The Development of Design Rules for Selective Laser Melting

Transcription

1 The Development of Design Rules for Selective Laser Melting Daniel Thomas Ph.D. Thesis

2 The Development of Design Rules for Selective Laser Melting Daniel Thomas BSc. (Hons.) Computer Aided Product Design Director of Studies Supervisor Second Supervisor Dr. Huw Millward Senior Research Officer The National Centre for Product Design & Development Research University of Wales Institute, Cardiff Dr. Richard Bibb Senior Lecturer Department of Design & Technology Loughborough University Professor Alan Lewis Director National Centre for Product Design & Development Research University of Wales Institute, Cardiff This dissertation is being submitted in fulfilment for the requirements for the degree of Doctor of Philosophy for the University of Wales Institute, Cardiff. October 2009

3 Declaration This dissertation is the result of my independent research. Where it is indebted to the work of others, acknowledgement has been made. I declare that it has not been accepted for any other degree, nor is it currently being submitted in candidature for any other degree. I hereby give consent for my dissertation, if accepted, to be available for photocopying and for inter-library loan, and for the title and abstract to be made available to outside organisations. Candidate.. Mr. Daniel Thomas Date Director of Studies... Dr. Huw Millward Date Supervisors.... Dr. Richard Bibb Date Professor Alan Lewis.. Date

4 Acknowledgements I would like to thank the following: My parents for the support and encouragement they have given me throughout my time at university. My research supervisors Dr Huw Millward, Dr Richard Bibb and Professor Alan Lewis for their moral support, guidance and editorial input to this thesis. All colleagues and fellow researchers at The National Centre for Product Design and Development Research (PDR) for their support, and for creating a sociable environment to complete this research. I would also like to thank all staff at PDR for allowing me access to all facilities and technical expertise required during the research. The industrial and academic professionals who took part in research activities, for their expert input and feedback to the research. Finally, I would also like to thank Kevin Hart and his colleagues at MTT-Group for their technical assistance and support during the course of this research study. i

5 Abstract The research reported in this thesis focuses on assisting the design process in respect of end use metallic products produced using the Selective Laser Melting (SLM) technology. The advancements in layer additive manufacturing technologies such as SLM have enabled the manufacture of end use products directly from Computer Aided Design data. Many companies and researchers are exploring the application of SLM in industry for specific applications, such as the mass customisation of biomedical implants and novel lattice structures. However, bridging SLM from research into mainstream manufacturing is not straightforward, as demanding industry standards and compliance need to be fulfilled. Additive Manufacturing (AM) technologies are often perceived by designers as being able to generate all conceivable geometries. However, SLM is not completely freeform as the inherent process difficulties can distort many part geometries, and designers often lack an understanding of these process issues and their effect on the final SLM product. The aim of this research is to address this lack of design knowledge, by developing a set of design rules to allow for more predictable and reliable results when manufacturing parts with SLM. This thesis documents how the design rules were created. Firstly, the geometric limitations of SLM were evaluated through a quantitative cyclic experimental methodology. Part orientation, fundamental geometries and compound design features were explored until self-supporting parts with the optimum part accuracy were achieved. Design rules were then created and evaluated through a series of interviews with industrial and academic design professionals. ii

6 Table of Contents Acknowledgements..i Abstract ii Table of contents..iii List of figures and tables....iv Glossary of terms... v Chapter 1: Introduction 1.1 Introduction...1 Chapter 2: Literature review 2.1 Overview Additive Manufacturing Principles of Additive Manufacturing Rapid Prototyping Overview History Definition of Rapid Manufacturing (RM) Selective Laser Melting (SLM) Overview iii

7 2.5.2 SLM background Principles of SLM Existing SLM research Process development and materials research Applications and novel geometries Need for SLM design rules Design rules Summary 36 Chapter 3: Methodology 3.1 Introduction Aim and objectives Methodology overview Phase 1: Quantitative research methods Selected methods Summary of quantitative research undertaken Phase 2: Qualitative research methods Selected methods Other methods considered Summary of qualitative research undertaken Research novelty.54 iii

8 Chapter 4: Experimental framework 4.1 Introduction Apparatus and parameter optimisation Pilot studies Pilot study 1 surgical guide Pilot study 2 unspecified medical device Pilot study 3 mixed geometry test part Pilot study 4 holes Pilot studies summary Experimental protocols: process and design Experimental conditions Test part design Support generation and removal Post processing: grit blasting Post processing: machining Experimental protocols: measurement and analysis Considered methods unused: Selected measurement protocols: Quantifying part deformation during the build process De-lamination Curl Summary iii

9 4.8 Experimentation overview Supports and orientation Experiment 1: Simple orientation Experiment 2: Surface roughness Experiment 3: Ledges Experiment 4: Angular overhangs (chamfers) Experiment 5: Convex radii Experiment 6: Concave radii Experiment 7: Holes and channels Experiment 8: Tapping and reaming self-supporting holes Experiment 9: Minimum slot and wall thickness Experiment 10: Shrinkage Experiment 11: Stock on material / sacrificial geometry Chapter 5: Experimental results 5.1 Overview Results part A: Fundamental geometries: Experiment 1: Simple orientation Experiment 2: Surface roughness Experiment 3: Minimum slot and wall thickness Experiment 4: Parallel ledge Experiment 5: Angular overhangs (chamfers) iii

10 5.2.6 Experiment 6: Fillet radii (convex radii) Experiment 7: Fillet radii (concave radii) Results Part B: Compound shapes: Experiment 8: Holes and channels Experiment 9: Tapping and reaming a self-supporting hole Experiment 10: Shrinkage Experiment 11: Stock on material / sacrificial geometry Summary Chapter 6: Design rules initiation / development 6.2 Research synthesis: Creating design rules from the experimental data Part orientation and surface roughness Chamfers and Radii: Concave fillet radii and convex fillet radii Curved surfaces Analysis Analysis of results Analysis of negative responses: Summary. 152 Chapter 7: Design rules 7.1 Introduction Additive Manufacturing iii

11 7.1.2 Selective Layer Melting (SLM) Geometry Deformation Support Structure Background Knowledge Orientation Background Knowledge Design Rules: Part orientation and surface roughness Material allowance / geometric compensation Up-facing surfaces: Side wall surfaces: Down-facing surfaces: Minimum size between features Minimum wall thickness Chamfers and radius Chamfers Concave fillet radius Convex fillet radius Curved surfaces Round holes built flat Maximum upright holes without support Self supporting holes Pilot drilling Reaming accurate holes Threads iii

12 Chapter 8: Discussion 8.1 Introduction Literature review Methodology and test methods Two phase methodology Fixing parameters Experimental conditions Test methods Accuracy requirements Experimental Results minimum surface orientation Self-supporting radii Self-supporting holes Variability and repeatability Design rules development Initiating the design rules Design rules evaluation Future design rule optimisation Chapter 9: Conclusions 9.1 Introduction Research objectives iii

13 9.3 Contribution to knowledge Recommendations for future work References Appendices Appendix 1 Appendix 2 Appendix 3 Appendix 4 Appendix 5 Material file containing the parameters used.219 Completed questionnaires..223 Result tables..249 Transcribed interview feedback Related publications by the author iii

14 List of Figures and Tables Figure/Table Legend Page Figure 2.1 Image of the layers of a AM part, and the principle of the support structure. 5 Figure 2.2 Support structures created in Magics support generation software (Materialise, Belgium). 7 Figure 2.3 An example SL physical prototype part assembly. 9 Figure 2.4 SLS aircraft part made using EOS Polyamide material (MWPower, 2007). 10 Figure 2.5 Rapid Manufactured air intake on a Red Bull Air Race performance aircraft (Wood, 2009). 14 Figure 2.6 On the left is an engine intake made using SLS, and the right shows the same part made using hand made methods (3D Systems, 2009). 15 Figure 2.7 Showing a schematic diagram of the SLM process ( 18 Figure 2.8 Surgical cutting guide fitted to patient during surgery (Bibb & Eggbeer, 2005). 23 Figure 2.9 RPD framework fitted to patient cast (Bibb et al, 2006) 23 (Left) X-Ray image showing lower jaw with natural tooth and tooth Figure 2.10 root cavity for replacement, (Right) Sintered / melted structure of the Ti (cross-section) the laser power is 60W, from Tolochko et al 24 (2002). Figure 2.11 Heat sink test parts made using SLM (Wong et al, 2007). 25 Figure 2.12 Benchmark model created by Kruth et al, (2005). 27 Figure 2.13 Further benchmark test parts designed to identify limitations and measure the accuracy of SLM (Vandenbroucke & Kruth, 2007). 28 Figure 2.14 Examples of a 3-dimesional image (Sapa Ltd, 1997) and a detailed 2-dimensional Image (Groover, 2002). 34 Figure 2.15 Table and Image illustrating dimensional variations of a single design (Spencer, 1994). 35 Figure 2.16 Shrinkage that has occurred during Injection Moulding. Both parts shown are the same and there are two possible ways for the 35 distortion to occur (Rees, 1996). This image shows the poor geometry design that caused the Figure 2.17 distortion (A) and the design improvements to avoid the distortion 36 (B, C & D) (Rees, 1996). Image showing the effect of different corner angles of a rotationally Figure 2.18 moulded part. Excessive material causes porosity in the material 36 when the angle is too small (Beall, 1998). Figure 3.1 Flow chart of research methodology. 42 Figure 3.2 Deming s cycle for leaning and improvement (Deming, 1994). 45 Figure 3.3 Flowchart describing the experimental cycle used in this research. 46 iv

15 Figure 3.4 Example of a test part. The angle X is the limitation to be identified. The angle of the geometry was changed until the lowest 47 angle that would build was identified. Table 3.1 Showing the varying backgrounds in the design industry. 52 Figure 4.1 Image of the SLM Realizer system used in this study. 57 Figure 4.2 (Left) Image of the support structures in the small slot, (Right) the surgical guide still supported on substrate plate (Right). 59 Figure 4.3 Surgical guide fitted on SLA model of patient s skull. 60 Figure 4.4 The modified Surgical Guide. 60 Figure 4.5 Illustration of the medical device with supports. 61 Figure 4.6 Illustration of the poor surfaces roughness on the down-facing surfaces. 62 Figure 4.7 Side view of the poor step resolution. 62 Figure 4.8 Surgical device built for a second time, the parts are shown on the substrate plate still supported. 63 Figure 4.9 Image of the surgical device with the step on the rod improved. 63 Figure 4.10 Image of the mixed geometry test part. 64 Figure 4.11 Picture of the hole before (left) and after (right) the support removal. 64 Figure 4.12 Image showing the sink marks along the side of the hole. 65 Figure 4.13 Microscope images of mixed geometry test part. 65 Figure 4.14 Illustration of completed test parts. 66 Figure 4.15 Illustration of the curl occurring on holes, the gaps in the melt are where the hole centre is. 67 Figure 4.16 Shadow graph images of round hole pilot test pieces. 67 Figure 4.17 Illustration of the test part design. 71 Figure 4.18 Illustration of block supports created in Magics Software (Materialise, Belguim). 72 Table 4.1 This table shows the dimensions used for the block supports. 73 Figure 4.19 Illustration of the support structure variables. 73 Figure 4.20 Image of the shadow graph machine used. 76 Figure 4.21 Illustration of the Shadow Graph/CAD measuring process. 77 Figure 4.22 Taylor Hobson Sutronic 3p surface roughness gauge. 78 Figure 4.23 Photo of de-lamination where the layers haven t merged. 79 Table 4.2 This table shows the scores given for curl, and the reason why that score was given.. 79 Figure 4.24 Illustration of curl values 0 to 3, the dotted lines replesent the layers and triangles represent support structures. 80 Figure 5.1 Photograph of the orientations; left is along x-axis, and right is along y-axis. 87 Figure 5.2 Illustration of the 75µm maximum layer overhang at Figure 5.3 Illustration of the three surfaces measured in this experiment. 89 Figure 5.4 Graph illustrating the results from the second roughness measuring test. 90 Figure 5.5 Illustrating of the staircase effect changing at different orientations, 91 iv

16 the higher the orientation the more step occur, resulting in a lower surface roughness image from (Kruth & Vandenbroucke, 2007). Figure 5.6 Image of the hollow cube used for measuring the minimum wall thickness, and the two parts for identifying a minimum slot size. 92 Figure 5.7 Image showing the minimum wall thickness measured on the shadow graph. The top image is the cubed slots and the bottom 93 image is the round slots. The results are clear in this image. Figure 5.8 The completed SLM parts on the substrate plate. 93 Figure 5.9 The shadow images of the parts that had completed the test build. 94 Figure 5.10 The curl occurring during the SLM builds. The left image shows no curl effect as right shows the curl starting point. 95 Table 5.1 Showing the results of the chamfer measurements. 95 Figure 5.11 Images of the shadow graph measurements and microscope images. The visual alignment at 55 is 1.5, and at 60 there is no deviation. 96 Figure 5.12 Shadow graph images compared against the actual CAD data. 97 Figure 5.13 Illustrates the three different size parts built. The overhang sizes are 5mm, 10mm and 15mm, the tangents changed for each test piece. 98 Table 5.2 The accuracy and the curl level of the convex test parts, the blank boxes were not measured as they did not build. 98 Figure 5.14 Shadow images of the three different curl values. This image illustrates that the less curl the more accurate the part is. 99 Figure 5.15 Illustration showing the number of layer increase as the convex radius size increases, the number of layers below 45 also increases. 100 Figure 5.16 Images showing the shadow graph images of the tangential radius, the images are measured against the original CAD data. 101 Figure 5.17 Illustration showing the overhanging ledge sizes and the position where the tangent is changed. 102 Table 5.3 Showing the accuracy related and curl level of the concave test parts, the greyed out boxes are not shown as the geometry was 103 unsuccessful and failed. Figure 5.18 Shadow images of the three different curl values. This image illustrates that the less curl the more accurate the part is. 103 Figure 5.19 Illustration showing the number of layer increase as the concave radius size increase, the number of layers below 45 also increases. 105 Table 5.4 Showing the increase in height when the tangent increases. 105 Figure 5.20 Shadow images and original hole designs. 107 Figure 5.21 Showing shadow image of the equilateral and isosceles triangles. 108 Table 5.5 Results and geometrical details of the equilateral and isosceles triangle test parts. 109 Figure 5.22 Shows the possible changes in peak height when under constrained. 110 Figure 5.23 Graph of the results taken from attempt Figure 5.24 Actual SLM test pieces on substrate plate. 111 Figure 5.25 Graph showing the centre point of the max and min tangents. 112 Figure mm and 2mm radius holes built using SLM. 112 Figure 5.27 The previously concluded curves and lines where the curves were 113 iv

17 sectioned. Figure 5.28 Shows that 3.65mm overhang will give a 3.48mm height for both small and larger curves. 114 Figure 5.29 An illustration of the 28 curve having a lower height than at 40 when the same overhang values are used under 3.65mm. 115 Figure 5.30 Showing the height is the same as the offset at 5.18mm on the 40 curve. 115 Figure 5.31 Illustration of the 45 position at a maximum 5.18mm. 116 Table 5.7 Overhang sizes for peaks and the associated geometry feature. 117 Figure 5.32 Three example holes (R4, R7 and R10) with their associated peak geometries. 117 Figure 5.33 Picture of completed holes, R1 to R Figure 5.34 Shadow of 4mm (± 0.2) and 13mm (± 0.1) radii. 118 Figure 5.35 Shadow images of 1mm and 2mm radii. 119 Figure 5.36 Image showing the M6 (left) and M12 (right) holes after being tapped Figure 5.37 Image showing the M6 (left) and M12 (right) holes after reaming. 120 Figure 5.38 Image showing the shrinkage / distortion on the pilot test pieces. 121 Figure 5.39 Shadow graph images showing the distortion on wall thicknesses of 1mm to 10mm. 122 Figure 5.40 Shadow images of parts with curl 0, 1 and 2, the distortion is evident and increases with the curl values. Figure 5.41 Photo of the three cuboids after they were machined on all surfaces. 125 Table 5.7 Results from stock-on-material experiment, showing how much material was removed from each part. 125 Figure 6.1 Illustration showing what experiments were used to develop each design rule. 134 Figure 6.2 Illustration showing the relationship between orientation and surface quality. 136 Figure 6.3 An illustration of the three concave curves with intersection lines in increments of 1mm. This was created during the self-supporting 137 holes experiment. Figure 6.4 An illustration of the three convex curves with intersection lines in increments of 1mm. This was established after as part of the design 138 rule synthesis. Figure 6.5 Illustration of curved surfaces broken into individual less complex radii. 139 Table 6.1 Describing the professional profiles from each respondent. 141 Table 6.2 This table summarises the responses of the semi-structure interviews. Each response is categorised as positive or negative 145 together with the Respondent number. Figure 7.1 Illustration of a solid block sliced into layers. 154 Figure 7.2 Photograph of SLM during processing. 156 Figure 7.3 Three choices of orientation depending on requirements. 159 iv 124

18 Figure 7.4 Illustration of the relationship between surface quality, orientation and angled surfaces. Surface and the facing direction are identified 160 by hatched line cross sections. Figure 7.5 Illustration of the surfaces described in design rules. 162 Figure 7.6 Showing the three sizes of the parts described in Table Table 7.1 Shows how much material to add to the CAD data on down-facing surface to make the material removal in Table 7.2 possible. 164 Table 7.2 Shows how much material is required for removal to achieve flat and dense surfaces. 164 Figure 7.7 Example of a 0.3mm minimum gap feature. 165 Figure 7.8 Illustration of minimum wall thickness. 166 Figure 7.9 Illustration of a ledge and alternative geometries, chamfer, and radii. 166 Figure 7.10 Example of chamfers, 45, 55 and Figure 7.11 Illustration of angles relative to the substrate plate. 168 Figure 7.12 illustration of a tangential radius (traditional radius). 169 Figure 7.13 Illustrations of the details presented in Table Table 7.3 Table of dimensions for designing self supporting radii. 170 Figure 7.14 Illustration of tangential convex radius. 171 Figure 7.15 Illustration of the details presented in Table Table 7.4 Table of dimensions for designing self supporting convex fillet radii. 172 Figure 7.16 Illustrates simple radii as a curved surface. 173 Figure 7.17 Illustrates two complex parts with upward and downward facing curved surfaces made of both concave and convex radii. 173 Figure 7.18 Diagram showing the segmentation of a complex curve into convex and concave radii. 174 Figure 7.19 Illustrates the smallest hole at a vertical orientation. 175 Figure 7.20 Illustrates all round holes that can be built with no support. 176 Figure 7.21 Shows holes larger than 7mm curling away from the powder bed causing an obstruction to the powder recoat system. 176 Figure 7.22 Example of curl distortion after the SLM build. 177 Figure 7.23 Illustration of 0.5mm sag (distortion) on Ø7mm hole. 178 Figure 7.24 Illustration of the hole geometry split into 2 components. 179 Figure 7.25 Examples of self-supporting holes designed using the following design rules. 180 Figure 7.26 Illustrates the dimensions given in Table Table 7.5 Table of holes from 2mm to 30mm, all dimension of the holes are highlighted. 181 Table 8.1 Each step of the shadow graph protocol and the possible errors that could accumulate and cause inaccuracy. 189 Table 8.2 The reduced steps of the shadow graph protocol. 190 Figure 8.1 Illustration showing weld spatter affecting the touch probe CMM. 191 Figure 8.2 Minimum orientation angle with 0.075mm layer thickness and 0.075mm layer overhang. 193 iv

19 Figure 8.3 Residual stress distortion caused by scanning strategy as explained by Mercelis & Kruth (2006). 194 Figure 8.4 Down-facing sections of a convex and concave tangential radius. 195 Figure 8.5 Two minimum geometric limitations stacked together. 198 Figure 8.6 Weld spatter occurring during the build of an SLM test part. 200 Figure 8.7 Flow Chart showing that the designer applying the rules is the most efficient way of applying the rules. 205 Figure 8.8 Process showing the most effective implementation of the design rules. 206 iv

20 Glossary of Terms 3D Systems 3DP 3M ESPE 3M Lava Scan ST Design System 3TRPD AM ASME ASTM Boeing A USA based company who produce Stereolithography equipment. 3-Dimesional Printer Specialist supplier of dental products A dental 3d-Scanner that assists the design of patient specific dental products A Rapid Manufacturing / Prototyping bureaux Additive Manufacturing: a general term used for manufacturing parts in layers. AM is an umbrella term for RP and RM. American Society of Mechanical Engineers American Society for Testing and Materials The Boeing Company is a major aerospace and defence corporation CAD Computer Aided Design Ciba-Geigy Company that collaborated with 3D-Systems to develop SL resins and the STL file format. CMM Coordinate Measuring Machine CNC Computer Numerically Controlled machining. Concept Laser A company who produces Laser Cusing Machines GmbH DFM Design for Manufacture DMLS Direct Metal Laser Sintering (Trade name for EOS GmbH process) DTM Rapid Prototyping Vendor EOS GmbH Electro Optical Systems. A German based producer of rapid prototyping machines. F&S Fockele and Schwartz FDM Fused Deposition Modelling. An RP process that builds in plastic FEA Finite Element Analysis IR Infra-Red Laser Cusing Trade name for Concept Laser GmbH process LOM Laminated Object Manufacture. An RP process that builds in paper. LS Laser Sintering (same as SLS; SLS is a registered trademark of 3-D Systems) Magics STL manipulation software by Materialise. Marcam CAD / CAM software developer Material File A file that contains all process parameters on the SLM Realizer. Materialise A software developer and supplier of rapid prototyping services based in Belgium. Maxillofacial Surgery to correct a wide spectrum of diseases, injuries and defects in the head, neck, face, jaws and the hard and soft tissues Maxillofacial Surgical specialty concerned with the diagnosis and treatment of v

21 (surgery) MTT (previously MCP) NPD Objet ODM Orthopaedic Osseointegration PDS Photopolymer Photoshop RM Rolls Royce Aero- Engines RP RPD Sandvik Opsrey SL SLA SLA-250 SLA-500 SLM SLS STL TGM Ti UV VisCAM diseases affecting the mouth, jaws, face and neck. SLM developer and manufacturer New Product Development Objet Geometries RP system that manufactures in a photopolymer. On Demand Manufacturing Surgery concerned with conditions involving the musculoskeletal system. A fixed, bone anchored retention method for attaching external or oral prostheses. Product Design Specification A polymer that cures and solidifies when exposed to light Digital image manipulation software by Adobe Systems Inc. Rapid Manufacturing Aerospace engine division of Rolls Royce. Rapid Prototyping. Removable Partial Denture framework. A company that produces gas atomized metallic powders Stereolithography Stereolithography Apparatus A 3D Systems Stereolithography machine. A 3D Systems Stereolithography machine. Selective Laser Melting. An RP process that builds in metal. By MCP-HEK, Germany. Selective Laser Sintering. An RP process. Stereolithography file format. Defines 3D volumes in faceted triangles. Temperature Gradient Mechanism Titanium metal. Ultraviolet light spectrum. CAD software created by Marcam v

22 Chapter 1: Introduction Product development and manufacturing technologies are all aimed at improving productivity by improving efficiency and product quality through the entire manufacturing process, from concept and detail design to quality control and packaging. The introduction of Additive Manufacturing (AM) technologies provided new insight to the design and development of products, as visual aesthetic and ergonomic models could be produced with minimal labour and tooling requirements. Since the introduction of what were originally referred to as Rapid Prototyping (RP) processes, the technologies have advanced and the availability of manufacturing-grade materials has increased. This has resulted in a growing interest in using the techniques to produce end-use parts, often referred to as Rapid Manufacturing (RM). RM is still an emerging manufacturing paradigm that has led to vast research areas in academia and industry. Selective Laser Melting (SLM) is an AM process, which has advanced from Selective Laser Sintering (SLS) into a process potentially capable of manufacturing high-value, low volume, end-use parts from an increasing inventory of metals and alloys. Research into SLM has been focused on the physics, metallurgical, and advanced applications of SLM rather than research from a designer s perspective. There exists a gap in the available knowledge, in that there is very little design experience informing SLM part creation. Designers are unaware of the limitations and possibilities of the SLM process and therefore, design parts to be made using incompatible design rules from familiar manufacturing processes such as casting, milling and injection moulding. This has been recognised as a barrier to the growth of AM and in particular SLM (Wohlers, 2009). 1

23 This research was conducted to address this gap in knowledge, in particular to develop design rules that address the lack of design knowledge and design experience relating to SLM. The research undertaken has been conducted from a designer s perspective to design within the limitations of SLM through initiating draft design rules. This research has contributed to new knowledge as it is the first study that fully evaluates the geometrical limitations of SLM, and the first to develop design rules that are specifically for SLM to be used as a RM process. The research presented in this thesis was completed using an MCP Realizer 250 SLM machine and the material used was 316L stainless steel. The aim of this research is to develop a first draft of process-specific design rules for SLM. To accomplish the aims of this research, the following objectives were addressed: Evaluate the SLM process, and identify the geometric limitations of SLM; Perform analysis and synthesis of the experimental results and create a first draft of design rules appropriate for use in design practice; Evaluate the design rules to establish their effectiveness and inform their further development. All research undertaken to address the aims and objectives of this study are described as chapters two to eight. The following chapter describes a review of relevant existing literature. 2

24 Chapter 2: Literature Review 2.1 Overview This chapter presents an introduction and background to Additive Manufacturing (AM) developments and technologies. This chapter also explains the technological advancements, research and industrial applications that have progressed the use of AM from RP to RM. SLM has become a key AM process; an introduction to SLM is presented, along with an in-depth review of the existing SLM literature. This review is described in two sections: the first explaining existing research on the processing and physics of the SLM technology, followed by a review of research into applications of SLM. Finally, this chapter provides evidence from existing literature that product designers need to be educated with design rules to advance the use of SLM as a RM process, in order to exploit all possible applications and benefits in high-value product design. 2.2 Additive Manufacturing AM is an umbrella term given to all technologies that manufacture parts by adding material in layers, as apposed removing material in more traditional subtractive processes. During the time of this research, the terminology for manufacturing layer by layer has been standardised as AM by the ASTM committee F42 on Additive Layer Technologies ( 3

25 Initially, AM technologies were restricted to the manufacture of models and prototypes leading to the widely accepted term, RP, which was for many years a term used to refer to all layer additive manufacturing processes. More recently, advances in materials, processes and machine hardware meant that parts could be produced with sufficient mechanical properties to allow functional applications. This enabled layer additive technologies to be used to manufacture end-use components, which led to the term RM being adopted to differentiate the fully functional nature of the parts being manufactured from the previous RP models and prototypes. Now, AM is the general term used and RP and RM are used to describe the respective application of AM technologies Principles of Additive Manufacturing AM technologies produce parts by the polymerisation, fusing or sintering of materials in predetermined layers without the need for tools, as opposed to subtractive processes where material is removed from stock billet. Using AM enables production of geometries that are almost impossible to produce using other machining or moulding processes. The process does not require predetermined tool paths; draft angles or undercuts and internal part details can be designed and produced. The layers of all AM parts are created by slicing CAD data with specialist software (see Figure 2.1). All AM systems work using this principle; however, the thickness of the layers is dependent on the parameters and the system used. Machines that are used commercially in widespread applications typically use layers from 16µm up to 200µm. In all AM processes, the layers are visible on the surface of the part, which influences the surface quality of the finished part. This is known as the staircase effect and is a relationship between the layer thickness and the orientation of a surface. The thinner each layer is, and the more vertical the orientation of a part wall, the smaller each step will be, 4

26 which results in the smoothest surface quality. Also the thinner the layer is the longer the processing time and the higher the part resolution (Jacobs, 1992). Figure 2.1: Image of the layers of an AM part, and the principle of the support structure. The layers of an AM part are built up one on top of another in the z-axis. As one layer has been processed, the work platform and substrate plate is lowered by one layer thickness in the z-axis and a new layer of material is recoated by using a number of different methods. With resin-based systems the parts submerge in the resin by one layer thickness and a traversing edge flattens the resin before the material is processed. With powder based systems, powder is deposited and spread using a traversing edge or a roller, or the part material is deposited through a print cartridge or a nozzle, which only deposits the support and part material required. The time that it takes to recoat each layer with new material can be equally long, or even longer than the time it takes for the layer to be processed. For this reason, the most efficient way to produce parts is to build multiple parts together so that the material recoating time remains the same as building just one part. For optimum build efficiency, nesting software has become available to orientate and position parts within a virtual representation of the AM system s build envelope so that the maximum amount of parts 5

27 can be built together. Currently available examples of nesting software include VisCAM RP (Marcam, Denmark), and SmartSpace that is used within Magics (Materialise, Belgium). Some parts made using AM technologies require a support structure (Figure 2.1) for fixing the part into position on a substrate plate during the build process. This support structure also prevents the part layers from lifting away from their axis as a result of material shrinkage. All AM systems use a different type of support structure that are designed to be most effective with a specific material or technique that the system uses to build parts. The most commonly used support structures are thin, scaffold-like structures with small pointed teeth for minimising the amount of part contact so that they can be broken away from the part easily using hand tools. Examples of different scaffold support structure designs are shown in Figure 2.2; the supports shown were created in Magics support generation software (Materialise, Belgium). Figure 2.2: Support structures created in Magics support generation software (Materialise, Belgium). 6

28 This scaffold-like support style is used in AM systems like Stereolithography (SL), SLS, SLM and Fused Deposition Modelling (FDM). FDM can also produce supports in a soluble material that can be dissolved in a water-based solution after the parts are produced. Three-Dimensional Printing (3DP) sometimes requires a larger support structure to be used, and in some cases the downward facing surfaces of the parts are completely encased in support material that can be removed by breaking away or using a high-pressure water jet. 2.3 Rapid Prototyping Overview RP is the name given to the use of AM technologies to produce physical models with limited functionality directly from three-dimensional CAD data. RP parts often have low functionality and are commonly used as visual aids within product development. However, some RP parts have a limited amount of short-term functionality for more extensive prototype testing depending on the material selection. The materials used for RP are nearly all polymer-based, which allows for limited part functionality. Also starchbased materials or paper are used which allows for very little functionality. The ability to manufacture single or multiple physical models has revolutionised product development, enabling the reduction of the product life cycle time in a range of industries from automotive to medical. 7

29 History RP began in the 1980s with the development of SL, which involved curing photopolymer resins by means of an ultra violet spectrum (UV) laser. This process was patented in 1986 by C. Hull, and together with F. Reed, he created the company known today as 3D Systems (Jacobs, 1992). Collaboration between 3D Systems and Ciba-Geigy was formed to develop SL photopolymer resins, and also the 3D Stereolithography (STL) file format, which is still used today to provide data to all AM systems. The first commercially available Stereolithography Apparatus (SLA) was introduced at a show in Detroit in November Further developments and the increasing recognition of RP resulted in the release of the SLA-250 in 1989, and the SLA-500 in 1990, which allowed the manufacture of larger prototypes (Jacobs, 1992) ). An example SL part assembly is shown in Figure 2.3. Figure 2. 3: An example SL physical prototype part assembly. 8

30 Research at the University of Texas resulted in the development of the SLS technology, which was followed by the first commercialised SLS in 1992 through a company named DTM. In 2001 DTM were bought out by 3D Systems, which secured 3D Systems place as market leaders within the AM industry. Since SLS was introduced to industry, 3D Systems have been up against several competitors with the largest being EOS GmbH, who has named their system Laser Sintering (LS) because SLS is a registered trademark of 3D Systems. The SLS technology works by selectively sintering fine powder materials directly from CAD using an infrared (IR) laser. SLS mainly processes a number of thermoplastic materials such as nylon (polyamide), glass filled nylon, aluminium filled nylon (Alumide) and polystyrene, which means that the parts produced have good mechanical properties and can therefore be used as functional parts as well as visual prototypes. Polystyrene models can be used as a sacrificial pattern for investment casting as it leaves very little ash, and the filled polyamide materials have added strength and can be used in rapid tooling applications. An example SLS aircraft part made using EOS Polyamide material is shown in Figure 2.4 9

31 Figure 2.4: SLS aircraft part made using EOS Polyamide material (MWPower, 2007). The infrared laser scans the cross sections of each layer, sintering the powder particles and layers together. After the first layer is sintered, the substrate drops one layer thickness in the z-axis for recoating. The build chamber is preheated and all oxygen is replaced with inert gas such as argon or nitrogen to ensure that there is no oxygen in the sintering process. This prevents oxides from burning material, and eliminates the inherent risk of fire caused by handling fine powder. The build time for sintering each layer and depositing material may be rapid, but the overall process times are increased because of the need to heat the powder which can take around two hours, and then waiting for it to cool down before the parts can be removed from the machines, which can take up to twelve hours. Another drawback of SLS is that 50% of the processed material must be mixed with virgin material after every process cycle because it degrades as a result of the high temperatures required. The ability to sinter polymer powders using an infrared laser advanced into the sintering of metal powders in the late 1990s. Specially developed alloys and metals with polymer 10

32 based binding materials were developed to use SLS for Rapid Tooling. The materials had poor mechanical properties when compared to the traditionally used tool steels and were only suitable for limited small to medium production volumes. The material required post processing to create dense parts. What was known as the green part firstly needed to be heated to remove the polymer binder leaving what is called the brown part. A secondary material with a lower melting point such as copper or bronze was then infiltrated into the porous part in a furnace, which resulted in dense material parts that were constructed of two materials. The accuracy of the metal SLS part was difficult to control since stresses were introduced to the parts during each processing stage. A special alloy powder mix material was created by Electrolux (Electrolux, Finland) in conjunction with EOS GmbH that did not develop shrinkage distortions; however, this material was still not comparable to tool steels. The next major development in creating metal parts occurred with the introduction of fibre laser technology. This allowed the introduction of Selective Laser Melting (SLM) since the fibre laser enabled the direct sintering of manufacturing grade metals that were fully melted into dense parts without the need for post process infiltration (see SLM background in 2.4). Since 1991, a number of other systems have been commercialised, including Laminated Object Manufacturing (LOM), FDM, and 3DP. Today, the availability of RP systems has increased, along with the capabilities and materials that can be produced. The large selection of materials that replicate the properties of manufacturing materials has increased the effectiveness of producing physical prototypes in new product development (NPD) (Jacobs, 1996; Wohlers, 2009). 11

33 2.4 Definition of Rapid Manufacturing RM has been described as the use of a CAD-based automated AM process to construct parts that are used directly as finished products or components (Hopkinson et al. 2006, p. 1). Since the turn of the century, research into materials and AM technologies has advanced, and the production of functional, low volume products is becoming increasingly viable within many industries. Many companies are investigating technologies for RM, and are exploring design possibilities to increase the functionality of high-value components, and to decrease the product time to market. The ability to build end use components directly from CAD without the use of tools and multiple process stages as well as the everexpanding inventory of manufacturing grade materials, e.g. titanium, aluminium, ABS, and nylon, has been the key driver in the ongoing development of technologies for RM, The focus of RP has progressed into research and development of RM technologies and applications. However, the advent of RM is not as simple as it first appears. The transition from RP to RM is still ongoing because there are a number of issues that need to be addressed to bridge RM from research to mainstream manufacturing. When parts were made for prototype purposes with RP, quality control and part repeatability were not considered. Since RM parts have intended functionality, industry certification and compliance requirements must be addressed. These requirements include surface finish, accuracy, part repeatability, material control, process data traceability, speed, and machine/material cost. The regulations of high-value industries differ from country to country, which is why Titanium SLM implants are implanted in countries other than the UK, and why safety- 12

34 critical RM parts have not yet taken flight on commercial aeroplanes (Ruffo, 2009). According to Richard Rogers of Rolls Royce Aero-Engines UK (Rolls Royce, UK), through private communication, the potential for RM parts being used within aircraft engines has been recognised. Safety critical parts have been used in test rigs but have yet to be used in flight on passenger aircrafts; however, non-safety critical parts have been used in commercial flights. The decision of introducing RM parts into an engine is based on the potential level of failure and the consequence of that failure. That is, an assessment of whether any particular failure is likely to have catastrophic impact. There are many successful examples of RM within the medical, automotive and aerospace industries (Wohlers, 2009). On Demand Manufacturing (ODM) is a subsidiary of Boeing based in Oxnard, California, that has undertaken certification to produce flight qualified hardware using SLS. ODM are a certified manufacturer for the F/A-18 fighter jet for which they have produced over 14,000 parts. The manufacturing system used is a 3D- Systems Sinterstation SLS machine which has also been certified to meet aerospace and material compliance. The material properties for each production run are tested using tensile test bars that are built alongside the parts (Davidson & Glock, 2009). Designers have very little knowledge of how to design for given AM processes, and existing CAD systems do not allow for the freedom in design and part complexity that is possible with AM. In fact, AM systems can produce parts with such high detail that commercial CAD packages cannot yet cope with the amount of data that needs to be processed. This is a recognised concern within industry because it prevents designers exploring RM design possibilities and exploiting potential advantages (Hague et al. 2006). As previously stated, no safety critical parts have been installed on a commercial aircraft; however, the potential for using RM for such parts are being explored, and there are 13

35 several successful cases of parts being used on Formula One race cars, racing catamarans and the Red Bull Air Race performance aircraft as shown in Figure 2.5. The part shown is an air intake cover that has been rapid manufactured. Figure 2.5: Rapid Manufactured air intake on a Red Bull Air Race performance aircraft (Wood, 2009). Figure 2.6 shows an example RM SLS part made by 3D Systems in the USA for the Oregon State University Formula SAE Team. This part is an engine intake that was made from polyamide, and was lighter and less time consuming to manufacture than their previous manufacturing process of bonding pipes and composite materials. The ability to optimise the design through internal features meant that there was a more efficient use of fuel and air within the part. The part was installed into a racing car and used in competition. 14

36 Figure 2.6: On the left is an enginee intake made using SLS, and the right shows the same part made using handmade methods (3D Systems, 2009). 2.5 Selective Laser Melting Overview SLM is an AM process that is evolving into a viable RM process. SLM has advanced from SLS since the availability of fibre laser technologies and is constantly developing through vigorous in-house and university-based research. Powder particles can now be fully melted to form dense parts rather than less functional sintered parts as with previous metal SLS. The materials that can be processed using SLM are potentially capable of manufacturing end-use parts with industrially acceptable mechanical properties directly from CAD, such as 316L stainless steel as used throughout this research. Since the developments in materials, hardware, technical knowledge and SLM building techniques, many industries have recognised opportunities for fully functional parts to be manufactured using SLM. Several university spin out companies have been established 15

37 that rely solely on producing SLM parts, such as dental crowns. Exploiting the use of SLM can allow the design of components to be optimised to create functionalities beyond the capabilities of any previous technologies. However, for the SLM technology to be placed into a RM category and for the process to gain greater acknowledgment in industry, many process changes and developments need to be fulfilled and the process needs to prove itself as being a viable, reliable, repeatable and cost-effective manufacturing solution SLM background The first SLM system was introduced by Fockele and Schwarze (F&S) of Germany in This was a stainless steel powder-based SLM system that was developed in cooperation with the Fraunhofer Institute for Laser Technology (Wohlers, 2009). Later, F&S teamed up with MCP (now MTT Technologies Group) to be the first to commercially release a direct metal system with the MCP Realizer 250 Selective Laser Melting machine in This machine was then replaced by an updated SLM Realizer 250, and in 2005 the SLM Realizer100 higher resolusion machine was released. Since the release of the MCP Realizer SLM, other machine manufacturers (Trumph, EOS, and Concept Laser) released systems naming their process with the different name of Direct Metal Laser Sintering (DMLS) and Laser Cusing. SLM is the name that has become generic to all processes, and the name used in this research. In 2001, Concept Laser (GmbH) released the M3 Linear, and M1 Cusing, plus more recently the M2 Cusing systems for manufacturing reactive materials such as aluminium and titanium alloys. In 2003 EOS released a machine named the EOSINT M 270 DMLS. This is the most commonly installed machine for direct metal fabrication (Wohlers, 2009). In 2008, 3D 16

38 systems and MTT announced an agreement to distribute SLM machines in North America. In 2008/9 MTT revealed new versions of their SLM with a new SLM 250 and SLM Principles of SLM The SLM process involves the slicing of CAD data and adding support structures just like all other AM processes. Building SLM parts must be completed in an inert gas atmosphere, replacing all oxygen in the building chamber with gases such as argon. A schematic of the SLM process is shown in Figure 2.7. The hardware of the machine used in this study functions in a similar way to SLS systems. The metallic powder material (316L stainless steel in this research) is deposited in a predetermined thickness ranging from 50µm to 75µm across a substrate platform of similar material. A fibre laser then draws each layer of the sliced CAD data fully melting the powder particles together. The substrate platform then drops one layer thickness in the z- axis before the material is recoated, and the process is repeated until the entire build is complete. Newer versions of the SLM system use layer thicknesses starting from 20µm. 17

39 Figure 2.7: Showing a schematic diagram of the SLM process ( Once the SLM process is complete, the substrate is removed from the build chamber and the supports and parts are removed. The supports need to be carefully designed because they can be difficult to remove as they are the same dense metallic material as the part Existing SLM research SLM is still considered a new technology and the process is constantly developing due to research into materials, part mechanical properties and process methods. As quoted by Sutcliffe (Nathan, 2006) there is worldwide interest in developing a standard rapid manufacturing process based on selective laser melting. However, very little research has been published on the production of design rules for SLM. The existing SLM research and literature is based on process developments and applications of SLM as an RM technology. Most research was published during the development of the SLM process in collaboration with universities, which resulted in the release of commercial SLM machines. Since the 18

40 introduction of the SLM Machine, the research focus has shifted from just material and process developments, to research into SLM applications. In addition to university-based research, the SLM process is now perceived as a viable production-capable process in the biomedical and aerospace markets because of the ability to manufacture complex designs directly from CAD using materials that are already used in both fields, such as titanium, aluminium and various steels (Wohlers, 2009). The inhouse research is confidential and has limited published literature in comparison to other areas of research, as the machine suppliers are protecting their potentially valuable intellectual property Process development and materials research SLM research is mainly based on engineering principles and identifying the physics behind the SLM process. Researchers using the SLM machine have investigated how to control the melting of the powder, and the physics of the melt-pools generated during the process. Thesis work by Rombouts (2006) investigated a way to expand the number of materials that can be processed in SLM systems. This was done by identifying the physical behaviour of materials during SLM processing, and understanding the processes at a microscopic level. The main quality attributes for SLM parts such as surface roughness, strength, accuracy, hardness, density and residual stresses have to be examined thoroughly when a new process is being assessed for industrial applications (Rehme & Emmelmann, 2005). Rehme and Emmelmann (2005) stated that the most important part attribute is that they achieve the highest density possible. It appears that the main aim of the parameter and process optimisation research is to control the density of SLM parts, including work by 19

41 Childs & Hauser (2005) and Wright et al. (2006). They were able to control part density by ensuring optimised scan quality characteristics of single melted scan lines, by measuring and eventually controlling the stability of the melt pools and the heating/cooling temperature gradient. An advancement of just building single scan lines and layers was manufacturing whole components whilst controlling the SLM parameters and scan strategies. This research was completed to identify the effects that different parameters and scan strategies had on part geometry, material porosity (Morgan et al. 2004), (Kruth et al. 2005), and also to identify a method to control the surface roughness of SLM parts (Mumtaz & Hopkinson, 2007). The control of process parameters in this study was important in achieving repeatability of the components created during the SLM process. Kruth et al. (2005), state that fewer product deformations and higher process accuracy is achieved when using optimal parameters. According to Rehme and Emmelmann (2005) there are more than 130 process parameters affecting the SLM process, however, only 13 of these parameters are crucial towards gaining optimum part quality. Theoretical calculations of crucial parameters are used to control the amount of energy that is required to fully melt the metal powder being processed. The main parameters are the laser scanning speed, the laser power, layer thickness and hatching distance. Theory suggests that when the values are set for each variable in the equation, the sum should equal the energy density value for the specific processed material. The use of equations for selecting parameters for specific energy density is a limited, inaccurate method, as proven by Tsopanos et al. (2005). In many cases the equations designed to help select process parameters should only be used as a guide to the relationship between the parameters. Therefore, it is important that the theoretical calculations are accompanied by practical 20

42 testing by the user, as small process parameter amendments may be required to optimise the parameter values resulting in an optimum quality of part produced by SLM. All laser/heat based processes are known to induce large amounts of residual stresses due to the temperature gradient that occurs during the process. Residual stress is a recognised drawback that occurs in SLM. According to research by Mercelis et al. (2006) and Kruth et al. (2004) investigating scanning strategies and how residual stress develops within SLM parts, the stress is caused by the temperature gradient mechanism (TGM) that occurs in every melt-pool as the powder melts. Each melt-pool is heated rapidly and then cooled down rapidly causing the material to expand and shrink. As one scan line is melted all melt-pools cool and shrink separately causing tensile stress between subsequent melt-pools and scan lines. As the parts build up in the z-axis, the thicker the material becomes. This prevents the shrinkage distorting the parts, which consequently causes a build-up of stresses that may affect the mechanical and geometrical properties of the part when the parts are removed from the supports, or even at a later date when the parts are used for their intended application. Evidence shows that the environment of the SLM needs strict control as it reflects on part quality. The SLM build chamber is a sealed compartment, which is filled with an inert gas. The flow of the inert gas controls the amount of oxygen in the chamber limiting oxidisation during the process. This was demonstrated in an experiment by Badrossamay and Childs (2007) which resulted in the best part density from a 99.9% argon atmosphere. Oxides that develop on the parts surfaces prevent sufficient wetting, the wetting (powder melting) behaviour improves by eliminating the oxides from the base metal surface; this is why soldering requires a flux to get rid of any oxides (Rombouts, 2006). 21

43 2.5.6 Applications and novel geometries Increasing the inventory of materials that can be processed using SLM will expand applications in industy. Research has progressed from the development of parameter optimisation and SLM physics of 316L stainless steel as used in this study, into research of more advanced specialist alloy materials such as inconel, aluminium, titanium, tool steels and cobalt chrome. This reseach is driven by application possibilities, such as work by Kruth and Vandenbrouke (2007) who investigated possiblilties of procesing biocompatable metals for medical parts, such as implants and prostheses. Other research into materials included testing the wear behavior of tool steel mould materials using various SLM processes, compared to conventional machining methods (Voet et al. 2005). In the UK, the medical industry currently use SLM to manufacture surgical cutting guides as shown in Figure 2.8 (Bibb & Eggbeer, 2005), the cutting guides facilitate pre-surgery planning to reduce the time of operations. In the dental industry, a company named 3TRPD are producing commercially available dental crowns and bridges with a system named 3T Frameworks (3TRPD, Berkshire). This system incorporates the use of a 3M Lava Scan ST design system (3M EPSE, UK) and EOS M270 (EOS GmbH) to produce a service that promises a turn-around of three days. The possibility of building Removable Partial Denture frameworks (RPD) has been demonstrated by Bibb et al. (2006) (see Figure 2.9). This research has shown that it is feasible to produce RPD frameworks from scan data of patient anatomy, but this has not yet become commercially available. Other higher risk medical devices, including titanium and cobalt chrome implants, have been implanted outside the UK in North America (Medical Modelling, USA), Europe, and Australia (Tolochko et al. 2002; Sercombe et al. 2008). Standards and compliance in high-value industries have become a constraint in the advancement in applying SLM to medical devices. 22

44 Figure 2.8: Surgical cutting guide fitted to patient during surgery (Bibbb & Eggbeer, 2005). Figure 2.9: RPD framework fitted to patient cast (Bibb et al. 2006). The findings of research into manufacturing implants using SLM is thatt the surface finish can be manipulated to be selectively porous, and lattice-likee structures can be addedd to areas to promote osseointegration (bonding between the bones and the implant) when implanted. Over 1000 surgeries have been completed between the years 2008 and 2009 with excellent feedback (Ruffo, 2009). A good example of creating this graded density is 23

45 reported by Tolochko et al. (2002), where both full melting and sintering was used to produce a dental root implant. In this case the laser power was dropped from 100W that produced dense metal, to 60 W to only bond the powder particles together in areas requiring a porous finish (sintering) (see Figure 2.10). In the UK no class 3 medical implants of SLM parts have been carried out because the process has yet to meet the ISO standard that represents the requirements for a comprehensive management system for the design and manufacture of medical devices. Figure 2.10: (Left) X-ray image showing lower jaw with natural tooth and tooth root cavity for replacement, (Right) Sintered / melted structure of the Ti (cross-section) the laser power is 60W, from Tolochko et al. (2002). Internal geometries can be designed into products that cannot be produced by any other means of manufacturing. Such features have been demonstrated to have lower manufacturing costs and added performance values over conventional processes. A good example is shown by Tsopanos et al. (2005) with the manufacture of cross-flow heat exchangers made from stainless steel and copper, and also the manufacture of heat sinks made from Aluminium (Wong et al. 2007) (see Figure 2.11). Work by Dotcheva et al. (2009) also investigates the design and testing of lattice cellular structures. The research presented in this thesis contributed to the design of the lattice structures and was proven to be valid through expert review (the relevant paper is presented in Appendix 5). The aim of Dotcheva s research was to investigate the tensile strength of various lattice formations for 24

46 possible application within injectionn mould tooling to optimise mould cooling by using cellular internal designs (Dotcheva et al, 2009). Figure 2.11: Heat sink test parts made using SLM (Wong et al. 2007). One disadvantage of SLM is the significant processing time it takes to melt parts of larger volumes. Research into manufacturing and computer coding novel lattice structuress has investigated the possibility of replacing dense volumes of material with lattices that have the same stiffness and strength properties, and will reduce the time and the cost of processing large volume parts (Rehme & Emmelmenn, 2006). The interest in cellular structures has increased with research efforts being made to identify ways to design and to produce both open and closed structures with a regular and irregular structure (Brooks et al. 2005). The lattice structures created by Brooks et al. (2005) were lightweight with only 6% of the relative density of the equivalent solid shape. These parts were small cuboids and were able to absorb energy from compressive loads of over 2.5kN. Possible applications of the lattice structures include heat exchanges (Wong et al. 2007), orthopaedic implants (Mullen et al. 2009) and ultra-light aerospace components. 25

47 The software and computer coding created to develop the lattice structures have become available as a commercial software package and a plug-in to Magics RP preparation software (Materialise, Belgium). However, due to the complexity of the lattice structures computer hardware requires high processing capabilities that are currently impractical of reach with larger and more detailed structures. This has resulted in more research into methods of reducing the STL file sizes and creating alternative file formats. Research by Ramin & Harris (2008) and Brooks et al. (2005) has investigated automated methods of creating various cellular structures for tissue engineering applications by using an advanced CAD package that does not require additional software like Microsoft Windows because it uses large amounts of processing power. The methods used by Ramin & Harris (2008) resulted in a more efficient generation of the complex shapes with up to 75% time saving. Research into SLM applications, process developments and materials has identified limitations and constraints of the SLM process. However, very little research has been designed to explicitly identify the geometrical boundaries of SLM. Previous research focused on the design of more complex geometries such as surgical implants, lattice structures and heat transfer devices. Kruth et al. (2005) and Rehme and Emmelmann, (2005) both published an investigation into identifying the limitations of the SLM process. The research by Rehme and Emmelmann included the design of simple test pieces to test mechanical propetries of SLM parts. However, the work by Kruth et al. (2005), created a benchmark model that was made on five different metal AM processes, ranging from sintering to fully melting the material. The research completed by Kruth has been firmly rooted in Process Enginneering and Materials Science. The focus of his work has been on optimising the whole SLM process; seeking to develop a more complete understanding of the basic phyics of selective powder 26

48 metling by lasers, thus enabling the optimisation of process parameters. The geometry investigations published by Kruth and his colleagues were completed to evaluate parameter selections by measuring and comparing mechanical properties and material structure, and also to compare different metal AM processes. Kruth s geometric evaluations spanned no further than what is described in this literature review. The focus of the research published within this thesis is different from that of Kruth s. Whilst Kruth has sought to influence actual processing parameters, this research has focused on the geometrical capability of commercially available SLM machines. Therefore the test geometries are the main priority in this research and their purpose is to identify geometrical design capabilities that will enable the development of comprehensive design rules. However, the process parameters were the main focus in Kruth s research, and the geometry test parts he reported were only made as a method to quantify the effect of varoius process parameters with a view to identifying the optimum paramters for manufacturing. Figure 2.12: Benchmark model created by Kruth et al. (2005). The benchmark model shown in Figure 2.12 is constructed of several geometrical features, these include thin walls, overhangs in holes, self-supporting holes, one angled surface and 27

49 a large radius. The geometrical features were chosen so that the limitations of each process could be identified. To improve the geometries only parameter changes were used, that is, there were no changes to the geometries. This ensured that all machines were producing parts to their full potential so that the limitations of each process, and potential manufacturing applications could be identified. Figure 2.13: Further benchmark test parts designed to identify limitations and measure the accuracy of SLM (Vandenbroucke & Kruth, 2007). Further studies by Vandenbroucke and Kruth, (2007) developed more benchmark models (Figure 2.13) that tested a larger range of sizes for holes, overhangs and slots. These test parts were built and measured to identify part accuracy using a Taylor Hobson Form Talysurf surface roughness meter, a tactile probe coordinate measuring machine (CMM) and an optical micro-measurement machine to measure the smallest details. The test pieces designed in Kruth s work were considered in the main experimentation of this study because the design was created to measure fundamental geometries. This included an investigation of simple orientation and other simple geometries that were the building blocks to more complex constructions. The test pieces did not include a full range of sizes for each design feature for example only one hole and square was used to investigate overhanging geometries. Another drawback of using Kruth s benchmark designs was that if one feature on each test part failed because it was outside of the limitations of SLM, the 28

50 entire part would not continue. This would have resulted in no parts to be measured and an excessive use of material. As this study identified the limitations of SLM, it was inevitable that several test parts would fail so that the geometric limitation could be identified. For this reason all design features were designed as individual test parts. It is clear from work by Kruth (2005), that the accuracy of different machines and different parameter settings, results in different geometric limitations. The test parts were not replicated in this thesis study as they were too large to build each time a small geometric change was made on just one feature on the part. Also the design of the parts would have been difficult to measure using the apparatus available within this study (the apparatus selection is shown in Chapter 4). 2.6 Need for SLM design rules Hague et al argue that the advancement of SLM towards a robust RM process is held back by process inherent drawbacks such as part accuracy, surface finish, repeatability and material limitations. Since this citation, advancements in technology, part accuracy and surface roughness have occured, but the inherent drawbacks are still apparent, and designers using SLM are still designing for conventional processes. Design freedom is a major driver of SLM being accepted as a manufacturing process and could result in a reduction in lead-time, overall manufacturing costs, and improve part performance (Hague, 2006). However, very little literature exists that can be used to inform designers of the design limitations and design possibilities of SLM. To obtain repeatability in SLM parts and to control or avoid the process limitations of SLM, it is of high importance to have adequate product design guidelines (Rehme & Emmelmann, 2005). 29

51 Many researchers within the AM industry have identified the potential benefits of design rules. However, there is no previous research that has been aimed at addressing the lack of design rules for SLM, or other AM processes. The work that has been completed to date has been based on the physics and applications of SLM in producing novel geometries. Succesful examples of SLM parts have been designed by people who not only know the SLM process, but also use and study it day-to-day. If a product designer were to use SLM as a manufacturing process, there would be no design rules for them to follow, which may result in a failed part, and/or a part that could have been optimised further for RM. Without designers being able to access design guidance for SLM, they will not be familiar with the limitations of the process. This may result in the SLM process being excluded as an option when manufacturing processes are selected, even though it may be the most suitable process for a particular design. Pullin (2009) reported that none of the design considerations of conventional manufacturing processes apply to AM for part production. One common theme identified from literature by Wohlers (2009), and Pullin and Offen (2008) is that designers in today s industry have become adept at creating products that can be manufactured using conventional processes. Their education and design experience have conditioned them to subconsciously tailor their designs to suit today s common production processes, making it challenging to design outside of normal boundaries. Further evidence of designing for the wrong process was quoted through personal communication with Dr. Wilhelm Meiners of the Fraunhofer Institute of Laser Technology (ILT, Germany) - 90% of parts aren t designed for SLM, they are designed for casting or milling. It is only luck that the other 10% have good geometries for building with SLM. Of the 90%, no parts are designed considering the benefits of RM; therefore, compromising on surfaces is always required. 30

52 The limitations of RM need to be openly expressed to prevent any misconceptions of the process prohibiting future growth of the AM industry. One barrier for growth is the lack of design guidelines (Wohlers, 2009). Wohlers (2009) and Pullin and Offen (2008) reported that design guidelines are needed to encourage designers to explore the advantages of AM, allowing the manufacture of parts with increased functionality and value. It was suggested that design rules need to outline the limitations of the process, such as minimum wall thicknesses. It is recognised that when parts are intelligently designed for AM the full benefits are noticeable (Pullin, 2009). The design of a part determines the number of support structures that need to be used. Out of all the AM processes, SLM has the most difficult supports to remove as they are dense metal. Support structures are needed to anchor the parts to a substrate, to prevent movement during the process and to prevent surfaces from curling up away from the correct geometry. The use of the support structures in AM are reported as being the main restriction on part geometries, and the support placement is equally as important as the part design (Pullin & Offen, 2008). Burton (2005) has taken a different approach to producing design rules, by creating a knowledge transfer tool that enables industrial designers to establish greater understanding of applying design practice to AM processes. This project was established from recognising the need to inform designers of how to design for AM processes. The rules created were principle design guidelines based on a series of design matrices for selecting AM processes, and, ensuring that designers have considered the features that contributed to the processes selection in the first place. This approach to design rules was an appropriate method for ensuring that the correct AM process would be chosen. However, no process specific detailed design rules were developed for any AM process (Burton, 2005). 31

53 2.7 Design rules In 1941, a survey by the American Society of Mechanical Engineers (ASME) revealed the need for designers and engineers to have ready access to technical information on material properties for metals. A series of publications were released, including in 1958 the first design rules publication, Metals Engineering Processes, edited by Roger W. Bolz. Bolz was one of the first to introduce Design for Manufacturing (DFM) principles. The book was used as an aid for designers in enhancing manufacturability of metal parts for a number of metal processing techniques (Bralla, 1998). The use of instructional guidelines advanced into Design for Manufacturing and Assembly (DFMA) principles in the late 1960 s by Boothroyd and Dewhurst (Boothroyd et al. 2002). The principles were aimed at all stages of design, ensuring consideration of manufacturing throughout the design process. The advantages of applying DFMA principles during product design were: reductions in part count and cost; reduction in manufacturing and assembly times; improved part quality; and, product time-to-market. Now that RM has become available, the foreseeable benefits of efficient design are even greater. Design rules play a major role in the improvement of the efficiency and the optimisation of the quality of parts that the process produces. Some design rules are focused specifically on a product that has been identified as being suitable for the manufacturing process. Since their introduction, design rules have evolved with the advancement of manufacturing processes and they are used as an effective way for a designer to select a manufacturing process and to create optimised designs. Design rules are also used as an effective method to deliver new process capabilities to designers and engineers so that a process can be fully explored. 32

54 Detailed general engineering books such as The Pocket Ref (Cromwell Group Holdings Ltd, 2008) and Zeus Precision (Zeus Precision Charts Ltd, 1995) are used as a reference by designers, toolmakers and engineers. These books consist of details such as tapping drill sizes, limits and fits, material, and engineering drawing symbols. The content within the reference books are not focused at any particular manufacturing process and are aimed at being a quick guide and a breakdown of the key points within British Standards that can be applied and work hand in hand with all process specific design rules. Existing process-specific design rules are mainly for conventional processes such as casting, injection moulding and aluminium extrusion. The rules are presented through both a printed book format, and data available on the website of companies whose businesses are based on supplying manufacturing services or manufacturing equipment. The presentation of the process-specific design rules needed to be considered whilst developing the SLM design rules within this study. In particular, there are several attributes from existing design rules approaches that show how information may be presented most effectively through a series of illustrations and descriptions. These points are described as follows: Both two-dimensional and three-dimensional images are used and both are effective in delivering a different depth of detail. The two-dimensional images are more commonly used to illustrate detailed dimensional data. Three-dimensional images are used for assembly, less detailed dimensions, product examples, and to add more realism to the design feature. See Figure 2.14 for examples taken from aluminium extrusion design rules published by Sapa Ltd (1997) and Groover (2002). 33

55 Figure 2.14: Examples of a 3-dimesional image (Sapa Ltd, 1997) and a detailedd 2- dimensional Image (Groover, 2002). Tables in design guidelines are illustrative when there are several dimensional variations of a single geometry. Although tables can appear to be complex, some design rules are dependent on them as long as there is an image to which it corresponds in. Some cases tables do not have an associated image making them ineffective for someone new to a process. An example table and image taken from aluminiumm extrusion design rules is shown in Figure (Spencer, 1994) Figure 2.15: Table and Image illustrating dimensional variations of a single design (Spencer, 1994). 34

56 All design rules show the possible distortions and part deformations thatt are inherent to the manufacturing process. These distortions are caused by poor design and showing them is useful to emphasise why the design rules need to be used. An example of a part distortion presentationn is shown in Figure Figure 2.16: Shrinkage that has occurred during Injection Moulding. Both parts shown are the same and there are two possible ways for the distortion to occur (Rees, 1996). When the distortions are unavoidable they are shown together with a design solution or an alternative design to prevent the distortion occurring. This is shown with an image of a poor design and an image of good design (see Figure 2.17 and Figure 2. 18) Figure 2.17: This image shows the poor geometry design that caused the distortion (A) and the design improvements to avoid the distortion (B, C & D) (Rees, 1996) ). Figure 2.18: Image showing the effect of different corner angles of a rotationally moulded part. Excessive material causes porosity in the material when the angle is too small (Beall, 1998) ). 35

57 An introduction to the manufacturing technology is shown in all design rules. At this stage, illustrations of how the process works are presented to set the scene for the content of the design rules. The introduction to the design rules always presents enough information for the designer or engineer to be assured that the process they have chosen is the correct process. When large blocks of text are used without illustrations in the design rules, they are discouraging to read and can be difficult to follow especially when they are concerning geometrical features. 2.8 Summary The introduction of the RM concept has become an ongoing process since there are many issues including accuracy, build time, mechanical properties and repeatability that need to be addressed to complete the transition from RP to RM. Within the AM industry, advancements in laser technologies and the ever-increasing inventory of manufacturing grade materials has allowed the commercial introduction of SLM. SLM has become a common research theme in engineering research. The majority of the research is based on the physics, material properties and process improvements of SLM, and in many cases, the research is driven by industrial collaborators. Since earlier research efforts, many process improvements have been made because a greater understanding of the process has been gained. However, there has become noticeable competition amongst machine manufacturers, which has resulted in much of the research becoming in-house as companies wish to claim intellectual property rights (IP) on their process developments. Also, collaborative research with industry requires confidentiality agreements, which has reduced the availability of published research. 36

58 Many industries have recognised the benefits and possibilities of applying SLM within their manufacturing process. This has driven research into the development of novel geometries with the aim of enhancing product performance, and exploiting the freedom of the process. This includes the development of lattice and cellular structures, and the production of custom fitting implants. In many cases, the concept of building the geometries is proven; but, in some cases, standards and industry compliance have restricted applied cases of safety-critical SLM parts. For any new manufacturing process to succeed, the design limitations must be considered to enable designers to exploit the possibilities of the process within its boundaries. The benefits of creating design rules for SLM are recognised as being essential to its success. One major problem is that designers only know how to design for conventional processes, and have little or no understanding of SLM. Very little research exists that is aimed at identifying the geometrical limitations of SLM with the exception of work by Kruth et al. (2005, 2007). This work gave a good insight to the fundamental geometries that needed investigation. Kruth s research was focused on improving geometrical accuracy through the optimisation of process parameters, as well as identifying simple process limitations. No attempt has been made to develop design rules that are specific to SLM. The only work on developing design rules was aimed at encouraging full exploitation of the design abilities of general RM processes by using a series of matrices (Burton, 2005). The following chapter describes the methodologies used to investigate the SLM process, and to develop detailed design rules for SLM. 37

59 Chapter 3: Methodology 3.1 Introduction This chapter describes the research methodology used to generate and evaluate SLM design rules. The aim and objectives of this research led to the methodology being divided into quantitative and qualitative methods. This chapter describes the methodology underlying two phases of the research, and the stages that were required to fulfil the research objectives within each phase. Previous literature has identified that no design rules have been created for the SLM process. Therefore, careful consideration given to the methodology choice for gathering the data required for developing design rules within this research. The methodology selection was imperative for completing the quantitative and qualitative research effectively. The merits and disadvantages of all appropriate methodologies were reviewed and this chapter describes the reasoning behind the chosen methodologies, and why other contending methods were not selected. 3.2 Aim and objectives The aim of this research was to develop a first draft of process-specific design rules for SLM. To accomplish the aim of this research, the following objectives were set: Evaluate the SLM process, and identify the geometric limitations of SLM; Perform analysis and synthesis of the experimental results and create a first draft of design rules appropriate for use in design practice; Evaluate the design rules to establish their effectiveness and inform their further development. 39

60 3.3 Methodology overview The initial literature review was completed before the research methods were selected. After the initial literature review, a continuous literature review was undertaken over the duration of the research. The scope of this research entails the creation of design rules for a state-of-the-art technology. The selected methodology allowed the geometric limitations of SLM to be identified and presented to design professionals as design rules for manufacturing using SLM. To develop the design rules, the research objectives were split into two separate research phases. The first phase is quantitative and the second phase is qualitative. Each phase is illustrated in the research methodology flowchart in Figure 3.1. Phase 1: Quantitative research Phase 1 outlines the experimental framework which was the first research objective. As very little previous literature was available for developing the design rules for SLM, the geometric data required for identifying constraints and possibilities of SLM parts needed to be generated within this research. This quantitative approach included pre-experimental SLM machine calibration and SLM process parameter optimisation. The pilot studies and main experimentation required the use of a cyclic research methodology, which allowed iterations of a single experiment, with controlled parameter changes to establish geometrical data of design features. After the completion of pilot studies, definitive measurement and analysis instrumentation were chosen and a series of protocols were produced. Patton (1990) states, The validity of quantitative research is on careful instrument construction to be sure that the instrument measures what it is supposed to measure. This research phase was used to identify the inherent geometrical limitations of 40

61 SLM manufacture and to address the limitations by performing experiments and analysis to produce raw geometrical data. This raw data was refined in Phase 2 for developing the design rules that needed to be comprehensive for all common geometries (Phase 2). Figure 3.1 illustrates the stages within each phase of the research. Phase 2: Qualitative Research The aim of the second phase of the research was to develop design rules based on the raw data gathered in Phase 1. To initiate the design rules, a synthesis of the results from Phase 1 was required to ensure that the data could be interpreted and used as design rules. This synthesis was the critical bridge from Phase 1 to Phase 2 as the success of the design rules was dependent on whether or not the rules were able to be used by design professionals in practice. Therefore, further developments were required for the rules to be analysed by design professionals and to gain feedback that was used to conclude the design rule development in this research (See Figure 3.1). 41

62 Figure 3.1: Flow chart of research methodology. 42

63 3.4 Phase 1: Quantitative research methods Selected methods As many research methods were available, it was necessary to make a selection that would be most advantageous to the research objectives in Phase 1. The following methods are described in three stages, and were selected to complete Phase 1 of the research as illustrated in Figure Stage 1: Initial surveys: At the beginning of this research study, less than 15 MTT SLM machines were installed within Europe and very little literature existed. Therefore, questionnaires (see Appendix 2) were an obvious method for finding the answers to issues of interest and concern about SLM in preparation for further research. Questionnaires were presented to the recipients at the SLM User Group Conference in Paderborn, Germany, during September 2007 to gain an understanding of existing SLM users knowledge and experiences. The respondents expertise provided a wealth of knowledge for this initial study, as the recipients included key individuals that developed the first commercial SLM machines. Other recipients expertise included SLM development research and the commercial use of SLM from universities and sevice bureaus. As suggested by Blaxter et al. (2003), the questionnaires were given to and collected from the respondents in person, with many questions open ended. This created an opportunity to gather additional comments and opinions on the overall study. One possible disadvantage of the initial survey is that the research and commercial use of SLM was competitive amongst the respondents, and therefore, the depth and accuracy of the responses may have been affected as machine capabilities may have been embellished. The results of the initial surveys were considered in the experimental planning in Chapter 4, and the completed questionnaires are in Appendix 2. 43

64 Stage 2: Pilot studies: The aims of the pilot studies were to inform and make quantitative observations about the experimental methods proposed for the main experimentation. These studies were completed at an early stage in the research process (see Figure 3.1). A range of parts with a range of design features where selected for the pilot studies. At this stage of the research the results from building the pilot study test pieces could not be predicted. Therefore, the results provided data that informed the main experimentation of geometries that needed further investigation. Each pilot study was completed iteratively using elements of Deming s cycle (Deming, 1994) described in Figure 3.2. The advantages of the research cycle became evident as some pilot study builds failed, but were repeated successfully with controlled changes such as orientation. After each pilot test part was made, they were analysed using hand measurement tools, a shadowgraph, and a microscope with a digital camera attachment. The measurements were used to identify the most effective analysis method that provided repeatable and accurate geometrical data, which was then used to create a measurement protocol for all main experimentation. Along with the measurement protocol, the pilot studies were used to create a standard test part design protocol. The parts were designed so that they could be measured effectively using the measurement protocol. To see pilot studies, refer to experimental framework, in Chapter Stage 3: Main experimentation The experimental requirements came from both the research objectives and the pilot study conclusions. Action Research (AR) is a systematic approach to defining, solving and evaluating problems and concerns, as explained by Blaxter (2003). AR is commonly used 44

65 as a social-science focused method of solving problems, although, the iterative process cycle that is used in AR can be applied across a wider range of research initiatives. Research into the use of AR led to the selection of another systematic method developed by Deming (1994). This methodology is a manufacturing focused design-of-experiments cycle named the Deming s Cycle for Learning and Improvement. This cycle (Figure 3.2) is described as a Plan-Do-Study-Act (PDSA) process and can be used as a technique for performing a test aimed to make improvements in a manufacturing context. Figure 3.2: Deming s cycle for leaning and improvement (Deming, 1994). Figure 3.3 illustrates the experimentation cycle as used in this research, and is adapted to ensure that the experimentation would identify limitations. The geometrical featuree to be investigated was selected and test parts were created for analysis. The results were then reviewed and the experiment was repeated until the limitations of the geometry in question were identified, and the data could be used to inform subsequent experiments. If further testing was required then the appropriate dimensional values were changed and the cyclic process was repeated until the objectives were met. The experiments were planned so that the results would inform a starting point for the following experiment in the series. 45

66 Figure 3.3: Flowchart describing the experimental cycle used in this research The benefit of the experimental cycle (Figure 3. 2) for this research was clear, as the geometrical investigations would need to be addressed step-by-step to record the in-build accordingly. This step is described as adopt, amend or abandon by Deming (1994). The parts would be abandoned when the geometrical limitations of each test could be identified and conclusions could be made, or the experiment would be repeated if the results did not identify the geometric limitation of a test part and the resultss could not inform subsequent experiments. An example of a dimension that was invetigated is shown in Figure 3..4, where the angle labelled X is the dimension for which the limitation was investigated. Various values for X were built until the lowest possible value for X was identified. distortions and measured accuracy, and to make dimensional amendments 46

67 Figure 3.4: Example of a test part. The angle X is the limitation to be identified. The angle of the geometry was changed until the lowest angle that would build was identified. This systematic approach to the research encouraged the order of the experiments to be planned so that the data gained in one experiment could be used most effectively to inform a starting point for the plan of subsequent experiments in series. The advantage of the Deming s cycle was that it encouraged the data to be evaluated to ensure that it was complete before moving onto a following experiment, which resulted in all data being evaluated in stages to ensure that the research objective was met and a coherent set of data could be used to initiate the design rules Other methods considered There are many research methodologies available for the first phase of this research. Phase 1 involved three logical stages that were required to complete the research objectives. To select an appropriate methodology for achieving the required study outcome, a range of alternative methodologies were considered before being discounted. Large Distribution Surveys: The intention of a large survey distribution would be to identify the existing knowledge of SLM among users. An advantage of using large surveys is that there will be a large number of respondents (Blaxter, 2003). However, very 47

68 few SLM experts and users existed at the time of the research. Therefore, distributing a large scale survey was not possible. Case Studies: As part of phase 1, the methodology chosen must enable the dimensional limitations of SLM parts to be identified. To do this, a number of planned and carefully designed parts were necessary to create the data from required geometrical features. Using case studies would require direct observations of SLM parts being made in the field of research or commercial manufacture through field visits to other SLM users across Europe and the UK (Yin, 2003). As existing SLM parts being manufactured were designed according to design rules of traditional manufacturing processes, the results would be unsuitable and unlikely to be focused on the geometries required to create design rules for RM using SLM. Also, many other materials and hardware variations would lead to very little control of process parameters within this research. Using case studies would also require cooperation from many people, which may inncur confidentiality problems. Taguchi Methods: Taguchi methods (Taguchi, 1986) are a series of complex quantitative experimental design techniques predominantly used in quality management. One obvious benefit of Taguchi methods is that they can be used as tools for controlling parameters of an investigation so that un-necessary experiments are eliminated (Bendell & Disney, 1989). This method of controlling parameters is described as product variabilty by Taguchi (1986). It is also used to determine the effects of part tolerance deviation on the quality of products, which was considered for highlighting the limitations of the SLM process. After pilot studies were completed, the methodological requirements became clearer. Due to the relatively small number of controlling parameters, it was decided that Taguchi methods were unnecessary and too complex for the scope of this study, as all process 48

69 control parameters were standardised and the variable parameters were logical geometrical changes according to the results analysis of each test part. Many of the geometries in this research were built so that the measurement and analysis results would inform the subsequent experiments in series. This alone would limit the number of experiments in a less complex way as the experimentation advanced. Many results were expected to be unpredictable and could have more than one reason behind them. The results and conclusions from the pilot studies demonstrated that Taguchi methods were not needed to control parameters as the more complex parameters would be fixed, so that the geometric limitations of SLM would be the main subject for the study Summary of quantitative research undertaken This section describes the components of the quantitative experimental framework phase (Phase 1), and the methodologies used. Some methodologies were considered, but were not selected as they would not produce the detailed dimensional data required to develop the design rules within Phase 2. The first stage involved creating questionnaires that were presented face-to-face at the SLM User Group Conference in Germany, The results gave an insight to the current knowledge for designing SLM components. This event was chosen as an appropriate venue, as all available SLM expertise was grouped together. The SLM machine s hardware and software were calibrated prior to pilot studies. The pilot studies were used to test the experimental methods and to define measurement, part design and parameter protocols for the main experimentation. Another role of the pilot studies was to inform the experiments of the geometries that needed in-depth investigation. The AR approach to the main experimentation was successful in producing optimised parts in the pilot studies. The main experimentation was strategically chosen in a logical order starting with fundamental geometries, so the results would inform and create a starting 49

70 point for the following experiment. The success of the experiments was determined when the data created would allow the design rules (Phase 2) to be comprehensive of all common geometries, without any need for further test parts. 3.5 Phase 2: Qualitative research methods The objectives of Phase 2 were to initiate design rules based on the review of SLM limitations, and raw experimental data created in Phase 1, and, to evaluate the design rules for future development. The following section describes the methodologies that were considered and selected to complete the objectives, based on a review of their merits and disadvantages Selected methods Stage 4: Research synthesis The quantitative experimental data collected in Phase 1 consisted of dimensional values that relate to the limitations and capabilities of the MCP Realizer 250 SLM machine used in this research. The raw experimental data created in Phase1 required interpretation into fully explanatory text, tables and diagrams that could be understood by designers in practice. Each iteration of the quantitative experimentation in Stage 3 was planned so that the results from each geometry experiment could be interpreted successfully. Each iteration was intended to create results that would inform the initiation of a specific design rule, and in some cases they would inform more than one rule. The presentation of the design rules was based on the merits of previous design rules for traditional manufacturing processes. To see the results of the research synthesis, refer to Chapter 6. 50

71 Stage 5: Semi-structured interviews The design rules must inform and give guidance to the designer for when he or she is designing products for RM using SLM. Therefore the designer s input was crucial in evaluating the effectiveness of the rules. Six design professionals were selected to review the design rules. The candidates were chosen according to their varying backgrounds and expertise in the design industry (see Table 3.1) to ensure good cross-disciplinary results. This was so that the feedback would not be biased towards a specific design role. This approach considered that the design rules may be used for various reasons, from actual detailed design, to informing process selection and guiding AM technicians. The participants included an industrial designer, a mechanical engineer, a design lecturer and an AM practitioner. No more than six respondents were chosen because a qualitative expert insight to the design rules was needed from a small number of in-depth interviews, rather than a quantitative statistical analysis of the design rules that would provide less depth. Also, six was the number that covered each of the professional roles that were selected, and, if a duplicate of one role was chosen, bias may have occurred in the results. 51

72 Respondent Number Professional Role Within Design Product design management researcher: Application of design development technologies Implementation of management processes into industry Rapid manufacturing development manager: Qualified production engineer Development engineer Manages a AM department with a large company Senior product design lecturer: Experienced product development manager Application of concept development and design for manufacturing principles. Knowledge in the use of design rules Product development manager/ mechanical engineer: Mechanical engineer and product development engineer at a medical device manufacturer Controls New Product Development Develops marketing strategies Industrial designer: Front end concept design and product styling Experienced and award winning product designer Senior industrial design engineer: Responsible for all elements of the design process Engineers concept products for manufacture Regular customer of AM bureaus Table 3.1: Showing the varying backgrounds in the design industry. A semi-structured interview method was used as it allowed the respondents to provide a fresh commentary on the topic, as suggested by Blaxter et al. (2003). Yin (2003) described focused interviews as being conversational, which leads the respondent into providing fresh commentary as long as the questions are not suggestive in any way. This was an important criterion of using focused interviews, as fresh commentary provided new insights and approaches to the design rules that would not otherwise be apparent in questionnaires. The interviews were completed face-to-face, during site visits to the workplaces of all six respondents so that there were no influences on the responses that would perhaps occur in focus groups. The results were transcribed, and, as the design rules were based on geometry illustrations, the respondents were given the opportunity to sketch 52

73 when words could not portray a clear explanation. The results of the interviews are shown in Chapter 6. Also, the full transcriptions are shown in Appendix Stage 6: Data analysis The results from Stage 5 were analysed. The responses from the semi-structured interviews were analysed and common themes were extracted. The trends in the responses were separated into positive and negative comments. The analysis considered that the results were expected to be subjective and biased towards the different job roles of each designer. Differences between responses were identified for contribution to a thesis conclusion and discussion of future work to progress the design rules development. To see full data analysis refer to Chapter Other methods considered Questionnaires: Questionnaires were considered to review the first draft of the rules and to develop a preliminary set of design rules. This methodology was not used as a conversational feedback method was required to provide a fresh and detailed depth of response. In addition, the length of the initiated design rules were too long for them to be reviewed based on a questionnaire. The respondents needed to work through the design rule document with the interviewer for more comprehensive feedback, and to give the interviewees opportunity to expand and clarify their responses. Focus Groups: Using focus groups as a qualitative, observational feedback method was considered to establish design rule effectiveness and inform future developments. A focus group may have caused the respondent responses to be influenced by other respondents. Focus groups were used successfully by Burton (2005) to test principle design guidelines 53

74 of AM. Participants were asked to apply a suggested design approach to the design of a given product for fabrication via AM. The results proved the effectiveness of the guidelines to promote the use, and to exploit the advantages of AM. The testing required feedback of designers interpretation and understanding of the content and the presentation methods. Focus groups would be more suited to testing future developments of the design rules. Testing the rules by using focus groups and a design exercise as in Burton s work (2005) would only highlight how complex designing for SLM would be and test the design abilities of the participants instead of gaining expert opinion on the rules themselves Summary of qualitative research undertaken This section describes the stages that were within Phase 2 of the research. Phase 2 was completed in three stages of qualitative research. The first stage was to interpret the raw data created in Phase 1, into an illustrative format that can be used in design practice. Once the design rules were created they were analysed in context by design professionals. Semi-structured interviews were used to interview designers, and the interview structure was planned with a series of questions that were used as prompts to ensure that the conversational interviews remained focused. The final stage of Phase 2 was to analyse the semi-structured interview responses. The results from this analysis created part of the research conclusions, and identified areas for future development of the design rules. 3.6 Research novelty The success of more conventional manufacturing processes, such as casting, extrusion and injection moulding, relies on designers understanding the limitations and capabilities of the process (Tres, 1998). Research on the development of design rules for RP and RM 54

75 technologies is limited. More specifically, no design rules have been developed for SLM since the first machine was introduced in the UK (MTT-Group) in However, the need for SLM design rules has been recognised in previous literature as being a requirement for bridging RP into RM (Pullin & Offen, 2008; Wohlers, 2009). As stated in the literature review, all previous SLM research is based on metallurgical engineering principles, novel lattice structures and biomedical implants. Meanwhile, research into the application of SLM for the design and manufacture of less complex common geometries is limited and only undertaken to optimise parameters rather than to develop comprehensive data for the development of design rules. The research undertaken here is the first step in developing process-specific design rules for SLM. Instead of using physics-based research to overcome the inherent geometrical discrepancies of SLM, the issues relating to design for RM using SLM will be addressed through identifying geometric constraints and creating rules to maximise SLM part quality using existing commercially available SLM machines. The following chapter describes the quantitative experimental framework, described as Phase 1 of the methodology. The experimental framework chapter illustrates the preparation and planning of the main experimentation, including machine calibration, pilot studies and the development of experimental protocols. Chapter 4 is concluded by the rationale and explanation of the main experiments. 55

76 Chapter 4: Experimental Framework 4.1 Introduction The experimental framework is the quantitative phase of this research. This chapter describes the pre-experimental planning, such as software and hardware set-up, and includes a description of the pilot studies, how the pilot studies informed experimental protocols, and the main experimentation. Finally, this chapter describes the rationale behind each individual experiment and how the experimental protocols made the experiments a success. The experimental framework is illustrated in the methodology (Chapter 3, Figure 3.1). The parameter and apparatus optimisation, together with other elements of the experimental framework were exposed to peer review. This was done by presenting at conferences to get expert peer review feedback. The expert reviews confirmed that the research proposed was valid (See Thomas & Bibb, (2007) in Appendix 5) 4.2 Apparatus and parameter optimisation Any discrepancies in machine accuracy will reflect on the results of any experimentation and give poor accuracy and validity of the experimental results. To ensure the apparatus used throughout the experiments were used in an optimal state before any test parts were built, all hardware and software were calibrated. The calibration of the SLM also included the benchmarking of process parameters which remained consistent throughout the study. The controllable parameters were set according to previous literature, guidance from SLM suppliers and the results of optimisation experiments that were completed on the SLM 56

77 Realizer 250 machine used during this research (see Figure 4.1). The final parameters were validated by comparing them against the same parts made with the same parameters on an identical SLM Realizer 250 machine based at MTT (Staffordshire, UK). All parameters are presented in the material file, which is the coded data file that the SLM machine reads (Refer to Appendix 1 to see the material file used). All parameters were as follows: Figure 4.1: Image of the SLM Realizer 250 system used in this study. SLM Machine: MCP (now MTT) Realizer 250 SLM (Figure 4.1) Laser System: Ytterbium Fibre Scanning System : Analogue Point Distance = 0.08 mm (the point distance dividedd by the exposure time equals the scan speeds) Scan Speeds: Boundary solid = 0.1m/s (with 1 exposure) Hatch solid = 0.12m/s (with 1 exposure) Skin Hatch = 0.1m/s (with 0 exposures) Fill Contour Solid = 0.1m/s (with 1 exposure) Support = 0.13m/s (with 1 exposure) Hatch Distance (x and y) = 1.3 mm 57

78 Inert gas used = Pure Shield Argon (supplied by BOC) Oxygen Level = < 0.5% Laser Focus: Smallest Laser Spot (2650µm, this value is the position of the laser s lens, and may be different from machine to machine. The 250 based at MTT and the SLM used in this study had different values) Laser Spot Size = 0.08µm Laser Power: Full Power 100 Watts Chamber Pressure: 0.6 bar ± 0.2 bar Material: 316L Stainless Steel supplied by Sandvik Osprey (Neath, UK), maximum particle size- 45µm 4.3 Pilot studies The following pilot studies were completed to establish a clearer understanding of further experimentation requirements, and were completed using the parameters and material described in Chapter 4.2. The results of the pilot studies contributed to the planning of experiments and confirmed the appropriateness of the experimental methods. The test parts chosen were selected based on opportunities from other research and actual medical cases. Other small designs were created with the intention to promote and identify geometrical deformation. The material used for the pilot studies is 316L stainless steel, as used throughout the research. 58

79 4.3.1 Pilot study 1 surgical guide This medical device was a patient specific surgical cutting guide used to guide cutting tools during maxillofacial surgery. This device lowers the risk of human errorr and decreases time in theatre and it can be tested on a Stereolithography (SL) model of the patient prior to the surgery (Bibb & Eggbeer, 2005). The build orientation was chosen so that the surfaces thatt would be in contact with the patient were built upright and therefore had the best surface finish (See Figure 4.2). Figure 4.2: (Left) Image of the support structures in the small slot, (Right) the surgical guide still supported on substrate plate (Right). Small supports were used in the slots shown labelled in Figure 4.2. The supports were difficult to remove as the slot was narrow and the overall geometry could easily distort. All supportss were removed using pliers and an electric hand disk cutter. After the supports were removed the surgical guide didd not fit when tested against the SLA skull model. The poor fit of the device was the result of the force required to remove the supports causing the device to bend at the thinnest wall sections until it eventually fractured (shown in Figure 4.3). 59

80 Figure 4.3: Surgical guide fitted on SLA model of patient s skull. This device was re-designed with a stiffening bar that would not change the functionality, fitting, or get in the way of the surgeon s work. The thin wall sections that fractured were also made thicker so that the chance of part distortion was minimised. Once the revised device was built again, it fitted perfectly against the SLA model when removed from the substrate plate; although, it was difficult to remove supports from the two cutting slots. A small file and electric hand tool were eventually used to remove the supports. The design modifications are shown in Figure 4.4. Figure 4.4: The modified Surgical Guide. 60

81 The surgical guide was sterilised before it was used on a patient. During surgery this device fitted very well against the patient. The slot was designed so that there was enough clearance for the cutter to fit without any obstructions. However, the supports weree difficult to remove and during surgery the slot was found to be 0.5 mm too narrow, interfering with the surgical cutter. This was caused by the down-facing surfaces at the top of the slot building oversize, which occurred because down facing surfaces are poor where added material adheres to the surface in between the supports Pilot study 2 unspecified medical device Once this device had been built using SLM, it was difficult to remove from the substrate plate without damaging the part because of the support structures shown in Figure 4.4. After the supports were removed, it was clear that the geometry had distorted. Figure 4.5: Illustration of the medical devicee with supports. surface of the diameter. It was clear that the down-facing surfaces had a very poor surface finish (Figure 4.6). The rod section that is protruding from the part was meant to be round but was in fact oval. Figure 4.7 shows a small step on the rod section was no longer visible on the down-facing It had become obvious that support structures are very difficult to 61

82 remove from small intricate components. In this case extra care on support removal was taken but the part was still damaged. Figure 4.6: Illustration of the poor surfaces quality on the down-facing surfaces. Figure 4.7: Side view of the poor step resolution. The part was rebuilt at a different orientation to increase the angles of the down-facing surfaces (Figure 4.8). The angle from the rod to the substrate was increased from 45 to 75. The new orientation allowed the part to be removed from the substrate plate more easily, without any part deformation. 62

83 Figure 4.8: Surgical device built for a second time, the parts are shown on the substrate plate still supported. The surface quality on the rod had improved since the first attempt. The step in the rod was well defined and the profile shape of the rod was round (see Figure 4.9). The flat down-facing surfaces still had a poor surface quality. Figure 4.9: Image of the surgical devicee with the step on the rod improved. 63

84 Pilot study 3 mixed geometry test part This part (Figure 4.10) was constructed with multiple simple design features, from thin walls to inclusive angles and holes. This part was designed specifically for testing machine settings, and to identify geometric limitations. The overall envelope of the part was 20mm x 20mm x 20mm. Figure 4.10: Image of the mixed geometry test part. Support structures were used inside of the hole, and at the base of the part to attach the part to the substrate plate. The base supports were difficult to remove, they needed to be crushed in a bench vice until they separated from the part. This caused no damage to the part as the supported surface consisted of a solid, dense volume of material. The hole supports were easily removed with a small pin punch. However, there was evidencee that material had melted down in between the supporting teeth, leaving a rough surface at the top of the holes. This suggested that the supports do not improve the surface quality at distorted areas (see Figure 4.11) Figure 4.11: Picture of the hole before (left) and after (right) the support removal. 64

85 The surface quality of the part walls appeared to be the same, except on one surface as a small sink mark was present alongside the length of hole (See Figure 4.12). The top surface quality was not as smooth as the side walls. Figure 4.12: Image showing the sink marks along the side of the hole. Digital Microscopy was used to see a magnified image of the part features. The images are shown below in Figure 4.13 (a, b, c, and d). The images show more detail of the part features. Figure 4.13: Microscope images of mixed geometry test part. 65

86 I was not po It ossible to use the microsscope to meaasure the feaatures as theyy were p photographe ed in true perrspective. H However, thee detail proviided by the microscope m w was v very high annd gave a cleear indicationn of the defo ormations onn the test parrt. The 45 internal i c corners show wn in Figuree 4.13(a) werre not as shaarp as they w were in the CAD data andd the s surface quality at the topp of the holee was clear in n Figure 4.133(d) Pilot studdy 4 holes P Previous pilo ot studies id dentified thatt down-facinng surfaces and a support structures s caan cause m many probleems with thee geometry oof a part. Thhis test part w was designedd to identify what h happens to round r holes when w they arre built in a non self-suppporting posiition withouut s supports. A 5mm thickk bar with holes of 1mm m to 10mm diameter d (seee Figure 4.1) were built in i both t x and y-aaxis. The onnly support sstructure thaat was used w the was to ancho or the whole part to t substratee plate. the Figgure 4.14: Illustration of completedd test parts. 66

87 The results showed that as hole size increased the material began to curl away from the powder of the bed (Figure 4.15). This deformation is best described as part curl. Up to 7mm diameter, the amount of part curl was very small and did not seem to interfere with the powder recoating system. From 8mm and above the curl increased and fouled the recoating system as new powder layers were deposited. This pilot study was used to inform a scoring system for quantifying curl during the main experimentation (see 4.6.2). Figure 4.15: Illustration of the curl occurring on holes, the gaps in the melt are where the hole centre is. The distortion was difficult to see this by eye, therefore the pilot test pieces were viewed through a shadowgraph at 10 times magnification with scaled printouts overlaid onto the projected shadow. It was clear that the circularity of the holes decreasedd as hole size increased, and the holes were not symmetrical as the geometry had distorted towards the greater volume of material. This can be seen at the largest hole in Figure Figure 4.16 also shows that the tops of the holes are not round as the material has sagged into the hole. Figure 4.16: Shadowgraph images of round hole pilot test pieces. 67

88 The shadowgraph images provided good visual evidence of the deformation that occurs in round holes. The perspective of the 3D component does not show on the shadowgraph images, and allows a 2D profile image of the hole cross section. The test parts in Pilot Study 4 were 5mm thick. The shadowgraph results appeared to match the visual inspections of the hole as all distortions were clear. Pilot Study 1 identified that thin parts can bend and distort when being removed from the supports and substrate plate. The 5mm parts did not distort when they were removed. Wall thicknesses greater than 5mm were not used in this pilot study as they would require additional build time and use more material than necessary Pilot studies summary The pilot studies were essential in identifying geometric distortions for further investigation. Not all distortions could be highlighted as they were unknown at this stage in the research. The experimental methods were successfully used during pilot studies 1 and 2. The experiments were revised and repeated with appropriate design changes, until the parts were built successfully. Part deformations were identified throughout all of the pilot studies. These deformations were specific to the geometric features that were built, and provided the underpinning data to plan the investigations for further experimentation. The vulnerable geometric deformations were identified as follows: Down-facing flat surfaces Down-facing surfaces that are at an angle Round holes built at 90 to the substrate plate (upright orientation on z-axis) 68

89 The deformations that have been identified will have an affect on a wider range of geometries. For example, the inability to build holes will also reflect on radii at the same orientation, as they are essentially a section of a hole. Measurement methods used in the pilot studies were used not only to identify poor geometries, but were also used to assess the measurement method and the results that can be achieved. The microscopy produced high quality detailed images of the deformations that occurred in Pilot Study 3. The microscope images were in perspective and could not be used for measuring. The shadowgraph images created an accurate two-dimensional representation of the hole cross sections as shown in Pilot Study 4. The two-dimensional images were compared against the original geometry using transparent paper so that parts could be measured, and deformation could be identified. The pilot studies gave a good understanding for further experimentation requirements such as the design of the test parts and limiting the size of support structures to ensure parts are not distorted when they are removed. Measurement and build styles were also piloted which highlighted key areas of importance for creating experimental protocols. 4.4 Experimental protocols: process and design The experiment protocols were created from previous knowledge gained in the pilot studies, previous literature and general SLM user experience. The protocols were required for all experimentation to create a consistent approach to the research. Protocols for test part design, SLM process procedures and methods of measurement and analysis are described as follows. 69

90 4.4.1 Experimental conditions The SLM machine was deliberately positioned in typical machine shop to create a normal working environment for the experimentation. Therefore, certain elements including the ambient temperature and humidity of the working environment were not controlled. Procedures were established to control certain elements to add uniformity to the work. The powder was hand filtered and also stored in the machine in a workshop environment without air-conditioning. This meant that it could absorb moisture from humidity and become dirty from hand filtering, resulting in the powder losing its flow ability and consistency. Virgin 316L Stainless Steel powder purchased from Sandvik Osprey (Neath, UK) was used at the start of the experiments. All material was then removed from the machine once every four weeks and baked in an oven at 200 C for two to three hours to ensure that general dirt and any moisture build up from hand filtering in a workshop environment was removed. The SLM build chamber was cleaned using an alcohol solution between each experiment in a further attempt to eliminate contaminants Test part design A 5mm maximum part thickness was used for all experiments that investigated design features. Pilot Study 4 showed that a 5 mm thickness is sufficient for identifying geometric deficiencies by eye and when using a shadowgraph. A thicker part would also use unnecessary amounts of material and take longer to build. Parts greater than 5mm would also impair the measurement of a shadow image because the outline of the shadow will include more material which may hide small deficiencies. The deeper parts become, the more difficult it becomes for the shadowgraph optical system to focus on part details. Pilot study 1 resulted in the test part wall thicknesses needing to be scaled-up to prevent distortion when the part was removed from the substrate plate. Therefore, to reduce the 70

91 risk of bending/damaging test parts during their removal, no parts thinner than 5mm were used. The test part design is illustrated in Figure Figure 4.17: Illustration of the test part design. The design details labelled A, B and C in Figure 4.17 are described as follows: Label A: This surface was only used to align to the part into a squared position when measured using the shadowgraph protocol. Label B: This dimension was used to ensure that the calibration of the SLM was consistent throughout the experimentation. The data was not recorded as the geometry being investigated. Label C: These points were used to align the 2D hardcopy against the test part. Vandenbroucke & Kruth, (2007) created test pieces to identify simple geometric limitations. The test pieces were constructed of many different geometries which resulted in them being much bigger than the test parts designed in the research described in this thesis. In this research a comprehensive range of common geometries were tested. The cyclic experimentation methods used encouraged the use of small individual test pieces, designed to allow part geometry iterations and design changes, and to use as little material as possible to save on cost and to minimise build time. 71

92 4.4.3 Support generation and removal All test parts were supported using one block support structure for anchoring the test pieces to the substrate plate. The supports were designed using Magics software version 11.1 (Materialise) (See Figure 4.18). The use of the support structures has been reported as the the main restriction that RM imposes on part geometries, and consequently support placement is as important as the part design (Pullin & Offen, 2008). No structures were used to support the design features being investigated, this is because if they were used the geometry would have poor surfaces. Therefore, specific features were included in the design of the test parts to support the main parts without interfering with the surfaces being investigated. This enabled the self-supporting dimensions to be identified. The supports were removed from the test parts using a vice, a pliers and a hammer and chisel. They were not removed from the parts if the part was likely to be damaged. Figure 4.18: Illustration of block supports created in Magics Software (Materialise, Belgium). Table 4.1 and Figure 4.19 describe the support teeth dimensions used. The dimensions allow for strong anchorage during the build and enough leverage and room for support removal using hand tools. 72

93 Upper Teeth Lower Teeth Height 2 mm 1.5 mm Top Length 0.15 mm 0.15 mm Base Length 1.5 mm 1.5 mm Base Interval 0.2 mm 0.2 mm Table 4.1: This table shows the dimensions used for the block supports. Figure 4.19: Illustration of the support structure variables Post processing: grit blasting This was completed using a grit blasting machine to remove small partially melted powder particles that were loosely attached to part surfaces. This ensured that no anomalies jeopardised the quality of the measuring methods. Grit blasting is a common and quick finishing process and was also used in other SLM research by Vandenbroucke & Kruth, (2007) before measuring test parts. The grit media used was a medium glass bead ( grit). 73

94 4.4.5 Post processing: machining Post process machining was used on one experiment only, to investigate stock on material. The machining was performed on a CNC Vertical Milling Machine using a standard machinist s vice and parallel bars to ensure that the parts were square to the cutting axis. No significant findings or observations were made when milling the SLM parts, with exception to the experiment results presented in Chapter Experimental protocols: measurement and analysis After completion of the pilot studies, definitive measurement protocols were created. Various inspection and quality control methods were examined before deciding on the most suitable methods. The following protocols explain the measurement and analysis apparatus and techniques used during this study Considered methods unused: Coordinate Measuring Machine: Coordinate Measuring Machines (CMM) are a useful measuring technology that is more suited for measuring complex geometries that have been made using more traditional manufacturing methods such as milling, pressing and injection moulding. Using CMM on small SLM components would be very time consuming and outside the available budget for the overall research. CMM would also be unduly affected by local poor or uneven surface quality. However, this approach has been used successfully on a small number of test parts by Vandenbroucke & Kruth (2007) to measure the accuracy of simple SLM test parts. 74

95 D Scanning: 3D scanning is commonly used as a reverse engineering tool for making virtual replicas of 3D objects. This technology is also used for measuring components in production and creating FEA analyses of stresses and other deformations that develop during manufacturing. To use 3D scanning to measure the test pieces anticipated in this study would be extremely time consuming and expensive. Optical scanning would produce thousands of data points resulting in masses of unnecessary data. Surface quality and reflectivity would also result in a large amount of noise in optically scanned data that would lead to inaccuracies. The high measuring accuracies would not be of benefit to the results as many test parts were non-complex in design and had surface quality that would require a specialist surface profile scanner Selected measurement protocols: Microscopy: Microscopy was used in the pilot studies to create detailed images of part features. The images were taken using a digital camera add-on to an optical microscope. Whilst it is possible to take measurements of part features, the level of accuracy at the magnification used was more than required, this would have been very time consuming and would have required the use of an expensive travelling microscope. Perspective issues made it impossible to create reference points for accurate measurements. However, the images taken with the microscope proved very useful and gave a clear visual representation of part surface quality. Microscopy was used in this study at an early stage to depict and explain part deformation that would occur throughout all tests. 75

96 Shadowgraph imagery: The shadowgraph (Figure 4.20) creates a 2D representation of a 3D part at magnification. At first the parts were measured using the shadowgraph s optical measurement system. This took a long time to measure a small number of parts as many surfaces of the pilot parts were rough, making it difficult to select a reference point to measure to. Figure 4.20: Image of the shadowgraph machine used. The final method was to do all measuring in CAD software and just use the shadowgraph to capture the shadow image. Scaled (10x Magnification) images of the test parts were printed onto 90 gsm semi-transparent paper, including the part number. When the parts were placed squarely onto the shadowgraph using a small vice, the printed image was lined up against a known reference point and a photograph was taken. A long focal length was used to ensure that no barrelling effect developed in the photo. The photo was then imported onto the appropriate plane of the original CAD geometry in the software. The image was then scaled against the printed image in the photo, and the outer lines of the CAD image were offset to the extremities of the deformation. Dimensions were then 76

97 added to the CAD image and the results were recorded. The measurement process is shown in Figure Figure 4.21: Illustration of the Shadowgraph/CAD measuring process. The cross section of the parts in shadow images shows projections of alll uneven part discrepancies along a part surface. Measuring geometries with un-even surfaces with instruments that make contact with the part will not give the same measurements each time due to the surface quality that was expected of some of the least accurate test parts. Therefore, the shadowgraph images were appropriate for identifying the extents of all material deviations on a surface. The shadowgraph used was a Sigmascope HF 500. Calibration of the Shadowgraphh was completed within Arvin Meritor under their ISO 9001 quality management accreditation Surface roughness: The surface roughness of the SLM test parts were measured using a surface roughness gauge (Taylor Hobson Sutronic 3p) as shown in Figure

98 Figure 4.22: Taylor Hobson Sutronic 3p surface roughness gauge. The surface roughness gauge is an appropriate and simple method of measuring the mean roughness (Ra) of SLM parts. A stylus is used to traverse a sample distance across the part surface, and will display a digital readout of the Ra value. 4.6 Quantifying part deformation during the build process De-lamination Some part deformation happens during the SLM build that prevents the part building process from completing. De-lamination (Figure 4.23) takes place when subsequent layers do not melt together. As the metal cools, the material shrinks which causes it to lift away from the powder surface if it has not fully bonded with the layer below. This is caused by poor parameter selection. More specifically, de-lamination is the result of insufficient laser power to melt the layers together, or a scanning speed that is too fast for the laser power to be absorbed by the material. However, all parameters were benchmarked to minimise the risk of this happening. This work was published in Thomas and Bibb (2007). 78

99 Figure 4.23: Photo of de-lamination where the layers haven t merged Curl A common cause of part distortion is curl (see Figure 4.23), which occurs when the harsh temperature gradient melts and solidifies the powder in areas where the material is not self- supporting and has no underlying dense material. In this study, curl was expected because few supportss were used. The amount of curl that occurs during the build was quantified using the scoring system shown in Table 4.2 and is also illustrated in Figure Curl Level 0 No curl 1 Curl Amount Up to 1 layer thickness (<0.075mm) Explanationn No Curl Slight visible curl, no threatt to build finishing safely 2 3 More than 1 layer thickness (>0.075mm) Curl increased at least 2layers every layer Bad curl, build safety is at risk Fail, serious curl which is guaranteed to impair the powder re-coaterr Table 4.2: This table shows the scores given for curl, and the reason why that score was given. 79

100 Figure 4.24: Illustration of curl values 0 to 3, the dotted lines replesent the layers and triangles represent support structures. 4.7 Summary Once the research methodologies were established all hardware and software used in the experimentation were optimised and calibrated, and a series of pilot studies were completed. During these pilot studies a series of measurement methods and part designs were tested. The outcomes of the calibration and pilot studies contributed to the development of definitive experimental protocols that ensured all experimental parameters, test part designs and analysis techniques were controlled within set boundaries. The experimentation was completed to establish raw experimental data to inform the design rule development that follows. 80

101 4.8 Experimentation overview The experimental overview describes all quantitative experiments that were planned and completed to inform the design rules. Each experiment was required to produce dimensional data of specific design features, including part accuracy and dimensional limitations. The geometrical features that were investigated were planned based on noncomplex design features that are fundamental in the construction of more complex shapes. It has been established that if support structures are used, it is because the geometry is designed outside of the limitations of SLM and the surface is unsuitable for building. Therefore, each experiment was planned to identify the dimensions of several geometries that allowed them to build without the use of supports. Initial questionnaires were carried out to identify existing SLM users knowledge on design limitations of SLM. The results of the questionnaires were considered when developing the experiments required to develop fundamental results for the design rules. The process was iterative following Deming s plan, do, study, act approach. The results of each experiment were used to inform the design of the subsequent experiment Supports and orientation The selection of a support structure is closely linked to selecting a part orientation in preparation for building using SLM. The number of supports can be dependent on the orientation of a specific geometry. If a geometry is orientated so that 90% of the surfaces are self supporting, then only 10% of the geometry will need to be supported. It is impossible to eliminate the use of supports entirely as they are needed to anchor the parts to a substrate during a build. However, if a designer considers orientation and support at an early stage the amount of support and the accuracy of SLM parts will greatly improve. 81

102 All test pieces were designed to identify self-supporting geometries, which were used to show what needs support and what does not. The geometry of all future designs was unknown; therefore, the optimum support and orientation cannot be suggested and each SLM part would need to be treated case-by-case. Through the experience gained in the main experiments it became possible to suggest considerations for support and orientation within the design rules Experiment 1: Simple orientation The aim of this experiment is to identify the lowest self-supporting orientation of a simple flat surface that can be built using SLM. Simple cuboids were designed and supported on one surface only and were orientated at 90 to 25 at both x and y-axis directions. The test parts where designed so that the surface roughness could be measured on all surfaces in the following experiment Experiment 2: Surface roughness The aim of this study is to identify if there is a correlation between part build orientation and the surface roughness. The first experiment was completed to identify the minimum self supporting orientation of a surface. This experiment was designed using the test parts from Experiment 1 to identify what happens to the surface roughness average (Ra) as self supporting orientation angles changed Experiment 3: Ledges This experiment was designed to identify the effect of building ledges that are parallel to the substrate plate. It is not possible to build a surface parallel and facing the substrate 82

103 plate unless supports are used. This experiment investigated if there was a size restriction to when supports need to be introduced. Simple geometries were designed with overhanging ledges of 0.5mm to 5mm in increments of 0.5mm. No part larger than 5mm was built as it had been demonstrated to be unstable when considering the need for support structures during all pilot studies Experiment 4: Angular overhangs (chamfers) A chamfer is a common design geometry typically used for bevelling corners and eliminating unnecessary sharp corners. Experiment 1 explored building simple flat cuboids at various orientations. The results from these experiments were used to design several test parts with chamfers of 45, 50, 55, 60 and 65. The accuracy of the chamfered surface was measured in this experiment Experiment 5: Convex radii Convex radii are fundamental in the construction of holes and channels, they are commonly used to blend sharp external corners and in the construction of more complex curved surfaces. A typical convex radii curve has a varied tangent that causes the base of the radii curve to become overhanging. This experiment investigated the effect on part accuracy when making both tangential and partially tangential radii, and identified the critical dimensions of radii that need to be considered to make self-supporting radii. The results of the orientation experiments were used to plan and analyse this experiment. 83

104 4.8.7 Experiment 6: Concave radii This experiment was designed to identify the limitations of building concave radii in the same way as Experiment 5. The tangent of a concave radius varies and is smallest at the top of the radius curve. Results from the orientation experiments were used to plan this experiment also. The results of this experiment were used to develop design rules for selfsupporting radii Experiment 7: Holes and channels The objective of this experiment was to highlight the geometric constraints of holes. It was identified that round holes collapse at the top of the hole when built with or without supports, and holes could not be built larger than 7mm diameter without supports. The intentions were to develop new designs and to modify simple round holes so that they can be produced reliably and accurately without requiring support structures Experiment 8: Tapping and reaming self-supporting holes Previous experimentation has proven the difficulty involved with making simple round holes in an unsupported position. This experiment is aimed at identifying the best possible way of adding an internal thread and reaming self-supported holes (Experiment 7). Threads are common to the design of components in assemblies and reamed holes are commonly used to act as an accurate location hole for mating components. 84

105 Experiment 9: Minimum slot and wall thickness The aim of this experiment is to determine the smallest slot and the thinnest wall thickness that can be built using the SLM machine used in this research. Test parts were designed with an array of slot sizes to identify the smallest slot possible and a wall was built with a single pass of the laser. This wall was the thinnest possible wall that could be built with the SLM used in the research Experiment 10: Shrinkage Shrinkage and similar deformations are common with more conventional processes like injection moulding. A small amount of shrinkage appears on the outer walls of SLM parts, this was first identified in the pilot studies. The aim of this experiment was to identify when shrinkage occurs, and how to eliminate it through avoiding vulnerable geometries. Test parts were designed to promote the shrinkage so that its development could be monitored. Previous test parts from all experiments were also analysed to identify links between accuracy and shrinkage Experiment 11: Stock on material / sacrificial geometry Stock on material is the name given to adding extra material to specified surfaces of a design so that post processing can be completed without making the part undersized. Otherwise known as geometric compensation and material allowance, stocking on material is common place when parts are designed to be manufactured by sand and investment casting followed by conventional material subtraction processes such as milling and turning. This experiment was designed to quantify how much material compensation is required to remove the outer skin of an SLM part, revealing only dense unblemished metal. 85

106 Chapter 5: Experimental Results 5.1 Overview This chapter describes the results from the experimentation to gather the data for developing design rules. The geometries that were investigated to inform the design rule development were selected by deconstructing compound geometries to expose the fundamental geometries that are used in the construction of all design features, ranging from simple cuboids to complex lattice structures. The specific design rules were not identified at this stage, however, these are the fundamental geometries that will inform the development of the design rules. The cyclic Plan-Do-Study-Act (PDSA) (Deming, 1994) methodology was used to generate the results shown in this chapter. This PDSA approach to the experiments encouraged each result to be reviewed after each experiment was completed to identify if the dimensional data required from each test part was achieved. The order of the experiments was strategically planned so that they became more complex as they progressed, and so that the data gathered could inform subsequent experiments. Each experiment was repeated with revised geometrical changes until the required dimensional data was obtained and could be used to inform the test part design for subsequent experiments. The results are divided into part A and part B. Part A describes the results and significant findings of fundamental geometry experiments. These fundamental experiments were devised as a breakdown of the building blocks that construct more 86

107 complex geometries. Part B describes the results of compound experiments that were constructed from the results in part A. The results shown in this chapter were exposed to peer review. This was done by presenting at conferences to get expert peer review feedback. The expert review feedback stated that the results were valid (See Thomas & Bibb, (2008a) Thomas & Bibb (2008b), and Dotcheva et al (2009) in Appendix 5) 5.2 Results part A: Fundamental geometries: Experiment 1: Simple orientation Several experiments were completed to identify the minimum build orientation of a flat surface. Cuboids were built in orientations of 90, 85, 80, 75, 70, 65, 60, 55, 50 and 45 in alignment with the x-axis and y-axis as shown in Figure 5.1. Figure 5.1: Photograph of the orientations; left is along x-axis, and right is along y-axis. 87

108 In both the X and Y direction 45 proved to be the lowest orientation with a curl of 1 (see curl scoring criteria in Table 4.2). The curl was not significant enough to obstruct the build efficiency. The experiment was repeated with lower orientations and no parts below 45 built successfully along the y-axis. Angles of 40, 41, 42, 43, 44 and 45 were built in attempt to refine the result to within an accuracy of 1. Only 40, 43 and 44 built successfully across the x-axis parallel to the powder recoating system. This result was not repeatable in further attempts, but it showed that the x-axis was more stable than the y- axis, with the direction of the wiper system being the only difference. All further test parts were orientated so that curl distortion would occur on the weaker y-axis to prevent limitations being restricted to just the x-axis. Kruth & Vandenbroucke (2007) have previously stated that the staircase effect decreases proportionally with the cosine of the sloping angle. As the angle decreases, the staircase effect increases, meaning that the size of each step increases. This was evident in this experiment. As the minimum orientation is 45, the maximum overhanging step (layer) size was the same size as the layer thickness, at 75µm (See Figure 5.2). It was apparent that this is the maximum layer overhang, because greater than 75µm caused residual curl distortion during the SLM build. Experiment 4 (Section 5.2.4) shows the residual curl occurring as a result of excessive over-hanging geometries. 88

109 Figure 5.2: Illustration of the 75µm maximum layer overhang at Experiment 2: Surface roughness The average of the roughness profile (Ra) of all previously completed parts was measured (45, 50, 55, 60, 65, 70, 75, 80, 85, and 90 ). Three surfaces from each part were measured in increments of 5 on all parts, see Figure 5.3. Surface A was measured from 90 to 45 up-facing. Surface B allowed measurement of surfaces from 90 to 45 down-facing. The top surface (C) was perpendicular to the side walls (A and B), this allowed all up-facing orientations from 0 (parallel to substrate) to 45 to be measured. Figure 5.3: Illustration of the three surfaces measured in this experiment. 89

110 All results were gathered using a Taylor Hobson Sutronic 3p surface gauge (see surface roughness measurement protocol). The aim of this experiment is to identify correlation of orientation on surface quality. Initially the test parts were to be measured only once, however, the first resultss consisted of a large amount of variability and an irregular surface pattern caused by a small amount of weld spatter that had adhered to part of the wall during the build. To identify if the variability was a result of the measurement method used or whether it is process inherent, a pen line was drawn across the clearest path along the part surface avoiding the surface voids. Only parts from the y-axis were measured, as the results were similar across x and y-axis in the first test. The final results are shown in Figure 5.4. Figure 5.4: Graph illustrating the results from the second roughnesss measuring test. The results showed theree was variability in the data. However, the results do match the visual appearance of the test parts. When trend lines were added to the graphs, the slope of the lines suggested that the staircase effect was present, which correlated with work by Kruth &Vandenbroucke, (2007), this is illustrated in Figure

111 Figure 5.5: Illustrating of the staircase effect changing at different orientations, the higher the orientation the more step occur, resulting in a lower surface roughness image from (Kruth & Vandenbroucke, 2007) Experiment 3: Minimum slot and wall thickness The aim of this experiment was to determine the smallest slot, and the thinnest wall thickness that can be built by SLM. To measure the minimum wall section, a boundary of a 10mm tall cube was built, this ensured that only one pass of the laser would melt each layer to ensure minimum melting which results in a minimum thickness (see Figure 5.6). The boundary was built directly onto a substrate plate, with no support, and was measured using a vernier calliper. The wall thickness measured 0.4mm with a tolerance of ±0.02mm because of the accuracy of the vernier. 91

112 Figure 5.6: Image of the hollow cube used for measuring the minimum wall thickness, and the two parts for identifying a minimum slot size. To identify the minimum slot thickness, two test parts were created and measured using the shadowgraph (see Figure 5.6 and Figure 5.7). One part consisted of 5mm diameter cylinders with gaps from 1 to 0.05mm thick, and the other was constructed of small cubes with slots of the same sizes. The two different tests show what the smallest slot is, and to see if the depth of the slot makes a difference. The minimum slot size was found to be 0.3mm, for both types. The 0.3mm slot appears to be partially blocked because the surface quality of the parts protrudes and overlaps in the two-dimensional shadow image. A feeler gauge was used to ensure the slot was fully open. 92

113 Figure 5.7: Image showing the minimum walll thickness measured on the shadowgraph. The top image is the cubed slots and the bottom image is the round slots. The results are clear in this image. The slot was built vertically at 90. However, a change in the orientation will affect the size of this slot, and the lower the orientation is the larger the minimum slot will become. At orientations lower than 45 the slot will not self support and will therefore need a support structure. However the support may not be able to be removed if the slot is less than 4 or 5mm Experiment 4: Parallel ledge This experiment was undertaken to identify how large a ledge could be built with no supports, and remain parallel to the substrate plate. Ten testt parts were designed and built, with overhangs from 0.5mm 5mm, in increments of 0.5mm.. The final part images and curl distortion values are shown in Figure 5.8. Figure 5.8: The completed SLM parts on the substrate plate. 93

114 This experiment was completed twice with the same results. Only overhangs 0.5mm, 1mm and 1.5mm completed the build. All other parts had distortion values (curl) that were too large to finish, causing interference with the wiper system recoating procedure. It was identified by using a shadowgraph, that the only acceptable geometry was the 0.5mmm overhang (see Figure 5.9). Overhangs larger than 0.5mm did not represent the original square geometry. Experiment 1 (Chapter 5.2.1) shows that the minimumm overhanging step size of a single layer is 75µm. The 0.5mm result in this experiment is for one single step, and shows examples that an overhang size larger than 75µm will promote curl distortion. Figure 5.9: The shadow images of the parts that had completed the test build. During the build it was observed that the part would only curl when the raster scan was scanning perpendicular to the curl direction. The start of the curl is illustrated in Figure When the laser scanned the x-axis no curl occurred. When the laser scanned along the y-axis, the unsupported material curled up from the surface powder. This correlates with work by Mercelis & Kruth (2006) who have identified the influence of the scanning direction. 94

115 Figure 5.10: The curl occurring during the SLM builds. The left image shows no curl effect as right shows the curl starting point Experiment 5: Angular overhangs (chamfers) A chamfer is a typical feature found on many part designs. It is commonly used to remove sharp edges of a part, especially on metal parts, making the parts safe and easier to handle. Chamfer angles 45 to 65 were chosen for investigation in increments of 5. Shadowgraph imagery was then used to measure all parts and microscopy was used to visually inspect the parts. Using an outline from the correct original CAD geometry, the visual alignment of all parts was measured and any deviation was recorded. The results are shown in Table 5.1 and Figure Angle Curl Angle Tolerance Negative Material Positive Material mm 0mm mm 0mm mm 0mm mm 0mm mm 0mm Table 5.1: Showing the results of the chamfer measurements. 95

116 Figure 5. 11: Images of the shadowgraph measurements and microscope images. visual alignment at 55 is 1.5, and at 60 there is no deviation. The The test geometries were not repeatable to within a tolerance smaller than 1.5 or 0. 2mm. Microscope images show that all layers curl away from the original CAD geometry on the chamfer, which is why all inaccuracies were negative (undersize). This suggested that further experimentation will result in part deformations being undersize rather than oversize Experiment 6: Fillet radii (convex radii) A fillet radius is a common geometry used for both aesthetic and functional reasons. It is considered that a radius is part of the construction of any design using curved surfaces, and multiples of a simple radius are used in more complex shapes and forms. This experiment investigated the constraints and possibilities of building self-supporting convex radii. The lowest point of a traditional convex radius is vulnerablee to curl, as it is down-facing and exceeds the overhanging size limit identified in previous experimentation. The first trial during Experiment 6 proved this, as only 1mm and 2mmm radii out of five largerr parts 96

117 built successfully. The accuracies of the radii were poor and the radius did not resemble the original CAD data (see Figure 5.12). Figure 5.12: Shadowgraph images compared against the actual CAD data. Further experimentationn was undertaken which incorporated a tangent change at the lowest point of the radius so that the part would become self-supporting (see 30 examples in Figure 5.13) ). The tangent at the top of the radius remained tangential so that the transition of the radius to a different feature would have a blended edge. It was confirmed that the tangent was the dimension that determined the success of the build, and more curl distortion occurred as the radius size increased. Further experiments were then undertaken investigatingg different sized overhangs of 5, 10 and 15mm as shown in Figure The tangent angles tested were 22 to

118 Figure 5.13: Illustrates the three different size parts built. The overhang sizes are 5mm, 10mm and 15mm, the tangents changed for each test piece. Once the final test was completed, all parts were measured using a shadowgraph, and plotted in a table along with the amount of curl distortion and the dimensional values. This table shows the tangent angle where each size part built successfully, and most accurate, allowing for a general tolerance to be achieved. Parts with a curl of 3 did not finish and were not measured (See Table 5.2). Table 5.2: The accuracy and the curl level of the convex test parts, the blank boxes were not measured as they did not build. 98

119 The results showed that the larger the overhang, the larger the starting tangent of the radius needs to be. This was the same for the earlier radii experiments. It appeared that the part accuracy and the amount of curl distortion were related. Parts with curl level of 2 were significantly less accurate than parts with no curl at all. This is illustrated in Figure Figure 5.14: Shadow images of the three different curl values. This image illustrates that the less curl the more accurate the part is. Although the accuracy was not consistent for all parts with a curl of 2, 1 or 0, from Table 5.2 and the shadowgraph images (Figure 5.14), it was clear that there is an accuracy difference between each curl value. The accuracy is undersize more frequently than oversize because the distortions curled the geometry of the SLM part away from the correct geometry. The surface quality at the lowest point of the radii is the poorest, and improves as the surface tangent angle increases. If there is a small amount of curl, the curl value is 1, and the overall tolerance of the convex radius will be ±0.4mm, the tolerances vary from ±0.1mm to ±0.4mm for each part built within a curl of 1 for the 5, 10 and 15mm overhangs. 99

120 To achieve a tighter tolerance there must be no residual curl occurring throughout the build, tolerances of ±0.25mm can be achieved, although this tolerance varies from ±0.0mm to ±0.25mm. At the first layer of a convex radius, the tangent angle is at its lowest and increases to become more self supporting as the layers progress on the z-axis. As found in Experiment 1, the layer overhang is 75µm maximum which is at 45. The tangents of the lowest layers of the radius in this experiment were smaller than 45. This can only happen because the consecutive layers become self supporting, and the curl distortion does not get a chance to accumulate. However, as the radius size increases, the tangent of the first layer of the radius need to increase because there are more layers between the tangent and the self supporting 45 tangent to accumulate an excessive curl distortion, see Figure Increasing the tangent of the lowest layer on the radius also increases the overall height of the radii, which may become a design limitation. Figure 5.15: Illustration showing the number of layer increase as the convex radius size increases, the number of layers below 45 also increases. 100

121 5.2.7 Experiment 7: Fillet radii (concave radii) A concave radius is the opposite curve to a convex radius and a simple design feature that can be incorporated into designs to eliminate sharp corners and can be found in the constructionn of curved surfaces. The results from the previous experiment on convex radii were considered in analysis of the test parts. This experiment investigated the limitations of manufacturing concave fillet radii using SLM. Previous findings identified that a tangential convex radius could not be built successfully. The first experiment identified that tangential concave radii could not build successfully as the top of the radius is not self-supporting, because the tangent decreases to 0 as the part grows in the z-axis, resulting in excessive layer individual overhangs. Radius sizes 1, 2 and 3mm didd build but the geometry did not resemble a radii, and the 3mm had an unsafe curl value of 2 (see Figure 5.16). Figure 5.16: Images showing the shadowgraph images of the tangential radius, the images are measured against the original CAD data. To ensure that concave radii could be built with self supporting dimensions, the top tangent of the curve was increased (see 30 tangent example in Figure 5.17). Three experiments were undertaken to identify the minimum tangents required to build a concave radius 101

122 without the use of support structures. It was identified that as the tangent angle increases, the curl distortion decreases and the accuracy of the curve was improved. When measured, all parts were oversize rather than undersize as with the convex radii. The following experiments were completed at 5mm, 10mm and 15mm overhangs. Figure 5.17: Illustration showing the overhanging ledge sizes and the position where the tangent is changed. Once the final test was completed, all parts were measured using a shadowgraph, and plotted in a table along with the amount of curl distortion and the dimensional values. This table shows the tangent angle where each size part built successfully, and most accurate, allowing for a general tolerance to be achieved. Parts with a curl of 3 did not finish and were not measured (see Table 5.3). 102

123 Table 5.3: Showing the accuracy related and curl level of the concave test parts, the greyed out boxes are not shown as the geometry was unsuccessful and failed. The level of curl distortion appeared to become a reliable visual representation that occured during the build, of how accurate the parts actually are. It appeared that the part accuracy and the amount of curl distortion were related. Parts with curl level of 2 were significantly less accurate than parts with no curl at all. This is illustrated in Figure Figure 5.18: Shadow images of the three different curl values. This image illustrates that the less curl the more accurate the part is. 103

124 Although the accuracy was not consistent for all parts with a curl of 2, 1 or 0, from Table 5.3 and the shadowgraph images (Figure 5.18), it was clear that there is an accuracy difference between each curl value. When no curl occurred the radius is complete and ends with a sharp corner. The shadowgraph measurements show that the larger the overhang, the larger that the tangent needs to be. The surface quality on the concave radii degraded as the tangent of the radii increased when the parts built up in the z-axis. The part tolerance when building with the recommended tangents will be -0.2mm and +0.1mm, therefore a safe tolerance will be +/- 0.2mm. Parts that are built with smaller tangents than recommended, will either fail, or will have a curl value of 2. Many of the results were lower than the minimum surface orientation of 45 (see Experiment 1). As the radius experiments progressed it became clear that this phenomenon relates to the number of layers that were lower than 45 that exceeded the minimum layer overhang size. If only a few layers are lower than 45 then the curl will not have enough layers to accumulate and cause a failure. As the parts get larger the number of layers increases resulting in more layers vulnerable to accumulate high levels of curl. See Figure

125 Figure 5.19: Illustration showing the number of layer increase as the concave radius size increase, the number of layers below 45 also increases. One possible design limitation occurs as a result of increasing the tangent to create a selfsupporting radius. The height from the bottom to the top of the radius increases, which also results in the radius value on the curve increasing. See Table 5.4 for examples. Tangent (X) 5mm Overhang Height of Radius 10mm Overhang 15mm Overhang Table 5.4: Showing the increase in height when the tangent increases. 105

126 Both concave and convex radii will be affected with the radius height increasing in exactly the same manner. This may become a design constraint if a tangential radius cannot be made at a down-facing orientation. 5.3 Results Part B: Compound shapes: Experiment 8: Holes and channels Pilot studies completed earlier in this research identified that round holes cannot be built upright without failure or distortion. Holes smaller or equal to 7mm build safely but will have an average of 0.5mm of sag at the top of the hole, which is uncontrollable even with supports, and prevents the overall hole from being perfectly circular. Larger holes above 7mm show evidence of excessive curl distortion which impairs the build safety and results in larger inaccuracies. Holes are traditionally round due to drilling and reaming, however, since holes that are manufactured using SLM are built layer-by layer, there is an opportunity for holes to be made any shape. When the hole experiments where planned, it seemed necessary to set the dimensions of the holes according to cross sectional areas of 5mm 2, 10mm 2, 20mm 2 and 30mm 2. This was so that the holes could also be used as channels that could manage calculated amounts of fluid flow. Designing the holes based on area became a challenge because as one dimension was set to meet the area values, it had an impact on the other feature dimensions. This then required modifying the dimensions that were previously applied to the holes, which then resulted in other dimensions on the same feature needing to be changed to match the modified dimensions. 106

127 The achieved results had dimensions that were driven by setting the required area on each hole. The dimensions all fell between the 0 to 5mm, 10mm and 15mm overhangs that were tested. The initial thought was that these results would fill in the gaps between 5mm and 10mm and 10 to 15mm. On completion of the experiment it was found that the final results could not be used to design new holes, this was because designing the holes to set areas removed an element of control from the experiments. The following shows the results of all hole experiments, and significant findings that need to be considered in further developments of the holes. The shadow images and original geometries are shown in Figure Figure 5.20: Shadow images and original hole designs. The geometries of the triangle with radius, ellipse, and circle with radius peak were designed so that no sharp corners would cause stress concentration zones when applied to actual useable parts. When self-supporting dimensions were applied to the 107

128 geometries, the radius became a sharp point. The circle with peak was created exactly the same as expected from the original design (see Figure 5.20). All parts built with a curl distortion value of 1, and the accuracies were all within 0.2mm which was also predicted from previous experimentation. When two features of a part are built on top of one another, the lower feature must have no distortion where it connects to the top feature. If distortion is present then it creates a poor foundation for the top part, which will result in distortion within the first few layers of the top feature. If the top feature is built at the lower self-supporting limit the distortion may accumulate further from the lower feature and cause a build failure. Alternatively, if the top part is not built at the lower self-supporting limit the distortion will be corrected within the first few layers. This was identified when building the circle with radius, and the circle with a 45 peak at the lower self-supporting angles. The equilateral and isosceles triangle holes proved to be less complex to build than all other holes. This was because only self-supporting flat surfaces were required; the results are as shown in Figure 5.21 and Table 5.5. Figure 5.21: Showing shadow image of the equilateral and isosceles triangles. 108

129 Accuracy Hole Type Hole size Angle Curl - + Equilateral 5mm mm Equilateral 10mm mm Equilateral 20mm mm Equilateral 30mm mm Isosceles 5mm mm Isosceles 10mm mm Isosceles 20mm mm Isosceles 30mm mm Table 5.5: Results and geometrical details of the equilateral and isosceles triangle test parts. The size of each triangular hole does not affect the accuracy as the walls are self supporting. The result was predictable when considering the little difference between a curl of 0 and 1 in previous experiments Critical analysis: Investigating the limitations of self-supporting hole designs using hole area had proven to be unsuccessful, therefore, it was relevant to repeat the investigation with a different approach. Rather than creating many hole designs, the geometries were designed to be as close to a round hole as possible to maximize the surface area of the actual hole round, and to retain the maximum functionality where a traditional round hole is used. To do this further investigations of the hole labelled circle with radius peak was required as it had the most circle surface area, and the peak height could be minimised if tangent limitations were applied. 109

130 Previously, the tangents at the base of the radius peak were unconstrained on every test piece, which affected the height of the peak, and therefore, its ability to build as a selfsupporting geometry. This is illustrated in Figure Figure 5.22: Shows the possible changes in peak height when under constrained. The first experiment was to identify the maximum height that the circle could be made without supports. 1mm to 15mm radius test parts were created and as they began to curl the layer number was recorded and the part was stopped from building further. Only the top half of each hole was built at this stage as the bottom half is self supporting. After the first test build, the experiments proved to be challenging as the point of the build where the curl that was safe, to the point of the curve where the curl became unsafe, was spread over many layers. As only judgement was used when deciding to stop the parts, it was possible that some parts were stopped too early and some were stopped too late which caused the variability shown in Figure The results are shown in Figure 5.23 and Figure

131 Figure 5.23: Graph of the results taken from attempt 1. Figure 5.24: Actual SLM test pieces on substrate plate. To ensure that the maximum height of the hole was correct, the middle point of the layer that the first sign of curl occurs and where the curl clearly became unsafe (curl 2) was investigated. This ensured that the maximum height that alll holes could be built to was consistent. The results and centre points are shown in Figure

132 Figure 5.25: Graph showing the centre point of the max and min tangents. The parts from the centre point on the graphs were rebuilt successfully with no curl. The 1mm and 2mm radius built as with no curl as shown in Figure The holes were measured using the shadowgraph, the accuracy of both these holes shows that there was a sag of 0.3 to 0.4mm, this suggested that a peak was required to enhance the accuracy. Figure 5.26: 1mm and 2mm radius holes built using SLM Developing a self-supporting peak: The maximum height, and tangents of the circle part of the hole were identified, the following explains how the peak was designed. Previous experimentation on radii identified a minimum tangent at the highest point of various concave radii (see Figure 112

133 5.27). The results of the three radii were driven by the size of the overhang, which is effectively the size of the ledge that required a radius. A mirror of one radii was used to create the hole peak. Figure 5.27: The previously concluded curves and lines where the curves were sectioned. As the tangents and layer number that the circle built up to is known, it was possible to calculate the size of the gap at the top of the circles, and therefore the size of the overhang for the peak radius was known. The curves were split up into 1mm sections that represented different overhang sizes as shown in Figure The line and curve intersections are the dimensions required for the radius peak. To keep the hole size to a minimum, the overhangs were selected starting from the highest point (lowest tangent) of the curve. If the overhangs were selected from the lowest point of the curve the tangent angle would start as 90 tangential, which means that the curve will fully blend into the vertical surface and the height of the peak will be the largest possible. 113

134 It was decided that the 10mm overhang curve was not required. This was because it had no benefit over the larger 15mm overhang curve, as the top tangent was 40 on both. The height of the curve will be lower if the largest curve is used, and up until a certain height the smallest curve will give a lower height. Based on the intersected curves shown in Figure 5.27, the following theoretical design rules were established that would create selfsupporting peaks that have the lowest height so that the hole remains as close to the original round geometry as possible. The theories were tested and all holes are shown following this theory. If a curve has an overhanging geometry of less than 3.65mm, the 28 curve gives the lowest height compared to the 40 curve. Therefore the smaller curve should be used in this below overhangs of 3.65mm. Figure 5.28 shows the 28 curve at 3.65mm, and Figure 5.29 illustrates the height difference on the 28 and 40 curves with a smaller overhang. Figure 5.28: Shows that 3.65mm overhang will give a 3.48mm height for both small and larger curves. 114

135 Figure 5.29: An illustration of the 28 curve having a lower height than at 40 when the same overhang values are used under 3.65mm. Between an overhanging distance of 3.65mm and 5.18mm, the larger 40 curve will have a lower height. Therefore this curve should be used if the overhang of the radius falls in between 3.65mm and 5.18mm. Also the 28 curve is not large enough. See Figure Figure 5.30: Showing the height is the same as the offset at 5.18mm on the 40 curve. 115

136 After 5.18mm, the peak becomes higher than if a 45 angle was used. Therefore a 45 angle peak should be used instead of a radius for all sizes larger than 5.18mm. See Figure Figure 5.31: Illustration of the 45 position at a maximum 5.18mm Experiment - Applying the self-supporting peak theory to actual holes: Using the rules shown above, self-supporting peak designs were created for holes of radius 1mm to 15mm. The peak designs and the hole circles were constructed together and then built using SLM. Table 5.6 describes the peak geometry details that was used for each hole size, and Figure 5.32 illustratess an examplee of each peak design used. The overhang of peak describes one half of the peak, which is the mirrored to construct the peak. 116

137 Radius Overhang of Peak (Half the Gap) Peak Used mm 28 Curve mm 28 Curve mm 28 Curve mm 28 Curve mm 28 Curve 6 3.9mm 40 Curve mm 40 Curve mm 45 Degree Angle mm 45 Degree Angle mm 45 Degree Angle mm 45 Degree Angle mm 45 Degree Angle mm 45 Degree Angle mm 45 Degree Angle mm 45 Degree Angle Table 5.6: Overhang sizes for peaks and the associated geometry feature. Figure 5.32: Three example holes (R4, R7 and R10) with their associated peak geometries. The 1mm and 2mm radius holes built without requiring a self supporting peak. As described earlier, sag occurs on round holes even as small as 1mm and 2mm radius. Therefore, a peak will be added in this experiment to eliminate the sag causing 117

138 inaccuracies. All holes were built complete and successfully (see Figure 5.33), at no point during the build did any curl on the test pieces exceed a value of 1. Figure 5.33: Picture of completed holes from R1 to R15. All parts were measured using the shadowgraphh method. It appeared that the largerr the hole size, the more accurate the hole became. Radius 3mm and 4mm built with a tolerance of ± 0.2mm.. All holes larger than this up to a radius of 15mm measured to a tolerance ± 0.1mm, these results corresponded to previous experimental results. See Figure 5.34 for an example of two shadowgraph images. Figure 5.34: Shadow image of 4mm m (± 0.2) and 13mm (± 0. 1) radii. 118

139 Hole sizes of radius 1mmm and 2mm previously built withoutt a peak but it gave a poor geometry because of sag collapsing in the hole. In this experiment, a peak was added to try to eliminate this sag causing the inaccuracies. The results showed a very small improvement (See Figure 5.35). The peak was barely recognisable; there was still excess material inside the circle of the hole, although the sag was less apparent. These holes measured to a tolerance of ± 0.3mm. Figure 5.35: Shadow images of 1mm and 2mm radii Experiment 9: Tapping and reaming a self-supporting hole This experiment consisted of tapping and reaming the alternative hole designs developed in Experiment 6. Threads and holes with high accuracy are a common occurrence in the design of a functional metal component. If the design rules are to suggest a new type of hole design to be used, it is important that a guideline for tapping and reaming the holes is also illustrated. One large (M12) and one small (M6) hole was selected. Both holes corresponded to the dimensions used in the holes experiment, however the circle section of the holes to be tapped were designed at the tapping drill sizes (M6 = 5mm diameter and M12 = 10. 2mm diameter) and the holes to be reamed were made 0.3mm smaller to allow the reamer to 119

140 cut. All parts were held in a machine vice and were tapped by hand, and reamed by a pillar drill. Figure 5.36: Image showing the M6 (left) and M12 (right) holes after being tapped. Both test parts weree tapped and a M6 and M12 bolt screwed through. The thread was not even, as both tapping drills pulled towards the self supporting peak. At the side of the holes the thread was at the correct depth. At the base of the hole the thread depth was too small. The uneven tapping effect is worse on the M12 bolt because the peak is larger than at M6. This was clear when the hole was being tapped as the M12 intermittently cut, and the M6 did not. See Figure 5.36 for image of the tapped holes. Figure 5.37: Image showing the M6 (left) and M12 (right) holes after reaming. Both holes reamed, however the larger hole chattered as the reamer was cutting. The reamer intermittently cut because one cutting edge of the reamer was located in the 120

141 peak as the others were cutting. This resulted in the hole not being round (See Figure 5.37). Both holes failed to ream centrally as the reamer pulled towards the self-supporting peak. This was evident as witness marks of the SLM generated surface quality were visible at the lower point of the hole Experiment 10: Shrinkage This experiment was completed to investigate the shrinkage that becamee apparent on many test parts throughout the experiments. This was first observed on pilot studies investigatingg hole design issues (Figure 5.38), and was consistent throughout the experiments. This experiment was completed in two parts. Part 1; simple test pieces were designed to promote the shrinkage, and Part 2; a retrospective analysis of all previous test parts was completed, to identify when and why the shrinkage occurs. The following describes the results of this experiment. Figure 5.38: Image showing the shrinkage / distortion on the pilot test pieces. 121

142 Part 1: Test parts analysis The test parts consisted of a 10mm round hole that was positioned in offsets of 1mm from the centre of 5mm thick cuboids. The offsets created wall thicknesses of 1-10mm. The round hole was built without support to promote the distortion. It was clear that distortion had occurred on the outside walls when they were measured using a shadowgraph. However, the shadow images do not show the distortion as clearly as when inspected by eye. The images are shown in Figure Figure 5.39: Shadowgraph images showing the distortion on wall thicknesses of 1mm to 10mm. 122

143 All test parts had a planned curl of 2 at the top of the holes. The shrinkage was evident on the side of the parts with a wall thickness of 1mm to 7mm and no shrinkage was evident on parts of 8mm wall thickness and above. The smaller the wall thickness, the worse that the shrinkage appeared to be. 1mm had the worst shrinkage whereas 7mm was only a small amount. The shrinkage occurred at the same height as the deformations at the top of the holes Part 2: Analysis of previous test parts The occurrence of curl on previous test parts was quantified using a scoring system in previous experimentation. The results were revisited to identify correlation between the curl distortion and shrinkage. Four parts from each category of curl, 0, 1 and 2 were selected and analysed using the shadow images of those previously taken during experiment 8. The only information that was required at this stage was whether or not the distortion had occurred, and how much distortion there was. The results showed that the shrinkage increased the larger the curl value was during the build, as highlighted in Figure It was clear that when the distortion does occur, it is consistently around the same layer as the very top of the holes. 123

144 Figure 5.40: Shadow images of parts with curl 0, 1 and 2, the distortion is evident and increases with the curl values. The study aimed to highlight the distortion that has occurred and create enough information for it to be avoided; therefore, no definitive conclusion has been achieved. However, it is possible to suggest that geometries using self supporting features contribute to minimizing the shrinkage, and the severity of the shrinkage caused by other features decreased when the walll thickness is increased Experiment 11: Stock on material / sacrificial geometry Many applications of metallic parts require highh accuracy, and a polished surface finish. This experiment identified how much material would need to be removed from the top, side, and bottom surfaces of an SLM part for it to be machined to dense metal. Three 124

145 cuboids (30x30x15mm, 20x20x15mm and 10x10x15mm) were built. Before the parts were removed from the substrate plate, the top surface was milled using a vertical milling machine. This allowed the parts to be held square in a vice to machine the side and bottom surfaces. The parts weree then removed and the side and bottom surfaces were machined whilst recording how much material was being removed, as shown in Figure The quantities of material removed are shown in Table 5.7. Figure 5.41: Photo of the threee cuboids after they weree machined on all surfaces. Block Size 30 x 30 x 15mm 20 x 20 x 15mmm Height before material removal Material Removed until flat (Bottom) Material Removed until dense (Bottom) 15.7mmm +/ mm 1.6mm / mm 1mm Material Removed until flat (Top) Material Removed until dense (Top) Material Removed until dense (Side) 0.3mm 0.7mm 0.12mm 0.2mm 0.6mm 0.12mm Final Height / / / x10x15mm / / mm 0.8mm 0.2mm 0.6mm 0.12mm / / Table 5.7: Results from stock-on-material experiment, showing how much material was removed from each part. All blocks built 0.7 to 0.8mm oversize on the down-facing surface, which was caused by the laser penetrating more than one layer of powder on the first layer. The down- 125

146 facing surface was avoided throughout this research study as it is the main cause of inaccuracy. The larger the surface area of the down-facing surface, the more material needed to be removed. This is because it is caused by shrinkage of the thin first few layers of the build. As parts grow taller the material becomes thicker which prevents the layers shrinking. However, this creates residual stresses. Each top surface was not flat before they were machined; the top surface had a higher boundary than the hatched material. The side surfaces completed flat, and measured to within a tolerance of 0.05mm before they were machined. 5.4 Summary The first part (A) of the results chapter was an investigation of fundamental geometries that were key to the construction of compound shapes. The results of the experimentation identified the geometrical constraints, and accuracies of the fundamental geometries, which proved to improve when the geometrical constraints were known and applied. The results of the orientation and surface roughness experiments identified that the staircase effect was present with SLM parts. There appeared to be variability in the roughness average values of the parts measured, which resulted in surface roughness trends to be identified rather than accurate values. The minimum surface orientation was identified, which resulted in the minimum overhang size of each layer being the same as the layer thickness at 0.075mm, which was considered in the design of all overhanging test 126

147 pieces. The thinnest possible wall thickness achievable was identified in one simple experiment, along with the minimum slots size that can be produced. The final fundamental geometry tested was concave and convex radii. The results showed that designing radii for manufacturing by SLM is more complex than conventional processes as tangential radii could not be built without support structures. Tangent changes were applied to the radii and the constraints were identified in order to design self supporting radii. The compound experiments in Part B were created considering all results from Part A. Shrinkage was witnessed throughout all experimentation; therefore a retrospective experiment measuring all test parts was completed. The results identified that the wall thickness and neighbouring distortion elsewhere on the same layer caused the shrinkage marks, and were preventative by using self supporting geometries. The results from designing various hole profiles using previous radii results, when combined with achieving area values proved to be unsuccessful because the dimensions could not be fully controlled since the holes needed to be specific area. The results could not be used to create design rules, therefore the experiment was repeated with a different approach based on keeping a round hole as close to round as possible by using a selfsupporting peak. Holes were successfully designed and had repeatable accuracies and provided data that can be used as a design reference for creating self-supporting holes of 1mm to 15mm. 127

148 The ability to post-process SLM parts was considered and the results showed exactly how much material would need to be added to a design if the surface were to be machined. Tapping and reaming self-supporting holes is inaccurate as the cutters intermittently cut. The following chapter describes how the design rules were initiated from the experimental data created in this chapter, and how the design rules were evaluated to identify future developments and challenges. 128

149 Chapter 6: Design rules initiation / development 6.1 Introduction Previous quantitative experimentation was used to identify all limitations and constraints of SLM components. Individual design features were investigated, and results have been created to be used in the development of design rules for SLM. This chapter describes how the design rules were initially created from the raw experimental data gathered in the main experimentation, and describes the semi-structured interviews used for evaluating and developing the rules. Finally, this chapter shows the analysis of the designer responses to the rules, and how future requirements for developing successive design rules for SLM were identified. 6.2 Research synthesis: Creating design rules from the experimental data Developing the design rules required interpretation of the raw data gathered from the quantitative experiments, to ensure that the rules could be understood by a product designer without the need for further interpretation. Presenting the experimental results to a designer would not provide clear rules and illustrative guidelines to designing for SLM, because they would be unformatted, too complex and difficult to interpret. The methodology used to identify the geometrical limitations of SLM created a logical step between raw data and the design rule initiation by providing results that needed very little interpretation to become illustrative. The Plan, Do, Study, Act (Deming, 1994) approach 129

150 to research, encouraged each experimental result to be studied after each test part was built, which ensured that the experimental objective of creating geometrical data for use in design rules was met. Each experiment was based on fundamental geometries that are part of the construction of compound geometries. Since there are a vast number of compound shapes that could be presented in the design rules, only fundamental geometries were conveyed that could inform the construction of the more complex forms. Therefore, associating the experimental data to each design rule became a simple task of re-writing the results and refining the presentation. On completion of the experiment, if the results could not be used to inform the design of the subsequent experiment, the experimental parameters were reconsidered, and the experiment cycle was repeated and reviewed until the results could be used to inform further experiments. The methodology flow diagram is shown in Figure 3.1 in the methodology chapter. The systematic methodology approach also encouraged the results of one experiment to inform subsequent experiments in series. This ensured that the results were evaluated as the experimentation progressed. If an experiment was complete, the results would successfully inform the design of subsequent experiment test parts, and therefore indicate that the results are able to inform the design rules. If this was not possible and test parts could not be designed, it meant that the data was inconclusive and was not comprehensive enough to inform the design rules. Further experimental iterations were required until the results could be used to inform the subsequent experiment. When the design rules were first created, existing design rules on traditional processes were reviewed (see literature review Chapter 2). These processes included injection moulding (Rees, 1996), aluminium extrusion (Sapa Ltd, 1997), metal injection moulding (MIM)(Groover, 2002), rotational moulding (Beall, 1998) and also the DFMA principles 130

151 by Boothroyd and Dewhurst (2002). From reviewing these rules, a number of significant points for both the introduction and main content of the rules were identified. These points were considered in the development of the design rules within this research, and are described as follows: (for the completed design rules refer to Chapter 7) Design rule introduction: The design rules introduction must describe how the technology works and any underlying issues that will affect the way that the rules are perceived. The introduction must also provide enough background information to set the scene for the design rules to facilitate the understanding of each design rule. The introduction must also include enough information for the designer or engineer to ensure that the process they have chosen is the correct process. It is important to describe any technical terminology at the start of the design rules. The rules are aimed at all levels of designer with various levels of experience, therefore, some designers may not be familiar with some terminology especially if they are not familiar with AM. Presenting the design rules: Within the content of the design rules, it was necessary to illustrate the occurrence of any process inherent distortions and explain why they occur. The design rule associated with the distortion should illustrate how good design practice will eliminate or avoid the distortion. This can be done by showing an illustration of a good design together with a poor design. An example of this is shown in Figure 2.15 and 2.16 within the literature review. 131

152 Both two-dimensional and three-dimensional images can be used to illustrate each design rule. The two-dimensional images should be used to show detailed dimensional data, where a three-dimensional image shows a greater resemblance of the design feature in context. When several variations of a single geometry needs to be described, it is not necessary to show an illustration of them all, because a single, fully labelled two-dimensional image can be used alongside a table that refers to each dimensional variation of the features shown in the two-dimensional image. Large blocks of text should not be used amongst the design rules as they can be discouraging to read. Instead, precise and smaller bullet points, or short concise paragraphs should be used to ensure that the rules are clear and simple to read. Best practice (Sapa Ltd (1997); Groover (2002)) indicated that design rules needed to be first put into context. Accordingly, an overview of AM is needed in order to inform the reader how different AM is compared to traditional subtractive and moulding processes. An overview of SLM is also needed to introduce SLM to designers who have no previous experience of the technology. The introduction to SLM included an image of the laser during processing; an explanation to how powder layers are recoated and how layers merge to produce parts; and, background information about geometry deformations that occur as a result of exceeding process limitations. A trade off was illustrated which explained how optimum orientation, support and part quality effects a designer s choice when selecting orientation. The research that informed 132

153 the supports and orientation and geometric deformation background information was taken from the experience gained of both topics throughout all experimentation. However, one experiment was completed to establish a greater understanding of geometry deformation, so that key design considerations could be presented. Figure 6.1 shows a chart of all design rules and the experiments that informed them. Some rules were informed by just one experiment, for example, the minimum size between features and minimum wall thickness. However, other design rules such as material allowance/ geometric compensation were informed by two experiments. Two informing experiments were unintentional, although the results indirectly informed the planning of the second experiment that was the main contributor to the rules. Detailed reports of the experiments are shown in the results section of this thesis. 133

154 Figure 6.1: Illustration showing what experiments were used to develop each design rule. 134

155 As explained previously, many experimental results did not need an extensive interpretation to make them illustrative when presented in the design rules. However, the following design rules required reconfiguration so that they would convey the data effectively Part orientation and surface roughness The first design rule that was reconfigured illustrates the link between surface roughness and the orientation of a surface. This design rule can be found at within the design rules in Chapter 7. The rule describes the optimum surface orientation and angles where support structures must be used. This design rule incorporated two different experiments that influenced one another, into one rule. The results were combined as they would be confusing and overcomplicated if they were presented separately. The first experiment identified all possible build orientations of a flat surface, and the second experiment measured the surface roughness at the different orientations. Different ideas were considered to establish an illustration combining both sets of results. Figure 6.2 shows the concluded illustration used in the design rules. The different shades of grey represent the surface qualities at different orientations. 135

156 Figure 6.2: Illustration showing the relationship between orientation and surface quality Chamfers and Radii: Concave fillet radii and convex fillet radii This design rule is part of a number of rules thatt illustrate overhanging design features. This rule provides a dimensional breakdown of how to create self-supporting convex radii. This is numbered as in the Design Rules shown in Chapter 7. The concave and convex radii experiments weree concluded with three curves that represent the limitations of building radii, as shown in Figure 6.3. The concave curves were analysed in greater depth during the holes experiment as part of the construction of self supporting hole designs. The curves were drawn in 2D using CAD software and intersected at 1mm intervals to establish dimensional values for different overhang sizes of radii (see Figure 6.3). The analysis identified that any section of the curves can be used as a radius. From this, rules were established and were successfully tested on self-supporting holes. 136

157 Figure 6.3: An illustration of the three concave curves with intersection lines in increments of 1mm. This was created during the self-supporting holes experiment. Although concave and convex radii were investigated in the main experimentation, only concave radii were analysed in greater depth, because they were required in the construction of self-supporting holes. Therefore further analysis of the convex radii results was required for creating both concave and convex design rules (see Figure 6.4). The results of this additional analysis contributed to the design rules by illustrating a breakdown of the critical dimensions of both convex and concave radii. 137

158 Figure 6.4: An illustration of the three convex curves with intersection lines in increments of 1mm. This was established after as part of the design rule synthesis Curved surfaces This design rule describes how to construct a surface from more than one radius. The rule illustrates how to identify and apply the appropriate dimensional limitations to create curved surfaces. This design rule is numbered as in the Design Rules shown in Chapter 7. The geometrical features chosen for investigation throughout the quantitative experimentation were chosen because they are the most common features and are the building blocks to more complex geometries. Experiments were focused on investigating the limitations of building flat surfaces; no experiments directly investigated curved surfaces. 138

159 Figure 6.5: Illustration of curved surfaces broken into individual less complex radii. A radius is a design feature that is commonly used to replace sharp corners. The results of the radii experiments informed the need for curved surface design rules. The radii experiments were used for illustrating how to breakdown a complex curve into individual less complex radii and to emphasize the location of critical surface tangents (see Figure 6.5). The dimensions given in the radii design rules were then referenced within the curved surfaces design rule. This allowed the construction of self-supporting curved surfaces. After all design rules were created, they were analysed by the end user to discover their effectiveness and requirements for future work. The following section describes the analysis process. 139

160 6.3 Analysis Six design professionals agreed to review a hard copy of the design rules, and participate in semi-structured interviews to evaluate the design rules. The six participants were selected because they had different roles within the design industry, including both academic and industry based initiatives. None of the respondents were existing users of SLM. However, the need for a designer was a key point, since designers are key users of design rules. An AM expert was also required, because the rules are based on a AM technology. A design academic is someone who has understanding of transferring academic design research into industry, and therefore was also chosen for the research. A design for manufacturing expert was used to be able to asses the suitability of the design rules to inform manufacturing practice. The criteria for selecting the potential participants to evaluate the design rules were based on four professional roles, which are described as follows: Designers: Industry based designers who are active in product development and design for manufacture. The majority of the respondents needed to be design practitioners as they would be the most likely to use and benefit from the design rules. AM Expert: An AM expert must know the AM field and any inherent technical issues. They should also be fully aware of the possible advantages and disadvantages of AM technologies. Design Academic: The academic must be involved in product development technologies but not necessarily AM. The academic will give a broadened perspective of how the technology is portrayed in the design rules. 140

161 Design for Manufacturing (DFM) Expert: A DFM expert will be familiar with the use of design rules and will therefore be qualified to provide constructive feedback from an industry point of view. Through personal communication and recommendations made, all selected participants met the criteria to provide critical feedback. Selecting professionals across different fields of design offered a range of perspectives based upon a range of specialties. The following describes the professional roles of the six respondents selected (see Table 6.1). Respondent 1 Respondent 2 Respondent 3 Respondent 4 Respondent 5 Respondent 6 Professional Profile Product Design graduate and has been a researcher for over five years. Area of interest and expertise include the application of product design and development technologies and management processes in small manufacturing companies. Particular knowledge is of design and development processes. A production engineer with a degree in mechanical engineering. Previously a senior development engineer creating software and hardware for NC machining, and has more recently set up and now manages a Rapid Manufacturing department within a large company. Respondent 3 has several years of industrial experience as a product development manager prior to becoming a senior design lecturer. Specialist areas include concept development and design for manufacture, which include process and material selection, and requires an in-depth understanding of design constraints and considerations of manufacturing processes. Respondent 4 is a mechanical engineer and project manager for a large medical device manufacturer. Primary roles include the development of new products, through liaising with surgeons and developing design ideas into manufactured products, and also developing marketing strategies and regulatory requirements for new product development (NPD). Industrial designer with expertise in front-end design and product styling. Respondent 5 is part of a design team that has won 10 international design awards. Respondent 5 has worked on over 70 commercial projects with clients that include inventors, Small to Medium Enterprises (SMEs) and multinational companies. Respondent 6 is a senior industrial designer and is responsible for every element of the product s design to manufacture, ensuring they comply with international safety requirements, whilst also managing the Rapid Prototyping side of the business. Table 6.1: Describing the professional profiles from each respondent. 141

162 The design rules were ed to the recipients four weeks before the semi-structured interviews were arranged. A series of questions were designed to analyse the design rules and to evaluate the clarity of the rules, ensuring that the content of each rule was clear, illustrative and comprehensive to what each respondent would typically expect within design rules. Questions were planned to assess how the introduction to the SLM technology set the scene to the design rules, and if this introduction was detailed enough for the respondents who were unfamiliar with SLM. Although questions were set to identify whether the rules were clear and illustrative, it was known before any interviews that many rules were more complex than existing rules for traditional subtractive processes. For example, the holes required peaks that had several dimensional constraints. Questions were designed to identify whether the complexity of the rules would discourage the use of the rules and the SLM process, or whether they would be encouraged to explore SLM and exploit the process within their industry. The questions were piloted with Respondent 1. The response was used to refine and finalise the questions, and was then used as part of the constructive analysis. The questions are shown as follows: Question 1: Overall, were the design rules a good insight to the SLM process? Question 2: Were the rules clear and illustrative? Question 3: Is the introduction to the technology useful? Question 4: Were the rules complete? Are there any design features missing? Question 5: Would you feel confident to use the design rules in design practice? Question 6: Since reading the rules, did your feelings change about using the SLM process in the near future? 142

163 Question 7: What is your opinion of the SLM process since reading the design rules? Question 8: The benefits of SLM, such as hollow and freeform geometries are not described within the rules; would this prevent you from considering unique design features that cannot be made using other processes? Question 9: Many geometrical features such as radii and holes are now more complex than other processes such as milling and casting. Are the descriptions of each rule clear or are they too complex for what are traditionally simple geometries? Question 10: What do you think about the length of the rules? Question 11: Are any rules unnecessary? Each interview took no longer than one hour and was completed at the convenience of the respondents at their place of work during January All responses were recorded in the format of notes made by the author during the interviews, and the respondents were given access to a pencil and paper so that they were able to express themselves using sketching when it was difficult to use words only. All respondents were asked to write in the design rules documents, critical feedback on individual rules. The design rule documents were then collected together with comments at the time of the interview. 6.4 Analysis of results The following are the results and analysis of the semi-structured interviews. Each respondent responded with two degrees of response. The first degree was a direct response to interview questions which were not directed at specific rules. The second degree was detailed comments that were specific to individual design rules. These included comments on illustrations, requests for illustrations, requests for further explanations and example 143

164 components. Each response was summarised and the responses were categorised as positive or negative together with the respondent number (see Table 6.2). All respondents are labelled as R1 to R6 (Respondent 1 to Respondent 6). Further details of each response are shown in Appendix

165 Interview Response Analysis Positive Negative All six respondents stated that the design rules were a good insight to the SLM process. However, suggestions for improvements were given including: Examples of the rules being applied to products. 4 out of 6 respondents found the design rules to be clear and illustrative, 2 stated that the rules were unclear. Detailed improvements were suggested in the design rule document 5 out of 6 respondents found the introduction to the technology useful. 1 of 6 found the introduction was not descriptive enough. All Respondents described improvements: More process information A breakdown of terminology Add Information about common orientation to the substrate plate A comprehensive description of all possible part deformation together with images. Images of commercial examples Comparison of SLM to conventional processes. Show a process schematic image Emphasise Implications of the limitations 5 out of 6 respondents thought the rules were comprehensive to all common geometries. 1 out of 6 stated that information on assembling SLM parts is missing. 5 out of 6 respondents stated that they feel confident to use the design rules in practice. 1 respondent stated that showing applications of the design rules would boost confidence to apply the rules in design practice. 5 of 6 respondents stated that the design rules were more complex than conventional processes. However, the responses show that they are not discouraged from using the design rules as a result of complexity. However 1 Respondent stated that the rules were too complex and would be discouraged to use SLM to manufacture. 5 of 6 respondents thought that the design rules were not too long, including a statement saying that the rules were short and concise. However, 1 of these 5 respondents found the rules had too many words. 1 Respondent stated that the rules were too long, although they were comprehensive and had no unnecessary rules. 11 All respondents stated no design rules were unnecessary. R1, R2, R3, R4, R5, R6 R1, R2, R4, R5, R6 R1, R2, R3, R4, R5 R1, R2, R4, R5, R6 R1, R2, R4, R5, R6 R1, R2, R3, R4,R5 R1, R2, R3, R5, R6 R1,R2, R3, R4, R5,R6 Table 6.2: This table summarises the responses of the semi-structure interviews. Each response is categorised as positive or negative together with the Respondent number. R3, R6 R6 R3 R3 R6 R4 145

166 The responses of questions 6, 7 and 8 could not be categorised into positive or negative responses, the questions are as follows: Question 6: Since reading the rules, did your feelings change about using the SLM process in the near future? 5 out of 6 respondents stated that they would consider using SLM as a manufacturing technique since reading the design rules. The responses stated that the benefits of SLM are clearer since reading the rules. Respondent 1 said that SLM would be used if it was short listed as the best manufacturing process for a specific design. Respondent 6 stated he could not understand why it would be used for any application other than organic, freeform geometries. Question 7: What is your outlook of the SLM process since reading the design rules? E.g. limiting, useful etc 5 out of 6 respondents have the same opinion that the SLM process is too constrained. Respondent 6 stated that too much needs to be considered when designing an SLM component, and cannot imagine any engineering application. Respondent 1 stated that the design rules were a good insight to SLM, however, his/her opinion of manufacturing functional parts using SLM, was that lots of post processing would be required. Respondent 5 did not think that the process was too constrained; he/she stated that it requires a new learning curve as the process is relatively new. He/she would apply the rules by designing a part as normal and use the rules to make 146

167 amendments on the final design. Respondent 5 also sees the design rules as a lesson in SLM design. Question 8: The benefits of SLM, such as hollow and freeform geometries are not described within the rules; would this prevent you from considering unique design features that cannot be made using other processes? Two respondents (R2 and R4) would still consider geometries that are unique to SLM / Additive Manufacturing (AM) capabilities. Respondent 2 stated that many examples of these geometries are used as a unique selling point by machine vendors. Respondent 4 stated that the advantageous design possibilities of layer manufacturing would still be considered if the process limitations are known. Respondent 1, 3, 5 and 6 are not familiar with these geometries and would need to see examples of them in the design rules Analysis of negative responses: There was a wide range of positive and negative responses as highlighted in Table 6.2. The positive responses help to validate the design rules but they do not inform their future development. All negative responses were reviewed in greater detail. The negative responses are the points that would need consideration in future design rule development. 147

168 Two respondents thought that the design rules were not clear or illustrative: Question 2 Respondents 3 and 6 stated that the design rules needed to be clearer and more illustrative. Respondents 3 and 6 both specialise in developing designs from concept to a manufactured product. A major part of the respondent s jobs includes preparing designs whilst considering accuracy, tolerances, assembly, and design feature limitations of manufacturing processes. The design rules for SLM must be clear so that all the requirements of making manufactured parts can be applied to SLM designs without significant interpretation prohibiting the design time. To improve the clarity of each design rule, the respondents suggested giving examples of commercial products designed and made with SLM. This was a common issue raised by all respondents. To select SLM as a manufacturing process, the rules would need to be clear and fully illustrative to all design professionals. The designers that felt the rules were unclear work closest to the manufactured product. The general opinion about the design rules was that they were only clear when read in their entirety. The respondents identified this because terminology and explanations of distortion occurrence were not all grouped together, but staggered throughout the document. This suggests that all terminology explanations should be addressed at an early stage in the design rules, together with illustrations where possible, as well as the descriptions staggered throughout the body of the rules, so that the rules can be flicked through and understood equally individually as well as when read cover to cover. 148

169 One respondent did not find the introduction useful: Question 3 Respondent 6 did not find the introduction to SLM useful because no commercial examples were shown. Respondent 6 stated that he/she is familiar with designing for CNC milling when metal parts are needed. He/she can predict the outcome when parts are returned complete after sending them for manufacture, but would not know what to expect with SLM parts. Therefore, respondent 6 stated that he would prefer to use CNC milling over SLM. Examples of parts shown in future design rule developments will allow designers to foresee SLM part outcomes, and will help decide on whether or not the processes is suitable for their manufacturing needs. If the designer still chooses alternative processes to SLM, it should be because SLM was not an appropriate process for the particular product designed, and the chosen process was short listed as better suited One respondent did not think that the design rules were comprehensive to all common geometries: Question 4 Respondent 3 stated that the design rules were not comprehensive to all common geometries as there were no rules on part assembly. Respondent 3 teaches design for manufacture and assembly principles (DFMA) that were first introduced by Boothroyd and Dewhurst during the 1970 s (Boothroyd, Dewhurst, & Knight, 2002), and has a great understanding of design guidelines and their importance. The aim of the design rules for SLM is to instruct designers how to avoid process inherent geometrical deformation. This was done by suggesting alternative designs, by only addressing designs for standard features like radii, angles and chamfers. These geometric features are the building blocks to the construction of more complex geometries, and with 149

170 careful consideration of the accuracies of each design feature, the assembly of two or more SLM parts could be planned One respondent stated that they would not feel confident using the design rules: Question 5 Respondent 3 stated that he/she would not feel confident using the design rules. The reason given was that no examples were present in the design rules to associate the design rules with. Respondent 3 also stated that if examples were present within the design rules document he/she would feel confident to use them. It was a common response from all respondents, which suggests that further design rule developments should include examples of all parts for individual rules One respondent said that the complexity of the rules would discourage him from using them: Question 9 Respondent 6 is an industrial designer who works towards strict time constraints within commercial design projects. The complexity of the rules would discourage him/her from using them and therefore using SLM. This was because the rules would take too much time to follow, and would be a new learning curve when applying the rules in NPD. The rules would encourage him to choose a manufacturing process that he/she is familiar with rather than SLM when tight timelines are involved. 150

171 One interviewee stated that the design rules were too long: Question 10 Respondent 4 stated that the length of the design rules were too long. However, this comment was contradicted because the response also stated that the length of the rules was necessary because they were comprehensive to all common geometries, and no rules were unnecessary Responsibility of using the design rules within the design process. When a product is designed it is engineered so that it can be produced via the chosen manufacturing process. To do this design rules are required, for example draft angles are a requirement of injection moulding. Out of all the interviews, two industrial designers who actively apply design for manufacturing rules stated that they would leave the SLM design rules for the SLM technician to apply once they have finalised their design. Respondent 4 stated that he/she would expect the SLM part bureaus to be the experts and therefore, expect the bureaus to apply the design rules. Respondent 3 stated that he/she would design parts to how he/she wanted it to be, and create control drawings so that the SLM technician within the AM bureau would edit the part to suit the manufacturing process. This would only be possible if the bureau has the capabilities and offers a design service. This result suggests that designers can be discouraged when applying new design rules that they are unfamiliar with. All other respondents understood that the design rules were aimed at them to apply during the detailed design stage of the design process. 151

172 6.5 Summary This chapter described how the synthesis between raw experimental data and the design rules were established. The methodology ensured that the data gathered in the experimentation phase is used directly in the design rules without major changes. Several design rules did need extensive interpretation to ensure that they were clear and illustrative to the reader. The orientation and surface roughness experiments were inter-reliant on each other, and were combined into one design rule rather than two separate rules. The convex radii results needed further analysis as only the concave radii required analysis in great depth during the creation of self-supporting holes. This ensured that design rules were created for convex radii, rather than just the concave radii. Design rules on how curved surfaces can be designed were as also extracted from the results of the radii experiments to complete the design rules. The design rules were assessed by six design professionals from different design areas and disciplines. Semi-structured interviews were completed and the results were analysed into positive and negative responses. Many smaller detailed points were made by all designers directly in the design rule document. All negative responses were analysed in greater depth to identify areas and common themes that require consideration in future developments of design rules. Major points included more examples of each design rule and commercial designs made by SLM should be shown. This was said to increase the confidence of the designer and to improve the design rules, making them easier to follow and refer to when designing within commercial timelines. After reading the rules, all participants established a new understanding of what is required of them to design for SLM. All agreed that the process is too restricted; however, they 152

173 understood the process and would be able to consider SLM in the selection of manufacturing processes and would not be discouraged to use it if SLM was identified as the most appropriate process for a specific design. It was found that the benefits of using SLM over contending metal manufacturing processes were apparent to the respondents, but it was concluded that principle design examples showing possible design features that are unique to layer manufacturing would clearly describe the process benefits. The aim of this research study was to initiate a first draft of design rules and inform future developments of this work. Therefore, it was not necessary to action any of the specific feedback highlighted within this chapter; a second evaluation of design rules is outside the scope of this research study. The first draft of the design rules is presented as follows in Chapter

174 Chapter 7: Design rules 7.1 Introduction Additive Manufactur ring Additive Manufacturing (AM) is a recent manufacturing initiative whichh involves very little tooling. Otherwisee know as additive freeform fabrication, AM is the opposite to subtractive manufacturin ng processess which involve the removal of material. AM is a method of manufacturing and prototyping components directly from 3D Computer-Aided Design (CAD) data. The CAD data is prepared for manufacture in dedicated software where the orientation of the desired build position is chosen. Support structures are added to the CAD data, primarily to hold the part into position as a scaffold, and secondly to anchor down any process inherent distortions as the material changes state. Finally, the CAD data is sliced into the layers ready for AM. See Figure 7.1. Figure 7.1: Illustration of a solid block sliced into layers. 154

175 Manufacturing layer by layer leaves visible traces of the layers. This is known as the staircase effect. The layers become less apparent as the layer thickness becomes smaller, and each step size increases or decreases as angles change. As the angle of a surface changes the appearance of the surface also changes. Designing for layer manufacturing allows more geometrical freedom than with traditional processes. Geometries such as undercuts, internal and complex surfaces can be manufactured in AM, these are difficult to achieve with many other processes. Despite this, many process limitations apply and need to be considered in the same way as more traditional manufacturing techniques Selective Layer Melting (SLM) SLM has been introduced to industry, enabling a Rapid Manufacturing (RM) approach to the production of end use, solid, functional metal products. Selective Laser Melting (SLM) is a layer additive rapid fabrication process which builds components in metallic powders using a high powered laser in an Argon controlled environment. The laser draws pre-determined slices of a part onto a powder bed at 75µm layer thicknesses, where the powder particles fully melt and merge together. Subsequent layers of powder are then swept across the top of the melted layer and the laser melts another layer, this cycle is continued until all layers of the part are complete. See Figure

176 Figure 7.2: Photograph of SLM during processing. Designing for SLM enables a freedom in design that conventional metal processes, such as milling cannot offer. Because of such high temperatures and rapid cooling rates of the powder melting and re-solidifying, many process and geometric limitations apply Geometry deformation Process limitations of SLM that occur during processing, often result in some part deformation. Many of these deformations can be avoided by following process specific guidelines and rules. Down-facing surfaces such as ledges and tops of holes are common features that promote a sag part deformation. This sag is caused by the intense power of the laser penetrating through the un-melted powder at greater depth than if densee metal was beneath the layer being melted. These include surfaces with support structures. The result is an unstable mass of material that is oversize. 156

177 To melt materials such as 316L stainless steel, high temperatures are required. During the build of an SLM component, rapid melting and rapid cooling of the metal occurs at localised areas. The contraction of the material when cooling causes non-self-supporting material to curl away from the powder surface. Support structures are used to stop this curl from occurring, although the support structures will not stop the sag from occurring. Thin wall sections that are close to a geometry that either curls or sags, can have a small amount of shrinkage which occurs as a secondary effect to the distortions. To avoid this shrinkage, using self-supporting geometries is advised whenever possible. The design rules highlight the geometries that commonly distort and suggest self-supporting alternatives Support structure background knowledge Support structures are used to anchor components to the substrate plate during an SLM build. Many process complications can occur because of residual stresses, therefore support structures are required to prevent any geometry movement occurring during the process. Supports are created in specialist software for preparing components for manufacture on various AM technologies. It is possible to design away the use of support structures as explained throughout the design rules. Reducing support usage will reduce the processing time of a component and the also the post-processing time required to remove them. 157

178 When building layer by layer, it is important that the part is connected to a substrate to stop any movement during processing. This is usually done by using a base support structure. Alternatively, parts can be built straight onto the substrate and be removed by a secondary process such as wire erosion. Adding a secondary process also adds time and cost, which needs to be considered throughout the SLM design process. When a part needs to be supported, the supported surfaces will be degraded not only by the supports, but also by the selection a poor surface orientation. It is always best to select an orientation that minimises the use of support. In many situations it is not possible to choose an orientation that reduces the need for support. In these instances it is important that the supports that are used do not obstruct neighbouring geometries that are self-supporting. For example, if a support has no teeth it will be harder to remove, it is then likely that parts will be damaged when they are removed. A careful selection of support can reduce the overall processing time and could mean the difference between a successful SLM component and a failure Orientation background knowledge Part orientation must be considered at all stages when designing new products or redesigning existing products. A good selection of orientation will limit the number of supports required. The following bullet points should be considered throughout the design process on new and revised components. When orientating an existing design, there are three choices to the orientation depending on the required outcome of the component that requires manufacture (see Figure 7.3). 158

179 Figure 7.3: Three choices of orientation depending on requirements. The part orientation can be built in the optimum orientation for the highest quality surface quality and accuracy (see Figure 7.3C). Critical geometries are to be identified to ensure that they are considered in the orientation. The part can be built in the lowest orientation for speed of build and limiting cost (see Figure 7.3A). Excessive support structures and poor surface quality may be the result of this choice. An average of the first two orientations may be selected where time and cost is not constrained and the surface quality is not particularly critical (see Figure 7.3B). 159

180 7.2 Design rules Part orientation and surface roughness Surfaces under an angle of 45 require the use of support structures to avoid build failure. Surfaces with an orientation less that 45 will also consist of a poor surface quality. Depending on the application, this poor surface quality may need to be removed after SLM manufacture (see material allowance). The support structures will not improve the surface quality for surfaces that require them. The supports will add post-processing time as they will need to be removed from the substrate plate. Figure 7.4: Illustration of the relationship between surface quality, orientation and angled surfaces. Surface and the facing direction are identified by hatched line cross sections. As illustrated in Figure 7.4, the vertically orientated surface had the best surface quality. Therefore, the optimum orientation is at 90 perpendicular to the substrate plate. 160

181 45 up-facing surfaces degrade until they suddenly improve at 0. Critical Geometries must be identified and aimed to be built within the light grey zones illustrated in Figure 7.4. The optimum orientation will expect to give a roughness average of around Ra 17µm. A poor surface roughness will be around Ra 34 µm Material allowance / geometric compensation High accuracies, tight tolerances and specific surface finishes may be required on a final product, rather than the grainy surface quality that is direct from SLM. Down-facing surfaces of an SLM component will not be smooth; they will also be inaccurate. The orientation selection in Figure 7.4 suggests avoiding these surfaces, this is not always possible. Combinations of near net shape SLM components and post-processing can achieve any desired accuracy. This design rule illustrates how much material allowance will be required on surfaces of SLM parts for post-process material removal; this will allow the achievement of high accuracies. Figure 7.5 illustrates the surfaces described in the rules. 161

182 Figure 7.5: Illustration of the surfaces described in design rules Up-facing surfaces The surfaces will be dimensionally accurate without removing material, although slightly uneven. Add 0.3mm to up-facing surfaces to remove material for a flat, but not completely dense surface. Traces of the SLM inherent finish will be evident. This will remove uneven discrepancies from top surfaces. Add 0.7mm to the up-facing surface so that material can be removed to leave a fully dense metal surface Side walll surfaces No material will need to be added to side walls to ensure that they are flat; the skin of the walls will be flat and accurate to within ±0.05mm. 162

183 To remove the skin for a fully dense surface, 0.12mmm will need to be added to the surfaces and removed. This is mainly useful for aesthetic and cleaning purposes of the component, and where tight tolerances are critical Down-facing surfaces Down-facing surfaces should always be built oversize by 0.8mmm (this is due to no underlying solid material). At the extremes of the surface the material can be undersize by approximately 0-0.3mm depending on the size of the part. Material compensationn described in Table 7.1 overcomes this problem. Sizes are illustrated in Figure 7.6. Support structures do not reduce the amount oversizee material on the down-facing surfaces. Down-facing surfaces are exposed to process inherent stresses resulting in a convex surface. The larger the surface, the worse this effect becomes. Table 7.1 shows how much material allowance is to be added to the CAD data for post process material removal. This will allow flat and dense surfaces to be achieved. The values in this table show how much material to add in CAD, not the amount to remove. Figure 7.6: Showing the three sizes of the parts described in Table

184 Surface Dimensions: 30 x 30mm 20 x 20mm 10 x 10mm Material to be added for removal: 0.3mm 0.2mm 0mm Material added for dense metal surface: 0.8mm 0.2mm 0mm Table 7.1: Shows how much material to add to the CAD data on down-facing surface to make the material removal in Table 7.2 possible. Table 7.2 shows how much material needs to be removed after the build is complete to achieve flat and dense surfaces. The values in Table 7.2 are the sum of the 0.8mm that builds oversize + the values in Table 7.1. Surface Dimensions: 30 x 30mm 20 x 20mm 10 x 10mm Material removal until flat surface: 1.1mm 1mm 0.8mm Material removal until dense metal surface: 1.6mm 1mm 0.8mm Table 7.2: Shows how much material is required for removal to achieve flat and dense surfaces. The values of material allowance are minimum values of material that can be added to a part for post-machining. It is possible to add more material; this should be considered together with the post process removal technique. Adding more material may be suitable for milling, where it will add many hours for hand finishing Minimum size between features When designing detail into a component such as slots, keyways and channels, a minimum gap size must be applied to prevent surfaces merging together. 164

185 Minimum size between features = 0.3mm. see Figure 7.7. Figure 7.7: Example of a 0.3mm minimum gap feature. It is not recommended to have gaps as small as 0.3mm if the orientation of that feature is less than 45, supports will be required and the slot will be closed Minimum wall thickness Thin walls cannot be designed smaller than the capabilities of the SLM hardware. The smallest wall thickness possible is 0.4mm as shown in Figure 7.8. The accuracy of the 0.4mm wall will be ± 0.02mm. Walls thinner than 0.4mm will not build on the final metal part. 165

186 Figure 7.8: Illustration of minimum wall thickness Chamfers and radius It is not possible to build 0 overhangs (shown in Figure 7.9) that are parallel with the substrate plate without the aid of support structures. If support structures are used the geometry will still be inaccurate with a poor surface quality. The following rules describe alternative design features with self-supporting dimensions. Figure 7.9: Illustration of a ledge and alternative geometries, chamfer, and radii. 166

187 Alternative orientations can eliminate the need for support and can improve the surface quality of the down-facing surfaces (for the geometry shown in Figure 7.9). Unless a functionn of the sharp edge is critical, either a chamfer or a radius should be used instead of the ledge. This will eliminate the need for supports, improve the surface quality and improve the accuracy of the design feature. The following shows the correct use of chamfers and both convex and concave fillet radii Chamfers A chamfer is a simple geometry that can be made via SLM without the use of support. Figure 7.10: Example of chamfers, 45, 55 and

188 Chamfers can be built at any orientation above 45, the surface quality will improve as the orientation becomes larger (see orientation). Figure 7.10 illustrates various chamfer angles. 45 is a typical chamfer angle. This chamfer can be built using SLM without any process complications. All angles are relative to the substrate plate as shown in Figure Figure 7.11: Illustration of angles relative to the substrate plate. If height limitations are relevant, a 45 chamfer can be the shortest height (Z-axis) rather than a radius, see Concave and Convex Radii Concave fillet radius It is not possible to build tangential radii larger than 3mm on SLM without supports at the vertical orientation shown in Figure This is because much of the radius becomes down-facing as the radius grows upwards in the z-axis. 168

189 Small radii up to 3mm with no supports can be built, although the accuracy on such small radius willl be so poor that they will not visually resemble a radius. Figure 7.12: illustration of a tangential radius (traditional radius). To make a tangential radius build, support structuress will be required or an alternative orientation will need to be used. Using supports will prevent the radii from curling up during a build, but willl not improve the surface quality of the down-facing be changed to suit a tangential radius, alternative radius construction should be considered. See Figure 7.13 and Table 7.3. sections of the radius. If the orientationn of a part cannot To design a radius to build accurately and successfully, the amount of down-facing overhang on the curve needs to be reduced by decreasing the tangent at the highest position of the curve. Table 7.3 illustrates the construction of a radius that fillets overhanging ledges. Traditional radii cannot be built self supporting; therefore only dimensional values are given rather than a straight forward radii value. 169

190 Figure 7.13: Illustrations of the details presented in Table 7.3. Overhang / Ledge Size (A) Bottom Tangent (B) Top Tangent (C) Height Of Radius (D) 0.5mm mm 1mm mm 1.5mm mm 2mm mm 2.5mm mm 3mm mm 3.5mm mm 4mm mm 4.5mm mm 5mm mm 5.18mm + 45 chamfer Table 7.3: Table of dimensions for designing self supporting radii. When a radius is used as a fillet for an overhanging ledge larger than 5.18mm, the height of the radius becomes larger than the width of the radius. This is a possible design limitation; therefore, it is recommended that a 45 chamfer is used from 5.18mm upwards (see Chamfers). 170

191 Convex fillet radius Just like concave fillet radii, it is not possible to build a traditional tangential radius at the orientationn shown in Figure Small radii up to 2mm with no supports can be built, although the accuracy on such small radius willl be so poor that they will not visually resemble a radius, with a tolerance of ± 1mm. Figure 7.14: Illustration of tangential convex radius. To make a tangential radius build, support structuress will be required or a selection of an alternative orientation will be required. Using supports will prevent the radii from curling up during a build, but willl not improve the surface quality of the down-facinaccurately and successfully, the amount of down-facing overhang on the curve needs to be reduced by decreasing the tangent at the highest sections of the radius. To design a radius to build position of the curve. Table 7.4 and Figure 7.15 illustrate the construction of radius that fillets overhanging ledges. 171

192 Figure 7.15: Illustration of the details presented in Table 7.4. Overhang / Ledge Size Bottom Tangent Top Tangent (C) Height Of Radius 0.5mm mm 1mm mm 1.5mm mm 2mm mm 2.5mm mm 3mm mm 3.5mm mm 4mm mm 4.5mm mm 5mm mm 5.5mm mm 6mm mm 6.5mm mm 6.73mm + 45 chamfer Table 7.4: Table of dimensions for designing self supporting convex fillet radii. When a radius is to fillet an overhanging ledge larger than 6.73mm the height of the radius becomes larger than the width of the radius. It is recommended that a 45 chamfer is used from 6.73mm onwards to reduce the height of the feature (see Chamfers). 172

193 Curved surfaces Curved Surfaces are made up of varied radii in both the convex and concave directions. Figure 7.16 illustrates simple curved surfaces made up of just one radius. Figure 7.16: Illustrates simple radii as a curved surface. The more radii that make up the curved surfaces, the more complex that the surface becomes. This is illustrated in Figure and in more detail in Figure 7.18, both convex and concave radii make up the curved surfaces. Figure 7..17: Illustrates two complex parts with upward and downward facing curved surfaces made of both concave and convex radii. 173

194 Figure 7.18: Diagram showing the segmentation of a complex curve into convex and concave radii. When building complex curves, the combination of radii should be segmented as shown in Figure 7.18, the segmented radii values should be taken from rules for designing radii for both convex and concave fillet radii (see concave and concave fillet radii design rules) Round holes built flat As described earlier, there is a minimum gap size that can be built between features before that gap closes becoming solid. The same rule applies to holes when built flat and in a self-supporting position parallel to the substrate plate. This design rule illustrates the smallest holes that can be built before the holes become closed and no longer a hole. The smallest hole possible to be built parallel to the substrate plate is 0.7mm diameter (Ø 0.7mm), see Figure

195 Holes smaller than 0.7mm diameter will leave a small witness mark on the surface of the part, this dot can only be seen when the skin of the SLM component is removed. Figure 7.19: Illustratess the smallest hole at a vertical orientation Maximumm upright holes withoutt support Round holes are not self-supporting when built in an upright orientation. It is possible to build small holes without support structures in this orientation because there are few layers to accumulate a curl deformation. The accuracy of these holes will be very poor as the highest layers will sag into the round geometry, obstructing the accuracy of the hole. The largest round hole that can be built without support structures is 7mm diameter (Ø7mm) as shown in Figure The smallest hole that can be built without support structures is 1mm diameter (Ø 1mm). Smaller holes will not be recognisable at this orientation as shown in Figure

196 Figure 7.20: Illustrates all round holes that can be built with no support. Holes above 7mmm diameter will curl away from the powder bed and will cause an obstruction to the powder recoat system. This is illustrated in Figure Figure 7.21: Shows holes larger than 7mm curling away from the powder bed causing an obstruction to the powder recoat system. 176

197 Figure 7.22 illustrates the distortion on an unsupported hole larger than 7mmm diameter. Figure 7.22: Example of curl distortion after the SLM build. Any size hole can be built with support structures; they will need post-process removal. The accuracy of all holes, with and without the support structure will be poor as the tops of the holes will collapse into the bore of the hole. This sag will be approximately 0..5mm. An example diameter with 0.5mm sag is shown in Figure

198 Figure 7.23: Illustration of 0.5mm sag (distortion) on Ø7mmm hole. Depending on the functionally of the holes, the holes can be designed as near net shape pilot drilling holes for post-processing. This is typical of a hole that requires a high accuracy, and secondary processing such as tapping a thread or reaming. The general tolerance of round holes is ± 0.5mm. This tolerancee will be different at various points of the circumference of the hole Self supporting holes Accuracy and repeatability are both desirable when a design is being created for manufacture. Round holes are difficult to build in an upright position without using support structures. Not only do these support structures need to be removed from the holes, they may be impossible to remove as they are in a position that cannot be accessed. The 0.5mm sag that obstructs the bore of a hole will always occur even with supports. 178

199 The following rules describee partially round holes that have been developed as being self supporting so that they are accurate, repeatable and do not need support structures. The following self supporting hole designs are split into two sections, peak and circle, see Figure Figure 7.24: Illustration of the hole geometry split into two components. Some peaks are constructed with curves and others with flat 45 lines. This is a result of minimising the overall height of the peak retaining the geometry as close to a round hole as possible. Figure 7.25 shows example self-supporting holes. 179

200 Figure 7.25: Examples of self-supporting holes designed using the following design rules. All dimensions of holes from 2mm to 30mm diameter are illustrated in Table 7.5. The holes have been optimised to keep as much of the circumference of the circle as possible, and the height of the peak to a minimum. All dimensions of the holes are illustrated below in Figure Figure 7.26: Illustrates the dimensions given in Table

201 Hole Ø Circle Height Peak Peak Start Peak Top Area of Hole 2mm 1.85mm 0.3mm mm 2 4mm 3.38mm 0.95mm mm 2 6mm 5.63mm 0.95mm mm 2 8mm 7.31mm 1.66mm mm 2 10mm 8.95mm 2.6mm mm 2 12mm 10.58mm 3.7mm mm 2 14mm 12.22mm 4.58mm mm 2 16mm 13.97mm 5.325mm 45 Angle n/a 214.6mm 2 18mm 15.76mm 5.942mm 45 Angle n/a mm 2 20mm 17.36mm 6.769mm 45 Angle n/a mm 2 22mm 18.96mm 7.591mm 45 Angle n/a mm 2 24mm 20.63mm 8.338mm 45 Angle n/a mm 2 26mm 22.27mm 9.114mm 45 Angle n/a mm 2 28mm 23.76mm 10.04mm 45 Angle n/a mm 2 30mm mm 10.75mm 45 Angle n/a 751.1mm 2 Table 7.5: Table of holes from 2mm to 30mm, all dimension of the holes are highlighted here. Holes with a radius of 1mm and 2mm will have a tolerance of ± 0.3mm. Holes with a radius 3mm and 4mm will be accurate within a tolerance of ± 0.2mm. All holes larger than this up to a radius of 15mm measured to a tolerance ± 0.1mm Pilot drilling If round holes are required that can not be built in a self-supporting position, it is possible to post-process drill the holes to a desired size. Build all holes for pilot drilling 7mm diameter or smaller. These holes do not require support and are sufficient for post process drilling at any size. 181

202 Reaming accurate holes If accurate holes with a high surface finish are required, material should be added to the bore of round holes for post-process reaming. Shapes other than round holes will not ream correctly, this will cause inaccurate holes with poor concentricity. Reaming non-circular holes is not recommended as it is likely to damage the reamer. The rules for building holes for reaming apply to all holes (see holes design rules) Threads It is possible to tap a round hole as a post process operation before or after pilot drilling, as long as the correct tapping drill size is used. It is possible to include the thread in the design of a component, as long as the correct tap is run through the hole as a secondary process. The rules for building holes for threads apply to all holes (see holes design rules). Tapping holes other than round hole will give inaccurate threads. The tap will pull towards the area with the lowest cutting resistance (least material), this will result in the thread forming in the wrong position. This concludes the end of the Design Rules. 182

203 Chapter 8: Discussion 8.1 Introduction The aim of this research was to create design rules for SLM. To accomplish the aims of this research, the following objectives were addressed as presented throughout this thesis: Evaluate the SLM process, and identify the geometric limitations of SLM. Perform analysis and synthesis of the experimental results and create a first draft of design rules appropriate for use in design practice. Evaluate the design rules to establish their effectiveness and inform their further development. The methods used to address the aims and objectives were appropriate and successful in terms of developing the first design rules draft for SLM. The methods for the experimentation created the data required to produce design rules. The methods then led to a smooth transition between the results and the design rules. To identify areas to improve the methodologies a retrospective view was taken. If this research was to be repeated, several protocols and experimental methods would need to be reconsidered to improve the repeatability and accuracy of the experimental results, and therefore, the accuracy of the design rules. 183

204 8.2 Literature review Since the introduction of RP in the early 80 s, the trends in manufacturing prototype products have progressed into an increasingly popular research field and industry of RM. SLM is a popular research topic for metallurgists and engineers because of the ability to produce parts from engineering grade metals. This has led to SLM being recognised for being potentially advantageous to high-value industries such as aerospace and medical. Metals are complex materials that have properties that suit many applications. Metals have numerous chemical and mechanical properties that may change during processing, which may be advantageous, or disadvantageous depending on the application. Almost all previous literature is based on the physics of SLM. However, since industrial interest has increased, the number of published articles has remained the same. One possible explanation to this is that companies considering the use of SLM within their business, and the companies that are producing the machines (EOS, Concept Laser, MTT) are protecting their potentially valuable IP. Research into materials and process parameters has driven the developments of SLM technologies from just sintering powder particles together (SLS), to fully melting (SLM) an inventory of materials such as aluminium, titanium and tool steel. As metallurgists have increased their knowledge of the physics involved in the SLM process, the inventory of available materials is increasing and the technologies are developing alongside. This has made it possible to create functional parts and has informed the research within this study. The review of existing literature has identified that there is a clear need for design rules. Not only was the literature from academic backgrounds (Hague. R, 2006), AM 184

205 practitioners and research from industries who are investigating the use of SLM as a RM process within their manufacturing facilities have also recognised the need for design rules (Pullin & Offen, 2008; Wohlers, 2009). It appeared that designers were either avoiding SLM, or were designing for SLM using guidelines for conventional processes such as casting, milling or even injection moulding. Wohlers, (2009) stated that there is very little design experience in RM, and for RM to succeed as a viable manufacturing process, designers will need to be educated, and become less reluctant to design for new RM processes as well as the conventional processes they have become used to. The possibilities offered by advanced and complex SLM designs have identified the potential for SLM to become a valuable process for new business applications. It was found that the lack of design experience needs addressing to prevent it becoming a bottleneck in the development of SLM. If designers know how to design for SLM they will become less reluctant to use it as a manufacturing process which will increase the success of SLM within industry until superior technologies are developed. 8.3 Methodology and test methods Two phase methodology To address the aims and objectives of the research, the methodologies were separated into two phases. Phase one was a quantitative evaluation of the SLM process that identified geometrical limitations. The second phase was qualitative and was used to initiate the design rules, and to establish their effectiveness and to inform further development. 185

206 Dividing the research into two phases meant that the overall research goal could be addressed in logical steps that could be addressed individually, rather than one larger step that consisted of multiple methodologies and too much data to review and analyse. After the first phase was completed, the design rules were initiated by a synthesis of the experimental results which was successful because the results that were created in Phase 1 could be used directly in the design rules. This was encouraged by the Plan-Do-Study-Act (PDSA) cycle (Deming, 1995) used as the experimental methodology. The experiments were only complete once the results could inform the design of the following experiment in series. An alternative method to the two phase approach may have consisted of initiating the rules as each associated experiment was completed. This would not allow the design rules to include any unexpected research findings, and rather than encouraging the use of SLM, the rules would be negative by only highlighting the process limitations Fixing parameters All process parameters were fixed throughout the experimentation. This makes the study independent from other SLM research where the build parameters have been the main focus. Fixing the parameters ensured that the study was focused on gathering the required dimensional data to enable the design rules initiation, rather than optimising process parameters to reduce geometrical deformations. At the beginning of this study, time was dedicated to benchmarking the SLM parameters used so that the highest quality build of the MCP SLM Realizer 250 was achieved throughout the experimentation. These qualities included surface finish, density, and the parameters that gave the most repeatability to the process. 186

207 If the process parameters were not optimised the machine would under-perform which would be unrealistic in comparison to an SLM within a commercial environment. One reason for developing design rules is to bridge research into mainstream manufacture, which will contribute to SLM becoming a true RM process. Therefore, it was necessary that all parameters were optimised to what was acceptable within industry for the MCP SLM Realizer 250 machine available at the start of this study. One limitation of fixing the parameters is that the rules will only be accurate to the SLM machine and parameters used in this study. However, the principles and features of the design rules will be applicable to all SLM machines with all variations of materials and parameters. If this study was repeated with different parameters such as layer thickness and laser power, the design rules would remain the same, and the dimensional details would be different Experimental conditions The ambient air temperature and humidity of the environment where the SLM machine was situated were not fixed, so that the experimentation would be completed within a realistic workshop environment. To identify if the uncontrolled environment affected the repeatability of the test parts, one test could be built several times at different ambient conditions. When the curl categories were established to quantify the heat induced distortions during SLM processing, the test parts became repeatable when the curl value was 0 or 1, which suggests that controlling the environment would not improve the accuracy of the design rules. Even if the environment was controlled, and the results were more accurate and repeatable, the shadowgraph measurement method used to measure all test pieces would not have been accurate enough for the improvement to be observed. Further discussion on accuracy requirements and measurement test methods is explained throughout this chapter. 187

208 8.3.4 Test methods The dimensional accuracy of all test parts was measured using a shadowgraph. A shadowgraph protocol was created that consisted of several steps to measure the accuracy of all geometrical features, regardless of the scale of the surface quality. The test parts were designed to be measured using the shadowgraph. The parts consisted of the design feature being investigated, and a feature with a known dimension that was used as a reference for scaling. The shadowgraph was a suitable method considering the process repeatability and the surface quality of the test parts. The shadowgraph protocol led to several significant findings, including a relation between distortions that were visible during the build, and the accuracy of the test parts. This finding was then considered in the analysis of all experimental test parts. After building the test parts, the shadowgraph protocol consisted of printing a semitransparent profile, aligning the print with test parts on the shadowgraph display, and importing the image into CAD to compare against original geometry. To measure a part, a total of six stages were required which were all open to accumulating error. According to experienced quality engineers, the tolerance of the measurement technique was expected to be within a tolerance of 0.05mm. This tolerance is typical when measuring ground surfaces, therefore, SLM parts may have less accurate tolerances as the surface quality of part is lower. Table 8.1 describes each step and the possible errors that the shadowgraph method may have been exposed to. 188

209 Step 1 Print scaled semitransparent CAD profile of test part. Possible Error Inaccurate print out. The tolerance of the scaled print out is unknown. 2 Hold part on shadowgraph The part needed to be held squarely in a grinding vice, this could have been slightly off square by one or two degrees. 3 Overlay and align semitransparent printout This relied on the shadowgraph user s knowledge to align the image over the part correctly. 4 Take photo of the shadow The image needed to be taken standing back from the shadow with the camera zoomed in to reduce any barrel effect in the image. A small barrel effect may still have been present. 5 Import image in CAD and Scale to size using known dimension. 6 Offset lines on the original CAD image to compare against the SLM part shadow image The accuracy of aligning the geometries was dependant on the user s visual judgement. The known geometry could have been slightly inaccurate, and the scaling of the image may not have been precise. This was completed using visual alignment of two lines, in many cases there was weld spatter on the part which would cause inaccuracy, and the person completing the task could misalign the CAD against the SLM part image. Table 8.1: Each step of the shadowgraph protocol and the possible errors that could accumulate and cause inaccuracy. The shadowgraph test method was appropriate for this study. However, the test methods would need refinement if the study were to be repeated. To improve the number of steps required to measure each part would need to be reduced so that there is less opportunity for accumulative error. The first step to be removed is printing the scaled geometry profile, which would avoid possible printer inaccuracies, and also inaccuracies of overlaying the image against the shadowgraph display. Taking the digital photograph of the shadowgraph display is the when the greatest amount of error could occur. If there is a barrelling effect in the photograph, the scale of the image 189

210 will be distorted, which will lead to inaccurate measurements. To eliminate this, a wider camera lens could be used, or a shadowgraph with an onboard ability to capture digital images. Many of the geometries measured had rough edges, so measuring point to point using the on-board optical system would be inaccurate. Table 8.2 shows the reduced shadowgraph protocol with less opportunity for error. Step 1: Hold part on shadowgraph Step 2: Step 3: Take photo of the shadow and import directly to PC at correct scale Offset lines on the original CAD image to compare against the SLM part shadow image Table 8.2: The reduced steps of the shadowgraph protocol. Since a retrospective view of this research could be taken, it appeared that the shadowgraph protocol was not the only method that could be refined if the research was to be repeated, but the number of test parts could be reduced. The large number of test parts in this study was required, although, if the amount of parts were to be reduced a more accurate and time consuming measurement method could be considered. Using a Coordinate Measuring Machine (CMM) would be a more accurate single-phase method of measuring the SLM test parts. CMM requires much more time to operate than using a shadowgraph, but is capable of measuring complex 3-dimensional geometries at higher accuracies. It was deemed unnecessary to use a more accurate measuring technique when the SLM process consists of such rough surfaces and many unrepeatable test parts. If CMM is used, only the self-supporting parts that built with a curl of 1 or 0 would need to be measured, because all other parts have already proven to be unrepeatable and detrimental to the build completing successfully. The poorest part surface qualities with a curl of 2 and 3 will cause the CMM probe to touch at different heights of the peaks and 190

211 troughs within the rough surface, and also on weld spatter which will cause inaccurate and unrepeatable measurement results (see Figure 8.1). Figure 8.1: Illustration showing weld spatter affecting the touch probe CMM Accuracy requirements The methodologies used within phase one of this research were open to measurement inaccuracies that may have affected the accuracy of the results and subsequently the design rules. Greater accuracies could have been achieved, however, the main aim of this research was to initiate design rules, rather than to create an optimised measurement process for measuring SLM parts. The measurement methods used were accurate enough to gain an understanding of the SLM process, and to identify trends in geometries such as the curl distortion relating to accuracy. During this research, many process-inherent geometry deformations were identified, and to avoid these problematic geometries, alternative designs were developed. It is possible to measure more accurately than ±0.05mm, although, the increased accuracy was not necessary at such an early stage in the design rules development, and the rules developed with the scope of this research are a first draft. 191

212 The design rules could be produced at a higher accuracy if the process repeatability improves, and the surface quality capabilities of newer generation machines are higher than the MCP SLM Realizer 250 that was used in this research. As this was the first attempt at creating process-specific design rules for SLM, one important result was that the methods used were successful in creating the design rules, and evaluating them to inform further development. The full methodology and measurement protocols are described in Chapters 2 and Experimental Results minimum surface orientation The first experiment in this research was aimed at identifying the lowest orientation a flat surface could be built at. The results showed that 45 was the lowest orientation, and when angles between 40 and 45 were built, there was variability in the results as 40, 43 and 44 built on one occasion only. 45 was chosen as the lowest orientation because it was the only dimension that was repeatable, however, it is possible that the smallest orientation will be less than 45 with more accurate readings of 44.9 or even This amount of detail was not considered during the orientation experiment because the accuracy of downfacing surfaces is affected by poor surface quality. The layer thickness used throughout this research was 0.075mm. The minimum orientation was identified as 45, therefore, the overhang distance of each layer was also 0.075mm, as shown in Figure 8.2. This leads to question whether or not the minimum 192

213 surface orientation would be reduced if the layer thickness was smaller, and will the 0.075mm layer overhang remain the same for all layer thicknesses? Figure 8.2: Minimum orientation angle with 0.075mm layer thickness and 0.075mm layer overhang. Work by Vandenbroucke & Kruth (2007) found that the minimum orientation was 20, and orientations less than 20 required support structures. This result was recorded with a layer thickness of 0.03mm, which was two and a half times less than the 0.075mm layer thickness used in this study. As the layer thickness was 0.03mm and the minimum surface orientation was 20, the overhang of the downfacing layers was calculated at 0.08mm, which is close to the 0.075mm overhang identified in this research. This suggests that a common overhang size for all layer thicknesses is approximately 0.075mm to 0.08mm. Not only was a different layer thickness used, the 20 angle was achieved at three different hatch space distances (100, 120 and 140µm) and three different scaning speeds (90, 140 and 190 mm/s). The machine that was used was a Concept Laser M3 Linear, which was different to the MCP SLM Realizer 250 used in this research. Research findings by Sutcliffe et al. (2005) identified a minimum surface orientation of 38, and Brooks et al. (2007) identified a minimum orientation of 40. The research was completed using an MTT SLM simular to the SLM used in this study. The layer thickness used is not mentioned, however, Sutcliffe concluded that the parameters for an 193

214 overhanging feature would need to be controlled independantly from the rest of the part, with a different laser power and a scan strategy that reduces residual stress in the layer. Sutcliffe stated that the scan vectors should be increased to build low orientations, this was also repeated in greater detail by Mercelis & Kruth (2006). Figure 8.3: Residual stress distortion caused by scanning strategy as explained by Mercelis & Kruth (2006). Research by Mercelis and Kruth identified that the scan strategy has a major influence on the residual stress of a part, and therefore, the amount of distortion that occurs during a build. Scanning in short lines across a narrow part creates more residual stress than when scanning longer lines along a perperdicular axis (See Figure 8.3). The more lines that are scanned in one direction the more the curl distortion occurs. Therefore, the fewer scan lines there are in one axis, the less distortion will occur. It appears that the minimum surface orientation of a flat surface correlates to the layer thickness. As the layer thickness of an SLM part decreases, the lower the surface orientation can be. By reviewing other research, it was identified that when the layer size is reduced, the size of the layer overhang remains close to 0.075mm, which results in the ability to build at lower orientation without the use of support structures. However, the 194

215 accuracy of the layer overhang, and the build orientation is subject to carefully controlled parameters. The availability of pre-heated build platforms in SLM machines has also improved the ability to build at lower orientations. This will increase the overall part quality as all geometric features that are constructed of overhanging geometries will become less constrained Self-supporting radii Both convex and concave radii were investigated. The first major finding was that traditional tangential radii were not self-supporting when built in the upright position. This was because the top of a concave radius becomes down-facing, and the bottom of a convex radius starts building down-facing as shown in Figure 8.4. Figure 8.4: Down-facing sections of a convex and concave tangential radius. Previous investigations have identified that when a component is self-supporting, there is a reduction in the part distortion that occurs during the build, and subsequently, an increase in the accuracy of the geometric feature in question. Experiment 10 was completed to identify the limitations of convex and concave radii so that they can be designed to be selfsupporting. The tangents of the down-facing radii section were changed (the down-facing 195

216 surfaces shown in Figure 8.4), and after several experimental iterations the minimum tangent for each test part was identified for making self-supporting radii. When the tangent on the radii increases, the overhanging size of each layer is reduced creating a steeper curve. This experiment followed the investigation into the minimum surface orientation where the 45 minimum orientation was identified. This result was expected to be common to all geometries, including the tangent of a curve. However, this was not the case as many minimum tangent results were lower than 45. The larger the radius became, the larger the tangent angle needed to be. For example, the smallest radius required a 28 tangent and the largest radii required a 40 tangent. These unexpected results showed how a simple geometry such as a radii can become so much more complex when it is made using SLM. The minimum surface orientation of 45 still remained the same, although curved surfaces were an exception. To explain why such low tangents were required, the number of layers for all parts was analysed. The tangent of the curve changes as the layers build up along the z-axis. On convex radii, the tangent of the curve increases, and on concave radii the tangent decreases. Depending on how many layers there are less than 45, curl will accumulate. The more layers of a radius that are less than 45, the more chance curl has to accumulate, which is why larger parts require a larger tangent angle. The number of layers may change if different process parameters and materials are used. The number of layers that can be built under 45 without curl occurring was not measured. However, if the SLM is to become more accurate, distortions may need to be measured in greater detail. To increase the accuracy and to reduce the minimum self-supporting angle, it is possible that the layer thickness needs to be reduced, which will be a compromise with added build time. 196

217 There was a small amount of variability in the results once they were measured using the shadowgraph method; although, it was still possible to identify that the less curl distortion that occurred during the build, the more accurate the parts were. When curl had occurred, the parts were undersize as a result of the curl pulling the SLM part boundary away from the correct geometry. The results from the radii experiments gave a good insight to building more complex geometries that are constructed from surfaces other than flat. The constantly changing tangent showed how it is possible to build outside the limitations of SLM if it is redeemed within a few layers Self-supporting holes The aim of the hole experiments was to create self-supporting alternatives to a round hole design, and so that intricately detailed components can be constructed of holes and channels without needing access for support removal. The results of the radii experiments were then used to develop a hole design with a peak that was optimised to be as close to a round hole as possible. One finding from the hole experimentation was if the top layer of a design feature has evidence of curl, it will accumulate onto another design feature that is built on top in the z- axis, only if the top part is built at the lowest possible orientation. This may cause a failure on a feature that would normally be self-supporting. This proved that the elimination of curl and the use of self-supporting geometries are important to create a smooth transition between two or more geometries stacked together as shown in Figure

218 Figure 8.5: Two minimum geometric limitations stacked together. The accuracy of the self-supporting holes developed in this study could be predicted based on previous results. The smallest holes appeared to be the least accurate at ±0.3mm and holes at 5mm radius and above were all within a tolerance of ±0.1mm. The accuracy difference between the smaller and larger holes was an unexpected result. One possible reason for this can be explained as a reaction to the heat building up in parts with a small volume. It is possible that the smaller parts could not cool and disperse heat away as quickly as the larger parts. Therefore, the localised heating of the melt-pool had very little dense material to conduct away the heat, resulting in excess material melting against the part boundaries. This occurrence could not be avoided through design changes unless the parts were made larger. This problem would need to be controlled by tailoring the design feature parameters independently to the rest of the part, according to the size of the geometry being melted Variability and repeatability The results of the surface roughness experiments were not repeatable, and there was a large amount of variability present. When two parts with the same design were measured the results were not the same. This also occurred in other geometry experiments. It is possible 198

219 that this was caused by inaccurate measurement methods or it may have been inherent to the SLM process. Once curl values were established, it became possible and repeatable to predict the accuracy of a part. However, in some cases when the same or a similar geometry was built, the surface quality appeared to be inconsistent. To identify the true repeatability of all geometries, it would be necessary to build more than just one or two parts. Although improving the accuracy of the results would improve the accuracy of the design rules, it may not have improved the results of this study because the rules were a first version that are valued because of the geometries, clarity, and comprehensiveness that would inform designers. As the experiments progressed, the test parts became compound and were constructed of geometries from earlier experiments within this research. This resulted in accuracies from the fundamental geometries being applied to the compound designs. The geometries used to create the compound test parts were repeatable to within ±0.1mm as they had no or very little curl distortion. The surface roughness experiments were completed twice as the first attempt showed high amounts of variability between each measurement. This variability appeared to be inherent to the SLM process and was caused by the adherence of weld spatter to the part surfaces (See Figure 8.6). This experiment was repeated with marker pen lines added to the test parts to indicate the best measurement position for the device (Taylor Hobson Sutronic 3p). The device was calibrated to within aerospace compliance at Gardner Aerospace Wales Ltd, using a surface roughness gauge. After the second measurements the results showed that the variability was still present. The staircase effect was apparent, however, in some places this was only visual rather than what was measured. Therefore, only an indication to the effect of part orientation was presented in the design rules, along with the highest and lowest mean roughness (Ra) values taken from the results. 199

220 Figure 8.6: Weld spatter occurring during the build of an SLM test part. One possible method of controlling the variability and repeatability was used when creating self-supporting holes in Experiment 8. In this experiment the variability in the results was caused by the uncertainty of when curl distortion had exceeded a level that jeopardised the building process. Round holes from a radius of 1mm to 15mm weree built and stopped at the point where they were no longer self-supporting and curl distortion had occurred. After the first attempt, it proved difficult to select which layer to stop the build at because there was a fine line between safe and unsafe amounts of curl. The experiment was repeated with smalll changes that controlled the variability by creating a zone where the actual results lie. As the first sign of curl occurred, the layer number was recorded, and the part was left to continue building until it was clearly unsafe as too much curl had occurred. The minimumm and maximum curl layer heights were then plotted in a graph and the mid-point between each result was selected as a predicted maximumm self-supporting height for holes with a radius of 1mmm to 15mm. The parts were then rebuilt and measured to an accuracy that was repeatable from earlier experiments. If the variability was not controlled, the design rules for holes could not have been created with minimum geometries, as the results would have been too random. 200 However, if

221 variability is present in the SLM process it is important that it is identified so that it can be reflected in the design rules and considered within the accuracy of the overall SLM process. 8.5 Design rules development Initiating the design rules To complete this research the methodology was split into two phases. The first phase was aimed at collecting all of the data required to inform the design rules that were created in Phase 2. The first task in Phase 2 was to initiate the design rules from the experimental data so that the rules could be used and understood in design practice without the designer needing to interpret the data themselves. To select an appropriate presentation technique throughout the design rules document, and to understand the level of detail required, existing design rules for more traditional and well established manufacturing processes were reviewed. These processes included casting, extrusion and injection moulding. This synthesis of data was not as difficult and time consuming as originally expected because the Plan-Do-Study-Act (PDSA) cycle (Deming, 1994) used to complete the experiments ensured that the experimental results were comprehensive, and were completed to a level that could be used to inform the following experiment, therefore, they would inform the design rules without extensive interpretation required. The PDSA cycle proved to be a successful methodology selection during the experimentation. It was used within individual experiments and for the bigger picture over all experiments. Within each experiment, the PDSA cycle ensured that the required results were achieved by completing and analysing one test part, and then repeating it with 201

222 controlled changes until the required geometric limitations were identified. Over the entirety of the experiments the PDSA cycle only encouraged progression from one experiment to another if the results of one could inform a dimensional starting datum to the design of the subsequent experiment. Each experiment was based on fundamental geometries that were addressed in the design rules. Therefore, each experiment was associated with a particular design rule. This resulted in all experimental results being directly linked to foreseeable design rules, and presented ready to be used in the rules as they had already been optimised for designing test parts. The methodology is explained in greater detail in Chapter Design rules evaluation The design rules were initiated from the quantitative experimental data, as described in Chapter 6. The first draft of the design rules document (see Chapter 7) was evaluated by six people, each with a different role within the design industry. The results gave a good feedback to the structure and content of the design rules, and also a good indication of further design rule development requirements. The six individuals chosen to evaluate the design rules were selected because of their capability to provide a constructive feedback from different areas in design. If more than six interviewees were selected, more constructive feedback would be recorded and the accuracy of the results would improve. However, six interviewees that were highly experienced within their design speciality were more relevant than a larger number of interviewees with less experience. The more experienced interviewees would provide more detailed and in-depth feedback that would have a maximum benefit on the accuracy of the design rules at such a critical stage in their development. 202

223 Large amendments to the design rules would need to be addressed early in their development so that later revisions could be focused on refinements. Further design rule developments may require a different evaluating method such as focus groups. Using a larger number of people in focus groups would allow for accurate results when the design rules are being refined for a wider audience. If the rules need to be tailored specifically for just industrial designers, the focus groups would need to consist of several industrial designers rather than participants from several different design disciplines. If later developments of the rules were to remain generic to all roles within design, a mixed focus group, as used in this study would be required. Focus groups were not used in this study as the feedback required was to be individual to each of the six design professionals, without any dominating respondents who would influence the others. Further design rule development will require user testing so that the rules are fully evaluated within their intended application. A design challenge could be presented to designers so that the rules could be testing in actual design practice. The feedback would determine whether or not the design rules are qualified for use in industry. A design challenge was considered as inappropriate in this research study because it may have tested the design ability of each designer, rather than the content of the rules. The results of the design challenge would not indicate whether the rules are difficult to follow, or whether it is just the process that is difficult to design for. An example of a design challenge is shown in thesis work by Burton (2005). 203

224 8.6 Future design rule optimisation When the design rules are developed further, they will need to be proven effective in industry to encourage their use, and to drive the success of the technology outside of research. Currently, SLM knowledge lies with the SLM users and machine suppliers rather than with product designers. The design rules created in this research are restricted to using the MCP SLM Realizer 250 used in this study. Newer SLM machines provide greater accuracy than the first generation machine that was used in this research. Different materials and process parameters also produce different results, however, the principles of each design rule will remain the same and only the dimensional details will vary, for example down-facing surfaces will be the least accurate. The results of the design rule evaluation showed that some designers are reluctant to apply the SLM design rules, and would expect them to be applied by the SLM technician. For the most effective use of the rules they should be applied at the detailed design stage of the design process before any manufacturing begins. Once a product has been designed and the designer then relies on the manufacturer to apply the rules, the functionality and aesthetics of the component will need to be re-evaluated to ensure that they remain the same. This adds two additional stages to the design process as illustrated in Figure

225 Figure 8.7: Flow Chart showing that the designer applying the rules is the most efficient way of applying the rules. To apply the design rules at the detailed design stage of the design process, designers will need to break away from their usual habits and become less reluctant to learn new design methods. Part of the designer s reluctance may be caused by the time involved in learning new processes, especially when they are unsure of the outcomes of SLM in the first place. Initially, it is more realistic that the design rules will be used at the RM and STL file preparation stage of the design process. An ideal scenario to apply the design rules most effectively is if they are used as a tool for cross-disciplinary communication, to fill the gap in knowledge between design and manufacture. If SLM is selected as an appropriate manufacturing process, it must be specified by the designer together with the chosen 205

226 material, during the development of the Product Design Specification (PDS). Firstly, to select SLM as a manufacturing process, the designer must be aware of the principle design rules through a generic design rule document. To complete the detailed design of a part, the designer must then be aware of the detailed geometric limitations. A two way communication must then be formed with the agreed manufacturer who will supply the detailed design rules that represent dimensional limitations of their particular commercial process. Once the designer has applied the detailed design rules the part can be sent for RM. A process diagram is shown in Figure 8.8. Designer Detailed Design Rules Design Process RM Bureau Selection Manufacturing Process Selection Figure 8.8: Process showing the most effective implementation of the design rules. For SLM to be chosen as a manufacturing process, and for designers to become familiar with SLM, it will need to be driven by commercial decisions. For the AM bureaux to become commercially successful and to gain a larger customer base, they should be able to offer design rules. For the sales of SLM machines to increase, the AM bureaux and industries must have a successful customer base and many successful industrial applications. Therefore, to achieve this it would be in the interest of the SLM machine 206

227 vendors to publicly provide design rules that may advance the use of SLM. The design rules may also be delivered through CAD software or be available on the internet, with an option to input the parameters that are provided by the RM service provider. 207

228 Chapter 9: Conclusions 9.1 Introduction The purpose of this research was to identify the geometrical limitations of SLM, and to develop and evaluate process-specific design rules for SLM. The content of this chapter presents the conclusions of each research objective, and describes the contributions that the research has made to new knowledge. Finally this chapter presents a description of future research to implement process-specific design rules, and to advance SLM into becoming a valued manufacturing solution. To accomplish the aims of this research, the following objectives were completed. 9.2 Research objectives Evaluate the SLM process and identify the geometric limitations of SLM The methodologies used for the experimentation have provided a process model for geometrically analysing SLM, which includes test piece design, and a number of process protocols. The shadowgraph technique allowed the required geometrical process data to be gathered for creating the design rules. The general tolerance and repeatability of SLM parts improved as the parts become more self-supporting. The first significant finding was a minimum layer overhang of 0.075mm occurs at a minimum orientation of 45, which was also used to identify a correlation between surface quality and surface orientation. The orientation results were then used to develop other geometries, such as a radii where the tangents needed to be minimised so that they are self-supporting, and an alternative hole design to overcome the difficulties of building round holes. The research showed that 208

229 designing simple geometries like radii and holes for SLM becomes increasingly complex when compared to conventional processes Perform analysis and synthesis of the experimental results, and create a first draft of design rules appropriate for use in design practice This was the start of the second research phase aimed at creating the SLM design rules. The experimental data was interpreted into a format that was presentable within the design rules. This synthesis of experimental data was not the extensive process it was originally expected to be. The cyclic methodology used to create the geometrical data encouraged each experiment to be complete only when the results could be used to inform the design of the following experiment. This lead to all experimental results being analysed and interpreted to be used directly within the design rules. A set of design rules were created that addressed all geometrical limitations, and presented design guidelines for SLM Evaluate the design rules to establish their effectiveness and inform their further development Once the design rules were created they were evaluated. The feedback was positive because the design rules gave designers a new insight to SLM, and increased the designer s confidence in designing for SLM in the near future. The content of the design rules was found to be comprehensive and all of the design rules were relevant, however, the evaluation concluded a number of points that need consideration to advance the design rules in further developments. Commercial examples of the design rules applied used on actual parts are required, and an emphasis of the implications of all process limitations need to be shown. As the designer may not be familiar with SLM a breakdown of terminology and more process background information is also required. 209

230 9.3 Contribution to knowledge In completing the research presented in this PhD thesis, the work has created new contributions to knowledge in the fields of product design, RM and the SLM process. The contributions are described as follows: An effective method for categorisation of SLM layer performance based on grading curl. The first geometrical analysis of SLM that is comprehensive to all fundamental geometries. Previous to this study, research into SLM has identified some geometrical limitations for a small number of geometries. However, this research is the first to analyse all possible fundamental geometries within one piece of work, with focus on identifying geometrical limitations, and suggesting design guidelines. The development of a protocol for establishing the geometrical capabilities of SLM that can be repeated and applied to other machines and materials. The overall experimental methodology was configured to facilitate the delivery of a practical set of design rules. The development of the first design rules specifically for SLM to be used as a manufacturing process. Through qualitative targeted interviews, information gathered was used to validate the applicability of the design rules in practice. 9.4 Recommendations for future work The research objectives within this research were successfully completed. The following describes several recommendations for future work that could be addressed to expand and advance on the research completed within this study. 210

231 Improve the consistency and repeatability of the surface quality of SLM parts. This will require efforts to eliminate the weld spatter that occurs during the SLM process, and will reflect on the accuracy of all geometries. Investigate if the overhang size of 0.075mm is consistent to all layer thicknesses and process parameters other than those used within this study. If the overhang size is the same for all layer thicknesses then the freedom in design will increase as the layer thickness is reduced. Develop the design rules further through a series of focus groups and design exercises. The feedback may be used to refine the design rules, so that they can be used as a benchmark for individualising the rules based on the limitations of other SLM machines with different process parameters, and also different materials. Refine the experimental model that was created within this study to produce the design rules. This model included the measurement methods, the experimentation and how the design rules were created. The following points may require further refinement to establish an efficient design rules development model that can be repeated within the constraints of a commercial environment. o Reduce, or combine the number of test parts. o Implement a more accurate dimensional analysis method. Develop a method for delivering the design rules to designers. This may include investigating the possibility of delivering the rules through online software, CAD software and even RP preparation software. The aim of this would be to identify the most effective way to encourage designers to use SLM and exploit design possibilities. In summary, there appears to be three main strands of research that require further investigation to advance SLM within the RM field. These strands focus on: technological advancement that improves SLM performance; investigations into how 211

232 the limitations vary within different experimental parameters; and, case study development to assess the design rules performance for real products in a real working environment. 212

233 References 1. Badrossamay, M., & Childs, T.H. (2007). Further studies in selective laser melting of stainless and tool steel powders. International Journal of Machine Tools and Manufacture Vol. 57, pp Beall, L. G. (1998). Rotational Molding, Design, Materials, Tooling and Processing. Munich: Hanser/Gardner Publications, Inc. ISBN Bendell, A., & Disney, J. (1989). Taguchi: Applications in World Industry. London: IFS Publications. 4. Bibb, R.J, & Eggbeer, D. (2005). Rapid manufacture of custom fitting surgical guides. 6th National Conference on Rapid Design and Manufacture., (pp ). Buckinghamshire, ISBN Bibb, R.J., Eggbeer, D. and Williams, R. (2006). 'Rapid manufacture of removable partial denture frameworks', Rapid Prototyping Journal, Vol. 12, no. 2, pp ISSN , 6. Blaxter, L., Hughes, C., & Tight, M. (2003). How To Research. London: McGraw, Hill Education. 7. Boothroyd, G., Dewhurst, P., & Knight, W. (2002). Product Design for Manufacturing and Assembly: Revised and Expanded (Second Edition). (I. Marinescu, Ed.) New York, USA: Marcel Dekker, Inc. 8. Bralla, G. J. (1998). Design For Manufacturing (second edition, ISBN x). New York: McGraw-Hill. 9. Brooks, W. K., Todd, J., & Sutcliffe, C. J. (2005). Automatic lattice geometry generation for rapid manufacture. 6th National Conference on Rapid Design and Manufacture., (pp ). Buckinghamshire, ISBN Burton, M. J. (2005). Design For Rapid Manufacture; Developing an Appropriatte Knowledge Transfer Tool for Industrial Designers, PhD Thesis, Loughborough: Loughborough University. 213

234 11. Childs, T. H., & Hauser, C. (2005). Raster scan selective laser melting of the surface layer of a tool steel powder bed. Proceedings of the Institution of Mechanical Engineers, Vol. 219, Part B, pp Cromwell Group Holdings Ltd. (2008). The Pocket Ref for Designers, Toolmakers and Engineers. Fourth Edition. 13. Davidson, J. and Glock, C. (2009). Critical Success Factors for High Quality, Functional, Additive Manufactured Products, Proceedings of the Rapid Manufacturing International Conference., July 09, Loughborough University. (Digital Proceedings) 14. Deming, W. E. (1994). The New Economics For Industry, Government and Education. London: The MIT Press. 15. Dotcheva, M., Thomas, D., Millward, H. (2009). An evaluation of rapid manufactured cellular structures to enhance injection molding tool performance. International Journal of Materials Engineering and Technology, Vol. 1, no. 2, pp Groover, P. M. (2002). Fundamentals of Modern Manufacturing (second edition),.new Jersey: Wiley. ISBN Hague, R.J.M. (2006). Unlocking the design potential for Rapid Manufacture. In Hopkinson, R. Hague, & P. Dickens, Rapid Manufacturing: An industrial revolution for the digital age. West Sussex: John Wiley and Sons ltd. pp Hague, R.J.M., Mansour, S., & Saleh, N. (2004). Material and design considerations for rapid manufacturing. International Journal of Production Research, Vol. 42, no. 22, pp Hopkinson, N., Hague, R. J. M. and Dickens, P. (2006) An Industrial Revolution for the Digital Age, John Wiley, Chichester. pp Jacobs P.F (1992), Rapid Prototyping & Manufacturing. First edition, second printing. Society of Manufacturing Engineers, Dearborn, USA 21. Jacobs P.F (1996), Stereolithography and other RP&M Technologies. ASME Press, New York, USA. 214

235 22. Kruth, J.P, Vandenbroucke, B., & Van Vaerenbergh, P. (2005). Benchmarking of different sls/slm processes as rapid manufacturing technologies. International Conference of Polymers and Moulds Innovations (PMI), (p.7). Gent. 23. Kruth. J. P., Froyen. L., et al. (2004). Selective Laser Melting of iron-based powder. Journal of material processing technology. Vol. 149, no. 1-3, pp Kruth, J. P., &. Vandenbroucke, B, (2007). Selective Laser Melting of Biocompatable Metals for Rapid Manufacturing of Medical Parts. Rapid Prototyping Journal, Vol. 13, no.4, pp Mercelis, P., & Kruth, J.P. (2006). Residual stresses in selective laser sintering and selective laser melting. Rapid Prototyping Journal, Vol. 12, no. 6, pp Morgan, R., Suttcliffe, C. J., & O'Neil, W. (2004). Density analysis of direct metal laser re-melted 316L stainless steel cubic primitives. Journal of Materials Science, Vol. 39, no, 4, pp Mullen, L., Stamp, R., Brooks, W., Jones, E., & Sutcliffe, C. (2009). Selective laser melting: A regular unit cell approach for the manufacture of porous, titanium, bone in-growth constructs, suitable for orthopedic applications. Journal of Biomedical Materials Research, Vol. 89b, no. 2, pp Mumtaz, K., & Hopkinson, N. (2007). Top surface and side roughness of inconel 625 parts processed using selective laser melting. Rapid PrototypingJournal, Vol. 15, no,2, pp Nathan, S Structural Change. Design Engineering, 16 th January, p Patton, M. Q. (1990). Qualitative Evaluation and Research Methods. London: Sage Publications. 31. Pullin, J., & Offen, A. (2008). Back to the Drawing Board - Addressing the design issues of RM. Time Compression Technologies Rapid Manufacturing Conference 2008, Rapid News Publications Plc, Coventry, West Midlands, October 2008, [CD-ROM]. 32. Pullin, J. (2009, April). Stop worrying, additive layers will save us all. TCT Magazine, Vol. 17, no. 2, p

236 33. Ramin, E., & Harris, R. A. (2008). Advanced computer-aided design for bone tissue-engineering scaffolds. Proceedings of the Institute of Mechanical Engineers. Vol. 223, Part H, pp IMechE. 34. Rees, H. (1996). Understanding Product Design for Injection Molding (Vols. ISBN: ). (E.H.Immergut, Ed.) Cincinnati: Hanser/Gardner Publications Inc. 35. Rehme, O., & Emmelmann, C. (2005). Reproducibilty for properties of selective laser melting. Third International WLT-Conference in Laser Manufacturing, Munich. 36. Rehme, O., & Emmelmenn, C. (2006). Rapid manufacturing lattice structures with selective laser melting. Laser-based Micropackaging. San Jose, CA, USA. 37. Rombouts, M. (2006). Selective laser sintering / melting of iron based powders. PhD Thesis, Katholieke University Leuven, Belgium. 38. Ruffo, M. (2009, March). RM and the Medical World. (M. Excell, Ed.) MWP, Vol. 153, no. 2, pp Sapa Ltd. (1997). Aluminium Extrusion Manual. Derbyshire: J&L Annonsbyra. 40. Sercombe, T., Jones, N., Day, R., & Kop, A. (2008). Heat treatment of Ti-6AI- 7Nb components produced by selective laser melting. Rapid Prototyping Journal, Vol. 14, no, 5, pp Sower, V. E., Savoie, M. J., & Renick, S. (1999). An Introduction To Quality Management and Engineering. London: Prentice Hall, Inc. 42. Spencer, H. (1994). Aluminium Extrusions a technical design guide St Edmundsbury Press Ltd. Suffolk. The Shapemakers: Hanser/Gardner Publications Inc. 43. Thomas, D., & Bibb, R. (2007). Baseline build-style development of Selective Laser Melting high density functional parts. Eighth National Conference on Rapid Prototyping and Manufacturing (pp ). Buckinghamshire. ISBN

237 44. Thomas, D., Bibb, R. (2008a). An investigation into the geometric constraints of Selective Laser Melting for the development of design rules. Ninth National Conference on Rapid Design, Prototyping and Manufacture (pp ). Buckinghamshire. ISBN Thomas, D., Bibb, R.J. (2008b), Identifying the Geometric Constraints and Process Specific Challenges of Selective Laser Melting, Time Compression Technologies Rapid Manufacturing Conference 2008, Rapid News Publications Plc, Coventry, West Midlands, October 2008, [CD-ROM]. 46. Tolochko, N., Savich, V., Laoui, T., Froyen, L., Onofrio, G., Signorelli, E., et al. (2002). Dental Root Implants produced by the combined selective laser sintering/ melting of titanium powders. Proceedings of the institute of Mechanical Engineers; Design and Applications. 216 Part L. IMechE. 47. Tres, A. P. (1998). Designing Plastic Parts for Assembly. Hanser. 48. Tsopanos, S., Sutcliffe, C.J., Owen, I., (2005). The Manufacture of Cross-Flow Heat Exchangers by Selective Laser Melting. Enhanced, Compact and Ultra- Compact Heat Exchangers, Science and Technology, Hoboken, NJ, USA, Engineering Conferences International. 49. Voet, A., Dehaes, J., Mingneau, J., Kruth, J., & Van Vaerenbergh, J. (2005). Study of the wear behavior of conventional and rapid mould materials. Proceedings of the PMI, p Wohlers, T. (2009). Wohlers Report; State of the industry. Fort Collins: Wohlers Associates. 51. Wong, M., Tsopanos, S., Sutcliffe, C., & Owen, I. (2007). Selective laser melting of heat transfer devices. Rapid Prototyping Journal, Vol.13, no. 5, pp Wood, D. (2009). Additive Layer Manufacturing at Airbus - Reality Check or View into the Future?. The TCT Magazine, Vol. 17, no.3, pp Wright, C., Youseffi, M., Akhtar, S., Childs, T., Hauser, C., Fox, P., et al. (2006). Selective laser melting of prealloyed high alloy steel powder beds. Material Science Forum, Vol. 514, pp

238 54. Yin, R. K. (2003). Case Study Research: Design And Methods. London: Thousand Oaks. 55. Zues Precision Charts Ltd. (1995). Zeus Precision. Metric Revision. Bibliography 1. McNiff, J., & Whitehead, J. (2006). All you need to know about action research. London: SAGE. Websites 1. ASTM International Committee F42 on Additive Manufacturing Technologies Retrieved July 2, 2009 from 2. An example RM SLS part made by 3D Systems in the USA for the Oregon State University Formula SAE Team [Image] Retrieved 9, July from 3. SLS aircraft part made using EOS Polyamide material [Image] Retrieved September 9, 2007 from TRPD Rapid Product Development DT Frameworks Retrieved, July 2, 2009 from 218

239 Appendix 1: Material File Containing the Parameters Used 219

240 The following are sections of the material file that contains all process control parameters. Throughout this section the parameters described are embedded within the code, which is split up into the various scanning paths the laser takes to process each layer. For each section these paths are as follows: Boundary Solid - Hatch Solid - Skin Hatch - The outline of the layer being processed. The core/main volume of material that makes up the part. An optional layer that second melts the top of a part to improve the surface quality. The effect was not evident during parameter optimisation experiments in this study. Also, second melting layers significantly increases the build time. Fill Contour Solid - A thin area between the Hatch Solid and Boundary Solid. Support - Inner Support - Refers to all support structures minus the outline/border of the supports. The outline supports that locate around the border of the supports. [EXPOSURE] exposure_75=75µlayer - Layer Thickness. [75µLAYER] exposuresequence=1 expoboundarysolid=800 expohatchsolid=680 exposkinhatch=800 expofillcontoursolid=800 - This section refers to the time the laser/scanner takes to scan each melt point. The units are in microseconds. exposupport=600 expoinnersupport=600 pdistboundarysolid=80 pdisthatchsolid=80 pdistskinhatch=80 pdistfillcontoursolid=80 pdistsupport=80 - This section refers to the distance between each melt point. The units are in microns. pdistinnersupport=80

241 laserpowerboundarysolid=1200 laserpowerhatchsolid=2560 laserpowerskinhatch=2000 laserpowerfillcontoursolid=2560 laserpowersupport=2000 laserpowerinnersupport= The Laser power that is used for each scan path. The values are not linked to the actual Wattage that the laser produces. The numbers are set according to a digital readout = 50W, 2000 = 80W 2260 = 100W lensposboundarysolid=2650 lensposhatchsolid=2650 lensposskinhatch=2650 lensposfillcontoursolid=2650 lenspossupport= This section refers to the lens position of the laser system, which is more commonly known as laser focus. The Units are in microns. lensposinnersupport=2650 numberofexposuresboundarysolid=1 numberofexposureshatchsolid=1 numberofexposuresskinhatch=0 numberofexposuresfillcontoursolid=1 - This section refers to the time the laser/scanner takes to scan each melt point. The units are in microseconds. numberofexposuressupport=1 numberofexposuresinnersupport=1 [HATCH] - This section consists of more detailed parameters for the Hatch Solid raster scanning strategy. Only the non-default parameters are described spotsize= default_withbeamcompensation=1 default_hatch_shrinkfactorsolid=0.19 minhatchlength= Laser spot size in a single melt pool. - Prevents the laser scanning over boundaries. - Compensation value for material shrinkage. - Minimum hatch scanning length. hatchstripesrightleft=0 hatchstripesupdown=0

242 hatchsortblocksize=800 - Size of chequerboard/square that is scanned default_hatch_weaksizelimit= default_hatch_sorttype=3 default_hatch_innersupport=1 default_hatch_innersupport_xdistance=1.8 default_hatch_innersupport_ydistance=1.8 default_hatch_wallthickness= This section consists of more detailed parameters for the Support scan Paths. default_wallhatched=1 default_hatch_contour_fill_offsetsolid= default_hatch_contour_fill_countsolid=1 default_hatch_typesolid=1 default_hatch_offsetsolid=0.07 default_hatch_xdistancesolid= default_hatch_ydistancesolid= The hatch distance for x and y-axis for Hatch Solid and Fill Contour Solid default_hatch_withstripessolid=0 default_hatch_stripesizesolid=5

243 Appendix 2: Completed Questionnaires 223

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269 Appendix 3: Result Tables 249

270 Result tables from experiments 1 and 2. The first experiment was to identify the minimum surface roughness, and the surface roughness measurements were from the second experiment Orientation X/Y Build Number Build /Fail A, Up-facing B, Down-facing C, Top surface (Ra - µm) (Ra - µm) (Ra-µm) 90 Y 1 Build Y 1 Build Y 1 Build Y 1 Build Y 1 Build Y 1 Build Y 1 Build Y 1 Build Y 1 Build Y 1 Build X 1 Build X 1 Build X 1 Build X 1 Build X 1 Build X 1 Build X 1 Build X 1 Build X 1 Build X 1 Build Y 2 Fail Y 2 Fail Y 2 Fail Y 2 Fail X 2 Fail X 2 Fail X 2 Fail X 2 Fail Y 3 Fail Y 3 Fail Y 3 Fail Y 3 Fail Y 3 Fail Y 3 Fail X 3 Build X 3 Fail X 3 Fail X 3 Build 44 X 3 Build X 3 Build

271 Results from the repeated surface roughness measurements in experiment 2. The same parts were used to measure the second time around. Orientation X/Y Build Number Build /Fail A, Up-facing (Ra - µm) B, Down-facing (Ra - µm) C, Top surface (Ra-µm) 90 Y n/a Build Y n/a Build Y n/a Build Y n/a Build Y n/a Build Y n/a Build Y n/a Build Y n/a Build Y n/a Build Y n/a Build

272 Results of the first section of experiment 8 The following table describes the successful builds from the first section (5.3.1) in experiment 8. Since these results proved to be difficult to use within the design rules, and not possible to inform the design rules, they were only used to inform the need for further experiments. Hole size Geometry Minimum Tangent Overhang size peak Overhang Size Circle / Radius Accuracy (MM) - + 5mm 2 Triangle with Radius 28 no peak 3.07mm mm 2 Circle with 45 Peak mm 1.3mm mm 2 Circle with Radius mm 1.28mm Peak 5mm 2 Elliptical 28 no peak 2.04mm mm 2 Triangle with Radius 28 no peak 6.08mm mm 2 Circle with 45 Peak mm 2.59mm mm 2 Circle with Radius mm 2.59mm Peak 10mm 2 Elliptical 28 no peak 4.07mm mm 2 Triangle with Radius 40 no peak 12.94mm mm 2 Circle with 45 Peak mm 3.46mm mm 2 Circle with Radius mm 4.28mm Peak 20mm 2 Elliptical 34 no peak 7.71mm mm 2 Triangle with Radius 42 no peak 19.91mm mm 2 Circle with 45 Peak mm 3.87mm mm 2 Circle with Radius mm 6.35mm Peak 30mm 2 Elliptical 40 no peak 10.89mm 0 0

273 Appendix 4: Transcribed Interview Feedback 253

274 Question 1: Overall, were the design rules a good insight to the SLM process? (R2) The rules were a definite and useful insight to SLM, R2 could see that lots of work had gone into the rules. (R2) After reading the rules, R2 still thinks that SLM is a limiting process although it is good to see what the limitations are and what the alternative solutions could be. (R4) The approach of the design rules introduction is good, the orientations tradeoff was a good insight to the rest of the rules. (R6) The rules are a fairly good insight to SLM, it shows that there are lots of limitations. R6 said that they would control the designer rather than the designer controlling the design. (R5) The rules are a good insight to SLM, although the rules feels negative compared to what R5 has seen of the process (the respondent was impressed with previous parts) would prefer to see more images at the beginning, and more words. (R5) Needs to see examples of the rules, with the rules and example parts at the end. This would make it more positive. (R3) The rules give a good insight to the SLM process, and made the respondent aware of how complex designing for the process can be, and that the design rules are completely different to standard manufacturing processes Question 2: Were the rules clear and illustrative? (R2) All text was clear and illustrative, except for holes when talking about the hole orientation, it got confusing when parallel, perpendicular and horizontal orientations were discussed (R4) The text is illustrative; there is a good balance between words and pictures. The pictures are clear. (R4) R4 didn t understand the threads part in rule 14; it is possible to tap a round hole as a post process operation before or after pilot drilling. (R4) R4 didn t understand the holes that well, R4 thought that the peak were just for build safety and that it is recommend that they are drilled out all times. The peak in this case should be small enough so that the whole hole can be drilled out. (R1) R1 would like to see more figures in some place to set the scene with holes with figures for pilot holes and threads. (R1) The rules are clear, Except for 10b, show the parts growing and show the down facing section of the overhang, and where is would be too large and overhang past a certain tangent will require supports. R1 would like an image of this. Although R1 still finds the information good enough to use, the respondent knew what to do but couldn t grasp why. R1 also highlighted repetition. (R6) Inconsistent unclear text, tables used in one rule and not the other. The respondent described the text as Fluffy and not concise to the point. R6 was confused by the surface roughness and surface finish comments; R6 said make it more layman s use words like smoother surface. The high roughness / poor finish are confusing.

275 (R6) The respondent highlighted lots of repetition within the same rule. (R6) The rules would need to be easier to flick though and follow. Would prefer to see the rules as a quick crib sheet. (R5) The test was clear, the diagrams didn t add anything to the text, the rules should be more visual and need to look like something drawn by a graphic designer. (Respondent is taking them as if they need to have shelf appeal), contrary to this, the diagrams do make good sense. (R5) Too wordy and a bit of repetition. (R3) In places the text assumed that the reader was more aware of the process, some terminology was taken for granted. (R3) Requires more examples of part deformation, explained fine but need examples. (R3) Unsure of what you are understanding until you read the whole document. (R3) All distortions should be shown together and earlier. (R3) Respondent suggested that the author should show a best and worse case for each rule. (R3) R3 didn t understand the radii tables, is the sharp corner on the convex and concave supposed to be there? It is the nature of the geometry. Need to show the reference dimension of which the rule geometry relates to and to explain the tangent more. Question 3: Is the introduction to the technology useful? (R4) The approach to the design rules introduction is good, the trade off R4 could understand. (R1) R1 found the intro useful, the respondent would like to see more process information upfront. And also a breakdown of the terminology used. (R1) R1 would like to see an early description of the common orientation as a reference to for all rule basic orientation. The substrate diagram is half way through and would need to be at the beginning to set the scene. (R1) Respondent would expect to see a figure for each rule. (R1) More figures would be good to show a support structure properly, and to show the sag at an early stage as an implication of limitations. Need to set the scene more, so that the reader understands what they are reading about all the way through. (R1) R1 knew what sag and curl was only from reading it all, R1 would like to see this all together at the beginning. (R6) The design rules would need to see proper commercial applications of SLM in the introduction. (R6) The three choices of orientation (trade off) does not work. It contradicted the orientation diagram. (R6) R6 would like to see example and comparisons to CNC, and what the benefits of SLM actually are. (R5) Show all the distortions and why together at the beginning, to set a scene for the rules.

276 (R3) Intro is useful, although needs to show more examples of parts, some photos need to be labelled more. Also the laser processing photo could benefit from a schematic picture of the process. Question 4: Were the rules complete? Are there any design features missing? (R2) R2 thought that the rules were comprehensive to all geometries. (R4) The rules were comprehensive, they considered everything. (R1) The rules are fully comprehensive; no other geometries come to mind. (R6) The rules were very comprehensive to common geometries. (R5) The rules were complete, wanted to see more complex parts. And examples of the features (R3) It is comprehensive and complete to common geometries; however screw holes were missing, and joining parts together. Question 5: Would you feel confident to use the design rules in design practice? (R2) R2 would feel confident, to use the design rules. Respondent would use them as aid to train staff on the process on his new machine. (R4) R4 would feel confident to use the design rules, they would be a good guide. R4 would expect the people R4 subcontracts the SLM parts to for being experts, although R4 would consider the rules. The bureaus would let R4 know the limitation, and would compromise design changes to suit the process and part function. (R1) R1 would feel confident using the rules. (R1) R1 would use the rules to help decide and select a manufacturing process, the rules would help R1 identify if the process is suitable for a known application. They would help the respondent identify whether or not critical geometries would work. (R6) R6 would feel confident to use the rules (R5) R5 would feel confident to use the design rules, although not confident about doing it well. (R3) Showing applications would give R3 more confidence. Question 6: Since reading the rules, did your feelings change about using the SLM process in the near future? (R1) If R1 used SLM it would be because the respondent had identified that the SLM is the best choice.

277 (R6) Respondent s outlook on the process is negative, not sure why anyone would want to use it. The only application R6 can imagine is bespoke amorphous forms that you want in metal or when a rough finish is needed. Although the respondent thinks you can get the same results as a heavy shot blast. Process has a long way to go before it is commercially viable. (R5) R5 would also use it as a manufacturing process in the future as R5 recognises the benefits more now that before. (R3) Had no previous experience or opinion. If R3 were to use the process, they would like to try the limitations out for themselves. Question 7: What is your opinion of the SLM process since reading the design rules? (R2) No feeling towards the SLM process changed, R2 is aware that the process has limitations and knows about part accuracy. (R4) R4 thinks that the process is very limited especially on the holes. (R1) R1 now thinks that the process will need lots of post processing for functional parts. (R6) R6 can t imagine the rules having any engineering applications, and cannot see why you would choose SLM over traditional methods. There is too much to think about, like adding material etc. (R6) R6 was over whelmed by the limitations. The respondent realises that all processes have boundaries but this is very restricted. (R5) It feels like a new learning curve, as it is a new process. R5 said you would have to design a part as normal, and then design it for SLM. (R5) R5 now has a better understanding of the process and what is involved to design for it (R5) The main outlook is that there are a lot of points to consider, but R5 can use the rules to design parts after reading the document. The respondent sees it as a lesson in SLM. (R3) Doesn t think the process is too constrained, gives freedom and wouldn t prevent R3 from using hollow and freeform geometries, because that is the nature of LM. Would be good to prompt you though. Question 8: The benefits of SLM, such as hollow and freeform geometries are not described within the rules; would this prevent you from considering unique design features that cannot be made using other processes? (R2) Thinks that there is no need to show the hype of hollow, lattice and other desirable possibilities. R2 says that this is hyped enough as a sales pitch from vendors. If you were using or buying a machine, you would know about what you can do with it in the first place. Designers would consider all advanced possibilities that you cannot do other ways.

278 (R4) R4 said that the beneficial layer manufacture possibility would still be considered even though they aren t in the rules, (R1) Any would use the beneficial geometries such as hollow parts with holes, as the respondent isn t familiar with SLM and would need to research further what is possible. This would be good in the intro. R1 knows of lattice structures only through 3 rd party info. (R6) R6 didn t know that you could do hollow holes etc, and suggested that a small hole would be needed anyway to remove the powder. The respondent knew about the freeform shape capabilities as they have seen examples through 3 rd party. (R5)Wouldn t know about using specialised geometries like hollow and lattices, it would be good to show them in an example section. (R3) Would like to see examples of how all the features turn into parts, a summary of all features in one part. Question 9: Many geometrical features such as radii and holes are now more complex than other processes such as milling and casting. Are the descriptions of each rule clear or are they to complex for what are traditionally simple geometries? (R2) The complexity is not daunting; R2 would not be put off using the rules. (R2) Other comments would like to see the implications of the limitations, e.g. support witness, etc. Would like to see that you cannot remove supports from internal channels. R2 would like to see some examples. (R4) R4 would not be put off by complexity, as long as R4 knows that they work, the respondent would use them no matter how hard they were. You have to design for the process. (R1) R1 knows already that there are some things you can do in SLM and not in other processes. If R1 used SLM it would be because R1 had identified that the SLM is the best choice. Therefore the complex rules would have to be used as they are what really works. (R6) The rules are complicated, with a new complex learning curve, R6 would prefer to use other processes. Although R6 didn t understand the whole thing. (R6) The complexity of the rules would discourage R6 to use the rules. (R5) The rules would discourage R5 to use SLM because of the complexity. (R3) The complexity wont put R3 off, but realises there is a lot to learn. Feels confident to use the rules, would be a challenge as it encourages you to design in a different way. Question 10: What do you think about the length of the rules? (R2) The length of the rules is not to long at all, seems about right considering the content,

279 (R4) R4 thought the rules were too long, although they were comprehensive and had no pointless rules. (R1) R1 thought that the document was short and concise; the respondent can envisage them as a working document. (R6) The rules were not too long, but were too wordy. Question 11: Are any rules unnecessary? (R1,2,3 & 6) No pointless rules.

280 Appendix 5: Related Publications 260

281

282 AN EVALUATION OF RAPID MANUFACTURED CELLULAR STRUCTURES TO ENHANCE INJECTION MOULDING TOOL PERFORMANCE MARIANA DOTCHEVA, DANIEL THOMAS and HUW MILLWARD The National Centre for Product Design and Development Research (PDR) University of Wales Institute Cardiff (UWIC) Cardiff CF5 2YB U. K. Abstract The work presented in the paper aims to investigate the capabilities of some cellular-type structures as fundamental elements for creating new internal material configurations for injection moulding tools. The idea is to design and manufacture cellular structures, such as truss structures and honeycombs that have equivalent performance compared to parts with bulk material. These structures satisfy not only the mechanical and geometrical requirements of the injection tooling design but will also contribute to the thermal management of the moulding process. The versatility of layer-additive manufacturing allows fabrication of these complex unit cell structures which could provide design freedom for creating new micro and macro internal geometries. Selective Laser Melting (SLM) technology has been used for producing the experimental samples. Test samples have been designed with and without skin surface and then experimentally verified. Cellular structures possess valuable characteristics such as low density, high strength, and good thermal properties. Four different designs of cellular structures that are suitable for application in injection moulding tools and for rapid manufacturing have been investigated. It has been illustrated that a lattice core is capable of supporting significant structural and process loads, while also facilitating cross flow heat exchange. The capabilities of SLM have been tested, and limitations of the technology have been established. Keywords and phrases: cellular structures, selective laser melting, injection moulding tools. Communicated by M. Misbahul Amin Received December 13, Introduction Cellular metal structures have been used in different industrial applications such as light constructions, heat exchangers, reconstructive surgery, chemical, automotive and aerospace industries. They possess valuable characteristics such as low density, high strength, good energy absorption, good thermal and acoustic properties [1]-[2]. These advantageous characteristics of the cellular structures make them desirable but the difficulties of producing them limit their applications. The new layer-additive technologies such as Direct Metal Additive Manufacturing technologies, Electron Beam Melting, Direct Metal Laser Sintering, and Selective Laser Melting (SLM), enable manufacturing free-form solid parts using laser technology and layers of metal powder. SLM gives the opportunity to produce complex parts from engineering materials (stainless steel, tool steel, titanium alloy, cobalt-chromium) in a short lead time, directly from 3D CAD data. This gives the designers freedom to use the cellular materials where they are needed creating better product functionality without sacrificing their mechanical quality. The work presented in this paper explores the suitability of SLM-built cellular structures as a core material for injection moulding tools. Each year the number and complexity of plastic parts grow and the pressure to produce the injection moulding tools in shorter time and to achieve the required part quality increases. Injection moulding tools are expensive to manufacture and they are usually only used in mass production. Normally several operations/machines are required, and it takes a

283 long time, from several weeks to a few months, to produce them. Sometimes, mistakes in the design or in the mould machining can extend the production time. Any development that leads to reducing the production time of moulding tools or improving the moulding process and its efficiency has significant economic effect. The SLM process could be a key technology providing cost and performance efficient solutions for the mould making industry. Using the SLM technology as a manufacturing process for producing injection moulding tools or inserts has been shown to be beneficial in three areas: The freedom of the part s geometry that SLM provides [8] and [11]; The reduction of the production cycle [5], [7] and [9]; The ability to build conformal cooling channels along with the 3D geometry of the tools [3] and [6]. Creating a conformal cooling system is restricted by the geometry of the injection moulding tools and inserts. In tools or tool areas where it is not possible to design conformal cooling channels, the cooling process depends mainly on the thermal conductivity of the mould s material. Alternative material or design solutions are needed to help in these cases for making the moulding process more efficient and accurate. The weakness of all laser-based freeform manufacturing processes is the huge consumption of time for building large solid parts, since only considerably small quantities of material can be processed per unit time. This can be improved significantly by employing the ability of the SLM technology to build cellular structures. SLM can create specific internal cellular structures simultaneously when producing moulding tools with complex external design and conformal cooling channels. A cellular type interior will reduce the production cycle and can contribute to the thermal management of the moulding process. The cooling portion of an injection moulding time can represent up to 75 % of the total cycle time. This number shows how important any reduction of the cooling time is. Better cooling also contributes to the quality improvement of the plastic parts. With SLM built skin/core structures, material is only used where the designer requires it in order to correspond to the load. The use of a cellular material for injection moulding tools is motivated by the desire to utilise the capabilities of SLM technology for fast building of 3D complex geometry and to improve the thermal management of the injection moulding process. This paper presents the preliminary results of the design and analysis of selected SLMbuilt cellular structures that are considered suitable as core material for injection moulding tools. The experimental work tried to establish the SLM limitations that need to be taken into consideration in the design of the cellular structures to make them manufacturable. 2. Cellular Structures for Injection Moulding Tools Solid cellular materials are classified as stochastic and ordered [1]. Stochastic materials, such as solid foams can have excellent heat- and sound-insulation properties and possess the ability to absorb a lot of energy. Ordered cellular materials, such as honeycombs and truss structures have great mechanical properties. They are widely used to enable the design of light structures, for creating unidirectional fluid flows, for absorbing the energy of impacts, and to facilitate thermal transport across faces. The advantages of ordered cellular structures that make them desirable in injection moulding tools are their strength-to-mass ratio, and their high surface-to-mass ratio that defines their excellent heat transfer characteristics [2] and [4]. The ordered cellular structures provide consistent mechanical parameters of the part and can be implemented in moulding tool design. Two types of ordered cellular structures were analysed in this research: honeycombs and truss. Honeycombs are often used because of their excellent compressive strength-to-weight ratio and high bending stiffness [5]. Truss structures, or as they are also called lattice structures, appear to be mechanically competitive alternatives to prismatic honeycombs. It has been foreseen that these lattice

284 structures could be of particular interest for injection moulding design because of their fully open interior structure which could assist multifunctional applications. The lattice core of injection moulding tools could be capable to supporting significant structural loads while also facilitating cross flow heat exchange. Figure 1 illustrates parts with internal honeycomb or truss cellular structures. Figure 1. Examples of parts with square honeycomb and with truss cellular structures Cellular structures design analysis The design of cellular core structure for injectionn moulding tools has to combine the material mechanical properties and the functional requirements. This approach integrates design requirements and manufacturing capabilities. The chosen cellular structures have to correspond to three groups of requirements: 1. Geometry Cellular architecture: truss or honeycombs; Low relative density; 2. Mechanical properties Adequate strength; Maximum displacement of the pressure contact surfaces; 3. SLM manufacturability Cellular elements not inclined at less than 45º [see Section 3.1]; Diameter of the strut more than 0.1 mm [10]; To avoid hangingg flat surfaces that required support. The geometry requirements have been defined from the general characteristics of cellular materials and from the desire to introduce the cellular structure as a thermal management element during the moulding process. The mechanical requirements are derived from the moulding process and from the quality requirements for the tooling. The SLM limitations are derived from the capabilities of this layer- additive technology. Rehme and Emmelmann [10] investigated the manufacturability and scaling laws for mechanical properties of periodic lattice structures. The layer-additive manufacturing process used in this work is SLM. Eight different unit cell types are presented that can be produced by SLM technology. Following the results reported in [10] we selected two designs that demonstrate the best compression stress results and that are SLM feasible. These two designs shown in Figure 2(a) and (b), posses low relative density, they response well to compressive load and they do not have struts that are orientated at an angle less than 45º with respect to the build plane. Lattice truss structures have several of these

285 flow directions and very high specific surface area for contact with a coolant flow. As a result, they can provide efficient heat removal which is valuable for the injection moulding process. On the other hand honeycomb structures shown in Figure 2(c) and (d) although posses higher density than the truss cell structures, they have excellent mechanical characteristics under compression loading and high bending stiffness [5]. They are suitable for SLM production if they are vertically orientated as shown in Figure 2. The four structures are named as follows: (a) wall centered truss structure, (b) half-wall centered truss structure, (c) square honeycomb structure, and (d) hexagonal honeycomb structure. Figure 2. The selected cellular units Sample design The four selected designss (Figure 2) were analysed for their abilities to respond to the pressure during the moulding process. There were two recommendations to the preliminary design that has been generated for all four selected cellular structures. Firstly, the strut diameter or the honeycomb wall thickness has been defined as 2 mm [10]. The second design recommendation concerns the width of the cellular units. In order to coordinate the width of the cells with the maximumm diameter of a cylindricall surface that SLM can produce without support, it has been accepted that the cell size has to be smaller than 7 mm. The second recommendation has been introduced such that the cellular structure can accommodate additional design elements that do not require support for SLM production. Any additional support will significantly reduce the fluid flow through the cellular structure. Commercial CAD software was used to construct the samples. The cellular structures weree created by patterning a unit sell and the generated models can be combined with other models. The sample models are cubic shape featuring 3 cells in each direction as illustrated in Figure 3(a). The overall sample dimensions are mn. As the modeled cellular materials will be used as internal structures in injection moulding tools, a skin element has been introduced to each cellular sample as it is shown in Figure 3(b). In order to investigate the influence of the depth of the skin on the mechanical behavior of the skin-cellular structures several samples were modeled with skin depth of 2 mm, 3 mm, 4 mm, 5 mm and 6 mm. The cellular samples have the same overalll element cell dimensions, but different relative density. Figure 3. Models of wall centered samples without and with skin.

286 3. SLM Manufacturability Rapid Manufacturing (RM) is perceived as being able to manufacture any geometry. However, in practical application SLM is not capable of complete geometrical freedom as there are inherent process issues, which can lead to defect in certain part geometries. The part orientation and style of the support used can be manipulated to improve the quality of a particular geometrical feature. However, as a given product becomes more complex, building all of its geometrical features to an acceptable standard becomes increasingly difficult. When using SLM, downward facing surfaces and angled surfaces that need support are inherently degraded compared to vertical walls and as such typically require post-process finishing processes. Adding stock material to these surfaces to allow for this surface removal during post-process finishing was investigated. This study investigated specific categories of geometrical feature, notably various overhanging geometries to make the SLM build of cellular structures reliable and repeatable. A common cause of part distortion is due to curl, which occurs when the harsh temperature gradient melts and solidifies the powder in areas where the material is not self-supporting. In this study, various overhanging and angular overhanging geometries were built at their limits; therefore, some instances of curl were anticipated. The amount of curl that occurs during the build was quantified using a scoring system as shown in Table 1. The aim of this scoring system was to identify how much curl is safe during SLM builds. All process parameters used in this study have been benchmarked and reported previously [12]. Table 1. This table shows the scores given for curl, and the reason why that score was given On completion of all test builds, the parts were measured by comparing the SLM parts geometry against the original CAD data. To do this, all parts were measured at 10 times magnification using a shadowgraph. The shadow imagery was then imported into the original 3D CAD geometries and was scaled using a known datum dimension. The original CAD file was then compared to the SLM geometry, lines and curves were offset to measure the accuracy of the SLM parts. Previous benchmark designs built by Vandenbroucke and Kruth [13] were used to measure the accuracy across the x- and y-axes with self supporting parts; the accuracy was found to be around 0.04 mm Orientation of non-complex geometry To test SLM manufacturability of inclined features are the struts of the lattice structures and the walls of the honeycombs, experiments were designed and implemented. The first experiment aimed to establish the minimum build angle and the optimum build angle of the features. To achieve these several cuboids were created at various orientations with overall dimensions of mn using Magic s version 11.0 (Materialize, Belgium). The parts were orientated between 45 and 90 vertical at increments of 5. All these parts were built across the x- and y-axes and built successfully with no process complications as shown in Figure 4(a).

287 Figure 4. (a) SLM parts built from 45 to 90 in both the x- and y-axes. (b) Failed SLM parts built between 25 and 40 in both the x- and y-axes. Further experimentation was performed, this time concentrating on lower angles of 25 to 40. The results show that parts with orientation 40 and below all failed as demonstrated in Figure 1. Further experimentation was required at this point as the critical minimum build angle was presumed to be between 40 and 45. The results from this were non-linear; all parts built along the y-axis had failed due to an excessive level of curl. In the x-axis parts built successfully at 40 built yet those built at 41 and 42 failed, whilst those built at 43 to 45 also built successfully, as illustrated in Table 2. Table 2. Result in X and Y for orientations 40 to 45 When building with the angled feature parallel to the x-axis, the SLM parts build more successfully than when built across the y-axis. This coincides with the direction of the powder recoating system. As the parts are scanned using a raster strategy, each part melted across both x and y directions, the only difference is the powder recoating. These experiments indicated that the minimum overhang angle that is safe to build is 45. Building lower angles is not disregarded, although it is clear that as the angle decreases below 45, building the geometry successfully becomes increasingly less likely. Theoretically, each step on overhanging surfaces increases in length as the angle decreases. With a layer thickness of mm as used in this study, the step size overhanging cannot exceed mm at 45. If the step size is larger than mm, curl distortion occurs Radius This investigation identified the limitations of building concave radii and how the size of the radius relates to part failure. Firstly, tangential radii were built from 1 mm to 5 mm, only 1 mm to 3 mm built successfully. As can be seen in Figure 5, none of the completed parts was entirely successful. This is because the inaccuracy in such a small geometry makes the geometry non-identifiable or measurable as a curve. A small inaccuracy on such a small part, such as 0.35 mm on a 1 mm radius becomes significant and degrades the appearance of the whole geometry. The inaccuracy is caused by the layers overhanging too excessively for the part to build successfully. As there will be such a small number of layers at a 1 mm radius, the detail of the curve would have also distorted during the slicing operation.

288 Figure 5. Poor accuracy of small tangential radii. The tangent of the radius became the primary factor in this experiment so that the geometry is no longer tangential; the radius was no longer the primary dimension. Various different sized overhanging radius parts were built with different overhanging dimensions 5, 10 and 15 mm, as shown in Figure 6 and Table 3. Tangents from 5 to 44 were built on all parts. Parts only started to build successfully for the 5 mm overhang greater than 22. Figure 6. Illustration of the overhangs.

289 Table 3. Results showing curl and accuracy values for all parts Parts rated as N/A weree not built as lower values previously built successfully. The amount of curl that occurred whilst building all test parts was recorded; all parts with a curl rating of 3 did not complete. Alll parts with a curl score of 0, 1 and 2 built, although, parts with a curl of 2 are declared unsafe to build as the parts excessively protruded through the powder. The resultss of the shadowgraphh imagery and measuring show that the parts with a curl of 2 were not as accurate as parts with the curl values of 0 and 1. Curl values 0 had no curl, 1 consisted of very small amounts of curl that is acceptable to the build without jeopardising successful completion. When a radius is concave, it becomes increasingly difficult to manufacture using SLM when the tangent of the curve is smaller than 42 for 10 and 15 mm overhangs and 34 for the 5mm overhang. These results dependd on the part size and the relative position of the part to the substrate plate Parts support To produce the sample shown in Figure 7(a) in order to avoid the introductionn of support structure for the flat overhanging feature, the part has to be positioned upside-down (Figure 7(b)), and the build starts with the skin. The major problem with this orientation is the significant shrinkage that the first layers experience. To eliminate the inaccurate portion of the skin element of the part, additional layers have to be added and later removed mechanically by a cutting operation. This solution will make the whole production process longer and lesss efficient. If it is decided to build the skinned truss part starting with the cellular structure, a support will be required for the flat horizontal features as it is illustrated in Figure 7(c). This conventional support Figure 7(d) will reduce the flow rate of the cooling fluid and will reduce the functionality of the lattice structure as an active and important element of the thermal management of the moulding process. (a) (b) (c) Figure 7. The skinned truss part with conventional support. (d) The other answer to the overhanging feature problem that has been tested within this study is to introduce additional design elements into the part structure that will be able to support the horizontal surfaces without the need of the conventional supporting elements. One of the initial design requirements is the cell size to be smaller than 7 mm in order to accommodate the maximum diameter of a cylindrical surface that SLM can produce without support. Figure 8 shows the design of the test samples created by using wall centered truss structure with additional designed supporting features and the required support for SLM production. The design of new supporting features usess the experimental results for manufacturability of inclined features and radius presented in previous sections.

290 Figure 8. Wall-centred truss structure part with additional support elements for the flat horizontal features and the simple support system. 4. Mechanical Simulation and Results As the cellular structuress provide less support than the equivalent solid part, the mechanical behavior of the modeled structures needs to be analysed. After selecting the SLM manufacturable cellular structures, the second step for introduction of cellular structures in injection moulding applications requires testing the capabilities of the structure to satisfy injection moulding process demands. Finite Element Analysis (FEA) is used for the structure analysis. The cellular structures were analysed on compressive loading similar to the loading in injection moulding tools. The average peak pressuree that the contact surfaces experience is accepted in this study as 3.5 MPa. We applied a uniform pressure on the top surfaces of the samples. The samples are fixed in all 6 degrees of freedom. The material is Stainless Steel 316L with Modulus of Elasticity E = 193GPa, and yields strength 170 MPa. The FEA results of the four cellular structures were compared when applying the same boundary conditions. The clamped cellular samples were tested in compression using FEA simulation. For computational efficiency, due to the symmetry, only one quarter of the samples needs to be modeled. The maximum stress and the maximum displacement of the selected types of cellular structures are presented in Table 4. The FEA analysis results show that the maximum stress of the two lattice and two honeycomb structuress are similar due to the close area size of the contact surfaces. The four structures have comparable maximum displacement which is low and demonstrates appropriateness of the chosen structures to support the compressive loading. Then the skin element was introduced in the sample s design of the four types of cellular structures. Samples with 2 mm, 3 mm, 4 mm, 5 mm and 6 mm horizontal skin have been analysed and the results for the maximum stress and displacement are presented in Figure 9 with respect to the skin depth. Figure 9. Maximum stress and displacement of the four structures with skin with respect to the skin depth.

291 Although the stress and displacements levels shown in Figure 9 of all four cellular structures are higher with applied skin compared to the same parameters in Table 4 when the structures are without skin, the trend is similar. Honeycomb structures demonstrate lower stress and displacement than the two truss structures. After 4 mm, skin depth, the stress and displacement for each structure change a little. The FEA simulation with thicker skin up to 10 mm was conducted, and the trend is maintained. Taking this fact into consideration it could be concluded that there would not be a significant advantage for the mechanical characteristics of a part with cellular internal structure that has skin thicker than 4 mm for the presented design. This fact is important for SLM produced parts, because the smaller quantity of solid part s areas means shorter machining time, less material, and less energy. The FEA analysis and results have been used for establishing the trend in mechanical properties of the cellular structures, and the physical data has been obtained from mechanical testing. Table 4. Comparison of the four cellular structures Table 5. Comparison of the four cellular structures with the density. In the second design of all four cellular structures it was required that their relative density would be the same. The relative density was set to 0.3, equal to the relative density of the half wallcentered structure. These equal mass structures were analysed under the same boundary conditions as the previous set. The results from FEA simulation are presented in Table 5. Even with thinner walls in order to achieve the required relative density, the honeycomb cellular structures demonstrate better mechanical properties. However, the two truss structures also satisfy the mechanical requirements according the simulation. Which type of the cellular structure is more appropriate for injection moulding tool design depends on the design ideas and on which of the structures will demonstrate better conditions for coolant flow during the injection moulding process.

292 5 Samples Production 5. P an nd Testing To validate thhe manufacturability of thee modeled celllular structurres all describbed designs (F T Figure 2) w were produceed by SLM an nd some of thhem are preseented in Figurre 10. The lay yer depth of the t SLM p process was set s to 75μm, the laser pow wer was 100 Watts, and bboundary and d contour speeeds were 0 m/s, and hatching speeed 0.12 m/s. The Stainlesss Steel 316L powder partiicles have a statistical 0.1 s d distribution of o size from 5 μm to 45 μm m. Figurre 10. SLM prroduced partss. Samples havee been producced using the two differentt orientation oof the part as it has been ddiscussed S i Section 3.2 in 2 and illustratted in Figure 7 and Figure 8. Figure 11 shows the beetter geometryy that the d design with created c archeed support eleements providdes: the edgees are sharperr, the dimenssions are m more accuratte, and the skkin feature iss flat and horrizontal. The relative den nsity of the suupported d design is a litttle higher thaan the relativee density of thhe conventionnal part designn with horizoontal skin f feature, but the design wiith arched suppporting featuures providess less stress and a less displlacement w which is an ad dvantage for part s functioonality. Figu ure 11. Sampples with arched and conveentional skin.

293 The producedd cellular samples were tested in compression with the aim to create realistic working conditions of a typical injection moulding process. The compression tests were conducted using Lloyd Twin Column Testing Machine LS100 Plus. First all structures were subjected to 3.5 MPa pressure and then they were tested to higher loading. Figure 12( (a) illustrates the compression test results performed on the selected four cellular structures shown in Figure 2. The four structures have identical dimensionss of the strut diameter and the honeycomb wall thickness but because of the different cell design, the relative density is different (Table 4). The results demonstratee similar trends as the FEA simulated data presented in Table 4. The half wall-centered lattice structure experiences the largest displacement and the square honeycombs is the most rigid structure. Figure 12. The compression results for the four structures with (a) different density and (b) with equal relative density. When the geometrical parameters of the four structures have been changed in order to possess equal relativee density, the most rigid structured appears to be the hexagonal honeycombs as shown in Figure 12( b). Again the design of the half wall centered truss provides lesss support and strength compared to the other three cellular structures. The wall-centered lattice structure demonstrates similar strength as the square honeycombs under the applied load of 3..5 MPa. Taking this into account, it just be highlighted the advantage that the wall-centereprocess compare to the square and hexagonal honeycombs structures. Further load increase during the compression test demonstratess the superiorr strength of the two honeycombs structures, as shown in Figure 13. These two cellular structures can support significant load and display excellent load-to-weight capacity. The wall-centered trusss can experience 35-40kN load without plastic deformation but half wall-centered lattice structure is less rigid and the tested design cannot be put under more than 15-20kN compression load. Their load supporting lattice type structure can bring to the cooling phase of the injection moulding

294 ability is limited but their other functional benefits make them valuable for injection moulding tools. Figure 13. Compression load capacity results for the four cellular structures with equal density. Compression tests of the wall-centered lattice structures with 2 mm, 3 mm, 4 mm, 5 mm and 6 mm solid skin features were performed and the trend shown in FEA simulation has been confirmed to high extend. The most flexible is the sample with 2mm skin (Figure 14). The displacement of the rest samples varies little. It could be concluded that horizontal skin with 4-5 mm depth provides for the given cellular structures the required strength. Larger depth of the horizontal skin will increase the production time without contributing to the mechanical characteristics of the structures. Further testing will be done in order to analyse the mechanical behavior of non-horizontal solid skin elements. Figure 14. Compression test results for the wall-centers structures with 2 mm, 3 mm, 4 mm, 5 mm and 6 mm solid skin features. These structures are capable of supporting the pressure that is experienced during the moulding process. They create repeatable and predictable parts characteristics. At the same time the density of the cellular material can contribute to the better cooling of the moulding. Some design optimisation approaches ([14], [15]) create variable topology of the cellular structures in order to

295 put material only in these positions where it is required from the load. This approach is efficient when the cellular structure is used only to correspond to certain loading. In our case the idea is to use the cellular structures as an active element of the cooling phase of the injection moulding process it is required that the cellular structure creates uniform coolant flow. If the density of the cellular material is variable, it will create more material resistance in some areas and will slow the fluid flow. 6. Discussion The work presented in this paper demonstrates the suitability of Selective Laser Melting built cellular structures as a core material for injection moulding tools. The idea is to apply the cellular structures in a new design concept for injection moulding tooling in order to integrate new functions into the tools and to improve their performance. Four different unit cell types were evaluated in terms of their suitability for fabrication by Selective Laser Melting and their compression mechanical behavior. The capability to support significant load (approximately 20 to 50 GPa) while also could facilitating cross flow heat exchange is seen as the key characteristic of the selected cellular structures. FEA and mechanical test results demonstrate that the lattice cellular structures appear to be mechanically competitive alternatives to prismatic honeycombs structures. These characteristics make these structures desirable for the injection moulding tools design. The SLM built parts are produced by complex process that melts metal powder with controlled scanning speed and laser power and their mechanical properties are expected to differ from the mechanical properties of the same parts made from bulk material. The FEA treats and analyses the structures following the solid-state mechanical approach. The comparison of the FEA and experimental data shows that the experimental data vary from the predicted data. However, the FEA results demonstrate the same trend as the results from mechanical testing. That is why the FEA outcomes have been used only as a trend during the design process. The choice of cellular structure will rely on the data from mechanical testing until an empirical correlation model is established. This model will adjust the FEA simulated data taking into consideration the real material properties built by SLM. The FEA and mechanical tests result sets to show that the relative density affects the mechanical parameters, but the architecture of the cellular structures is the most dominant. FEA analysis indicates that for the selected cellular structures after 4 mm skin depth the stress and displacement for each structure show little change. This tendency has been confirmed by the mechanical tests. Taking this fact into consideration it could be concluded that there would not be a significant advantage mfor the mechanical characteristics of a part with presented cellular internal structure that has skin thicker than 4 mm. In order to save machining time, material and energy this critical depth of solid skin for a given cellular structure needs to be defined. The experimental results lead to some initial design guidelines that are intended to improve the manufacturability and quality of SLM manufactured cellular structures. When a simple geometry, such as a radius, is to be produced via layered manufacturing process, the geometry becomes more complex, and depends on its dimensions, as proven in the presented experiments. The general rule for building any part geometry is to identify: how much can each layer overhang without distortion jeopardising the quality of the part. If any curl distortion is introduced into the build because of inappropriate design and orientation, the final part will be distorted, undersized or the entire build could fail. The first experiment on part orientation showed that a simple, angled, flat feature could only be built as low as 45. At 45, the maximum step overhang is mm, which is the same as the layer thickness. The lattice struts have been orientated at 45º, horizontal struts and walls have been eliminated, and this SLM constraint limited the possible structures design. The concave radius experiments relate to the staircase effect. These experiments address what curve tangents can be built at without the parts failing. It is clear that the larger the tangent the better the accuracy and surface of the radius, and the cylindrical surfaces with small diameters are

296 less accurate. As small cylindrical features are going to be used as an internal support elements for horizontal or large cylindrical or spherical surfaces with tangent less than 45º, their accuracy is not that critical. These cylindrical shape supports will be part of the cellular core structure of the injection moulding tool, and their inaccuracy and large surface roughness can be in fact beneficial for the cooling of the moulding tools. The roughed surfaces of the cellular structure elements will help in the heat transferring process and will make the cooling process more efficient. Part orientation is very important for SLM process. The part functionality is the dominant factor for selection of the best position for SLM production of injection moulding tools. The quality requirements to injection moulding tools are high and if the SLM part is close to the final part s geometry, then less finishing will be needed, and more efficient the production process will be. The upside position with incorporated cylindrical support elements for the horizontal skin feature eliminates the need of extensive support system inside the cellular structure. This position of the skin/cellular core samples shows good quality and geometry of the external part features. The comparison between the upside and upside-down positions demonstrates the advantage of the upside position. When using this position, not only the quality of the skin element is better, but the SLM building process is shorter as less additional layers are required to overcome the effect of the support construction on the skin and to eliminate the unavoidable shrinkage and curl in the first layers of the part. Additionally, the upside part position is more efficient as less powder material and energy will be used and the post machining will be quicker. Taking into consideration all the advantages of the upside position of the part in SLM builds, we are planning to produce an injection moulding test tool directly on a solid plate avoiding the need of supporting system. The plate will be used for fixing and clamping the part on machines for post processing and as a datum for CNC programming and inspection. The combination of structural strength with the potential for better thermal conductivity of the cellular core structure and the ability of SLM technology to build complex 3D shapes makes the proposed in/core approach very advantageous for injection moulding tooling. References [1] L. J. Gibson and M. F. Ashby, Cellular Solids: Structure and Properties, 2nd Edition, Cambridge University Press, [2] N. C. Haydn and G. Wadley, Multifunctional periodic cellular metals, Philosoph. Trans. Roy. Soc. A 364 (2006), [3] G. N. Levy and R. Schindel, Overview of layer manufacturing technologies, opportunities, options and applications for rapid tooling, Proc. IMechE, Part B: J. Engng. Manufact. 216(12) (2002), [4] C. Z. Liu, E. Sachlos, D. A. Wahl, Z. W. Han and J. T. Czernuszka, On the manufacturability of scaffold mould using a 3D printing technology, Rapid Prototype. J. 13/3 (2007), [5] P. Mognol, M. Rivette, L. Jegou and T. Lesprier, A first approach to choose between HSM, EDM and DMLS processes in hybrid rapid tooling, Rapid Prototyp. J. 13/1 (2006), [6] A. J. Norwood, P. M. Dickens, R. C. Soar, R. Harris, G. Gibbons and H. Hancel, Analysis of cooling channels performance, Int. J. Comput. Inte. Manufact. 17(8) (2004), [7] L. E. Rännar, A. Glad and C. G. Gustafson, Efficient cooling with tool inserts manufactured by electron beam melting, Rapid Prototyp. J. 13/3 (2007),

297 [8] P. Regenfuss, L. Hartwig, S. Klötzer, R. Ebert, Th. Brabant, T. Petsch and H. Exner, Industrial freeform generation of micro tools by laser micro sintering, Rapid Prototyping J. 11/1 (2005), [9] O. Rehme and C. Emmelmann, Rapid manufacturing of lattice structures with SLM, Proceed. SPIE 6107 (2006), [10] O. Rehme and C. Emmelmann, Cellular design for laser freeform fabrication, Proceedings of the Forth International WLT-Conference on Lasers in Manufacturing 2007, Munich, June [11] E. C. Santos, M. Shiomi, K. Osakada and T. Laoui, Rapid manufacturing of metal components by laser forming, Int. J. Mach. Tools & Manufact. 46 (2006), [12] D. Thomas and R. Bibb, Baseline build-style development of selective laser melting high density functional parts, 8th National Conference on Rapid Design, Prototyping and Manufacture, Buckinghamshire Chilterns University College, CRDM Research/ Lancaster University, [13] B. Vandenbroucke and J. P. Kruth, Selective laser melting of biocompatible metals for rapid manufacturing medical parts, Rapid Prototyp. J. 13(4) (2007), [14] H. V. Wang, C. Williams and D. W. Rosen, Design synthesis of adaptive mesoscopic cellular structures with unit truss approach and particle swarm optimisation algorithm, Solid Freeform Fabrication Proceedings, Solid Freeform Fabrication Symposium, Atlanta, Texas, 2006, pp [15] D. M. Watts and R. J. Hague, Exploiting the design freedom of RM, Solid Freeform Fabrication Proceedings, Solid, Freeform Fabrication Symposium, Atlanta, Texas, August, 2006, pp

298 AN INVESTIGATION INTO THE GEOMERTIC CONSTRAINTS OF SELECTIVE LASER MELTING FOR THE DEVELOPMENT OF DESIGN RULES

299 D. Thomas and R. Bibb National Centre for Product Design and Development Research, University of Wales Institute, Cardiff, U.K. ABSTRACT To be successful in a commercial Rapid Manufacturing (RM) environment the Selective Laser Melting (SLM) process must deliver predictable and reliable results. This study aims to provide design solutions for avoiding the inherent geometric process limitations of SLM. A description of early experimentation for the creation of the design rules is presented, enabling the application of new Design for Rapid Manufacturing (DFRM) principles for the manufacture of low risk medical devices using SLM. Common geometric features have been sub-categorised into individual, less complex categories. This particular study shows the effect of building overhangs and angular overhangs; these include 90 ledges plus various chamfers and fillets. The results will be used in the development of design rules and will set constraints for specific geometries. Keywords: Design Rules, Selective Laser Melting, Rapid Manufacturing, Geometry, Limitations. 1.0 INTRODUCTION As research into Selective Laser Melting (SLM) advances, medical case studies are becoming increasingly available. Rapid Manufacturing (RM) of fully dense medical devices from medical grade materials is becoming more appealing to the commercial market for New Product Development and end use, the effect that RP/RM can have during NPD has been previously reported [1]. RM is perceived as being able to manufacture any geometry. However, in practical application SLM is not capable of complete geometrical freedom as there are inherent process issues, which can lead to defects in certain part geometries. The part orientation and style of the support used can be manipulated to improve the quality of a particular geometrical feature. However, as a given product becomes more complex, building all of its geometrical features to an acceptable standard becomes increasingly difficult. The aim of this study was to investigate specific categories of geometrical feature, notably various overhanging geometries to ascertain design criteria for successful RM. To complete this study, multiple geometries were designed in CAD software and built using SLM from 316L Stainless Steel. Only one orientation was be used throughout the experiments with exception of the first test, which specifically investigated part orientation. Only one supported surface was used in a position that did not interfere with the geometries being investigated. When using SLM, downward facing surfaces and angled surfaces that need support are inherently degraded compared to vertical walls and as such typically require post-process finishing processes.

300 Adding stock material to these surfaces to allow for this surface removal during post-process finishing was investigated. A common cause of part distortion is due to curl, which occurs when the harsh temperature gradient melts and solidifies the powder in areas where the material is not self-supporting. In this study, various overhanging and angular overhanging geometries were built at their limits; therefore, some instances of curl were anticipated. The amount of curl that occurs during the build was quantified using a scoring system as shown in Table 1. The aim of this scoring system was to identify how much curl is safe during SLM builds. All process parameters used in this study have been benchmarked and reported in previously [2]. Curl Level Explanation 0 No Curl 1 Slight visible curl, no threat to build finishing safely 2 Bad curl, build safety is at risk 3 Fail, serious curl which is guaranteed to impair the powder re-coater Table 1: This table shows the scores given for curl, and the reason why that score was given. On completion of all test builds, the parts were measured by comparing the SLM parts geometry against the original CAD data. To do this, all parts were measured at 10 times magnification using a shadowgraph. The shadow imagery was then imported into the original 3D CAD geometries and was scaled using a known datum dimension. The original CAD file was then compared to the SLM geometry, lines and curves were offset to measure the accuracy of the SLM parts. Previous benchmark designs built by Kruth et al, 2007, were used to measure the accuracy across the x and y-axes with self supporting parts; the accuracy was found to be around 0.04mm [3]. 2.0 Action Research An action research approach was taken for these experiments. Each experiment in this study was an empirical investigation, and was planned before any part was built. The order of the experiments was planned so that the results from one experiment would provide information that would inform the following experiments [4]. 3.0 INVESTIGATIONS 3.1 Orientation of non-complex geometries. For this first experiment, cuboids were built at various orientations. This experiment identified the minimum build angle and the optimum build angle. To complete the experiment several 20 x 20 x 5 mm parts were created using Magic s version 11.0 (Materialize, Belgium), the parts were orientated between 45 and 90 vertical at increments of 5. All parts were built across the x and y- axes and built successfully with no process complications as shown in Figure 1a.

301 Figure 1a: SLM parts built from 45 to 90 in both the x and y-axes. 1b: Failed SLM parts built between 25 and 40 in both the x and y-axes. Further experimentation was performed, this time concentrating on lower angles of 25 to 40. The results show that parts with orientation 40 and below all failed as shown in Figure 1b. Further experimentation was required at this point as the critical minimum build angle was presumed to be between 40 and 45. The results from this were non-linear; all parts built along the y-axis had failed due to an excessive level of curl. In the x-axis parts built successfully at 40 built yet those built at 41 and 42 failed, whilst those built at 43 to 45 also built successfully, see table 2. Axis Angle Build or Fail Axis Angle Build or Fail Y 40 Fail layer 130 X 40 Build Y 41 Fail layer 130 X 41 Fail layer 130 Y 42 Fail layer 130 X 42 Fail layer 142 Y 43 Fail layer 130 X 43 Build Y 44 Fail layer 130 X 44 Build Y 45 Fail layer 142 X 45 Build Table 2: Results in X and Y for orientations 40 to Results When building with the angled feature parallel to the x-axis, the SLM parts build more successfully than when built across the y-axis. This coincides with the direction of the powder recoating system. As the parts are scanned using a raster strategy, each part will be melted across both x and y directions, the only difference is the powder recoating. This issue was not addressed further in this study, as there were many other variables to consider. These experiments indicated that the minimum overhang angle that is safe to build is 45. Building lower angles is not disregarded, although it is clear that as the angle decreases below 45, building the geometry successfully becomes increasingly less likely. Theoretically, each step on overhanging surfaces increases in length as the angle decreases. With a layer thickness of mm as used in this study, the step size overhanging cannot exceed 0.075mm at 45. If the step size is larger than 0.075mm curl distortion occurs. 3.3 Chamfers A chamfer is a common geometrical design feature used for bevelling 90 corners and eliminating unnecessary sharp corners of designs. Chamfer are commonly found on medical devices to avoid unnecessary sharp corners that may be cause unwanted tissue damage or make parts more difficult to clean effectively.

302 The results of the previous experimentation weree applied to the investigation of what angle chamfers can be built. Various parts were built to find out which angles produced the best accuracy and flatness of the surface. No chamfers larger than 65 were built, as they would not represent a chamfer. 3.4 Results On completion of this experiment, the parts were measured and the accuracy was recorded in Table 3. The offset of the angular line and the angle of the chamfer were measured. This highlighted the amount of material that was undersize or oversize. All parts except for the 60 angle were undersize (negative material) by 0.2 mm and at 65, 0.15 mm. No parts were oversize (positive material). Figure 2 showss examples of a shadowgraph and microscope images for parts at 55 and 60. Angle Accuracy Negative / Positive Material (± mm Angle Complete Curl (+/- Degrees) Yes Yes Yes Yes Yes Table 3: Results from building chamfers at 45 to 65. Figure 2: Shadow images showing accuracy of SLM chamfers. The accuracies of geometries that have overhangingg steps were not repeatable; therefore, the z-axis must be the least accurate axis. This is why we apply tolerances to processes, so that we allow for this sort of inconsistency. Overhanging geometries cover a larger range of geometrical features rather than just simple angular overhangs. Radii and holes are obvious geometries that incur overhanging angles. The next set of experiments investigated what happens when other overhanging geometries with gradients of angles need to be made.

303 3.5 Radius Through previous experiments, limitations and optimal surface angles have been identified. As the angle of a flat, simple geometry can be orientated between 90 vertical, and 45 from the substrate plate, the staircase effect of each layer becomes increasingly evident as the angle decreases. During the following experiments, more complex overhanging and angled geometries were investigated. The intended outcome was to gain an understanding of what the limitations are for building overhanging convex and concave radii. 3.6 Concave Radius (Fillet) This investigation identified the limitations of building concave radii and how the size of the radius relates to part failure. 3.7 Results Firstly, tangential radii were built from 1 to 5 mm, only 1 to 3 mm built successfully. As can be seen in Figure 3, none of the completed parts was entirely successful. This is because the inaccuracy in such a small geometry makes the geometry non-identifiablee or measurable as a curve. A small inaccuracy on such a small part, such as 0.35 mm on a 1 mm radius becomes significant and degrades the appearance of the whole geometry. The inaccuracy is caused by the layers overhanging too excessively for the part to build successfully. As theree will be such a small number of layers at a 1 mm radius, the detail of the curve would have also distorted during the slicing operation. Figure 3: Poor accuracy of small tangential radii. The tangent of the radius became the primary factor in this experiment so that the geometry is no longer tangential; the radius was no longer the primary dimension. Various different sized overhanging radius parts were built with different overhanging dimensions, 5, 10 and 15 mm, see Figure 4. Tangents from 5 to 44 were built on alll parts. Parts only started to build successfully for the 5 mm overhang greater than 22.

304 Figure 4: illustration of the overhangs Table 4: Results showing curl and accuracy values for all parts. Parts with N/ /A were not built as lower values previously built successfully. The amount of curl that occurred whilst building all test parts was recorded, all parts with a curl rating of 3 did not complete. All parts with a curl score of 0, 1 and 2 built, although, parts with a curl of 2 are declared unsafe to build as the parts excessively protruded through the powder. The results of the shadowgraphh imagery and measuring show that the parts with a curl of 2 were not as accurate as parts with the curl values of 0 and 1. Curl values 0 had no curl, 1 consisted of very small amounts of curl that are acceptable to the build without jeopardising successful completion. When a radius is concave, it becomes increasingly difficult to manufacture using SLM when the tangent of the curve is smaller than 42 for 10 and 15 mm overhangs and 34 for the 5mm overhang. These results depend on the part size and the relative position of the part to the substrate plate. 3.8 Convex Radius A convex radius is the exact opposite of the concave radius. Where a concave radius will fillet a corner internally, a convex radius will fillet a corner externally. The following experiment investigates convex radii the same way as concave radii. 3.9 Results Only very small tangential convex radiii can be built using SLM. Only radii sizes of 1 and 2 mm built showing very little sign of curling during the build. However, the geometries were un- recognisable as a curve. The poor accuracy would not be so obvious on a larger part, but as the geometries are so small in this case, the entire geometry becomes un-recognisable, Figure 5 shows the geometry discrepancy.

305 Figure5: Shadowgraph images compared against the actual CAD data. Further experimentation with tangents from 5 to 44 were carried out; various sized overhangs were tested (5, 10 and 15 mm) as they were for concave radius, see Figure 6. Figure 6: illustration of the overhangs Table 5: Results showing curl and accuracy values for all parts. As the tangent increases, the curl decreases and accuracy of the parts improve. Parts with a curl value of 2 are less accurate, consisting of the most negative material. This demonstrates that the curl distorts the SLM geometry away from the correct CAD data. As the curl reduces to 1 and 0, the material inaccuracies are mostly positive, therefore, there was no or very little distortion. Positive material is more acceptable than negative material in a manufacturing environment, as this excessive material can be removed. The amount of positive material here is very small and not consistent, therefore, general tolerances for these surfaces should allow for unexpected positive material. With a curl score of 1, the overall tolerance of the convex radius will be ±0.4 mm, the tolerances vary from ±0.1 to ±0.4 mmm for each part built within a curl of 1 for the 5, 10 and 15 mm overhangs. To achieve this tolerance 5 mm overhanging parts must be made with a minimum tangent of 32, where 10 mm is 36 and and 15 mm is 38 and above. Figure 7 is a visual example of curl scoress of 0, 1 and 2.

306 Figure 7: Examples of the geometry discrepancy caused by curl To achieve a tighter tolerance there must be no curl occurring throughout the build, tolerances of ±0.25 mm can be achieved, although, this tolerance varies from ±0.0 mm to ±0.25 mm. The top end of the tolerance must be chosen so that the designs will not by chance build undersize. For 5 mm and 10 mm overhangs, the tangents required to achieve thesee tolerances are greater than or equal to 38. As the part size increases to 15 mm, the tangent increases to a minimum of Stock on Material Stock on material, is the name given to adding extra material to specified surfaces of a design so that material-removing post processing can be completed without making the part undersize. Otherwise known as geometric compensation and material allowance, stocking on material is common place when parts are designed to be manufactured by processes such as sand and investment casting, and conventional material subtraction processes such as milling and turning. The results and measurements can be seen in Table 6. It is clear from the height of the part before material removal that the first melted layer has melted approximately 0.8 mm deep into the support structure, as there was no other solid material underneath this layer to prevent this from happening. CAD Data Block Size 30 x 30 x 15mm 20 x 20 x 15mm 10x10x15mm Height beforee material removal 15.7mm +/ / / /- 0.2 Material Removed until flat (Bottom) 1.1mmm 1mm 0.8mm Material Removed until dense (Bottom) 1.6mmm 1mm 0.8mm Table 6: Showing the measurements of the parts before and after material removal. When the parts are first removed from the supports, it is clear thatt the downward facing base surface is not flat on the 30 x 30 x 15 mm test piece. This is because the process inherent temperature gradient has caused the first layers of the part to curl upwards at the edges, and particularly in the corners to create a dish shaped affect. In these experiments, 1.1 mm needed to be machined off to leave a flat surface and 1.6 mm was removed to reveal a fully dense surface. As this dish effect did not occur on the smaller surfaces no extra material needed to be removed to

307 make it flat. For the 20 x 20 x 15 mm sample, 1 mm was removed to make the surface dense, and 0.8 mm was removed from the 10 x 10 x 15 mm sample to make the make the surface dense. 4.0 CONCLUSIONS The results show possibilities and limitations of designing medical devices for manufacturing using the SLM process. The findings indicate some initial design guidelines that are intended to improve the manufacturability and quality of SLM manufactured parts. When a simple geometry such as a radius is to be produced via layer manufacturing process, the geometry becomes more complex, as proven in these experiments. The features examined are simplified geometries that make up most products whether freeform or dimensionally constrained. They have been broken down into less complex individual geometries for investigation. The general rule of building any part geometry is, how much can each step overhang without distortion jeopardising the quality of the part. If any curl distortion is introduced into the build because of inappropriate design and orientation, the final part will be distorted undersize or the entire build could fail. The first experiment on part orientation showed that a simple, angled, flat feature could only be built as low as 45. This means that the maximum step overhang is 0.075mm, which is the same as the layer thickness. The concave and convex radius experiments relate to the staircase effect, these experiments address how much less than a tangent 45 can be built without the parts failing. It is clear that the larger the tangent the better the accuracy and surface of the radius. As the radii geometries are gradients of different step overhang sizes, a few degrees less than 45 can be built as only a few layers will consist of overhangs larger than 0.075mm. One design problem of building radii with such large tangents is the height of the radius increases, for example a standard 15 mm overhang would usually be filleted by a 15 mm tangential radius, with the height and width the same at 15 mm. If a radius has a 15 mm overhang with a 42 overhang, the height of the radius will be mm. However, alternative geometries such as a chamfer could be used to replace the radius. These experiments have suggested some geometrical criteria for successful manufacture using SLM. The investigation of the completed parts has suggested manufacturing tolerances and also geometric compensation to part features to allow for post processing. This research will be used for further investigating design features that incorporate up, side and down facing surfaces such as blind holes, through holes and channels. The findings will be used to suggest and investigate alternative cross-section shapes that can achieve the design intent whilst being accessible to RM through SLM. Future work will lead to the development of fully illustrated design rules. References 1. Rombouts, M. (2006), PhD Thesis, Selective Laser Sintering / Melting of Iron-Based Powders, Department of metallurgy and material engineering, Katholieke University Leuven, Leuven, Belgium, pp Thomas, D., Bibb, R. (2007), Baseline Build-Style Development of Selective Laser Melting High Density Functional Parts, 8 th National Conference on Rapid Design, Prototyping and Manufacture, Buckinghamshire Chilterns University College, CRDM Research / Lancaster University 3. Vandenbroucke, B. and Kruth, J. P. (2007), "Selective laser melting of biocompatible metals for rapid manufacturing medical parts", Rapid Prototyping Journal 13(4): Yin, R. K., ed. (2003), Case study research: design and methods, Applied social research methods series, Thousand Oaks, London

308

309 Baseline build-style development of Selective Laser Melting high density functional parts Authors: Daniel Thomas and Richard Bibb National Centre for Product Design & Development Research University of Wales Institute, Cardiff Western Avenue, Cardiff, CF5 2YB, UK Abstract Many companies are exploring the implications of adopting a "Rapid Manufacturing" (RM) approach. The availability of Selective Laser Melting (SLM) has enabled the rapid manufacture of solid, fully functional metal components directly from Computer-Aided Design (CAD) data. Whilst SLM is being explored for a number of specific and advanced applications such as biomedical implants and novel lattice structures, the application of the process in the manufacture of less exotic devices and products has been neglected. The research at PDR aims to address this by exploring the development and optimisation of the SLM process in the rapid manufacture of functional medical devices. To be successful in a commercial environment the process must be predictable and reliable. Appropriate design rules or guidelines would facilitate generating designs that take full advantage of the process whilst minimising potential difficulties. This paper will describe the first steps of this research programme with the establishment of a baseline build style on which future experiments will be carried out. This paper describes a possible approach to creating a good initial build-style setup for new users of the technology before commencing onto new research. This paper will also describe future work that will address issues such as build orientation, support strategy and the development of appropriate design rules for common geometric features. Keywords: Rapid Manufacturing, Selective Laser Melting, Design Rules, Stainless Steel, New Product Development. Introduction Recent advances in Rapid Prototyping (RP) have led to the introduction of additive layer manufacturing of solid metallic parts. Not only can RP be used for building prototypes to aid the design process but it can also build end-use, full specification parts from a range of materials that are used everyday in manufacturing environments. This ability has given RP/Rapid Manufacture

310 (RM) a completely new level of freedom in designing and manufacturing end use metallic products. The Product Design Process is an important tool used during New Product Development (NPD) ensuring that the final product designed is created to the best possible standard. SLM can have a productive impact on this design process, the effect that RP/RM can have during NPD has been previously reported [1]. The aim of this project is to develop rules, which can be used and considered when designing and manufacturing medical devices to be manufactured using the SLM. Through a series of previous pilot case studies, PDR s experience with SLM has indicated that the development of design rules could significantly accelerate the application of SLM to genuine RM. Research by Sutcliffe, C. J. (2003) has stressed how important the relationship between each process parameter is when building high-density components. In-depth behavioural analysis of powder melting during the SLM process has been reported by Rombouts, M. (2006). This research helped generate a good starting point for further research [2]. Before any experimentation, it is vital that a benchmark build style was created to work from and use as a datum. In this case, the benchmark is based on a text file, which is read by the software before every layer is scanned or melted. This text file, known as a Material file contains the parameters that are used to control the build style during the SLM process. This study uses the material file to control the scanning speeds, laser power, laser focus, and scan spacing. These aren t the only parameters that need controlling, although they are the key non default, controllable parameters that affect the part quality during SLM. The relationships between each of the parameters must be understood for the SLM machine to work at its full potential. The values for the parameters and their relationships must be set carefully as there is a fine line between good and poor results. An Energy Density equation can be used to gain an understanding of the relationships between the key parameters in the SLM process, this equation has been used in this study only as a reference to the initial parameter relationship understanding. The equation cannot be used to fine tune individual SLM s, as each machine s finely tuned values are different from one another, resulting in different energy density values (see equation 1). Ep = P πr 2 2r v Equation 1: Equation for energy density, energy Density (E p ), laser power (P), scan speed (V) Creation of a Benchmark Build Style - Step 1: The first step in creating the benchmark file was to understand how a material file works. To do this, knowledge was shared between other researchers and technicians of SLM, and existing literature was revised. The findings were as follows: Private communication with Marcus Eisen of MCP Germany, highlighted that each laser-scanned line is made from a series of laser melted points at predetermined distances apart. Each point is a melt-pool that merges with the successive melt-pool at desired speeds, thus creating a scan line.

311 The wetting of the melt pool is an important factor when determining a good Material File. As recommended by the SLM manufactures, the melt-pool must be between 130µm and 180µm for an 80µm laser spot size melting Stainless Steel (316L) as used throughout this paper. If the melt-pool is larger than 180µm, the material is over exposed to the laser s power. If it is smaller than 130µm the material does not fully melt as it is under exposed to the laser s power. The melt-pool can both be controlled by the key parameters addressed in this study. A good melt pool re-melts the layer below the top freshly coated powder layer so that the layers merge well, limiting layer curl possibilities, inter layer porosity or even de-lamination. The Layer thickness used in this paper is 75µm. At PDR, the fibre laser is capable of producing a circular 80µm spot, this is a default setting of the lasers hardware. The melt pool should be an estimated 160µm in diameter because the material surrounding the 80µm laser spot also melts. Therefore, the minimum wall thickness capability is 160µm. The size of the melt pool will differ depending on the material s thermal conductivity, the power of the laser and the amount of time that the material is exposed to the laser. Stainless Steel 316L as used throughout this paper will melt a pool an estimated 50% larger than the laser point when melting at a 100W laser power. The distance between the points is referred to as Pdist (point distance) within the material file. According to Marcus Eisen of MCP Germany, the point distance should be the same as the smallest laser point size. The melt pool is 160µm for an 80µm laser spot, an overlap is required between each subsequent point melted. The overlap should be around 50-60% for the smoothest merge between each point resulting in a good single scan line, visual inspection of these scan lines can determine the quality of each line. If the distance between the points is too large then the points will not have enough overlap and the scan line will appear as dots. If the point distance is too small, the weld line will be too thick because there is too much overlap. Both of these will result in porosity issues when building parts (fig.1). Fig.1: examples of both good and poor scan weld line and point overlap. In the material file, the laser exposure time is referred to as Expo. The values for exposure are measured in microseconds (µs). The exposure determines the amount of time the scanner takes moving the laser during powder melting. If the laser is exposed to the material for too long then the melt-pool will expand larger than the recommended 130µm to 180µm. Exposure is relevant because it is calculated with the point distance resulting in the scan speed (equation 1). Equation.1: Scan speed calculation

312 The Scan Speed in meters per second (m/s) should be between 0.1 m/s and 0.2 m/s for an analogue scanner as used at PDR. Marcus Eisen of MCP advised that the laser power (Watts) should always be 100W to ensure full melting of the steel according to. This is full power on the MCP Realizer s fibre laser at PDR. When full power, together with the correct scan speeds are used, solid hatching sections will build at the maximum density possible. When building supports the Power can be dropped to 80W, this ensures that the supports remain fine, detailed and level with the powder recoating not catching on the wiper blade. Although the SLM is set to 100W, 316L will only absorb 66% of the energy, the remaining 34% will be reflected and not used, this is different for all materials. Another critical parameter is the focus of the laser, referred to as the Lens Position in the material file. The focus defines the position of the mirrors reflecting the laser onto the powder. Measured in micrometers (µm), the focus is set depending on the scan geometry. When scanning boundaries supports and boundary contours the focus should be set so that the laser point is at its smallest on the Z-axis zero position. This is because the smallest focus point results in greater accuracy in part detail as it delivers all the lasers power in the smallest area. Hatching can require an offset focus so that the full power concentration is spread over a slightly larger laser point, according to Marcus Eisen and Rombouts, M. (2006) this will prevent the powder surrounding the melt pool from being blown away by the power of the laser before it is melted, decreasing the density of the part. The amount of focus offset is a variable and differs per material and SLM hardware, if it is required, the focus can be set by melting a single layer on a powder bed, the porosity and form of the single layer can be inspected by eye, the densest, flattest layer will be the best offset. [3] The hatch distance is the distance between the centre points in consequent parallel hatch lines of solid sections. It is important that there is a sufficient overlap between the melt-pools to ensure that no gaps remain causing porosity issues. The overlap is an approximate value and is different for most materials different thermal conductivities create different melt-pool sizes. The manufacturer specifies that when melting 316L Stainless Steel with 100W of energy and an 80µm laser spot size, the hatch distances for both X and Y directions should be set at 130µm giving a 30µm overlap. When melting a new hatch line, 75% should be new powder and 25% of the line should be re-melting the previous line (Fig.2). These percentages may fluctuate from one machine to another depending on the hardware used. Fig.2: Diagram of 1 point from 2 hatch lines overlapping using 316L Stainless Steel. If the hatch distance is too large there will be no overlap, therefore, the porosity will be poor throughout the part as not all powder would have melted. If the hatch distances are too small there will be too much overlap as when the laser melts it always melts a small amount of the powder outside the melt-pool. If the material is melted with too much overlap on every scan line the

313 powder will run out for consequent lines, this will result in a poor burnt finish and high porosity according to the SLM manufacturers Creating a Benchmark Build Style - Step 2: After the parameters in the material file were set according to best advice and previous research, different file combinations were tested on the MCP Realizer SLM machine and the results were then analysed. To make these tests accurate, multiple geometries were tested. A part was designed (Fig.3) using Solid Works at PDR that included, sharp corners, radii, horizontal holes, deep slots, a 1mm thin wall section and an inclusive angled dovetail. Fig 3: Test part used for creating the benchmark build-style. After a full understanding of the material file key parameters was achieved, the following tests show how the parameters were finely tuned so that the SLM at PDR has a reliable benchmark build style to advance onto further research. Test one: Three material files were created (material file 1,2 and 3), as the scan speeds were recommended to be between 0.1m/s and 0.2m/s, each file had different speeds, 0.1m/s, 0.15m/s and 0.2 m/s. The power used was 92.5W, 100W should have been used but the laser was incapable of reading its maximum rated power. The test part from figure 2 was built on a mild steel substrate (fig.4).

314 Fig.4: The three test pieces completed on the substrate. On completion of this test each part was visually assessed. It was clear that the 1m/s scan speed material file had given the best results. The surface finish was made up of horizontal lines; this means that each hatched layer is well defined. The supports were also the best of all three parts with a better finish and density, the best skin hatch (skin hatch = upper surface) was at 0.15m/s. although they were all equally as difficult to remove from the substrate plate. One surface on each of these parts was hand ground and polished using different grades of emery paper to a high mirror finish (fig.5). The level of porosity was then clear and very similar for all three parts; since the material file knowledge had been used the porosity had clearly improved from previous builds. Fig.5: Photograph of the three parts together after polishing one surface. Left 0.1m/s, Middle 0.15m/s, Right 0.2m/s Laser Change: The original fibre laser was not capable of producing 100W. Therefore, the laser was changed and a refurbished laser was fitted. Before the experimentation to create a benchmark build style, parts were being built at 70W. A huge improvement was made to both the finish and density when the power increased 92.5W was implemented. Further improvements were expected when melting at 100W. Test 2: After the laser change, the laser power was set to melt at 100W and test 1 was repeated for a second time. As expected, the 100W full power made further improvements. Again, the scan speed of 0.1m/s had the best finish and supports; the best skin hatch was at 0.15m/s. These parts were polished to reveal that the there was no visible porosity meaning that the parts density had improved to a very high level (Fig.6). It was now clear that the slowest scan speeds produced the best results, no accurate density measuring tools were used to measure the porosity, but when compared to conventional solid material there was no difference.

315 Fig.6: Polished surfaces on 0.1 m/s scan speed, Test 3: To further develop the material file and process parameters the following test parts were designed (material file 4). The parts included non hatched wall thicknesses as smalll as 0.5mm and solid cubes at 10mm 2 x 8mm high. Using the scan speed of 0.1m/s from test 2 these parts weree built at 100W, 0.15m/s was used for the skin hatch as it provided the better top surfaces in test 1 and 2. The results were similar to those in test 2 on the solid sections, but the 0.5mm thin wall sections had failed, they were brittle, porous and at some points you could see through them (Fig.7).. Fig.7: Poor 0.5mm wall sections on substrate plate. Test 4: In tests 1, 2 and 3 the boundaries were scanned once, since the results from test three weree poor on the 0.5mm wall section an extra boundary had been added to the material file (material file 5). The extra boundary scanned the outer layer of the part twice ensuring that it was as dense as possible. This extra boundary improved the density, rigidity and surface finish of the 0..5mm wall thickness. The wall thickness was expected to be oversize because of the extra boundary. However when measured they were the same as when using a single boundary on test 3 (Fig.8).

316 Fig.8: High quality 0.5mm wall sections on substrate plate. Tests revealed that if a wall section less than an estimated 1mm thick is being built it helps to use a double boundary, although this takes slightly longer because boundaries are scanned twice, larger complex parts will be affected more noticeably than small simple geometries. With the current boundary settings a melt pool near to 130µm is being produced, therefore lines/walls smaller than 130µm cannot be produced. Test 5 - Surgical guide: The primary research aim is to optimize the SLM process for use with medical applications. It became necessary to build a surgical guide designed by the CARTIS Group (Fig.10) using procedures described by Bibb, R. and Eggbeer, D. in The geometries of this part were more generic with no flat simple surfaces to build. The material file from test 4 was used [4]. During the first attempt the supports separated from the substrate plate, because excessive power caused part curl therefore the power was dropped to 80W the for supports (Material file 6). The support teeth (Fig.9a and b) were created in a more frequent pattern with fewer sharp points. The power drop and the supports alterations ensured that the part built successfully on the second attempt. Fig.9: Left, 1 st attempt with large 2mmx-hatching and large default teeth. Right, 2 nd attempt with a small 1mmx-hatching and smaller more frequent teeth

317 The surface finish was not quite as good as when this material file was used on the test piece, because there was a more multifaceted triangulation structure on the STL file used. The text (blurred for patient confidentiality) was very clear. The surgical guide was placed against an SLA model of the patients skull created from a CT scan. This confirmed that the form was an accurate fit against the anatomy and therefore fit for purpose (Fig.10) Fig.10: Drilling guide (left) SLA model of scull section to test fit (right) Conclusions Previous in depth research had made it possible to create parameters suitable for the production of good quality, small dense parts. A test part was created to run tests that guided the research and advice taken from other SLM experts into a verified build style. This build style will be the benchmark for the development of design rules. The benchmark material file is to be used together with design of experiments to develop the process rules such as, build orientation, support strategy and the development of appropriate design rules for specific geometric features whilst considering post processes and the economics of SLM. There are further possibilities to improve the current build style. Now that a good benchmark build style has been created, the process parameters can be tweaked during further experimentation for the suitability of individual parts and purposes during further experimentation, although the benchmark created is a good build style there are great possibilities make further improvements. The design rules will be used as an aid to designing for SLM, they will be considered throughout the design process, revising the steps in the design process for new product development and advancing the use of SLM as a RM tool for medical applications. References 1. Burton. M, J. (2005). PhD Thesis, Design for Rapid Manufacture: Developing an Appropriate Knowledge Transfer Tool for Industrial Designers. Department of Design and Technology. Loughborough, Loughborough University: 239.

318 2. Pogson, S. R., Fox, P., Sutcliffe, C. J., O'Neill, W.O. (2003). "The production of copper parts using DMLR." Rapid Prototyping Journal 9(5): Rombouts, M. (2006). PhD Thesis, Selective Laser Sintering / Melting of Iron-Based Powders. Department of metallurgy and material engineering. Leuven, Katholieke University: Bibb R, Eggbeer D, Bocca A, Sugar A, Rapid design and manufacture of custom fitting stainless steel surgical guides, in 6 th National Conference on Rapid Design, Prototyping & Manufacture, June 2005, CRDM/Lancaster University, ISBN: , pp