University of Southern California Dissertations and Theses

Energy control and material deposition methods for fast fabrication with high surface quality in additive manufacturing using photo-polymerization

(USC Thesis Other)

Energy control and material deposition methods for fast fabrication with high surface quality in additive manufacturing using photo-polymerization

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ENERGY CONTROL AND MATERIAL DEPOSITION METHODS FOR FAST FABRICATION WITH HIGH SURFACE QUALITY IN ADDITIVE MANUFACTURING USING PHOTO-POLYMERIZATION by Yayue Pan A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (INDUSTRIAL AND SYSTEMS ENGINEERING) May 2014 Copyright 2014 Yayue Pan i Acknowledgements First and foremost I want to thank my advisor Dr. Yong Chen. I thank Dr. Chen very much for giving me the opportunity to join his research group and giving me all the help during these four years. I appreciate all his contributions of time, ideas, and funding support. Dr. Chen set an excellent example for me as a successful professor and a wonderful advisor. I would benefit from these four years with him for my whole life. I would also like to express my sincere gratitude to Dr. Behrokh Khoshnevis for his continuous support, constructive guidance and overwhelming encouragements both in my personal and professional time at USC. I am especially grateful for the fun and the knowledge I gained as his teaching assistant for the last three years. His willingness and patience in advising students let me learn how a distinguished researcher could also be a great educator. Next special thanks go to Dr. Wei Wu, who was willing to participate in both my qualifying and final defense. In addition, I owe thanks to another committee members, Dr. Jouni Partenen who was very helpful to my Ph.D. research work in chapter 3, and Dr. Qiang Huang who taught me the course Design of Experiments which contributed a lot during my Ph.D. research. I feel very lucky to have worked with Prof. Charlie CL. Wang from The Chinese University of Hong Kong. He often discussed interesting problems with me and encouraged me exploring further in relative research topics. Part of my work in Chapter 4 was developed under his advising. It was also wonderful time of working in 3D Systems, with my manager Dr. Mehdi Mojdeh. He gave me hands-on opportunities to study the ink-jet printing systems and learned many performance improvement approaches. He was very helpful, patient and supportive during my Ph.D. research work. I feel honored of having worked with Chuck Hull, the great inventor of ii stereolithography. I appreciate his time, guidance and support in my work in 3D Systems. I would like to thank many other colleagues: Pingyong Xu and John Fong taught me many basics about photopolymers used in additive manufacturing; John Stockwell provided a lot of supports on many technical problems; Ray Soliz helped me on many mechanical related tasks. I wish I had worked longer time with such an excellent team. An important group has not been mentioned yet, but they deserve their own part: the RTH319 members: Chi Zhou, Xuejin Zhao, Yongqiang Li, Xuan Song, Kai Xu, Dongping Deng, Pu Huang, Jing Zhang, Xiao Yuan, Matthew Petros, Mahdi Yoozbashizadeh. I specially thank Chi Zhou, who provided generous help in starting my research after I joined the lab. My research work presented in Chapter 3 and Chapter 5 was accomplished by collaborations with him. I also want to express my sincere gratitude to Xuejin Zhao, who contributed in the work presented in Chapter 4 greatly and who helped a lot in hardware setup. I also sincerely thank my other friends, Yongqiang Li, Matthew Petros, Xuan Song, Kai Xu, Dongping Deng, Jing Zhang and Xiao Yuan, for the friendship and all the support they offered. It would be the most wonderful time in my life working with those reliable, diligent, friendly and smart persons. Lastly, I would like to thank my parents and dear Peidong for all their love. Thank my parents who raised me and supported me in all my pursuits. Thank Peidong for believing me and accompanying me all the way. Thank you all. iii Table of Contents Acknowledgements ........................................................................................................................... i List of Tables ................................................................................................................................... vi List of Figures ................................................................................................................................ vii Abstract ......................................................................................................................................... xiv Chapter 1 Introduction ................................................................................................................ 1 1.1 Research Background and Motivation ................................................................................... 1 1.1.1 Additive Manufacturing .................................................................................................. 1 1.1.2 Photo-polymerization AM process [1] ............................................................................ 3 1.1.3 Mask Image Projection based Stereolithography Process ............................................... 5 1.1.4 CNC Accumulation Process ............................................................................................ 7 1.1.5 Surface Finish in Meso\Micro Photo-polymerization AM .............................................. 8 1.1.5 Build Speed in Meso\Micro Photo-polymerization AM ............................................... 12 1.2 Research Problem and Hypothesis ....................................................................................... 16 1.2.1 Research Scope, Problem and Strategy ........................................................................ 16 1.2.2 Hypotheses .................................................................................................................... 19 1.3 Research Content and Process .............................................................................................. 22 Chapter 2 Energy Control in Photo-polymerization AM Systems ........................................... 26 2.1 Principle of Photo-polymerization ....................................................................................... 26 2.2 Energy Sources for Photo-polymerization ........................................................................... 26 2.3 Energy Delivery Distinctions between CNC Accumulation and MIP-SL ........................... 28 2.4 Energy Distribution in Mask Image Projection Stereolithography ...................................... 29 2.4.1 Lateral energy distribution and lateral resolution .......................................................... 29 2.4.2 Vertical energy distribution and vertical resolution ...................................................... 31 2.5 Energy Distribution in Multi-axis CNC Accumulation Process .......................................... 31 2.6 Concluding Remarks ............................................................................................................ 33 Chapter 3 Fast Energy Delivery in CNC Accumulation through Process Optimization .......... 35 3.1 Design and Development of Multi-tool and Multi-axis CNC Accumulation System ......... 35 3.2 Curing Performance of Multiple Tools in CNC Accumulation ........................................... 38 3.3 Process Settings of the Multi-tool CNC Accumulation System ........................................... 41 3.3.1 Design of experiments ................................................................................................... 42 3.3.2 Experimental result and analysis – line curing .............................................................. 43 3.3.3 Experimental results and analysis – point curing .......................................................... 45 3.3.4 Verification experiments and results ............................................................................. 47 3.4 Motion Planning for Building Features on Curved Surfaces ............................................... 48 3.4.1 Tool motion for different curved surfaces ..................................................................... 49 3.4.2 Building process for features on a spherical surface ..................................................... 52 iv 3.5 Tool Selection for Building Features on a Curved Surface .................................................. 55 3.5.1 Selection of a straight accumulation tool ...................................................................... 55 3.5.2 Test of an angled tool for fabricating features on a vertical surface ............................. 56 3.5.3 Test of accumulation tools with mask patterns ............................................................. 57 3.6 Experimental Results and Discussions ................................................................................. 58 3.6.1 Test 1: a spiral curve on a flat surface ........................................................................... 58 3.6.2 Test 2: an inverse conic cup .......................................................................................... 59 3.6.3 Test 3: fluidic channels on a cylindrical surface ........................................................... 60 3.6.4 Test 4: rods on a spherical surface ................................................................................ 62 3.6.5 Comparison with layer-based approaches ..................................................................... 64 3.7 Concluding Remarks ............................................................................................................ 65 Chapter 4 Accurate Energy Delivery in CNC Accumulation by Tool Path Planning .............. 67 4.1 Challenges of Energy Control in CNC Accumulation ......................................................... 67 4.2 Pipeline of an Integrated CNC Accumulation System ......................................................... 69 4.3 Design and Development of an Integrated 3D Scanning Unit ............................................. 70 4.3.1 Dual-axis mirror based 3D Scanning Unit .................................................................... 70 4.3.2 Calibration of 2D Camera ............................................................................................. 75 4.3.3 Calibration of Laser Scanning Lines ............................................................................. 76 4.3.4 Computing 3D Points .................................................................................................... 77 4.3.5 3D Scanning Verification .............................................................................................. 78 4.4 Processing of Scanning Points ............................................................................................. 80 4.5 CNC Accumulation Tool Path Planning .............................................................................. 87 4.6 Experiment Results and Analysis of Energy Control in CNC Accumulation System ......... 88 4.7 Concluding Remarks ............................................................................................................ 93 Chapter 5 Accurate Energy Delivery in MIP-SL using Gray Scale Approach ......................... 95 5.1 Energy Control Principle of Mask Image Projection based Stereolithography ................... 95 5.2 Principles of Gray-Scale Approach ...................................................................................... 95 5.3 Process Planning using Gray-Scale Approach ..................................................................... 98 5.4 Verification in Top-down Projection based MIP-SL ......................................................... 100 5.5 Verification in Bottom-up Projection based MIP-uSL ....................................................... 103 5.5.1 Prototype System ......................................................................................................... 103 5.5.2 Verification in bottom-up projection configuration .................................................... 104 5.5.3 Verification in microscopic fabrications ..................................................................... 106 5.6 Concluding remarks ........................................................................................................... 108 Chapter 6 Accurate Material Deposition in MIP-SL using Meniscus Approach ................... 111 6.1 Challenges of up-facing Surface Finish in MIP-SL ........................................................... 111 6.2 Meniscus Modeling in Top-down Projection based MIP-SL ............................................. 113 6.2.1 Mathematical Modeling .............................................................................................. 113 6.2.2 Process Parameter Calibration ..................................................................................... 115 6.2.3 Meniscus Shape Analysis and Simulation ................................................................... 116 6.3 Process Planning for Building Smooth Up-facing Surfaces .............................................. 119 6.4 Optimization of Meniscus Process Parameters .................................................................. 121 6.4.1 Approaches for modifying meniscus profile ............................................................... 123 6.4.2 New Sub-slice Planning for negative approximation error ......................................... 124 v 6.4.3 New sub_meniscus point planning for positive approximation error ......................... 125 6.5 Meniscus Projection Image Calibration and Planning ....................................................... 126 6.6 Experimental Setup for Meniscus Approach in Top-down Projection System .................. 129 6.7 Tests of Meniscus Approach in Top-down Projection based MIP-SL ............................... 130 6.7.1 Tests on Straight Up-facing Surfaces .......................................................................... 130 6.7.2 Tests on Curved Up-facing Surfaces – Concave Cases .............................................. 132 6.7.3 Tests on Curved Up-facing Surfaces – Convex Cases ................................................ 133 6.8 Extension of Meniscus Approach in Bottom-up Projection based MIP-SL ...................... 135 6.8.1 Meniscus Modeling ..................................................................................................... 135 6.8.2 Process Parameter Calibration ..................................................................................... 138 6.9 Experimental Verification of Meniscus Approach in Bottom-up System ......................... 139 6.9.1 Experimental Setup ..................................................................................................... 139 6.9.2 Results and Discussion ................................................................................................ 140 6.10 Concluding Remarks ........................................................................................................ 146 Chapter 7 Fast Material Deposition in MIP-SL using Fast Recoating Method ...................... 148 7.1 Material Deposition Speed In Photo-polymerization AM ................................................. 148 7.1.1 Non-layer Photopolymerization AM ........................................................................... 148 7.1.2 Layer Photopolymerization AM .................................................................................. 149 7.2 Fast Recoating Approaches for meso-scale MIP-SL ......................................................... 150 7.2.1 Recoating in Top-down and Bottom-up Projection configurations ............................ 150 7.2.2 Models of Part Separation Forces in Recoating Process ............................................. 152 7.2.3 Fast Recoating Process Design for Meso-scale MIP-SL process ................................ 156 7.2.4 A Fast Meso-scale MIP-SL Process and Its Building Speed Analysis ....................... 160 7.3 Experimental Study of Fast Recoating Approach for Meso-scale MIP-SL process .......... 164 7.3.1 Hardware System ........................................................................................................ 164 7.3.2 Software System .......................................................................................................... 165 7.3.3 Materials ...................................................................................................................... 166 7.3.4 Experimental Results and Discussion ......................................................................... 166 7.3.5 Building Time Analysis ............................................................................................... 169 7.4 Fast Recoating Approaches for Micro-scale MIP-SL ........................................................ 171 7.4.1 Models of Part Separation Forces in Micro-scale Fabrication .................................... 171 7.4.2 Fast Recoating Process Design for Micro-scale MIP-SL process ............................... 177 7.4.3 Experimental Study of Support Settings in micro Fabrications .................................. 179 7.4.4 Experimental Study of Fast Recoating Approach for Micro-scale MIP-SL ............... 183 7.4.5 Building time statistics ................................................................................................ 190 7.5 Concluding Remarks .......................................................................................................... 191 Chapter 8 Conclusion .............................................................................................................. 194 8.1 Answering Research Questions and Contributions ............................................................ 194 8.2 Limitations and the Future Work ....................................................................................... 200 Reference ...................................................................................................................................... 201 vi List of Tables Table 3.1: Factors and their settings in the curing experiment ...................................................... 43 Table 3.2: Curing experimental result for the line curing case ...................................................... 44 Table 3.3: Factorial effects in curing experiment for the line curing case ..................................... 44 Table 3.4: Curing experiment for the point curing case ................................................................. 46 Table 3.5: Factorial effects in curing experiment for the point curing case ................................... 46 Table 4.1: Statistical analysis of scanning tests ............................................................................. 79 Table 6.1: Roughness of the built surfaces in Fig. 6.18-6.20. ..................................................... 134 Table 6.2: Residue Modeling Parameters. ................................................................................... 139 Table 6.3: Accuracy of the built geometries with different building strategies. .......................... 144 Table 6.4: Roughness of the built surfaces in Fig. 6.27-6.28. ...................................................... 145 Table 7.1: Building time statistics ................................................................................................ 170 Table 7.2: Performance of MIP-uSL systems presented in literatures and our developed system ...................................................................................................................................................... 190 Table 7.3: Performance of our new developed MIP-uSL system. ............................................... 191 vii List of Figures Figure 1.1: Adapted from [1] ........................................................................................................... 2 Figure 1.2: An illustration of two pixels in DMD chip and DMD based MIP-SL System. ............. 6 Figure 1.3: A schematic illustration of the CNC accumulation system. .......................................... 7 Figure 1.4: An illustration of the layer-based AM processes and related stair-stepping effect. ...... 9 Figure 1.5: An illustration of the MIP-SL AM process and related approaches for improving surface finish. ................................................................................................................................. 10 Figure 1.6: A schematic illustration of additive manufacturing process. ....................................... 12 Figure 1.7: An illustration of bottom-up projection configuration and separation forces.............. 14 Figure 1.8: Illustrations of built pillars. .......................................................................................... 15 Figure 1.9: Development of research work to address the research problem in this dissertation .. 18 Figure 1.10: Relationship between hypothesis and research contents ........................................... 24 Figure 1.11: Overview of the dissertation ...................................................................................... 25 Figure 2.1: Relationship between platform size, pixel resolution and energy distribution. ........... 30 Figure 2.2: An illustration of the light distribution in vertical direction. ....................................... 31 Figure 2.3: An illustration of light energy distribution in CNC accumulation .............................. 32 Figure 2.4: Research progress by this chapter and the following research work ........................... 34 Figure 3.1: A schematic illustration of the multi-tool and multi-axis CNC accumulation process. ........................................................................................................................................................ 36 Figure 3.2: The hardware setup of a multi-tool and multi-axis CNC accumulation system. ......... 37 Figure 3.3: An illustration of multi-tool stations of a CNC machine and a CNC accumulation system. ............................................................................................................................................ 38 viii Figure 3.4: The relation between curing depth and exposure time for the accumulation tools. .... 40 Figure 3.5: The relation between curing width and motion speed for the accumulation tools. ..... 40 Figure 3.6: Half-normal plots of the curing experiment for the line curing case. .......................... 45 Figure 3.7: Half-normal plots of the curing experiment for the point curing case. ........................ 47 Figure 3.8: Comparison between different tools for a 2D dragon pattern and some characters. ... 48 Figure 3.9: A schematic illustration of the 5-axis motion configuration for curved surfaces. ....... 51 Figure 3.10: An example of building features on a spherical surface. ........................................... 52 Figure 3.11: An illustration of the relationship between tool size and surface curvature. ............. 56 Figure 3.12: Test of an angled tool on repairing a feature on a vertical wall. ................................ 57 Figure 3.13: Adding textures on a vertical plane using tools with different mask patterns. .......... 58 Figure 3.14: A spiral curve pattern built by the small tool. ............................................................ 58 Figure 3.15: The building of an inverse conical cup. ..................................................................... 60 Figure 3.16: Fluid channels on a cylindrical surface. ..................................................................... 61 Figure 3.17: Rods on a spherical surface. (a) Planned tool path; (b) built physical object. ........... 63 Figure 3.18: A tilted rod fabricated by the CNC accumulation and SLA Processes. .................... 63 Figure 3.19: Roughness measurements of built surfaces by SLA and CNC accumulation processes. ........................................................................................................................................ 65 Figure 3.20: Research progress by this chapter and the following research work ......................... 66 Figure 4.1: A comparison of (a) the CNC machining and (b) the CNC accumulation processes [40] ........................................................................................................................................................ 68 Figure 4.2: A test case of building-around-inserts using the CNC accumulation processes. ......... 69 Figure 4.3: A pipeline of an integrated CNC accumulation system for automatic energy control 70 Figure 4.4: An illustration of the triangulation method and a scanning head with a fixed laser and sensor. ............................................................................................................................................. 72 ix Figure 4.5: A schematic illustration of the dual-axis mirror based scanning unit.......................... 73 Figure 4.6: 3D scanning unit with the dual-axis mirror in the 5-axis CNC accumulation system . 74 Figure 4.7: Laser calibration using the top plane (layer 10) and the bottom plane (layer 0). ........ 77 Figure 4.8: Point retrieval by calculating the closest point ............................................................ 78 Figure 4.9: Scanning results of the top surface of a mechanical part. ............................................ 79 Figure 4.10: Overview of the point processing framework. ........................................................... 82 Figure 4.11: Framework of the APSS fitting using the LDNI array points. .................................. 86 Figure 4.12: Tool path generation and CNC controller. ................................................................. 87 Figure 4.13: Multi-axis motion configuration of the CNC accumulation system. ......................... 88 Figure 4.14: Surface reconstruction for the CNC accumulation process. ...................................... 89 Figure 4.15: Fabrication process based on the scanned gear surface. ............................................ 90 Figure 4.16: Scanning and post processing of the scanned teapot surface. .................................... 92 Figure 4.17: The tool path and modified teapot. ............................................................................ 92 Figure 4.18: Research progress by this chapter and the following research work ......................... 94 Figure 5.1: Principle of the gray scale image method .................................................................... 97 Figure 5.2: Relation between the cure depth and the gray scale value for SI 500 resin................. 98 Figure 5.3: Comparison between the traditional and the gray scale image method. ...................... 99 Figure 5.4: A gray scale image generation algorithm with an illustration example. ................... 100 Figure 5.5: Hardware and software setup of the developed MIP-SL testbed. .............................. 101 Figure 5.6: A comparison of the built down-facing surfaces based on different methods. .......... 102 Figure 5.7: Hardware and software setup of the bottom-up projection MIP-uSL testbed. .......... 104 Figure 5.8: Built curved surfaces based on different methods in bottom-up projection MIP-uSL system. .......................................................................................................................................... 105 Figure 5.9: Surface measurement results of the fabricated parts shown in Fig. 5.8. .................... 106 x Figure 5.10: Cured micro-surfaces built by different methods in bottom-up projection MIP-uSL system. .......................................................................................................................................... 107 Figure 5.11: Surface measurement results of the fabricated parts shown in Fig. 5.10. ................ 108 Figure 5.12: Research progress by this chapter and the following research work ....................... 110 Figure 6.1: An illustration of the stair-stepping effect in up-facing surfaces fabricated by layered AM . .............................................................................................................................................. 111 Figure 6.2. An illustration of the meniscus approach for improving the surface finish of MIP-SL ...................................................................................................................................................... 112 Figure 6.3: Meniscus wetting to intersecting plane surfaces and fluid interface profile. ............. 114 Figure 6.4: Two built parts with different b and h values for measuring surface tension parameters. ...................................................................................................................................................... 116 Figure 6.5: Plotting results of the meniscus profile in case 1. ...................................................... 117 Figure 6.6: Plotting results of the meniscus profile in case 2. ...................................................... 118 Figure 6.7: Plotting results of the meniscus profile in case 3. ...................................................... 118 Figure 6.8: Plotting results of the meniscus profile in case 4. ...................................................... 119 Figure 6.9: A comparison between the traditional and the meniscus equilibrium methods. ....... 120 Figure 6.10: The process planning problem for the meniscus equilibrium method. .................... 121 Figure 6.11: Framework of meniscus algorithm. ........................................................................ 122 Figure 6.12: An illustration of meniscus process optimization. ................................................... 123 Figure 6.13: An illustration of influence of sub-slice planning on formed meniscus shape. ....... 124 Figure 6.14: An illustration of influence of sub-meniscus point on formed meniscus shape. ..... 126 Figure 6.15: The meniscus projection image calibration. ............................................................ 127 Figure 6.16: An illustration of the meniscus image planning. ..................................................... 129 Figure 6.17: The developed MIP-SL testbed for fabricating smooth surfaces. ........................... 129 xi Figure 6.18; A comparison of the built concave up-facing surface based on different methods. 131 Figure 6.19: A comparison of the built concave up-facing surface based on different methods. 133 Figure 6.20: A comparison of the built convex up-facing surface based on different methods. . 134 Figure 6.21: Meniscus wetting to intersecting plane surfaces and fluid interface profile. ........... 135 Figure 6.22: Meniscus profile in case 3 and case 2. ..................................................................... 136 Figure 6.23: Meniscus profile in case 4. ...................................................................................... 136 Figure 6.24: Plotting result of the meniscus profile in case 5. ..................................................... 137 Figure 6.25: Bottom-up built parts with b= 4mm and h=5: Left: SI-500; Right: E-Shell. .......... 138 Figure 6.26: The developed MIP-SL testbed for fabricating smooth surfaces. ........................... 140 Figure 6.27: A test case of concave surface. ............................................................................... 141 Figure 6.28: A test case of micro lens using SI 500. .................................................................... 142 Figure 6.29: A test case of micro lens using E-Shell. .................................................................. 143 Figure 6.30: Geometry Profile of the models in Fig. 6.27 ........................................................... 144 Figure 6.31: Surface measurement of area A2 in parts built by M1 and M2 2dippings in Fig. 6.27 ...................................................................................................................................................... 145 Figure 6.32: Surface measurement results of part A and B in Fig. 6.28 ...................................... 145 Figure 6.33: Research progress by this chapter and the following research work ....................... 147 Figure 7.1: Illustration of two non-layer photo-polymerization AM processes ........................... 148 Figure 7.2: A schematic diagram of MIP-SL system: (a) top-down projection (b) bottom-up projection ...................................................................................................................................... 151 Figure 7.3: Experimental setup for studying part separation forces in the MIP-SL process. ...... 153 Figure 7.4: Pulling-up forces of a cured layer from a PDMS film in different settings. .............. 155 Figure 7.5: The MIP-SL process based on the two-way movement design with PDMS. ............ 157 xii Figure 7.6: Pulling-up forces of a cured layer based on the two-way movement design in different settings. ......................................................................................................................................... 159 Figure 7.7: Shearing force verification test. ................................................................................. 160 Figure 7.8: The movement time in the X and Z axes in our prototyping system. ........................ 163 Figure 7.9: The building time of a layer in the two-way movement based MIP-SL process. ...... 163 Figure 7.10: The prototype hardware system ............................................................................... 165 Figure 7.11 Related software system ........................................................................................... 165 Figure 7.12: A test of a gear: (a) CAD model; (b) built objects in two liquid resins. .................. 167 Figure 7.13: A test of a head: (a) CAD model; (b-c) two views of the built object. .................... 167 Figure 7.14: A test of a statue: (a) CAD model; (b-e) two views of the built objects in two liquid resins. ............................................................................................................................................ 168 Figure 7.15: A test of teeth: (a) CAD model; (b-c) built objects in two liquid resins. ................. 168 Figure 7.16: A test of a hearing aide shell: (a) CAD model; (b-c) two views of the built object.168 Figure 7.17: A test of a brush: (a) CAD model; (b) built object. ................................................. 169 Figure 7.18: Layer building time of the test cases. ...................................................................... 170 Figure 7.19: Forces during moving up process and sliding process ............................................ 172 Figure 7.20: Pressure distribution ................................................................................................ 174 Figure 7.21: Illustrations of built pillars. ...................................................................................... 174 Figure 7.22: An illustration of the separation forces in x and z directions with different sizes ... 176 Figure 7.23: Test results for identifying the minimum moving distance and delay time. ........... 179 Figure 7.24: The flow filling time with different gap height. ..................................................... 179 Figure 7.25: Contour-based support generation method: (a) Given layers; (b) Layer analysis result. ...................................................................................................................................................... 181 xiii Figure 7.26: Experimental study of support settings: (a) Parameter values (b) Fabricated support anchors .......................................................................................................................................... 182 Figure 7.27: Setup for fast recoating approach verification. ........................................................ 183 Figure 7.28: Fabrication capability tests. (a)-(c): pillars; (d)-(f): gaps. ....................................... 184 Figure 7.29: A gear with micro features: (a) CAD model (b) built object (c) microscopic image. ...................................................................................................................................................... 185 Figure 7.30: A hearing-aid shells with micro-scale features. ....................................................... 185 Figure 7.31: A threaded pipe test case ......................................................................................... 186 Figure 7.32: A turbo fan test case ................................................................................................ 186 Figure 7.33: A scaled-up frame test case .................................................................................... 188 Figure 7.34: A test case on a pipe with overhanging micro features ........................................... 188 Figure 7.35: A test case on the surface stiction problem. ............................................................. 190 Figure 7.36: Research progress by this chapter ............................................................................ 193 Figure 8.1: A pictorial overview of the dissertation and the contributions of each chapter ........ 199 xiv Abstract More and more research prototypes and commercial systems have been developed to pursue faster build speed and better part quality in different manufacturing scale levels. Compared to traditional prototyping approaches that take days, additive manufacturing (AM) can build physical objects with any complicated structures in hours. Many industries have profited from AM and now AM is considered as an alternative to traditional manufacturing processes when the existing manufacturing methods cannot create a product practically, efficiently or affordably. However, to meet such high expectations, many challenges still remain. The primary challenge to overcome is the conflict between the build speed and the surface quality. Since the surface finish and approximation error depend on the layer thickness used in AM, the dominant approach to achieve a high surface quality is to reduce the layer thickness. However it would significantly slow down the building process. On the contrary, if a faster build speed is desired, a bigger layer thickness should be adopted and hence the surface quality would be worse. A tradeoff between the surface quality and building time is usually needed in AM processes. The thrust of this research is to contribute to the advancement of AM by addressing such a dilemma of the surface quality and build speed, to develop novel approaches to achieve both the high surface quality and the high build speed goals in photo-polymerization based AM processes. In this research work, photo-polymerization AM processes from meso-scale to micro-scale were investigated. Novel approaches and strategies were developed to deliver energy and polymerize material at the target position quickly and accurately, thus reaching high build speed and high surface quality simultaneously. xv CNC accumulation process and Mask Image Projection based Stereolithography (MIP-SL) process, are two typical photo-polymerization AM processes: the non-layer based and the layer based photo-polymerization AM, respectively. Studies of these two processes were performed to show the feasibility of applying the developed approaches in any photo-polymerization AM systems. In CNC accumulation process which is non-layer based, an accumulation tool head is made of optic fibers and the light is delivered by the optic fibers to polymerize liquid resin. Multiple accumulation tools were designed for fabrications with different resolutions and speeds. The accumulation tools are driven by multiple stages and thus are able to move or rotate along multiple axes. To control the energy power of each accumulation tool in the multi-tool and multi- axis CNC accumulation system, statistical methods were applied to optimize the manufacturing process settings and force analysis were performed, thus to achieve the optimal build speed and part quality. In addition, in order to successfully fulfill a manufacturing task, another challenge related to energy control has to be overcome. In CNC accumulation, it is critical to accurately deliver the energy to the exact position with a good normal direction. Otherwise, the fabricated result may not be able to reach the accepted geometry accuracy, or the material may attach on the tool tip instead of the base surface, which would cause failures to the build task. To address this problem, a novel dual-axis 3D scanning unit was developed and integrated. A point processing method based on the Algebraic Point Set Surface (APSS) fitting and Layered Depth-normal Image (LDNI) representation was developed for converting the scanning points into a 3D surface model. Based on the constructed surface model, a multi-axis tool path could be generated for building tasks of repairing or remanufacturing of any geometry using the developed system. With those developed energy control approaches, the developed multi-tool and multi-axis CNC accumulation system is able to fabricate the part accurately and rapidly. xvi In MIP-SL system which is a layer based photo-polymerization AM system, a DMD chip is usually used to dynamically generate a mask image. The mask image is used to define the light pattern hence control the energy distribution. As a layer based photo-polymerization AM process, MIP-SL is much more complicated on material deposition than CNC accumulation process. Related challenges include ultra-thin layer recoating, liquid surface profile, and so on. Accordingly, methods both on energy control and material deposition aspects were developed to achieve the two syngenetic goals of fast build speed and high surface quality. In order to address the notorious stair case effect in AM which damages the surface quality significantly, on energy control aspect, a gray scale image approach was developed by modeling and controlling the cure depth with different light intensities. On material deposition aspect, a meniscus equilibrium approach was developed by modeling and controlling the liquid profile with different liquid/gas/solid equilibrium conditions. With the two approaches, extreme high surface finish could be achieved without sacrificing build speed or even with higher build speed. In addition, a novel fast material recoating approach was proposed and implemented in MIP-SL system. With the developed material recoating approach, our system is able to realize an order of magnitude faster build speed comparing to the existing additive manufacturing systems in the market. Different building scales share some common physics in the fluid flow and polymerization process but also differ in many fields, such as separation force effects and optic system requirements. When it comes to micro-scale fabrication, the optical systems and machine designs were modified, and the corresponding optimal process parameter settings were investigated to build micro features in both CNC accumulation system and MIP-SL system. Besides, when the scale comes down to micro, the fast recoating technique used in meso-scale cannot be copied directly. Different recoating mechanisms should be applied due to the change in separation forces. xvii The meniscus equilibrium approach developed in meso-scale systems was also investigated and tuned to overcome the stair-case effect in micro-scale fabrication. Testbeds were developed and experiments were performed to verify the effectiveness and efficiency of the proposed approaches in the two photo-polymerization AM systems. The test results demonstrated that the proposed approach could achieve an order of magnitude faster build speed than any other AM systems while the surface finish could be improved by ~80%. Unlike other technologies which can only accomplish one goal at the cost of the other, the approaches proposed in this research can improve the building speed and surface finish to the largest extent at the same time in both meso-scale and micro-scale fabrications. The build speed and surface finish are not conflicting goals any more as they were in the conventional AM systems. This work would be meaningful in advancing AM from "rapid prototyping" to truly "rapid" manufacturing technology for not only prototypes but also end-use products. 1 Chapter 1 Introduction 1.1 Research Background and Motivation 1.1.1 Additive Manufacturing Additive manufacturing (AM) is a technology that can fabricate parts directly from computer aided model by accumulating material together, usually in a layer by layer manner. AM involves a number of steps that move from the virtual CAD description to the physical resultant part. As shown in the following figure, most AM progresses involve, to some degree at least, the following eight steps [1]: (1) Conceptualization and CAD (2) Conversion to STL ( triangular representation of the CAD model) (3) Transfer and manipulation of the STL file to the AM machine (4) Machine setup (5) Build (6) Part removal and cleanup (7) Post-processing of the part (8) Application According to the raw material used, it can be classified into four main categories [1]: liquid polymer such as stereolithography, discrete particles like Selective Laser Sintering and 3D printing, molten material such as fused deposition molding, and laminated sheets such as Solido PLT (KIRA). Compared with conventional manufacturing processes, such as CNC machining and molding, AM processes have their unique advantages in terms of material\energy efficiency, simple operating style, more design flexibility, and so on. 2 Figure 1.1: Adapted from [1] Nowadays AM processes are adopted in more and more areas not only for rapid prototyping, but also for parts production also. Intensive research has been conducted in adopting AM in fabricating plastic or metal parts for various applications [2-8], including medical components, architectures, and so on. Despite these design and manufacturing advantages, the adoption of AM technologies as a means for fabricating end-use components has historically been dampened by the technologies inherent drawbacks [9]: (1) Size limitations [10, 11]. AM processes often use liquid resin, melted polymer\metal, or powder, to build 3D parts by stacking 2D layers up one by one. Factors including the size of material particle or fused deposition, the material mechanical property and the unit amount of energy input, render AM unable to produce large sized objects with high resolution. Furthermore, large-sized objects also often are impractical due to the extended amount of time need to complete the build process. There need to be a tradeoff between building size, part resolution and building time. 3 (2) Imperfections. Parts produced using layer-based AM processes often possess a rough and ribbed surface finish [12-19]. This appearance is due to the inherent mechanism of layer- based AM processes, which builds a physical part by stacking sliced 2D layers together. Such manufacturing principle brings the end product an unavoidable unfinished look. (3) Cost[20]. AM equipment is considered an expensive investment. Entry level 3D printers average approximately $5,000 and can go as high as $50,000 for higher-end models. 1.1.2 Photo-polymerization AM process [1] Among various types of AM technologies, we focus on photo-polymerization AM processes that use light to solidify liquid resin one layer at a time. To better understand photo- polymerization AM, it is necessary to learn the generals of photo-polymerization process. Photopolymerization processes make use of liquid, radiation curable resins, or photopolymers as their primary materials. Photopolymers are materials that change from a watery liquid state to a strong, plastic-like solid almost instantaneously when light of the right wavelength shines on them. Photopolymers are usually composed of several types of ingredients: photoinitiators, reactive diluents, flexibilizers, stabilizers, and liquid monomers. Broadly speaking, when light radiation impinges on the resin, the photoinitiators undergo a chemical transformation and become “reactive” with the liquid monomers. A “reactive” photoinitiator reacts with a monomer molecule to start a polymer chain. Subsequent reactions occur to build polymer chains and then to cross-link – creation of strong covalent bonds between polymer chains. Polymerization is the term used to describe the process of linking small molecules (monomers) into larger molecules (polymers) composed of many monomer units [1]. Photopolymers were developed in the late 1960s and became widely applied in commercial areas like coating, printing and dentistry. In these areas, coatings are solidified by radiation 4 without needs for patterning either the radiation or the material. This changed with the invention of stereolithography system, which was introduced by Chuck Hull in the mid-1980s. By curing the photosensitive material layer by layer and stacking them together, he found that a solid 3D part could be fabricated. This is the beginning of photopolymerization AM technology, which is also called as stereolithography (SL). Since then, various photopolymerization AM processes and technologies have been developed. In those systems, UV and visible light radiation are most commonly used. A typical stereolithography system includes the following subsystems: light source and optics system, the platform system, the control system, the vat system, and the recoating system[1]. Typically, recoating is done using a shallow dip and recoater blade sweeping. The recoating process used in most systems can be described as follows [21, 22]: (1)After a layer has been cured the platform dips down by a layer thickness. (2)The recoater blade slides over the whole build depositing a new layer of resin and smoothing the surface of the vat. For resin with low viscosity, a deep-dip recoating approach has also been developed to replace the surface sweeping approach. After the up and down movements in the Z axis, a sufficient waiting time is required for the liquid resin to settle down into a flat surface. The optics system includes a laser, focusing and adjustment optics, and two galvanometers that scan the laser beam across the surface of the vat. Modern SL machines have solid-state lasers that have more stable characteristics than their predecessors, various gas lasers. SL machines from 3D Systems have Nd-YVO4 lasers that output radiation at about 1,062 nm wavelength (near infrared). With the development of technologies, in addition to the laser based stereolithography, more photopolymerizaiton AM processes were proposed and developed. Based on the 5 consolidating manner, there are four configurations for photopolymerization processes in a vat that has seen some research interest: (1) Two-photon approaches that are essentially high resolution point-by-point approaches [23-29]; (2) Vector scan, or point-wise, approaches typical of commercial SLA machines [30, 31]; (3) Mask projection, or layer-wise approaches, that irradiate entire layers at one time; (4) CNC accumulation that use fiber optics to transmit UV light. Note that in the vector scan and two-photon approaches, scanning laser beams are needed, while the mask projection approach utilizes a large radiation beam that is patterned by another device, which nowadays is Digital Micromirror Device TM (DMD) at large. In this research, Mask Image Projection Stereolithography (MIP-SL) and CNC Accumulation Process will be studied as two typical examples of layer based and non-layer based photopolymerization AM processes, respectively. 1.1.3 Mask Image Projection based Stereolithography Process In mask-image-projection-based Stereolithography process (MIP-SL), projection light is patterned by a mask image to selectively cure the liquid photopolymer resin. Since an entire 2D layer of the part is cured at one shot of projection using dynamic mask images, unlike the point or line curing in stereolithography using a laser beam, the main advantage of mask projection method is speed. Dynamic masks can be realized by liquid crystal display (LCD), or digital micromirror device (DMD). A DMD is a microelectromechanical system (MEMS) device that enables one to simultaneously control ~1 million small mirrors that turn on or off a pixel at over 5 KHz. Mask image projection based stereolithography (MIP-SL) was first introduced in the early 1990s by researchers who wanted to develop special SL machines to fabricate micro-scale parts[2, 6 32-36]. Some earlier systems utilized LCD displays as their dynamic mask [37], while the most systems that developed in recent years are all used DMDs as their dynamic masks [32, 34, 38, 39] due to many good properties of DMDs like high contrast ratio and high light transmission ratio. Using DMD technology, a light projection device can project a dynamically defined mask image onto a resin surface to selectively cure liquid resin into layers of the object. Consequently, the related AM process, Mask-Image-Projection-based Stereolithography (MIP-SL), can be much faster than the laser-based SLA process by simultaneously forming the shape of a whole layer. An illustration of the MIP-SL process is shown in Figure 1.2. Similar to conventional SLA, the MIP-SL process starts with the CAD model of the part, which is then sliced at various heights. Each resulting slice cross section is stored as bitmaps to be displayed on the dynamic mask. Light radiation reflects off of the “on” micro -mirrors and is imaged onto the resin surface to cure a layer. An automated XYZ stage is used to translate the vat of resin in three dimensions. Controlling the area to be exposed using digital micro-mirror devices (DMD) obviates the need for any scanning at all, thus increasing throughput and reducing the number of moving parts. (a) DMD chip (b) DMD based MIP-SL System Figure 1.2: An illustration of two pixels in DMD chip and DMD based MIP-SL System. 7 1.1.4 CNC Accumulation Process Recently, a CNC accumulation process which is also a SL-related process was proposed and developed by our research group[40]. CNC accumulation process is a photopolymerization AM process based on UV-curable liquid resin such as the ones used in the SLA process. Different from the SLA process and MIP-SL process, the accumulation tool in the process is merged under the liquid resin, thus it is able to cure resin in various directions. CNC accumulation system utilizes ball lenses to focus the UV light and transmit the focused light beam through a quartz fiber optic light guide. The end side of the light guide is merged in a resin tank as the curing tool. A 5-axis motion system is used to control the rotation and movement of curing tool and the tank. The configuration is shown as Figure 1.3. Figure 1.3: A schematic illustration of the CNC accumulation system. Recently the development of non-layer-based methods was identified as an important research direction for future AM development[41]. A few of attempts have been made in other AM systems. Examples include laser aided Direct Metal Deposition [42-45], and Directed light Fabrication (DLF) [46], which deposit molten metal powders in an arbitrary direction with the assistance of 5 axis movements. Those techniques are serving for fabrications using metal powder Computer control O X Y Z A B LED with power supply Focusing lens Fiber optic cable Teflon film + - Resin Tool Attaching force on built part Attaching force on tool tip Newly cured resin portion Built part 5-axis motion 8 and laser beam, so that the AM processes can be used to repair and rebuild worn or damaged components and to apply wear-and corrosion-resistant coatings. Hence a multi-direction accumulation process of liquid polymer is desired for applications in plastic parts fabrication. In SLA or MIP-SL system, it is challenging to manipulate the accumulation of solidified layer in multiple directions, because the liquid surface is always flat in gravity field. Because of the multi-axis configuration and the merge-in-liquid accumulation tool, CNC accumulation process demonstrates a promising capability in building-around-inserts and part repairing. MIP-SL system and CNC accumulation system are layer-based and non-layer based AM processes respectively, and they both have promising potential to be high throughput and good surface finish photo-polymerization AM technologies. Based on their different light energy delivery principle, MIP-SL system has strength on direct digital manufacturing and CNC accumulation is beneficial for remanufacturing. Therefore, we investigated the unique challenges in MIP-SL and CNC Accumulation process respectively, and then explored the energy input control and material deposition mechanism in meso\micro scale MIP-SL and CNC accumulation system to address those challenges. 1.1.5 Surface Finish in Meso\Micro Photo-polymerization AM -Layer-based Photo-polymerization AM: MIP-SL Compared with other layer-based AM technologies, MIP-SL system has already improved part surface quality, but it still suffers from the notorious stair stepping effect problem [47-49]. As shown in Figure 1.4, a given three-dimensional (3D) model is first sliced into a set of two- dimensional (2D) layers. By stacking the 2D layers together, a physical part can be fabricated in an AM process to approximate the original computer-aided design (CAD) model. Due to the use 9 of 2D layers, the fabricated part surfaces especially the ones whose normal is close to the building direction (Z axis) may have big approximation errors. Such poor surface quality limits the use of AM in applications that require smooth surfaces, e.g. the fabrication of 3D micro channels in micro fluidic systems or various assembly features in high-precision products. Since the approximation error depends on the layer thickness used in the slicing of a 3D model, a dominant approach for addressing the stair-stepping effect in AM is to reduce the layer thickness. For example, the layer thickness typically used in a Stereolithography Apparatus (SLA) system is 0.1 mm while the layer thickness used in the latest developed inkjet-based systems (e.g. the ones from Objet, Geometries Ltd) can be as small as 0.012mm. While the thinner layers in comparison always have less stair-stepping problems, the use of such ultra-thin layer thickness also significantly slows down the building process. In addition to the big sacrifice in building speed, it is also very difficult for AM processes such as SLA to significantly reduce its layer thickness since the spreading of liquid resin into uniform ultra-thin layers is well-known challenging. Figure 1.4: An illustration of the layer-based AM processes and related stair-stepping effect. In the past decades, researchers extensively studied the MIP-SL processes and various kinds of approaches were proposed to improve the performance. Techniques including controlled cure 10 depth[35, 50, 51], post-processing[13, 17, 52], part orientation[49, 53-56], and meniscus methods[57] were introduced to improve the surface finish of AM fabricated parts. Yet they are only effective for a certain type of surfaces and the build time is elongated significantly with those approaches. In this research, we present an alternative approach for achieving smooth surfaces in the photo-polymerization AM processes. The AM process considered in this research is the MIP-SL system, but it is also applicable to other photopolymerization AM systems. An illustration of the MIP-SL process is shown in Figure 1.5 a. Figure 1.5: An illustration of the MIP-SL AM process and related approaches for improving surface finish. Based on the Z axis and its surface normal N, all the surfaces in a 3D model can be classified into: (1) Vertical surfaces (N • Z = 0), (2) down -facing surfaces (N • Z < 0), and (3) up-facing surfaces (N • Z > 0). There is no need to specially consider the vertical surfaces since they generally do not contribute to the stair-stepping effect. For the other two types of surfaces, our approaches for improving their surface finishes are shown in Figure 1.5 b. b b h b Liquid resin Liquid resin Meniscus equilibrium Up-facing surfaces Down-facing surfaces Liquid resin Gray scale images Curing depth 1 Curing depth 2 Curing depth 3 Curing depth 4 Input CAD model Built physical model N D N U Z (a) (b) 11 (2) In this research, a gray scale image approach has been developed for the fabrication of smooth down-facing surfaces. As shown in Figure 1.5 b-(top), the key idea of the approach is to use a fine Z resolution to sample a given down-facing surface. Hence the curing depth at each pixel can be represented by a cure depth that has a higher resolution than the layer thickness. Accordingly a gray scale value at a pixel can be set to provide the desired energy input for the related curing depth. Consequently a gray scale image will be computed for all the down-facing surfaces in the MIP-SL AM process. (3) A meniscus equilibrium approach has been developed for the fabrication of smooth up- facing surfaces. As shown in Figure 1.5 b-(bottom), the key idea of the approach is to closely match the fluid interfaces at the corners of intersecting planes to the related curved surfaces in the input geometry. Hence, liquid meniscus will be formed at the corners when the cured layers emerge out from the liquid. The shapes of the meniscuses depend on the contact horizontal and vertical surfaces, and can be modeled by considering the effects of capillarity, suction by gravity, multilayer adsorption and the boundary conditions [58-62]. Accordingly a process optimization problem can be formulated in order to match the meniscuses to the given curved surfaces. Compared to existing solutions like the use of ultra-thin layers or post-processing that significantly elongates the building time, the developed techniques enable a much bigger layer thickness to be used while achieving similar surface finish requirement. Hence significantly reduced building time could be expected in the MIP-SL AM process. -Non-layer based Photo-polymerization AM: CNC Accumulation The intrinsic cause of stair-case effect in layer based AM is that, a physical part is fabricated by stacking the sliced 2D layers together in a uni-direction to approximate the given CAD model. For such layer-based AM processes, only the linear motions in the X, Y and Z axes 12 are required. If the material could be accumulated in arbitrary direction, the stair-case effect could be eliminated totally. An example of a tilted rod (AB) is shown in Figure 1.6. Figure 1.6: A schematic illustration of additive manufacturing process. As mentioned before, CNC Accumulation process is a non-layer based AM process. The accumulation tool is able to move along or rotate around multiple axes. Hence the building direction can be set in a way that gives perfect smooth surface with no stair-case effect. In order to do that, tool path together with the optimal building direction needs to be studied in CNC Accumulation process. 1.1.5 Build Speed in Meso\Micro Photo-polymerization AM -Layer-based Photo-polymerization AM: MIP-SL: Unlike traditional prototyping approaches that take days, AM-based rapid prototyping can build physical objects in hours. Due to such time and cost benefits, AM processes have been widely adopted in the product development process for building prototypes of a design. Although the speed of AM systems has significantly increased over the years, the building process of a moderate sized 3D model is typically measured in hours. In a recent NSF workshop on 13 developing the roadmap for AM[41], the development of AM machines with higher throughput was identified to be critical for future rapid manufacturing requirements. Future high-speed AM systems require new approaches, evolving from point-processing or line-processing methods such as a laser or an extruding nozzle, to area-processing or volume-processing methods. Consequently, the related AM process, Mask-Image-Projection-based Stereolithography (MIP-SL), can be much faster than the laser based SLA process by simultaneously forming the shape of a whole layer. In the MIP-SL process, the building time of each layer consists of spreading liquid resin into a uniform thin layer and curing the formed liquid layer into a solid layer. Compared to a laser beam that is used in the SLA process, the DMD used in the MIP-SLA process can dramatically decrease the curing time of a layer. Hence, the bottleneck for achieving a fast building speed is the spreading of liquid resin into uniform thin layers. Most of the developed systems are based on the top-down projection as shown in Figure 1.5. Suppose d LT is the layer thickness. After a previous layer has been cured, the platform in such a system usually moves down a certain distance d and then up by d-d LT in order to spread liquid resin into a uniform thin layer. In addition to the Z movement, a recoating process is usually required to sweep through the platform such that the top surface can be flattened. For resin with low viscosity, a deep-dip recoating approach has also been developed to replace the surface sweeping approach. After the up and down movements in the Z axis, a sufficient waiting time is required for the liquid resin to settle down into a flat surface. However, such recoating methods typically take over a minute, which limits the building speed of the MIP-SL process. Consequently, the building time of such MIP-SL systems is still measured in hours. To address the building speed limitation of the MIP-SL process, we present a novel approach for quickly spreading liquid resin into uniform thin layers. Our approach is based on a 14 two-way movement design in a meso-scale bottom-up projection system, and a pull-up directly design in micro-scale bottom-up projection system. The force occurred during the separation process in z and x direction is shown in Figure 1.7 b. There is large attaching force between the cured layer and the bottom surface of the resin vat in the bottom-up projection system. With too big separation force in z or x direction, the part will be failed to be built, as shown in Figure 1.8 (a) and (b). A bigger moving acceleration and velocity is preferred to fasten the building speed, yet it would result in bigger separation force and hence probably building failure. (a) (b) Figure 1.7: An illustration of bottom-up projection configuration and separation forces Vz △P F Z h stationary Liquid Air Cured part PDMS Film A A’ z x F X Vx 15 (a) Failed pillars due to insufficiently curing (b) pillars built by sliding (left) and up-down motions (right) Figure 1.8: Illustrations of built pillars. Similar to the bottom-up projection based MIP-SL system, the separation force is an important factor in CNC accumulation system that affects the building speed significantly. Instead of attaching to the bottom surface of resin vat, with over-large attaching force, the cured resin will attach to the accumulation tool in CNC accumulation system, since the tool is merged in the resin and in direct contact with resin during the building process. Thus a critical problem to be addressed is how to ensure the newly cured resin will attach to the base or previously built part instead of the accumulation tool, and make the fast curing speed as fast as possible. To overcome this detaching performance and building speed problem, we studied the attaching forces in the CNC accumulation process. Assume F Base is the attaching force between the newly cured resin portion and the previously built part, and F Tool is the attaching force between the newly cured resin portion and the curing tool. A proper gap needs to be maintained to ensure F Base > F Tool . The gap and the bonding force are highly related to the curing energy. Based on the curing equations and design of experiment, the attaching force related to the gap and the exposure energy was modeled with critical time-gap states identified to ensure the cured resin can be safely detached from the tool. -Non-layer based Photopolymerization AM: CNC Accumulation: 0.5mm b b A-A F P A A (a) Failed pillars F B L A A 0.5mm 0.5mm (b) 0.5mm b b A-A F P A A (a) Failed pillars F B L A A 0.5mm 0.5mm (b) 16 As mentioned in previous section, the CNC accumulation process is a layerless additive manufacturing process that directly uses a tool inside a tank to selectively cure liquid resin. Two main properties of such an AM process include: (1) The allowable motions between the tool and the work piece can be significantly increased; and (2) the feeding of liquid resin for curing is straightforward that can ensure the material accumulation process. It could address the intrinsic problems in planar layer based AM systems like stair stepping effect and inconsistent material properties. However, the resolution and speed is dependent on the tool size. With a bigger tool, the cured spot size would be bigger yet the curing speed would be slower due to the lower unit energy, and vice versa. So in order to improve the build speed of CNC accumulation system, the curing mechanisms need to be studied and the machine designs need to be adjusted. In this paper, a multi-tool station is designed and applied in the CNC accumulation process, and process parameters are optimized to improve the build speed and throughput. 1.2 Research Problem and Hypothesis 1.2.1 Research Scope, Problem and Strategy Research Scope: As stated in the above section, the research scope is defined as the Photopolymerization based AM processes. Obviously examining all the photo-polymerization AM processes which would be an impossible mission. So two typical examples of photo- polymerization AM processes, MIP-SL is studied as an example of layer based processes, and CNC accumulation is studied as the example of nonlayer based processes. All our discussions in this dissertation will focus on these two systems. Research Problem: As introduced in the previous section, in layer based photopolymerization AM processes, like SLA and MIP-SL, due to the use of the 2D slices, a staircase effect usually damages the surface quality greatly and limits the use of the AM in many 17 applications. Since the approximation error and the stair case effect depend on the layer thickness used in the slicing of a 3D model, a dominant approach is to reduce the layer thickness or adding post-processing steps. Yet both methods would significantly slow down the building process. On the contrary, if a faster speed is desired, a bigger layer thickness should be adopted and hence the staircase problem would be worse. The same conflict of improving surface quality or build speed exists in nonlayer based photopolymerization AM systems like CNC accumulation and two- photon technologies. The surface quality and build speed in nonlayer based photopolymerization AM systems have contrary requirements on the cure point shape and sizes. It is a well-known problem that high surface quality or fast build speed is always achieved at the cost of the other. Thus the goal is to solve the conflict between build speed and surface quality in photopolymerization AM. Figure 1.9 shows how we develop our research work based on the defined research scope and the identified research problem. The research problem Q1 could be further divided into two sub problems Q1.1 and Q1.2. Research Strategy and Content: As shown in Figure 1.9, the two main aspects of the manufacturing processes: energy control and material deposition, will be studied to explore the approaches to fundamentally solve the research problem. The research could be categorized into four classes to solve the two sub problems Q1.1 and Q1.2: (1) Fast Energy Delivery; (2) Accurate Energy Delivery; (3) Fast Material Deposition; (4) Accurate Material Deposition. Through controlling the energy delivery and depositing the material accurately and rapidly, it is hypothesized that the system could achieve fast build speed together with high surface quality without any compromise. 18 Additive Manufacturing using Photopolymerization Layer-based processes Non-layer based processes SLA: point-wise laser scanning, micro-scale fabrications. MIP-SL: layer- wise irradiation, micro- to meso- scale fabrications Two-photon: laser beam intersections, nano- to sub-micron scale fabrications CNC Accumulation: accumulation tool, micro- to meso- scale fabrications Research Scope: Research Problem: Layer thickness Cure Point shape & size Build Speed Surface quality: Surface finish, Geometry accuracy Surface quality: Surface finish, Geometry accuracy Build Speed ? ? Q1. How to achieve high build speed together with high surface quality in photo-polymerization based additive manufacturing processes? Q1.1 How to achieve high build speed without sacrificing the part quality? Q1.2 How to achieve high surface quality without compromising the build speed? Research Strategy: Energy Control Material Deposition Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research question: Research Content: Figure 1.9: Development of research work to address the research problem in this dissertation 19 1.2.2 Hypotheses As discussed in previous section, the research goal of this work is the development of energy control and material deposition approaches for photo-polymerization based additive manufacturing processes to achieve high build speed and high part surface quality simultaneously. The according primary research question is stated as follows: Primary Research Question: Q1. How to achieve high build speed together with high surface quality in photo- polymerization based additive manufacturing processes? Accordingly, this question could be subdivided to the following to sub-questions. Q1.1 How to achieve high build speed without sacrificing the part quality? Q1.2 How to achieve high surface quality without compromising the build speed? As discussed in previous section, to achieve this goal, we examined both the energy control and material deposition aspects, which could be further divided into four categories: Fast energy delivery and accurate energy delivery, fast material deposition and accurate material deposition. To find solutions in these four categories, the following hypotheses are investigated: H1. A multi-tool design and optimized process could make the energy delivery fast in CNC accumulation system, thus to improve the build speed and surface quality together for building on platform tasks in CNC accumulation system. 20 H2. A 3D surface reconstruction approach and automatic optimal tool path generation could make the energy delivery accurate in CNC accumulation system, thus to improve the build speed and surface quality together for building around inserts tasks in CNC accumulation system. H3. A gray scale image method could make the energy delivery accurate in MIP-SL, thus to fabricate smooth down-facing surfaces in MIP-SL system together with the same or even faster build speed. H4. A meniscus approach in MIP-SL system could deposit material with an accurate profile, thus to fabricate smooth up-facing surfaces in MIP-SL system together with the same or even faster build speed. H5. A fast material deposition method in MIP-SL process could further improve the build speed of MIP-SL systems without affecting the surface quality. For the first question Q1.1, our goal is first to develop scientific understanding of the relationship of curing performance with accumulation tool movement velocity in CNC accumulation, and the recoating process in MIP-SL. Based on the understanding, we will develop and implement optimized models for manufacturing process parameters in both CNC accumulation and MIP-SL processes, and then use analytic and experimental method to intellectually solve the challenge in fast material deposition and hence increase the build speed. A multi-tool CNC accumulation process is designed and developed. Statistical methodologies are applied to perform the accumulation tool selections and set the optimal build speed of each accumulation tool under different settings. The complex relationship between 21 various parameter settings in the CNC accumulation system are studied, such as curing speed, curing gap, UV power, base material and curing strategy. As shown in the similar work for the machining and AM processes [18, 63-66], the statistical method has a good capability in identifying appropriate manufacturing process settings. With the help of the statistical model, the optimal curing speed under various parameter settings is identified in the multi-tool CNC accumulation system. Therefore, a desired curing resolution and fast curing speed can be guaranteed at the same time. Multiple experiments have been conducted and it's verified that the multi-tool CNC accumulation system has a very good capability in efficiently fabricating parts with a very smooth surface finish as well as isotropic mechanical property. As an non-layer photopolymerization AM process, the CNC accumulation can address the stair stepping issue perfectly with a well-planned tool path. The curing tool can change its orientation with the axis of the part to get good part quality and surface finish. The aforementioned process planning model using statistical method is a very helpful tool for setting process parameters and tool movement speeds of multi-tools in CNC accumulation system, for tasks that build an object directly from the scratch. Thus the process planning model enables the CNC accumulation system to achieve the research goal for building directly on platform tasks. However, for another type build task which is another important application of the nonlayer photopolymerization AM, building-round-inserts, the tool path planning is still a big challenge. To facilitate the automatic tool path planning, a scanning unit and a 3D surface reconstruction algorithm are developed and integrated in the CNC accumulation. With the generated tool path, the CNC accumulation is capable of delivering energy to the accurate position with a proper tool orientation and a proper gap distance, thus is able to achieve fast build speed together with high surface quality in building-around-inserts tasks as well. 22 In layer based photopolymerization AM, stair stepping effect is a notorious problem that causes the research problem. The existing approaches like the use of ultra-thin layers or post- processing process are used to mitigate the problem, at the expense of longer building time. Gray scale method and meniscus approach are proposed to fabricate smooth surfaces with the same or even bigger layer thickness for layer-based photopolymerization AM processes in this research. It can eliminate the stair-stepping effect fundamentally and completely. And because it uses the same or even bigger layer thickness, the build speed would not be slowed down or even could be increased. Using the proposed meniscus approach, we can build highly smooth surface using MIP-SL process without compromising the build speed, and with even higher build speed in some conditions. To further improve the build speed without impacting the surface quality in MIP-SL system, the two-way movement approach and pull-up directly method are novel fast recoating approaches that are proposed in this dissertation. There is difference in the separation forces between meso- scale and micro-scale fabrications, and thus different separation mechanism should be adopted for different building area sizes. No work has been reported about different methods of fast separation and recoating in different manufacturing scales. Using the proposed fast recoating method, we can build meso-scale parts and micro-scale parts in minutes, instead of hours like other reported MIP-SL systems. 1.3 Research Content and Process The research procedure is developed according to the hypotheses step by step, as shown in Figure 1.10. Chapter 2 first describes the principle of photo-polymerization, energy distribution in MIP-SL and CNC accumulation systems. Characteristics and unique challenges in energy control in these two processes are discussed. After that, each hypothesis is verified in one of the 23 following chapters. In Chapter 3, hypothesis 1 is investigated and verified. To deliver the energy faster hence achieve a higher build speed, a multi-tool station is designed and developed in CNC accumulation process. And then a systematic process parameters optimization method is developed for the multi-tool CNC accumulation system. Additionally, in Chapter 4, a dual-axis 3D scanner is developed and integrated to assist tool path planning for accurate energy control in the CNC accumulation process. The design and development of 3D scanner, the enabled automatic tool path generation, and the experiments on building around inserts are presented. In Chapter 5, we extended the study to energy control in MIP-SL system. Gray-scale Image is adopted to manipulate the pixel-to-pixel energy distribution. Experiments show that the developed gray-scale method is able to fabricate smooth down-facing curved surface without compromising the build speed. In Chapter 6 and Chapter 7, novel material deposition methods are investigated and developed to achieve fast build speed and high surface quality. In Chapter 6, a novel material deposition approach, the controlled meniscus approach mentioned in hypothesis 4, is proposed to address the stair stepping effect in up-facing surfaces without sacrificing build speed. The meniscus approach is implemented in both top-down projection and bottom-up projection system to fabricate smooth curved surface. The effectiveness and efficiency of the proposed meniscus approach are validated through various test cases in Chapter 6. The last hypothesis, H5, is verified in Chapter 7. Different projection modes are discussed, and separation forces are modeled in this chapter. Based on the separation force study, fast recoating approaches for meso- scale and micro-scale fabrications are developed. It is shown in Chapter 7 that the developed fast recoating method is capable of improving the build speed of MIP-SL process significantly. 24 Lastly, Chapter 8 summarized the whole research on energy control and material deposition study for high build speed and surface quality. Conclusions are drawn and suggestion is made for future research. To demonstrate the structure of this dissertation better, an overview of the chapters in this dissertation and their connections is shown as Figure 1.11. Research Strategy: Energy Control Material Deposition Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research Content: 5 Energy Control Material Deposition Fast Delivery Accurate Delivery Fast Deposition Accurate Deposition CNC Accumulation MIP-SL H2: 3D surface reconstruction, Optimal tool path (Chapter 4) H1: Multi-tool, Process optimization (Chapter 3) H3: Gray- scale image (Chapter 5) H5: Fast recoating method (Chapter 7) H4: Controlled meniscus Approach (Chapter 6) √ √ √ Figure 1.10: Relationship between hypothesis and research contents 25 Figure 1.11: Overview of the dissertation 26 Chapter 2 Energy Control in Photo-polymerization AM Systems 2.1 Principle of Photo-polymerization As mentioned in Chapter 1, polymerization is the process of linking small molecules (monomers) into larger molecules (polymers) composed of many monomer units. Two main types of photopolymer chemistry are widely used in commercial applications: (1) Free-radical photopolymerization – acrylate; (2) Cationic photopolymerization – epoxy and vinylether. Epoxy SL resins typically have much smaller shrinkages, and much less tendency to warp and curl. The free radical-initiated polymerization process can be described as follows: (1) Free radical formation: P-I --- I* , where P-I is a photoinitiator, I* is a free radical. (2) Initiation: I* + M --- I-M*, where M is a monomer. (3) Propagation: I-M* --- I-M-M-M-M...-M* (4)Termination: --- I-M-M-M-M...-M-I Cationic photopolymerization has the same broad structure. A photoinitiator generates a cation as a result of laser energy. Acrylate photopolymer usually has high photospeed when it is exposed to UV radiation, but it has bigger shrinkage rate and tends to warp and curl[67]. On the other side, the epoxy resins have disadvantages like slow photospeed and brittleness of the cured parts. Additionally, the photopolymerization would be inhibited when it is exposed to the oxygen[68]. In most commercial SL resins, the acrylates are added to reduce the brittleness of the epoxy parts. 2.2 Energy Sources for Photo-polymerization The knowledge on photopolymerization principle and energy sources enables us to investigate the solutions to the aforementioned challenges in photo-polymerization AM processes. 27 We will begin with an investigation into the fundamental energy sources and the relative energy distribution characteristics. As introduced in Chapter 1, various types of radiation could be used to cure SL photopolymers, including UV, electron beams, visible light or even X-rays and gamma rays. Among those options, UV and visible light radiation are most widely used in commercial photopolymerization AM systems. Different photopolymer reacts to different light wavelength ranges, thus exhibit different curing performances. After decided what energy to use, various configurations could be designed to deliver the radiation. Two main categories of configuration are used in photo-polymerization AM system, that is, vat configuration in which liquid photopolymer is contained in a vat, and non-vat configuration in which liquid photopolymer is dropped on demand like ink-jet printing. The non- vat configuration is out of the research scope of this paper. We are only dealing with the vat configuration in this research. As discussed in Chapter 1, in vat configuration, there are mainly the following sub-categories: (1) Layer based process: Vat scan, in which a laser beam is focused and shoot on the photopolymer liquid surface. The laser dot size determines the resolution and the scanning speed of the laser dot determines the build speed directly. (2) Layer based process: MIP-SL, in which a UV lamp or visible light lamp is used and a dynamic mask image generator is used to pattern the light to irradiate entire layers at one time. So the resolution is dependent on the mask and the build speed is a result of the interaction between curing speed and energy power. (3) Non-layer based process: Two-photon, in which two scanning laser beams intersected and only the intersection point can be cured. Therefore, unlike the vector scan and mask projection, the curing process can occur in any point under the liquid surface and there is no need 28 to recoat a new layer of resin. The resolution is dependent on the intersection point size and the build speed is dependent on the two beams' scanning velocities. (4) Non-layer based process: CNC accumulation, in which a LED is used as the light source and optical fiber is used to deliver the light under the liquid surface. Same as two-photon technology, the curing process could occur inside the liquid and the material could be accumulated in any direction. Note that the two-photon technology has much higher requirements on laser beam control, including the beam size, scanning direction and speed, thus it needs very precise and costly control system and optics, while the CNC accumulation is much more costly effective and have more choices in materials. Another distinction is that CNC accumulation could have multiple tools with different tool sizes and shapes which could give a good balance between resolution and speed, while the two-photon technology is a point-by-point process and it has an intrinsic problem of light shading in building-around-inserts. Hence, in this research, MIP-SL is studied as a representative of layer-based photo- polymerization AM and CNC accumulation is studied as a representative of non-layer photo- polymerization AM process. 2.3 Energy Delivery Distinctions between CNC Accumulation and MIP-SL As introduced in previous section, the fundamental difference between CNC accumulation and MIP-SL are caused by the way how the light energy is delivered. As mentioned in the previous chapter, fiber optics is used as curing tool in CNC accumulation process. It is able to merge into the liquid resin, thus it is able to deliver the energy in any arbitrary direction. Therefore it has the freedom to accumulate material in whatever direction. It's a non-layer-based AM process. However, MIP-SL process can only deliver the energy in the gravitational direction, either top-down or bottom-up. Hence it is a uni-direction layer-based fabrication approach. The 29 orientation of the part matters a lot to the surface quality and anisotropy quality of the fabricated part because of the limitation of the unique energy delivery path. In MIP-SL process, the unique energy delivery path makes the algorithm development, software coding, and hardware implementation simpler, and the allowable motions between the tool and the work piece is less than CNC accumulation process. In most cases, only translational motions in the X, Y, and Z axes are allowed. The simplicity makes MIP-SL widely adopted. Also as an area-processed approach, MIP-SL has the advantage at fast building speed. However, embedding existing components in such processes is usually difficult. For example, Kataria and Rosen[69] identified some major problems related to building-around-inserts in such process including light shadowing and vat recoating. Compared with MIP-SL system, the tool path of the CNC accumulation process can have multi-axis motions; consequently, the CNC accumulation process is more flexible in delivering light energy into any position with any direction, which makes building-around-inserts feasible. But due to the size of curing tool, CNC accumulation process is less efficient than MIP-SL process and it is not good at building complicated geometry directly. These two processes are not competing with each other, but would be a good complement to each other. They both have strengths and are very helpful in certain manufacturing areas. 2.4 Energy Distribution in Mask Image Projection Stereolithography 2.4.1 Lateral energy distribution and lateral resolution Since the energy delivery provided by a DMD is controlled by a certain amount of discontinuous pixels, the building size is limited by the pixel numbers and the resolution is now limited by the pixel size of the mask image. Suppose the mask image resolution is 1024 x 768, then the pixel size is: 30 Pixel_Size_x= Platform_Size_x/1024; Pixel_Size_y= Platform_Size_y/768; If we want to increase the resolution, the pixel size should be smaller and then the building size would be shrank down, and the vice versa. Thus a tradeoff between the building size and the resolution has to be made. Meanwhile, the light energy distribution on the pixel would be more concentrated with a smaller image size. The effect of image size on resolution and energy distribution on the surface is illustrated on the following figure. Figure 2.1: Relationship between platform size, pixel resolution and energy distribution. To achieve the ideal lateral resolution, it is critical to ensure that the curing surface is exactly on the focus plane. Besides, a proper exposure time is also important to get the ideal 100mm 100µm 1mm 1µm E (mJ) X(um) X Y Platform size (a) Size vs. Resolution (b) Size vs. Energy lateral distribution Light energy of one pixel 31 lateral resolution since the details will be dampened by overcuring. In order to control the lateral energy distribution and thus the lateral resolution, the mask image should be well prepared. 2.4.2 Vertical energy distribution and vertical resolution Similar to the lateral energy distribution, a Gaussian distribution model is usually used as the first order approximation to describe the flux-density contribution of light from the image plane[34]. An illustration of the light distribution in vertical direction is showed in figure 2.2[34] (a) (b) Figure 2.2: An illustration of the light distribution in vertical direction. As illustrated on the above figure, the vertical resolution is primarily relied on the light penetration. Although using thinner layer could make the edges smoother, the vertical resolution of overhanging features will still be damaged if the light penetration is not controlled well. To achieve higher resolution in vertical direction, either the energy vertical distribution or the power should be manipulated or the photopolymer properties should be altered. In order to better control the vertical resolution, the relationship between energy power and photopolymer curing properties need to be investigated. 2.5 Energy Distribution in Multi-axis CNC Accumulation Process 32 (a) Point curing (b) line curing in CNC Accumulation (Adapted from[40]) (c) Cured point with 1s, 5s, and 10s curing time Figure 2.3: An illustration of light energy distribution in CNC accumulation According to the Beer-Lambert exponential law of absorption, the light exposure will decrease exponentially with depth z. As shown in Figure 2.3 (a) (b), the material is cured along the axis of the accumulation tool. It is also shown in (c) that the cured shape, width and length varies greatly with the curing time. To conclude, the cured shape and size is dependent on factors including the size and shape of the accumulation tool tip, the gap between the tool tip and the base surface, the moving direction and the moving speed of the tool. To achieve a good resolution Cured resin Liquid resin Curing tool depth (a) (b) V Cured resin Liquid resin Curing tool gap width Base (cured resin) Coating (Teflon film) SF Tool SF Base A Tool A Base 1s 5s 10s 33 in lateral and vertical direction, process parameters including a proper moving speed and orientation of the tool, and an appropriate gap distance need to be identified. To achieve a fast build speed or high throughput, multiple tools are investigated and adopted in the following chapter. Following with the developed multi-tool CNC accumulation system, the optimal process parameters are explored and identified in the same chapter. 2.6 Concluding Remarks In this Chapter, the discussion is extended from the background introduction in Chapter 1 to energy distribution characteristics in the two specific photo-polymerization AM processes: MIP- SL and CNC accumulation. The energy distribution on lateral surface and vertical direction in MIP-SL is analyzed. It is dependent on the mask image properties. The resolution is influenced by the energy, pixel size and also other factors like optical properties and curing time. Unlike MIP-SL system, in CNC accumulation system, the light is able to be directed in any arbitrary direction due to the orientation of the tool. Thus the energy distribution is influenced and also could be controlled by more parameters, including tool size and shape, tool orientation and moving speed, and the gap distance. Through adjusting those parameters in CNC accumulation, the energy magnitude and distribution map could be controlled and then the curing performance could be optimized. Based on the thorough analysis of those energy related issues, we are now ready to further develop state-of-the-art energy control methods to solve the key problems and then improve the performance of MIP-SL and CNC accumulation process. In Chapter 3, we will present the design and development of multiple tools in CNC accumulation process. With multiple tools, the energy could be delivered much faster and therefore the total build time could be shortened greatly. Additionally, we also show a systematic process parameter optimization approach in the multi-tool CNC accumulation system to control the energy in a way that desired 34 resolution could be achieved with a high build speed. In Chapter 4, a 3D scanning unit and a 3D reconstruction pipeline are developed and integrated to plan the tool path and orientation of the accumulation tools in CNC accumulation. With the assistance of the automatic tool path generation, the energy could be controlled and distributed accurately and the fabrication capability is greatly improved as a result. In Chapter 5, a gray scale image method is developed and the experimental results illustrate that the developed approach is able to fabricate smooth surface with a well-controlled energy distribution on the lateral surface and in the vertical direction. As shown in Figure 2.4, we'll first develop the fast energy delivering method in CNC accumulation system to verify the hypothesis H1 in the following chapter. Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research Content: 9 Energy Control Material Deposition Fast Delivery Accurate Delivery Fast Deposition Accurate Deposition CNC Accumulation MIP-SL √ √ √ ? H1 Figure 2.4: Research progress by this chapter and the following research work 35 Chapter 3 Fast Energy Delivery in CNC Accumulation through Process Optimization 3.1 Design and Development of Multi-tool and Multi-axis CNC Accumulation System As discussed in Chapter 1 and 2, the material deposition is straightforward and very simple, while the energy control is the primary factor that affects the performance of CNC Accumulation system. How fast that we could deliver the energy with the accurate amount to the accurate position with a proper orientation determines the surface quality of the fabricated part and the build time of the task. We will first investigate the use of multiple tools with multi-axis motions to achieve fast energy delivery in CNC accumulation in this chapter. In the proposed multi-axis CNC accumulation system[40], a single accumulation tool was used. The light spot size of the tool is ~1.57mm. Such a resolution is usually insufficient for fabricating fine features on curved surfaces. Motivated by the commonly used multi-tool stations in CNC machines [70-72], to improve the build speed and resolution, multiple tools and multi- axis tool motions in the CNC accumulation process are investigated and developed: (1) Multiple tools. A single accumulation tool was used in our previous work[40]. The light spot size of the tool is ~1.57mm. Such a resolution is usually insufficient for fabricating fine features on curved surfaces. Motivated by the commonly used multi-tool stations in CNC machines [70-72], alternative accumulation tools with higher resolution and in various shapes are investigated in this paper. Accordingly a CNC accumulation system with multiple tools has been developed. The tool path planning based on multiple accumulation tools are also discussed for building features with varying resolutions in a single component. 36 (2) Multi-axis tool motion: In addition to translational motions in the X, Y, and Z axes that are commonly used in the layer-based AM processes, the tools in the CNC accumulation process can have multiple rotational motions as well. Inspired by the multi-axis CNC machining process, the tool path planning for building features on curved surfaces such as cylindrical and spherical surfaces is investigated. The desired tool motion is required to consider both built features and the underlying surface in order to achieve desired shapes. Figure 3.1 shows a schematic illustration of the multi-tool and multi-axis CNC accumulation process. An accordingly developed CNC accumulation system is shown in Figure 3.2. Figure 3.1: A schematic illustration of the multi-tool and multi-axis CNC accumulation process. 2 Fast Energy Delivery in CNC Accumulation Design of multiple tools with different shapes and sizes: Small tool big tool Schematic illustration of the multi-tool and multi-axis CNC accumulation: Computer control O X Y Z A B LED with power supply Focusing lens Fiber optic cable Teflon film + - Resin Tool Attaching force on built part Attaching force on tool tip Newly cured resin portion Big tool 5-axis motion + - Small tool C A Cured resin x Y Z A C 5-Axis motion: angled tool 37 Figure 3.2: The hardware setup of a multi-tool and multi-axis CNC accumulation system. In the system, two sets of accumulative tools are used to demonstrate the multi-tool station concept. The ultraviolet (UV) LED of the larger tool is from Nichia (NCSU033A). A sapphire ball lenses (NT43-831 from Edmunds Optics) is used to focus light on a quartz fiber optic light guide (NT38-954 from Edmunds Optics). The spot size of the provided ultraviolet (UV) light has a similar size to the core size of the optical fiber, which is 1.57mm. The UV-LED used in the smaller tool is from Mightex System (FCS-0365-000). The optical fiber is also from Mightex System (FPC-0200-22-02SMA). The core size of the optical fiber is much smaller (~0.3mm or 300m). A Teflon film is applied on the tip of both tools. A 5-axis motion system is designed to allow the accumulation tools to be able to provide UV light in various orientations. The 5-axis motion includes the X, Y, Z translations and the A, C rotations. The X, Y, and C motions are applied to the workpiece while the other two are applied to the tool. The experimental results 38 illustrate the unique properties of the CNC accumulation process and its use in fabricating conformal features on curved surfaces. The remainder of this chapter is organized as follows. Section 3.2 presents the curing properties of two accumulation tools. Section 3.3 presents the process settings of the accumulation tools based on a statistical study. Section 3.4 discusses the multi-axis tool motion planning for building features on curved surfaces. Section 3.5 presents tool selection methods for building features on curved surfaces, followed by four test cases and related experimental results shown in Section 3.6. Finally, conclusions from this study and future work are presented in Section 3.7. 3.2 Curing Performance of Multiple Tools in CNC Accumulation Similar to the CNC machining process, it is desired to have a wide variety of accumulation tools in the CNC accumulation process that may have different shapes, sizes, light intensities, and resolutions (refer to Figure 3.3). In addition, a tool changing station may be developed to enable the dynamic change of the accumulation tools during the building process. Hence an appropriate tool can be used in fabricating a given geometric shape. Figure 3.3: An illustration of multi-tool stations of a CNC machine and a CNC accumulation system. 39 In the CNC accumulation process, sufficient energy has to be provided to initiate the polymerization process. As extensively studied in the SLA process, a critical energy exposure threshold can be found for a given type of liquid resin. According to our previous study of the gap between the tool and the built part, a specific gap as well as the curing depth can be identified based on the critical energy[40]. For fixed curing depth and penetration depth, the curing width is proportional to the spot diameter ( 2 w d p L B C D  ). Thus, a highly focused UV spot can cure a feature with high resolution while a tool with a much larger spot can cure a large area with an increased building speed. A single tool with a spot diameter around 2mm was used in our previous work[40]. In this study, a second tool is added with a spot diameter of 0.3mm. In the paper they are named large tool and small tool respectively. By integrating the two accumulation tools in the system, a good balance between the building resolution and fabrication speed can be achieved. The curing performances of the two accumulation tools based on the aforementioned UV LEDs and fiber cables are measured and compared. For a photocurable resin (Accura 60 from 3D Systems), a set of experiments have been conducted to model the curing performance of the accumulation tools. 40 Figure 3.4: The relation between curing depth and exposure time for the accumulation tools. Figure 3.5: The relation between curing width and motion speed for the accumulation tools. Figure 3.4 shows the relation between curing depth and curing time for the point curing case (i.e. the curing tool is moving in the curing direction[40] ). As shown in the figure, the fitted 41 curves are close to the polymerization equations for both tools. In addition, for the same exposure time, the small tool will always cure deeper than the large tool. Results also indicate that the cured shape by using the small tool has a smaller size than the large tool. Figure 3.5 shows the relation between curing width and curing speed for the line curing case (i.e. the curing tool is moving in a direction that is orthogonal to the curing direction[40] ). As shown in the figure, the large tool has much larger curing width than the small tool for the same motion speed. When the curing speed is fast (>0.75 mm/s), nothing will be cured by using the large tool while the small tool can still cure lines with a high resolution. Based on our experiments, the minimum line width (i.e. lateral resolution) that can be achieved by the small and big tools is around 0.2mm and 1.5mm respectively. So in order to achieve fast build speed and good part quality, process parameters need to be studied and optimized in the multi-tool and multi-axis CNC accumulation system. 3.3 Process Settings of the Multi-tool CNC Accumulation System As shown in Figs. 3.4 and 3.5, the small tool has a higher motion velocity, a bigger curing depth and a smaller line width. Although the curing performance results provide a good reference for setting appropriate parameter values, it should be noted that the tests were carried out by moving the tools that are positioned above liquid resin. There are several process variables involved in the CNC accumulation that have not been considered, e.g. gap distance, base material, and interaction with liquid flow, etc. One critical process setting in the accumulation process is the tool velocity. If a tool moves too fast, the accumulation process may fail because the newly cured resin may not be able to attach to the base or previous cured parts. On the contrary, a slow tool velocity will lead to longer building time. It will also enlarge cured features, which leads to built parts with lower resolutions. 42 The process settings for the large tool have been studied based on trial-and-errors in [40]. Note the resolution of the large tool is much lower than that of the small tool. In addition, as shown in Figure 3.5, the feasible velocity range of the small tool is much bigger than that of the large tool. To identify the optimal curing speed of the small tool under different parameters settings, such as curing gap, UV power, base material, temperature, and curing strategy, we adopt a statistical methodology by using Design of Experiments (DOE) techniques. 3.3.1 Design of experiments In this study, the relation between the curing speed and other processing parameters is to be established. The important process parameters that may affect the optimal curing speed include curing gap, UV light intensity, base material, resin temperature, and curing strategy, etc. As mentioned before, there are two major types of curing motions in the CNC accumulation process, line curing and point curing[40]. For both cases, experiments are designed to identify the fastest curing speed, at which the desired building resolution and success rate can be achieved. Response. The response of the statistical study is the curing speed V. The highest speed Vmax in our hardware setup is 3.556 mm/sec. To simplify the recording and calculating process, we will use a coefficient value max V V to represent a curing speed V. Factors. Four factors including UV power, gap distance, curing strategy, and base type, are chosen in the study. (a) UV power: Based on the availability of the light intensity setting given by the UV LED system, this factor has two levels: 0.495w, denoted as +1 and 0.231w, denoted as -1. (b) Gap distance: The thickness of the cured portion is equal to the distance between the tool head and the base. According to our previous experience, this factor has two levels: 1.27mm denoted as +1, and 0.762mm denoted as -1. (c) Base type: Two types of material bases are tested including a plastic film and a cured resin surface. The formal one is set as +1 and the latter -1. (d) 43 Curing strategy: Two curing strategies were tested including (i) the tool head is above the resin surface; and (ii) the tool head is merged in the liquid resin. In both cases a Teflon coating is applied on the tool head. The formal one is set to be the first level (denoted as +1), and the latter one is denoted as -1. The factor settings are summarized in Table 3.1. Table 3.1: Factors and their settings in the curing experiment Factor Low level (-) High level (+) A-UV light power 231 mw 495 mw B – Gap distance 0.762mm 1.27mm C – Base type cured resin clean plastic film D- Curing strategy in the resin above the resin Experiment design. In the line curing experiments there are four factors in the line curing study. A 2 (4-1) fractional factorial design is planned to study the defining relationship of I=ABCD. In the point curing experiments, there are three factors in the point curing study since the curing strategy (D) must be in resin. A 2 3 full factorial design experiment is designed for the three factors (A, B and C). Three replicates are carried out for each designed setting to analyze the dispersion. Experiment procedure. For the line curing study, a 2D line on a planar base is fabricated by moving the tool head horizontally. For the point curing study, a 3D rod on a planar base is fabricated by moving the tool head along the normal of the base surface. For each given parameter setting, experiments are performed by slightly varying the curing speed V. For each tested speed, ten replicated experiments are performed. The maximum speed is identified such that the related curing line resolution is less than 0.3mm and the success rate is larger than 80% (i.e. eight or more results satisfy the given requirements). 3.3.2 Experimental result and analysis – line curing 44 The experimental result of the line curing case is shown as in Table 3.2. As discussed before, the defining relation is I=ABCD, where A is UV light power, B is gap distance, C is base type, and D is curing strategy. The aliasing relationships are: A=BCD; B=ACD; C=ABD; D=ABC; AB=CD; AC=BD; AD=BC. All the four main effects are clear. According to the hierarchical ordering principle, the main factors A, B, C and D are likely to be more significant than the three way interaction factors BCD, ACD, ABD and ABC. In addition, two-way interaction effect AB is more significant than CD; AC is more significant than BD; and BC is more significant than AD. The factorial effects can be obtained as shown in Table 3.3. Table 3.2: Curing experimental result for the line curing case A B C D Y y s 2 lns 2 + + + + 0.65 0.638 0.638 0.642 5.2 10 -5 -9.863 - + - + 0.375 0.4 0.4 0.392 2.0810 -4 -8.476 + - - + 0.65 0.775 0.785 0.737 5.6610 -3 -5.175 - - + + 0.591 0.591 0.555 0.579 4.2210 -4 -7.770 + + - - 0.188 0.2 0.2 0.196 5.2110 -5 -9.863 - + + - 0.113 0.12 0.115 0.116 1.4610 -5 -11.136 + - + - 0.7 0.73 0.711 0.714 2.310 -4 -8.376 - - - - 0.193 0.208 0.208 0.203 7.510 -5 -9.498 Table 3.3: Factorial effects in curing experiment for the line curing case Effect y 2 ln s A 0.2498 0.9009 B -0.2216 -2.1298 C 0.1308 -1.0330 D 0.2802 1.8973 AB -0.0848 -0.9575 AC 0.0806 -0.5677 45 BC -0.0458 -0.2966 The half-normal plots (refer to Figure 3.6) were used to analyze the significant effects. It is indicated that main effects D and A are significant to location, while B and D are significant to dispersion. The regression equations for location and dispersion, respectively, are: ˆ 0.44707 0.14011 0.12489 DA y x x    (3.1) 2 ˆ ˆ ln 8.7694 1.0649 0.9487 -0.5165x B D C z s x x      (3.2) where ˆ y is the expectation of the curing speed, and ˆ z is the variance. Since the critical point of the optimal speed is identified as the maximum feasible speed, it can be considered as a Larger-the-Better problem here. Consequently both +1 levels for A and D are chosen to maximize the average expected optimal velocity E(y), and B and C are set to +1 as well to minimize the variance of optimal velocity Var(y). Thus the optimal setting for the line curing case should be (+,+,+,+), which means that light power is 495 mw, gap size is the big level 1.27 mm, the base material is a plastic film, and the tool is above the liquid resin surface. Figure 3.6: Half-normal plots of the curing experiment for the line curing case. 3.3.3 Experimental results and analysis – point curing The experimental result of the point curing case is shown as in Table 3.4. The factorial effects and half-normal plots are shown in Table 3.5 and Figure 3.7, respectively. 46 Table 3.4: Curing experiment for the point curing case A B C Y y s 2 lns 2 + + + 0.145 0.18 0.18 0.168 4 4.08 10   -7.803 + + - 0.132 0.127 0.125 0.128 5 1.36 10   -11.207 - + + 0.05 0.053 0.056 0.053 6 7.58 10   -11.790 - + - 0.065 0.06 0.069 0.065 5 1.83 10   -10.911 + - + 0.15 0.18 0.15 0.160 4 3.00 10   -8.112 + - - 0.12 0.130 0.12 0.123 5 3.01 10   -10.412 - - + 0.114 0.1 0.09 0.101 4 1.45 10   -8.837 - - - 0.102 0.1 0.1 0.101 7 7.50 10   -14.103 It is shown that A and AB are significant location effects, and main effects C and A are significant dispersion effects. The regression equations for location and dispersion are given as follows: ˆ 0.112312+0.032521 0.012187 A AB y x x  (3.3) 2 ˆ ˆ ln 10.39674 1 .26144 1 .01340 CA z s x x      (3.4) Thus the optimal setting for the point curing case should be (+,+,-), which means that light power is 495mw, gap size is the big level 1.27 mm, and the base material is cured resin. Table 3.5: Factorial effects in curing experiment for the point curing case Effect y 2 ln s A 0.0650 2.0268 B -0.0179 -0.0620 C 0.0166 2.5229 AB 0.0243 -0.1814 AC 0.0220 0.3286 BC -0.0022 -1.2604 ABC 0.0040 1.8121 47 Figure 3.7: Half-normal plots of the curing experiment for the point curing case. 3.3.4 Verification experiments and results Two test cases based on a dragon pattern and several characters are performed to verify the identified process settings and to compare the building resolution difference of the two accumulation tools. The shape and related tool paths of the dragon pattern and the characters are shown in Figure 3.8a. The bounding box size of the dragon pattern is 25.4× 12.7 mm. Figure 3.8b shows the cured dragon pattern using the large tool. Figure 3.8c shows the cured dragon pattern using the small tool with the settings of (+,,+,+), that is, the gap size of 0.762 mm (B=-1). In comparison, Figure 3.8d shows the cured dragon pattern using the small tool with the optimal setting identified in the statistical study (+,+,+,+). The testing results verified that using the small tool with the optimal setting identified in the statistical study can lead to a building resolution that is less than 300m. Another test case based on several characters is also shown in Figure 3.8a. Figure 3.8e shows the cured characters using the large tool. In comparison, Figure 3.8f shows the cured pattern using the small tool with the setting identified in the statistical study. The comparison results also illustrate that the small tool has a much higher resolution and, consequently, can achieve clear outlines from the tool path. C 48 (a) Tool paths for a dragon pattern and characters (b) dragon pattern cured by large tool (c) Dragon cured by small tool before optimization (d) Dragon cured by small tool after optimization (e) Characters cured by large tool (f) Characters cured by small tool after optimization Figure 3.8: Comparison between different tools for a 2D dragon pattern and some characters. 3.4 Motion Planning for Building Features on Curved Surfaces Extensive work has been done on tool path planning for CNC machining (e.g.[73-77]). In comparison, few attempts have been made on multi-axis tool path planning for additive manufacturing processes. In a work done by Sundaram and Choi[78], a direct slicing procedure for the 5-axis Laser-aided Metal Deposition process is presented. A neutral exchange format IGES was used as the slicing format. The SURFCAM system developed for 3- to 5-axis milling process was customized for the material deposition operation. An adaptive slicing method for the 1 mm 49 multi-axis Laser-aided Manufacturing Process (LAMP) was investigated in [79-81]. A software system was developed using Visual C++ and OpenGL and integrated with their hardware system. The slicing direction can change with the surface curvature of the part. The layer thickness can also be changed to make the overhang of all adjacent layers be within the maximum overhang range. Thus the sliced part can be built without any support structure. The aforementioned multi- axis tool path planning methods are all designed for the metal-based AM processes. In addition, only a single tool was considered in the studies. For the purpose of building conformal features on curved surfaces, the tool motion and building process of the multi-tool and multi-axis CNC accumulative process are discussed as follows. 3.4.1 Tool motion for different curved surfaces In order to achieve a better lateral resolution and accuracy, it is generally desired to plan the circular path by using rotation motion instead of the interpolation of X-Y translations as illustrated in the 5-axis CNC machining process[82, 83]. For a given curved surface especially cylindrical and spherical surfaces, a five-axis motion including three translations along the X, Y, Z axis and two rotations A and C around the X and Z axes can be utilized. Figure 3.9a shows the 5-axis motion configuration incorporated in our CNC accumulation system. In addition to the X, Y, Z translations, the rotations C is used for controlling the tools to scan around spherical or cylindrical surface; and the rotation A is used to position the tool to the normal direction of a curved surface. Consequently, a desired surface quality and material consistency can be achieved by such a 5-axis CNC accumulation process. To simplify the tool path planning problem, a coordinate system is constructed with three constraints including: (i) the rotation centers O 1 and O 2 are aligned on the same height along the Z axis; (ii) the tool is aligned to be parallel to the Z axis; and (iii) the tool is pointing to the rotation center O 1 . 50 As shown in Figure 3.9b, an arbitrary curved surface could be described by two radiuses of curvatures at a given point, R 1 and R 2 . In this paper, three types of surfaces (flat, cylindrical, and spherical surfaces) are considered. (1) Flat surface: When R 1 and R 2 are both infinite, the surface is flat. Figure 3.9c presents an illustration of fabricating features on a flat surface. In the figure, the blue color lines represent the cured resin on a base surface; and a gray block denotes the tool head. Accordingly, the material curing path direction is denoted as T 2 , which is an interpolation of the X and Y axis motions. The material accumulation direction is denoted as T 1 , which can be achieved by the Z axis motion. Thus the tool path planning for the flat surface case is the same as conventional AM processes based on the translations along the X, Y and Z axes. (2) Cylindrical surface: When one of the curvatures is infinite and another one is finite, the curved surface is a cylindrical surface. Two different types of cylindrical surfaces are shown in Figure 3.9d and 3.9e. To fabricate conformal features on such surfaces, multiple axis movements including rotations are required. In both cases, the tool needs to be tilted by  degree using the rotation A such that it can be aligned to the normal direction of the cylindrical surface. Thus the material accumulation process can be facilitated either along the accumulation direction T 1 , or along the curing directions T 2 and T 3 . For the cylindrical surface as shown in Figure 3.9d, the Y and Z translations are combined to define T 1 and T 2 . In addition, the C rotation is used for curing around the cylindrical surface. For the cylindrical surface as shown in Figure 3.9e, the Y, Z translations are combined to define T 1 . The X translation is used to define the curing path T 2 along the cylindrical surface. 51 Figure 3.9: A schematic illustration of the 5-axis motion configuration for curved surfaces. (3) Spherical surface: When both curvatures R 1 and R 2 are finite and equivalent, the surface is a spherical surface. For fabricating features on a spherical surface, the rotation A is used to X Y Z C A O 2 O 1 tool Y Z X R 2 R 2 R 1 R 1 (a) (b) Z X Y Tool head T2 T1 Θ C A Z Y X T1 T2 (c) (d) Θ A Z Y X T1 T2 T3 Θ C A Tool head T1 T2 Z Y X (e) (f) 52 position the tool in the normal direction of the spherical surface. As shown in Figure 3.9f, the accumulation direction T 1 and the curing directions T 2 can be defined by combining the A, C rotations with the Y, Z translations. 3.4.2 Building process for features on a spherical surface As a cylindrical surface can be viewed as a special case of a spherical surface by setting one dimensional curvature as infinite, its tool path planning can be derived from the more general case of a spherical surface. The building of features on a spherical surface is discussed as follows. Figure 3.10: An example of building features on a spherical surface. Figure 3.10 shows an example in which a set of rods are fabricated on a spherical surface. In the multi-tool CNC accumulation process, the large tool is first used in building the base spherical surface. After the base surface has been built, the small tool can then be used in building the rods on the surface. A description of the building process based on both tools is given as follows. 1. Initialize the position for the large tool and the motion for 5 axes; Θ Tool head 1 st layer 5 th layer Base R d l a D A Z X Y C (a) (b) 53 2. Rotate the tool (Rotation A) by  to point the tool in the normal direction of the spherical surface; 3. Turn on the UV LED of the large tool; 4. Rotate the platform (Rotation C) by one full round; 5. Turn off the UV LED of the large tool; 6. Go to 2 if the base surface is not finished, otherwise go to 7; 7. Initialize the position for the small tool; 8. Control the motion back to the home position; 9. Rotate the small tool (Rotation A) by  to point the tool in the normal direction of the spherical surface; 10. Rotate the platform (Rotation C) by  , which is determined by the adjacent rods at the height related to d; 11. Turn on the UV LED of the small tool; 12. Move the small tool in the normal direction of the spherical surface (using the Y-Z translations). 13. Turn off the UV LED of the small tool if the rod is finished. 14. Go to 10 if not all the rods in the same Z level are finished; otherwise, go to 9 if not all of the rods are finished; 15. Stop the building process when all the rods have been built. For a given test case and the related CNC accumulative process, the design and manufacturing parameters and their constraints can be identified as follows. (1) Identifying design parameters:  : the A rotation angle between two neighboring rods; 54 d : the distance between two neighboring rods; i  : the A rotation angle between the bottom with the rods in the i th round; i  : the C rotation angle between two neighboring rods in the i th round; N : the number of rounds. (2) Identifying manufacturing parameters: D : diameter of the tool head; R : radius of the spherical base; a : diameter of the rod; l : height of the rod;  : the maximum rotation angle that the tool can achieve. (3) Parameter constraints: Some constraints may exist due to the tool and part collision and the hardware setup. For example, the tool head may collide the neighboring pillars if  is too small. Hence the minimum value of  can be derived based on D , a , R and l. As another example, suppose  is the maximum angle that the tool head can be rotated. Accordingly the maximum number of rounds (N) will be limited by  . A list of such constraints on the aforementioned parameters is given as follows. (i) 11 22 2 tan sin 2( ) 4( ) aD Rl a R l             ; (ii) i i i d R      ; (iii) 2 sin( ) ii n R d    ; (iv) 2 ii n   ; and (v) N     . With some adjustments, the parameters and constraints can also be applied to features on other curved surfaces. For example, to fabricate a dragon pattern on a spherical surface, the pattern can be decomposed into multiple layers on the surface. The parameter a and  can vary in different layers and be combined with the Y, Z motions to define a set of close loops. Note l 55 would not be a constant for each layer. Similarly, the building process can be extended to fabricate other types of features on a spherical or a cylindrical surface. 3.5 Tool Selection for Building Features on a Curved Surface 3.5.1 Selection of a straight accumulation tool For an accumulation tool that has a cylindrical shape with a diameter d t , two types of collision may exist in the fabrication process. In the first case (refer to Figure 3.11a), a tool may be too big to reach a curing depth c d in a curved surface with a curvature r s (i.e. 01 2 t d PP  ). In the second case (refer to Figure 3.11b), a tool may be too big for a rotating angle  related to a curved surface (i.e. 23 OP OP  ). The tool size based on the related constraints can be derived as: 2 (2 ) t d s d d c r c    , and (3.5) 2( ) sin t s d d r c     ; (3.6) where d t is the diameter of the tool, r s is the smallest surface curvatures in all dimensions, and c d is the curing depth. Therefore, for fabricating features on a given cylindrical and spherical surface with given surface curvature, the tool size d t can be selected by setting the related r s in the equations. 56 Figure 3.11: An illustration of the relationship between tool size and surface curvature. 3.5.2 Test of an angled tool for fabricating features on a vertical surface Materials can be accumulated in a variety of directions with the aid of tool rotations. However, the feasible rotation angle  max is limited for a straight tool as shown in the previous section. In order to build features on a surface that requires a large rotation angle (e.g. a vertical surface), a non-straight accumulation tool can be used in the fabrication process. As shown in Figure 3.12a, an angled accumulative tool was made for testing in our multi-tool and multi-axis CNC accumulation system. Similar to the straight tools as discussed before, the angled accumulation tool is also made of a fiber optics guide with a Teflon film coating. The only difference is that the tool head is bent by a certain angle (e.g. 45 o ). Tool head P 0 P 1 O c d d t r s  O P 3 P 2 r s d t (a) (b) 57 (a) Tool head with coating (b) physical model before repair (c) tool path (d) repaired physical model Figure 3.12: Test of an angled tool on repairing a feature on a vertical wall. A test case of repairing a slot on a vertical surface is performed to illustrate the capability of the CNC accumulation process on building features along the Y axis. Such accumulation process would be impossible for a straight tool as shown in Figure 3.9. In the test, a physical model is first built by a commercial SLA machine (refer to Figure 3.12b). The built cube has four slots on one side. Suppose the bottom one needs to be enclosed. The planned tool path based on the X and Y axis motions is shown in Figure 3.12c. The built physical model is shown in Figure 3.12d. Hence angled accumulation tools can enable building features on curved surfaces that are vertical or even facing downwards. 3.5.3 Test of accumulation tools with mask patterns In addition to different bending angles, an accumulation tool used in the CNC Accumulation process can have different mask patterns for curing various shapes. A simple test to illustrate the idea is shown in Figure 3.13a. A heart-shape physical mask applied on the end of the angled tool is shown in Figure 3.13b. By switching the physical mask with different shapes, the light pattern can easily be changed. Four physical masks with the shapes of star, square, heart and disk, were tested. Figure 3.13d shows the added textures on a vertical wall based on the angled tool with the Fixture Coating Optic Fibers Z X Y (a) (b) (c) (d) 58 four different masks. The thickness of the textures can be controlled by moving the tool with desired distances. Figure 3.13: Adding textures on a vertical plane using tools with different mask patterns. 3.6 Experimental Results and Discussions Four test cases have been performed to verify the capability of the CNC accumulation process in building features on various curved surfaces. The tool paths of the small and large tools that are used in the tests are denoted in the figures. (a) Planned tool paths (b) built object. Figure 3.14: A spiral curve pattern built by the small tool. 3.6.1 Test 1: a spiral curve on a flat surface A 2D spiral curve on a flat surface was first tested. In this test case, only the small tool was used in building the curve. The process only involves the X-Y translations. The planned tool path Fixture Coating & Mask Optic Fibers Working Plane Coating & Mask ★ ■ ● ★ ■ ● ★ ■ ● (a) (b) (c) (d) Path of small tool 1cm (a) (b) 59 is shown in Figure 3.14a. Accordingly the built physical object is shown in Figure 3.14b. In the test, different moving speeds were used due to the different lengths of the spiral ring. Based on our previous experiments, a speed of 1.8mm/sec was used in building the outermost ring, and a speed of 1.3 mm/sec was used in building the innermost ring. Based on the built result, the speed of the outside ring seems to be too large, which lead to some defects as shown in the figure. 3.6.2 Test 2: an inverse conic cup An inverse conical cup as shown in Figure 3.15a was built to verify the capability of the process in building 3D cylindrical surfaces. The thickness of the cup wall is 2.3mm. In the test, only the large tool was used in building the part. The planned tool paths are shown in Figure 3.15b. In order for the cup wall to achieve better surface quality, the tool was inclined by an angle  as defined in the conical cup. During the building process, the rotation A is used to define the tool inclination; the rotation C is used for the tool to cure a whole layer in the same height; and the translation Z move the tool to a desired height. As shown in Figure 3.15b, the base surface is built first by combining the rotation C and the translation X. In the test, a dent was also added on the cup wall. Its sizes are 1mm in height and 2.3mm in width. Accordingly, the layer thickness in the building process is set at 1 mm, except for the first layer. As shown in Figure 3.15c, the first layer of the cup wall is non-uniform due to its connection to the bottom surface. For achieving the wall thickness of 2.3mm, a moving speed of 0.4 mm/s was applied based on the test results as discussed in Section 3.4. The built physical object is shown in Figure 3.15d. Note that the surface finish of the cup wall is smooth even for such a large tool size and layer thickness (1mm was used in the test. In comparison, the layer thickness of the SLA process is ~0.1mm). This is mainly due to the tool orientation that was used in the fabrication process. 60 Figure 3.15: The building of an inverse conical cup. 3.6.3 Test 3: fluidic channels on a cylindrical surface A test case of building fluidic channels on a cylindrical surface was performed to illustrate the tool path planning based on multiple tools. As shown in Figure 3.16a, several fluidic chambers and channels are designed on a cylindrical surface. The main purpose of the fluidic channel design is to transport liquid from two chambers A and B into a mixing chamber C. P Path of big tool Path of big tool Θ 1 st Cured layer 5 th Cured layer Cross-section of cup wall Tool head Cup base 10 mm (a) (b) (c) (d) 61 Figure 3.16: Fluid channels on a cylindrical surface. Since the cylindrical base is used for supporting the fluid channels, its uniform thickness is important while its lateral resolution is not. In comparison, the chambers and channels will define how liquid flows on the cylindrical surface. Consequently their resolution will determine the complexity of the fluidic system that can be achieved. In the test, the wall thickness is set at 0.5mm (or 500µ m), and the smallest gap of the channels is set at 1.0mm. Accordingly, the large tool is selected in building the base surface while the small tool is used in building the fluidic chambers and channels. The planned tool paths are shown in Figure 3.16b, in which the paths of the large and small tools are shown with related notations. Θ A Cured layer by big tool Base Cured layer by small tool T1 T3 Small tool T2 Big tool (a) (b) (c) (d) big tool small tool base Cured by small tool Cured by big tool 62 The building processes using the small and large tools on a cylindrical surface are close to the ones as discussed in Section 3.5.1 and 3.5.2 respectively. The only difference is that the base surface and features are built by a combination of X, Y, Z translations and A rotation. An illustration of the cured layers in a cross-section plane is shown in Figure 3.16c. The gray cylinders represent the large and small tool heads. T 1 , T 2 and T 3 define the three moving directions involved in the building process. The cured areas based on the large and small tools are denoted in the figure. Accordingly the built physical object is shown in Figure 3.16d. The fine features of the walls and the coarse features of the base can be observed from the object since they were built by the small and large tools, respectively. All the fabricated features are conformed to the given cylindrical surface. 3.6.4 Test 4: rods on a spherical surface A test case of building rods on a spherical surface was performed to illustrate the capability of the CNC accumulation process on building features on a pre-fabricated surface. The CAD model of the test is shown in Figure 3.10a. The building process is the same as the one discussed in Section 3.5.2, except that the base surface is an existing spherical shell that is fabricated by another manufacturing process. The rods were built by using the small tool. The planned tool paths are shown in Figure 3.17a. The fabricated object is shown in Figure 3.17b. The diameter of the fabricated rods is around 1.2 mm. The fabricated rods have a good surface finish mainly because each rod is built by moving the tool along its axial direction continuously. No stair stepping effect is observed (refer to Figure 3.18b for a microscopic image). As discussed in previous sections, the diameter of a rod can be controlled by moving the tool in different speeds. Accordingly, non-uniform diameters along a rod can be achieved by dynamically adjusting the tool’s moving speed, which is also shown in the built physical object. 63 Figure 3.17: Rods on a spherical surface. (a) Planned tool path; (b) built physical object. (a) A CAD model of a tilted rod (b) Rod fabricated by CNC accumulation process (c) Rod fabricated by SLA (d) Rod fabricated by SLA using smaller layer and yellow resin Figure 3.18: A tilted rod fabricated by the CNC accumulation and SLA Processes. Path of big tool Path of small tool (a) (b) Z X Y D=0.7mm 45 0 H=10mm (b) 0.5mm l 0.5mm 0.5mm After removing supports 10m m 64 3.6.5 Comparison with layer-based approaches The surface finish of a tilted rod using the multi-axis CNC accumulation process is quantitatively compared with the same rod that is fabricated by the SLA process. A tilted rod taken from the built object and its microscopic image are shown in Figure 3.18b. A CAD model of a tilted rod with the same design parameters is constructed as shown in Figure 3.18a. Two setups, a research system[84] based on the same resin and a commercial system (Ultra TM from EnvisonTec Inc.) based on a different resin (SI500 from EnvisonTec Inc.), were used in fabricating the designed CAD model. For a layer thickness of 0.1mm and 0.05mm, the built physical objects and their microscopic images are shown in Figure 3.18c and d, respectively. Note that, in addition to the stair-stepping effect on the up-facing surface, both parts have extra materials on the bottom-facing surface due to the resin over-curing in the Z axis, especially for a resin that is more translucent. Some support structures are also added in order for the part to be built in the SLA process, which leave some marks on the bottom-facing surface. All the aforementioned effects make the built rods have much worse surface finish than the ones that are built by the multi-axis CNC accumulation process. The surface roughness of the fabricated up- facing surfaces is measured using a digital gauge mounted on a XZ linear stage. The measured results are shown in Figure 3.19, in which the microscopic images of the measured surfaces are also shown. In addition to better surface quality, the CNC accumulation process can build the tilted rod in a much shorter time as well. The rod length in the CAD model as shown in Figure 3.18a is 12mm. The fabrication time for the CNC accumulation process is 16 seconds using the moving velocity of 0.75mm/s. In comparison, the fabrication time using the mask-projection-based SLA process as shown in Figure 3.18c is 25 minutes by using a layer thickness of 0.1mm and a curing time of 1.2 seconds per layer). The fabrication time using the SLA process as shown in Figure 3.18d is 65 even much longer (60 minutes) by using a layer thickness of 0.05mm and a curing time of 9 seconds per layer. (a) Measurement results (b) Microscopic images of the measured surfaces Figure 3.19: Roughness measurements of built surfaces by SLA and CNC accumulation processes. 3.7 Concluding Remarks To fasten the energy delivery and hence the build speed, a multi-tool design is implemented in the CNC accumulation system. Accumulation tools with different sizes and shapes have been investigated with their parameter settings. The use of multiple accumulation tools improve the performance of CNC accumulation in fabricating features with different resolution requirements and faster build speed. In addition, based on the study of tool selection and multi-tool motion planning, features on curved surfaces can be fabricated with good surface finish, consistent material property and fast build speed. Using a developed prototype system, the building processes for common types of curved surfaces including flat, cylindrical and spherical surfaces have been tested. The experimental results illustrate that the multi-tool and multi-axis CNC accumulation process can be beneficial for building conformal features on curved surfaces. The hypothesis H1 is verified in this chapter: (a) (b) 0 0.5 1 1.5 2 2.5 3 -0.05 0 0.05 L (mm) D (mm) Surface Roughness Measurements M1 M2 CNCA M1 CNCA 0.2mm M2 0.2mm 0.2mm 66 H1. A multi-tool design and optimized process could make the energy delivery fast in CNC accumulation system, thus to improve the build speed and surface quality together for building on platform tasks in CNC accumulation system. Considerable work remains to mature the newly developed AM process. As shown in Figure 3.20, following work on fabricating accurate shapes and geometries by accurately controlling the energy delivery path and direction is presented in the following Chapter. Work on exploring the integration of a compact dual-axis 3D scanner, and new applications in building round inserts, will be discussed in the following Chapter. 8 Energy Control Material Deposition Fast Delivery Accurate Delivery Fast Deposition Accurate Deposition CNC Accumulation MIP-SL H1: Multi-tool, Process optimization (Chapter 3) √ √ √ ? H2 More complicated cases, building-around-inserts Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research Content: Figure 3.20: Research progress by this chapter and the following research work 67 Chapter 4 Accurate Energy Delivery in CNC Accumulation by Tool Path Planning 4.1 Challenges of Energy Control in CNC Accumulation After investigating the use of multiple tools with multi-axis motions to achieve fast energy delivery in CNC accumulation in the previous chapter, now it comes to how to make the energy delivery accurate in this chapter. Only together with the work presented in this chapter, we could be able to achieve fast fabrication with high surface quality in CNC accumulation for various build tasks, including the common tasks that builds objects directly on platform, as well as building around inserts which is almost impossible to finish by layer based AM processes. As mentioned in previous chapters, the CNC accumulation process has great similarity to the CNC machining process. As shown in Figure 4.1a, the CNC machining process uses a machining tool to remove the material that is in touch with the subtracting tool. Hence, for a given work piece (W) and tool path (Si) with tool orientation in each cutter location (Oj), the constructed shape (M) will be () ij SO M W T   . In contrast, the CNC accumulation process, as shown in Figure 4.1b, uses an accumulation tool to add material that is in touch with the curing tool. Hence, for tool path (Si) with tool orientation (Oj), the constructed shape (M) will be () ij SO MT   . 68 Figure 4.1: A comparison of (a) the CNC machining and (b) the CNC accumulation processes [40] Two good properties of the CNC accumulation process are: (1) The allowable motions between a tool and a work piece are significantly increased; and (2) the feeding of liquid resin to facilitate the material accumulation process is straightforward since an accumulation tool is merged inside liquid resin during the building process. As introduced in previous chapters, a very meaningful application of CNC accumulation system is building-around-inserts. However, a critical challenge is how to generate tool paths for an inserted component whose digital model and relative position are unknown. Tool path is the major information input for energy delivery control. With the planned tool path, the system is able to deliver the energy at the correct position with proper amount to successfully cure the material. Without a well-planned tool path, the building job is almost impossible to be finished successfully. To be more specifically, Figure 4.2 shows a test case based on a gear with a broken tooth. Suppose the plastic gear is fabricated by the injection molding or other AM processes. During its usage, one of the teeth is broken. Instead of throwing the part away, the broken gear can be repaired using the CNC accumulation process. However, the tool path planning for adding the desired tooth can be challenging since (1) the physical object to be repaired may be positioned (a) CNC Machining (b) CNC Accumulation Accumulation Tools (T) Machining Tools (T) Work Piece (W) Liquid Resin Tool Path (S i + O j) Tool Path (S i + O j) () ij SO M W T   () ij SO MT   69 arbitrarily on the platform; (2) the surface of the broken tooth, which will be used as the base surface for the missing tooth, can be irregular. Figure 4.2: A test case of building-around-inserts using the CNC accumulation processes. If successfully addressed the challenge of automating the tool path planning for any given objects to be built on, the capability of remanufacturing parts will not only reduce the material and energy waste, but also expand the functionality of product components. Additive manufacturing processes such as Tungsten Inert Gas (TIG) welding and Laser Engineered Net Shaping (LENS) have been developed before for repairing and modifying metallic parts; however, automatic repairing and modification of plastic parts are still lacking. Such product sustainability issues attract increasing attentions in recent years. We are motivated to develop an accurate energy control method for CNC accumulation process to address such needs. 4.2 Pipeline of an Integrated CNC Accumulation System To address the tool path issue for proper energy control, an in-situ scanning system that can capture the shape of inserted objects and related geometric processing methods have been developed for the multi-axis CNC accumulation process. Accordingly, a set of geometric processing methods, including the APSS point processing, LDNI-based surface retrieving and offsetting, tool path planning, and 5-axis accumulation motion control, have also been developed. With the assistance of the integrated scanning unit, the process is always able to plan and control 6.35mm An insert built by other processes Feature built around the insert CAD model 70 the energy delivery and input in any complicated fabrication cases, which enables it offering an effective and efficient fabrication solution for repairing and modifying plastic parts. Figure 4.3 shows the data flow pipeline of an integrated CNC accumulation system for building-around-inserts. The in situ scanner returns scanning data of a surface portion where material needs to be added. For the scanning result that is usually sparse, the algebraic point set surface (APSS) algorithm is adopted in processing the scanning points for constructing related surface models. Automatic tool path generation based on the scanning results is then performed. The tool path is then transformed to G-code. The movement of curing tool and the on/off states of the light power is directly controlled by the G-code. Figure 4.3: A pipeline of an integrated CNC accumulation system for automatic energy control The following section presents the in situ laser scanner based on a MEMS device, and the point processing algorithms for constructing the digital model of a scanning surface. Two test cases are also presented to show the successful energy control with the integrated 3D scanning unit. 4.3 Design and Development of an Integrated 3D Scanning Unit 4.3.1 Dual-axis mirror based 3D Scanning Unit A desired in-situ 3D scanner needs to be compact on size. It also needs to be flexible on accommodating the position and orientation of a mounted component in the tank. Also note that, 71 in order to use the CNC accumulation process in building-around-inserts, only the concerned surface model needs to be captured; there is no need to reconstruct the 3D model of the whole object. Current 3D scanning systems can be classified into contact and non-contact systems. The contact 3D scanners use a probe to physically touch the surface of an object. In addition to a firm fixture to hold the object, a motion system is needed to translate and orientate the touch probe related to the target surface. In comparison, non-contact 3D scanners emit radiation or light beams, and use detectors to receive the signals reflected from the object [85]. A triangulation method is one of the most commonly used 3D scanning approaches[86]. In the triangulation method, a prepared light pattern is projected onto a surface in order to capture the point positions on the surface. The light pattern can be a dots, lines, or grids. A camera or light detector is then used to capture the projected light. Consequently, the 3D coordinates of the projected light on the surface can be calculated by a triangulation model[86]. Figure 4.4 shows a typical 3D scanning head based on the triangulation method. Both the laser and sensor are fixed inside the head. Some scanners have a rotating hexagon mirror to convert a laser dot into a laser line[87]. However, the limited freedom of the laser scanning path requires the use of an external motion system to position the scanning head in order to scan a surface[87-90]. 72 Figure 4.4: An illustration of the triangulation method and a scanning head with a fixed laser and sensor. Micro-electo-mechanical system (MEMS) is a rapidly growing, emerging technology. Products such as direct digital micromirror (DMD) devices have been used in AM processes [84, 91]. Recently a dual-axis analog MEMS pointing mirror device (TALP1000B), has been developed by Texas Instruments (Dallas, TX) [92]. The gold coated mirror (~3.2× 3.6mm) of the MEMS device is constructed of single crystal silicon. It can be driven with precise pointing resolution. In addition, each rotation axis of the device is individually and independently controlled. Such a compact dual-axis rotation mirror enables a laser dot to scan in a certain range. Hence it is possible to cover a surface by using the controlled movement of such a micromirror instead of translating the scanning head that is much larger and more bulky. In this research, a 3D scanning unit based on the MEMS technology is developed. Figure 4.5 shows the schematic of the developed 3D scanning unit based on the dual-axis rotation mirror. A dot laser shoots a laser beam toward the mirror. The laser dot is then reflected by the mirror and shoots on the surface of a scanning object. By rotating the mirror along the NS and WE axes, the laser dot can scan any position within the scanning area. The scanning area that can be covered in one setup is 73 dependent on the maximum rotation angles of the mirror (± 5 o ). An appropriate scanning depth is determined by a calibration database. A camera is installed with the mirror to detect the laser dot reflected by the part surface. Based on the planned laser line and the captured image data, the 3D position of the intersection point can be calculated based on the aforementioned triangulation method. Figure 4.5: A schematic illustration of the dual-axis mirror based scanning unit. Figure 4.6 shows a prototype laser scanner that is integrated into a 5-axis CNC accumulation system. In the 3D laser scanning unit, (1) a red dot laser by Laser Glow Technologies (Toronto, Canada) is selected based on the wavelength range of the coated mirror (700nm – 10m). Its brightness and focus range can be adjusted. (2) The TALP1000B is a high- performance MEMS device designed for light steering applications. The rotation range of the two-axis mirror is ± 5 degrees and the mirror size is about 9mm 2 . The switching time of the mirror is less than 5ms. The device provides position feedback with 13-bit resolution. (3) A small web camera with an image resolution of 640× 480 is used as the light sensor. Working piece Scanning area Dual-axis Mirror Small Camera Dot Laser 1 2 3 4 7 x y 2 7 E N S W Laser line Laser line Dual-axis Mirror Scanning Area 4 8 74 Figure 4.6: 3D scanning unit with the dual-axis mirror in the 5-axis CNC accumulation system The dot laser and the dual-axis mirror are installed on the two ends of an aluminum bar. The laser dot reflected by the mirror falls on the platform, which is covered by a chessboard paper. Camera is installed side away from the mirror to capture the image of the reflected laser dot. Other than the laser scanning unit, there are three translation stages X, Y and Z, and two rotation stages A and B. These five stages are designed for the 5-axis accumulation tool. A control board is used to control the 5-axis movements. Based on such an integrated system, a given workpiece is first fixed in a resin tank that is placed on the platform. The laser scanning unit is then used to collect the scanning data related to the deposition surface. With the scanning data, tool paths for the accumulation process are computed using the point processing algorithm and the tool path generation operations. The accumulation tool, driven by the 5-axis motion stage, then adds material on the inserted workpiece. Camera Mirror Dot Laser Accumulative Tool Control Board Platform Power Supply 18mm 30mm Accumulation tool Dual-axis mirror 75 The developed scanning unit can be mounted on a separate flexible arm. The unit can also be fixed on the motion system that is designed for the CNC accumulation tool. Consequently, the workpiece can be moved up and down using the Z stage to enlarge the scanning volume in the Z direction, and moved sideways using the X and Y stages to enlarge the lateral scanning area. For a fixed setup, as shown in Figure 4.5, the scanning area depends on the rotation angle of the dual-axis mirror, and the distance between mirror and the scanning plane. In addition, the camera image range needs to be larger than the laser scanning area. In the prototype system as shown in Figure 4.6, the defined scanning range has a Z height of 25.4mm. On the top plane where the Z value is equal to 25.4mm, the scanning area is set at 25× 25mm to ensure sufficient accuracy. If a surface is outside of the scanning range, either the scanner or the workpiece needs to be moved. Accordingly, the related transforming matrix can be applied to the scanning result. 4.3.2 Calibration of 2D Camera Camera calibration is required for identifying the intrinsic and extrinsic parameters of a CCD camera. Accordingly, the relation between the captured 2D image and the 3D position of a point P can be established. The camera calibration is a time-consuming task despite the availability of some camera calibration toolbox. For example, Zhang [93] introduced a camera calibration method, which is used in the OpenCV image processing library and many other toolboxes. However, the calibrated camera parameters and related computation models cannot fully capture the nonlinearity in a working volume. To capture such nonlinearity for improved accuracy, the working volume (refer to Figure 4.5) is subdivided into 11 layers in our study with a distance of 2.54mm between two neighboring layers. A printed chessboard is fixed on the platform, which can be moved in the Z axis. 76 Accordingly the 11 layers from z=0mm to z= 25.4mm with z = 2.54 mm can be calibrated separately. For a layer K at height z w , all the corners on the chessboard have known coordinates (x w , y w , z w ). An image of the chessboard is captured. Based on it, the image coordinates (X i , Y i ) of all the corners can be identified. Accordingly the relationship between an image coordinate of each layer and the related world coordinate can be established as: _ ( , ) ( , , ) i i Layer K w w w X Y P x y z  . By repeating the process, a database between the 3D points within the working volume and the image points that are captured by the camera can be built as: _ ( , ) ( , , ),0 10. i i Layer K K w w X Y P x y K z K     (4.1) Such relations combine both intrinsic and extrinsic matrixes of the camera. Since the layer distance is small, a bilinear interpolation approach is used to approximate the world coordinates (x, y, z) based on two neighboring calibrated planes P K (x, y, K× z) and P K+1 (x, y, (K+1)× z), when [ ,( 1) ] z K z K z     , and the four corners of a checker box. 4.3.3 Calibration of Laser Scanning Lines The rotation angles of the mirror are controlled by the output values (NS and WE) provided by two Digital-Analog-Converters (DAC) in a control board. To eliminate the DAC output noises and other hardware construction errors, the laser line positions for different signal values also need to be calibrated. Different from the camera calibration, the reflected laser line is straight from the top to bottom planes. Consequently, a signal value (NS, WE) can be calibrated by the laser dots on the top and bottom planes that define a laser line. As shown in Figure 4.7, suppose (x i_10 , y i_10 ) and (x i_0 , y i_0 ) are the image position of a laser dot on the top and bottom planes. The world positions of the 77 laser points (X w_10 , Y w_10 , Z w_10 ) and (X w_0 , Y w_0 , Z w_0 ) can uniquely define the laser line inside the scanning volume. Figure 4.7: Laser calibration using the top plane (layer 10) and the bottom plane (layer 0). A database consists of the positions of all the laser lines and the related DAC signal values can be built as follows: f (NS, WE)= (X w_0 , X w_10 , Y w_0 , Y w_10, 0, 10× z), where NS and WE are the DAC numbers of the two axes of the mirror. (X w_0 , Y w_0 ) and (X w_10 , Y w_10 ) are the related world positions of the laser line on the bottom and top planes respectively. Since a related laser line is straight, an intersection point at a level Z can be computed by the linear interpolation between two related points L 0 , and L 10 . 4.3.4 Computing 3D Points In the scanning process, the dual-axis mirror is rotated based on a given DAC number (NS, WE). The laser is first turn off to capture an image of the scanned object (without the laser dot). The laser is then turned on for capturing an image with the laser dot. Consequently, an image X Z Y N S W E Dual-axis mirror (xi_10,yi_10,Xw_10,Yw_10,25) (xi_0,yi_0,Xw_0,Yw_0,0) (ns, we) Dot Laser Control Board 78 with the laser dot can be obtained by the subtraction between the two images. Based on the calibration methods the 3D computation is modified from the standard triangulation process. Figure 4.8: Point retrieval by calculating the closest point Let (X i , Y i ) denotes the image position of the laser dot. Accordingly, a set of world coordinate points P K (x w , y w , z w ) related to (X i , Y i ) for each layer K× z can be identified from the camera calibration database (refer to Equation 1). Ten camera line segments P i P i+1 can be computed by linear interpolation. Similarly, the corresponding laser line related to the current DAC number (NS, WE) can be identified through the laser point calibration database. Accordingly, the laser points L i L i+1 can be calculated by linear interpolation. As shown in Figure 4.8, the eleven layers divide the working volume into ten regions. Each region contains a camera line segment and a laser line segment. The closest point between the two line segments can be calculated in each region. For example, X 0 and X 10 are the closest point in regions Layer0-1 and Layer9-10, respectively. The point with the smallest distance is selected as the object point in the 3D space. For example, as shown in Figure 4.8, X 4 in region Layer4-5 would be the computed 3D position. 4.3.5 3D Scanning Verification Scanning tests are performed to verify the accuracy of the developed scanning system. In the first test, the platform was raised by the z-stage to different Z heights (0.0mm, 2.54mm, and P0 P20 P21 P40 P41 P60 P61 P80 P381 P400 P N : (X W ,Y W ,Z W ) (X D ,Y D ) T0 T19 T380 T399 P401 P420 Grids: 20× 20 DAC Step: 20 Layer 9 Layer 7 Layer 8 Layer 6 Layer 5 Layer 3 Layer 4 Layer 2 Layer 1 Layer 0 Layer 10 Camera Line (xi, yi) Laser Line (ns, we) P5 P6 P7 P8 P9 P10 L10 L9 L8 L4 L7 L3 L5 L6 L2 L0 L1 P4 P3 P2 P1 P0 X4 X10 X0 79 5.08mm). The planar surface is scanned and reconstructed based on the calibration databases. In the second test, a mechanical part with a height of 3.8mm was positioned at a height of 3.9mm. Its top surface is a planar plane painted in black. The scanning results of its top surface are recorded. Figure 4.9a is the computed scanning points related to the average plane. Figure 4.9b shows the deviation of all the points in the Z axis. (a) Distribution of scanning points (b) deviation of all the scanning points on Z axis. Figure 4.9: Scanning results of the top surface of a mechanical part. Table 4.1: Statistical analysis of scanning tests Test Median (mm) AVG (mm) Max (mm) Min (mm) VAR (mm) 1.Platform (z = 0) 0.202 0.203 0.487 0.0 0.0075 2. Platform (z = 2.54) 2.814 2.816 3.319 2.515 0.014 3. A part (z = 3.9) 4.148 4.146 4.393 3.844 0.007 4. Platform (z = 5.08) 5.080 5.072 5.080 4.934 0.0005 Table 4.1 presents the statistics of the scanning results, the computed point Z heights, in the verification study. The average height deviations of the computed 3D points in the four test cases are less than 0.3mm. The scanning errors may come from the sources such as the resolution of ~3.8mm 80 the webcam, the control resolutions of the dual-axis mirror, and the errors in the line segment approximation and the laser dot computation. They are discussed in more details as follows. (1) A low-cost webcam is used in our prototype system. The camera has a limited resolution (640× 480 pixels). In our setup, it is about 0.18mm/pixel. In addition, the brightest image position of the laser dot is used to represent the laser dot. Both camera resolution and the image point identification may introduce errors in the final scanning result. (2) Two digital analog convertors (DAC) are used in controlling the voltage and current of the coils that drive the dual-axis mirror. The resolution of the DAC in our control board limits the rotation resolution of the mirror. In addition, there may be vibration and other mechanical noises in the hardware setup. (3) The laser dot used in our system has a certain focus distance. The spot size slightly changes from the top to bottom planes. In addition, the linear interpolation between two neighboring camera points may also introduce errors. Approaches, such as adopting a better camera, using more layers in building the calibration database, and computing 3D points based on non-linear interpolation, can be developed to further reduce the scanning error. Nevertheless, the experimental results show that the scanning error of the developed 3D scanner is generally less than 0.5mm in the Z axis. Such accuracy is satisfactory for the CNC accumulation process. 4.4 Processing of Scanning Points The scanning results obtained by the aforementioned point-by-point based 3D scanner have the following properties. (1) The sampling points are relatively sparse compared to other scanning systems such as the structured light based systems. Instead of generating sampling points that are dense with various noises and outliers, a small number of key points on the surface 81 are sampled for surface reconstruction. (2) The point clouds of a surface may have uneven densities in some areas. (3) The scanning result can be a portion of a surface. (4) The scanning results may be distorted due to measurement noises. (5) A base surface to be scanned for the CNC accumulation process is usually continuous since the accumulation tool path is usually a set of continuous curves. Hence a 3D surface reconstruction method that can process sparse sampling points of a continuous surface is required in the related point processing software system. The algebraic point set surfaces (APSS) algorithm proposed by Gross et.al. [94, 95] is a good choice in handling high curvatures and sparse point cloud [96, 97]. In this paper, we extended the APSS algorithm and used the layered depth-normal image (LDNI) representation [98] in generating triangular meshes of the scanned surface. Figure 4.10 shows the framework of the developed point processing method. The scanning data computing process is divided into the following steps: (1) preprocessing points including denoising, normal estimation, and orientation propagation; (2) constructing an implicit surface based on the APSS projection; and (3) generating sampling points of the implicit surface for constructing triangular meshes. Each step is described in details as follows. 82 Figure 4.10: Overview of the point processing framework. Normal estimation and orientation play an important role in the mesh surface reconstruction. For a given set of sampling points, normal is chosen as the Eigen-vector corresponding to the smallest Eigen-value. It can be computed by the co-variant matrix of points: '( ) ( )( ) i T ii p N p p p p p    (4.2) where N’(p) is the point set of the k nearest neighbors of a query point p. Z X Normal Direction Estimation Z X Normal Orientation Propagation X LDNI Arrays Projection using APSS Z Y Y Y c x Pi Pi Pj X LDNI Arrays Projection using APSS Z Y c x Z X Normal Direction Estimation Z X Normal Orientation Propagation X LDNI Arrays Projection using APSS Z Y Y Y c x Pi Pi Pj X LDNI Arrays Projection using APSS Z Y c x (a) (b) (c) (d) 83 However, the Eigen-vector analysis cannot guarantee a correct orientation. Re-orienting the normal vectors is necessary based on a direct flipping strategy. The orientation of pi is flipped by the closest point: (4.3) Algebraic Point Set Surfaces algorithm is a newly developed point set surface definition based on moving least squares (MLS) fitting of algebraic spheres. The surface representation can be expressed by either a projection procedure or in implicit form. Compared to existing planar MLS, the key advantages of the APSS approach include significantly improved stability of the projection under low sampling rates and in the presence of high curvature. The method can approximate or interpolate the input point set and naturally handle planar point clouds. The following weighting function is used in the fitting process[94]: || || ( ) ( ) () i i i px w x f hx   , ( ) , 24 (1 ) ; 1 () 0; . xx fx otherwise     (4.4) where pi is a point in the input point set and x is a query point that is near the surface; () i wx represents the weight of the point pi for the query point x; () i hx is defined by a constant factor K timing the average distance D of pi’s k nearest neighbors. The weighting function f is a smooth and decreasing function. The weighting function proposed in [94] is used in our study. An algebraic sphere is defined as the 0-isosurface of the scalar field ( ) [ ] , where 01 [ ,..., ] T d V v v   is the vector of scalar coefficients to describe the sphere. For 1 d v  ≠ 0, the corresponding center O and radius R can easily be computed[94]: 84 1 1 1 [ ,..., ] 2 T d d O v v v   , and 01 /; T d R O O v v   (4.5) with d being the dimension. In degenerate cases, V corresponds to the coefficients of a plane equation with v0 representing the plane’s distance from the origin and [v1, …, vd] T being its normal. With normal constraints, V(x) can be computed by the following equations [94]: ( ) ( ) ̂ ( ) (4.6) ( ) ( ) ̂ ( ) ( ) (4.7) ( ) [ ]; [ ] [ ] (4.8) where {e k }time represent the unit basis vectors of the coordinate system. Scalar β should be carefully chosen to weight the normal constraints. In our study, it is calculated based on: 6 ( ) ( ) 10 () ii i i i w x h x wx    (4.9) The LDNI representation[98] is selected for converting the implicitly defined APSS surface into triangular meshes. An “almost” or thogonal projection[99] is used in computing LDNI points. That is, the coordinate of a query point used in the LDNI projection procedure is calculated as following: 85 0 ' min (0) ; x x Ext I x    (4.10) 1 ' min (1) ; y x Ext I y    (4.11) 2 ' min (2) ; x Ext i z    (4.12) (max (2) min (2)) / ( 1) z ceil Ext Ext s     ( ) / n x O r  (4.13) where x, y denote resolutions in the X and Y directions, respectively. min (0) Ext , min (1) Ext , and min (2) Ext represent the minimum value of the X, Y, and Z coordinates of the input point set. max (2) Ext represents the maximum value of the Z coordinate of the input point set. Ix, Iy are the index of the X and Y axis, respectively. S is the number of points per ray, and i is an integer between 0 and S. Figure 4.11 shows the LDNI-based APSS projection algorithm. 86 Figure 4.11: Framework of the APSS fitting using the LDNI array points. Input: (1) A non-empty point set P={X, n}, which contained preprocessed scanning points X, and its normal vector n. Both X and n are 3 dimensional vectors; (2) LDNI resolutions ∆x, ∆y in the X and Y directions, respectively; and the number of LDNI query points per ray S; Output: A set of points and normals (Ps and Ns) that describe a projected surface. Steps: (1) A point set P’={X’}, where X’ is a 3 dimensional query point calculated using Equations (7) to (9). Set K be the number of points in P’. (2) Set Ps={}; Ns={}; ε =0.000001; (3) j=0; (4) Repeat until j = K (4.1) set X’= P’[j]; (4.2) Compute V(X’) using Equation (5); (4.3) If ( v 4 < ε) go to plane fitting (4.3.1); else go to sphere fitting (4.3.2) (4.3.1) Plane fitting: (4.3.1.1) p(X’)= ( v 0 ×v 1 , v 0 ×v 2 , v 0 ×v 3 ); (4.3.1.2) n(X’)= (v 1 , v 2 , v 3 ); normalize n(X’); (4.3.1.3) X” = X’  n(X’) T × (p(X’)  X’) × n(X’); (4.3.1.4) If ||n(X’) T × (p(X’)  X”)||<ε, save X” to Ps and n(X’) to Ns; else, go back to (4.2). (4.3.2) Sphere fitting: (4.3.2.1) Compute O and R using Equation (4); (4.3.2.2) X”= (R/||X’  O||) × (X’ O) + O; (4.3.2.3) If ||X”  X’||< ε, compute n(X”) using Equation (10), save X” to Ps and n(X”) to Ns. else, go back to (4.2). (4.4) j = j+1; (5) output Ps and Ns. 87 3D Model March3 CNC control application G-Code CNC Accumulation Controller X, Y, Z, A, B axes motion & LED on/off Figure 4.12: Tool path generation and CNC controller. 4.5 CNC Accumulation Tool Path Planning Based on the scanned surface processed after APSS fitting and the CAD model of to-be- added geometry, a general tool path planning method for the 5-axis CNC accumulation processes consists of the following major steps. (1) Thicken the scanned surface by using an offsetting operation with an offset distance determined by the curing depth of the tool. A general offsetting method for an input polygonal model and an arbitrary offset distance is presented in[100] and used in our study. (2) Subtract the target geometry and the offset model to compute the shape of a thin layer. Robust Boolean operations based on the Layered Depth-Normal Image (LDNI) are described in[98] and used in our study. (3) Generate a set of contours to fill the computed layer shape with an interval distance defined by the curing spot size. (4) Use an off-the-shelf CNC machining software system to generate numeric control G-code based on the planned tool path. (5) Use a CNC accumulation controller to load the generated G-code, and accordingly send motion 88 commands to a controller to drive the 5-axis motion. At the same time, a micro-controller is used to control the curing state (i.e. turning the LED ON/OFF). Figure 4.12 illustrates the aforementioned steps. Two test cases were designed to verify the effectiveness of the integrated CNC accumulation system in building-around-inserts. 4.6 Experiment Results and Analysis of Energy Control in CNC Accumulation System Based on the planned tool paths, accumulation tools can selectively cure liquid resin into desired shape on the scanned surface. Two test cases were designed to verify the effectiveness of the integrated CNC accumulation system in building-around-inserts. (a) A schematic drawing (b) an accumulation tool used in the system Figure 4.13: Multi-axis motion configuration of the CNC accumulation system. The hardware setup of our prototype integrated CNC accumulation system is shown in Figure 4.6 and 4.13. A light guide is used to provide UV light from a LED to the tip of the fiber optics. A Teflon film is applied on the tool tip as the coating media[40]. A 5-axis motion system including X, Y, Z translations and A, B rotations is designed to allow the tool to cure resin in various positions and orientations, as shown in Figure 4.13. Z X Y A B Resin Tank Accumulative Tool Camera 5-axis CNC Accumulation Accumulation Tool Teflon film Tool Newly cured resin portion Previously cured resin portion 89 During the building process, an inserted component is mounted on a resin tank that is placed on the platform. The surface to be used as the base is first scanned by the developed 3D scanner. The tank is then filled with UV curable resin. The building process can start after the tool paths have been planned. After the accumulation process finishes, the built object is taken out for cleaning. A test case on repairing an existing plastic part is shown in Figures 4.14 and 4.15. In the test, a gear was built using a commercial SLA machine. We manually break one tooth of the gear (refer to Figure 4.14a). Note that the broken surface on which material will be deposited is irregular. With the integrated CNC accumulation system, it is possible to repair the broken tooth. Figure 4.14: Surface reconstruction for the CNC accumulation process. First, the working surface, on which a new tooth will be built, is scanned. As shown in Figure 4.14b, the scanning point set is sparse with noises. The scanning points are processed using the point processing algorithms as described in previous sections. The computed LDNI point set after the point processing step is shown in Figure 4.14c. Finally a triangular mesh as shown in Figure 4.14d is constructed from the LDNI model. 2 (a) Broken gear (b) Scanned data Scanning Area (c) LDNI Model (d) Mesh Model 90 Assuming the scanned surface has been aligned with the target CAD model (refer to Figure 4.15a). A set of offsetting operations can be performed based on the scanned surface and a given curing depth. Accordingly multiple layers can be computed based on the offset models and the target tooth model. As shown in Figure 4.15b, there are 15 layers to be built in order to repair the missed tooth. Note all the layers have a uniform thickness. For each layer, continuous tool paths can be planned based on the layer model as shown in Figure 4.15b. Figure 4.15c shows the tool path of one layer in the CNC accumulation process. Note that the planned tool path is not on a planar plane, which requires 5-axis motions. Based on the planned tool paths, the repairing work of the broken gear can be performed by the CNC accumulation process. Figure 4.15d shows the repaired gear. For demonstration purpose, a different resin is used in the CNC accumulation process. If the same type of resin was used, the repaired tooth would be identical to other teeth. Figure 4.15: Fabrication process based on the scanned gear surface. As discussed in previous section, each manufacturing process has its own advantages and disadvantages. For example, complex geometries such as surface textures and characters are difficult to be fabricated by forming processes such as injection molding and extrusion. Such complex features will significantly increase the tooling cost in the forming processes. Assume a (a) Alignment (c) Tool Path (b) Offsetting (d) Repaired part 91 teapot with no surface textures or characters is fabricated by the injection molding process (refer to Figure 4.16a). In the test, the model was still fabricated by a SLA machine due to the lack of injection molding tools. However, such simple shape can easily be fabricated by the forming process in a much shorter time and at a much lower cost. Further assume the final design needs to have complex features such as 3D textures, which are difficult to be injection molded. In order to add such features, the integrated CNC accumulation system can be used in modifying an existing part that is fabricated by the forming processes. Suppose some characters such as “USC” are required on the exterior surface of the teapot. Similar to the previous test case, the portion of the surface where the new feature will be added is scanned. The scanning result is shown in Figure 4.16b. The scanning points are then processed by the point processing algorithms as described in previous section. A much smoother and denser LDNI point set is generated as shown in Figure 4.16c. A mesh surface can be constructed from the LDNI points (refer to Figure 4.16d). Assume the CAD model of “USC” pattern and its position on the scanned surface are known (refer to Figure 4.17a). Accordingly related layers of the "USC" pattern can be computed by offsetting the pattern surface (refer to Figures 4.17b). The corresponding tool paths for each layer can be generated as shown in Figure 4.17c. The teapot modified by the integrated CNC accumulation system is shown in Figure 4.17d. Note that, in such a component, two different manufacturing processes have been used in its fabrication. In addition, each process utilizes its unique properties: the injection molding processes is fast and low-cost; while the CNC accumulation process is capable of fabricating complex shapes on arbitrary surfaces. 92 Figure 4.16: Scanning and post processing of the scanned teapot surface. Figure 4.17: The tool path and modified teapot. Experimental results illustrate that the novel 3D laser scanning system gives the CNC accumulation system good control on the energy delivery position and path, and the integrated CNC accumulation system has the capability on building-around-inserts. Its applications on (a) Teapot (b) Scanned points (c) LDNI model (d) Mesh surface (a) Designed model (b) One layer (c) Tool path (d) Fabricated part 93 plastic part repairing and modification have been demonstrated, which may provide a fabrication option for future sustainable manufacturing and heterogeneous component design. 4.7 Concluding Remarks An integrated CNC accumulation process has been developed based on a novel 3D laser scanning system that uses a dual-axis mirror. Compared to conventional laser scanning systems, the use of the MEMS device reduces the size and increases the scanning freedom of the system from one-dimensional line segment to a two-dimensional surface area. Related point processing algorithms have also been developed based on the APSS fitting and the LDNI representation. Experiments verify the accuracy and compactness of the 3D scanning system. Accordingly, a 5- axis CNC accumulation system with the in situ 3D scanner has been developed. With the automated generated tool path, the multi-axis CNC accumulation system is able to place the accumulation tool on the accurate place with a proper orientation, hence deliver the energy accurately, which is critical for complicated job tasks like building-around-inserts. Its applications on plastic part repairing and modification have been demonstrated, which may provide a fabrication option for future sustainable manufacturing and heterogeneous component design. The hypothesis H2 is verified in this chapter: H2. A 3D surface reconstruction approach and automatic optimal tool path generation could make the energy delivery accurate in CNC accumulation system, thus to improve the build speed and surface quality together for building around inserts tasks in CNC accumulation system. Together with the work presented in the previous chapter, it is able to completely solve the primary research questions in the CNC accumulation systems for all types of build tasks: 94 Q1. How to achieve high build speed together with high surface quality in photo- polymerization based additive manufacturing processes? As shown in Figure 4.18, after exploring the non-layer based photopolymerization AM process, we will investigate the MIP-SL system in the following chapters, in which the energy control and material depositions are very different. Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research Content: 8 Energy Control Material Deposition Fast Delivery Accurate Delivery Fast Deposition Accurate Deposition CNC Accumulation MIP-SL H2: 3D surface reconstruction, Optimal tool path (Chapter 4) H1: Multi-tool, Process optimization (Chapter 3) √ √ √ ? H4 ? H5 ? H3 Chapter 5,6,7 Figure 4.18: Research progress by this chapter and the following research work 95 Chapter 5 Accurate Energy Delivery in MIP-SL using Gray Scale Approach 5.1 Energy Control Principle of Mask Image Projection based Stereolithography The energy input in Mask Image Projection based Stereolithography (MIP-SL) process is a light beam patterned by mask image which is generated by the DMD chip. Since the light energy is delivered as a pattern to solidify one layer in a shot ( in micro seconds to seconds), this area- processing energy delivery manner is already very fast compared to the point-wise or line-wise energy delivery methods. So there is no research needs to study fast energy delivery method in MIP-SL systems. The energy input for curing one unit area is determined by the light source power, mask image pattern and projection time. By adjusting any one of these three factors, the energy input amount could be changed and thus the curing shape would be changed. In an existing MIP-SL system, the light source is usually fixed. Besides, because MIP-SL system is an area-processed approach, the projection time for each pixel in one layer is also consistent. Therefore, in order to make varying energy input within one layer, the only factor that we can play with is the mask image pattern. If different energy input amount can be assigned for different pixels within the same layer, it means that varying thickness can be cured within one layer and thus the stair- stepping effect can be eliminated. 5.2 Principles of Gray-Scale Approach Motivated by the above idea of the optimized mask image projection, a gray scale mask image method has been developed to improve the surface finish of down-facing surfaces. As extensively studies in the SLA process[17], the critical threshold (Ec) is a primary parameter for a 96 given photosensitive resin. The liquid resin will be solidified when the exposed energy is bigger than Ec; otherwise, the resin will remain as liquid or gel. In addition, the curing depth (Cd) is directly related to the energy input E if E > Ec. The dependence of Cd as a function of the exposure E is generally log-linear[1]. Their relation follows the equation: ln( ) dp c E CD E  (5.1) One way to control E is to adjust the exposure time for a binary projection image. Another approach is to adjust pixels’ gray scale values while fixing the exposure time of the projection image. In comparison, the gray scale image approach is easier to control. Hence, in the MIP-SL process, full light intensity will be projected for a pixel of a mask image if its gray scale value g = 255; when 0 < g < 255, only partial light intensity will be projected for curing the resin area related to the pixel. Corresponding to a given Ec and an exposure time T, a threshold gray scale value gc can be identified. An illustration of the gray scale image method is shown in Figure 5.1. Suppose gc = 75. Hence the exposures at pixels B, C and D whose gray scale values are set as 102, 153 and 204 respectively will lead to cured shapes with different cure depth. No resin will be cured at pixel A due to its insufficient energy input. In addition, the cure depth of a pixel is required to be bigger than a layer thickness such than sufficient over-cure exists. Hence the current layer can be bonded to the previous layers. 97 Figure 5.1: Principle of the gray scale image method For the liquid resin and the light source used in our testbed, designed experiments have been performed to calibrate the relations between the cure depth and the gray scale values. The CAD model of a designed part is shown in Figure 5.2a. In the test, the exposure time was set at 3 seconds. Mask images with different gray scale values were used in curing the top layer. The part was then taken out and the related cured layer thickness Cd was measured. The recorded Cd values for different gray scale values are shown in Figure 5.2a. The cure depth increases with the set gray scale value. In addition, the gray scale value threshold gc is around 76 when T = 3 seconds. For a pixel with g < gc, liquid resin was not cured or the cure depth was too small for a given layer thickness (i.e. < 0.1mm). The relation between Cd and ln(g) has also been computed as shown in Figure 5.2(b). The relation matches Equation 5.2 quite well and can be approximated by the following equation: 0.521*ln( ) 2.252, ; () 0, c c g g g Fx gg       (5.2) Critical exposure E c Z X Resin Surface Cure depth Layer n Layer n-1 Over Cure Pixel B g=102 Pixel A g=51 Pixel D g=204 Pixel C g=153 E(102) E(153) E(204) Cured geometry Layer thickness 98 Figure 5.2: Relation between the cure depth and the gray scale value for SI 500 resin Hence by slightly changing the gray scale values of neighboring pixels in a projected mask image, the cure depth related to the pixels can be gradually changed. Thus a higher Z resolution of down-facing surfaces can be achieved, which will lead to improved surface finish. In the process planning for a given 3D model with down-facing surfaces, the cure depth at each pixel needs to be accurately controlled by adjusting its gray scale value. In addition, an upward Z offsetting distance needs to be considered since the range of achievable cure depths (i.e. 0.1mm < d < 0.65mm) is bigger than a layer thickness to ensure the over-cure between layers. 5.3 Process Planning using Gray-Scale Approach The schematic illustration of the traditional and the gray scale image methods are shown in Figure 5.3. Several layers are reserved as compensation for over-cure. In the traditional method, a set of layers are sliced based on the Z offsetting result. Accordingly binary mask images are used in the MIP-SL process. In the gray scale image method, a compensation Z thickness is calculated based on the minimum and maximum curing depth for gray scale pixels (i.e. 0.1mm and 0.65mm respectively in our test). Hence the desired thickness in each layer is bigger than the minimum achievable curing depth and smaller than the maximum achievable curing depth. Based on the Z offsetting result and the established relation between curing depth and the gray scale 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 255 239 223 207 191 175 159 143 128 112 96 Gray Scale Value Cure Depth(mm) Cure Depth 4.5 5 5.5 0 0.1 0.2 0.3 0.4 0.5 0.6 Ln (Gray Scale Value g) Cure Depth (mm) Figure 8. Working curve for SI 500 resin: the relationship between gray scale value and cure depth with exposure time = 3 s. Physical Cure Depth Fitted Cure Depth A Top layer A (a) Experiments on c d and g values (b) Relation of c d and ln(g) values Cure depth (mm) Gray scale values 99 values, a mask image with desired gray scale values at all the pixels can be computed. The mask image can then be used in the MIP-AM process for building smooth down-facing surfaces. Figure 5.4a shows the related algorithm for generating gray scale mask images. An example is shown in Figure 5.4b to illustrate the two main steps, Z offsetting and gray scale image setting. Figure 5.3: Comparison between the traditional and the gray scale image method. Step 2. Mask image planning Binary Image Step 1. Image Generation Step 2. Building Traditional Method Gray Scale Image Method Step 1. Modeling 0, g < g c F(x) = 1 2 1 2 3 X Z Input geometry Z Cured Shape Slicing plane Y X Z X Cured layer    0.521× ln(g) – 2.252, g > g c Z Cured Shape Slicing plane Gray scale Image Step 3. Building Y X 4 100 Figure 5.4: A gray scale image generation algorithm with an illustration example. 5.4 Verification in Top-down Projection based MIP-SL A prototype system has been built for verifying the presented gray scale image approach for fabricating smooth down-facing surface. The hardware setup of the developed MIP-SL system is shown in Figure 5.5a. In the designed system, an off-the-shelf projector (Acer H5360) was used as the projection device. The use of a commercial projector can significantly reduce the prototype cost and simplify the system design. The optical lenses of the projector were modified to reduce the projection distance. Various projection settings including focus, key stone rectification, brightness and contrast were adjusted to achieve a sharp projection image on the designed projection plane. The DMD resolution in our system is 1024× 768 and the envelope size is set at 110× 82 mm. Two linear motion stages from Thomson Industries Inc (Radford, VA) are used in the system. One is used as the Z elevator for lifting the build platform; another is used as the sweeper for moving a blade to flatten the resin surface. A high performance 8-axis motion control Compute sampling points & normals Calculate Z offset d for all the down-facing points Mask image planning CAD model Cure depth range Layer thickenss Relation of cure depth and gray scale value Offset boundary points and normals Gray scale mask images Boundary points and normals Max C d Δd Z C d1 X g 1 Layer Thickness C d3 g 5 g 4 g 3 g 2 (a) Main steps in the gray scale image method (b) An illustration example Input boundary Z offset boundary Slicing planes Gray scale mask image Portion to be cured 101 board KFLOP+SnapAmp 1000 (Dynomotion Inc., Calabasas, CA) was used for driving the linear stages. Commercially available photo-curable resin (envisonTEC SI500, Ferndale, MI) was used in the experiments. The exposure time was set at 3 second based on the curing depth analysis. A mask imaging planning and control software system has been developed by using the C++ programming language with Microsoft Visual C++ compiler. The graphical user interface (GUI) of the developed software system is shown in Figure 5.5b. The system can load in a STL model, perform process planning and synchronize the mask image projection with desired linear stage motions. Figure 5.5: Hardware and software setup of the developed MIP-SL testbed. As shown in Figure 5.6a, a CAD model with a down-facing surface was used in this test. The same process parameters were used in building the model by using the traditional and newly developed methods. To generate gray scale images for building the part, the CAD model was offset by 0.3mm. Accordingly the gray scale value at each pixel was calculated based on the (a) (b) Resin tank 1 DLP projector Z Stage Reflecting mirror Motion controller Joy stick Building Platform (inside resin) Sweeper Power supply Control panel Motion parameters Building process report Geometry slicing Geometry (a) (b) Resin tank 1 DLP projector Z Stage Reflecting mirror Motion controller Joy stick Building Platform (inside resin) Sweeper Power supply Control panel Motion parameters Building process report Geometry slicing Geometry (a) (b) 102 corresponding curing depth. Consequently the computed gray scale images can cure a layer of liquid resin with uneven thicknesses. The related build results M1 and M2 are shown in Figure 5.6a. The Z height values of the down-facing surface are shown in Figure 5.6b. The microscopic images of a portion of M1 and M2 are shown in Figure 5.6c. It is obvious from the images that the traditional method can only cure layers with an increase on the given layer thickness. In comparison, the gray scale image method can achieve varying Z heights within one layer. Therefore, the gray scale image method can significantly improve the surface smoothness of down-facing surfaces in the MIP-SL process. Figure 5.6: A comparison of the built down-facing surfaces based on different methods. The differences in the build results illustrate the effectiveness of energy control by gray scale image in improving the surface finish of down-facing surfaces in the MIP-SL process. Consequently, with a good control of energy input by optimized mask image, it is possible to build smooth down-facing surfaces without sacrificing building time. The hardware setup shown in Figure 5.5 has a relatively large building volume 110× 82 mm which has a pixel resolution of around 0.1mm. In order to investigate the effectiveness of the 29 Tests & Results – Down-facing M1 M2 Top View Side View CAD Model 5.5 5.7 5.9 6.1 6.3 6.5 6.7 6.9 7.1 7.3 7.5 0 50 100 150 200 250 M1 M2 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm 0.5mm (a) CAD model and built parts (b) Surface quality measurement results (c) Microscopic images of the built surfaces Sampling points Z (mm) M1 M2 Down-facing surfaces Down-facing surfaces 103 proposed energy control method in micro-scale fabrication, we built a bottom-up projection based MIP-uSL system and performed a set of different test cases. The prototype and test cases will be discussed in the following section. 5.5 Verification in Bottom-up Projection based MIP-uSL 5.5.1 Prototype System As shown in the above section, the gray-scale image approach is applied in top-down projection based MIP-SL system. The energy input is controlled carefully by manipulating the gray scale values of each pixel. Its capability in fabricating smooth down-facing surfaces is verified by experiments. To verify the capability of this proposed approach in bottom-up projection configuration and microscopic fabrication cases, experiments of building curved surfaces in a bottom-up projection based MIP-uSL system have been carried out by using the proposed gray-scale image approach. Experimental results show that curved up-facing surfaces with accurate geometries could be also fabricated using the proposed gray-scale image method in bottom-up configurations. Moreover, micro-scale curved features could also be obtained by applying the developed method. A prototype system has been built to verify the presented process. The hardware setup is shown in Figure 5.7a. An optical system is designed to focus the mask image of a DMD onto the building plane with an envelope size of 12.7× 8 mm. Similar to the previous prototype, the DMD resolution is 1024×768. A blue light filter (410nm) is used for the tested resin (Perfatory™ SI500 from EnvisionTec Inc.). A linear stage from Aerotech Inc. (Pittsburgh, PA) with 1m resolution is used as an elevator in the Z axis. The layer thickness in our tests is between 10 to 75m. A 4- axis motion controller with 28 Bi-directional I/O pins from Dynomotion Inc. (Calagbasas, CA) is used to drive the linear stage and to synchronizethe movement and projection. The resin tank is a 104 clear glass Petri dish with PDMS coating on the bottom. In addition, similar to the previous setup, a mask planning software system has been developed using the C++ language to control the developed bottom-down projection based MIP-SL system, as shown in Figure 5.7 b. It integrates the geometry slicing, image planning for applying gray-scale image method in bottom-up configuration, projection and motion controlling. Figure 5.7: Hardware and software setup of the bottom-up projection MIP-uSL testbed. 5.5.2 Verification in bottom-up projection configuration A simple curved surface was tested first with the developed bottom-up projection based MIP-uSL system. Its dimensional size is 10mm by 6.7mm by 5.26mm (xyz). The CAD model is shown in Figure 5.8a. In order to get a good visional effect, a relatively big layer thickness, 75 microns is used to fabricate the parts. A fabricated part was shown in Figure 5.8b. A purchased photosensitive resin (SI500) is used. Two products were fabricated with exactly the same process parameters, including projection time, material, layer thickness, moving parameters, etc. The only difference is that one product is fabricated with the proposed gray-scale image approach and the other one is without the approach. Figure 5.8c and 5.8e are the microscopic images of area A Power supply Control board Tank Resin Z stage Projector Platform (a) Bottom-up projection based MIP-uSL (b) Software User Interface 105 and B of the product fabricated without the proposed approach, while Figure 5.8d and 5.8f are the microscopic images of the two areas of the product fabricated with the proposed approach. Figure 5.9 is the quantitative measurements of the surface profiles. Curve M1 is the surface profile of the part fabricated without gray-scale image method and M2 is the one with. Figure 5.8: Built curved surfaces based on different methods in bottom-up projection MIP-uSL system. Similar to the results shown in previous section, it is very obvious that the traditional energy control method in MIP-SL will lead to z over-cure and stair-stepping problems. Yet by controlling the energy distribution accurately through gray-scale image method, the z over-cure problem and stair-stepping problem could be both significantly alleviated or even eliminated. Accordingly, the surface quality and geometry accuracy could be improved greatly. 0.01inch 0.2mm 0.01inch 0.01inch (a) (b) (c) (d) (e) (f) Area B Area B Area A Area A Area A Area B 106 Figure 5.9: Surface measurement results of the fabricated parts shown in Fig. 5.8. 5.5.3 Verification in microscopic fabrications All the test cases discussed above are all simple meso-scale curved surfaces. To verify the effectiveness of the proposed energy control method, a more complicated test case with microscopic curves was tested. The CAD model is shown in Figure 5.10a. It is 3.5mm by 2.5mm by 2.185mm in x, y and z direction. The model has a waved down-facing surface. 0 0.5 1 1.5 2 2.5 3 3.5 2 2.5 3 3.5 4 4.5 5 5.5 M1 M2 X (mm) Y (mm) 107 Figure 5.10: Cured micro-surfaces built by different methods in bottom-up projection MIP-uSL system. Two parts were fabricated using the CAD model with the same process parameters, except the difference of the energy control method. As shown in Figure 5.10b, the part was fabricated with the conventional energy control method, while the part shown in (c) was fabricated with the gray-scale image method. A 70 microns layer thickness was used in both cases. More detailed microscopic images of two different areas A and B of the two fabricated parts are shown in (d)- 0.2mm 0.2mm (d) (e) 0.5mm 0.5mm (a) (b) (c) Area A Area B Area A Area B Area A Area A 0.2mm 0.2mm (f) (g) Area B Area B 108 (g). The quantitative measurements of the surface profiles of these two parts are shown in Figure 5.11. The green dots represent the CAD profile, the red squares plot the profile of the part fabricated without the use of the gray-scale method (M1), while the blue line segment plot is the profile of the part fabricated with the use of the proposed method (M2). Figure 5.11: Surface measurement results of the fabricated parts shown in Fig. 5.10. Again, compared to the part fabricated with conventional method, the part built with gray- scale image has much better surface finish and also has smaller approximation error. The test case showed that the proposed gray-scale method is able to control energy input well in a micro scale. Hence it is also effective in improving the part quality in micro-scale fabrications. 5.6 Concluding remarks Compared to the point-wise or linear-wise photopolymerization AM processes like SLA or CNC accumulation, MIP-SL holds its unique advantage in fast energy delivery since the light could be patterned and projected as a dynamic mask image with a very high frequency by using 0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 M2 M1 X (mm) Y (mm) ● CAD Profile  M1 M2 109 the DMD chip. Hence there is no further research needs for fast energy delivery in MIP-SL systems. Hence this dissertation focuses on accurate energy delivery methods in MIP-SL and the work is presented in this chapter. In this chapter, a gray scale image method is proposed and developed to control energy input accurately, hence to address the z over-cure problem and the stair stepping problem in MIP- SL systems. A relationship between the cure depth and the gray scale values has been calibrated and built. Accordingly, a gray scale image planning process is developed to design the gray scale values for each pixel, so that the fabricated model could approximate the CAD model better. A top-down projection based MIP-SL system and a bottom-up projection based MIP-SL system have been both developed to verify the proposed energy control method. Three test cases with different geometries and different dimensional scales have been performed. Parts have been built with the conventional black-white image method and the proposed gray-scale method. The microscopic images and comparisons between the quantitative measurements showed that the proposed gray-scale image method is able to control the energy input finely to fabricate both meso and micro scale down-facing curved surfaces with higher surface finish, smaller approximation errors, and without the cost in build speed. The hypothesis H3 is verified in this chapter: H3. A gray scale image method could make the energy delivery accurate in MIP-SL, thus to fabricate smooth down-facing surfaces in MIP-SL system together with the same or even faster build speed. But due to the constraint in light path, this energy control method is only applicable to down- facing surfaces. For up-facing curved surfaces, it is not applicable because the face starts beneath the liquid surface. Therefore, another question follows with this approach: How to make smooth up-facing surfaces without sacrificing building time? 110 As shown in Figure 5.12, we will look for solutions for up-facing surfaces fabrications in the material deposition aspect in the following chapter. Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research Content: 7 Energy Control Material Deposition Fast Delivery Accurate Delivery Fast Deposition Accurate Deposition CNC Accumulation MIP-SL H2: 3D surface reconstruction, Optimal tool path (Chapter 4) H1: Multi-tool, Process optimization (Chapter 3) H3: Gray-scale image (Chapter 5)  Down-facing surfaces √ √ √ ? H4 Up-facing surfaces? Figure 5.12: Research progress by this chapter and the following research work 111 Chapter 6 Accurate Material Deposition in MIP-SL using Meniscus Approach 6.1 Challenges of up-facing Surface Finish in MIP-SL All the surfaces in a 3D model can be classified based on the Z axis and its surface normal N into: (1) vertical surfaces (N  Z = 0), (2) down-facing surfaces (N  Z < 0), and (3) up-facing surfaces (N  Z > 0). There is no need to specially consider the vertical surfaces since they generally do not contribute to the stair-stepping effect. We addressed the problem of down-facing surfaces in the previous chapter, yet the problem in fabricating up-facing surfaces remains unsolved. As shown in Figure 6.1, due to the use of 2D layers, the fabricated up-racings surfaces especially the ones whose normal directions are close to the building direction (Z axis) usually have big approximation errors. Figure 6.1: An illustration of the stair-stepping effect in up-facing surfaces fabricated by layered AM . Stair-stepping effect on up-facing surfaces 112 In this chapter, different with the previous chapter in which the energy control is examined, we present an alternative approach for achieving smooth up-facing surfaces in the layer based photopolymerization AM systems, by looking at the material deposition side. Figure 6.2. An illustration of the meniscus approach for improving the surface finish of MIP-SL In this research, we developed a meniscus method for the fabrication of smooth up-facing surfaces (refer to Figure 6.2) in both top-down projection configuration and bottom-up projection configuration. The key idea of the meniscus approach is to closely match the fluid interfaces at the corners of intersecting planes to the related curved surfaces in the input geometry. The shape of the meniscus depends on the contact of horizontal and vertical surfaces, and can be modeled by considering the effects of capillarity, suction by gravity, multilayer adsorption and the boundary conditions[111, 112]. Accordingly, a process optimization problem can be formulated in order to match the meniscus to the given curved surfaces. The presented method has a better shape controllability over approaches of dipping and flush-curing the build parts or post-processing such as abrasive flow machining [113]. b b h b Liquid resin Liquid resin Meniscus equilibrium Z-Stage Light Source Digital Micromirror Device Tank X Z Lens Light Source Digital Micromirror Device Lens 113 Compared to other existing solutions including the use of ultra-thin layers or any post processing methods, the developed techniques enable a higher up-facing surface finish with the use of a much larger layer thickness. Hence significantly improved up-facing surface finish together with a reduced building time is expected in MIP-SL using the meniscus approach. The remainder of the chapter is organized as follows. Section 6.2 presents the models of formed meniscus in various cases in top-down projection configuration. Accordingly, the process planning for building smooth up-facing surfaces in top-down projection configuration is presented in Section 6.3. Section 6.4 presents the meniscus shape control methods and algorithms, followed by Section 6.5 which describes the calibration and planning of meniscus projection images for a given process plan. The experimental setup for performing physical experiments in top-down projection configuration is discussed in Section 6.6. The test results for various curved surfaces are presented in Section 6.7. Next we present how to extend the meniscus method into bottom-up projection configuration in Section 6.8 and Section 6.9. Finally, conclusions are drawn in Section 6.10. 6.2 Meniscus Modeling in Top-down Projection based MIP-SL 6.2.1 Mathematical Modeling The formation of the interfacial profile of the equilibrium meniscus that is attached to intersecting surfaces is first discussed. Such understanding and related mathematical models will provide the theoretical basis for the process planning and smoothness control in building the up- facing surfaces. As shown in Figure 6.3.a, given a pair of walls, suppose the length of the horizontal plane is b, and the height of the vertical plane is h. With a finite horizontal plane, the maximum range of the fluid that can rise up on an infinite vertical plane y(x=0, x length = b) is h c . 114 Similarly, with a finite vertical plane, the maximum range of the fluid that can extend along an infinite horizontal plane x(y=0, y length = h) is b 0 . Figure 6.3: Meniscus wetting to intersecting plane surfaces and fluid interface profile. Previous research on determining equilibrium shapes of fluid interfaces has been performed [58-62]. The Young-Laplace (YL) equation is the most widely accepted physical analysis of equilibrium meniscus shape. In the analysis, two types of forces, surface tension and pressure, are taken into account. The surface tension is defined with respect to a specific photopolymer resin. It is a thermodynamic quantity measuring the energy required to expand the area of the gas-liquid interface. As shown in Figure 6.3.b, the YL equation relates the pressure difference ΔP across a fluid interface within a gravitational field to the curvature of the interface and the interfacial tension by: 1 2 12 12 11 () P P P gy RR          (6.1) where P 1 and P 2 are the pressures on either side of the interface between resin and air, g is gravity acceleration, y is the height of the meniscus above the horizontal plane surface,  12 is the interfacial tension, and R 1 and R 2 are the radii of curvature of the fluid interface. If R 1 can be x y z P 1 P 2 R 2 R 1 R 2 R 1 (a) (b) 115 considered infinite, the problem is simplified to a 2D case. Thus the following Eq. (6.2) can be derived from Eq. (6.1): ̈ ( ̇) ̇ ̈ ̇ (6.2) The interfacial tension  has a relation with the capillary height c h and the contact angle  as shown in Eq. (6.3). 2 2 (1 sin ); c h g     (6.3) So Eq. (6.4) can be developed according to Eq. (6.2) and (6.3). ̈ ( )( ̇ ) (6.4) 6.2.2 Process Parameter Calibration The contact angle  in Eq. (6.4) is the angle at which the liquid resin interface meets the solidified resin surface. Specific to the given liquid and solid system, the contact angle is determined by the interactions between the liquid resin, solidified resin and air interfaces. Researchers reported that  is observed to be independent of volume and gravity and depends only on the surface tension [114, 115]. The capillary height h c is the maximum height that the fluid can reach on an infinite vertical wall. h c is a characteristic length for the fluid subject to gravity and surface tension. Both parameters ( and h c ) can be experimentally measured. Based on different b and h values, a set of experiments has been designed. The MIP-SL process was used in building the test parts with intersecting horizontal and vertical surfaces. After the horizontal and vertical surfaces have been built, the part is first merged inside liquid resin; it is then lifted up slowly until the horizontal plane is totally outside the liquid resin. A liquid meniscus in contact with the intersecting surfaces will be formed on the part surfaces. After a long waiting time, the liquid 116 volume will reach equilibrium over the horizontal wettable surface area. A mask image is then projected on the meniscus area to cure the liquid resin. The shape of the formed meniscus can thus be captured in the built part. A set of parts with different sized horizontal and vertical planes has been built. Two of such built parts are shown in Figure 6.4. Based on the experiments, the contact angle  and the capillary height h c can be measured. For the liquid resin used in our experiments (Perfactory SI500 from EnvisionTEC, Ferndale, MI), it is estimated that  =25 o , and h c =1.40mm. Figure 6.4: Two built parts with different b and h values for measuring surface tension parameters. 6.2.3 Meniscus Shape Analysis and Simulation In the MIP-SL process, suppose h denotes the height of the vertical plane, b denotes the length of the horizontal plane, and h c and b 0 denote the maximum values that the liquid can reach on the vertical and horizontal planes, respectively. The formed meniscus shapes for different b and h values can thus be analyzed based on the following four scenarios: (1) Case 1: 0 ; c h h b b  ; (2) Case 2: 0 ; c h h b b  ; (3) Case 3: 0 ; c h h b b  ; and (4) Case 4: 0 ; c h h b b  . For each scenario, different boundary conditions can be derived. Hence the formed meniscus shapes are different, and are discussed as follows. (1) Case 1: both h and b can be considered infinite. The related boundary conditions are: Top view h 1 Side view mm mm 2.47 mm 0.99 mm h 2 Side view Top view (1.4 mm) (0.3 mm) (a) (b) 117 ̇( ) - - ( ) (6.5) Using the governing Eq. (6.4) and boundary conditions Eq. (6.5), the height that the fluid can reach on the vertical wall is calculated to be 1.398372mm, which agrees well with the experiment result of 1.40 mm. We used Matlab (MathWorks, Natick, MA) to solve this problem and plotted the result as shown in Figure 6.5. Figure 6.5: Plotting results of the meniscus profile in case 1. (2) Case 2: h is smaller than h c (x=  ), and b can be considered infinite. Hence the curvature of the meniscus is decided by h. The boundary conditions are as follows. ̇( ) ( ) ̇( ) (6.6) In this case, the length of the fluid extended on the horizontal plane b 0 is dependent on the value of h, namely, b 0 = f(h). The meniscus shapes with different h values were estimated by solving the governing Eq. (6.4) and the boundary conditions described in Eq. (6.6) using Matlab. The plotting results of different h values are shown in Figure 6.6. mm mm 118 (a) 0.5 < h < 1.4 (b) 0.1 < h < 0.5. Figure 6.6: Plotting results of the meniscus profile in case 2. Figure 6.7: Plotting results of the meniscus profile in case 3. (3) Case 3: h is bigger than h c (x=  ) and b is smaller than b 0 . Thus the curvature of the meniscus is decided by b. The boundary conditions are as follows. ̇( ) ( ) ̇( ) (6.7) Based on the governing Eq. (6.4) and the boundary condition Eq. (6.7), the meniscus shapes can be computed in Matlab. Figure 6.7 shows the plotting results for different b values. (4) Case 4: h is smaller than h c (x=  ), and the length of the horizontal plane b is smaller than b 0 . Hence the curvature of the meniscus is decided by both b and h. The problem follows the boundary conditions: mm mm mm mm (a) (b) mm mm mm mm (a) (b) 119 (0) , ( ) 0 y h y b  (6.8) The meniscus shape can be calculated for different b and h values. Three examples using Matlab to solve the mathematic model and the related boundary conditions for different b and h values are shown in Figure 6.8. Figure 6.8: Plotting results of the meniscus profile in case 4. 6.3 Process Planning for Building Smooth Up-facing Surfaces Suppose an input 3D CAD model with curved up-facing surfaces is given as shown in Figure 6.9.a. The traditional additive manufacturing method will slice the model into multiple 2D layers based on a given layer thickness. One mask image will be generated for each layer in the MIP- SL process to cure the related 2D shape. Consequently the built part will inevitably have stair- stepping defects. In comparison, Figure 6.9.b illustrates the major steps of the meniscus equilibrium method for achieving smooth up-facing surfaces. For a specific layer i, c(x, y) is computed which represents the ratio of the Z-height at the location (x, y) to the layer thickness L T . Similar to[116], a sliced image with gray scale values can be generated for each layer. Accordingly the thickness at pixel (x, y) is c(x, y)× L T . Hence instead of a binary layer image, the sliced image has pixels with ratios c(x, y). Note c(x, y) is continuous and c(x, y)  [0, 1]. Suppose the original gray scale image can be split into two mask images, image #1 and #2. As shown in Figure 6.9.b, the mask image 1 can cure a portion of layer i that will be used in forming meniscus; the mask image 2 can be used in (b) (a) (c) mm mm mm mm mm mm b = 2, h = 1.2 b = 0.5, h = 0.2 b = 2, h = 0.3 (b) (a) (c) mm mm mm mm mm mm b = 2, h = 1.2 b = 0.5, h = 0.2 b = 2, h = 0.3 (b) (a) (c) mm mm mm mm mm mm b = 2, h = 1.2 b = 0.5, h = 0.2 b = 2, h = 0.3 (a) (b) (c) 120 curing the formed meniscus based on image 1 . Note that it is not necessary to use the meniscus equilibrium approach in building each layer. That is, depending on the given geometry, meniscuses can be formed by image 2 after multiple layers of image 1 have been built. This is different from the approach presented in [57, 117] for the SLA process, in which the formed meniscuses are cured in each layer. In addition to slow building speed, such an approach was found to only work for up-facing surfaces that are 40 o or less from the horizontal. Such a limitation can be explained by the presented meniscus shape modeling. Figure 6.9: A comparison between the traditional and the meniscus equilibrium methods. As discussed in the meniscus mathematical model, two important parameters in determining the shape of the meniscus profile are h and b. Thus the planning of image 1 and image 2 can change the setting of h and b, which leads to different approximation errors between the meniscus Binary Image Meniscus Modeling Step 2. Mask image planning Step 1. Image Generation Step 2. Building (a) Traditional Method Air Fluid Interface 1 X Z b h 1 X Z Air 1 F(x,y) f(x,y) X Y Air Resin Step 3. Building Z Air Our Method Z X X Y X Cured Shape Binary Image #1 Binary Image #2 Slicing plane Mask projection Cured layer Step 1. Image Generation Z X Cured Shape Slicing plane Gray scale Image Y X Mask projection Mask projection Cured layer 121 profiles and the desired shapes. An illustration example is shown in Figure 10.a. For the given shape boundary defined by black lines, two different sets of mask images (image 1 and image 2 ) can be used, which define two vertical walls as shown in brown lines. Accordingly the formed meniscuses will have different profiles due to different b or h values as shown in Figure 10.a. (a) Different width and height of vertical planes (b) An illustration example. Figure 6.10: The process planning problem for the meniscus equilibrium method. In order to generate the MIP-SL process plan that can ensure the meniscus shapes match the given curved geometry, the process planning for the up-facing surfaces can be mathematically defined. The shape approximation error needs to be controlled in an accepted range by manipulating the meniscus shape. Detailed meniscus shape control and optimization methods are described in the following section. 6.4 Optimization of Meniscus Process Parameters As shown in Figure 6.11, suppose a boundary curve y = f(x) ( ( ), ( )) is given, where is smaller than the maximum vertical height h c as discussed in the meniscus model (Case 1). An algorithm based on the greedy heuristic is developed as shown in Figure 6.11. It is different from the approach presented in[115] for the SLA process, in which the formed meniscuses are cured in each layer. Z h 1 Image 1 Image 2 b 1 h 1 b 2 Image 1 Image 2 X Z X Z Image 1 Image 2 X h 2 b 1 Meniscus profile Shape boundary Vertical Wall (a) 3 2 1 X Z L T h b b i h i 0 f(x) F 0 (b i , h i ) F 1 F 2 h max (b) 122 Figure 6.11: Framework of meniscus algorithm. Figure 6.12 shows an example using this algorithm. The meniscus shape in the area of ( ) is first estimated and then compared with the input geometry. As shown in Figure 10(b), the meniscus curve is the orange curve while the black curve is the given CAD model profile. If the error is within the acceptable range, the two points ( ) ( ) can be selected as the meniscus points M 0 and M 1 . Otherwise, two strategies will be applied according to the value of approximation error. If the meniscus curve is outside of the CAD profile which represents by a positive , the point where the biggest approximation error happens will be selected as an additional meniscus point, as M 0-1 in Figure 10(c). On the contrary, if the meniscus curve is inside the CAD profile, for example, the orange curve between M 0-1 and M 0 in Figure Input: Boundary curve segment y=f(x) (yϵ(0,y0), xϵ(0,b0) ) and Slices {S} Simulate the Meniscus Curve y’= F(x) ε a too big? Y ε a positive? N Insert new meniscus point (b1, y1) to split the input segment into: y=f(x) (yϵ(0,y1), xϵ(0,b1) ) y=f(x) (yϵ(y1,y0), xϵ(b1,b0) ) Update input boundary curve segment: y0=y1, b0=b1 Output: Meniscus point (b0,y0) and slices {S} Insert new slice in the negative approximation error area: Update slices {S} Calculate approximation error ε a between y’ and y Y N 123 10(c), new slices would be added to push the meniscus curve out to better approximate the CAD profile. Curves S0-1-1 and S0-1-2 are the examples. Figure 6.12: An illustration of meniscus process optimization. Based on the updated meniscus points and slices, the new meniscus profile and the corresponding approximation error are calculated. Such process is iterated until the approximation error is within a defined satisfactory range. With the output meniscus points, dipping points and numbers are decided. Also a set of projection images (image 1 and image 2 ) can then be computed for the building process with the output slices and dipping strategies. Accordingly, image 1 can be used in solidifying resin in each layer. When the current layer number comes to h i related to the meniscus point M hi , the fabricated model will be raised from the liquid resin and moved to the meniscus projection area. After a certain waiting time, related mask image 2 will then be projected on the portion of the meniscus surface to cure the formed meniscus shapes in such areas. 6.4.1 Approaches for modifying meniscus profile (a) (b) (c) (d) M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 (a) (b) (c) (d) M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 M0-1 M1 M0-1 M1 M0-2 S0-1-1 M0 S0-1-2 M0 M0 M1 124 The following algorithm is used to decide if the meniscus profile needs to be improved according to the approximation error and what strategy needs to be used for the modification of meniscus profile. Algorithm1: Input: A segment of curve surface f(x, h), a layer thickness. Find: A set of b i and h i Satisfy: ( ) ( , ) i i i F x F b h  Calculate: ∫ ( ( ) ( )) ∑ ∑ if , use algorithm 1.1 to plan the new sub-slice images. if , use algorithm 1.2 to plan the new meniscus points. 6.4.2 New Sub-slice Planning for negative approximation error As shown in Figure 6.13, the black curves represent the CAD model profile and dotted yellow curves represent the formed meniscus profile. With the two slices as shown in Figure 6.13a, it is demonstrated that there is a negative approximation error as the meniscus curve is inside the CAD model profile. By inserting a new sub-slice in between, as shown in (b) and (c), the formed meniscus shape is changed due to additional stair generated by the sub-slice. And with different positions for inserting additional sub-slice, the modified meniscus profiles are different. Figure 6.13: An illustration of influence of sub-slice planning on formed meniscus shape. The following algorithm is used to decide the inserted sub-slice position y_ s : Algorithm1.1: Controlled Meniscus Approach (k, f)=(0,2) (k,f)=(0.5,2) (k,f)=(-0.5,2) k=0.3 b k =0.3 b t b t k= -0.1 b t b k = -0.1 b t Binary Image #1 Geometry profile Geometry profile Cured by Image #1 Cured by Image #2 Layer thickness f 1 f 2 f 3 (a) (b) (c) Fig. 1: An illustration of influence of sub-slice planning on formed meniscus shape. To decide the inserted sub-slice position y _s: Introduction | Meniscus Modeling | Meniscus Control | Test Cases y0 y1 yi1 yi2 Controlled Meniscus Approach (k, f)=(0,2) (k,f)=(0.5,2) (k,f)=(-0.5,2) k=0.3 b k =0.3 b t b t k= -0.1 b t b k = -0.1 b t Binary Image #1 Geometry profile Geometry profile Cured by Image #1 Cured by Image #2 Layer thickness f 1 f 2 f 3 (a) (b) (c) Fig. 1: An illustration of influence of sub-slice planning on formed meniscus shape. To decide the inserted sub-slice position y _s: Introduction | Meniscus Modeling | Meniscus Control | Test Cases y0 y1 yi1 yi2 Controlled Meniscus Approach (k, f)=(0,2) (k,f)=(0.5,2) (k,f)=(-0.5,2) k=0.3 b k =0.3 b t b t k= -0.1 b t b k = -0.1 b t Binary Image #1 Geometry profile Geometry profile Cured by Image #1 Cured by Image #2 Layer thickness f 1 f 2 f 3 (a) (b) (c) Fig. 1: An illustration of influence of sub-slice planning on formed meniscus shape. To decide the inserted sub-slice position y _s: Introduction | Meniscus Modeling | Meniscus Control | Test Cases y0 y1 yi1 yi2 125 Input: The segment of curve surface f(x, y) (y0<y<y1) which gives negative approximation error. A set of y i which are candidates for the inserted sub-slice position y _s. y i is sampled from the range of y0 and y1. Calculate { } using the algorithm described in algorithm 1. Output: i that gives the smallest . y_ s =y i With the output y_ s , we first slice the input CAD model at the position y= y_ s . And then the new slice is added into the slices group {S} for layer by layer curing. So the resulted part has non- uniform layer thickness to achieve more accurate meniscus profiles. 6.4.3 New sub_meniscus point planning for positive approximation error As shown in Figure 6.14, the black curves represent the CAD model profile, and the dotted black curves are layers. As shown in Figure 6.14 (b), a part is built layer by layer. The meniscus is then formed after it reaches the layer of M1. The formed meniscus is represented by the yellow curve. It is shown that the formed meniscus between M0 and M1 is outside of the CAD profile, giving a positive approximation error. Instead of forming meniscus until the part is built to position M1, we can solidify multiple meniscuses sequentially to improve the accuracy. As shown in Figure 6.14(c), when the building part is at the level of M0-1, it is moved out from liquid to form a meniscus between M0-1 and M0. After solidified the meniscus between M0-1 and M0, we continue to build the following layers and form the second meniscus between M0-1 and M1. The fabricated result of two meniscuses would give smaller approximation error than the one time meniscus shown in (b). Algorithm 1.2 is used to determine the position of the inserted sub-meniscus point y_ d . 126 Figure 6.14: An illustration of influence of sub-meniscus point on formed meniscus shape. Algorithm1.2: Input: The segment of curve surface f(x, y) (M0<y<M1) which gives positive approximation error. A set of y i which are candidates for the inserted sub-meniscus position y_ d. y i is sampled from the range of M0 and M1, i=1,2,..m. Calculate { } using the algorithm described in algorithm 1. Output: i that gives the smallest . y _d =y i 6.5 Meniscus Projection Image Calibration and Planning As shown in Figure 6.2, the built part is raised up above the liquid surface when the resin related to the desired meniscus shape is cured. Accordingly the mask image (image 2 ) that will be used in the meniscus equilibrium approach needs to be planned by considering the distortions associated with the different Z heights of the desired meniscus shapes. Note the resin surface is the focus of the projection image. The formed meniscuses after lifting up for a certain Z height are actually out of focus. To identify a meniscus mask image with controlled exposure for all the meniscuses, the projection light modeling and calibration is first performed. Based on established pixel positions in the topmost and bottom planes, the inverse calculation of the mask image is presented for any given Z height. Hence the computed mask image 2 will provide the desired projected light to exactly cover the portion of resin related to the meniscus shape that may located in different Z levels. M0-1 M1 M0 M0 M1 (a) (b) (c) y1 y2 y3 y4 y5 ym M0-1 M1 M0 M0 M1 (a) (b) (c) y1 y2 y3 y4 y5 ym M0-1 M1 M0 M0 M1 (a) (b) (c) y1 y2 y3 y4 y5 ym 127 6.5.1 Projection Image Modeling and Calibration The projection light of each pixel (i, j) is assumed to be straight within the Z height range in which the formed meniscuses will be located. As shown in Figure 6.15a, if P 1 (i, j) and P 2 (i, j) on the top and bottom Z planes can be identified, the position P(i, j) for any given Z height can be easily computed based on the linear interpolation of P 1 and P 2 . Figure 6.15: The meniscus projection image calibration. In this research, our projection image calibration method is based on the use of a high resolution camera (Canon SX10) to capture the relative position of a projected checkerboard and a physical checkerboard that is positioned at top and bottom Z levels. We then used image processing techniques to build a top and bottom Z level databases, in which, the physical positions of all the pixels on the two calibrated plane are recorded (refer to Figure 6.15b). Based on them, a linear interpolation approach can be used in computing the physical position of a pixel Z Stage Platform Bottom Z level Top Z level 20 mm Image Pixel Position Physical Position ( X P0 , Y P0 ) … … ( X P1023 , Y P767 ) Image Pixel Position Physical Position ( X 0 , Y 0 ) … … ( X 1023 , Y 767 ) Bottom Z level database Image Pixel Position Physical Position ( X P0 , Y P0 ) … … ( X P1023 , Y P767 ) Image Pixel Position Physical Position ( X 0 , Y 0 ) … … ( X 1023 , Y 767 ) Top Z level database Given Z height Given Z level Database h Pixel (1023, 767) Pixel (0, 0) Pixel (1023, 767) Linear interpolation Camera Checkerboard in a calibration plane (a) (b) (c) Bottom Z level Top Z level (1023, 767) (0, 0) (0, 0) (1023, 767) Resin surface Maximum raising level h DMD P(i, j) P 2 (i, j) P 1 (i, j) Z Stage Platform Bottom Z level Top Z level 20 mm Image Pixel Position Physical Position ( X P0 , Y P0 ) … … ( X P1023 , Y P767 ) Image Pixel Position Physical Position ( X 0 , Y 0 ) … … ( X 1023 , Y 767 ) Bottom Z level database Image Pixel Position Physical Position ( X P0 , Y P0 ) … … ( X P1023 , Y P767 ) Image Pixel Position Physical Position ( X 0 , Y 0 ) … … ( X 1023 , Y 767 ) Top Z level database Given Z height Given Z level Database h Pixel (1023, 767) Pixel (0, 0) Pixel (1023, 767) Linear interpolation Camera Checkerboard in a calibration plane (a) (b) (c) Bottom Z level Top Z level (1023, 767) (0, 0) (0, 0) (1023, 767) Resin surface Maximum raising level h DMD P(i, j) P 2 (i, j) P 1 (i, j) 128 in any given plane that is inside the calibration volume. Figure 6.15c shows the calibrated databases in red and blue lines for the top and bottom layers considered in our setup, respectively. The projection results of some intermediate Z levels were also processed, which verified the interpolation results. 6.5.2 Meniscus Projection Image Planning In the process planning results of a given CAD model, a set of mask image 1 and image 2 have been computed for each layer of the model (refer to Figure 6.9). In addition, a set of meniscus points P hi have been the determined. Based on them, the number of layers between two neighboring meniscus points and the related mask image 2 for curing meniscuses related to all the layers are known. Accordingly, a single meniscus projection image can be computed by considering the distortion related to their Z heights. That is, since the platform will be moved upward for a certain height for forming meniscus, the mask image 2 in each layer can be converted into a transformed image based on the aforementioned projection calibration results. Finally all the computed mask images can be combined to form a single projection image for the related meniscus points. Figure 6.16 illustrates the mask image 2 planning process based on an example similar to the test case as shown in Figure 14. The CAD model of the test part is shown in Figure 6.16a, which is sliced into 76 layers. The meniscus equilibrium method will be applied at layers 20, 40, 60, and 76. Based on the calibration results as shown in Figure 6.15, the computed meniscus projection image for the first 20 layers (i.e. P h1 ) is shown in Figure 6.16d. The computing results of three individual layers (1 st , 10 th , and 20 th ) are also shown in the figure. 129 Figure 6.16: An illustration of the meniscus image planning. 6.6 Experimental Setup for Meniscus Approach in Top-down Projection System As introduced in Chapter 5, a top-down projection based MIP-SL prototype system has been built for verifying the presented methods in this research. As a reminder here, the hardware setup and the software system of the developed MIP-SL system is shown in Figure 6.17. The components of the system have been described in details in Chapter 5. So it is not repeated here. Figure 6.17: The developed MIP-SL testbed for fabricating smooth surfaces. (a) (b) (d) Mask image #2 CAD Model (c) Layer 1 Layer 10 Layer 20 + + I 2 (L 1 ) I 2 (L 10 ) I 2 (L 20 ) I 2 c (L 1 ) I 2 c (L 10 ) I 2 c (L 20 ) I 2 c (L 1-20 )=È I 2 c (L i ) ... ... Meniscus points at layers 20, 40, 60, 76. P h1 P h2 P h3 P h4 Resin tank 1 DLP projector Z Stage Reflecting mirror Motion controller Joy stick Building Platform (inside resin) Sweeper Power supply (a) Control panel Motion parameters Building process report Geometry slicing Geometry (b) 130 A set of test cases were designed to test the meniscus equilibrium method. The experimental results are presented as follows. 6.7 Tests of Meniscus Approach in Top-down Projection based MIP-SL 6.7.1 Tests on Straight Up-facing Surfaces As shown in Figure 6.18a, a CAD model with a slanted surface M was used in the test. To compare the differences between the traditional and the newly developed methods, the surface M was split into two halves, M 1 and M 2 . The traditional method was used in building the upper half M 1 , and the meniscus equilibrium method was used in building the bottom half M 2 . The same layer thickness (0.1mm) was used in the test. Accordingly a set of mask images was generated by slicing the given CAD model. However, one meniscus point every ten layers was added in set {M} during the process of building M 2 . That is, after every 10 layers have been built, the built part will be raised up by a certain distance (above the resin surface by 1.5mm). After waiting for several minutes, liquid will reach equilibrium. A mask image 2 related to the portion of M 2 within the built 10 layers is then projected to cure the formed meniscuses. The building result is shown in Figure 6.18a. It is obvious that the appearance of the M 2 portion is much smoother than that of the M 1 portion. 23 Tests & Results Up-facing Surface ----Test 2: M1 M2 Figure 11. A comparison between the surfaces built by traditional method (M1) and by meniscus equilibrium method (M2) Top View Side View M2 Application (a) CAD Model Top View Side View M1 M2 M1 M2 M (a) Z (mm) Sampling points (b) 131 Figure 6.18; A comparison of the built concave up-facing surface based on different methods. Quantitative measurements have been performed to better understand the surface quality difference. The Z height of a set of uniformly sampled points on M was measured using a digital height gauge with a 0.1mm probe tip. To automate the measuring process, two motion controlled linear stages from Velmex Inc (Bloomfield, NY) were used. In the measurement setup, the built part was fixed on a linear stage to be translated in the X direction. The digital height gauge was fixed on another linear stage to be translated in the Z direction. Every time the probe touches a sampling point on the surface, a height value will be recorded from the digital gauge. To avoid the friction between the probe tip and the slanted surface, the gauge moves away from M in the Z direction first before the part is translated by a small distance in the X direction. The recorded Z height values of the sampling points on surface M are plotted in Figure 6.18b. The X axis denotes a set of uniformly spaced sampling points, where ‘0’ denotes the starting point of the measurement and ‘400’ denotes the 400 th measuring point. The Z axis denotes the readouts from the digital height gauge. The recorded data of the upper half M 1 and the bottom half M 2 are plotted using blue and red points respectively. The quantitative comparison shows that the stair- stepping effect of M 1 is much greater than that of M 2 . We also marked three portions, a 1 , a 2 and a 3 , on the surface as shown in Figure6.18b. They represent M 1 , the transition between M 1 and M 2 , and M 2 respectively. A precision measurement machine (Sol system from Micro Vu Inc., 24 M1 M2 Figure 11. A comparison between the surfaces built by traditional method (M1) and by meniscus equilibrium method (M2) Top View Side View M2 Application (a) 0.5 mm 0.5 mm 0.5 mm a1 a2 a3 a1 a2 a3 (c) 132 Windsor, CA) was used in taking microscopic images of these three portions. The captured images are shown in Figure 6.18c. The measured surface finishes are given in Table 6.1. The results illustrate that the traditional method will lead to a ragged surface, while the meniscus equilibrium method can effectively eliminate the stair-stepping effect in the MIP-SL process. 6.7.2 Tests on Curved Up-facing Surfaces – Concave Cases As shown in Figure 6.19a, a CAD model with a curved surface M was used in the test. The same process parameters were used in the traditional and the newly developed methods for building the CAD model. The related build results M 1 and M 2 are shown in Figure 6.19a. The same measurement procedure as described in Section 6.1 was followed for a quantitative comparison of their surface finishes. The Z height values of the curved up-facing surface are shown in Figure 6.19b. The microscopic images of a portion of M 1 and M 2 are shown in Figure 6.19c. The measured surface roughness is given in Table 6.1. 25 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error M2 Side View CAD Model Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error M1 Top View M2 Discontinuous error (a) 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 0 100 200 300 400 500 600 M1 M2 Z (mm) Sampling points (b) 133 Figure 6.19: A comparison of the built concave up-facing surface based on different methods. The meniscuses in this test case are formed based on a cylinder and an intersecting planar surface. The test results verify the capability of the meniscus equilibrium method to fabricate smooth concave up-facing surfaces. One limitation of the developed method is that the given geometry is required to have an outlet for the liquid resin to flow down to the tank such that meniscuses can be formed. In the process planning of the meniscus equilibrium method, there may also be discontinuous error between neighboring meniscuses as shown in Figure 6.19a. In order to reduce such error, the generated masks image 1 and the layers at which image 2 are projected should be adjusted. However, reducing the discontinuous errors may increase the shape approximation error. Hence a balance between the two types of errors needs to be considered. 6.7.3 Tests on Curved Up-facing Surfaces – Convex Cases As shown in Figure 6.20a, a CAD model with a curved surface M was used in the test. The same process parameters were used in both the traditional and newly developed methods to build the model. The related build results M 1 and M 2 are shown in Figure 6.20a. The same measurement procedure as described in Section 6.1 was followed for a quantitative comparison of their surface finishes. The Z height values of the curved up-facing surface are shown in Figure 6.20b. The microscopic images of a portion of M 1 and M 2 are shown in Figure 6.20c. The (c) 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 (c) 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 (c) 134 measured surface roughness is given in Table 6.1. The test results verify the capability of the meniscus equilibrium method to fabricate smooth convex up-facing surfaces. Figure 6.20: A comparison of the built convex up-facing surface based on different methods. The profiles of the built surfaces were sampled with surface finishes measured. The computed roughness measures are given in Table 6.1. Table 6.1: Roughness of the built surfaces in Fig. 6.18-6.20. Figure Surface Roughness measures R a R q R z Figure6.18 M 1 0.028 0.034 0.120 M 2 0.010 0.012 0.021 Figure6.19 M 1 0.031 0.038 0.118 M 2 0.008 0.010 0.041 27 Tests & Results Up-facing Surface ----Test 4: M1 M2 Side View CAD Model Top View (a) 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5 0 50 100 150 200 250 300 350 M1 M2 Z (mm) Sampling points (b) (c) 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 M2 M1 (c) 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 M2 M1 (c) 135 Figure6.20 M 1 0.027 0.033 0.090 M 2 0.012 0.014 0.028 6.8 Extension of Meniscus Approach in Bottom-up Projection based MIP-SL 6.8.1 Meniscus Modeling Figure 6.21: Meniscus wetting to intersecting plane surfaces and fluid interface profile. Our previous research[118] reported a smooth surface fabrication method based on applying meniscus wetting in the top-down based Streolithography for meso- and macro-scales features [118]. The approach is shown in Figure 6.2. A mathematical model has been derived from Young-Laplace (YL) equation for the residue modeling in top-down projection based MIP-SL process by considering pressure difference, surface tension, and gravitational field [118]. As discussed in previous sections, the developed meniscus shape equation relates the curvature of the interface with gravitational influence and interfacial tension that is represented by contact angle  and capillary height hc, as shown in Eq.(6.4). The capillary height h c is the maximum height that the fluid can reach on an infinite vertical wall. h c is a characteristic length for the fluid subject to gravity and surface tension. To model the meniscus shape, different boundary conditions are used in the following five cases. h b y x R 1 R 2 R 1 R 2 P 1 P 2 x y z (a) (b) h b y x R 1 R 2 R 1 R 2 P 1 P 2 x y z (a) (b) 136 (1) Case 1: both |h| and b can be considered infinite. The related boundary conditions are: ̇ ( ) ( ) (6.9) (2) Case 2: |h| is smaller than h c , and b is bigger than b 0 . Hence the curvature of the meniscus is decided by h. The boundary conditions are as follows. ̇( ) - - ( ) - ̇ ( ) ( ) (6.10) (3) Case 3: |h| is bigger than h c , and b is smaller than b 0 . Thus the curvature of the meniscus is decided by b. The boundary conditions are as follows. ̇( ) - - ( ) ; ̇( ) ( ) (6.11) The meniscus shape can be calculated for different b and h values. Figure 6.23 and 6.24 are example plots of Case 2 and Case 3: when h equals to -1 mm, b is a value bigger than b 0 ; when |h| is bigger than h c and b equals to 2.15mm. Figure 6.22: Meniscus profile in case 3 and case 2. Figure 6.23: Meniscus profile in case 4. -1.5 -1 -0.5 0 0 1 2 3 Case2: h=-1, b infinite Case 3: h>hc, b=2.15 mm mm -1.29 2.15 -1 0.8 h=-1,b>b0 mm mm -0.5 -0.4 -0.3 -0.2 -0.1 0 0 0.05 0.1 0.15 h=0.4 b=0.1 h=0.05 b=0.08 mm mm -1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0 2 4 6 case 5 mm mm h b y y x y x y x y x x -1.5 -1 -0.5 0 0 2 4 6 Case2: h=-1, b infinite case 3: h>hc, b=2.15 -1.29 2.15 -1 4.79 0.08 0.05 1.19 4.98 137 (4) Case 4: h is smaller than h c , and the length of the horizontal plane b is smaller than b 0 . Hence the curvature of the meniscus is decided by both b and h. The problem follows the boundary conditions: ( ) ( ) (6.12) Case 4 is one of the most common two cases in micro-scale fabrication. The meniscus shape can be calculated for different b and h values. Figure 6 plots the result of two examples of Case 4: when h equals to -0.4 mm, b equals to 0.1mm; when h equals to -0.05 mm, b equals to 0.08mm. (5) Case 5: h is smaller than h c , and the discontinuity is connected with a horizontal plane that the size is smaller than the liquid droplet size. The problem turns into a wetting puddle in the bottom- up projection system and the boundary conditions are: ̇( ) ( ) (6.13) Case 5 is another case that will happen in bottom-up projection based micro-scale fabrication. Matlab was used to solve the mathematic model of the puddle shape and the result is plotted in Figure 6.24, where the blue line represents the last layer of the object and a puddle is formed on the surface. Figure 6.24: Plotting result of the meniscus profile in case 5. mm mm -0.5 -0.4 -0.3 -0.2 -0.1 0 0 0.05 0.1 0.15 h=0.4 b=0.1 h=0.05 b=0.08 mm mm -1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0 2 4 6 case 5 mm mm h b y y x y x y x y x x -1.5 -1 -0.5 0 0 2 4 6 Case2: h=-1, b infinite case 3: h>hc, b=2.15 -1.29 2.15 -1 4.79 0.08 0.05 1.19 4.98 138 6.8.2 Process Parameter Calibration As discussed in our previous research [118], the contact angle  in Eq. (6.1) the angle formed by the liquid resin interface and the solidified resin surface. Specific to the given liquid and solid system, the contact angle is determined by the interactions between the liquid resin, solidified resin and air interfaces. Figure 6.25: Bottom-up built parts with b= 4mm and h=5: Left: SI-500; Right: E-Shell. Researchers reported that  is observed to be independent of volume and gravity and depends only on the surface tension [114, 115]. Similar to the parameter calibration work described in our previous research [118], parameters ( and hc) were experimentally measured. Based on different b and h values, a set of experiments has been designed. The MIP-uSL process was used in building the test parts with intersecting horizontal and vertical surfaces. After the horizontal and vertical surfaces have been built, the part is first lifted up from the liquid resin. A liquid meniscus in contact with the intersecting surfaces will be formed on the part surfaces. After a long waiting time, the liquid volume will reach equilibrium over the horizontal wettable surface area. It is then moved slowly in x direction to a new place. A mask image is then projected on the meniscus area to cure the liquid resin. The shape of the formed meniscus can thus be captured in the built part. A set of parts with different sized horizontal and vertical planes has been built. Two of such built parts are shown in Figure 6.25. Based on the experiments, the contact angle  and the 2.25mm 1.722 mm 0.5 mm 1.883 mm 1.567mm 0.5 mm 1.883 mm 1.567mm 0.5 mm 2.25mm 1.722 mm 0.5 mm 2.25mm 1.722 mm 0.5 mm 1.883 mm 1.567mm 0.5 mm 1.883 mm 1.567mm 0.5 mm 2.25mm 1.722 mm 0.5 mm 139 capillary height h c can be measured. For the liquid resins used in our experiments (Perfactory SI500 from EnvisionTEC, Ferndale, MI), it is estimated that  =25 o , and h c =1.88mm. Another liquid resin, E-Shell, is also from EnvisionTEC. Its measured  and h c are 21 o and 1.72 mm, respectively. Table 6.2 shows the material properties and model parameters of these two materials. Table 6.2: Residue Modeling Parameters. Property E-shell Bottom-up SI 500 Bottom-up SI 500 Top-down Viscosity 339.8cP 180cP 180cP Density 1.19 g/cm 3 1.10 g/cm 3 1.10 g/cm 3 Contact angle 21 ◦ 25 ◦ 25 ◦ Capillary height 1.72 mm 1.88 mm 1.40 mm Maximum wetting b 0 2.25 mm 1.567 mm 2.5 mm 6.9 Experimental Verification of Meniscus Approach in Bottom-up System 6.9.1 Experimental Setup As introduced in Chapter 5, a prototype MIP-uSL system has been built to verify the approaches presented in this dissertation. The same prototype MIP-uSL system that was presented in Chapter 5 is used here to verify the proposed fast recoating approach. As a reminder, the hardware setup and software is shown in Figure 6.26. 140 Figure 6.26: The developed MIP-SL testbed for fabricating smooth surfaces. In addition, a Matlab program is used to simulate and generate the meniscus points and additional slicing points. Although the user interface of the software as shown in Figure 6.26 b looks similar to the software introduced in Chapter 5, more functions have been added. In addition to the common functions like CAD file operations, geometry slicing, motion and projection control that have been mentioned in the software presented in previous chapters, the testbed has been upgraded to suit the use of meniscus approach on this research status. New functions like generating the meniscus images (Image 1 ) and regular layer projection images (Image 2 ) and meniscus curing operation have been added into the software. 6.9.2 Results and Discussion Tests on Concave Surface As shown in Figure 6.27, a CAD model ( ) with a curved surface M was used in the test. The traditional layer based MIP-SL process was used to build the part. The layer thickness was set as 20 . Figure 6.27(a), 6.27(e), and 6.27(h) show the microscopic Power supply Control board Tank Resin Z stage Projector Platform (a) Control panel Motion parameters Building process report Geometry slicing Geometry (b) Control panel Motion parameters Building process report Geometry slicing Geometry (b) Geometry slicing Control Panel Motion Parameters Building Process Report 141 images of the area A 1 and A 2 portions of the built part. As shown in Figure 6.27(e) and 6.27(h), although a small thickness was applied, the layer stepping effect is still visible. A one-time dipping meniscus method was used to fabricate the same CAD model. Figure 6.27: A test case of concave surface. Microscopic images of the same two areas A 1 and A 2 of the fabricated part are shown in Figure 6.27(b), 14(f), and 14(i). It is obvious that the one-time dipping meniscus method brings big approximation error. To reduce the approximation error, a controlled meniscus method with two dips was used. The fabricated result is shown in Figure 6.27(c), (g), and (j). It is shown that the fabricated part profile is smoother and closely approximates the input CAD model. Tests on Convex Surfaces 142 Figure 6.28 and Figure 6.29 show the fabricated results of a micro-lens with a convex surface ( ) using the resins SI500 and E-shell. Figure 6.28(a) and (b) show the CAD model and one fabricated physical model respectively. Figure 6.28(c) and (d) are the microscopic images of area A 1 and A 2 of the part fabricated by the traditional layer-based MIP- SL process. Figure 6.28(e) and (f) are the microscopic images of the part fabricated by the meniscus method. Figure 6.28: A test case of micro lens using SI 500. Same process parameters were repeated to fabricate the micro-lens using transparent resin E- shell. The results are shown in Figure 6.29. In order to demonstrate the effectiveness, a large layer thickness (100m) was used to fabricate Part A. Also the fabricated lenses were put on a paper, which has blue and red curves to compare the optical performances. Figure 6.29 indicates the defects of the fabricated micro-lens due to the stepping effect and the improvement of the optical performance using the meniscus method. The paper image is totally distorted by the rough surface of the fabricated part Lens B. A1 A2 (b) Physical Model 3mm 0.5mm (a) CAD Model (c) Side View A1-Part A 0.1mm 0.2mm (f) Top View A2- Part B 0.2mm (d) Top View A2- Part A 0.1mm (e) Side View A1-Part B A1 A2 (b) Physical Model 3mm 0.5mm (a) CAD Model (c) Side View A1-Part A 0.1mm 0.2mm (f) Top View A2- Part B 0.2mm (d) Top View A2- Part A 0.1mm (e) Side View A1-Part B 143 Figure 6.29: A test case of micro lens using E-Shell. Quantitative Study of Test Cases Quantitative measurements have been performed to better understand the surface quality difference. A microscope was used to record the surface profiles of the fabricated micro structures. Approximation Errors. Profiles of the area A1 of the built surfaces showed in Figure 6.27 were sampled and plotted in Figure 6.30. Because in the area A1, the angle between Z axis and its surface normal N is very small, there is no obvious stair-case effect. As shown in Figure 6.30, all three parts are smooth in this area. However, the 1 dipping meniscus method left a great amount of extra material in its meniscus area, hence caused a big approximation error. By inserting a sub-meniscus point, the fabricated part using 2 dipping meniscus method is much closer to the original model. Table 6.3 shows the approximation error statistics of the three fabricated parts. 3mm 0.5mm 0.2mm 0.2mm 0.2mm 0.2mm (a)Top View- Lens A (b)Top View- Lens B (c)Top View- Lines on Paper (d)View of Paper under Lens A (e)View of Paper under Lens B 0.2mm 144 Figure 6.30: Geometry Profile of the models in Fig. 6.27 Table 6.3: Accuracy of the built geometries with different building strategies. Part Accuracy (μm) Max-offset Ave- offset Figure 6.27- M1 12 4 Figure 6.27-M2: 2dippings Meniscus 6 2.3 Figure 6.27- M2 1 dipping Meniscus 375 258 CAD Model 0 0 Surface Roughness. Profiles of the area A2 of the part built by traditional MIP-uSL process M1 and the part built by 2 dipping meniscus method showed in Figure 6.27 were sampled and plotted in Figure 6.31. Profiles of Part A which was built by traditional MIP-uSL process M1 and the part built by meniscus method showed in Figure 6.28 were plotted in Figure 6.32. Figure 6.31 and Figure 6.32 show that the controlled meniscus approach addressed the stair- case effect for both concave and convex surfaces. The measurements showed in Table 6.4 verified its effectiveness in improvement of surface finish. 0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 M1 1d 2d mm mm 26 Introduction | Meniscus Modeling | Meniscus Control | Test Cases 1 Dipping Meniscus M2-1d M2-1d M1 M2-2d M1 M2-2d 145 Figure 6.31: Surface measurement of area A2 in parts built by M1 and M2 2dippings in Fig. 6.27 Figure 6.32: Surface measurement results of part A and B in Fig. 6.28 Table 6.4: Roughness of the built surfaces in Fig. 6.27-6.28. Surface Roughness measures Ra Rq Rz Figure 6.27- M1 0.0067 0.0088 0.016 Figure 6.27-M2 2dippings 0.0007 0.0009 0.0017 Figure 6.28- M1 0.0037 0.0044 0.0072 Figure 6.28- M2 0.0007 0.0011 0.0022 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.5 1 1.5 2 2.5 M1 M2 2Dippings mm mm 0 0.05 0.1 0.15 0.2 0 0.5 1 M1 M2 mm mm 146 6.10 Concluding Remarks This chapter presented a novel approach for building up-facing smooth curved surfaces in the mask-image-projection-based stereolithography process with both top-down and bottom-up projection configurations. A meniscus equilibrium method has been developed for building smooth meso/micro scale up-facing surfaces in both top-down projection configuration and bottom-up projection configuration. Based on the developed meniscus models in various scenarios, a process planning problem has been formulated. An optimization algorithm has been developed for identifying the optimal dipping numbers and meniscus points, as well as the corresponding mask images. With the optimization model, our approach is able to fabricate a part with minimum meniscus dipping numbers and satisfactory approximation error. A calibration and planning approach has also been developed for computing meniscus projection images that can accurately cure liquid resin in different Z heights. Both hardware and software were constructed to verify the meniscus-based approach in top-down and bottom-up projection MIP- SL systems. Experimental tests based on the developed methods have been performed and compared with the traditional methods. The differences in the build results illustrate the effectiveness of our method in improving the surface finish of up-facing curved surfaces in the MIP-SL and MIP-uSL processes. The hypothesis H4 is verified in this chapter: H4. A meniscus approach in MIP-SL system could deposit material with an accurate profile, thus to fabricate smooth up-facing surfaces in MIP-SL system together with the same or even faster build speed. Together with the gray scale image method presented in the previous chapter, it is able to completely solve the sub question in MIP-SL system: Q1.2 How to achieve high surface quality without compromising the build speed? 147 The gray scale image method presented in the previous chapter and the meniscus approach presented in this chapter both allows the use of a bigger layer thickness, while improving the surface finish greatly. The developed meso/micro MIP-SL systems are able to achieve fast build speed together with high surface quality by delivering the light energy fast and depositing the material accurately with the developed gray scale image method and meniscus approach. The progress of this research is shown in Figure 6.33. It is shown that we can further explore the research problem in MIP-SL systems by verifying the last hypothesis H5 about fast material deposition. 6 Energy Control Material Deposition Fast Delivery Accurate Delivery Fast Deposition Accurate Deposition CNC Accumulation MIP-SL H2: 3D surface reconstruction, Optimal tool path (Chapter 4) H1: Multi-tool, Process optimization (Chapter 3) H3: Gray- scale image (Chapter 5) ? H5 H4: Controlled meniscus Approach (Chapter 6) √ √ √ Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research Content: Figure 6.33: Research progress by this chapter and the following research work 148 Chapter 7 Fast Material Deposition in MIP-SL using Fast Recoating Method 7.1 Material Deposition Speed In Photo-polymerization AM As discussed in the previous chapter, although the research problem in MIP-SL system is fully solved by the gray scale image method and meniscus approach, we would like to explore further improvements of build speed and surface quality by investigating the fast material deposition methods in this chapter. 7.1.1 Non-layer Photopolymerization AM As discussed in Chapter 1 and Chapter 2, there are primarily two categories of non-layer photopolymerization AM process: (1) Two-photon Technology; (2) CNC accumulation. Figure 7.1: Illustration of two non-layer photo-polymerization AM processes Laser beam1 Laser beam 2 Resin surface Built part (a) Two Photon Technology Built part (b) CNC Accumulation Technology Optic Fibers LED Accumulation tool 149 As shown in Figure 7.1, there is no need for material recoating in the non-layer photo- polymerization AM systems. The material stay in the tank and get cured by the light intersection point directly in the two-photon technology. In the CNC accumulation process, the accumulation tool is moving inside the liquid tank and the liquid feeding is very simple, immediate and straightforward. Therefore, unlike the layer photo-polymerization AM processes, the performance of the non-layer photo-polymerization AM processes is primarily relied on the control of energy delivery. By delivering the energy rapidly and accurately, we could achieve high throughput and part quality in the non-layer photo-polymerization AM systems. There is no need to study the material deposition in non-layer photopolymerization AM systems. 7.1.2 Layer Photopolymerization AM In photo-polymerization process, the material is solidified very quickly in the area that is exposed to light with enough energy. After one layer solidification, a new layer of liquid material needs to be recoated on the solidified layer. Hence, aside from the energy input, how the material deposition is planned and controlled make significant influence in the fabricated part quality and building speed. In order to develop systems with higher speed and bigger fabrication capability, we need to have approaches for planning and controlling the energy input\delivery and material deposition both. In previous chapters, we used MIP-SL process as an example of the layer photo- polymerization AM process and presented methods for energy control in MIP-SL. In this chapter, we use MIP-SL process again as a typical example of layer photo-polymerization AM process to study the fast material deposition methods. In MIP-SL system, the material deposition speed is mainly dependent on material properties and layer thickness. With a smaller layer thickness and a more viscous liquid resin, the material 150 deposition would take longer and be more difficult. Compared to the non-layer photo- polymerization process, like CNC accumulation, the material recoating takes up a major portion of the building time. In the following sections, a fast material recoating approach is proposed and developed for the layer photopolymerization AM processes like MIP-SL. 7.2 Fast Recoating Approaches for meso-scale MIP-SL 7.2.1 Recoating in Top-down and Bottom-up Projection configurations In MIP-SL process, there are two ways of material deposition: (1) Free surface approach which usually uses top-down projection as shown in Figure 7.2a. There is no constraint on the liquid surface and the material recoating is done by dipping the cured part down and waiting for a while until the liquid settle down under gravity. (2) Constrained surface approach which usually uses bottom-up projection as shown in Figure 7.2b. The mask image is projected onto the bottom of a transparent tank. After a layer is cured at the bottom of the built part, the platform is moved up and then down to form a small gap with the bottom surface of the resin tank. A uniform thin layer can be achieved after the formed gap is filled with liquid resin. A bottom-up projection based system has several advantages over a top-down projection based system: (1) The container depth is independent of the part height. Thus a shallow vat can be used to reduce the required volume of the liquid resin. During the building process liquid resin can be added by a pump when needed. (2) Recoating is achieved by constraining liquid resin between the previously cured layers and the resin tank. Hence no additional sweeping is needed for flattening the resin surface. 151 (3) Much smaller layer thickness can be achieved since the gap size is only determined by the Z stage resolution regardless of the fluid properties of liquid resin. (4) The curing of liquid resin is sealed from the oxygen-rich environment. By eliminating the oxygen inhibition effect, the liquid photopolymer resin can be cured faster. Figure 7.2: A schematic diagram of MIP-SL system: (a) top-down projection (b) bottom-up projection Despite the advantages, the bottom-up projection based system has not been widely used in the SLA and MIP-SL processes. A main reason is that the separation of the cured part from the tank surface is difficult. That is, in the bottom-up projection based MIP-SL process, a cured layer is sandwiched between the previous layer and the resin vat. The solidified material may adhere strongly to the corresponding rigid or semi-rigid transparent solidification substrate, causing the object to break or deform when the build platform moves up from the vat during the building process. One approach to prevent the detachment of a cured layer from the built part is to increase its exposure such that the cured layer can strongly bond to the previous layers. However, such over- curing will also lead to poor surface quality and inaccurate dimensions. Another approach to address the problem is to apply a certain type of coating on the resin vat to reduce the attachment force of a cured layer. Suitable coatings, including Teflon and silicone films, can help the separation of the part from the vat [40, 101]. A coated Teflon glass has also been used in the D M D Lens M irror Tank Z stage Platform R esin Sw eeper C om puter Platform Z stage Tank Mirror Lens Fixture Resin Computer (a) (b) 152 machines of Denken[102] and EnvisionTec[103]. However, even with the intermediate material, the separation force can still be large. Huang and Jiang[101] investigated the attachment force for the coating of an elastic silicone film. Based on a developed on-line force monitoring system, test results indicate that the pulling force increases linearly with the size of the working area. Experiments indicate that, for a square of 6060mm, the pulling force to separate the part from the film is greater than 60 N. Such a large attachment force between the cured layer and the vat is a key challenge that needs be addressed in the bottom-up projection based MIP-SL process. We address the large separation force and the related speed problem by developing a two- way movement method. For the purpose of easy sliding between the built part and the resin tank, we used another type of coating material, Polydimethylsiloxane (PDMS, Sylgard 184, Dow Corning). The coating selection is based on the ability of PDMS in inhibiting free-radical polymerization near its surface, as shown by Dendukuri et al [68, 104]. In their research, it was identified that a very thin oxygen-aided inhibition layer (~2.5 m) is formed that can prevent the cured layer from attaching to the PDMS film. Thus cured layers can easily slide on the PDMS film. Based on the approach, a fast MIP-SL process has been developed that can build a CAD model in minutes. The following sections present the following contents: Part separation study based on the PDMS film, a two-way movement approach to reduce the part separation force, the process settings and the related building time analysis and the experimental results of multiple test cases. 7.2.2 Models of Part Separation Forces in Recoating Process In order to separate the cured part from the PDMS film, a simple and intuitive approach is to directly move the platform up a certain distance d and then down by d-d LT where d LT is the layer thickness. Since one layer thickness is usually very small (50-200 m), the distance d is 153 usually much larger in order for the resin to fully fill the gap (e.g. 5mm). We studied the part separation force of a cured layer from a coated PDMS glass based on such movements. The PDMS film thickness is set at 1mm. A set of physical experiments have been designed and performed to understand the separation force based on such an approach. Figure 7.3: Experimental setup for studying part separation forces in the MIP-SL process. Figure 7.3 shows the setup for measuring the pulling-up force. Two FlexiForce sensors (Tekscan, South Boston, MA) with a range of 0-25 lbs are sandwiched between the fixture and the vat. The two sensors are connected to a microcontroller, which can sample and record the sensors’ readouts at over 3 k Hz. Since the vat is free at the bottom and the side, and only fixed at the top, the pulling force by the part will be transferred to the sensors when the platform rises. In the experiments, we first use a given mask image to build a certain number of layers (e.g. 25 layers). The layer thickness is set at 0.2mm. We then begin to record the separation force in the building process of the next few layers. For each layer, after the designed mask image has been exposed for a certain time, the platform is raised up slowly at 0.6mm/sec for 5mm and the related readouts of the sensors are then recorded. Platform Z stage Tank Resin Part Fixture PDMS Different projection patterns with the same area Mask image projection Move the platform up by Z Force sensor Controller 154 In our study, we considered three factors that may affect the separation force including (1) exposure time, (2) image area, and (3) image shape. To understand the effects of these factors, designed experiments were conducted. Seven projection patterns were used for testing the effect of image shape. They are shown in Figure 7.3, which include circle band, hexagon, t-shape, square, star-shape, triangle, and u-shape. For comparison, all the projection patterns have the same area in the tests. The separation forces of a cured layer were measured based on each of the seven projection patterns. Figure 7.4 shows the measured separation forces of a sensor for different test cases. The horizontal axis indicates the distance in the Z direction (in the unit of 10 m), and the vertical axis indicates the measured pulling force (in ounces). (a) T= 1sec, Area = 625 mm 2 (b) T= 0.5 sec, Area = 625 mm 2 -20 0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 Position (* 10um) Force (oz) Triangle Square Hexagon Tshape Ushape Band Star -10 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 Position (* 10um) Force (oz) Triangle Square Hexagon Tshape Ushape Band Star 155 (c) T= 1sec, Area = 156 mm 2 Figure 7.4: Pulling-up forces of a cured layer from a PDMS film in different settings. It can be observed from the experimental results that: (1) As the Z stage moves up, the separation force increases gradually. After the cured layer is detached from the PDMS film, the separation force will drop rapidly from the peak value to 0; (2) Due to the flexibility of the PDMS film, the pulling-up force is rather small within a moving distance that is less than 200m. (3) The peak force gets larger when the same mask image is exposed longer; (4) The peak force gets larger when a larger image area is projected; (5) The image shape has more complex effects on the peak force. In addition, their effects may interact with the exposure time and the projection area; (6) With the coated PDMS film on the vat, the separation force is still considerably large (~100 oz or 27.8 N for an image area of 625mm2 with 1 second exposure). (7) As the Z stage moving acceleration and velocity gets bigger, the separation force increases significantly. The experiment results indicate that the suction force between the cured layer and the PDMS film is large during the pulling-up process. Such a large force on the cured layer may cause the building process to fail if the bonding force between the current layer and previous -10 0 10 20 30 40 50 0 10 20 30 40 50 60 70 Position (* 10um) Force (oz) Triangle Square Hexagon Tshape Ushape Band Star 156 layers is smaller than the suction force. In addition, after building multiple layers, such forces on the PDMS film may lead to cracks in the film due to material fatigue caused by the cyclic loading. Hence, huge separation force during the recoating process is the main problem for bottom- up projection mechanism. Also, a small acceleration and velocity is required for successful separation, which lower down the building process significantly. To facilitate a high-speed MIP- SL process based on the bottom-up projection, we develop a novel two-way movement design for material recoating. The two-way movement design, as discussed in the following sections, can effectively address the large separation force that is problematic, while achieving a fast building speed at the same time. 7.2.3 Fast Recoating Process Design for Meso-scale MIP-SL process The developed fast recoating approach based on two-way movement design is motivated by the following observations: (1) As shown in Figure 7.4, the pulling-up force is negligible for the Z movement of a small distance (e.g. 50 or 100 m) based on the 1mm thick PDMS film. (2) As demonstrated in[104], the oxygen-aided inhibition around the PDMS surface leaves a non-polymerized lubricating layer near the PDMS film. Therefore the cured layer can easily slide on the PDMS surface. X stage Platform Z stage PDMS Z Translation stage Cured resin Force sensor Vat Frame Motion Resin DMD Lens Mirror (1) (2) (3) (4) (2) (4) (3) DMD Lens Mirror DMD Lens Mirror X Translation stage X Translation stage LENS Mirror Step 1: Move up platform △z (1 layer) Tank Cured resin Liquid resin Layer thickness PDMS X stage Platform Z stage PDMS Z Translation stage Cured resin Force sensor Vat Frame Motion Resin DMD Lens Mirror (1) (2) (3) (4) (2) (4) (3) DMD Lens Mirror DMD Lens Mirror X Translation stage X Translation stage LENS Mirror Step 2: Move tank left/right △x Tank 157 Figure 7.5: The MIP-SL process based on the two-way movement design with PDMS. An illustration of the process is shown in Figure 7.5. In our method, a transparent PDMS film is first applied on the bottom surface of a transparent glass vat. (1) After a mask image is exposed to cure a layer, the platform is moved up in the Z axis for one layer thickness (e.g. 50 m). Accordingly, the regions of the PDMS film related to the shape of the cured layer will be pulled up by the suction force. However, the force is small due to the super elasticity of the PDMS film. Note that there is no liquid resin between the cured layer and the PDMS film at this moment. (2) The tank is moved along the X axis for a certain distance x. A good property of the PDMS film is that a very thin oxygen-aided inhibition layer (~2.5 m) is formed near the PDMS film that can provide a non-polymerized lubricating layer for easy sliding[104]. If the moving distance is sufficiently large (e.g. larger than the extent size of the cured layer in the X axis), the elastic deformation of the pulled-up PDMS film will be released by such a sliding movement. Hence, at the end of the X movement, liquid resin will be filled in the small gap between the cured layer and the PDMS film. (3) The mask image of a new layer can now be projected at the bottom surface to cure the next layer. These three steps can then be repeated by moving the tank in an opposite direction. Note that, to achieve the motion in the X direction, we only move the tank and the related frame. X stage Platform Z stage PDMS Z Translation stage Cured resin Force sensor Vat Frame Motion Resin DMD Lens Mirror (1) (2) (3) (4) (2) (4) (3) DMD Lens Mirror DMD Lens Mirror X Translation stage X Translation stage LENS Mirror Step 3: Projecting mask image Tank 158 There is no relative motion between the platform and the projection device. Hence the XY accuracy of the MIP-SL system will not be affected by the X translations. To verify the proposed two-way movement design, a set of experiments were conducted based on the setup as shown in Figure 7.3. The same set of mask patterns were used in building test layers. The same exposure time and layer thickness were used (1 second and 0.2mm respectively). The building process as shown in Figure 7.5 was first used in building a set of layers. In the tests, the tank was translated in the X axis by 20mm. The moving speed is set at 25mm/sec. After the layers have been built, the pulling-up forces in the Z axis during building the next layer were recorded. However, instead of curing a new layer as shown in Step 3, the part is moved up slowly at 0.6mm/sec for 2.5mm. The measured forces of a sensor in the Z axis during the aforementioned three steps are shown in Figure 7.6. In each figure the curves record the test results based on a sampling resolution of 80 milliseconds. (a) T= 1sec, area = 625 mm 2 -4 -2 0 2 4 6 8 10 12 14 16 18 20 0 10 20 30 40 50 60 70 Sampling time (* 20 msec) Force (oz) Triangle Square Hexagon Tshape Ushape Band Star Sampling time (x80 msec) Move the platform up by 0.2 mm Move the tank in X by 20 mm Move the platform up by 2.5 mm 159 (b) (b) T= 1 sec, area = 156 mm 2 . Figure 7.6: Pulling-up forces of a cured layer based on the two-way movement design in different settings. The figures show that the force in the Z direction is rather small when the platform is moved up by 0.2mm. During the remaining two steps (i.e. sliding on the PDMS film and the platform pulling-up), the peak separation forces are also relatively small (around 2-6 oz or 0.56- 1.67 N). Such measured forces are only 3-4% of the related ones as shown in Figure 7.4 Hence the two-way movement design can effectively reduce the large separation force in the bottom-up projection system. In the two-way movement design, cured layers can easily slide on the PDMS surface. The FlexiForce sensors were used in a modified setup to measure the shearing force in the X direction. However, no meaningful readouts were recorded from the sensors. To quantitatively estimate the value of the shearing force in the X axis, a set of square rods with different sizes were built using the two-way movement design. The built rods shown in Figure 7.26 are 10mm tall. The minimum cross section size is 0.4× 0.4mm. Note that we also successfully built rods with even smaller sizes. However, the rods were so fragile that they lost the mechanical strength to sustain themselves when the part was taken out of the resin vat and washed in isopropyl alcohol. -4 -2 0 2 4 6 8 10 12 14 16 18 20 0 10 20 30 40 50 60 70 Sampling time (* 20 msec) Force (oz) Triangle Square Hexagon Tshape Ushape Band Star Sampling time (x80 msec) Move the platform up by 0.2 mm Move the tank in X by 20 mm Move the platform up by 2.5 mm 160 Nevertheless, for a rod with a size of 0.4× 0.4mm, the maximum tangential force that can be added on it can be analytically estimated. As shown in Figure 7.7, the testing rods in the experiment can be modeled as a cantilever beam. Suppose the length of the beam is L, the size of the beam section is b×b, the force in tangent direction is F. The maximum bending stress at the end can be calculated as: / Mc I   , where I is the section modulus, 4 /12 Ib  , and /2 cb  . Substituting these values for their variables, the resultant equation is 3 6FL b   . Suppose the allowable blending stress is []  and the minimal beam section size is   b . We will have the following equation: 3 [ ][ ] 6 b F L   . The material used in our tests has the following parameters: []  =65MPa, [] b =0.4mm, L=10mm. According to the equation, the upper bound of the tangential force is only 0.07N or 0.25oz. Compared with the separation force in the Z direction, the shearing force in the X direction is rather small. Figure 7.7: Shearing force verification test. 7.2.4 A Fast Meso-scale MIP-SL Process and Its Building Speed Analysis The two-way movement design enables the quick spreading of liquid resin into a uniform thin layer. In addition, the DMD-based digital mask projection enables the fast curing of the spread liquid resin into a desired solid layer. Consequently, for a given 3D CAD model, a fast 0.8mm 161 MIP-SL process can fabricate a physical meso-scale object within a short building time. The curing characteristics and the two-way movement settings of the developed MIP-SL process are presented as follows. A detailed analysis of its building time is also discussed. Curing Characteristics There are two types of photopolymer systems, acrylate chemistry and cationic photopolymerization, in the SLA process[105]. Acrylate chemistry polymerizes via a free-radical mechanism while cationic photopolymerization undergoes ring-opening reactions in the presence of cationic photoinitiators. The monomer propagation for cationic reactions requires relatively higher activation energy. Consequently, the photopolymerization speed of acrylate-based photopolymers is higher due to the lower activation energy for free-radical reactions. Considering the photopolymerization speed difference, we selected the photopolymer resins based on acrylate chemistry for the developed fast MIP-SL process. As shown in previous section, our projection system can cure a layer within a short exposure time (<500 milliseconds). Such a fast curing time contributes to the desired fast MIP-SL process for building 3D objects in minutes. After an image is exposed for a certain time (T projection ), a waiting time, T wait_projection , is required before the layer can be moved up for one layer thickness (i.e. Step 1 in Figure 7.5). Such a waiting time is critical in order for the acrylate resin to complete the solidification process and gain sufficient strength for the Z movement. Otherwise, the building process may fail. The waiting time is dependent on the resin’s curing property. Due to the fast photo polymerization speed of the acrylate resins, the waiting time in our system is short (~300 milliseconds in our tests). Two-way Movement Parameter Settings 162 In the two-way movement design, the cured part is first moved up for one layer in the Z axis and the tank is then translated in the X axis for a certain distance. The two linear movements have different accuracy and speed requirements. (1) The Z movement needs to be accurate since it will determine the layer thickness of the next layer. The Z stage also needs to have a resolution that is much smaller than a layer thickness. Accordingly, to ensure the desired accuracy and resolution, we set small acceleration and velocity values in the Z movement. The slow movement of the cured part also enables the PDMS film to fully elastically deform for a small attaching force. However, the movement time in the Z axis (T Z ) is still reasonably short (~0.4 second in our tests) since only a small moving distance is required (e.g. 50 or 100 m). (2) The tank needs to be moved in the X axis for a certain distance to release the elastic deformation of the PDMS film. The X moving distance is related to the shape and size of the cured layer, and less than the extent size of the cured layer in the X axis. Since the relative position of the platform and the projection system will not change during the X movement, the accuracy and resolution requirements on the X movement are not as high as those on the Z movement. Hence a much larger acceleration and velocity can be applied in the X movement to reduce the movement time in the X axis (T X ). In our testbed, we used a Z linear stage with a thread of 0.5mm/round, and a X linear stage with a thread of 25.4mm/round. The moving time for different displacement distances in our prototyping system was calibrated for both linear stages. The results are plotted as blue and red lines in Figure 7.8 for the Z and X stages, respectively. As shown in the figure, the movement in the Z axis is much slower than that in the X axis. The movement time required to complete a moving distance in the X axis can be identified based on the calibration data. 163 Figure 7.8: The movement time in the X and Z axes in our prototyping system. After the X movement, another waiting time, T wait_X , is required in order for the flowing liquid resin to settle. Otherwise, the building process may fail. The waiting time caused by the X motions is related to the movement distance and the moving speed. Due to the small gap between the cured part and the PDMS film, the required waiting time is typically short (~100 milliseconds). After the waiting time of T wait_X , the liquid resin forms a uniform thin layer, which is ready for the next layer to be built. The process can then be repeated after the related mask image is exposed. Figure 7.9: The building time of a layer in the two-way movement based MIP-SL process. The Building Time of a Layer As shown in Figure 7.9, the building time of each layer is thus the sum of all the aforementioned steps: T Layer = T Projection + T wait_Projection + T Z + T X + T Wait_X . 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Movement Distance (inch) Required time (s) Moving Time vs. Distance (Z stage) Moving Time vs. Distance (X stage) Mask image projection T Projection Platform Z movement T Z Vat X movement T X T wait_Projection T wait_X T layer 164 The first two items, T Projection and Twait_Projection, are related to the curing characteristics of the photopolymer resins used in the MIP-SL process. The photopolymer resins based on acrylate chemistry can be quickly cured. A stronger light source used in the projection system can further reduce the projection time T Projection . The other three items, T Z , T X , and T wait_X , are related to the two-way movement design. T Z is related to the layer thickness and the moving velocity in the Z axis. T X is related to the size of the cured layer and the moving velocity in the X axis. A linear stage with a higher speed can be used to further reduce the movement time T X . T wait_X is determined by the gap distance between the PDMS and the cured layer, the moving velocity of the tank, the shape of the cured layer, and the flow properties of the liquid resin. For a typical layer thickness that is usually small, T wait_X is reasonably short (~100 millisecond in our tests). Note that the projection time T Projection for the first few layers is much longer (e.g. 3-4 seconds) to ensure the initial layers can be strongly bonded to the build platform, For all the other layers, the total building time of a layer is usually short (a few seconds in our tests). Hence a fast fabrication speed can be achieved in the developed process (e.g. building 3mm height per minute). 7.3 Experimental Study of Fast Recoating Approach for Meso-scale MIP-SL process 7.3.1 Hardware System A prototype system has been built to verify the developed process. The hardware setup of the fast MIP-SL system is shown in Figure 7.10. In the designed system an off-the-shelf projector (CASIO XJ-S36) was used. The optical lenses of the projector were modified to reduce the projection distance. Various projection settings including focus, key stone rectification, brightness and contrast were adjusted to achieve a sharp projection image on the designed projection plane. The DMD resolution in our system is 1024× 768 and the envelope size is set at 165 48× 36 mm. A precise linear stage from Aerotech Inc (Pittsburgh, PA) is used as the elevator for driving the platform in the Z axis. A fast linear stage from Servo Systems Co. (Montville, NJ) is used to drive the tanks back and force in the X axis. A high performance 4-axis motion control board with 28 Bi-directional I/O pins from Dynomotion Inc. (Calabasas, CA) is used for driving the linear stages. A flat and clear glass Petri dish is used as resin tank. A PDMS film (Sylgard 184, Dow Corning) is coated on the glass dish. 7.3.2 Software System A mask planning testbed has been developed using the C++ language with Microsoft Visual C++ compiler. The testbed integrates the geometry slicing and the motion controlling. It also synchronizes the image projection with the X and Z movements. The graphical user interface (GUI) of the developed software system is shown in Figure 7.11. The flowchart of the fast MIP- SL process is also shown in Figure 7.11. Figure 7.10: The prototype hardware system Figure 7.11 Related software system Z stage x stage Power supply Control board Projector Tank Resin Platform Load geometry Move platform to home Load next image Expose mask image Wait t wait_projection Move up platform for one layer Move the tank by the current layer’s extent size in the X axis Wait t wait_X Finished building? END Yes No Slice geometry Control panel Motion parameters Building process report Geometry slicing Geometry 1 Geometry 2 prepare the motion and images build base using single material process finished building? END Y N current layer only A only B A + B build A wash A and dry A build B next layer A or A+B clean B and Dry B dock at CA only B dock at CB build A next layer A or A+B clean A and dry A dock at CB only B dock at CA build B next layer A or A+B clean B and dry B dock at CA only B dock at CB Geometry Slicing Control Panel Motion Parameters Building Process Report 166 7.3.3 Materials Perfatory™ SI500 (yellow) and Acryl R5 (red) from EnvisionTec Inc. ( Ferndale, MI), were used in testing the developed fast MIP-SL process. Both resins belong to Acrylate. For curing depths of 0.05mm and 0.1mm, the exposure times for SI500 based on our projection system are set at 0.3 sec and 0.45 sec, respectively. The exposure times for Acryl R5 are set at 0.4 sec and 0.55 sec for curing depths of 0.05mm and 0.1mm, respectively. 7.3.4 Experimental Results and Discussion Tests have been performed to verify the building speed of the developed prototyping system. The results of the designed tests have demonstrated that the presented MIP-SL process can build 3D models in minutes instead of hours. A set of CAD models with different complexity was used in our tests. The screenshots of six input CAD models are shown in Figure 7.12.a-7.17.a. The related STL files have triangle numbers ranging from several hundreds to 1.2 million (refer to Table 7.1). Two different layer thicknesses commonly used in the MIP-SL process were tested. A 50μm layer thickness was used in the fabrication of a gear model. The mask image projection time was 0.35 second for each layer except the base. The projection waiting time was set at 0.1 second. For all the other models, a 100μm layer thickness was used in their building processes. Due to the larger layer thickness, a longer image exposure and projection waiting times were used (0.45 and 0.3 second respectively in the tests). Accordingly the Z movement will also take a longer time for a larger layer thickness. In the tests, the movement time in the Z axis (T Z ) is 0.32 and 0.42 second for the layer thickness of 50μm and 100μm, respectively. The required moving distance in the X axis is related to the size and shape of the cured layer. For a layer with a big cross-section area (e.g. the models of a head and a statue), the X translation 167 distance is set to a value that is close to the X extent size. Due to the large movement, the X waiting time was also set longer. In comparison, for a layer with a small cross-sectional area (e.g. the models of a hearing aid shell and the top portion of a brush), the X translation distance can be much smaller than the extent size of the layer in the X axis. However, due to the fast moving speed in the X axis, the differences on T X are usually small (less than 1 second as shown in Table 7.1). Two types of resins, SI500 and Acryl R5, were tested. Their curing characteristics are slightly different. For the same layer thickness, the curing of Acryl R5 takes ~0.1 second longer than that of SI500. The viscosities of the two resins are also slightly different. However, the same settings can be used in the two-way movement design based on the two resins. Figure 7.12: A test of a gear: (a) CAD model; (b) built objects in two liquid resins. Figure 7.13: A test of a head: (a) CAD model; (b-c) two views of the built object. 168 Figure 7.14: A test of a statue: (a) CAD model; (b-e) two views of the built objects in two liquid resins. Figure 7.15: A test of teeth: (a) CAD model; (b-c) built objects in two liquid resins. Figure 7.16: A test of a hearing aide shell: (a) CAD model; (b-c) two views of the built object. 1.6mm (a) (b) (c) 169 Figure 7.17: A test of a brush: (a) CAD model; (b) built object. Figure 7.12-7.17 show the built objects based on the developed fast MIP-SL process. The quality of the built objects was examined to be satisfactory. Both surface finish and dimension were analyzed to be acceptable. In our prototyping system, the nominal size of a pixel is 47μm. The fine image resolution enables the meso-scale features (i.e. in the range of 0.1-1 mm) to be well captured in the built physical objects, e.g. the lip of the human head, the cloth folds in the Beethoven statue, and the dentures in the teeth model. 7.3.5 Building Time Analysis All the models shown in Figure 7.12-7.17 were built within 12 minutes using our prototyping system. The models with less than 100 layers (e.g. the gear, the teeth, and the brush) only require 2-3 minutes to be built. A statistic of the building time is given in Table 7.1. A much larger exposure time (e.g. 4-5 seconds) was required for the first few layers in order to build a base. Consequently the built objects and the build platform can be well bonded. For all other layers, as shown in Figure 7.18, the building time of a layer (T Layer ) in our MIP-SL process is only 1.4-2.5 seconds. The variation on T Layer is mainly due to different layer thicknesses and the X moving distances. For an average of 2 seconds per layer and a layer thickness of 0.1mm, the building speed of the developed MIP-SL process is ~3mm per minute, or 180mm per hour. To the best of our knowledge, such a MIP-SL process is one of the fastest layer-based additive manufacturing processes that have been developed. A video of building the gear model as shown in Figure 7.12a can be found in [106]. 170 Table 7.1: Building time statistics Model Gear Head Statue Teeth Shell Brush Figure # Fig.7.12 Fig.7.13 Fig.7.14 Fig.7.15 Fig.7.16 Fig.7.17 Tri # 660 24190 5204 133806 32762 1259246 Size X (mm) 25.4 25 17.7 24 24 7.6 Thickness (mm) 0.05 0.1 0.1 0.1 0.1 0.1 T projection (sec) 0.35 (0.4) * 0.45 0.45 (0.55) * 0.45 (0.55) * 0.45 0.45 T wait_projection (sec) 0.1 0.3 0.3 0.3 0.3 0.3 T Z (sec) 0.32 0.42 0.42 0.42 0.42 0.42 Move X (mm) 7.6 25 12.7 2.5 2.5 6.35 (1.3) + T X (sec) 0.58 1.1 0.67 0.46 0.46 0.56 (0.39) + T X_Wait (sec) 0.05 0.1 0.1 0.05 0.05 0.05 T Layer (sec) 1.4 (1.45) * 2.37 1.94 (2.04) * 1.68 (1.78) * 1.68 1.78 (1.61) + Height Z (mm) 4.93 28.5 30.5 7.3 22.3 8.3 Layer # 98 285 305 73 223 83 T total_building (min) 2.31 (2.39) * 11.26 9.86 (10.4) * 2.04 (2.17) * 6.24 2.25 * The projection and waiting time for R5 resin; others are for SI500. + Different X moving distances were used in building the bottom (1 st -19 th layers) and the top (20 th -83 th layers) portions of the brush. Figure 7.18: Layer building time of the test cases. Brush_Top Gear Gear_red Statue_Red Teeth_Red head statue Teeth, Hearing Aid Brush_Bot 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 0 5 10 15 20 25 30 X Movement Distance (mm) Layer Building Time (sec) 171 It is illustrated in the experimental results that the separation force and the sliding force are relatively small during the two-way movement process. The motions related to the two-way movement design can also be performed quickly. The MIP-SL process developed based on such a recoating approach can achieve high fabrication speed for input CAD models. The experimental results demonstrate that the newly developed MIP-SL process can successfully fabricate 3D objects with satisfactory quality in a short time (usually in minutes). It is a big step of improvement in building speed of photo-polymerization AM process. 7.4 Fast Recoating Approaches for Micro-scale MIP-SL When the fabrication scale comes from meso-scale to micro-scale, the energy input control principle and methods keep the same because the fundamental curing mechanism doesn't change. So the energy control for micro-scale fabrication is not repeated in previous chapters. However, because of the difference in liquid flow mechanisms, material deposition for micro-scale MIP-SL process is distinct from the meso-scale problem. The fast recoating approaches for meso-scale MIP-SL is studied and extended for micro-scale fabrications in MIP- SL. The modified fast recoating approaches for micro-scale MIP-SL is presented in this section. 7.4.1 Models of Part Separation Forces in Micro-scale Fabrication To separate the cured part from the vat, the build platform may move up in the Z axis or slide in the X axis. In Section 7.2, we used tools measured the pulling-up force in meso-scale MIP-SL fabrication systems. However, when it comes to the micro-scale fabrication systems, the pulling-up force is not readable in the measurement tool but is still big enough to damage the micro features. The fabricated part quality is very sensitive to the forces. Therefore, a purely experimental methodology based on trial and error design is not desirable for the study of micro- scale photopolymerization processes. 172 In order to understand the related separation forces and recoating time, the governing equations of both direct pull-up and sliding forces are derived in this section. Forces in Pulling-up and sliding process After a layer is cured, the Z stage is moved up to separate the cured layer from the vat surface. A gap between the cured layer and the vat surface is formed and the liquid flow will fill the gap during the process. As shown in Figure 7.19, a pulling-up force FZ occurs during the separation process for a given velocity VZ (velocity in the X axis VX = 0). Huang and Jiang[101] identified the pulling-up force is a function of the area of the cured part. He further suggested that the force can be predicted by the pulling coefficient of the vat surface, and the value of plane- strain crack initiation toughness. Figure 7.19: Forces during moving up process and sliding process To simplify the model, we analyze the separation force in the pulling up process. As shown in Figure 7.19, the cured part needs to overcome force FZ, which is related to the pressure difference and pulling up viscosity. To analyze the analysis, we assume: (1) The cured layer is a circular plane; (2) The problem is an incompressible flow problem. The pressure driven flow Q and the flow due to pulling-up movement Q' can be described as: Vz △P F Z h stationary Liquid Air Cured part PDMS Film A A’ z x F X Vx 173 dr dp rh Q ×    12 2 3 (7.1) dt dh r Q 2 '   (7.2) Since the pressure difference is created by pulling-up movement, we can consider pressure driven flow equivalent to pulling-up flow, that is: ' Q Q  , thus: rdr dt dh h dp × ×  3 6 (7.3) By integrating the pressure, we got: ) ( 3 2 2 3 r R dt dh h P P R r  × ×    (7.4) where is the pressure at radius r of the circular plate, is the pressure on the edge of the circular plate which can be approximated as the atmosphere pressure , μ is the fluid viscosity, h is the gap size, is the pulling up speed. Set R as 5mm, PR as the atmosphere pressure 0.98 105Pa, μ as 180cP, h as 50um and as 160um/s, the pressure distribution calculated from Eq. 7.4 is shown as in Figure 7.20. Integrating the pressure on the plate yields: 3 2 3 4 0 48 . 0 2 3 2 ) ( h A v h R v rdr P P F z z R r a z     ×   ×    (7.5) where A is the dimension size R 2 . Note that the separation force Fz is nonlinear with the part size A: 2 A F z  (7.6) 174 In comparison, the Z stage can also be moved in the X axis before moving up in order to separate the cured layer from the vat surface [84]. During the sliding process, the shearing force in the X axis mainly comes from the viscosity, which can be modeled as: h Av dx z v F x X /        hence, A F X  (7.7) (a) Pressure distribution with R=5mm (b) Contour plot of the pressure shown in (a). Figure 7.20: Pressure distribution Maximally Allowed Forces in the X and Z Direction To prevent the detachment of a cured layer from the built part, the separation force Fz in the pulling-up process should be less than the bonding force of the cured layer and the built part. Otherwise, the cured layer will be detached and the related features will not be shaped. (a) Failed pillars due to insufficiently curing (b) pillars built by sliding (left) and up-down motions (right) Figure 7.21: Illustrations of built pillars. -5 0 5 x 10 -3 -5 0 5 x 10 -3 8 9 10 x 10 4 X(m) Y(m) Pressure(Pa) 83600 83600 83600 83600 86480 86480 86480 86480 86480 89360 89360 89360 89360 89360 89360 92240 92240 92240 92240 92240 92240 92240 95120 95120 95120 95120 95120 95120 95120 95120 X(m) Y(m) -5 0 5 x 10 -3 -5 0 5 x 10 -3 (a) (b) 0.5mm b b A-A F P A A (a) Failed pillars F B L A A 0.5mm 0.5mm (b) 0.5mm b b A-A F P A A (a) Failed pillars F B L A A 0.5mm 0.5mm (b) 175 Figure 7.21 shows some examples of failed pillars. For them, the force required to reach the break point can be modeled as: A F t p ×   (7.8) where σt is the bonding strength of the newly cured layer with the previous layer, which is determined by the curing conditions and A is the beam section area. Hence the maximally allowed separation force Fz before breaking follows: A F z  max_ (7.9) By considering the maximum bending strength in x direction experienced by the beam (refer to Fig. 4b): ) 4 /( / 4 R R FL I Mz   ×   (7.10) Thus the maximum allowed bending force FB can be expressed as: ) 4 /( 3 L R F B   ×  (7.11) where σ is the flexural strength, M is the bending moment, L is the length of the beam, z is the distance of the neutral axis from the point of interest and I is the area moment of inertia. By substituting those parameters by their maximum approximate values, we are able to approximate the maximum allowed force in x direction according to equation 7.11. It's shown that the maximum shearing force in the X axis follows: 5 . 1 max_ A F x  (7.12) As shown in Equations (7.6), (7.7), (7.9) and (7.12), the separation force and the maximum allowed forces in the Z and X directions have different relationships with the surface area. As shown in the above equations, the separation force and the maximally allowed forces along the Z and X directions have different relationships with the surface area A. 176 (a) and when the cured part is moved in the Z axis; (b) and when the cured part slides along the X direction. Substituting viscosity µ by 180cP, h by 50um, initial h which is the gap distance in the beginning of sliding by 2.5um, moving speed vz by 160um/s, sliding speed vx by 40mm/s, and assuming thatσt is 78100Pa andσis 20000Pa, the according trends of the forces along the Z and X directions are shown in Figure 7.22. For a decreasing part size A, the maximally allowed force decreases much faster than the separation force ; while the separation force decreases much faster than the maximally allowed force . Consequently, a recoating method based on the sliding movement in the X axis is preferred for the meso- and macro-scale MIP-SL process [84]. In contrast, a recoating method based on the direct pulling up-down in the Z axis is preferred for the micro-scale MIP-SL process. With the set values as described before, as shown in Figure 7.22(a), it indicates that when A is bigger than 0.7 10-3m2, that is, r is bigger than 15mm, sliding is necessary to assist the separation process since the pulling up force is bigger than the maximum allowed value. On the other hand, as shown in Figure 7.22(b), when A is smaller than 2.5 10-5m2, that is, when r is smaller than 2.8mm, the sliding will destroy the part, hence only pull-up movement is allowed to fabricate small features. Figure 7.22: An illustration of the separation forces in x and z directions with different sizes 0 0.5 1 1.5 x 10 -3 0 50 100 150 200 A (m 2 ) F (N) fz vs. A fmaxz vs. A 0 2 4 6 8 x 10 -5 0 0.1 0.2 0.3 0.4 A(m 2 ) F(N) fx vs. A fmaxx vs. A Pull up only Two-way Only Fz (N) Fx (N) (a) (b) 177 7.4.2 Fast Recoating Process Design for Micro-scale MIP-SL process Based on the aforementioned analysis result, a pulling-up movement design is used in developing a fast MIP-uSL process. Similar to the meso-scale MIP-SL system, a thin layer of Polydimethylsiloxane (PDMS, Sylgard 184, Dow Corning)is coated on the vat surface. This is mainly based on the ability of PDMS in inhibiting free-radical polymerization near its surface. Thus a very thin oxygen-aided inhibition layer (~2.5 μm) is formed near the PDMS film to prevent the bonding between the cured layer and the film. After a mask image is exposed, the platform is moved up slowly in the Z axis by a certain distance. Enough waiting time is necessary to ensure the complete filling of liquid resin in the gap. The mask image of a new layer can then be projected to cure the next layer. Since a significant portion of the MIP-uSL fabrication time is spent on the resin recoating, the process parameters including the moving-up distance and the delay time are investigated in order to achieve a fast building process. Experimental study of parameter settings A long waiting time and a big moving distance would guarantee the complete filling of liquid resin in the gap. However, it is desired to identify a short waiting time and a small moving distance in order to achieve a fast fabrication speed. A set of experiments based on different moving distances and surface areas have been performed to identify such parameter settings. In the experiments, cubes with various sizes are built by using different waiting time and moving distances. Figure 7.23 shows a cube of 6.96 × 6.96 × 2mm. If the built part has holes or deep shadows under a microscope, the waiting time is considered to be insufficient. Accordingly a longer waiting time will be used in rebuilding the part. A critical waiting time for void-free curing is identified for different Z moving distances (h s ). By changing the cube dimension, a new critical waiting time can be identified. Figure 7.24 shows the relation between the critical waiting 178 time, the cube size and the Z movement distance. In all the tests a same speed (0.002 in/s)is used in moving up the platform. It can be observed from Figure 7.24 that: (1) for the same gap height, the minimum waiting time increases with the dimensional size; (2) bigger gap height results in less minimum waiting time;(3) no waiting time is needed for a gap distance larger than 12 μm and a dimensional size smaller than 6 mm. Building time of a layer The building time of a layer is the sum of all the aforementioned steps: T layer = T projection + T wait_projection + T z + T wait_z , in which, T projection and T wait_projection are related to the curing characteristics of the photopolymer resins and the light intensity of the light source used in the system; T z is the time required for moving the platform one layer up in the Z axis; and T wait_z is the minimum waiting time in order for resin to fill the one layer gap. After an image is exposed for T projection , a waiting time T wait_projection is required before the layer being moved up by one layer thickness. Such a waiting time is critical for the acrylate resin to complete the solidification process and gain sufficient strength for the Z movement. Otherwise, the building process may fail. Due to the fast photocuring speed of the acrylate resin, T wait_projection in our system is short (~300 milliseconds in our tests). T wait_z is related to the gap distance, filling area size, moving velocity and conditions of the PDMS coating. A minimum of 0.1 second can be used to ensure the liquid resin to stay still. For various conditions, a minimum delay time can be added based on the fitted models as shown in Figure 7.24. 179 Figure 7.23: Test results for identifying the minimum moving distance and delay time. Figure 7.24: The flow filling time with different gap height. 7.4.3 Experimental Study of Support Settings in micro Fabrications Supports structures in the MIP-SL process are similar to fixtures in machining or scaffolds in construction. They are important for building parts with complex geometries. Appropriate support structures are especially critical for the bottom-up projection based MIP-SL process since all the parts are built upside-down in the fabrication process. Each cured layer needs to be securely anchored to previous layers; otherwise the building process would fail. In addition, note 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm fit (data1) data3: x=6.012mm fit (data2) fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm data3: x=6.012mm fit (data1) fit (data2) fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm data3: x=6.012mm fit (data1) fit (data2) fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm fit (data1) fit (data2) data3: x=6.018mm fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm fit (data1) fit (data2) data3: x=6.018mm fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) T vs hs Waiting time tw (s) Gap height hs ( x 10 -5 m ) 9 8 7 6 5 4 3 2 1 0 1 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm fit (data1) data3: x=6.012mm fit (data2) fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm fit (data1) data3: x=6.012mm fit (data2) fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm data3: x=6.012mm fit (data1) fit (data2) fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm data3: x=6.012mm fit (data1) fit (data2) fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm fit (data1) fit (data2) data3: x=6.018mm fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 x 10 -5 0 1 2 3 4 5 6 7 8 9 Gap height hs (m) Waiting time tw (s) T vs hs data1: x=6.985mm data2: x=6.452mm fit (data1) fit (data2) data3: x=6.018mm fit (data3) (1) Fitted result of data1: y(x) = (0.00004879 / 4 - 828.1994 k1 x^... k1 = 0.052664; k2 = 243.3 R = 0.9903 (lin) (3) Fitted result of data3: y(x) = (0.000041623 / 4 - 828.1994 k1 x... k1 = 0.046818;k2 = 2608.5 R = 0.99463 (lin) (2) Fitted result of data2: y(x) = (0.000046892 / 4 - 828.1994 k1 x... k1 = 0.045337; k2 = 1150.6 R = 0.99074 (lin) 180 that when the building volume of MIP-SL process come down to micro scale in MIP-uSL process, the part sizes are tens to hundreds times smaller than those of the macro-scale SLA process; however, the sizes of a support anchor in the MIP-SL process (e.g. 50μm rods are used in our micro-scale experiments) can only be several times smaller than those in the macro-scale MIP-SL process (e.g. 100-150μm rods are usually used). Consequently it is important to identify appropriate anchor positions for a given geometry and add a minimum amount of supports to avoid the support structures are too dense. No literature on the support generation for the SL process was found. For the macro-scale MIP-SL process, most previous work on the support generation is based on direct processing of polygonal models. A dominant support generation approach is to compare the orientation angle of a triangle in a polygonal model with a minimum supporting angle specified by a user and accordingly identify supporting regions [107-109]. However, the accordingly generated supports are usually dense since the approach does not take advantage of the self-supporting property of geometric features. For example, no supports are needed for a vaulted overhand or a small overhang as all the layers of such geometric features can be anchored to previously built layers in the building process. A contour-based support generation method and related algorithms have been developed for the MIP-SL process. Instead of analyzing polygonal triangles, our method is based on systematically analyzing the shape of sliced 2D layers and accordingly identifying support regions. Our method fully utilizes the self-supporting property of a geometric feature and can significantly reduce the number of supports that are required in the building process. The MIP-SL processes build physical objects layer-by-layer. Accordingly, the principle of our support generation approach is to analyze the sliced contours layer-by-layer for determining the required supports in the building process. Figure 7.25 shows a simple 1-dimensional example 181 to illustrate the contour-based support generation method. Suppose the relative sizes and positions of a current layer and its previous layer are known. All the previous layers have been built when the current layer is to be built. Obviously the portions of the current layer that directly contact the previous layer have been supported. In addition, certain neighboring areas are also supported by the previous layer (i.e. enlarging the previous layer by Dist Self_support as shown in Figure 7.25b). The value Dist Self_support of a region may be experimentally determined related to the size of the region. For the remaining portions of the current layer that have not been anchored by the previous layer (i.e. regions ii in Figure 7.25b), supports are required in order for them to be anchored. The additional pins that are added under a given part are called anchor supports in the paper. Such anchors have been fixed on the building platform or previously built layers when the current layer is to be built. Figure 7.25: Contour-based support generation method: (a) Given layers; (b) Layer analysis result. Anchor Anchors Current layer Previous layer (b) (a) Previous layer Self-supported regions by previous layer Dist self-support Dist self-support Current layer Regions in current layer to be supported by anchors Dist anchor Dist anchor Anchor Anchors Current layer Previous layer (b) (a) Previous layer Self-supported regions by previous layer Dist self-support Dist self-support Current layer Regions in current layer to be supported by anchors Dist anchor Dist anchor (a) (b) (i) (ii) 182 Assume when an anchor is added at position P, the circular region centered at P with a radius of Dist Anchor can be safely built. Accordingly all the regions (ii) are fully supported after a certain number of anchors have been added. The value of Dist Anchor can also be experimentally determined for anchors with given shapes and sizes. After all the layers of a 3D model have been analyzed, the CAD model of related supports including bases and reinforcements can be computed. Consequently the building process of given CAD models will be successful since all the layers are fully anchored either by the previous layer or the added anchor supports. In MIP-uSL support design, the support parameters cannot be simply scaled down directly from the meso and macro scale fabrication. Otherwise, it will be too dense and also too weak to support the structure. Therefore, a set of experiments were carried out to find the optimal support settings for the MIP-uSL process. Figure 7.26a shows values of some critical support parameters for MIP-SL process and MIP-uSL process in our setup. Figure 12b is a microscopic image of fabricated supports of a micro structure. Figure 7.26: Experimental study of support settings: (a) Parameter values (b) Fabricated support anchors (a) (b) 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.12 0.13 0.14 0.15 0.16 0 1 2 3 4 5 6 7 scaled-down values Micro Support Settings Macro Support Settings 1: SelfSupportDist 2: AnchorSupportDist 3: SupportDistRadius 4: AnchorCylinderRadius 5: DownFacingAnchorConeHeight 6: DownFacingAnchorConeTopRadius 0.5 mm Parameter 6: 50 um Parameter ID ( Inch) 183 7.4.4 Experimental Study of Fast Recoating Approach for Micro-scale MIP-SL As introduced in Chapter 5, a prototype MIP-uSL system has been built to verify the approaches presented in this research. The same prototype MIP-uSL system described in Chapter 5 is used here to verify the proposed fast recoating approach. As a reminder, the hardware setup and software is shown in Figure 7.27. Figure 7.27: Setup for fast recoating approach verification. Various tests have been performed to verify the capability of the develop system on micro- scale fabrication. The building speed of the developed process and the benefits of support structures have also been tested. Tests on Micro-fabrication Capability Several simple tests were carried out to verify the capability of the system on building micro structures or meso-scale parts with micro features. Two of the test cases are shown in Figure 7.28a and 7.28d. In the first test case, a set of pillars are designed between a top and bottom plate. The sizes of the pillars are gradually decreasing from left to right. With the aid of Slice geometry Finished building? Load geometry Move platform to home Load next image Expose mask image Wait twait_projection Move up platform for one layer Wait twait_filling End Yes No (a) GUI (b) Flow chart Geometry slicing Control Panel Building process report Motion Parameter Power supply Control board Tank Resin Z stage Projector Platform 184 the top and bottom plates, any successful built pillars can be kept after the post processing process. The test case is to identify the smallest pillars the prototype system can build reliably. In the second test case, a plate has a small gap in its center. A set of such plates with different gap sizes were built to identify the smallest gap sizes the prototype system can build successfully. Figure 7.28b and 7.28e shows two built objects with the microscopic images of the thinnest pillar (50 m) and the smallest gap (110 m). A test case of a gear was used to verify the capability of the MIP-SL system in fabricating micro-scale curved surfaces. The CAD model of a gear with a diameter of 3mm is shown in Figure 7.29a. The built physical object is shown in Figure 7.29b. The microscopic image of some gear teeth is given in Figure 7.29c, which shows nice curvatures of the pitch surfaces. A test case of a hearing-aid shell was used to test the capability of the MIP-SL system in fabricating micro-scale shell structures. Figure 7.30a shows the CAD model of the hearing aid shell and related support structures constructed by our support generation algorithm. Figure 7.30b shows the built physical object including the added support structures. As shown in Figure 7.29c, the curved shell with desired thickness (~80 m) was successfully built. A designed venting hole in the left side was also successfully fabricated. Figure 7.28: Fabrication capability tests. (a)-(c): pillars; (d)-(f): gaps. 1 mm 50 um 0.1 mm (a) (b) (c) (d) (f) (e) 1 mm 185 Figure 7.29: A gear with micro features: (a) CAD model (b) built object (c) microscopic image. Figure 7.30: A hearing-aid shells with micro-scale features. The tests as shown in Figure 7.28− 7.30 illustrate that our MIP-SL system can build parts with micro-features. Furthermore, in order to ensure the success rate and mechanical strength of built parts, (1) solid structures need to have at least 50 m thickness; (2) solid shells need to have at least 60 m thickness; and (3) holes or gaps need to have at least 110 m diameters or distances. Tests on objects with micro-features CAD models with more complex features were tested. The experimental results verify that, with proper supports constructed by the support generation system, all the micro-scale features that satisfy the aforementioned constraints can be successfully fabricated by our MIP-SL system. 0.2 mm (a) (b) (c) 1 mm 1 mm 0.5 mm (a) (b) (c) 80 um (c) 186 Figure 7.31: A threaded pipe test case Figure 7.32: A turbo fan test case Figure 7.31a shows the CAD model of a threaded pipe with embossed letters. Figure 7.31b shows the built object after the cleaning process. Figure 7.31c shows the built part after removing supports. Figure 7.31d and 7.31e are two microscopic images of the built object. Due to the big layer thickness (20 m) and resolution limits, the threads of the pipe are not as sharp as 1 mm 0.5 mm 0.5 mm (a) (b) (c) (d) (e) 1 mm 1 mm (a) (b) (c) (d) (e) (f) 0.5 mm 50 um 1 mm 187 those designed in the CAD model. However, the structure of the pipe is successfully fabricated. The embossed letters on the pipe are also successfully built. Figure 7.32a shows a turbo fan model. Figure 7.32b shows the built object after the cleaning process. Figure 7.32c shows the built supports after removing them from the built part. Figure 7.32d and 7.32e are the built fan in the top and bottom views respectively. The bottom surface has a lower surface finish due to the added support structures. Figure 7.32f shows a microscopic image of the thin blades of the fan (~50 m). Note that the contact areas of the part surface with support tips are rather small (<50 m).Most of the supports were disconnected with the built object in the cleaning process. Hence the supports can be easily removed. In addition, the marks of supports left on the surface are reasonably small. Tests on objects with overhanging features Supports are critical to facilitate the manufacturing of overhanging features in MIP-uSL process. To demonstrate the use of supports, a set of CAD models with different geometries was used. A CAD model of frame structure with generated supports is shown in Figure 7.33a. Built physical parts before and after removing supports are shown in Figure 7.33b and c, respectively. Figure 7.33d is the bottom surface which was connected with the supports. Figure 7.33e and 7.33f are the microscopic images of the part surface. As shown in 7.33(b), most supports were gone after cleaning. Also the masks of supports are negligible in 7.33(d). 188 Figure 7.33: A scaled-up frame test case Figure 7.34: A test case on a pipe with overhanging micro features Furthermore, a CAD model shown in Figure 7.34 (a) was used. A few of supports designs with different parameters were applied to build the part. Figure 7.34 (b) and (c) are the built objects with different supports. In Figure 7.34 (c), the lower pipe which is horizontally positioned (a) (b) (c) (d) (e) (f) 0.5 mm 0.5 mm 1 mm 1 mm 0.5 mm 0.5 mm 1 mm (a) (b) (d) (c) (e) (f) 1 mm 0.5 mm Area A 189 is deformed. The reason is that its supports were too weak to hold the pipe in the building process and some layers were failed. Figure 7.34 (d) and (f) is a comparison of the built structure of area A with and without supports, respectively. Figure 7.34 (e) is a microscopic image of the part showed in (f) after cleaning and before removing the supports. The support detached from the bottom surface after cleaning. The tests showed that when the sliced image is a disconnect graph, the geometry cannot be formed successfully without support. Also, the support parameters determined supports strength and hence its capability of holding cured layers, as well as the marks. Tests on stiction problem of thin shells When using the SL process in fabricating micro-beams or micro-shells, such thin beams or plates would suffer from the surface tension of liquid resin and cleaning solution in the processes of pulling it out from the resin tank or cleaning vat. Figure 7.35 a−c show an example, in which the three thin plates will be permanently deformed due to the stiction force introduced by the liquid surface tension. With the aid of support structures, it is found that the well-designed anchor supports can overcome such stitching forces in the building and cleaning processes. Figure 7.35 d−f show the built object with the thin plates as designed. 190 Figure 7.35: A test case on the surface stiction problem. 7.4.5 Building time statistics Table 7.2: Performance of MIP-uSL systems presented in literatures and our developed system Literature Light source mask Object size (x * y) Lateral resolution Layer thickness Speed Monneret et al. 1999,2001 Visible light LCD Diagonal: 3~10 mm 2 um 10 um 60 s/layer Bertsch et al. 2000 UV light DMD 3*2 mm 2 10 um 5 um 18 s/layer I.B.Park et al. 2011 Lamp (UV) DMD Not reported Not reported 50 um 180s/laye r A.S. Limaye et al. 2007 Lamp (UV) DMD 1.1 * 1.8 mm 2 6 um 100 um 90s/layer Young Myoung Ha et al. 2007 Lamp(UV) DMD 2 * 2 mm 2 2 um 5 um 100s/laye r: Jae Won Choi et al. 2006 Lamp(UV) DMD Not reported 5um 30 um 15s/layer Our system Visible light DMD 12.7 * 8 mm 2 40 um 20 um 2s ~6s/layer Table 7.2 is a summary of the building speeds of the proposed system and others' research reported in literatures. Table 7.3 shows the building time of our MIP-SL system in fabricating the related test parts. The developed process can build a layer in <5 seconds. Hence a CAD 1 mm (a) (b) (c) (d) (e) (f) 191 model with micro-scale features can be fabricated in minutes instead of hours. Such fabrication speed is much faster than all the previously reported SL work. Table 7.3: Performance of our new developed MIP-uSL system. Model Pipe1 Fan Frame Pipe2 Planes Tri # 249522 12678 32776 12608 5700 Sizex(mm) 12.6 * 5.867 11.511* 8.514* 8* Structure type shell curvature truss shell fence Thickness(um) 20 12 20 20 20 Tprojection(se c) * 0.55 0.4 0.55 0.55 0.55 Twait_projecti on(sec) 0.5 0.3 0.5 0.5 0.5 Tz (sec) 3.06 2 3.06 3.06 3.06 Twait_z (sec) 0.1 0.1 0.1 0.1 0.3* Tlayer (sec) * 4.21 2.8 4.21 4.21 4.41 Heightz (mm) 13.068 4.539 10.160 11.923 5.839 Layer # 653 378 508 596 292 Ttotal_buildin g (min) 46.27 17.97 36.10 42.27 21.93 * Note: although the X dimensional sizes of the parts are bigger than 6mm, the flow filling area is actually small. So T Z is still set as 0.1 second. * Tlayer is the building time of one layer except the layers for base. * Tprojection is the projection time of one layer except the layers for base. 7.5 Concluding Remarks Since the material deposition is very fast and straightforward in non-layer photo- polymerization AM processes like CNC accumulation, it is not studied in this dissertation. Research focus is placed on the material deposition of layer photo-polymerization AM processes. Fast material deposition methods in MIP-SL processes are studied. In this chapter, approaches for fast material deposition are demonstrated and discussed to achieve faster building speed on the base of the use of bigger layer thickness with gray scale image method and meniscus method. 192 To reach fast material recoating, a two-way movement design is adopted for meso-scale MIP-SL. With the developed approach, the separation forces are much smaller, so that the newly developed meso-scale MIP-SL process can successfully fabricate parts in a much shorter time. Compared with the average building time of traditional MIP-SL systems, the building time is reduced by 80% in our system. Considering the difference in liquid flow mechanisms when the scale comes down to microscopic, the fast material deposition study is extended to micro-scale MIP-SL. The bottom- up projection approach and fast recoating method proposed in meso-scale MIP-SL process are investigated in the micro-scale fabrication circumstances. Novel MIP-uSL process has been presented for fabricating 3D micro-scale structures with fast building speed and complex geometries. It is very interesting that the relationship between maximally allowed forces and separation forces in z and x direction changed with the building size. Hence, two-way movement design is not applicable in micro-scale MIP-SL systems and instead a directly pull-up design is applied. Furthermore, no waiting time is required for a cured area that is smaller than 6× 6 mm2 and a gap of 20m. The building time of each layer can be significantly reduced. The experiments show that the building time of a layer is about 2~4 seconds in our system, which is significantly faster than other systems reported in literatures. Moreover, the supports can also address the stiction problem of micro-scale beams and plates that is caused by surface tension. The experimental results illustrate that the newly developed MIP-uSL process can successfully fabricate complex 3D objects with micro-scale features with satisfactory quality in a reasonably short time. In conclusion, the methods presented in chapter 5 and 6 solve the stair-case effect problem and hence improve the surface finish greatly, while allows the use of bigger layer thickness to improve the build speed by a certain amount. With the fast recoating approach presented in this 193 chapter, the build speed could be further increased without impacting the high surface quality. So the conflict between the fast build speed and high surface finish is perfectly solved in MIP-SL systems. With the developed energy control and material deposition methods in Chapter 5, 6 and 7, it is able to fabricate meso/micro scale products with a high surface finish in a short time using MIP-SL processes. As shown in Figure 7.36, all hypothesizes have been verified and the research problem in both CNC accumulation and MIP-SL systems have been completely solved. So it comes to the closure of the whole dissertation, which is Chapter 8. Accurate Delivery: MIP-SL; CNC Accumulation Fast Delivery: MIP-SL; CNC Accumulation Fast Deposition: MIP-SL; CNC Accumulation Accurate Deposition: MIP-SL; CNC Accumulation Research Content: 5 Energy Control Material Deposition Fast Delivery Accurate Delivery Fast Deposition Accurate Deposition CNC Accumulation MIP-SL H2: 3D surface reconstruction, Optimal tool path (Chapter 4) H1: Multi-tool, Process optimization (Chapter 3) H3: Gray- scale image (Chapter 5) H5: Fast recoating method (Chapter 7) H4: Controlled meniscus Approach (Chapter 6) √ √ √ Figure 7.36: Research progress by this chapter 194 Chapter 8 Conclusion 8.1 Answering Research Questions and Contributions The research goal of this dissertation is developing approaches for photo-polymerization based AM to achieve high throughput and high part surface quality simultaneously. The whole dissertation is based on the research hypotheses and answering the research questions that are stated in Chapter 1. To make a clear and thorough summarization, it is necessary to revisit the research questions and check whether the research problems are properly addressed. As stated in Chapter 1, the research is motivated by the fact that the photopolymerization AM process play an important role in manufacturing field, while its build speed and part quality are conflicting sides and the research work to achieve these two goals without the cost of the other is not well conducted yet. To address this dilemma, we set a research goal for this dissertation that develop novel approaches for photopolymerization AM processes to fabricate parts with high part surface quality within a short building time. Thus the according primary research question is stated as follows: Primary Research Question: Q1. How to achieve high build speed together with high surface quality in photo- polymerization based additive manufacturing system? Accordingly, this question could be subdivided to the following to sub-questions. Q1.1 How to achieve high build speed without sacrificing the part quality? Q1.2 How to achieve high surface quality without compromising the building speed? 195 We examined both the energy control and material deposition aspects. Hence the two goals could be analyzed in two aspects: (1) To accurately control energy and material deposition in photo-polymerization AM process; (2) To rapidly deliver energy and deposit material in photo- polymerization AM process. To guarantee that the research results could be extended and applied in any photo-polymerization AM, two different types of photopolymerization AM processes, CNC accumulation and Mask Image Projection Stereolithography (MIP-SL) have been investigated as the layer based and non-layer based photopolymerization AM process examples. To better understand the difference between MIP-SL which is an example of the layer- based photopolymerization AM process, and the CNC accumulation which is an example of the non-layer-based photopolymerization AM process, we discussed the distinctions and characteristics of energy control methods in these two processes in Chapter 2. In Chapter 1 and Chapter 2, it is found that the as a non-layer based photopolymerization AM process, material deposition is very straightforward and simple in CNC accumulation system and there is no research values. Accordingly, the energy delivery is the key variable that we need to control to address the research goal in CNC accumulation system. So the fast and accurate energy control methods are studied and developed in CNC accumulation system. On the contrary, the energy delivery is already fast in MIP-SL system, while the material deposition plays a key role in the build speed and the surface quality of built parts in MIP-SL system. So accurate energy control method, fast and accurate material deposition methods are studied in MIP-SL system. Accordingly, the following hypotheses are investigated to respond the two sub research questions: H1. A multi-tool design and optimized process could make the energy delivery fast in CNC accumulation system, thus to improve the build speed and surface quality together for building on platform tasks in CNC accumulation system. 196 H2. A 3D surface reconstruction approach and automatic optimal tool path generation could make the energy delivery accurate in CNC accumulation system, thus to improve the build speed and surface quality together for building around inserts tasks in CNC accumulation system. H3. A gray scale image method could make the energy delivery accurate in MIP-SL, thus to fabricate smooth down-facing surfaces in MIP-SL system together with the same or even faster build speed. H4. A meniscus approach in MIP-SL system could deposit material with an accurate profile, thus to fabricate smooth up-facing surfaces in MIP-SL system together with the same or even faster build speed. H5. A fast material deposition method in MIP-SL process could further improve the build speed of MIP-SL systems without affecting the surface quality. The research is conducted by investigating and verifying the above five hypothesizes. Firstly, we studied the relationship of curing performance with accumulation tool movement velocity in CNC accumulation processes in Chapter 3. Multiple accumulation tools with different shapes have been developed and process parameters were optimized for each accumulation tool. With the optimal process parameters, the multi-tool CNC accumulation system is able to build any features in the fastest way by delivering energy rapidly. To deliver the energy accurately thus to reach a higher surface quality, in Chapter 4, a dual-axis mirror based 3D scanner is developed and integrated in the multi-tool CNC accumulation system. Experiments verified that the multi-tool design and optimized energy control methods could improve the throughput of CNC accumulation system. It is also illustrated that the optimized tool path planning in CNC accumulation system could improve the surface quality of the built part. 197 After investigating the energy control methods in CNC accumulation system in Chapter 3 and 4, a gray-scale method is proposed and developed to control the energy input in MIP-SL system in Chapter 5. Both top-down projection and bottom-up projection configurations were investigated. Experiments verified the effectiveness and efficiency of this method in improving the surface quality of curved features in both meso- and micro scales. In addition to the energy control methods, material deposition methods were also investigated in order to answer the two research questions. Challenges of z over-cure and stair stepping effects were addressed for down-facing curved surfaces with a well-controlled energy distribution presented in Chapter 5. However, for up-facing curved surfaces, the energy control method is not working due to the fact that the liquid surface is always get exposed to the light and be solidified first than the liquid beneath. Therefore, another solution, which is named as meniscus approach, is presented in Chapter 6 to improve the surface quality of up-facing curved surfaces by controlling the material depositions. Mathematical models and process planning methods were presented. A set of test cases were performed to verify the capability of this approach in improving surface finish of meso- and micro- parts. Additionally, material recoating is the bottleneck of build speed in layer-based photopolymerization AM processes like MIP-SL. In order to achieve even faster build speed without affecting the surface quality, a fast material recoating approach was developed for both meso-scale and micro-scale MIP-SL systems in Chapter 7. Experimental results illustrate that an order of magnitude faster building speed is able to be achieved with the same part quality using the developed fast recoating approach in MIP-SL systems. To conclude, due to the differences in energy delivery manner, CNC accumulation is non- layer based photopolymerization AM process because it doesn't require to recoat material for solidification layer by layer, while MIP-SL process is a typical layer based photopolymerization 198 AM process as it recoats and solidifies material in a layer by layer way. Both processes were studied in this research to answer the main research question Q1. A multi-tool design and tool path planning method were developed and integrated in CNC accumulation to assist the energy control. A gray-scale image method was developed and applied in MIP-SL to control the energy distribution. By controlling the energy well, experimental results illustrated that the build speed and part quality could be improved in these two photopolymerization AM processes. As a non- layer based photopolymerization AM system, the material deposition in CNC accumulation is very straightforward and simple. Hence there is no research value. In contrast, in the layer-based photopolymerization AM system like MIP-SL, material deposition is critical to the manufacturing performance. Hence, in addition to the energy control study, a fast material recoating method and a meniscus approach were developed to achieve fast and accurate material deposition, hence answering the main research question of this paper. Experimental results verified and effectiveness and efficiency of the developed approaches in this research. Based on the study of CNC accumulation and MIP-SL processes, the potential to extend and apply the developed approaches in any photo-polymerization AM system to reach fast build speed together with high surface quality is illustrated. A pictorial overview of the dissertation and the contributions is illustrated in Figure 8.1. It proceeds from the top to bottom, beginning with the introduction and process analysis in Chapter 1 and 2. Figure 8.1 provides a clear guide for the dissertation construction, the roles and contributions of each chapter. 199 Accuarate Energy Control (Chapter 4) Non-layer AM: CNC accumulation Fast Energy Control (Chapter 3) Layer AM: MIP-SL Energy Control Methods Fast Energy Control Material Deposition Methods Non-layer AM: CNC accumulation Not investigated Small tool big tool Cured resin x Y Z A C 5-Axis motion: angled tool ˆ 0.112312+0.032521 0.012187 A AB y x x  11 22 2 tan sin 2( ) 4( ) aD Rl a R l             2 sin( ) ii n R d    N     Multi-tool design: Process optimization Testing on surface quality and speed: Accurate Energy Control (Chapter 5) 0.521*ln( ) 2.252, ; () 0, c c g g g Fx gg       0.2mm 0.2mm (d) (e) 0.5mm 0.5mm (a) (b) (c) Area A Area B Area A Area B Area A Area A Modeling cure depth: Testing on surface quality with gray scale method: Layer AM: MIP-SL Fast Material Deposition (Chapter 7) Accurate Material Deposition (Chapter 6) Testing on surface quality and speed: 3D surface Reconstruction Algorithm: (a) Teapot (b) Scanned points (c) LDNI model (d) Mesh surface (c) 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 (c) 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 26 Tests & Results Side View Top View Application M1 M2 M2 Discontinuous error Side View Top View Application M1 M2 M2 Discontinuous error 0.5 mm M1 0.5 mm M2 (c) (c) 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 M2 M1 (c) 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 28 Tests & Results – Up-facing 0.5 mm 0.5 mm M1 M2 M2 M1 (c) Meniscus Method Principle Testing meniscus method and comparisons: Modeling recoating process: Model Pipe1 Fan Frame Pipe2 Planes Tri # 249522 12678 32776 12608 5700 Sizex(mm) 12.6 * 5.867 11.511* 8.514* 8* Structure type shell curvature truss shell fence Thickness(um) 20 12 20 20 20 Tprojection(se c) * 0.55 0.4 0.55 0.55 0.55 Twait_projecti on(sec) 0.5 0.3 0.5 0.5 0.5 Tz (sec) 3.06 2 3.06 3.06 3.06 Twait_z (sec) 0.1 0.1 0.1 0.1 0.3* Tlayer (sec) * 4.21 2.8 4.21 4.21 4.41 Heightz (mm) 13.068 4.539 10.160 11.923 5.839 Layer # 653 378 508 596 292 Ttotal_buildin g (min) 46.27 17.97 36.10 42.27 21.93 Testing and Statistics of build time with fast recoating approach Background, Research Problem, Hypothesis, Processes Analysis (Chapter 1, 2) Research Topic: Energy Control and Material Deposition Methods for Fast Fabrication with High Surface Quality in Additive Manufacturing using Photo-polymerization Figure 8.1: A pictorial overview of the dissertation and the contributions of each chapter 200 8.2 Limitations and the Future Work Although this dissertation provides many achievements and made contributions in answering the research question, there are still many interesting research areas that are not investigated in this dissertation. Photopolymerization AM has many different technologies. Based on the classification of layer based and non-layer based types, the same type technologies share many in commons but still have many differences in their ways to control energy and deposit material. Moreover, as discussed before, different scale fabrications have different phenomenon and related challenges. So there are still many research topics related to this dissertation could be explored in the future, including:  Test more tools with different shapes and sizes in CNC accumulation system;  Test more materials with different properties in CNC accumulation system, to validate the efficiency and effectiveness of the developed process optimization pipeline;  Test more materials with different properties in MIP-SL system, to validate the use of the developed approaches with various resins with distinct properties;  Extend the proposed methodologies in CNC accumulation process to other non-layer based photopolymerization AM systems, like two-photon technologies.  Extend the proposed methodologies in MIP-SL process to other layer based photopolymerization AM systems, like SLA systems.  Address the research problem in macro-scale photo-polymerization AM processes;  Integrate all the approaches developed for different scales fabrications, and apply to multi- scale photopolymerization AM processes. 201 Reference [1] I. Gibson, D.W. Rosen, B. Stucker, Additive Manufacturing Technologies: Rapid Prototyping to Direct Digital Manufacturing, 2010. [2] M. Vaezi, H. Seitz, S.F. Yang, A review on 3D micro-additive manufacturing technologies, Int J Adv Manuf Tech, 67 (2013) 1721-1754. [3] R. Soar, Additive Manufacturing Technologies for the Construction Industry, Rapid Manufacturing: An Industrial Revolution for the Digital Age, (2006) 249-273. [4] A.M. Christensen, Tactile Medical Modeling Using Additive Manufacturing Technologies, Faseb J, 24 (2010). [5] S. Junk, M. Trankle, Design for Additive Manufacturing Technologies: New Applications of 3d-Printing for Rapid Prototyping and Rapid Tooling, Int Conf Eng Des, 5 (2011). [6] L.E. Murr, S.M. Gaytan, D.A. Ramirez, E. Martinez, J. Hernandez, K.N. Amato, P.W. Shindo, F.R. Medina, R.B. Wicker, Metal Fabrication by Additive Manufacturing Using Laser and Electron Beam Melting Technologies, J Mater Sci Technol, 28 (2012) 1-14. [7] P.D. Dalton, C. Vaquette, B.L. Farrugia, T.R. Dargaville, T.D. Brown, D.W. Hutmacher, Electrospinning and additive manufacturing: converging technologies, Biomater Sci-Uk, 1 (2013) 171-185. [8] B. Paiva, R.J. Neto, J. Lino, Developing Ti jewelry through additive manufacturing and conversion technologies, Innovative Developments on Virtual and Physical Prototyping, (2012) 443-450. [9] S.H. Huang, P. Liu, A. Mokasdar, L. Hou, Additive manufacturing and its societal impact: a literature review, Int J Adv Manuf Tech, 67 (2013) 1191-1203. 202 [10] E. Yasa, J. Deckers, J.P. Kruth, The investigation of the influence of laser re-melting on density, surface quality and microstructure of selective laser melting parts, Rapid Prototyping J, 17 (2011) 312-327. [11] Y.M. Ha, J.W. Choi, S.H. Lee, Mass production of 3-D microstructures using projection microstereolithography, J Mech Sci Technol, 22 (2008) 514-521. [12] G. Strano, L. Hao, R.M. Everson, K.E. Evans, Surface roughness analysis, modelling and prediction in selective laser melting, J Mater Process Tech, 213 (2013) 589-597. [13] W. Morton, S. Green, A.E.W. Rennie, T.N. Abram, Surface finishing techniques for SLM manufactured stainless steel 316L components, Innovative Developments on Virtual and Physical Prototyping, (2012) 503-509. [14] K. Nyembwe, D.J. de Beer, J.G. van der Walt, S. Bhero, Assessment of Surface Finish and Dimensional Accuracy of Tools Manufactured by Metal Casting in Rapid Prototyping Sand Moulds, S Afr J Ind Eng, 23 (2012) 130-143. [15] G. Strano, L. Hao, R.M. Everson, K.E. Evans, Multi-objective optimization of selective laser sintering processes for surface quality and energy saving, P I Mech Eng B-J Eng, 225 (2011) 1673-1682. [16] M.A. Beard, O.R. Ghita, K.E. Evans, Using Raman spectroscopy to monitor surface finish and roughness of components manufactured by selective laser sintering, J Raman Spectrosc, 42 (2011) 744-748. [17] P.M. Pandey, N.V. Reddy, S.G. Dhande, Improvement of surface finish by staircase machining in fused deposition modeling, J Mater Process Tech, 132 (2003) 323-331. [18] M. Haider, L. Lessard, F. Trochu, M.A. Octeau, Experimental optimization of process parameters to obtain class A surface finish in resin transfer molding, De Sign, Manufacturing and Applications of Composites, (2003) 263-270. 203 [19] M. Alimardani, V. Fallah, M. Iravani-Tabrizipour, A. Khajepour, Surface finish in laser solid freeform fabrication of an AISI 303L stainless steel thin wall, J Mater Process Tech, 212 (2012) 113-119. [20] R.L. Gordon, A.E. Rosenbluth, Lithographic image simulation for the 21st century with 19th-century tools, Wave-Optical Systems Engineering Ii, 5182 (2003) 73-87. [21] D.T. Pham, C. Ji, A study of recoating in stereolithography, P I Mech Eng C-J Mec, 217 (2003) 105-117. [22] K. Renap, J.P. Kruth, Recoating issues in stereolithography, Rapid Prototyping J, 1 (1995) 4- 16. [23] Y. Chen, H.C. Guo, W. Gong, L.Y. Qin, H. Aleyasin, R.R. Ratan, S. Cho, J.X. Chen, S.S. Xie, Recent advances in two-photon imaging: technology developments and biomedical applications, Chin Opt Lett, 11 (2013). [24] S.F. Sun, Fabrication Technology of Involute Micro Gear Based on Two-photon of Femtosecond Laser, Frontiers of Manufacturing and Design Science, Pts 1-4, 44-47 (2011) 670- 674. [25] H.X. Lin, T.Y. Wang, R. Chen, Y.N. Wu, S.Y. Feng, Z.F. Huang, J.P. Lei, L. Ou, G.N. Chen, Y.Z. Li, A Processing Technology of Two-photon Microscopic Image of Nasopharyngeal Cancer Cells, Optics in Health Care and Biomedical Optics Iv, 7845 (2010). [26] A. Bullen, R.S. Friedman, M.F. Krummel, Two-Photon Imaging of the Immune System: A Custom Technology Platform for High-Speed, Multicolor Tissue Imaging of Immune Responses, Curr Top Microbiol, 334 (2009) 1-29. [27] M. Zhou, L.P. Liu, Q.X. Dai, C.P. Pan, Micro-fabrication technology by femtosecond laser direct scanning using two-photon photo-polymerization, Advanced Microlithography Technologies, 5645 (2005) 240-247. 204 [28] J.O. Koskinen, J. Vaarno, N.J. Meltola, J.T. Soini, P.E. Hanninen, J. Luotola, M.E. Waris, A.E. Soini, Fluorescent nanoparticles as labels for immunometric assay of creactive protein using two-photon excitation assay technology, Anal Biochem, 328 (2004) 210-218. [29] S. Kawata, Three-dimensional optics for optical data storage and optical MEMs using two- photon technology, 18th Congress of the International Commission for Optics: Optics for the Next Millennium, Technical Digest, 3749 (1999) 442-443. [30] X.Y. Zheng, J. Deotte, M.P. Alonso, G.R. Farquar, T.H. Weisgraber, S. Gemberling, H. Lee, N. Fang, C.M. Spadaccini, Design and optimization of a light-emitting diode projection micro- stereolithography three-dimensional manufacturing system, Rev Sci Instrum, 83 (2012). [31] J.W. Choi, R. Quintana, R.B. Wicker, Fabrication and characterization of embedded horizontal micro-channels using line-scan stereolithography, Rapid Prototyping J, 17 (2011) 351- 361. [32] R.D. Limaye A, Quantifying dimensional accuracy of a mask projection micro stereolithography system, Proc. Solid Freeform Fabrication Symposium, (2004). [33] A.S. Limaye, D.W. Rosen, Process planning method for curing accurate microparts using Mask Projection Micro Stereolithography, Virtual Modeling and Rapid Manufacturing, (2005) 441-447. [34] C. Sun, N. Fang, D.M. Wu, X. Zhang, Projection micro-stereolithography using digital micro-mirror dynamic mask, Sensor Actuat a-Phys, 121 (2005) 113-120. [35] A.S. Limaye, D.W. Rosen, Compensation zone approach to avoid print-through errors in mask projection stereolithography builds, Rapid Prototyping J, 12 (2006) 283-291. [36] A.S. Limaye, D.W. Rosen, Process planning method for mask projection micro- stereolithography, Rapid Prototyping J, 13 (2007) 76-84. 205 [37] S. Monneret, V. Loubere, S. Corbel, Microstereolithography using a dynamic mask generator and a non-coherent visible Light source, Design, Test, and Microfabrication of Mems and Moems, Pts 1 and 2, 3680 (1999) 553-561. [38] A. Bertsch, P. Renaud, C. Vogt, P. Bernhard, Rapid prototyping of small size objects, Rapid Prototyping J, 6 (2000) 259-266. [39] G.W. Hadipoespito, X.C. Li, Microstereolithography based on Digital Micromirror Device for complex meso/micro structures, Trans N Amer Manufac, 33 (2005) 105-111. [40] Y. Chen, C. Zhou, J.Y. Lao, A layerless additive manufacturing process based on CNC accumulation, Rapid Prototyping J, 17 (2011) 218-227. [41] D. Bourell, M. Leu, D. Rosen, NSF workshop - roadmap for Additive Manufacturing: identifying the future of freeform processing, (2009). [42] J. Choi, Y.S. Chang, Analysis of laser control effects for direct metal deposition process, J Mech Sci Technol, 20 (2006) 1680-1690. [43] B. Dutta, V. Singh, H. Natu, J. Choi, J. Mazumder, Direct Metal Deposition, Adv Mater Process, 167 (2009) 29-31. [44] J.Z. Ruan, L. Tang, F.W. Liou, R.G. Landers, Direct Three-Dimensional Layer Metal Deposition, J Manuf Sci E-T Asme, 132 (2010). [45] J.O. Milewski, G.K. Lewis, D.J. Thoma, G.I. Keel, R.B. Nemec, R.A. Reinert, Directed light fabrication of a solid metal hemisphere using 5-axis powder deposition, J Mater Process Tech, 75 (1998) 165-172. [46] M.A. McLean, G.J. Shannon, W.M. Steen, Laser generating metallic components, Xi International Symposium on Gas Flow and Chemical Lasers and High-Power Laser Conference, 3092 (1997) 753-756. 206 [47] B. Huang, S. Singamneni, O. Diegel, Construction of a Curved Layer Rapid Prototyping System: Integrating Mechanical, Electronic and Software Engineering, I C Mech Mach Vis Pr, (2008) 576-580. [48] R.M. Mahamood, E.T. Akinlabi, M. Shukla, S. Pityana, Revolutionary Additive Manufacturing: An Overview, Laser Eng, 27 (2014) 161-178. [49] H.S. Byun, K.H. Lee, Determination of the optimal part orientation in layered manufacturing using a genetic algorithm, Int J Prod Res, 43 (2005) 2709-2724. [50] B. Sager, D.W. Rosen, Use of parameter estimation for stereolithography surface finish improvement, Rapid Prototyping J, 14 (2008) 213-220. [51] J.W. Choi, R.B. Wicker, S.H. Cho, C.S. Ha, S.H. Lee, Cure depth control for complex 3D microstructure fabrication in dynamic mask projection microstereolithography, Rapid Prototyping J, 15 (2009) 59-70. [52] I. Mingareev, T. Bonhoff, A.F. El-Sherif, W. Meiners, I. Kelbassa, T. Biermann, M. Richardson, Femtosecond laser post-processing of metal parts produced by laser additive manufacturing, J Laser Appl, 25 (2013). [53] G. Lu, Q.S. Yan, X. Xiong, H. Wan, S.Y. Zhang, C.C. Liu, The study of influence laws of the part orientation on the dimensions of green part base on SLS, Advanced Research on Engineering Materials, Energy, Management and Control, Pts 1 and 2, 424-425 (2012) 1233- 1236. [54] K. Hildebrand, B. Bickel, M. Alexa, Orthogonal slicing for additive manufacturing, Comput Graph-Uk, 37 (2013) 669-675. [55] N. Volpato, A. Franzoni, D.C. Luvizon, J.M. Schramm, Identifying the directions of a set of 2D contours for additive manufacturing process planning, Int J Adv Manuf Tech, 68 (2013) 33- 43. 207 [56] J. Zhang, Y. Li, A unit sphere discretization and search approach to optimize building direction with minimized volumetric error for rapid prototyping, Int J Adv Manuf Tech, 67 (2013) 733-743. [57] H. Narahara, K. Saito, Study on the improvement of surface roughness of complex model created by three dimensional photofabrication – proposal of lift up irradiation method, Journal of the Japan Society for Precision Engineering, 61 (1995) 233-237. [58] P.C. Wayner, Interfacial Profile in the Contact Line Region of a Finite Contact-Angle System, J Colloid Interf Sci, 77 (1980) 495-500. [59] V. Jairazbhoy, Prediction of equilibrium shapes and pedestal heights of solder joints for leadless chip components, Ieee T Compon Pack A, 19 (1996) 224-233. [60] A.I. Hill, C. Pozrikidis, On the shape of a hydrostatic meniscus attached to a corrugated plate or wavy cylinder, J Colloid Interf Sci, 356 (2011) 763-774. [61] T. John, D. Rannacher, A. Engel, Influence of surface tension on the conical meniscus of a magnetic fluid in the field of a current-carrying wire, J Magn Magn Mater, 309 (2007) 31-35. [62] A.W. Adamson, and Zebib, A., Transition region between an infinite plane meniscus and an adsorbed film, The Journal of Physical Chemistry, 84 (1980). [63] J.P. Kruth, S. Kumar, Statistical analysis of experimental parameters in selective laser sintering, Adv Eng Mater, 7 (2005) 750-755. [64] P.B. Bacchewar, S.K. Singhal, P.M. Pandey, Statistical modelling and optimization of surface roughness in the selective laser sintering process, P I Mech Eng B-J Eng, 221 (2007) 35- 52. [65] A.K. Singh, R.S. Prakash, Response surface-based simulation modeling for selective laser sintering process, Rapid Prototyping J, 16 (2010) 441-449. 208 [66] T. Brajlih, B. Valentan, J. Balic, I. Drstvensek, Speed and accuracy evaluation of additive manufacturing machines, Rapid Prototyping J, 17 (2011) 64-75. [67] W. Li, L.J. Lee, Shrinkage control of low-profile unsaturated polyester resins cured at low temperature, Polymer, 39 (1998) 5677-5687. [68] A.S. Jariwala, F. Ding, A. Boddapati, V. Breedveld, M.A. Grover, C.L. Henderson, D.W. Rosen, Modeling effects of oxygen inhibition in mask-based stereolithography, Rapid Prototyping J, 17 (2011) 168-175. [69] A. Kataria, D.W. Rosen, Building around inserts: methods for fabricating complex devices in stereolithography, Rapid Prototyping J, 7 (2001) 253-261. [70] A.A. Afifi, D.R. Hayhurst, W.A. Khan, Non-productive tool path optimisation of multi-tool part programmes, Int J Adv Manuf Tech, 55 (2011) 1007-1023. [71] N. Baskar, P. Asokan, R. Saravanan, G. Prabhaharan, Selection of optimal machining parameters for multi-tool milling operations using a memetic algorithm, J Mater Process Tech, 174 (2006) 239-249. [72] D.G. Rao, C. Patvardhan, R. Singh, Part grouping and tool loading in versatile multi-tool machining centers, International Journal of Industrial Engineering Computations, 3 (2012). [73] W.J. Cao, D.J. Hu, Study of CNC Machining Auto-programming Technology Based on Design/Manufacturing Feature, Mater Sci Forum, 626-627 (2009) 657-662. [74] X.X. Gong, Y.N. Rui, G.F. Qian, X.R. Duan, CNC Machining Technology for the Sliding Block Parts with Special Shape in the Automobile Mold, Adv Mat Res, 102-104 (2010) 474-478. [75] Z.Z.C. Chen, D.J. Song, A generic and practical algorithm of isocusped tool path planning for three-axis sculptured part CNC machining, Transactions of the North American Manufacturing Research Institution of Sme, Vol 32, 2004, (2004) 143-150. 209 [76] Z.Z.C. Chen, G.W. Vickers, Z.M. Dong, A new principle of CNC tool path planning for three-axis sculptured part machining - A steepest-ascending tool path, J Manuf Sci E-T Asme, 126 (2004) 515-523. [77] S.G. Lee, S.H. Yang, CNC tool-path planning for high-speed high-resolution machining using a new tool-path calculation algorithm, Int J Adv Manuf Tech, 20 (2002) 326-333. [78] R. Sundaram, J. Choi, A slicing procedure for 5-axis laser aided DMD process, J Manuf Sci E-T Asme, 126 (2004) 632-636. [79] J. Ruan, Eiamsaard, K., Liou, F., Automatic Process Planning and Toolpath Generation of a Multiaxis Hybrid Manufacturing System, Journal of Manufacturing Processes, 7 (2005). [80] L. Ren, T. Sparks, J.Z. Ruan, F. Liou, Integrated Process Planning for a Multiaxis Hybrid Manufacturing System, J Manuf Sci E-T Asme, 132 (2010). [81] J. Zhang, F. Liou, Adaptive slicing for a multi-axis laser aided manufacturing process, J Mech Design, 126 (2004) 254-261. [82] R. Lin, Real-Time Surface Interpolators for Multi-Axis CNC Machine Tools, Ph.D. dissertation, University of Michigan, Ann Arbor, MI, (1994). [83] P. Waurzyniak, CNCs Speed Machining Processes, Manuf Eng, 148 (2012) 77-+. [84] Y.Y. Pan, C. Zhou, Y. Chen, A Fast Mask Projection Stereolithography Process for Fabricating Digital Models in Minutes, J Manuf Sci E-T Asme, 134 (2012). [85] F. Chen, G.M. Brown, M.M. Song, Overview of three-dimensional shape measurement using optical methods, Opt Eng, 39 (2000) 10-22. [86] F. A.K., Laser Scanners, Machine Vision Handbook, (2012). 210 [87] H. Golnabi, Design and operation of a laser scanning system, Opt Laser Technol, 32 (2000) 267-272. [88] A. Abuazza, D. Brabazon, M.A. El-Baradie, Analysis of surface defects using a novel developed fiber-optics laser scanning system, J Mater Process Tech, 143 (2003) 875-879. [89] Y.J. Li, Beam deflection and scanning by two-mirror and two-axis systems of different architectures: a unified approach, Appl Optics, 47 (2008) 5976-5985. [90] H. Golnabi, Image evaluation for the synchronised laser scanning systems, Opt Laser Technol, 31 (1999) 225-232. [91] C.Y. Zhou C., Additive manufacturing based on optimized mask video projection for improved accuracy and resolution, SME Journal of Manufacturing Processes, 14 (2012). [92] Texas Instruments Inc, in, http://www.ti.com/product/talp1000b. [93] Z. Z., Flexible camera calibration by viewing a plane from unknown orientations, in: International Conference on Computer Vision, Corfu, Greece, 1999, pp. 666-673. [94] G. Guennebaud, M. Gross, E. Zurich, Algebraic point set surfaces, Acm T Graphic, 26 (2007). [95] G. Guennebaud, M. Germann, M. Gross, Dynamic sampling and rendering of algebraic point set surfaces, Comput Graph Forum, 27 (2008) 653-662. [96] C.X. Xiao, G.P. Feng, Y. Chu, Z.Y. Du, X.H. Yang, Multi-level Partition of Unity Algebraic Point Set Surfaces, Mines 2009: First International Conference on Multimedia Information Networking and Security, Vol 1, Proceedings, (2009) 645-649. [97] C.X. Xiao, Multi-Level Partition of Unity Algebraic Point Set Surfaces, J Comput Sci Tech- Ch, 26 (2011) 229-238. 211 [98] C.C.L. Wang, Y.S. Leung, Y. Chen, Solid modeling of polyhedral objects by Layered Depth-Normal Images on the GPU, Comput Aided Design, 42 (2010) 535-544. [99] A.A. Alexa M., On normals and projection operators for surfaces defined by point sets, in: the Eurographics Symposium on Point-Based Graphics, 2004, pp. 149-156. [100] Y. Chen, C.C.L. Wang, Uniform offsetting of polygonal model based on Layered Depth- Normal Images, Comput Aided Design, 43 (2011) 31-46. [101] Y.M. Huang, C.P. Jiang, On-line force monitoring of platform ascending rapid prototyping system, J Mater Process Tech, 159 (2005) 257-264. [102] SLP-4000 Solid Laser Diode Plotter, in: D. Corporation (Ed.) Product Brochure 1997. [103] http://www.envisiontec.de, in: EnvisionTEC (Ed.). [104] D. Dendukuri, D.C. Pregibon, J. Collins, T.A. Hatton, P.S. Doyle, Continuous-flow lithography for high-throughput microparticle synthesis, Nat Mater, 5 (2006) 365-369. [105] J.G. Cham., Layer manufacturing with embedded components: process planning considerations, in: DETC99, Las Vegas, NV, 1999. [106] Fast MIP-SL process demo, in, Youtube. [107] a.D.D. S. Allen, Determination and evaluation of support structures in layered manufacturing, Journal of Design and Manufacturing, 5 (1995). [108] P. Kulkarni, A. Marsan, D. Dutta, A review of process planning techniques in layered manufacturing, Rapid Prototyping J, 6 (2000) 18-35. [109] a.K.L. J. Hur, Efficient algorithm for automatic support structure generation in layered manufacturing, in: DETC96, Irvine, CA, 1996. 212 [110] C. Zhou, Chen, Y., Waltz, R. A., Optimized mask image projection for solid freeform fabrication, ASME Journal of Manufacturing Science and Engineering, 131 (2009). [111] D.M. Elena, Fuentes, J., Cerro, R. L., and Savage, M. D., An analytical solution for a partially wetting puddle and the location of the static contact angle, J Colloid Interf Sci, 348 (2010). [112] A.W. Neumann, Benhabib, B., Szekely, J., Racz, L. M., and Rooks, S., Evaluation of two- dimensional and three-dimensional axisymmetric fluid interface shapes with boundary conditions, 7 (1991). [113] R.E. Williams, and Melton, V. L., Abrasive flow finishing of stereolithography prototypes, Rapid Prototyping J, 4 (1998). [114] L.R. White, On Deviations from Young's Equation, Journal of Chemical Society, Faraday Transactions, 73 (1977) 390-398. [115] W. Zisman, Relation of the equilibrium contact angle to liquid and solid constitution, (1964) 1-51. [116] C. Zhou, Chen, Y., Three-dimensional digital halftoning for layered manufacturing based on droplets, in: Transactions of the North American Manufacturing Research Institution of SME, 2009, pp. 175-182. [117] Stereolithography and Other RP&M Technologies, in: Society of Manufacturing Engineers, Dearborn, MI. [118] Y. Pan, X. Zhao, C. Zhou, Y. Chen, Smooth Surface Fabrication in Mask Projection based Stereolithography, Journal of Manufacturing Processes, 14 (2012) 460-470.

Abstract (if available)

Abstract More and more research prototypes and commercial systems have been developed to pursue faster build speed and better part quality in different manufacturing scale levels. Compared to traditional prototyping approaches that take days, additive manufacturing (AM) can build physical objects with any complicated structures in hours. Many industries have profited from AM and now AM is considered as an alternative to traditional manufacturing processes when the existing manufacturing methods cannot create a product practically, efficiently or affordably. However, to meet such high expectations, many challenges still remain. The primary challenge to overcome is the conflict between the build speed and the surface quality. Since the surface finish and approximation error depend on the layer thickness used in AM, the dominant approach to achieve a high surface quality is to reduce the layer thickness. However it would significantly slow down the building process. On the contrary, if a faster build speed is desired, a bigger layer thickness should be adopted and hence the surface quality would be worse. A tradeoff between the surface quality and building time is usually needed in AM processes. ❧ The thrust of this research is to contribute to the advancement of AM by addressing such a dilemma of the surface quality and build speed, to develop novel approaches to achieve both the high surface quality and the high build speed goals in photo‐polymerization based AM processes. In this research work, photo‐polymerization AM processes from meso‐scale to micro‐scale were investigated. Novel approaches and strategies were developed to deliver energy and polymerize material at the target position quickly and accurately, thus reaching high build speed and high surface quality simultaneously. ❧ CNC accumulation process and Mask Image Projection based Stereolithography (MIP-SL) process, are two typical photo‐polymerization AM processes: the non‐layer based and the layer based photo‐polymerization AM, respectively. Studies of these two processes were performed to show the feasibility of applying the developed approaches in any photo‐polymerization AM systems. ❧ In CNC accumulation process which is non‐layer based, an accumulation tool head is made of optic fibers and the light is delivered by the optic fibers to polymerize liquid resin. Multiple accumulation tools were designed for fabrications with different resolutions and speeds. The accumulation tools are driven by multiple stages and thus are able to move or rotate along multiple axes. To control the energy power of each accumulation tool in the multi‐tool and multi‐axis CNC accumulation system, statistical methods were applied to optimize the manufacturing process settings and force analysis were performed, thus to achieve the optimal build speed and part quality. In addition, in order to successfully fulfill a manufacturing task, another challenge related to energy control has to be overcome. In CNC accumulation, it is critical to accurately deliver the energy to the exact position with a good normal direction. Otherwise, the fabricated result may not be able to reach the accepted geometry accuracy, or the material may attach on the tool tip instead of the base surface, which would cause failures to the build task. To address this problem, a novel dual‐axis 3D scanning unit was developed and integrated. A point processing method based on the Algebraic Point Set Surface (APSS) fitting and Layered Depth‐normal Image (LDNI) representation was developed for converting the scanning points into a 3D surface model. Based on the constructed surface model, a multi‐axis tool path could be generated for building tasks of repairing or remanufacturing of any geometry using the developed system. With those developed energy control approaches, the developed multi‐tool and multi‐axis CNC accumulation system is able to fabricate the part accurately and rapidly. ❧ In MIP-SL system which is a layer based photo‐polymerization AM system, a DMD chip is usually used to dynamically generate a mask image. The mask image is used to define the light pattern hence control the energy distribution. As a layer based photo‐polymerization AM process, MIP-SL is much more complicated on material deposition than CNC accumulation process. Related challenges include ultra‐thin layer recoating, liquid surface profile, and so on. Accordingly, methods both on energy control and material deposition aspects were developed to achieve the two syngenetic goals of fast build speed and high surface quality. In order to address the notorious stair case effect in AM which damages the surface quality significantly, on energy control aspect, a gray scale image approach was developed by modeling and controlling the cure depth with different light intensities. On material deposition aspect, a meniscus equilibrium approach was developed by modeling and controlling the liquid profile with different liquid/gas/solid equilibrium conditions. With the two approaches, extreme high surface finish could be achieved without sacrificing build speed or even with higher build speed. In addition, a novel fast material recoating approach was proposed and implemented in MIP-SL system. With the developed material recoating approach, our system is able to realize an order of magnitude faster build speed comparing to the existing additive manufacturing systems in the market. ❧ Different building scales share some common physics in the fluid flow and polymerization process but also differ in many fields, such as separation force effects and optic system requirements. When it comes to micro‐scale fabrication, the optical systems and machine designs were modified, and the corresponding optimal process parameter settings were investigated to build micro features in both CNC accumulation system and MIP-SL system. Besides, when the scale comes down to micro, the fast recoating technique used in meso‐scale cannot be copied directly. Different recoating mechanisms should be applied due to the change in separation forces. The meniscus equilibrium approach developed in meso‐scale systems was also investigated and tuned to overcome the stair‐case effect in micro‐scale fabrication. ❧ Testbeds were developed and experiments were performed to verify the effectiveness and efficiency of the proposed approaches in the two photo‐polymerization AM systems. The test results demonstrated that the proposed approach could achieve an order of magnitude faster build speed than any other AM systems while the surface finish could be improved by ~80%. Unlike other technologies which can only accomplish one goal at the cost of the other, the approaches proposed in this research can improve the building speed and surface finish to the largest extent at the same time in both meso‐scale and micro‐scale fabrications. The build speed and surface finish are not conflicting goals any more as they were in the conventional AM systems. This work would be meaningful in advancing AM from ""rapid prototyping"" to truly ""rapid"" manufacturing technology for not only prototypes but also end‐use products.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Scalable polymerization additive manufacturing: principle and optimization

PDF

Multi-scale biomimetic structure fabrication based on immersed surface accumulation

PDF

Selective Separation Shaping (SSS): large scale cementitious fabrication potentials

PDF

Slurry based stereolithography: a solid freeform fabrication method of ceramics and composites

PDF

Enhancing the surface quality and dimensional accuracy of SIS-metal parts with application to high temperature alloys

PDF

Selective separation shaping: an additive manufacturing method for metals and ceramics

PDF

Deformation control for mask image projection based stereolithography process

PDF

Additive manufacturing of piezoelectric and composite for biomedical application

PDF

Motion-assisted vat photopolymerization: an approach to high-resolution additive manufacturing

PDF

Metallic part fabrication wiht selective inhibition sintering (SIS) based on microscopic mechanical inhibition

PDF

3D printing of polymeric parts using Selective Separation Shaping (SSS)

PDF

The extension of selective inhibition sintering (SIS) to high temperature alloys

PDF

Contour Crafting process planning and optimization

PDF

Fabrication-aware machine learning for accuracy control in additive manufacturing

PDF

Contour crafting construction with sulfur concrete

PDF

Hybrid vat photopolymerization processes for viscous photocurable and non-photocurable materials

PDF

Analyses of strength of layered structures fabricated by Contour Crafting

PDF

Hybrid vat photopolymerization: methods and systems

PDF

Statistical modeling and machine learning for shape accuracy control in additive manufacturing

PDF

Experimental and numerical techniques to characterize structural properties of fresh concrete

Asset Metadata

Creator Pan, Yayue (author)

Core Title Energy control and material deposition methods for fast fabrication with high surface quality in additive manufacturing using photo-polymerization

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Industrial and Systems Engineering

Publication Date 06/23/2014

Defense Date 05/13/2014

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag additive manufacturing,build speed,fast recoating,meniscus method,OAI-PMH Harvest,photo‐polymerization process,surface finish

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Chen, Yong (committee chair), Khoshnevis, Behrokh (committee member), Wu, Wei (committee member)

Creator Email yayuepan@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-422996

Unique identifier UC11285998

Identifier etd-PanYayue-2574.pdf (filename),usctheses-c3-422996 (legacy record id)

Legacy Identifier etd-PanYayue-2574.pdf

Dmrecord 422996

Document Type Dissertation

Format application/pdf (imt)

Rights Pan, Yayue

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA