top of each other. Each cross- section is cut to shape often using a laser cutter. Many materials can be additively manufactured in this way including paper, polymers, composites, ceramics, and metals [66].This easily automated process is relatively cheap and safe. However, it tends to suffer from accuracy problems (especially in the build direction) and does not offer the best part quality. Often post-process machining and/or drilling are necessary to produce accurate final parts. While typically an adhesive is used to bond layers together, LOM can also include the bonding of two thin sheets of metal via ultrasonic consolidation [97]. Selective laser sintering was developed in the late 1980s at the University of Texas at Austin by Carl Deckard and his advisor Joe Beaman [18]. In this process powder particles are fused together via either full or partial melting from a laser beam. A wiper deposits a thin layer of powder particles on the fabrication bed. A laser scans over the particles in the shape of the desired object causing the particles to melt and fuse together. The fabrication bed lowers a tiny bit and a new layer of powder is swept on top with the wiper blade. The laser scans over this layer causing the powder to fuse with the layer underneath it. The final part is built up layer-by-layer this way. Upon completion the final part is removed and adhering particles are cleaned off the final part. No support structures are needed due to the presence of the powder bed. Unsintered powders can typically be recycled and reused, though this may vary with material. Materials suitable for SLS processing include polymers, ceramics, and metals. Sometimes a binder material is used as well. SLS offers the advantage of being able to rapidly produce parts of complicated geometries in one step; however, many issues can arise if not performed correctly. As developing a computational model of SLS is the main focus of this dissertation, this process will be discussed in much more detail in later chapters. Also note here that in the case of full melting of metal powders, this technique is frequently referred to as selective laser melting (SLM) though the overall process is the same for SLS and SLM. Many other AM technologies exist besides the ones explicitly described in this section. However, many of these techniques are variants of the techniques used in the described AM technologies. For example laser cladding and direct metal deposition work by shoot- ing metal particles onto a surface while those particles are simultaneously melted via a laser beam. These techniques are used to additively manufacture and to repair exist- ing parts. Selective electron beam melting (SEBM) is identical to SLS/SLM except an electron beam provides the heat source and the process is performed in a vacuum. Three- dimensional printing deposits powder material on a substrate. This powder is selectively joined via a binder. Chemical vapor deposition (CVD) and lithography processes are also occasionally considered 3D printing techniques though they are not further addressed in this work. Since AM is such a recent field, there is still an ever-growing amount of research into these processes and ways to expand material capabilities, improve product perfor- mance, and decrease processing time. New technologies are continuously developed and commercialized. Some of these newer, not as proven, technologies are briefly discussed in 5 the Conclusions chapter.
1.2 Applications of AM
Additive manufacturing offers the ability to rapidly produce parts of complicated ge- ometries in ways previously unfeasible with traditional techniques. As such, AM is fre- quently used in rapid prototyping (RP) and rapid manufacturing (RM) of parts. While many people may think of 3D printing as a hobbyists endeavor; its uses are actually far greater. While only 25% of the AM market was involved the direct manufacture of end- use products as of 2011, it is the industry’s fastest growing segment with a 60% annual growth rate [13]. SLS in particular has applications in a vast number of industries includ- ing (but certainly not limited to) tooling, biomedical, aerospace, automotive, energy, and consumer products. SLS and other AM techniques have vast potential for rapid tooling (RT) due to their ability to quickly fabricate (1) sacrificial patters using polymers; (2) insert, tool, core, and molds using composites; (3) metallic molds with cooling channels; and (4) ceramic based molds [67]. According to the National Center for Manufacturing Science, AM can reduce die production time by 40% [87]. SLS/SLM has also shown the capability of producing hard metal parts such as tungsten carbide-cobalt (WC-Co) and titanium carbide-nickel (TiC-Ni) used in machining tools and abrasion resistant coatings [83, 27]. In the biomedical industry many researchers have demonstrated the ability for SLS to produce tissue engineering scaffolds, valves, stents, and many types of implants [127, 140, 111]. AM allows for easy customization of bio-products to fit each individual. Custom fit personal hearing aids and dental products (implants, crowns, bridges, etc.) can also be made faster and in higher quality using SLS [67]. Using UV light to cure polymers, LUXeXceL has shown the ability to 3D print functional lenses [107]. AlpZhi is similarly 3D printing lenses. AM techniques have also been shown to significantly decrease production time of surgical tools [87]. One of the biggest adopters of AM (and SLS/SLM in particular) has been the aerospace industry. In addition to using AM for rapid prototyping of design changes, entire parts are now fabricated using SLS, which have the benefits of being lighter, cheaper, and quicker to manufacture. GE has made headlines for their new fuel nozzle in the LEAP engines which are entirely made using SLM (see Figure 1.2) [4]. Many other casings, panels, and duct work is fabricated using SLS, again saving time, money, and weight (a huge factor in airplane design). Molds and templates for cast parts can be 3D printed in a matter of hours, not days or weeks as typically required using conventional machining [42]. Similarly, the automotive industry has demonstrated the potential for AM to reduce part build time, costs, and weight. Many clips, brackets, panels, and HVAC systems can quickly be customized for each individual and car model and produced using SLS [67]. Other companies, such as Local Motors and Divergent Microfactories, have taken this a step further and are 3D printing entire cars. Divergent Microfactories has been able to 6 significantly reduce car weight and environmental impacts without sacrificing safety or performance using modular, 3D printed components [16]. Other sectors of the energy industry (in addition to aerospace and automotive) also see significant benefits through the use AM. Companies such as Siemens and GE are testing SLM produced components in their gas turbine blades. AM allows them the opportunity to reconfigure turbine blade design to enhance cooling and improve efficiency, while simultaneously reducing build time [137]. Thermal barrier coatings also show the potential to be 3D printed onto turbine blades. However, 3D printed turbine blades still suffer from performance issues and thus have not yet been introduced into deliverable products; this is very much an active area of research. A technique known as laser-driven non-contact transfer printing has shown the ability to produce flexible inorganic solar cells by depositing prefabricated material from a stamp onto a substrate [73]. Roll-to- roll printing is additionally being tested for the fabrication of solar cells. 3D printing of optical materials allows for customized transparency and varying indices of refraction. With improvement such technology could also have applications for solar cells [107]. SLS has been able to improve the performance of fuel cells by being able to make more efficient flow channels in a carbon plate, increasing chemical reaction between flowing gas and the catalyst [67]. In the consumer industry, AM techniques have many applications in producing en- closures, fixtures, seals, gaskets, hinges, etc. Clothing, jewelry, and art have also taken to the design freedom enabled using AM. Additionally, shoes and athletic apparel have adopted AM as a way to quickly fabricate customized products [67, 45]. Fully functional, 3D printed electronics is another growing industry [139]. However, it is important to note here that while AM has clearly seen its use sky- rocket in a number of industries, there are still significant limitations preventing increased growth. Namely, AM techniques are currently limited by size of parts produced, material imperfections, cost, and the number of available materials. Currently only hundreds of materials are able to be 3D printed, compared to thousands available for conventional manufacturing processes [131]. As continued research into these limitations proceed and new techniques become commercialized, the applications of AM will only continue to grow.
1.3 Outline of this Work
This chapter was meant to introduce the reader to the basics regarding additive manu- facturing: techniques and some applications. Chapter 2 will delve further into the energy and societal impacts of AM. Chapter 3 will then begin to focus specifically on the SLS process. In Chapter 3 a more in depth discussion is provided about how the SLS pro- cess works, the different process parameters, properties of parts fabricated using SLS, other considerations and limitations, and previous modeling attempts. In Chapter 4, the deposition and laser heating model of the SLS process is described. A coupled discrete 7
Figure 1.2: Devices produced using AM techniques. SLS produced fuel nozzle for GE LEAP jet engines (left) [5] and acetabular cup for a hip implant (right) [3] 8 element - finite difference model is used to simulate the deposition and laser heating of a layer of powder particles. The mechanical model and thermal model are described, along the the numerical solution scheme and programming algorithm. Chapter 5 provides validation and results of the deposition and laser heating model described in Chapter 4. Different parameter studies are performed and numerical examples are shown. Chapter 6 presents the modeling of sintered parts as they cool down and thermal stresses develop. A background of residual stress and previous modeling attempts is first provided. Next, the finite difference scheme used to simulate the thermal stresses upon cooling is described. Results and numerical examples are provided for different cases and initial conditions. Chapter 7 concludes this work by discussing the main takeaways and observations from the modeling results. Future improvements and work to be done regarding the complete modeling of SLS is discussed along with extensions of the work to other AM techniques. 9
Chapter 2
Energy and Societal Impacts of Additive Manufacturing
The need to improve sustainability in all sectors of our society is perhaps more press- ing than ever before due to global climate change and temperature rise as a result of anthropogenic greenhouse gas emissions. There is no one magic technology that can help curb our CO2 equivalent emissions down to the required rates to limit temperature rise ◦ to the target of 2 C over pre-industrial levels. Figure 2.1 depicts global CO2 emissions under different scenarios. The 450 scenario is that necessary to limit total atmospheric CO2-equivalent to 450 ppm, which is considered necessary to limit global temperature rise to 2 ◦C [50]. Clearly, significant energy efficiency and conservation measures are needed in virtually every sector of our society to decrease emissions and meet this target. This section analyzes manufacturing’s impact on greenhouse gas (GHG) emissions and the role additive manufacturing may play in these emissions and on our society as a whole.
2.1 Manufacturing Energy Use and CO2 Emissions Manufacturing is an indispensable part of our society. It is necessary to produce goods and services which we all use on a daily basis. The industrial sector, including manu- facturing, mining, and construction, accounts for nearly 25% of all global employment [21]. Indeed manufacturing has also been listed as the most important cause of economic growth according to the Roosevelt Institute [106]. While clearly a strong driver for the global economy, manufacturing also is a major culprit in the climate change problem. Figure 2.2 shows the US GHG emissions by sector, in which it is clear that industrial emissions constitute a very significant portion of all US emissions [133]. According to the US Energy Information Administration (EIA), manufacturing accounts for nearly 20% of total energy usage in the United States [132]. In fact, manufacturing accounted for 90% of US industrial energy consumption and 84% of industry carbon dioxide emissions in 2002 [109]. 10
Figure 2.1: Global energy related CO2 emissions by scenario. 450 scenario is considered necessary to limit global temperature rise to 2 ◦C [50]
Figure 2.2: US greenhouse gas emissions by sector [133] 11
Just as a move away from fossil fuels to clean sources of energy is considered vital to reducing GHG emissions, a move towards more efficient and sustainable manufacturing processes is also necessary, especially as the amount of raw materials processed has steadily increased over the past decades. For example, the annual global production of aluminum more than doubled between 1980 and 2005 [21]. While the energy intensity (energy per $1 of goods) of US industry has been reduced by 50% over the past three decades, an alarming trend has been observed. As new manufacturing processes, which can work at finer scales and smaller dimensions, are commercialized, the specific electricity requirements (J/kg) for these processes is increasing and the process rate (kg/hr) is decreasing. Additionally, high exergy materials are increasingly used in inefficient ways [36]. Thus it is important to take a closer look at the impacts of AM before deciding upon its potential sustainability impacts.
2.2 Life Cycle Impacts of AM Technologies
When evaluating new technologies, such as additive manufacturing, it is important to analyze the total environmental impacts in an objective and quantifiable manner. A commonly used technique for such type of analysis is called the life cycle analysis (LCA). LCAs analyze the environmental impacts of a process as it relates to numerous different factors, including GHG emissions, deforestation, toxicity, ozone depletion, eutrophication, and others. Each of these factors is then weighted appropriately to determine a final environmental impact score. LCAs typically include complete cradle-to-grave analyses of processes/machines including the material extraction, manufacturing, transportation, use, and end-of-life phases [138]. As additive manufacturing technologies are still relatively recent, not too many com- prehensive LCA studies have been performed comparing the effects of AM vs. conventional manufacturing. However, the general consensus is that AM processes have better envi- ronmental characteristics as compared to traditional machining [81, 123, 45]. The main reasons lie in reduced material consumption, less tooling required, and the elimination of often toxic cutting fluids used in machining. One study by Serres et al. [114] found a 70% environmental impact reduction for CLAD (a laser based AM process) as com- pared to machining. Other studies for laser based AM found similar impact reductions [91, 88] stemming from the ability to remanufacture (fix) broken parts using direct metal deposition and the reduction in time necessary to produce certain parts of complicated geometries. Additionally, dies for injection molding can be created in much shorter time frames using laser based AM [87]. However, other studies have found conflicting reports regarding the environmental benefits of AM. One study found that casting is actually significantly better than AM in terms of energy usage, although AM beats casting in terms of other environmental impacts such as material use [45]. Another found that that milling performed comparably with inkjet printing and SLA in terms of total environmental impact [23]. This study actually 12 found that how the machine is used was more important than the fabrication technique, in that a lot of power is consumed when machines are simply idling. Thus having the machines constantly in use significantly reduced the impacts per amount of material processed. However, a follow up study found that AM did possess environmental benefits when producing parts of complicated geometries [24]. The main reason for this lies in the fact that the geometry of the part typically does not matter in the amount of time required for fabrication using AM techniques. However, when trying to produce a part with complicated geometry (such as a hollow part with small features) using conventional milling, many turns and cuts are needed which significantly increases the time required for fabrication. Baumers et al. [8] measured the energy use during the SLS process. In this study the authors mentioned the need to include utilization percentage when performing LCAs of these machines as it will have a significant affect on the environmental impacts. They measured that warm-up and cool down time can sometimes consume upwards of 20% of the total energy required to build parts in an SLS machine. Kellens et al. [57] performed a more detailed analysis of the lifecycle impacts of SLS/SLM machines. Figure 2.3 shows the main impact sources for the production of nylon 12 based PA2200 parts using an SLS machine and the production of stainless steel parts using an SLM machine. In the LCA for the production of nylon parts in an SLS machine, waste powder production is actually the leading source of environmental impacts, followed by energy use. This demonstrates the need for more efficient recycling of powder during SLS operations. In the LCA for the production of stainless steel using an SLM machine, energy use is the leading source of environmental impacts, followed closely by the production and use of the nitrogen gas atmosphere in the chamber. A better sealed chamber could decrease the impacts of nitrogen (or other inert gas) by reducing the amount needed to continually fill the process chamber. Powder waste is only a very small percentage of total impacts during this SLM process suggesting that metal powder is produced and recycled more efficiently than polymer powders in SLS. However, these results could vary depending on specific material, machine, and batch size [57]. In any case, it is worth further investigating powder and compressed air production practices as improvements in these areas will improve the sustainability of the SLS process as a whole. Overall, firm conclusions regarding energy use in AM vs. conventional machining are still unclear due to a lack of literature on the subject and the variability in energy measurement methodologies used by each researcher. While there have been attempts to quantify the direct lifecycle impacts of AM tech- nologies, there are many indirect effects which could significantly alter global manufactur- ing energy consumption. The ability to fabricate net-shape parts of virtually any geometry allows for the redesign of currently used parts. For example, it is easy to imagine formerly solid parts that could now be made hollow due to ease of manufacturing. This would not only save material but would reduce costs and energy use associated with manufacturing. Lighter parts in an airplane or car would also lead to large fuel savings. Parts could be designed with fewer components reducing total manufacturing time and material usage. In fact, AM has been able to reduce the number of steps required to produce surgical 13
Figure 2.3: LCAs showing largest impact sources for the SLS of PA2200 nylon (left) and SLM of stainless steel (right) [57] tools from 62 to 7 [88]. GE has been able to produce lighter fuel nozzles for their airplane gas turbine engines using SLM. They were able to reduce 18 separate components, which required multiple machining and welding steps, into one additively manufactured nozzle. This nozzle is 25% lighter and can be made in much less time [4]. In total, GE has been able to realize a weight reduction of over 500 pounds per engine through 3D printing of various components, which can amount to huge fuel savings and CO2 reductions over the life of the engine [42]. Divergent Microfactories has been able to reduce the weight of a car chassis from around 1000 pounds down to 100 pounds through the use of their node based assembly system made possible by AM and the use of carbon fiber. Not only have they dramatically cut down on material use in the manufacturing of their vehicle (a significant portion of a car’s LCA), but they claim their 3D printed car would have one-third the emissions of a comparable electric vehicle [16]. This is an impressive feat which illustrates the potential environmental benefits of AM, even if some of their claims may be partially exaggerated. AM could affect the supply chain since parts would no longer need to be made in large quantities as the price per part does not depend on scale when using AM. This would allow less overstocking of parts and decrease the need for huge manufacturing plants. More manufacturing could be performed locally and on-demand which would decrease energy used for transportation, packaging, and storage of components. All these indirect impacts are currently not reflected in LCAs of specific technologies and should be the subject of continued investigation.
2.3 Societal Impacts of AM
An article from The Economist described the digitization of manufacturing (made possible primarily by AM) as the third industrial revolution, a sentiment that has been echoed by many since then [85]. By being able to create a part on a computer and subsequently 3D print it, anyone can become a designer and manufacturer. Since using a 14
3D printer is often as simple as reading a manual, and a few trial and error prints, the need for highly skilled machinists will decrease as more people and companies purchase AM machines. Overall the way our society approaches manufacturing in the future could be significantly different from the conventional methods used in the past as AM technologies continue to develop. In a literature review of the societal impacts of AM technologies, Huang et al. [45] found that the main benefits of AM use can be summarized by: • Customized additive manufacturing of healthcare products • Reduced material use and energy consumption • Supply chain reconfiguration by on-demand and on-site manufacturing Many healthcare products are already being designed and built more efficiently using AM techniques [67, 140]. AM allows for mass customization of parts and quicker build times for things such as hearing aids, implants, tissue scaffolding, stents, and dental parts. Additionally personal protective equipment such as helmets, protective garments, and even athletic equipment can be custom fit to individuals providing safety benefits. In the previous section material use and energy benefits were discussed by noting how AM drastically reduces cutting waste and can greatly shorten the fabrication time required for parts with complicated geometries. Parts can be redesigned to have fewer components and more complicated structures. Design changes can be implemented quickly and without the need for significant process restructuring. Mass customization of parts becomes much more feasible as you would no longer need to overhaul an entire factory if the shape of one part is changed. The amount of time required for manufacturing of these parts can also be decreased. One of the biggest impacts of AM can be in how the supply chain is structured. AM can improve the efficiency of supply chains through just-in-time manufacture and waste reduction. Parts can be made at the shop floor instead of requiring delivery by vendors. Build-to-order strategies can be implemented ensuring that overstocking does not occur [130]. A concept known as distributed manufacture is gaining attention as AM allows for more small, localized manufacturing facilities [45]. This will reduce the need for large warehouses, packaging, and transportation of goods. Holmstrom et al. [44] suggested that integrating AM into the spare parts supply chain can be approached in one of two ways: 1. Using centralized AM to replace inventory holding 2. Distributed AM to replace inventory holding and conventional distribution The centralized approach is desirable when the parts requiring AM are limited in quantity and response time is not as critical. The distributed approach is more appropriate when there is a significantly high demand of AM producible replacement parts to justify the necessary investment in AM machinery [39]. 15
The growth of distributed manufacturing could also have significant health benefits if this causes a move away from large, centralized manufacturing facilities in remote locations. Many companies are predicting that AM could lead to more manufacturing re- turning to US as the need for cheap labor in China and other countries decreases [13, 85]. By moving manufacturing closer to company headquarters and end-use, companies can save on transportation costs and be able to more quickly change designs in products. Less large factories could improve health as many low-wage workers in developing coun- tries do not have appropriate personal protective equipment and adequate ventilation. Less conventional machining would also decrease the exposure to toxic cutting fluids for these workers. Additionally, noise reduction could improve hearing as AM machines are typically much quieter than milling counterparts [45]. However, AM is also not without its drawbacks. While toxic cutting fluids are not used, many resins used in SLA processes are toxic and pose health risks if touched or inhaled. For many materials, the biological effects are still not fully understood [45]. Additionally, inhalation of fumes released during FDM could be dangerous if proper ventilation is not present as ultrafine particle emissions released during FDM have been shown to be toxic to rats and mice [126]. Since many FDM machines are bought for use in private residences, there are no safety regulations in place to ensure they are used in areas with adequate ventilation. An additional concern could lie in the fact that the automization of manufacturing could cause massive job loss, an even more significant problem considering that the popu- lation will only continue to rise as the 21st century progresses. However, IEA projections actually show that many new jobs can be created with the use of more sustainable man- ufacturing techniques. Figure 2.4 (from a 2009 IEA report) shows projections of the CO2 emissions and employment per manufacturing sector in 2050 for two different business as usual scenarios and a G2 scenario representing increased efficiency. While the CO2 emissions are understandably much lower in the efficient G2 scenario, this forecast also predicts the creation of more jobs under the G2 scenario [102]. Also worth noticing in this figure are the large CO2 emissions impacts and also employment for the manufacturing of metals and plastic, materials currently used in AM techniques. While plastics may still be a source of health concerns due to the large impacts and toxicity of the resins used in SLA and even some of the FDM plastics, the emissions due to metals could certainly be reduced using techniques such as SLS/SLM by drastically reducing material waste and overstocking of parts. Also note that these are only impacts associated with manufacturing of these materials. However, transportation emissions can also be reduced due to AM since many parts can potentially be made locally and thus not require long-distance shipping. Finally, it is important to note here that AM will never completely replace conventional manufacturing techniques. In terms of cost and energy use, casting still significantly outperforms AM when it is required to produce large quantities of the same part [45]. Additionally, limitations still exist in AM regarding size, imperfections, performance, materials, and cost. Until all these issues can be addressed, the widespread adoption of 16
Figure 2.4: IEA projections of energy related CO2 emissions per manufacturing sector by 2050 (left) and employment per manufacturing sector by 2050 (right). Two business as usual (BAU) scenarios and the efficient G2 scenario are depicted [102] 17
AM will remain limited. However, the amount of research currently going into improving existing AM technologies and creating new ones is steadily increasing, showing significant promise for its growth [13]. While it clearly has the potential to significantly improve upon the sustainability of the manufacturing industry, more research is needed in this area to better quantify the impacts. Some of the potential economic impacts of AM are discussed in Appendix A. 18
Chapter 3
Selective Laser Sintering Process and Considerations
Sintering traditionally refers to the heating of a compacted powder material to a temperature below the melting point, but high enough to allow bonding of the individual particles. This typically involves heating the pre-compacted powder to 70 – 90% of the melting temperature of the particles. The primary sintering mechanisms, which allow bonding to occur, are diffusion and vapor phase transport, where metal atoms release to the vapor and then re-solidify at convergent geometries (often the interface of two particles). Alternatively, in a mixture of powder particles, it is common to melt the powder with a lower melting temperature, which will then surround the solid particle due to surface tension (liquid-phase sintering). The final density of sintered parts can reach over 99% of the theoretical density of the resultant alloy. However, varying amounts of porosity may remain due to trapped gases remaining in voids after compaction, which may be desirable depending upon the application [54]. Conventional sintering involved placing the powdered material into a die where it is compressed and placed in a furnace. Solid-state sintering, where the powders do not fully melt, requires many hours at temperatures close to the melting temperature to allow the powders to bond together. Figure 3.1 depicts the traditional sintering of metal powders in a furnace. Another common sintering technique is electric current assisted, or spark plasma, sintering. In this process, powders are placed into a die connected to electrodes. The die is mechanically compressed while an electric current is run through the electrodes. The powders heat up due to Joule heating arising from their internal resistance. This process has the advantage of producing net shape parts in a much shorter time, and can be more energy efficient than traditional sintering [95]. Finally, there is selective laser sintering (SLS), which was first invented in 1986 at the University of Texas, Austin [125]. Since it’s original invention, it’s use has continued to grow rapidly with potential applications is virtually every industry where goods/products are made. The remainder of this chapter will describe the SLS process in much more detail. 19
Figure 3.1: Different stages depicting the solid state sintering of metal powders [69]
3.1 SLS Process Description
Selective laser sintering differs from traditional sintering in that the powders are not consolidated prior to heating and often full melting occurs. Note that in the case of full- melting for metals, the process is frequently referred to as selective laser melting (SLM). SLS typically refers to partial melting for metals, or the cases where the process is applied to ceramics and/or polymers [68, 67]. However, the distinction between SLS and SLM is often vague. Note that many other names may be used to describe this process, including direct metal laser sintering (DMLS) or direct metal laser melting (DMLM), among others [90]. However, all of these terms refer to the same basic process, with possible slight differences in some of the process parameters. For the remainder of this work, the process shall be referred to as selective laser sintering (SLS) or occasionally selective laser melting (SLM) for simplicity. Figure 3.2 depicts a schematic of a typical SLS/SLM machine. An actual EOS SLS machine is shown in Figure 3.3. First a thin layer of powder particles, with sizes ranging from approximately 10 to 100 µm, is deposited onto a substrate via a roller. Each layer is typically on the order of 20 to 100 µm thick. Many different powder materials can be used including metals, polymers, ceramics, and composites. A laser is then directed via a scanner system over the powders in the pattern of the desired shape, either partially melting (SLS) or fully melting (SLM) the powders. As the laser selectively melts the particles, they fuse and bond together. Next, the fabrication bed lowers, a new layer of powder is placed on top by the roller, and the laser selectively scans over the surface again. This layer-by-layer, additive manufacturing process is repeated until the final part has been fabricated. The laser beam typically has a nominal diameter on the order of 0.5 mm or less. A 20
Scanner system Object being Laser Roller fabricated Unsintered Powder supply powder
Powder Fabrication delivery piston piston
Figure 3.2: Schematic of a typical SLS/ SLM set-up
Figure 3.3: EOS P 800 SLS machine [1] 21
pulsed or continuous wave CO2 laser is typically used for polymer materials, while an Nd:YAG or fiber laser is more typically used for metals. The reason for this lies in the fact that polymers and ceramics (oxides) absorb the 10.6 µm wavelength emitted by the CO2 laser better, while metals and ceramics (carbides) better absorb the 1.06 or 1.08 µm wavelengths emitted by Nd:YAG or fiber lasers, respectively [67, 41]. Typically, the laser power can vary anywhere from approximately 30 - 400 W depending on the material and desired part characteristics (although up to 1 kW or even higher power lasers have also been used) [90]. Typically higher powers are used for full melting in SLM while lower laser powers may be used for the partial melting which occurs in SLS. Additionally, scan speed may be varied from 2 to 700 cm/s or more [125, 33]. In SLS, a binder material with a lower melting temperature may be applied to the particles. This binder material is melted, bonding together particles with higher melting temperatures. The binder also helps ensure even distribution of each layer of powder particles [145]. To aid in the process, frequently the sintering chamber is heated to minimize required laser energy and expedite the process. Preheating the chamber has the added benefit of reducing thermal gradients and thus reducing the residual stresses present in the final part. Additionally, an inert gas, usually nitrogen or argon, is present to avoid oxidation or burning of the powders. Upon completion of the SLS process, the final part is removed and often bead blasted to remove unsintered, adhering particles [125]. In the case of SLS, the final parts frequently still have a significant amount of porosity. In these cases additional operations, such as post-sintering, hot isostatic pressing, or infiltration, may be required to increase the part density [64]. Post-sintering works by holding the post-SLS part at high temperatures (approximately 70% of the melting temperature) for hours or even days to allow solid- state sintering and densification to occur [67]. Hot isostatic pressing involves holding the post-sintered part at an elevated temperature while an inert gas, typically argon, is pumped in at a high pressure causing consolidation of the part by plastic deformation. Infiltration involves injecting epoxy or some other flowable material with lower melting point into the porous part, essentially plugging up the holes [78, 134]. Due to the high heating and cooling rates present in SLS (up to 106 oC/s) large residual stresses may build up in the final part [135]. This high cooling rate also causes anisotropy of the final part as long thin grains form in the z-direction (or the vertical build direction), and inhibits the formation of certain microstructures. Typically martensitic microstructures are formed in steels [67]. Post SLS heat treating, such as annealing, can be performed to relieve residual stresses and change the microstructure [113]. SLS has the advantage of being able to rapidly produce parts of relatively complex geometries, in a timely and cost effective manner. It is especially useful in its ability to make metallic parts, which would otherwise be impossible to form using other rapid prototyping, rapid tooling, and rapid manufacturing techniques as shown in the works by Simchi and Pohl [120, 121]. Maeda and Childs and Fischer et al. [83, 26] have demonstrated the ability for SLM to produce parts made from hard metal powders, such as WC-Co and TiC-Ni used in machining tools and abrasion resistant coatings. SLS has also seen uses in RP and RM for polymeric and ceramic materials. Tan et al., Williams et al., 22 and Schmidt et al. [127, 140, 111] have successfully laser sintered polymeric biomaterials which can be used to produce tissue engineering scaffolds and certain types of implants. SLS is also currently employed to produce parts in the automotive, aerospace, biomedical, and energy industries, among others.
3.2 Process Parameters
While there are clearly a vast, and growing, number of applications for the SLS pro- cess, if not applied correctly, the resulting parts can have many defects, especially high residual stresses, microcracking, delamination of layers, and/or high porosity [64]. An optimal set of process parameters need to be applied to ensure high quality of the fin- ished product. Such process parameters may include (but certainly aren’t limited to) laser power, scan speed, spot size, hatch spacing (distance between successive parallel passes of the laser), and scan strategy. Additionally powder material, size distribution, layer thickness, and conditions in the surrounding atmosphere can significantly affect final part quality. Clearly, the optimal set of parameters will vary depending on material and desired final part characteristics. This section discusses each of these process parameters in more detail.
3.2.1 Laser power Laser power typically ranges from 30 - 400 W depending on the material and desired part characteristics. SLS machines are typically rated up to 50 W as this is adequate to melt most polymers and for partial melting of metals. Metal SLM machines have laser powers on the order of hundreds of watts to ensure full melting at reasonable scan speeds [67]. Laser cladding machines and other variants of SLM machines designed to melt much thicker layers at a time can employ lasers on the order of kilowatts [115]. The laser energy density is a key consideration in ensuring adequate melting of powder particles and the underneath layer, while minimizing ablation of the particles, which can lead to increased porosity and uneven surface finishes. The laser energy density is typically defined as [67] Laser power Laser energy density = (3.1) Spot size ∗ Scan speed
3.2.2 Scan speed Similar to laser power, scan speed is another parameter which can be adjusted to ensure proper melting. A lower scan speed will increase the laser energy density and thus lead to a larger melt pool. Faster scan speeds will decrease run time. However, faster scan speeds have also been shown to decrease melt pool width, which can lead to balling (due to Rayleigh instabilities, discussed further in Section 3.4) and uneven surface finishes [74, 144]. Additionally very high scan speeds may not allow sufficient time for the heat 23 to diffuse across the powder bed. Typical scan speeds range from from 2 to 700 cm/s, though this can be varied.
3.2.3 Spot size Most SLS machines use a Gaussian laser where the power decreases exponentially away from the center of the laser. The spot size of a Gaussian laser is defined as the point where the beam intensity falls to 1/e2 of the peak intensity at the center. A larger spot size will allow for faster scanning of an individual layer by increasing the melt pool width after a single pass. However, larger spot sizes will also decrease resolution of the final part. Spot size can be changed on a laser with a technique known as laser defocus, which changes the focal point of the laser to a plane below the build platform. However, this will also change the spatial distribution of laser intensity, which will affect melt pool geometry [10, 122].
3.2.4 Hatch spacing The hatch spacing is defined as the distance between parallel laser passes (see Figure 3.4). The ideal hatch spacing will be dependent on the melt pool size and the laser spot size. It should be large enough to decrease run time but still small enough such there is some remelting between parallel laser passes. This is necessary to ensure proper bonding and to decrease the chance of balling [84].
3.2.5 Scan strategy The scan strategy can have a very large effect on the characteristics of your final part. A good scanning strategy can significantly decrease distortion, anisotropy, and porosity in the final part [117]. Typically laser scanning is done in parallel lines in a zig-zag pattern, either along the length or width of the part. Alternating the scanning strategy between adjacent layers can also help decrease anisotropy and increase final part density [63]. Generally it has been shown that shorter scan lengths are desired to decrease the chance of balling. Additionally shorter scan lengths helps ensure that cooling and solidification does not occur before the laser comes back for an adjacent pass. Solidification is undesired between adjacent laser passes as this can cause improper bonding between layers and an uneven surface. To ensure shorter scanning lengths an island strategy may be employed where the build layer is divided into small blocks, or “islands”, each of which is scanned independently in an alternating pattern. This strategy has been shown to reduce balling and additionally distortion due to uneven thermal contraction when cooling [67]. The typical parallel zig-zag and island scanning strategies are shown in Figure 3.4. Note that many other scanning strategies exit, such as spiral, diagonal, fractal path, and others [146]. It is also generally preferred to build objects in a manner such that the z-direction 24
Hatch spacing
Figure 3.4: Different scanning strategies used during SLS. Typical parallel zig-zag pattern (left) and island scanning strategy (right). Dashed lines refer to laser path.
(vertical build direction) corresponds to the smallest dimension of the part, if possible. This decreases fabrication time and increases part accuracy [67].
3.2.6 Powder material and manufacturing Clearly the powder material will have a big impact on the final part characteristics. Generally, powders which are suited for welding applications perform well when additively manufactured via SLS. More ductile materials are better able to withstand the thermal stresses that arise when cooling, whereas as brittle materials have a greater tendency to form microcracks when cooling. This is especially present in the SLM of Inconel (a superalloy frequently used in gas turbines among other applications) [55, 122]. More ductile materials are better able to withstand residual stresses by plastic deformation and thus resist forming microcracks. The manufacturing technique used to create the powders can also be important. Metal powders are typically manufactured via atomization, a process where molten metal is forced through a nozzle at moderate to high pressures. A gas is introduced into the metal stream just before leaving the nozzle to induce turbulence. The collection chamber is also filled with a fluid promoting turbulence of the molten metal. This process is capable of forming fine micron sized powders with spherical shapes. Gas atomization has been shown to produce more spherical powders for steel when compared against water atomiza- tion. These more spherical powders have better packing and wetting characteristics upon melting, leading to a more dense final part. Additionally less oxidation of the powders is present using gas atomization [75]. 25
Figure 3.5: Bimodal packing distribution
Note that powder particles may be formed by other techniques such as pulverization or chemical reactions. Sometimes these can produce particles of highly irregular shapes as compared with the spherical particles formed by the atomization technique (for metal powders) or co-extrusion processes (for polymer powders) [110]. These irregular shapes inhibit the flowability of such powders, leading to increased porosity. Additionally, poly- mer powders of less than 50 µm in size attract to each other due to inter-particle forces, further decreasing flowability. To combat this frequently a lubricant or some other small powder is added to the mixture, which can increase flowability and loose bed density [32].
3.2.7 Powder size distribution The powder size distribution has a very significant effect on the densification and certain material properties of the final part. Simulations and experiments by Korner et al. [60] on selective electron beam melting (SEBM), similar to SLM except that an electron beam acts as the heat source and the process occurs in a vacuum, found that loose powder bed density played the biggest role in densification of the final part. As a result there has been investigations into using bimodal, or even trimodal, packing distributions where smaller particles fill the interstitial spaces between larger particles to increase the loose bed density (as depicted in Figure 3.5). The problem with such powder distributions, however, lies in the fact that the smaller particles may evaporate before the bigger particles fully melt. These evaporated particles may then get trapped as gas bubbles in the final part, increasing the final porosity [56]. Using finer powders can decrease the surface roughness of the part. However, using too fine of powders decreases the packing density of the loose bed. Hence performing SLS with sub-micron sized powders is limited in application and use due to difficulties in producing dense parts when using such small powders [67]. 26
3.2.8 Layer thickness The thickness of each layer is an additional consideration when determining SLS pro- cess parameters. Obviously, the part can be built faster if thicker layers are used; however, this will also decrease the smoothness and resolution of the part. Meanwhile, thinner lay- ers have been shown to be optimal for allowing any trapped gas bubbles to escape before solidification occurs [63]. The thickness of each layer is additionally limited by the melt depth of the laser, as some remelting of the underneath layer is required for proper bond- ing to occur. The optimal layer thickness will depend on material, desired part resolution, powder size distribution, and input laser energy density.
3.2.9 Surrounding gas atmosphere The atmosphere of the surrounding environment is critical in producing usable parts via SLS. If oxygen is present in the atmosphere, metals will very quickly form oxides which will inhibit melting and consolidation of the powders. Oxides have also been shown to increase the prevalence of balling in the melt pool. Thus the SLS chamber is filled with an inert gas, typically argon or nitrogen, and the partial pressure of oxygen is closely monitored. Creating a vacuum in the SLS chamber is also avoided as this can lead to increased ablation of particles [67]. Increasing the pressure of the chamber can decrease ablation.
3.3 Material Properties of Parts Fabricated via SLS/SLM
If done correctly, many parts created via SLS/SLM can be used in industrial applica- tions. However, it is important to note that parts created via SLS tend to have slightly different material properties as compared to conventionally forged or cast parts.
3.3.1 Density Obviously the density of parts produced via SLS is a critical consideration. If optimal process parameters are used, often times up to 99.9% of the theoretical bulk density can be achieved for steels and titanium allows [147, 63]. Other metals, including superalloys, can also achieve nearly full density under proper processing conditions. The main sources of lingering porosity in a material are due to only partial melting of the powder and entrapped gas bubbles. Remelting of each layer, by essentially scanning the laser twice over each layer, has been shown as an effective technique for increasing final part density [64]. In the case of porous parts, post-processing techniques can be used to increase the density including post-sintering, hot isostatic pressing, and infiltration. 27
Figure 3.6: Microstructure of SLM-processed Inconel 718 [51]
3.3.2 Microstructure Another feature characteristic of parts produced via SLS/SLM is anisotropy. Due to the extremely high cooling rates present (up to 106 oC/s) in the build (or vertical) direc- tion, metals have a tendency to form long, very fine columnar grains. These grains solidify before dendrites can form, causing an anisotropic microstructure (typically martensitic) [10]. Due to this anisotropy, build direction has a significant effect on the final part strength, as this will vary in different directions [118]. Figure 3.6 depicts the microstruc- ture of SLM-processed Inconel 718 [51]. Another consideration is residual stresses and microcracking which may arise in more brittle materials due to the high cooling rates and uneven thermal gradients. Preheating the chamber and the base plate has been shown to reduce residual stresses by decreasing the thermal gradients [119]. Additionally post-SLS heat treating, such as annealing, can be performed to relieve residual stresses and change the microstructure [113].
3.3.3 Strength The yield strength of parts produced via SLS/SLM can often be higher than that of materials produced with conventional methods. This can be explained by the marten- sitic microstructure and the fine columnar grains formed due to the high cooling rates. Additionally, the tensile strength is often comparable to, or higher than that of con- ventionally produced materials [122]. However, the ductility (measured by the strain at breaking), is almost always less for parts produced via SLM. The spatially distributed melt pool boundaries restrict the ductility, explaining why SLM parts are typically more brittle [118]. Figure 3.7 shows a comparison against standard bulk material properties of the yield strength, tensile strength, and breaking at elongation for stainless steel parts 28
Figure 3.7: Comparison against standard bulk material properties of yield strength, tensile strength, and breaking at elongation for stainless steel parts produced via SLM. Bar color indicates direction tested [65] produced via SLM [65].
3.3.4 Hardness and surface roughness The hardness of a material is a measure of how resistant it is to shape change. While this will clearly depend on the porosity of the sample, for fully dense steel parts the hardness has been shown to be comparable with conventionally wrought products of the same alloy [128]. The surface roughness of SLS parts is strongly dependent on the particle size, as partially molten particles will typically adhere to the edges of any sintered part. Studies have shown that remelting of top layers can be used to reduce the surface roughness of the top surface at the expense of increased processing time [147].
3.4 Other Considerations during SLS
During the SLS/SLM process, care must be taken to ensure that balling does not occur. Balling is a phenomena attributed to Plateau-Rayleigh instabilities where the melt pool will tend to ball up rather than forming one continuous track if the length to width ratio of the track is too high. This can cause an uneven surface finish and uneven layer height, leading to unusable parts. In studies performed on SLM of stainless steel and pure nickel powder, Li et al. [74] found two main types of balling that occur. Type 1 is ellipsoidal balls with dimension of approximately 500 µm which occur due to poor wetting ability of the material. The other type are small spherical balls of approximately 10 µm size, which 29
Figure 3.8: SEM images depicting balling behavior of a single laser scan at different scan speeds [74] do not have a significant effect on final part quality. One of the primary factors which leads to balling is the length to width ratio of the track. If the ratio of track length to track width is greater than approximately π, balling will occur due to Plateau-Rayleigh instabilities [67]. The length to width ratio can be decreased by increasing laser power or spot size or decreasing the scan speed or scan length. Preheating the chamber can also reduce this ratio. Figure 3.8 depicts the balling phenomena as a function of scan speed for a given laser power and spot size [74]. Clearly the balling behavior becomes more prevalent at higher speeds. The oxygen content is another factor which has a significant effect on the amount of balling that occurs. More oxygen in the chamber leads to the increased formation of oxides which have poor wetting characteristics. Decreasing the oxygen percentage in the chamber has been shown to lead to decreased balling under the same processing parameters [74]. Other studies have found that increased melt pool temperature decreases balling, as higher temperature molten metals have better wetting characteristics [64]. Note that wetting is dependent on interfacial energies between the surfaces of the liquid, solid, and air. Another factor which has a strong influence on the melt pool shape is Marangoni convection, which is surface tension driven convection that occurs in the melt pool. For many materials surface tension is a strong function of temperature. Thus convection currents form in the center of the melt pool (where temperature is highest) either to or from the outer edges of the melt pool, depending on whether the material’s surface tension increases or decreases with temperature. The inclusion of oxides or any other impurities also has a very strong influence on surface tension, and hence Marangoni convection [76]. Depending on the nature of these convective currents, this can lead to the formation of a shallow and wide melt pool to a narrow and deep melt pool, or somewhere in between. A couple of final considerations to take into account during SLS are the delamination of layers or part warping if process parameters are not applied correctly. Delamination 30
Figure 3.9: Warping of parts due to uneven thermal expansion/contraction [89] of layers occurs due to insufficient bonding between successive layers. This can either be due to poor wetting characteristics in the material, most commonly due to the presence of oxides, or due to insufficient melting. By adjusting the laser energy density such that some of the previously melted layer gets re-melted during each pass, increased bonding between layers will occur and delamination can be avoided. Warping occurs due to thermal gradients in the part and thermal expansion mismatch. When heated the material will want to expand, but may be constrained from doing so due to surrounding material. Conversely, hot material will want to contract when cooling, but may again be constrained from doing so by surrounding material. Both of these scenarios can cause warping in the final part, as depicted in Figure 3.9 [89]. Preheating the chamber and re-melting can relieve these thermal stresses and reduce or even eliminate warping in the final part. Post-process machining or heat treatment may also be used.
3.5 Previous Modeling Attempts
Many researchers have already proposed different ways to simulate this process (see Zeng el al. [150] and Schoinochoritis et al. [112]for a very detailed review of many different approaches). These approaches can generally be lumped into three main categories:
1. Empirical models based off experimental results
2. Continuum based models (finite elements, finite differences, finite volumes, etc.)
3. Particle-scale, discrete models
Empirical models are useful in optimizing process parameters for a specific material and powder size; however, their applications are limited when trying new powders and/or materials. Continuum based models can offer decent part-level simulations but fail to capture powder bed inhomogeneities and the stochastic nature of the process. Particle- scale models can capture more of the physics at the size of individual particles, but can be very computationally expensive. 31
Nelson et al. [94] used empirical data to create a 1D heat transfer model of SLS that can predict sintering depths. Simchi and Simchi and Pohl [120, 121] used empirical results to determine a relationship between energy input and densification during SLS. Kolossov et al. and Dong et al. [59, 19] have created 3D finite element (FE) models for the temperature evolution during laser sintering. The latter model also predicts den- sification. Matsumoto et al. [86] proposed a FE method for calculating the temperature and stress distribution in a single layer of sintered material. Antony et al. [7] used FEM and experimental analysis to look at SLM of 316L stainless steel powders. Investigation was made into the effects of wetting angle and balling. Illin et al. combined the finite element method with empirical correction factors to determine the melt size and temper- ature distribution during laser melting [47]. Dai and Gu [17] used a commercial finite volume software to simulate SLM for copper alloys. Effects such as Marangoni convection and entrapped gas bubbles are included in this simulation, which is compared with exper- iments in predicting resultant part density. Gusarov and Kruth [35] provide an analytical equation for the penetration of a laser into a powder bed as a function of powder bed density and particle size. This work was followed with a finite difference (FD) simulation of heat transfer during SLM [34]. Korner et al. [60] used a modified Lattice Boltzmann approach to produce a 2D model of selective electron beam melting and followed up on this by successively assembling mul- tiple layers using the same approach in a later work [61]. This work showed that loose powder bed density had the biggest effect on final part density and captures the stochastic nature of the powder bed. Khairallah and Anderson [58] use a multiphysics Arbitrary La- grangian Eulerian (ALE) code to fully simulate the SLM process. This approach appears novel in that individual particles are modeled and meshed up in determining the melt pool size. Additionally, the effect of surface tension on melt pool geometry is demonstrated. The main drawback of this approach is the large computational effort required to simulate relatively short time scales. Kovaleva et al. [62] use a discrete element (DE) approach to model individual particles and determine the melt pool size during SLM by summing the total volume of all melted spheres, offering a qualitative depiction of the melt pool. In many of the previously described works, the material was treated as a continuum medium and effective material properties were used. Additionally, due to the difficulty of accounting for localized phase change in FE models, this effect was usually neglected. Empirical models have the drawback of being process and material specific. Meanwhile, particle scale models, such as the one by Khairallah and Anderson [58], are typically extremely computationally expensive. In the following chapter, a reduced-order discrete element model is presented to model the deposition and subsequent laser sintering/melting of the powdered particles. The interaction of a single layer of these DE particles with an underneath substrate is also modeled. The solid substrate is modeled via the FD method. An algorithm for dealing with the change in material properties due to phase changes is presented. This approach has the advantage of eliminating the need of using homogenized effective properties for the powder bed, can capture particle dynamics and the stochastic nature of the powder bed, and yet is reduced order so that computation time does not 32 exceed a few hours (on a laptop computer) for any of the simulations run. This allows for quick optimization of process parameters for different materials and/or powder size distributions. 33
Chapter 4
Powder Deposition and Laser Heating Model Description
A multiphysical modeling approach has been employed to simulate the SLS process for a single layer of particles. A discrete element approach was used to model particle-to- particle and particle-to-wall mechanical and thermal interactions. In this discrete element model, individual powder particles are modeled as discrete, thermally and mechanically interacting spheres as shown in Figure 4.1. Particle to underneath substrate interactions are modeled using the finite difference method for the solid substrate. The temperature of each particle is assumed to be uniform, due to the low Biot numbers and in an effort to reduce computation time. Additionally we assume that the particles are small enough so that the effect of their rotations with respect to their center of mass is negligible to their overall motion. The modeling approach can be characterized in two parts: (1) simulation of the deposition of the powder particles; (2) simulation of the temperature evolution of the particles and underneath substrate after a single pass of a laser beam. The model builds upon a previous work by the author [29] and adapts approaches previously developed in other works by Zohdi [156, 154] and Campello and Zohdi [11].
4.1 Particle Dynamics
A simple model of the deposition of the particles is described in this section. In an effort to simplify the layer deposition process, the authors assume that the particles are being dropped into the domain from a height of approximately 0.3 mm or less. These particles are allowed to settle themselves into a layer by gravity. Starting from Newton’s second law, we can determine an equation for the motion of the i-th particle (starting from a sample of N non-intersecting particles): total con fric env grav mix¨i = Fi = Fi + Fi + Fi + Fi , (4.1) where m represents mass, x represents position, and F represents force. Contact forces con fric env (Fi ), friction forces (Fi ), environmental forces (Fi ), and the force due to gravity 34
Figure 4.1: Discrete element representation of pre-sintered powder particles (SEM image from [151])
grav (Fi ) are considered. con The contact force, Fi , is modeled via a standard Hertzian contact model for inter- secting spheres [52]. This theory assumes that the contact area between the particles is small with respect to the dimensions of each particle and with respect to the relative radii of curvature of the surfaces. Additionally, the strains are considered small and within the elastic limit, and the surfaces are considered frictionless. These assumptions can certainly be justified for the case of elastically loaded, smooth metallic spheres, as is the case in the present work. Note that other contact models exist and may be used depending on the situation, such as the “hard” particle momentum-based model as described by Luding [80]. In the case of Hertzian contact, it follows that if the distance between two particles (i and j) is less than the combined radii of the particles than a particle-to-particle contact force exists: If kxj − xik ≤ Ri + Rj, then 4√ Fcon = − R∗E∗δ3/2n − dδ˙ n , (4.2) ij 3 ij ij ij ij
Nc con X con Fi = Fij , (4.3) j=1 ∗ ∗ where Nc is the number of particles in contact with particle i, R and E are the effective radius and Young’s modulus of the interacting particles given by 35