STEREOLITHOGRAPHY CURE PROCESS MODELING

A Dissertation
Presented to
The Academic Faculty

By

Yanyan Tang








In Partial Fulfillment
Of the Requirements for the Degree
Doctor of Philosophy in the
School of Chemical & Biomolecular Engineering








Georgia Institute of Technology

August 2005
STEREOLITHOGRAPHY CURE PROCESS MODELING


















Approved by:
Dr. John D. Muzzy, Advisor Dr. Clifford L. Henderson, Co-advisor
School of Chemical and School of Chemical and
Biomolecular Engineering Biomolecular Engineering
Georgia Institute of Technology Georgia Institute of Technology

Dr. David W. Rosen Dr. Peter J. Ludovice
School of Mechanical Engineering School of Chemical and
Georgia Institute of Technology Biomolecular Engineering
Georgia Institute of Technology

Dr. Rigoberto Hernandez
School of Chemistry & Biochemistry
Georgia Institute of Technology
Date Approved: July 18, 2005


iii
ACKNOWLEDGEMENTS

I would like to take this opportunity to thank my advisor Dr. John Muzzy for the
support and freedom he provided me to pursue this project as I envisioned it.
I would also like to thank my co-advisor, Dr. Cliff Henderson, for his advice and
insight.
I appreciate being around Dr. David Rosen who pays attention to students’ academic
growth, from which I benefit a lot.
Thanks also go to Dr. Peter Ludovice and Dr. Rigoberto Hernandez for serving on my
thesis committee.
I’m thankful to Dr. Jonathan Colton for his welcome gesture and helpful advice when
I borrowed his equipment. I’d also like to mention the help from Dr. C. P. Wong and his
postdoc Zhuqing Zhang for use of photo-DSC, Dr. William Koros and his graduate
student William Madden for use of the density gradient column, Dr. Laren Tolbert and
Kendra McCoy for use of the UV/VIS spectrometer, and Dr. Tongfan Sun for
measurement of liquid thermal conductivity.
I appreciate the opportunity to work in both Henderson’s and Rosen’s research
groups. I owe thanks to Augustin Jeyakumar for help with ellipsometer and optical
microscope, Cody Berger for sharing his experience in standing wave issue, Lovejeet
Singh for discussion regarding to CTE measurement, Benita Comeau and Mikkel Thomas
(ECE) for use of surface profilometer, and Trevor Hoskins for being a wonderful
officemate. Benay Sager (ME) has discussed with me regarding to the SLA minivat
operation, Ec & Dp measurement, and other SLA problems.

iv
It’s worth mentioning the assistance from Safdar Ali and Marshall Sloane (ChBE
undergraduates) as well as Andrew Mrasek (ME undergraduate) with part building in
SLA 250 and dimension measurement by SEM.
It shouldn’t be forgotten, either, that Ms. Yolande Berta has offered nice help
regarding to SEM measurement and carbon coating unit operation, and that Jeff Andrews
and Brad Parker in ChBE machine shop have kindly and carefully machined the light
guide custom unit and user-designed DSC pans.
I’m lucky to have had you around, and I don’t take it for granted.


v
TABLE OF CONTENTS

ACKNOWLEDGEMENTS............................................................................................... iii

LIST OF TABLES........................................................................................................... viii

LIST OF FIGURES ............................................................................................................ x

LIST OF SYMBOLS....................................................................................................... xiv

SUMMARY................................................................................................................... xviii

CHAPTER 1 INTRODUCTION..................................................................................... 1
1.1 Introduction to Stereolithography....................................................................... 1
1.2 Project Objective................................................................................................. 2
1.3 Project Strategy................................................................................................... 9

CHAPTER 2 MODEL DEVELOPMENT..................................................................... 11

CHAPTER 3 KINETIC CHARACTERIZATION........................................................ 17
3.1 Photopolymerization Kinetic Model................................................................. 17
3.2 Kinetic Experiments.......................................................................................... 27
3.2.1 DPC Pan.................................................................................................... 28
3.2.2 Standing Waves ........................................................................................ 28
3.2.3 DPC Experiments...................................................................................... 32
3.3 Kinetic Data Analysis & Model Parameterization............................................ 35
3.4 Kinetic Model Validation ................................................................................. 44

CHAPTER 4 MATERIAL CHARACTERIZATION................................................... 46
4.1 Specific Heat Capacity...................................................................................... 46
4.2 Glass Transition Temperature........................................................................... 50
4.3 Coefficient of Thermal Expansion.................................................................... 52
4.4 Density.............................................................................................................. 58
4.5 Thermal Conductivity....................................................................................... 59
4.6 Heat of Polymerization ..................................................................................... 61
4.7 Absorption Coefficient...................................................................................... 62
4.8 Summary........................................................................................................... 64

CHAPTER 5 SIMULATIONS ...................................................................................... 65
5.1 Single Laser Drawn Line .................................................................................. 67
5.2 Overlapping Lines............................................................................................. 74
5.3 Stacked Single Lines......................................................................................... 78



vi

CHAPTER 6 MODEL VERIFICATION...................................................................... 82
6.1 DOC Threshold Model ..................................................................................... 82
6.2 DOC Threshold Model Prediction.................................................................... 85
6.2.1 Single Line Part Prediction....................................................................... 85
6.2.2 Overlapping Line Part Prediction ............................................................. 88
6.2.3 Stacked Line Part Prediction..................................................................... 90
6.3 Exposure Threshold Model Prediction ............................................................. 91
6.3.1 Ec and Dp Determination.......................................................................... 92
6.3.2 Single Line Part Prediction....................................................................... 93
6.3.3 Overlapping Line Part Prediction ............................................................. 97
6.3.4 Stacked Line Part Prediction..................................................................... 99
6.3.5 Comparison of DOC and Exposure Threshold Model............................ 100
6.3.6 Model Prediction using Ec and Dp Evaluated by a Different Protocol .. 103
6.4 Summary......................................................................................................... 108

CHAPTER 7 MODEL APPLICATIONS.................................................................... 109
7.1 Parameter Effect Investigation........................................................................ 110
7.1.1 Sensitive Parameters for Width Resolution............................................ 113
7.1.2 Sensitive Parameters for Speed (Width Direction)................................. 118
7.1.3 Sensitive Parameters for DOC................................................................ 121
7.1.4 Sensitive Parameters for Temperature Rise............................................ 124
7.1.5 Sensitive Parameters for Depth Resolution ............................................ 127
7.2 Resolution and Speed Prediction by Regression Model ................................. 131
7.2.1 Regression Prediction Model for Depth Resolution............................... 131
7.2.2 Regression Prediction Model for Width Resolution............................... 132
7.2.3 Regression Prediction Model for Speed (Width Direction).................... 135
7.2.4 Regression Prediction Model for Maximum DOC................................. 138
7.2.5 Regression Prediction Model for Maximum Temperature Rise............. 140
7.3 Parameter Optimization .................................................................................. 143
7.4 Parameter Analysis using Exposure Threshold Model ................................... 146
7.4.1 Parameter Significance Investigation ..................................................... 146
7.4.2 Parameter Optimization .......................................................................... 150

CHAPTER 8 CONCLUSIONS & RECOMMENDATIONS...................................... 152

APPENDIX A NOVECURE OUTPUT WITH 365NM FILTER (EXFO)................. 155

APPENDIX B NEGLIGIBLE HEATING EFFECT OF LIGHT IN DPC
EXPEIRMENTS.................................................................................. 156

APPENDIX C A BRIEF LITERATURE REVIEW ON GEL POINT ESTIMATION
............................................................................................................. 157

APPENDIX D MINITAB REGRESSION OUTPUT OF REDUCED MODEL FOR
WIDTH RESOLUTION...................................................................... 163

vii
APPENDIX E MINITAB STEPWISE REGRESSION OUTPUT FOR WIDTH
RESOLUTION.................................................................................... 164

APPENDIX F REGRESSION PREDICTION MODEL FOR DEPTH RESOLUTION
............................................................................................................. 165

APPENDIX G REGRESSION PREDICTION MODEL FOR WIDTH RESOLUTION
............................................................................................................. 166

APPENDIX H REGRESSION PREDICTION MODEL FOR CURING SPEED
(WIDTH) ............................................................................................. 167

APPENDIX I REGRESSION PREDICTION MODEL FOR MAXIMUM DOC...... 168

APPENDIX J REGRESSION PREDICTION MODEL FOR MAXIMUM
TEMPERATURE RISE ....................................................................... 170

REFERENCES ............................................................................................................... 172

viii
LIST OF TABLES

Table 1 Material & Process Parameters Involved in the SL Cure Process Model ........ 26

Table 2
P t
k k / and
2 / 1
/
t P
k k Values Obtained from DPC Experiments........................ 37

Table 3 Determination of Rate Constants
t
k and
P
k .................................................... 38

Table 4 Kinetic Parameter Values ................................................................................. 42

Table 5 Characterized Material Properties .................................................................... 64

Table 6 Process and Laser Parameter Values Used for Simulations ............................. 67

Table 7 Dimensions of Single Line Parts Built at Two Laser-Scanning Speeds:.......... 86

Table 8 Single Line Part Prediction by DOC Threshold Model .................................... 87

Table 9 Dimension Measurements of Overlapping Line Parts...................................... 89

Table 10 DOC Threshold Model Prediction for Overlapping Line Parts........................ 89

Table 11 Dimension Measurements of 3-Layer Stacked Line Parts................................ 90

Table 12 DOC Threshold Model Prediction for 3-Layer Stacked Line Parts.................. 91

Table 13 Single Line Part Prediction Results Based on Exposure Threshold Model...... 93

Table 14 Exposure Threshold Model Prediction Results using Modified Beam Profile. 96

Table 15 Comparison of Prediction Results by Two Threshold Models....................... 101

Table 16 Exposure Threshold Model Prediction (high working range): 1. Single Line (1)
V
s
= 1.071 (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and 3. Stacked-line Parts
........................................................................................................................ 102

Table 17 Exposure Threshold Model Prediction (protocol): 1. Single Line (1) V
s
= 1.071
in/sec (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and 3. Stacked-line Parts .. 106

Table 18 Exposure Threshold Model Prediction (protocol; high working range): 1.
Single Line (1) V
s
= 1.071 in/sec (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and
3. Stacked-line Parts ....................................................................................... 107


ix
Table 19 Exposure Threshold Model Prediction (protocol; low working range): 1. Single
Line (1) V
s
= 1.071 in/sec (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and 3.
Stacked-line Parts ........................................................................................... 108

Table 20 Potential Sensitive Parameters and Their Level Values ................................. 111

Table 21 Significant Factors Identified from Screening Experiment ............................ 112

Table 22 Full Factorial Design and Response Values (Width Resolution) ................... 113

Table 23 Estimated Factorial Effects and Lenth’s Test for Width Resolution.............. 114

Table 24 Estimated Factorial Effects and Lenth’s Test for Speed (Width)................... 118

Table 25 Estimated Factorial Effects and Lenth’s Test for Maximum DOC................ 121

Table 26 Estimated Effects and Lenth’s Test for Maximum Temperature Rise ........... 124

Table 27 Estimated Effects and Lenth’s Test for Depth Resolution ............................. 127

Table 28 Significant Factors for Investigated Responses.............................................. 130

Table 29 Simulation Conditions to Test Predictive Ability of Regression Models....... 132

Table 30 Depth Resolution Predicted by Regression Model ......................................... 132

Table 31 Width Resolution Predicted by Regression Model......................................... 135

Table 32 Curing Time (Width Direction) Predicted by Regression Model................... 138

Table 33 Maximum DOC Predicted by Regression Model ........................................... 140

Table 34 Maximum Temperature Rise Predicted by Regression Model ....................... 142

Table 35 Conditions used for Test of Temperature Rise Regression Model ................. 142

Table 36 Parameter Range Used for Response Optimization........................................ 144

Table 37 Evolver Optimization Results for Investigated Responses............................. 145

Table 38 Parameters in Exposure Threshold Model and Their Level Values ............... 147

Table 39 Evolver Optimization Results using Exposure Threshold Model .................. 150


x
LIST OF FIGURES

Figure 1 Complex SL Process and Oversimplified Exposure Threshold Model............. 5

Figure 2 Cured Shape of Single Laser Drawn Line....................................................... 12

Figure 3 2D Domain for Single Laser Drawn Line ....................................................... 12

Figure 4 Absorbed Intensity at Point Q(x,y,z)................................................................ 19

Figure 5 Structure Formula of E4PETeA (Sartomer) .................................................... 27

Figure 6 Structure Formula of DMPA (Ciba)................................................................ 27

Figure 7 DPC Sample Pan ............................................................................................. 28

Figure 8 Standing Wave Intensity at 365nm.................................................................. 30

Figure 9 Standing Wave Intensity at 304-395nm.......................................................... 30

Figure 10 Isothermal DSC Runs to Detect the Onset Temperature of Thermal Cure ..... 33

Figure 11 DPC Experimental Curves (Continuous and Flash Exposure at 50
o
C)........... 34

Figure 12 Nonlinear Fit of Propagation Rate Constant
P
k vs. Conversion X (50
o
C)..... 40

Figure 13 Nonlinear Fit of Termination Rate Constant
t
k vs. Conversion X (50
o
C) ..... 40

Figure 14 Semi-log Plot of True Kinetic Constants
0 P
k and
0 t
k vs. 1/T........................ 41

Figure 15 Linear Fit of
c
f / 1 (Critical Fractional Free Volume) vs. 1/T.......................... 41

Figure 16 Comparison of the Experimental and Simulated Polymerization Rate Curves
(incident power = 0.1 mW): (a) 30
o
C, (b) 50
o
C, (c) 70
o
C............................... 45

Figure 17 C
p
-T Plot of Liquid E4PETeA Monomer Exported from MDSC Data........... 48

Figure 18 C
p
-T Plot of Cured E4PETeA Polymer Exported from MDSC Data.............. 48

Figure 19 Glass Transition of Liquid E4PETeA Monomer Detected by DSC................ 50

Figure 20 Glass Transition of Cured Poly(E4PETeA) Detected by DSC ....................... 51


xi
Figure 21 Effect of Heating Rate on Measured T
g
Value of Liquid Monomer ............... 51

Figure 22 Effect of Heating Rate on Measured T
g
Value of Cured Polymer .................. 52

Figure 23 Temperature Dependence of Liquid E4PETeA Monomer Film Thickness .... 54

Figure 24 Temperature Dependence of Cured E4PETeA Polymer Film Thickness ....... 54

Figure 25 CTEs of Poly(E4PETeA) below T
g
Determined by Linear Regression of
Curves Obtained by Fitting with Si Substrate Optical Data of 25
o
C (diamonds)
and of Curves Obtained by Fitting with Temperature Dependent Si Substrate
Data (triangles) ................................................................................................ 56

Figure 26 Effect of Temperature on Heat Generated by Polymerization ........................ 61

Figure 27 Absorption Coefficient Spectrum of DMPA................................................... 63

Figure 28 Three Basic Laser Drawing Patterns: Case I. Single Laser Drawn Line, Case
II. Overlapping Single-Layer Lines, Case III. Stacked Single Lines .............. 66

Figure 29 Transients of (a) Intensity, (b) Initiator Concentration, (c) Radical
Concentration, (d) Monomer Conversion, and (e) Temperature at Point (x, 0,
0)...................................................................................................................... 70

Figure 30 Distribution of (a) Monomer Conversion and (b) Photoinitiator Concentration
upon a Single Laser Scan................................................................................. 72

Figure 31 Monomer Conversion vs. Width at the Top Surface of the Single Line Part
(Plot Interval = 0.01 sec, except for t = 1860 sec)........................................... 73

Figure 32 Monomer Conversion vs. Depth along the Centerline of the Single Line Part
(Plot Interval = 0.01 sec, except for t = 1860 sec)........................................... 73

Figure 33 (a) Monomer Conversion (b) Initiator Concentration (mol/m
3
) Distributions
upon Two Overlapping Scans.......................................................................... 76

Figure 34 Monomer Conversion vs. Width at the Top Surface of Two-Overlapping-Line
Part (Plot Interval = 0.01 sec, except for t = 1860 sec) ................................... 77

Figure 35 (a) Monomer Conversion (b) Initiator Concentration (mol/m
3
) Distributions
upon Two Stacked Scans................................................................................. 80

Figure 36 Monomer Conversion vs. Depth at the Centerline of Two-Layer-Line Part
(Plot Interval = 0.01 sec, except t = 1860 sec) ................................................ 81

Figure 37 SEM Image of Cross Section of a Single Line Part ........................................ 83

xii

Figure 38 Degree of Cure Contour for Parts Built at V
s
= 1.071in/sec (with the measured
part contour shown in red)............................................................................... 84

Figure 39 Degree of Cure Contour for Parts Built at V
s
= 10.71 in/sec........................... 88

Figure 40 Working Curve from WINDOWPANE
TM
Experimental Data ....................... 92

Figure 41 Beam Intensity Profile of HeCd Laser in SLA-250/50 ................................... 94

Figure 42 Laser Movement when Drawing Overlapping Lines ...................................... 97

Figure 43 High Working Range for Model Acrylate Resin in SLA.............................. 102

Figure 44 Comparison of Working Curves Obtained by the 3D Systems
WINDOWPANE Procedure (labeled “SOP” in the figure) and by the Part
Building Protocol........................................................................................... 104

Figure 45 Ec and Dp Determined in the High Range using Part Building Protocol ..... 106

Figure 46 Ec and Dp Determined in the Low Range using Part Building Protocol ...... 107

Figure 47 Normal Plot for Width Resolution ................................................................ 114

Figure 48 Factorial Effects Plot for Width Resolution: (a) main effect (b) interaction. 116

Figure 49 Factorial Effects Plot for Speed (Width): (a) main effect (b) interaction...... 120

Figure 50 Factorial Effects Plot for Max DOC: (a) main effect (b) interaction ............ 123

Figure 51 Factorial Effects Plot for Temperature Rise: (a) main effect (b) interaction 126

Figure 52 Factorial Effects Plot for Depth Resolution: (a) main effect (b) interaction. 129

Figure 53 Curvature in Factors for Width Resolution ................................................... 133

Figure 54 Curvature Effect for Width Resolution: (a) main effect (b) interaction........ 134

Figure 55 Curvature in Factors for Speed (Width) ........................................................ 135

Figure 56 Nonlinear Behavior of Beam Radius for Speed (Width)............................... 136

Figure 57 Factors Curvature for Speed (Width): (a) main effect (b) interaction........... 137

Figure 58 Curvature in Factors for Maximum DOC ..................................................... 138


xiii
Figure 59 Factors Curvature for Maximum DOC (a) main effect (b) interaction ......... 139

Figure 60 Curvature in Factors for Maximum Temperature Rise ................................. 140

Figure 61 Factors Curvature for Max Temp Rise (a) main effect (b) interaction.......... 141

Figure 62 Main Effects Plot for Cure Depth.................................................................. 148

Figure 63 Main Effects Plot for Line Width.................................................................. 148






xiv
LIST OF SYMBOLS

Ep
A Pre-exponential Factor of Propagation Rate Constant
Dependence on Temperature

Et
A Pre-exponential Factor of Termination Rate Constant
Dependence on Temperature

p
A Parameter of Propagation Rate Constant Dependence on
Fractional Free Volume

t
A Parameter of Termination Rate Constant Dependence on
Fractional Free Volume

C
d
Cure Depth of Resin
P
C Specific Heat Capacity
M P
C
,
Specific Heat Capacity of Monomer
P P
C
,
Specific Heat Capacity of Polymer
D
M
Diffusion Coefficient of Monomer
D
P
Penetration Depth of Laser into Resin
D

Diffusion Coefficient of Polymeric Radical
D
S
Diffusion Coefficient of Photoinitiator
E Exposure
E
c
Critical Exposure
p
E Activation Energy for Propagation
t
E Activation Energy for Termination
f Fractional Free Volume
cp
f Critical Fractional Free Volume for Propagation

xv
ct
f Critical Fractional Free Volume for Termination
M
f Fractional Free Volume of Pure Monomer
P
f Fractional Free Volume of Pure Polymer
h Heat Convection Coefficient
h
s
Hatch Space, i.e., the Lateral Distance between Adjacent
Laser Scan Centerlines

I

Incident Laser Intensity
I
o
Laser Peak Intensity
I
a
Absorbed Light Intensity
k Thermal Conductivity
D
k Diffusion Limited Kinetic Constant
p
k Rate Constant of Propagation
0 p
k True Rate Constant of Propagation
r
k Reaction Limited (“true”) Kinetic Constant
*
r
k Rate Constant of Reaction Diffusion
t
k Rate Constant of Termination
0 t
k True Rate Constant of Termination
L Average Sample Thickness Over Temperature Range
f
L Average Free Path Length
L
w
Linewidth of the Cured Line
[M] Monomer Concentration
[M]
0
Initial Monomer Concentration

xvi
[P·] Polymeric Radical Concentration
P
L
Laser Power
) (t Q Heat Integral in DPC Experiment
tot
Q Reference Heat of Reaction
R Gas Constant = 8.314J/mol-K
i
R Rate of Initiation
N
R Normalized Rate of Propagation
p
R Rate of Propagation
rd
R Reaction Diffusion Parameter
t
R Rate of Termination
S Photoinitiator Concentration
t
e
Characteristic Exposure Time
T
b
SLA Resin Bath Temperature
T
gM
Glass Transition Temperature of Monomer
T
gP
Glass Transition Temperature of Polymer
T
inf
Ambient Temperature in SLA Chamber
U
R
Rao Function
V Volume of Material
m
V Molar Volume per Structure Unit of Polymer
V
s
Laser Scanning Speed
w
o
Half Width of Laser Spot (@ 1/e
2
)
X Monomer Conversion

xvii
α Volumetric Coefficient of Thermal Expansion
M
α Coefficient of Thermal Expansion of Monomer
P
α

Coefficient of Thermal Expansion of Polymer
β Linear Coefficient of Thermal Expansion
ε Molar Absorptivity
i
φ Quantum Yield of Initiation
M
φ Volume Fraction of Monomer
λ Laser Wavelength (325nm in SLA-250/50)
ν Poisson’s Ratio
ρ Density
M
ρ

Density of Pure Monomer
P
ρ

Density of Pure Polymer
P
H ∆ Heat of Polymerization


xviii
SUMMARY

Although stereolithography (SL) is a remarkable improvement over conventional
prototyping production, it is being pushed aggressively for improvements in both speed
and resolution. However, it is not clear currently how these two features can be improved
simultaneously and what the limits are for such optimization.
In order to address this issue a quantitative SL cure process model is developed which
takes into account all the sub-processes involved in SL: exposure, photoinitiation,
photopolymerizaion, mass and heat transfer. To parameterize the model, the thermal and
physical properties of a model compound system, ethoxylated (4) pentaerythritol
tetraacrylate (E4PETeA) with 2,2-dimethoxy-2-phenylacetophenone (DMPA) as initiator,
are determined. The free radical photopolymerization kinetics is also characterized by
differential photocalorimetry (DPC) and a comprehensive kinetic model parameterized
for the model material. The SL process model is then solved using the finite element
method in the software package, FEMLAB, and validated by the capability of predicting
fabricated part dimensions.
The SL cure process model, also referred to as the degree of cure (DOC) threshold
model, simulates the cure behavior during the SL fabrication process, and provides
insight into the part building mechanisms. It predicts the cured part dimension within
25% error, while the prediction error of the exposure threshold model currently utilized in
SL industry is up to 50%. The DOC threshold model has been used to investigate the
effects of material and process parameters on the SL performance properties, such as
resolution, speed, maximum temperature rise in the resin bath, and maximum DOC of the

xix
green part. The effective factors are identified and parameter optimization is performed,
which also provides guidelines for SL material development as well as process and laser
improvement.

1
CHAPTER 1
INTRODUCTION

In this chapter, the stereolithography (SL) technology is introduced, the objective of
this work is addressed, and the strategy to achieve the goal is demonstrated.

1.1 Introduction to Stereolithography
Stereolithography is currently the most widely used process in the rapid prototyping
and manufacturing (RP&M) field. “It translates computer aided designs (CAD) into solid
objects through a combination of laser, photochemistry and software technologies”
1
.
A basic printing process goes like this
2
:
• “A 3-D model of an object is created in a CAD program.
• The software (e.g. Lightyear, 3D Systems) slices the 3-D CAD model into a series
of very thin horizontal layers.
• The sliced information is transferred to an ultraviolet laser that scans the top layer
of the photosensitive resin, hardening it.
• The newly built layer attached to the platform is lowered to just below the surface
the distance of one layer, and a new layer of resin is then recoated and scanned on
top of the previous one. This process repeats layer by layer, with successive layers
bonding to each other, until the part is complete.”
2



1
‘Stereolithography’, Conceptual Reality L.L.C., 2001, http://conceptual-reality.com/stereo.htm.
2
‘How Stereolithography (3-D Layering) Works’, Howstuffworks Inc., 1998-2001,
www.howstuffworks.com/stereolith.htm.

2
“Traditional prototype production is a long, inefficient, expensive and fraught-with-
inaccuracy process that adds to the ultimate cost of a product, wastes manpower and
materials, and slows the production cycle”
3
. SL technology provides a solution to these
problems inherent in the traditional approach. “It is a technological breakthrough that
allows solid physical parts to be made directly from computer data in a short time using
an automated process.”
3

1.2 Project Objective
Although SL is a remarkable improvement over the conventional prototyping
production in many aspects, it still needs further improvement in speed and resolution to
meet the demands of industry. Resolution is particularly important as it indicates the
minimum feature sizes and surface finish achievable.
One important factor that affects SL resolution is inherent in the nature of the laser.
For example, for the case of a Gaussian laser and a resin obeying the Beer-Lambert law,
the resin will cure in a shape of a parabolic cylinder upon a single laser scan vector
(Jacobs, 1992). Using a smaller layer thickness can reduce this boundary effect, but it
also increases the build time. Resolution can be improved by shrinking the laser beam
size, but it also causes an increase in the building time. Increasing the laser intensity can
improve SL speed since both the rate and degree of cure increase with the intensity
(Maffezzoli et al., 1998). However, since the cure reaction is exothermic and SL resins
have low thermal conductivities, the heat of reaction associated with the local photo-
polymerization cannot be easily dispersed. When the laser intensity is increased in order

3
‘Benefits of Stereolithography – Higher Quality, Lower Costs’, Pure Fluid Magic Inc., 1999,
www.purefluidmagic.com/sl_bene.htm.

3
to increase the part building speed, it also unfortunately leads to faster heat generation.
Consequently, some thermally initiated polymerization might occur in the vicinity of the
exposed region, which would reduce the resolution of the prototype being constructed.
Furthermore, the temperature gradients built within the resin might cause considerable
thermal stresses and correspondingly thermal strains, which could deteriorate the
mechanical/chemical properties of the part, or even manifest themselves as part
distortion.
Can these two features in the SL process, resolution and speed, be improved
simultaneously or do they have to be compromised with each other? If there is an
optimized solution, what are the limits for such optimization given a photosensitive
material system? What are the most sensitive parameters that affect the resolution or
speed? In order to answer these questions, being able to simulate and predict part shape,
build time, and potential difficulties would be very beneficial.
Current models of the SL process assume that the extent of resin cure is a function of
only the amount of exposure to UV radiation (Jacobs, 1992). They utilize an exposure
threshold model that assumes a dose E(x,y,z) that is greater than a minimum “critical
exposure,” E
c
, causes the resin to solidify at point (x,y,z). Basically it derives an exposure
spatial distribution in the resin, e.g. Equation (1), for a single laser drawn line (Jacobs,
1992), and substitutes E
c
for E(y,z), then y* and z* obtained (Equation 2) describe the
cured shape of the part.

) / exp( ) / 2 exp(
2
) , (
2
0
2
0
P
s
L
D z w y
V w
P
z y E − − =
π
(1)


4
) / * exp( ) / * 2 exp(
2
2
0
2
0
P
s
L
c
D z w y
V w
P
E − − =
π
(2)

where
L
P ,
0
w , and
s
V are the laser power, beam radius, and scanning speed, respectively;
P
D is the penetration depth of the laser into the resin, the depth where the laser intensity
decreases to 1/e (about 36.8%) of the intensity incident at the resin surface.
P
D can be
expressed as ) 3 . 2 /( 1 S ε D
P
= (Jacobs, 1992), where ε and S are the absorption coefficient
and concentration of the photoinitiator, respectively. A Gaussian laser and a resin
obeying Beer’s Law are assumed here.
This exposure threshold model is an oversimplification of the SL process. As
demonstrated in Figure 1, it directly connects the exposure to the resin and the final solid
part shape. It ignores an important intermediate step: reaction. Therefore, how the
reaction, the resin kinetic characteristics, as well as the diffusion and thermal effects
influence the size, shape and properties of parts fabricated by SL cannot be investigated
by using this model. Its ability to predict the cured part outline is challenged especially
when part resolution is in demand.


5



Figure 1 Complex SL Process and Oversimplified Exposure Threshold Model

Another deficiency of the exposure threshold model currently used in industry is that
it also assumes the exposure is additive, i.e. when the laser draws multiple lines or layers
to form a part, it simply adds all the exposure deposited in the part building process to
determine the part dimensions (see section 6.3). The time delay between lines or layers is
ignored as well as the chemical effects (e.g. chemical reaction, material change, etc.)
during the delay. Again, the exposure threshold model is just an oversimplification of the
part building process. It ignores everything (reaction, diffusion, heat, etc) but the
exposure.
Saito (1993) conducted experiments varying laser power and scanning speed in SLA,
and claimed a relationship which is close to the power function between the cured depth
and laser scanning speed on a semi-log plot. Nagamori and coworkers (2001, 2003)
performed SL curing tests to investigate how the laser power, laser beam diameter, and
laser scanning speed affect the cured depth and width. They correlated the cured depth
with energy density (exposure) and found a linear relation on the semi-log graph. All

6
these studies were trying to directly connect the laser exposure to the part dimensions, as
in the exposure threshold model introduced above.
A lot of work has been done to investigate the effect of process parameters and
optimize the SL process, but they are all based on the exposure threshold model currently
used in industry. For example, Chockalingam and coworkers (2003) determined the part
shrinkage (by comparing the SL finished part dimensions with the part dimensions on the
CAD model) for an experimental set designed by genetic algorithm concerning the
effects of layer thickness, hatch spacing, hatch style, hatch over cure, and hatch fill cure
depth. They then performed an optimization and identified an optimal value set of these
parameters to obtain parts with the same shrinkage ratio in both depth and width
directions. Cho and coworkers (2000) also used a genetic algorithm based methodology
to determine an optimal value set for the process parameters, such as hatch spacing, hatch
overcure, border overcure, hatch fill cure depth, and layer thickness, to minimize SL part
building error. Schaub and coworkers (1997) identified four key variables that affect the
part dimensional accuracy among various control variables in the SL process. They then
used design of experiments and the ANOVA technique to analyze and compare the
significance of these four parameters, and concluded that layer thickness and part
orientation have more effects on the part dimensional accuracy. Onuh and Hon (1998a)
used the Taguchi method to design and conduct experiments concerning layer thickness,
hatch spacing, hatch style, hatch overcure, and hatch fill cure depth. They analyzed the
built results and optimized these building parameters to improve the surface finish of SL
parts. Onuh and Hon (1998b) added two new hatch styles to their previous work (1998a)
and studied the effects of these styles on the dimensional accuracy. Jayanthi and

7
coworkers (1994) performed a study on the influence of process parameters, such as layer
thickness, hatch spacing, hatch overcure, and fill cure depth, on curl distortion of the
cured part. This study was performed for two writing styles: hatch and weave. The
ANOVA procedure was utilized to identify significant factors for each writing style, and
it was concluded that the hatch writing style yields better results than weave style. All
these studies took the exposure threshold model for granted, used it to control the SL part
building, and analyzed the finished part property upon the variation of the process
parameters.
Eschl and coworkers (1999) tested and simulated the transient post-fabrication
shrinkage of SL parts to investigate the effect of the resin material type, acrylate or
epoxy, on the SL cure process. They found that the epoxy resin produces more accurate
parts because the stress due to shrinkage is smaller and the final stiffness is higher. Their
methodology of studying material effects is based on an investigation of the built results
rather than a direct study on the building process. This is a different perspective, which,
however, cannot address the curing dynamics or the heating issue in SL building process.
A more complete model is needed that accounts for reaction, heat transfer and mass
transfer in order to predict the cured shape and size more accurately, to investigate how
the chemical effects (e.g. resin properties, cure reaction, etc.) impact the SL fabrication
results, and to find the optimum combination of material and process parameters to
improve SL resolution and speed.
Flach and Chartoff (1995a,b) incorporated both reaction and heat transfer into an SL
process model and simulated the cure process when the laser is stationary and when it
moves along one line. Mass transfer, however, was not taken into account. Their

8
simulation results predicted that a substantial temperature increase (~90
o
C) occurs in the
resin bath under certain conditions. They also presented the profiles for monomer
conversion and photoinitiator consumption in the curing process. However, no
experimental verification of the model was provided. Furthermore, a systematic study of
how the various SL parameters affect the SL process was not performed. Therefore, their
work did not directly provide guidance on how to improve the SL process. Furthermore,
the diacrylate monomer (hexanedioldiacrylate, HDDA) used in their work does not form
well-made solid parts in SLA. Hur and coworkers (1997, 2000) further studied the part
deformation and the thermal stress formed in the built part when the laser is stationary
and moves along one line. However, in addition to suffering from the deficiencies in
Flach and Chartoff’s work (1995), their work also ignored the dark polymerization
reaction in the case of the laser moving.
In this study, a tetraacrylate monomer is used for both simulation and part building in
SLA. Its material properties and photopolymerization kinetics are characterized. The
process model established incorporates both an energy balance and mass balances for
multiple species. Since the chemical reactions are taken into account upon transient
irradiation, the new model discards the additive exposure assumption used by the current
exposure threshold model. The SL cure process is simulated and the process modeling is
verified experimentally. For several responses that characterize the SL performance, such
as temperature rise in the SLA vat, part resolution, and green part degree of cure,
significant factors which affect each of these responses are identified and optimized.


9
1.3 Project Strategy
In this work, a complex SL cure process model is established that captures effects that
are ignored in the exposure threshold model. It incorporates laser exposure,
photoinitiation, polymer chain propagation and termination, species diffusion in the
curing polymer network, and heat transfer via conduction in the exposed region and its
vicinity. This model investigates during the part building process the spatial and temporal
distributions of temperature, rate of polymerization, and degree of cure (DOC), which are
necessary to characterize the cured part. It gives a full description of the transient cure
behavior of the resin in the SLA bath, as well as a prediction of the cure behavior upon
the variation of material or process parameters. Therefore, a fundamental understanding
of the SL process that takes into account the detailed physics and chemistry of the
underlying process can be expected; the material and process modifications can be made
for SL technology improvement; and the SL applications which are currently limited by
poor prediction of the exposure threshold model can be activated. Additionally, the
sensitivity analysis of material parameters provides a guideline for developing new
photosensitive SL resins.
In Chapter 2, the SL cure process model is formulated as a set of coupled partial
differential equations describing mass and energy transport during the curing process,
incorporating exposure and dark reaction in one model. In Chapter 3, the
photopolymerization kinetics are characterized using differential photocalorimetry (DPC)
and a comprehensive kinetic model is parameterized for a model acrylate resin system.
The thermal and physical properties of the model material are characterized in Chapter 4.
Chapter 5 demonstrates the simulation results by solving the process model using the

10
finite element method with the software package FEMLAB (Comsol Inc.). Chapter 6
verifies the process model through part fabrication and measurement. In Chapter 7,
significant material and process parameters are identified and optimized for SL resolution
and curing speed. Conclusions and recommendations are made in Chapter 8.


11
CHAPTER 2
MODEL DEVELOPMENT

The simplest case of complex laser drawing patterns in SL is that the laser moves
along one direction and draws a single vector line. For a Gaussian laser and a resin
obeying Beer’s law used in this work, the cured shape upon a single laser drawn line is a
parabolic cylinder (Jacobs, 1992), as shown in Figure 2, where the x axis is the laser
moving direction. Considering the repetitive cure behavior along the x-axis (the very
ends of the line which may receive different amount of exposure are not of interest here),
only the cross section of the parabolic cylinder needs to be modeled. The heat and mass
transfer along x direction can be ignored due to infinitely small behavior difference
between neighboring planes (cutting the parabolic cylinder into infinite number of
parabolic planes) as well as low thermal conductivity and diffusion coefficients of the
curing system. A 3-dimensional problem is thus reduced to a 2-dimensional one.
Furthermore, since the cross section is symmetric about z axis, only a half section needs
to be modeled. This leads to a 2-dimensional rectangular domain in Cartesian coordinate
(Figure 3) which is used to simulate the resin cure behavior during the single line
drawing process.


12


Figure 2 Cured Shape of Single Laser Drawn Line




Figure 3 2D Domain for Single Laser Drawn Line

The shaded region in Figure 3, which corresponds to the half cross section of the
parabolic cylinder in Figure 2, is where most of the reaction occurs and the material
properties vary significantly. The size of this region increases with time as heat
conduction and/or molecular diffusion continues (Flach and Chartoff, 1995a). The
domain is chosen to be large enough to ensure ambient temperature and concentrations
outside the rectangle at any time.

13
Mass transfer by diffusion and heat transfer by conduction are the two transport
phenomena occurring in the SL cure process. Equation (3) is the energy balance of the
curing system. Equations (4)-(6) describe mass balances for monomer, polymeric radicals
(including monomer radicals), and photoinitiator, respectively.
i
R ,
P
R , and
t
R are the
rate of initiation, propagation, and termination, respectively;
i
φ is the quantum yield of
initiation.


P P P
R H
z
T
y
T
x
T
k
t
T
C ∆ +
)
`
¹
¹
´
¦


+


+


=


2
2
2
2
2
2
ρ
(3)

) (
] [ ] [ ] [ ] [
2
2
2
2
2
2
P M
R
z
M
y
M
x
M
D
t
M
− +
)
`
¹
¹
´
¦


+


+


=


(4)

2 2 2
2 2 2
[ ] [ ] [ ] [ ]
( )
P i t
P P P P
D R R
t x y z

¦ ¹ ∂ • ∂ • ∂ • ∂ •
= + + + −
´ `
∂ ∂ ∂ ∂
¹ )
(5)

2 2 2
2 2 2
( / )
S i i
S S S S
D R
t x y z
φ
¦ ¹ ∂ ∂ ∂ ∂
= + + + −
´ `
∂ ∂ ∂ ∂
¹ )
(6)

These equations are coupled with one another through the reaction terms as source(s)
or sink(s) and have to be solved simultaneously. The photopolymerization mechanism
and kinetics will be addressed in Chapter 3.
As shown in Equation (3) the heat generated by steps other than propagation is
assumed to be negligible. The heating effect of the laser (325nm wavelength) is

14
negligible (~10
1
J/mol or less) due to the low absorption of the curing resin (except the
photoinitiator) and very short exposure time. It can be safely ignored when compared
with the large amount of heat generated by reaction (~10
5
J/mol).
To take shrinkage effect into account, the convection term should also be
incorporated into Equations (3)-(6). Only diffusion and heat conduction phenomena are
considered here due to the minor difference (within 6%) between the density of liquid
monomer and cured polymer.
Attention should be paid when the assumption is made that the propagation and
termination only occurs in the dark. Although the exposure time for the resin is very short
in SL (~20ms in this study), this assumption is not valid for the photosensitive material
system studied in this work. The later simulation results demonstrate that significant
reactions and material property variations occur during this 20ms. Therefore, in Equation
(5) the source/sink term can not be limited to
t
R which only describes the radical
reaction in the dark. The radical initiation rate
i
R has to be incorporated in order to take
the exposure reaction into account. Since
i
R is proportional to the irradiance I, it is
beneficial to develop a time-varying description of I which integrates the two periods that
any point would go through (irradiation and dark) into one equation ÷ Equation (7). A
time parameter
0
t is introduced into the equation so that when time t goes from 0 to +∞, I
increases to a maximum (the investigated point is directly irradiated) and then decreases
till zero (the beam moves away from the investigated point). As the laser moves from -∞
to +∞ along x direction, any point (x,y,0) only receives limited time of exposure, which

15
in SL is defined as characteristic exposure time and expressed as
s o e
V w t / 3 . 4 = (Jacobs,
1992). Any value greater than half
e
t can be used as
0
t .

( ) | | { }
8
2 2 2
0 0
10 196 . 1
) nm (
) / exp( / ) ( 2 exp
×
− + − − =
λ
p o s
D z w y t t V I I (7)

where y and z axes are as shown in Figure 2, I
0
(W/m
2
) is the maximum intensity incident
at the resin surface, ) nm ( λ is the laser wavelength, and the last quotient term is adopted
to convert the unit of intensity from W/m
2
to mol/m
2
-s.
Neglecting the insignificant property variations along the laser scanning direction (x
axis) (Figure 2), the terms containing x variations can be removed from Equations (3)-
(6). The initial and boundary conditions corresponding to this 2D problem are established
as follows:

0
0
[ ] [ ] at 0, 0 5 , 2 0 (a)
0 at 0 2 0, 0 ( )
[ ] [ ] at 5 2 0, 0 (c)
0 at 0 0 5 , 0 (d)
[ ] [ ] at 2 0 5 , 0 (e)
i o d
d
o d
o
d o
Q Q t y w C z
Q
y , C z t b
y
Q Q y w , C z t
Q
z , y w t
z
Q Q z C , y w t
= = ≤ ≤ − ≤ ≤

= = − ≤ ≤ ≥

= = − ≤ ≤ ≥

= = ≤ ≤ ≥

= = − ≤ ≤ ≥
(8)

where Q represents temperature T, monomer concentration [M], polymeric radical
concentration [P•], or photoinitiator concentration S; their initial values are equivalent to
their boundary values
0
] [ ] [ Q Q
i
= . C
d
is cure depth, the maximum depth of the solidified
area (Jacobs, 1992); w
0
is the laser beam radius. The domain size is initially set based on

16
the values of C
d
and w
0
and adjusted accordingly to accommodate the transient
variations of the simulated properties.
For the temperature condition at z=0 boundary, heat transfer with the natural air
environment in the SLA chamber can be incorporated by replacing the temperature
condition (8d) with the following:

( )
inf
T
k h T T
z

= −

at z = 0, 0 ≤ y ≤ 5w
o
, t ≥ 0 (9)

where k is thermal conductivity of the curing resin system, h and T
inf
are the air-resin
heat transfer coefficient and ambient temperature in the SLA chamber, respectively. The
later simulation results show that the heat convection at the resin surface in the SLA
chamber doesn’t have a noticeable effect on the part building results.
In order to solve the governing equations (3)-(6), the reaction dependent source/sink
terms need to be defined. The kinetic model and its parameterization are detailed in
Chapter 3.



17
CHAPTER 3
KINETIC CHARACTERIZATION

In this chapter, a kinetic model of photoinitiated free radical polymerization is
described, the kinetic experiments are designed, the kinetic coefficient data are extracted
without using steady state assumption, a curve fitting method is utilized to analyze the
kinetic data, and the kinetic model is parameterized and validated for a photosensitive
acrylate system.

3.1 Photopolymerization Kinetic Model
As discussed in Chapter 2, all source/sink terms in the balance equations are related
to the resin cure kinetics. Ignoring chain transfer reactions, the photocure mechanism for
acrylate resin can be briefly described as follows:


¦
)
¦
`
¹
• ÷→ ÷ • +
• ÷→ ÷
1
P R M
R PI
i
k

(10)


• ÷→ ÷ + •
+1 n
k
n
P M P
P
Propagation
m n
k
m n
M P P
tc
+
÷→ ÷ • + • Termination by Combination
m n
k
m n
M M P P
td
+ ÷→ ÷ • + • Termination by Disproportionation
Q In R
in
k
÷→ ÷ + • Inhibition

Initiation

18
where PI, M, and In represent the photoinitiator, monomer, and inhibitor, respectively; R•
is the primary radical, P
n
• the polymeric radical with a chain length of n monomer units,
and M
n
the stable polymer molecule with a chain length of n monomer units.
Correspondingly, the rates of initiation, propagation and termination are expressed as
Equations (11), (12), and (13), respectively.

a i i
I R φ = (11)
4


] ][ [ M P k R
p p
• = (12)

2
] • [ = P k R
t t
(13)

where
p
k and
t
k are the rate constants for propagation and termination; ] [ • P and ] [M
are the polymeric radical concentration and monomer concentration; and
a
I is the
absorbed light intensity or rate of absorption (mol/m
3
-s). For a resin obeying Beer’s Law,
the expression of the absorbed intensity at any point Q(x,y,z) (Figure 4) can be derived as
in Equation (14). I is the intensity incident on the resin surface (mol/m
2
-s), and ε
(m
3
/mol-m) and S (mol/m
3
) are the absorptivity and concentration of the initiator.

) , , ( 3 . 2
) 1 )( , , ( ) , , ( ) , , (
) , , (
3 . 2
0 z 0 z y, x,
lim lim
z y x SI
z
e z y x I
z y x
y x z y x I y x z y x I
z y x I
z S
a
ε
ε
=


=
∆ ∆ ∆
∆ ∆ ′ − ∆ ∆
=
∆ −
→ ∆ → ∆ ∆ ∆
(14)

4
(Fouassier, 1995; Crivello, 1998)

19


Figure 4 Absorbed Intensity at Point Q(x,y,z)

Fouassier (1995) has claimed that the absorbed light intensity can be expressed as:

) 1 (
3 . 2 l S
a
e I I
ε −
− = (15)

where l is taken as 1cm in order for
a
I to have the unit of photons/cm
3
-s. This always
gives the absorbed intensity at a point 1cm lower than where the light is incident on, i.e.
for an irradiance at point ) , , ( z y x : ) , , ( z y x I , ) cm 1 , , ( + z y x I
a
rather than ) , , ( z y x I
a
is
evaluated. Equation (14), however, eliminates this spatial inconsistency.
The rate of initiation thus can be rewritten as:

SI R
i i
ε φ 3 . 2 = (16)

The decay of the photoinitiator can be approximated as:


20
SI I
dt
dS
a
ε 3 . 2 − = − = (17)

The temperature dependence of the kinetic constants,
p
k and
t
k , is assumed to follow
the Arrhenius form. The reaction is faster at higher temperature.

RT E
Ep p
p
e A k
/ −
= (18)

RT E
Et t
t
e A k
/ −
= (19)

where
Ep
A and
Et
A are pre-exponential factors,
p
E and
t
E are activation energies for
propagation and termination, respectively, and R is the gas constant.
During the polymerization, the reaction is expected to accelerate due to the
temperature rise caused by the heat of reaction; however, this is not what happens
throughout the reaction. Due partially to the consumption of monomers and radicals, a
rate decrease is observed in both propagation and termination reactions. Another reason
for this is the decrease of the rate constants themselves. The rate constants are not only
dependent on temperature but on the free volume of the reacting system. With the
polymerization going on, the curing system becomes more viscous, the free volume
decreases, and the mobility of the reacting species is reduced. The reaction becomes
diffusion controlled. At the same temperature, the values of
p
k and
t
k are expected to be
larger in an environment with more free volume and less diffusion limitation.
Marten and Hamielec (1979, 1982) related the kinetic constants
p
k and
t
k directly to

21
the diffusion coefficients of monomer and polymer radicals, respectively, with a
temperature dependent proportionality constant. They assumed distinct regions exist for
reaction and diffusion controlled polymerization, and divided the course of reaction into
three conversion intervals to evaluate
p
k and
t
k . Bowman and Peppas (1991) adopted
the same idea and coupled these intervals with volume relaxation during polymerization.
These models don’t take any transition region into account and the parameters have to
switch to different values in different conversion ranges, i.e. each stage in the
polymerization has to be treated separately. This problem has been solved by combining
the reaction-controlled rate constants for propagation and termination and the diffusion-
controlled mechanisms to incorporate the transition regions for both k
p
and k
t
. The rate
constants k
p
and k
t
are expressed in terms of reaction resistances (Anseth and Bowman,
1993).

D r t p
k k k k
1 1
or
1
+ = (20)

where
r
k is the reaction limited (“true”) kinetic constant, and
D
k is the diffusion limited
kinetic constant.
For propagation, the resistances to reaction simply come from the reaction itself and
the monomer diffusion. For termination, the diffusion resistance is not only from the
translational and segmental diffusion of polymer radicals (
D
k
1
, translational diffusion is

22
negligible for highly crosslinked chains), but from the reaction diffusion (
*
1
r
k
, parallel to
the segmental diffusion resistance).


D r r t
k k k k +
+ =
*
1 1 1
(21)

According to Buback et al. (1989) and Buback (1990), the concept of reaction
diffusion has been put forward by Schulz (1956) and has been refined and put into
quantitative terms by several groups. The reaction diffusion is inherently a propagation
step – the “frozen” polymer radical propagates via the reactive monomer matrix until
encountering a second macroradical, which is also called “residual termination”. The rate
coefficient of this process,
*
r
k , is proportional to
p
k and to monomer concentration [M].
The proportionality constant
rd
R (called reaction diffusion parameter) is independent of
temperature, pressure, and conversion (Buback et al., 1989; Buback, 1990).

] [
*
M k R k
p rd r
= (22)

Anseth and Bowman (1993) assumed the diffusion limited kinetic constant
D
k to be
proportional to the diffusion coefficient of the reacting species and modeled it using the
Doolittle equation (Bueche, 1962). Equations (18) and (19), the expressions of
p
k and
t
k
without diffusion consideration, define the true kinetic constant
r
k for propagation and
the true kinetic constant
r
k for termination, respectively. Substituting all the above

23
information into Equations (20) and (21), the dependencies of the rate constant (
p
k or
t
k ) on both temperature and fractional free volume are incorporated into one equation
(Goodner et al., 1997, 1998, 1999, and 2002) which describes the rate of propagation or
termination throughout the whole polymerization course without changing parameter
values.

) / 1 / 1 (
0
1
cp p
f f A
p
p
e
k
k

+
= (23)

) / 1 / 1 (
0
0
/ ] [
1
1
ct t
f f A
t p rd
t
t
e k M k R
k
k
− −
+
+
= (24)

with
RT E
Ep p
p
e A k
/
0

= (25)

RT E
Et t
t
e A k
/
0

= (26)

where
0 p
k and
0 t
k are the true kinetic constants for propagation and termination,
respectively, f is the fractional free volume of the curing system,
cp
f and
ct
f are critical
fractional free volumes for propagation and termination, respectively, and
p
A and
t
A are
parameters that determine the rate at which the propagation and termination rate
constants decrease in the diffusion-controlled region (Goodner et al., 1997, 2002). When
the free volume of the polymerization system is much larger than the critical free volume,

24
there is no diffusion limitation on propagation or termination, Equation (23) or (24) is
reduced to Equation (18) or (19). The free volume decreases with the curing reaction
going on. When it decreases to be smaller than the critical free volume, the reaction
(propagation or termination) becomes diffusion limited. The diffusion resistances have to
be incorporated in the kinetic constants as in Equations (23) and (24).
For a curing system comprised of pure monomer and pure polymer, the fractional free
volume f is related to monomer conversion X as follows (Goodner et al., 1997, 2002):

) - 1 (
M P M M
f f f φ φ + = (27)

) - ( 025 . 0
gM M M
T T f α + = (28)

) - ( 025 . 0
gP P P
T T α f + = (29)

X
P
M
M
ρ
ρ
φ
+
=
X - 1
X - 1
(30)

In the above equations,
M
f and
P
f are the fractional free volumes of pure monomer
and pure polymer,
M
φ is the volume fraction of monomer, and the α ’s, T
g
’s, and ρ ’s are
the volumetric coefficients of expansion, glass transition temperatures, and densities,
respectively, of pure monomer and pure polymer. The free volume of the polymerization
system is dependent on both temperature and composition (conversion).

25
Goodner and Bowman (2002) also described the critical fractional free volume for
propagation or termination as a function of temperature:

|
.
|

\
|
− + =
ref ref
c c
T T AR
E
f f
1 1 1 1
(31)

From the kinetic model described above and the SL process model established in
Chapter 2, all the parameters (except the kinetic ones) involved are listed in Table 1. The
process & laser parameters can be recorded during the SL part building (as shown in
Chapter 5). The determination of material properties will be addressed in Chapter 4. The
kinetic experiment has been conducted and the kinetic model for a model material system
parameterized in sections 3.2 and 3.3, respectively.

26
Table 1 Material & Process Parameters Involved in the SL Cure Process Model

Parameters Symbols Units
Process Parameters laser scanning velocity V
s
m/s
bath temperature Tb K
thermal convection coefficient h
W/m
2
-
K
chamber temperature Ta K
Laser Parameters laser power P
L
W
wavelength λ nm
beam radius w
o
m
Material Properties thermal conductivity k W/m-K
heat of polymerization ∆Η
P
J/mol
absorptivity (initiator) ε
m
3
/mol-
m
initiation quantum yield


diffusion coefficient (monomer) D
M
m
2
/s
diffusion coefficient (radical) D

m
2
/s
diffusion coefficient (initiator) D
S
m
2
/s

coefficient of thermal expansion
(monomer) α
Μ
1/K

coefficient of thermal expansion
(polymer) α
P
1/K
glass transition temperature (monomer) Tg
M
K
glass transition temperature (polymer) Tg
P
K
heat capacity (monomer) C
PM
J/kg-K
heat capacity (polymer) C
PP
J/kg-K
density (monomer) ρ
Μ
kg/m
3

density (polymer) ρ
P
kg/m
3

Resin Compositions monomer concentration [M] mol/m
3

initiator concentration [S] mol/m
3


i
φ

27
3.2 Kinetic Experiments
In order to simulate the polymerization behavior in the SLA bath, a model material
system is identified and its kinetics characterized. Representative of the acrylate
compounds commonly used in SL (Steinmann et al., 1995, 1999; Pang et al., 2000;
Melisaris et al., 2000), Ethoxylated (4) PentaErythritol TetraAcrylate (E4PETeA,
SR®494, Sartomer) was chosen as the model compound. The 2,2-dimethoxy-2-
phenylacetophenone (DMPA, Irgacure®651, Ciba) was selected as its initiator. The
inhibitor was removed from the received E4PETeA by a prepacked inhibitor remover
column (Aldrich). 0.2wt% DMPA was added in the acrylate.



Figure 5 Structure Formula of E4PETeA (Sartomer)



Figure 6 Structure Formula of DMPA (Ciba)



28
3.2.1 DPC Pan
The differential photocalorimetry (DPC) technique was used to monitor the
photopolymerization kinetics. The aluminum pans were machined specially to have a
0.15mm depression to hold the sample. The sample size was determined accordingly to
fill the depression, by which the thickness uniformity is ensured (Tryson and Shultz,
1979). The depth of the depression was proved small enough for heat to dissipate quickly
so that the temperature uniformity through the sample can be assured.


Figure 7 DPC Sample Pan

3.2.2 Standing Waves
To ensure uniform reaction occurring in DPC so that diffusion and heat conduction
can be ignored in the mass and energy balance, the uniformity of light intensity through
the sample thickness also has to be assured. Considering the reflectivity of the aluminum
pan, the light absorption of the photosensitive sample is not the only factor that affects
the intensity uniformity. When the light is incident on the sample surface, the transmitted
light travels through the sample thickness and strikes the aluminum substrate, the
reflectivity of which is 0.9642 (Bass et al., 1985). Therefore, most of light is reflected off
the aluminum and travels through the sample again to the top surface, where the

29
reflection by the air-sample interface makes the light travel down into the sample and
then reflected by the aluminum. This process continues until the light wave dies out due
primarily to absorption. The waves traveling in opposite directions form a standing wave
in the sample. The intensity of the standing wave is the intensity that exposes the sample,
which takes not only the absorption but reflection into account. Mack (1985, 1986, and
1994) addressed in detail how the standing wave is formed in a thin film of absorbing
material coated on a reflective substrate and described its intensity quantitatively.
According to the equation by Mack (1985, 1986, and 1994), the electric field of the
light in the photosensitive sample on a reflective substrate can be calculated. The
intensity that exposes the sample can be obtained by squaring the magnitude of the
electric field (Mack, 1986). The magnitude of the standing wave intensity can be
different from material to material, depending mainly on the absorptivity and thickness of
the photosensitive material. Figures 8 and 9 demonstrate the standing wave intensity
through a 150µm thick sample (E4PETeA with 0.2wt% DMPA) contained in the
aluminum pan. Figure 8 shows the variation of the intensity at 365nm wavelength
through the sample thickness; Figure 9 sums the intensities at individual wavelengths
from 304 to 395nm. This wavelength range is where DMPA absorbs (Chapter 4) and the
light source irradiates (Appendix A). The standing wave intensity outside the material
absorption range is not of interest since it doesn’t contribute to the polymerization
initiation.


30


Figure 8 Standing Wave Intensity at 365nm



Figure 9 Standing Wave Intensity at 304-395nm

Obviously, the light intensity varies with depth into the sample. From Figure 9, it can
be seen that for the investigated curing system the average of the light intensities (304-
395nm) over depth is 1.15I
0
(I
0
is the intensity incident at the sample top surface).
Assuming reciprocity, the cure result under the standing wave intensity (Figure 9) should
be the same as that under uniform intensity 1.15I
0
through the sample thickness.
If the pan bottom doesn’t reflect, according to the Beer’s Law, the intensity of 0.93I
0

would be expected at the bottom surface (150µm depth). In this case the top surface

31
intensity I
0
can be taken as the intensity the sample receives through the thickness
(negligible light attenuation). Most published work involving DPC experiments took this
assumption, ignoring the aluminum reflection and sample absorption. The aluminum
reflection as well as the reflection from the sample-air interface, however, causes
standing wave formed in the sample, the intensity of which as discussed above is more
than 10% different from the top surface intensity I
0
.
Anseth (1994a) and Mateo (1997) and their coworkers took the reflection of the
aluminum pan into account, but they assumed the sample only received two doses, one
dose from the incident beam and a second dose from the reflected beam by the aluminum,
and came up with (1+0.9642)I
0
for the intensity that exposes the sample in the pan
(Recall the aluminum reflectance is 0.9642). These two light waves interfere with each
other, the electric field being the sum of them. The intensity, however, is not a simple
sum. The intensity is the square of the magnitude of the electric field for plane waves
(Mack, 1994). Furthermore, as described earlier, rather than one single reflected wave,
there are an indefinite number of reflected waves bouncing up and down in the sample
and a standing wave is formed. The average intensity the resin receives was found to be
15% (not 96.42%) higher than the top surface intensity. The consideration of aluminum
reflection by simply adding intensities of two doses together deviated even further away
from the intensity the sample actually receives than the adoption of top surface intensity
without reflection consideration.

32
3.2.3 DPC Experiments
1.16(±0.05)mg sample was put in the aluminum pan using a micropipette to cover the
0.15mm deep depression. The differential scanning calorimeter DSC Q1000 with photo
calorimetric accessory (TA Instruments) was adopted to monitor the photopolymerization
of the model acrylate resin. The light source Novacure 2100 (EXFO Photonic Solutions)
was used with filtered wavelength at 365nm (Appendix A). The incident power was
adjusted and measured to be 0.06mW. In order for the intensity measured to be exactly
the intensity incident on the sample surface in the actual experiment, a custom mount for
the power probe (PM3, used with laser power meter EPM 2000e, Molectron) and twin
light guides of the light source was designed and machined to simulate the DSC cell
environment.
The model material system developed here can also be thermally initiated and
polymerized. A set of isothermal DSC experiments demonstrate that the material won’t
be initiated thermally below 140
o
C (Figure 10). DPC experiments should be performed
below this temperature to avoid thermal polymerization. Note that the heat flow
oscillation in the initial stage represents the temperature overshooting and equilibration
behaviour. The big exothermal peak in 140
o
C DSC curve, indicating the heat of
polymerization, illustrates that the polymerization can be thermally initiated at and above
this temperature. The absence of this peak in 130
o
C DSC curve shows that the material
system won’t polymerize at or below 130
o
C.


33

Figure 10 Isothermal DSC Runs to Detect the Onset Temperature of Thermal Cure

In this study, limited by the liquid light guide requirement on temperature, the DPC
experiments cannot be conducted above 70
o
C. Both continuous and flash exposure
experiments were carried out isothermally at three different temperatures (30, 50, 70
o
C).
During the continuous irradiation experiment, the light is on until the heat flow curve
drops to the baseline, i.e. the reaction is complete under the current temperature. The light
is on only for a very short time in flash exposure experiments. For each temperature, five
or more different flash times were used in order to extract the kinetic constants at
different conversions. A typical set of experimental data is shown in Figure 11.

34
-2
0
2
4
6
8
10
H
e
a
t

F
l
o
w

(
W
/
g
)
0 1 2 3 4 5
Time (min)
––––––– 50C(cont irrad).001
– – – – 50C(flash0.2m)
––––– · 50C(flash0.4m)
––– – – 50C(flash0.6m)
––– ––– 50C(flash0.8m)
Exo Up


Figure 11 DPC Experimental Curves (Continuous and Flash Exposure at 50
o
C)

The heat flow curves are only contributed by the reaction generated heat. As shown in
Appendix B, after the reaction is completed, no noticeable change in heat flow curve is
observed when the light is turned off. This demonstrates that the heating effect of the
DPC light source is negligible (Lecamp et al., 1997). A quantitative estimation also
shows that the heating by light is insignificant (~10
0
J/mol for continuous irradiation and
10
-1
J/mol for very short time exposure) comparing with the heat generated by reaction
(~10
5
J/mol for continuous irradiation and 10
4
J/mol for very short time exposure).


35
3.3 Kinetic Data Analysis & Model Parameterization
Assuming that the heat produced by the reaction is proportional to the amount of
monomer reacted (Burel et al., 1999, Cook, 1992, 1993, Lecamp et al., 1997, 1999, and
Maffezzoli and Terzi, 1995, 2001), the degree of cure, X, in DPC experiments can be
defined as follows:


tot
Q
t Q
X
) (
= (32)

Correspondingly, the rate of propagation normalized by the initial monomer
concentration is expressed as follows (the monomer consumption by initiation, ignored in
Equation 33, is negligible compared with the consumption during propagation):


tot
P
N
Q dt
dQ
dt
dX
M
R
R
1
] [
0
× = = = (33)

where ) (t Q is the heat developed at any time t during a DPC measurement, which is the
integration of the heat flow signal
dt
dQ
.
tot
Q is in principle the total heat of reaction when
all the monomers are converted.
tot
Q is approximately the heat of polymerization
measured at high enough temperature and light intensity.



36
For the dark reaction in the flash exposure experiment,

2
] [
] [
• − =

P k
dt
P d
t
(34)

Integrating the above equation and combining it with Equation (12), ] ][ [ M P k R
P P
• = ,
leads to:

1 1 2 2
] [
) (
] [
t
R
M
t t
k
k
t
R
M
P P
t
P
+ − = (35)

Therefore, the ratio
P t
k k / can be determined from the slope of the plot of
P
R M / ] [
(i.e.
N
R X / ) 1 ( − ) as a function of time t (Tryson and Shultz, 1979). The starting time t
1
is
a time point after the light is turned off. The ending time t
2
is taken well before the
reaction dies out.
At each temperature (e.g. 30, 50, and 70
o
C), from a series of flash exposure
experiments (e.g. 0.2, 0.4, 0.6, and 0.8 min irradiation) the value of
P t
k k / at different
conversions can be extracted, as shown in Table 2.


37
Table 2
P t
k k / and
2 / 1
/
t P
k k Values Obtained from DPC Experiments

Flash Exposure dark reaction Continuous Irradiation QSSA Effect
T(
o
C) Time (min) X kt/kp kp/kt^0.5 kp/kt^0.5 (QSSA) Difference (%)
30 0.2 0.150 111.25 0.37 0.37 0.2
0.4 0.339 21.66 0.16 0.18 12.3
0.6 0.378 20.78 0.12 0.13 12.2
0.8 0.424 18.94 0.08 0.09 15.2
50 0.2 0.163 182.72 0.34 0.34 0.2
0.4 0.342 19.46 0.17 0.19 10.5
0.6 0.404 14.92 0.12 0.14 13.4
0.8 0.493 8.12 0.06 0.08 26.9
70 0.2 0.277 41.96 0.38 0.39 2.5
0.4 0.455 11.43 0.15 0.18 21.2
0.6 0.491 10.48 0.12 0.14 21.5
0.8 0.567 7.09 0.07 0.09 27.3

For the continuous irradiation,

2
] [
] [
• − =

P k R
dt
P d
t i
(36)

Integrating and combining it with Equation (12), ] ][ [ M P k R
P P
• = , to obtain:

) ( ) ( 2 )
1
1
ln( )
1
1
ln(
1 2
2 / 1
1
1
2
2
t t k R
y
y
y
y
t i
− =

+


+
(37)

with
2 , 1
)
X - 1
(
2 / 1
2 / 1
2 , 1 t
N
i P
t
R
R k
k
y = (38)

where
i
R is evaluated using Equation (16), SI R
i i
ε φ 3 . 2 = , and Equation (17),

38
SI
dt
dS
ε 3 . 2 − = . Since both photoinitiator absorptivity and light intensity are a function of
wavelength,

2
1
) ( ) (
λ
λ
λ λ λ ε d I is substituted for I ε in Equations (16) and (17). Equation
(37) combines two kinetic constants together in terms of
2 / 1
/
t P
k k . For each temperature,
a trial and error analysis is performed using
P t
k k / data derived from the dark reaction to
evaluate
2 / 1
/
t P
k k values at corresponding conversions.
Table 2 lists the ratio
P t
k k / obtained from flash exposure experiments and the ratio
2 / 1
/
t P
k k obtained from continuous irradiation experiments. The values of
P
k and
t
k can
thus be determined separately at several different conversions for each temperature (as
shown in Table 3), from which the free volume and temperature dependence of
P
k and
t
k , i.e. the parameters in Equations (23) and (24), can be determined.

Table 3 Determination of Rate Constants
t
k and
P
k

T(
o
C) X kt/kp kp/kt^0.5 kt(m
3
/mol-s) kp(m
3
/mol-s)
30 0.15 111.3 0.37 1694 15
0.34 21.7 0.16 12 0.6
0.38 20.8 0.12 6 0.3
0.42 18.9 0.08 2 0.1
50 0.16 182.7 0.34 3860 21
0.34 19.5 0.17 11 0.6
0.40 14.9 0.12 3 0.2
0.49 8.1 0.06 0.2 0.03
70 0.28 42.0 0.38 254 6
0.45 11.4 0.15 3 0.3
0.49 10.5 0.12 2 0.2
0.57 7.1 0.07 0.3 0.04


39
As shown in Table 3, at low conversion, the magnitude of
t
k is two order higher than
P
k . Goodner and coworkers (1997) found that at similar conversion, the magnitude of
t
k
is three orders higher than
P
k for 2-hydroxyethyl methacrylate (HEMA). It’s possible
that the steric resistance in tetra-functional material studied here affects more
significantly the true kinetic constants for termination than for propagation. For both
HEMA and E4PETeA, the rate constants
t
k and
P
k drop dramatically with the
conversion increasing. The drop is more significant for tetraacrylate due to the
crosslinkage. For both HEMA and E4PETeA,
t
k drops at a lower conversion than
P
k .
The scientific plotting and data analysis software Origin (OriginLab Corporation) is
used for the nonlinear curve fitting of
P
k and
t
k data to determine the free volume
dependence, as shown in Figures 12 and 13. Linear curve fitting in Excel (Microsoft
Corporation) is performed to determine the temperature dependence of true kinetic
constants
0 P
k and
0 t
k (Equations 25-26), as demonstrated in Figure 14. Figure 15
illustrates the temperature dependence of critical fractional free volume (Equation 31) for
propagation and termination.


40
0.0 0.2 0.4 0.6
-2
0
2
4
6
8
10
12
14
16
k
p

(
m
3
/
m
o
l
-
s
)
Conversion X

Figure 12 Nonlinear Fit of Propagation Rate Constant
P
k vs. Conversion X (50
o
C)



0.0 0.2 0.4 0.6
-500
0
500
1000
1500
2000
2500
3000
3500
4000
4500
k
t

(
m
3
/
m
o
l
-
s
)
Conversion X

Figure 13 Nonlinear Fit of Termination Rate Constant
t
k vs. Conversion X (50
o
C)


41

Figure 14 Semi-log Plot of True Kinetic Constants
0 P
k and
0 t
k vs. 1/T


Figure 15 Linear Fit of
c
f / 1 (Critical Fractional Free Volume) vs. 1/T



42
The fractional free volume of the tetraacrylate system is found to be one order of
magnitude higher than that of HEMA (Goodner, et al., 1997) because the CTE of the
system is one order of magnitude higher than that of HEMA. The linear CTE of the
tetraacrylate system (see Chapter 4), however, is within the same range of the CTE used
for the HEMA system (Goodner, et al., 1997). Accordingly, the critical fractional free
volume for propagation and termination of E4PETeA (Figure 15) turns out to be one
order of magnitude higher than that of HEMA (Goodner, et al., 1997).
The values of the kinetic parameters obtained are listed in Table 4.

Table 4 Kinetic Parameter Values
Parameters Symbols Values Units
free volume parameter for propagation
(see Equation 23) Ap 6.1 N/A
free volume parameter for termination
(see Equation 24) At 6.4 N/A
reaction diffusion parameter
(see Equation 24) Rrd 0.013 m
3
/mol
pre-exponential factor for propagation
(see Equation 25) A
Ep
28.4 m
3
/mol-s
pre-exponential factor for termination
(see Equation (26) A
Et
8916 m
3
/mol-s
activation energy for propagation
(see Equation 25) Ep 1627 J/mol
activation energy for termination
(see Equation 26) Et 2103 J/mol


It should be mentioned that instead of integrating Equation (36) to obtain the
relationship of
P
k and
t
k , people usually apply the quasi steady state assumption
(QSSA) and Equation (36) is then reduced to:


2
] [ • = P k R
t i
(39)

43
Substituting the above equation into Equation (12), ] ][ [ M P k R
P P
• = , the ratio
2 / 1
/
t P
k k can be directly calculated as follows:


i
N
t
P
R X
R
k
k
) 1 (
2 / 1

= (40)

Table 2 also lists the values of
2 / 1
/
t P
k k determined with steady state assumption and
their comparison with the values obtained without this assumption. It turns out that QSSA
is valid only at low conversions; in this specific case, it has more than 10% deviation
when the conversion is greater than 30%.
To analyze the kinetic data and parameterize the kinetic model, Goodner and
coworkers (1997) have divided the polymerization process under continuous light
irradiation into four regions based on the free volume and treated the four regions
individually to determine the kinetic parameters. The advantages of this regional analysis
method are that only continuous irradiation experiments need to be conducted (flash
exposure experiments are not required) and unlike the nonlinear curve fitting, a unique
parameter set can be expected. However, for a highly crosslinked polymerization system
as investigated here, the first region where there are no diffusional limitations on either
propagation or termination and the third region where there are no diffusional limitations
on propagation but termination is reaction-diffusion controlled are often found to be ill-
defined. The second region, autoacceleration, could also be ill-defined or just have not
enough data to determine parameters. Often only the fourth region, autodeceleration, is
defined well. This limits the application of the regional analysis method which requires

44
the system have four distinct kinetic regions. To broaden its application, the flash
exposure experiment (so-called unsteady state analysis) has been proposed to find
0 t
k (or
0 P
k ) for systems that don’t have a well-defined first (or third) region (Goodner et al.,
1997). In this case, however, the regional analysis has lost one of its attractive
characteristics mentioned above. In addition to the continuous irradiation experiments,
flash exposure experiments also have to be conducted and analyzed to find
0 P
k or
0 t
k to
complement the regional analysis. Furthermore, the regional analysis method is
established based on the QSSA assumption throughout the reaction, which is not valid as
shown in Table 2.

3.4 Kinetic Model Validation
It’s not easy to find a unique solution for the nonlinear curve fitting. Other restrictions
should be applied for the fitted parameters in order to obtain a reasonable and unique
parameter set. The fitted parameters should not only achieve a best fit for the kinetic data,
but should be physically reasonable as well as capable of predicting kinetic behavior.
Figure 16 demonstrates the kinetic model simulations for a series of continuous
irradiation experiments conducted at different temperatures with the light power of 0.1
mW. The agreement between the predicted and experimental results validates the adopted
kinetic model and the determined kinetic parameters.


45
0
0.005
0.01
0.015
0.02
0.025
0 0.2 0.4 0.6 0.8 1
conversion X
d
X
/
d
t

(
1
/
s
)
predicted
experimental

(a)
0
0.005
0.01
0.015
0.02
0.025
0 0.2 0.4 0.6 0.8 1
conversion X
d
t
/
d
X

(
1
/
s
)
predicted
experimental

(b)
0
0.005
0.01
0.015
0.02
0.025
0 0.2 0.4 0.6 0.8 1
conversion X
d
X
/
d
t

(
1
/
s
)
predicted
experimental

(c)
Figure 16 Comparison of the Experimental and Simulated Polymerization Rate Curves
(incident power = 0.1 mW): (a) 30
o
C, (b) 50
o
C, (c) 70
o
C


46
CHAPTER 4
MATERIAL CHARACTERIZATION

As listed in Table 1, some process and material parameters need to be determined in
order to simulate the SL cure process. The process and laser parameters (such as laser
scanning speed V
s
, bath temperature T
b
, laser power P
L
, and laser beam radius w
o
) are
obtained from the actual part building process in SLA-250/50 (3D systems, laser
wavelength λ = 325nm
5
). h = 4.18 W/m
2
-K is taken as the value of heat convection
coefficient at the interface of the natural air flow and the resin (Pananakis & Watts,
2000).
i
φ = 0.6 is taken as the quantum efficiency of initiation for DMPA (Goodner et
al., 2002). The thermal and physical properties of the resin are evaluated for the model
compound system comprised of E4PETeA tetraacrylate and 2 wt% photoinitiator DMPA.
The initiator concentration is higher than in Chapter 3 to facilitate the SL cure process.
They are obtained from literature, theoretical approximation, or experimental
determination.

4.1 Specific Heat Capacity
The modulated differential scanning calorimeter (MDSC) option for DSC 2920 (TA
Instruments) was used to measure the specific heat capacity of the pure monomer and
pure polymer. MDSC provides unique capabilities besides those of standard DSC such as
separation of complex transitions, detection of weak transitions, accurate measurement of

5
SLA Systems Specifications, 3D Systems, http://www.3dsystems.com

47
polymer crystallinity, and direct determination of heat capacity and thermal
conductivity.
6

In addition to the standard DSC cell calibration (performed in calibration mode) for
cell constant, baseline and temperature, a heat capacity calibration (performed in
modulated mode) was also performed in order to obtain accurate heat capacity
measurement. The sapphire standard (provided by TA Instruments) was used as the
calibrant. The calibration procedure should be as close as possible to that of the following
measurements. The weights of two pans with lids were matched to within ±0.1mg. One
pair of pan and lid were sealed and used as reference; the other pair was used to hold the
weighed calibrant or sample and then sealed and placed in the sample position in the
DSC cell. The method which tells the machine what to execute was formulated by
combining the recommendations on heat capacity calibration and on heat capacity
measurements as well as general MDSC operating parameters.
6

The liquid nitrogen
cooling accessory (LNCA) was used for the optimum performance of the MDSC
measurement. The nitrogen gas was used to purge and circulate in the DSC cell at a rate
of 40ml/min before and during the experiment. A data sampling interval of 1.0
seconds/point was used. The MDSC directly measures the heat capacity and stores the
signal.
Figures 17 and 18 are exported plots of heat capacity signal for liquid E4PETeA
monomer and its cured polymer, respectively. Samples were weighed 12.09±0.03 mg
using an analytical balance (AG 245, Mettler Toledo).


6
‘Modulated DSC
TM
Option’, DSC 2920 Differential Scanning Calorimeter Operator’s Manual, Thermal
Analysis & Rheology, TA Instruments, 1995.

48

Figure 17 C
p
-T Plot of Liquid E4PETeA Monomer Exported from MDSC Data


Figure 18 C
p
-T Plot of Cured E4PETeA Polymer Exported from MDSC Data

49
The heat capacities were found to be functions of temperature as follows:

6 . 218 ) ( 6 . 5
,
+ × = K T C
M P
(41)


,
9.1 ( ) 1535.5
P P
C T K = × − (42)

where
M P
C
,
and
P P
C
,
are the heat capacities (J/Kg-K) of monomer and cured polymer,
respectively.
The molar heat capacity of liquid E4PETeA monomer was also calculated to be 947
J/mol-K (i.e., 1.8 J/g-K for specific heat capacity) at 25
o
C by the addition of group
contributions (Van Krevelen, 1990). This calculated result is within 5% of the
experimental value at the same temperature, which justifies the experimental
measurement. Furthermore, the heat capacity value of E4PETeA is close to those of other
acrylates such as methyl and butyl acrylates, etc (Yaws, 2003).
A weight-averaged heat capacity was used for the curing material, i.e., mixture of
monomer and cured polymer:

X C X C C
P P M P P , ,
) 1 ( + − = (43)

where X is monomer conversion.


50
4.2 Glass Transition Temperature
The glass transition temperatures of liquid E4PETeA monomer and its cured polymer
are determined using a standard differential scanning calorimeter (DSC 2920, TA
Instruments). The samples were weighed ~16mg for several heating rates: 5, 10, 15, 20
o
C/min. Figures 19 and 20 are the heat flow curves at 10
o
C/min heating rate and
demonstrate the glass transition of liquid monomer and cured polymer, respectively.
Figures 21 and 22 illustrate the effect of heating rate on the T
g
measurement. In the range
of heating rates tested, the measured T
g
value increases linearly with the heating rate. For
T
g
measurement, a heating rate within 10-20
o
C/min is recommended. A lower heating
rate leads to gradual material change and thus less obvious glass transition.



Figure 19 Glass Transition of Liquid E4PETeA Monomer Detected by DSC

51



Figure 20 Glass Transition of Cured Poly(E4PETeA) Detected by DSC


-67
-66
-65
-64
-63
-62
-61
0 5 10 15 20 25
y = -67.547 + 0.28267x R
2
= 0.98663
T
g

(
o
C
)
Heating Rate (
o
C/min)


Figure 21 Effect of Heating Rate on Measured T
g
Value of Liquid Monomer

52
220
222
224
226
228
230
232
234
236
0 5 10 15 20 25
y = 215.23 + 0.99617x R
2
= 0.99246
T
g

(
o
C
)
Heating Rate (
o
C/min)


Figure 22 Effect of Heating Rate on Measured T
g
Value of Cured Polymer

To eliminate the heating rate effect, the T
g
values obtained by extrapolating linear
curves in Figures 21 and 22 to 0
o
C/min, -67.5
o
C and 215.2
o
C, are adopted for liquid
monomer and cured polymer, respectively.

4.3 Coefficient of Thermal Expansion
The coefficients of thermal expansion (CTE) of monomer and cured polymer were
determined by using an ellipsometry technique to measure film thickness at different
temperatures. The linear CTE is defined as (Van Krevelen, 1990):

T
L
L ∂

=
1
β (44)


53
where
T
L


is the slope of the film thickness versus temperature plot, and L the average
thickness over the temperature range investigated.
The variable angle spectroscopic ellipsometer (VASE VB 250, J.A. Woollam) was
used to determine the film thickness at elevated (heating) or lowered (cooling)
temperatures. A hot plate is installed on the commercial ellipsometer. The temperature
controller (OMEGA CN 76000) can control temperature within ±0.2
o
C. Thermocouple
(HH 11, OMEGA) was used for temperature calibration. At each set temperature, the
ellipsometer scan starts after the film reaches thermal equilibrium. The film was spin-
coated on silicon substrate (with native oxide layer) from a 10 wt% propylene glycol
methyl ether acetate (PGMEA) solution. The monomer film was put in the vacuum oven
and baked at 90
o
C in vacuum for 1hr to remove the solvent without solidifying the
monomer. The solid polymer film was obtained by baking the liquid film containing
monomer and photoinitiator and solvent at 180
o
C in vacuum for 60 hrs. No phenomenon
such as discoloration or brittleness was observed, hence no apparent degradation occurred
Figures 23 and 24 demonstrate the temperature dependence of monomer film
thickness above T
g
and of polymer film thickness below T
g
, respectively. β (monomer) =
5.9×10
-4
1/K and β (polymer) = 0.96×10
-4
1/K are found from these two graphs.



54


Figure 23 Temperature Dependence of Liquid E4PETeA Monomer Film Thickness



Figure 24 Temperature Dependence of Cured E4PETeA Polymer Film Thickness

55
The heat treatment of the film before ellipsometric measurement reduced the
entrapped solvent enough that no solvent effect was observed during the heating or
cooling stages. Two L-T curves are found to almost overlap with each other in Figure 23.
In Figure 24, the fit for the first temperature scan (heating cycle) has a slightly higher
slope than that for the second scan (cooling cycle). This is probably due to the residual
unconverted monomer which has greater CTE entrapped in the polymer matrix.
The films with thickness above 1000 Å were made for measurement. For the films
with thickness below 1000 Å, the thermal fluctuation of the air above the film could
cause a big error in the thermal property quantification (Kahle et al., 1998).
The temperature dependence of Si substrate n & k spectra (complex refractive index:
) ( ) ( ) ( λ λ λ ik n + = n ) was taken into account when fitting the ellipsometric data to
determine the film thickness. A slower increase in CTE was observed with film thickness
decreasing, compared with the result from the fit with only the optical properties of Si at
25
o
C, as shown in Figure 25. The thickness variation (500-2400 Å) was achieved by
varying the spin speed and time. The CTE increases drastically for thickness below 2000
Å, but remains approximately constant for greater thickness.
Kahle and coworkers (1998) demonstrated that when temperature dependent substrate
data were used for the fit, there was no pronounced thickness effect for the CTE within
the thickness range of 500 to 10
5
Å for the poly(methyl methacrylate) (PMMA) film they
investigated. This is different from what we observed here for the poly(E4PETeA) film,
which might indicate that the trend discussed here depends on the material properties of
the film such as molecular weight or cross-link density. The CTE value at greater
thickness (2400Å here) was taken as the bulk CTE, β (polymer) = 0.96×10
-4
1/K.

56



Figure 25 CTEs of Poly(E4PETeA) below T
g
Determined by Linear Regression of
Curves Obtained by Fitting with Si Substrate Optical Data of 25
o
C (diamonds) and of
Curves Obtained by Fitting with Temperature Dependent Si Substrate Data (triangles)


The polymer CTE value, however, was measured under the constraint of the Si wafer
and therefore it overestimates the true value. The true CTE is related to the constrained
CTE by the following equation (Kahle et al., 1998):


ν
ν
β β
+

× =
1
1
d constriane ned unconstria
(45)


57
where ν is Poisson’s ratio. The
ν
ν
+

1
1
term converts expansion constrained by the Si
substrate to a true unconstrained CTE value. Taking ν below T
g
as 0.40 (typical Poisson’s
ratio value for polymers, Van Krevelen, 1990) for the poly(E4PETeA), the true CTE is
calculated to be 0.4×10
-4
1/K.
The liquid film, on the other hand, is not constrained by the substrate, therefore, the
measured value is the true bulk CTE, β (monomer) = 5.9×10
-4
1/K.
The volumetric CTE can be obtained by the following equation (Van Krevelen,
1990), assuming the bulk material is isotropic.


ned unconstrai
T
V
V
β α 3
1
=


= (46)

where V is the volume of the material over the temperature range investigated.
The volumetric CTEs of E4PETeA (above T
g
) and its polymer (below T
g
) are thus
determined to be 1.77×10
-3
, and 1.23×10
-4
1/K, respectively. These values are at the same
magnitude as CTEs of ethylene glycol dimethacrylate (EGDMA) and its polymer
(Bowman and Peppas, 1991). The polymer CTE value is also in the same range as
PMMA (Brandrup and Immergut, Ed., 1989). The monomer CTE is at the same order of
magnitude as other acrylates such as methyl and butyl acrylates, etc (Yaws, 2003).


58
4.4 Density
The density of the cured polymer was found to be 1200 Kg/m
3
at 35
o
C (column
control temperature) by using density gradient column (DC-4, Techne). Two water-
calcium nitrate solutions of different concentrations were used to fill the column and
form a linear density gradient from top to bottom.
The temperature dependence of density can be described as follows using the
volumetric CTE α :

) 308 ( 1
1200
) 308 ( 1
) 308 (
− +
=
− +
=
T T
K
P P
P
P
α α
ρ
ρ (47)

Similarly, we have the following for the monomer density:

) 298 ( 1
1128
) 298 ( 1
) 298 (
− +
=
− +
=
T T
K
M M
M
M
α α
ρ
ρ (48)

where
M
ρ (298 K) =1128 Kg/m
3
from the product technical data sheet.
The density of cured polymer was also calculated at 298 K. Using a group
contribution method (Van Krevelen, 1990), the molar volume per structural unit of the
polymer was calculated to be 404.42 cm
3
/mol at 298 K. With the unit molecular weight
of 528 g/mol, the density of the cured polymer was found to be 1290 Kg/m
3
at 25
o
C,
which is within 10% of the value obtained from Equation (47) for the same temperature.
This justifies the measurement and Equation (47) will be adopted.


59
The density of the curing material system can be expressed as:

) 1 (
M P M M
φ ρ φ ρ ρ − + = (49)

where
P
ρ and
M
ρ are described in Equations (47) and (48), respectively, and
M
φ is the
monomer volume fraction as described before.

4.5 Thermal Conductivity
The thermal conductivity of polymer can be calculated using the following equation
(Van Krevelen, 1990):


2 / 1
3
1
) 1 ( 3
(
¸
(

¸

+

|
|
.
|

\
|
=
ν
ν
ρ
m
R
f P
V
U
L C k (50)

where ρ ,
P
C ,
f
L , U
R,

m
V , and ν are density, specific heat capacity, average free path
length, Rao function, molar volume per structural unit and Poisson’s ratio of the cured
polymer, respectively. It can be obtained from Sections 4.1 and 4.3 that at 298 K, ρ =
1.25 g/ml,
P
C = 1.18 J/g-K, and
m
V = 404.42 cm
3
/mol.
f
L = 5×10
-11
m for PMMA is
taken. The Rao function, U
R
, is calculated to be 22,460 (cm
3
/mol)⋅(cm/s)
1/3
using a group
contribution method (Van Krevelen, 1990). The factor
2 / 1
1
) 1 ( 3
(
¸
(

¸

+

ν
ν
is nearly constant for

60
solid polymers (≈ 1.05) (Van Krevelen, 1990). The thermal conductivity of the cured
polymer is thus calculated to be 0.123 W/m-K at 298 K.
The thermal conductivity of polymer is temperature dependent. From a generalized
plot of k(T)/k(T
g
) as a function of T/T
g
based on available experimental data (Van
Krevelen, 1990), thermal conductivity of amorphous polymers can be evaluated at
different temperatures. The thermal conductivity of the cured E4PETeA polymer at its
glass transition temperature 230
o
C (Section 4.2) was thus found to be 0.135 W/m-K,
which is within 10 % of the value at 25
o
C and therefore the temperature dependence can
be ignored in the temperature range during the cure reaction.
The thermal conductivity of the liquid acrylate monomer was measured using the
relative transient hot-wire method (Sun and Teja, 2003). A U-shape Pyrex cell, with
capillary as part of it, filled with liquid mercury is inserted into the liquid sample. The
Pyrex capillary is employed as the wire. A Hewlett-Packard (Model 6213A) power
supply is used to provide the voltage for heating. A thermocouple is used to measure the
sample temperatures. Further details of the experimental apparatus and procedure as well
as theory were described by DiGuilio and Teja (1990). The thermal conductivity of the
liquid E4PETeA is found to be 0.161 W/m-K at 297.8 K by averaging the results of five
experiments. The value is reproducible within 0.5% and close to the thermal conductivity
values (~0.13 W/m-K at 297.8 K) of other acrylates such as butyl acrylate and methyl
acrylate, etc (Yaws, 2003). The temperature dependence is insignificant and thus ignored
within the SL cure temperature range (refer to other acrylates, Yaws, 2003).
The later modelling results demonstrate that thermal conductivity is not a sensitive
parameter. For approximation, the averaged value of the cured polymer and liquid

61
monomer (0.142 W/m-K) can be taken as that of the curing material system to be used in
the process model. This value is at the same order of magnitude as that used for
hexanedioldiacrylate (HDDA) curing system, 0.2 W/m-K (Flach and Chartoff, 1995a).

4.6 Heat of Polymerization
The isothermal standard DSC experiments performed on the model material show
that the thermally initiated polymerization doesn’t occur below 130
o
C. The DPC
experiments were performed at constant light intensity (0.36 mW/cm
2
) for several
different temperatures below 130
o
C. The heat generated due to polymerization was
found to increase with temperature linearly (Figure 26).

380
400
420
440
460
480
500
520
20 40 60 80 100 120 140
y = 352.36 + 1.1758x R
2
= 0.9928
R
e
a
c
t
i
o
n

G
e
n
e
r
a
t
e
d

H
e
a
t


(
J
/
g
)
Temperature (
o
C)


Figure 26 Effect of Temperature on Heat Generated by Polymerization


62
Additional standard DSC experiments were conducted at elevated temperatures (till
350
o
C at a rate of 10
o
C/min) for samples irradiated at 130
o
C. A small amount of
residual heat was detected and added to give the maximum total heat of 540 J/g generated
at light intensity = 0.36 mW/cm
2
.
The DPC and subsequent DSC experiments were repeated for higher light intensities
(30, 40, 50, and 60 mW/cm
2
) and no more heat due to reaction was detected.
Therefore, 540 J/g can be taken as the heat of polymerization of the model material
used.
The heat of polymerization was also calculated to be 650 J/g from the theoretical
enthalpy of 20.6 kcal/mol per acrylate double bond (Anseth et al., 1994b). This value is
within 20 % of the experimental result.

4.7 Absorption Coefficient
The absorption coefficient of photoinitiator, DMPA, was determined by using a UV-
VIS spectrometer (Lambda 19, Perkin Elmer) and Beer’s law. To obtain the absorption
spectrum of DMPA in its E4PETeA solution, spectral subtraction (Smith, 1996) was
performed.

A(DMPA)= A (solution) - subtraction factor A × (monomer) (51)

where A represents the absorption spectrum. The 0.05, 0.1, 0.2 wt% DMPA in E4PETeA
were used as sample and pure E4PETeA monomer as reference in the spectrometer. The
absorption spectrum thus obtained is the direct subtraction of the absorption of pure

63
monomer from that of solution. The subtraction factor the reference absorption is
multiplied by was taken as 1.0 due to the low concentrations investigated. The
investigated system assumed to obey the Beer’s law, the extinction coefficient spectra of
the three solutions of different concentrations overlap with one another (Figure 27).

0
5
10
15
20
25
30
300 325 350 375 400 425 450 475 500
0.05wt%
0.1wt%
0.2wt%
E
x
t
i
n
c
t
i
o
n

C
o
e
f
f
i
c
i
e
n
t

(
m
3
/
m
o
l
-
m
)
Wavelength (nm)
Concentration of DMPA
(wt% in E4PETeA)


Figure 27 Absorption Coefficient Spectrum of DMPA


64
4.8 Summary
In this chapter, the material thermal and physical properties are measured
experimentally and verified by the theoretical calculation and literature values for similar
materials. These values are listed in Table 5 and used in the SL process model established
in Chapter 2.

Table 5 Characterized Material Properties

Material Parameters Values Units
thermal conductivity 0.142 W/m-K
heat of polymerization 2.85e5 J/mol
absorptivity (initiator) 19.9 m
3
/mol-m
quantum yield of initiation 0.6
7

coefficient of thermal expansion
(monomer) 0.00177 1/K
coefficient of thermal expansion
(polymer) 0.00012 1/K
glass transition temperature (monomer) 205.65 K
glass transition temperature (polymer) 488.35 K
heat capacity (monomer)
6 . 218 ) ( 6 . 5
,
+ × = K T C
M P

J/kg-K
heat capacity (polymer)
,
9.1 ( ) 1535.5
P P
C T K = × −
J/kg-K
heat capacity (curing system)
X C X C C
P P M P P , ,
) 1 ( + − =
J/kg-K
density (monomer)
1128/(1 ( 298))
M
T α + −
kg/m
3

density (polymer)
1200/(1 ( 308))
P
T α + −
kg/m
3

density (curing system)
) 1 (
M P M M
φ ρ φ ρ ρ − + =
kg/m
3







7
Goodner et al., 2002

65
CHAPTER 5
SIMULATIONS

With the kinetic parameters determined in Chapter 3 (Table 4), material properties
evaluated in Chapter 4 (Table 5), and laser and process parameters recorded in the part
building process, the SL cure process model established in Chapter 2 is solved using the
multiphysics modelling and simulation code FEMLAB. FEMLAB is a product of the
COMSOL Group
8
and has many model types available for use (application models). It
also supports equation-based modelling, enabling users to enter their specific differential
field equations. Application models were used in this research.
The process model established earlier can be easily customized in the FEMLAB
environment. Since SL curing is a coupled mass and energy balance problem, two
application models, diffusion and heat transfer by conduction, have been employed to
accomplish the description of the cure process model. The transient analysis mode is
selected. The 2D geometry described in Chapter 2 is the domain in which the balance
equations apply when the laser draws a single line. As mentioned earlier, a small domain
size has been adopted initially, which has then increased until no significant deviation in
the modeling results from different domain sizes is observed, i.e., the domain should be
large enough to accommodate the phenomena occurring physically. The balance
equations established in the process model are consistent with those described in
FEMLAB application models. The initial conditions are applied to the domain and
boundary conditions applied to each boundary of the domain. The numerical values or

8
COMSOL Group, http://www.comsol.com/

66
formula descriptions of the material, process, and kinetic parameters also enter the
software. Triangular, quadratic, and Lagrange elements have been selected for domain
discretization. The area where the reaction occurs and the resin properties vary
significantly has finer mesh. The initial and upper limit of the time step size can be set
manually. The absolute tolerance has been set for each individual dependent variable
based on their initial values. The absolute and relative tolerances determine the limit for
the error estimated in each integration step
9
. The model is then solved using a time-
dependent nonlinear solver in the software.
Three basic cases of the laser drawing patterns in SL are simulated (Figure 28): a
single laser drawn line (also see Figure 5), overlapping single-layer lines with certain
spacing, and stacked single lines with certain layer thickness.



Case I Case II Case III


Figure 28 Three Basic Laser Drawing Patterns: Case I. Single Laser Drawn Line,
Case II. Overlapping Single-Layer Lines, Case III. Stacked Single Lines


For each case, the mesh convergency, time stepping convergency, and domain
convergency (i.e. the solution is converging to a stable value as the mesh is refined, the

9
“User’s Guide – FEMLAB 3.0”, COMSOL Group.

67
time step size is reduced, or the domain is enlarged) have been performed to ensure valid
and accurate solution.
All the simulations presented here have used ethoxylated (4) pentaerythritol
tetraacrylate loaded with 2 wt% 2,2-dimethoxy-2-phenylacetophenone as photoinitiator.
The values of the process and laser parameters used for the simulations are listed below.

Table 6 Process and Laser Parameter Values Used for Simulations

Parameters Values Units
Process Parameters laser scanning velocity 0.0272 m/s
bath temperature 304.55 K
thermal convection coefficient 4.18 W/m
2
-K
chamber temperature 300.48 K
Laser Parameters laser power 0.0288 W
wavelength 325 nm
beam radius 1.1×10
-4
m


5.1 Single Laser Drawn Line
The process model (consisting of governing equations, domain, initial and boundary
conditions) for this case has been established in Chapter 2. The profile of the transient
intensity exposed on the resin is also described in Chapter 2. The graphs in Figure 29
demonstrate how the monomer conversion, temperature, radical concentration, and
initiator concentration at a particular spatial point (x,0,0) (any point on centerline of the
cured line at the surface) vary with time. The curing reaction occurs immediately upon
the laser exposure. The temperature increases rapidly due to the rapid exothermic
reaction (approximately 30
o
C increase during the first 0.1sec), and then decreases as the

68
reaction slows down and heat conduction plays a role. Due to the very fast reactions, the
radicals are rapidly exhausted and the monomer is consumed significantly in the first 0.1s
as well. The transient intensity caused by laser movement (Figure 29a) has induced
“Gaussian” radical concentration profile. In Figure 29, the laser directly exposes the
investigated point at t = 16ms, which gives the highest intensity (mol/m
2
-s) and leads to
most consumption of the initiator and most generation of the radicals. The initial delay is
due to the absence of irradiation. The initiator is consumed and the radicals are generated
during the very short irradiation period. In the subsequent dark period, no more initiator
is consumed to produce radicals.


(a)


69

(b)


(c)


70

(d)


(e)

Figure 29 Transients of (a) Intensity, (b) Initiator Concentration, (c) Radical
Concentration, (d) Monomer Conversion, and (e) Temperature at Point (x, 0, 0)

71
Figures 30 shows when the radicals are used up and the temperature returns to the
bath temperature, the sectional view of the monomer conversion and initiator
concentration profiles. Upon a single laser scan (with scan conditions listed in Table 6), a
maximum of 25% monomer can be converted, which occurs at the center of the single
line where the resin receives the most exposure and the most initiator is consumed. From
the conversion contour in Figure 30 (a), it can be expected that the cross section of the
cured line would be of bullet shape. Figure 30 (b) demonstrates the consumption status of
the initiator after a single laser scan and half an hour of post-curing in the bath. A longer
stay in the bath wouldn’t lead to significant change to the initiator distribution due to the
dark environment and extremely low diffusion.


(a)


72

(b)

Figure 30 Distribution of (a) Monomer Conversion and (b) Photoinitiator Concentration
upon a Single Laser Scan

Figures 31 and 32 demonstrate the evolution of monomer conversion along the width
at the resin surface (y axis in Figure 2) and along the depth centerline (z axis in Figure 2)
of the single line part, respectively. It can be seen that the reaction starts rapidly upon
irradiation (as shown in Figure 29a, irradiation starts at t = 0.01 sec and ends at t = 0.025
sec). In about 20 ms, the reaction slows down in the dark. At the center of the irradiation
(y = 0) where more radicals are generated during exposure, the dark reaction contributes
to about 10% conversion of monomer.

73

Figure 31 Monomer Conversion vs. Width at the Top Surface of the Single Line Part
(Plot Interval = 0.01 sec, except for t = 1860 sec)



Figure 32 Monomer Conversion vs. Depth along the Centerline of the Single Line Part
(Plot Interval = 0.01 sec, except for t = 1860 sec)


74
5.2 Overlapping Lines
The governing equations established for single laser drawn line in Chapter 2 are also
valid for the overlapping line case, while the domain and boundary conditions need to
vary accordingly to accommodate multiple lines. The full cross section of the drawn
overlapping lines (rather than half cross section investigated in single line case where
symmetry can be easily determined) is considered here. As for the single line case, the
3D problem is reduced to a 2D one due to, ignoring the line ends, the repetitive cure
behavior along the laser scanning direction (x axis). The jump time of the laser from the
end of one line to the start of a second line is negligible (within 1ms). With the same laser
moving speed as in single line case (see Table 6), it takes about 940 ms to draw a single 1
inch line or pass the same x location for a second time.
The laser irradiation is imposed line by line. The curing situation upon previous
drawing(s) is employed as the initial condition for the next scanning. Equation (52)
describes the transient intensity profile for the nth line.

( ) | | { }
8
2 2 2
0 0
10 196 . 1
) nm (
) / exp( / ) ) 1 n ( 0025 . 0 ( ) 1)0.94 (n ( 2 exp
×
− − − − + − − − − =
λ
p o s s
D z w h y t t V I I

(52)

where I
0
is the maximum intensity incident on the resin surface (W/m
2
), I is the intensity
incident on any point (y, z) in the resin (mol/m
2
–s), h
s
is the hatching space (i.e. line
spacing), 0.0025 is used to position the first drawn line in a 0.005 m wide domain, and
the time point at which the laser starts to draw the first line is taken as t = 0.

75
Figure 33 demonstrates for two overlapping lines with hatching space
0
5 . 1 w h
s
=
when the radicals generated are used up and the temperature returns to bath temperature,
the distributions of monomer conversion and initiator consumption. The monomer
conversion and temperature rise are found to be up to 32 % and 40
o
C, while for the single
line drawn with same speed and same laser power, they are 25 % and 30
o
C, respectively,
as mentioned earlier. The size of the same conversion outline (indicating part size) in
Figure 33 (a) is obviously larger comparing with that in Figure 30 (a). The initiator
consumption profile in Figure 33 (b) represents the two adjacent laser irradiations, which,
considering the limited diffusion in the crosslinked network, is basically Figure 30 (b)
added with a second laser drawing at a
s
h distance. However, in Figure 33 (a), the two
scans cannot be easily distinguished from each other because the reaction also occurs in
the joint area of the two scans. As the hatching space varies from large to small, the cured
lines generated can be separated, with an uneven bottom surface, or with a flat bottom
surface (as in this case). More discussions on superposition of adjacent scans and the
effect of hatching space on the cured part shape can be found in Jacobs (1992).


76

(a)



(b)

Figure 33 (a) Monomer Conversion (b) Initiator Concentration (mol/m
3
) Distributions
upon Two Overlapping Scans


77
Figure 34 demonstrates the evolution of monomer conversion along the width at the
resin surface of the two-overlapping-line part. It can be seen that the second line grows
rapidly (a noticeable increase in conversion is observed at the right of the first line) upon
irradiation (as discussed earlier, the laser starts to draw the second line at t = 0.94sec).
More monomer is converted in the first line region near the second line due to the extra
exposure from the second line drawing. In about 20 ms, the reaction in all regions slows
down in the dark. Similar to the single line case, the dark reaction contributes to about 10
% conversion of monomer at the locus where more irradiation is received.



Figure 34 Monomer Conversion vs. Width at the Top Surface of Two-Overlapping-Line
Part (Plot Interval = 0.01 sec, except for t = 1860 sec)


78
5.3 Stacked Single Lines
In this case, single lines are built layer by layer with each line drawn on the top of the
previous one(s). The simulation of the first layer is exactly the same as of the single line
case. Additional sub-domains for subsequent layers (or lines) are added layer by layer at
the top of the previous domain. The depth of each sub-domain is the layer thickness. Also
as in the single line case, the 3D problem is reduced to a 2D one due to, ignoring the edge
effect, the repetitive cure behavior along the laser scanning direction (x-axis).
The laser irradiation is imposed layer by layer. The curing situation upon previous
drawn layer(s) is employed as the initial condition for the next layer. The time delay
between drawing two neighboring layers is estimated to be 40sec, including resin
recoating and laser beam analyzing time. Equation (53) describes the transient intensity
profile for the nth layer.

( ) | | { }
8
2 2 2
0 0
10 196 . 1
) nm (
] / ) 1) (n ( exp[ / ) 1)40 (n ( 2 exp
×
− − − + − − − − =
λ
p T o s
D L z w y t t V I I
(53)

where L
T
is the layer thickness; the time point at which the laser starts to draw the first
layer is taken as t = 0.
Figure 35 demonstrates for two stacked lines with layer thickness
T
L = 4 mils when
the radicals generated are used up (except the small amount of radicals trapped in the first
layer due to high crosslinking) and the temperature returns to bath temperature, the
distributions of monomer conversion and initiator consumption. The monomer
conversion and temperature rise are found to be up to about 40 % and 35
o
C, respectively,

79
comparing with 25 % and 30
o
C for the single line drawn with same speed and same laser
power. The same conversion outline (indicating part size) in Figure 35 (a) is obviously
wider and deeper than that in Figure 30 (a) as well. The maximum monomer conversion
and initiator consumption obviously occur in the first layer near the joint boundary where
the resin receives the maximum exposure during the first laser scan and still receives
extra exposure during the second scan. Two layers can be easily distinguished from each
other as shown in Figure 35, which, however, doesn’t necessarily mean that these two
layers are separate physically. The cure at the joint area could be enough to hold two
layers together. The layer thickness affects how well the two adjacent layers are attached
to each other. Too large of a specified layer thickness causes two layers to partially join
together or even separate; too small a layer thickness makes two layers well connected,
however, it increases the part building time.



80

(a)


(b)

Figure 35 (a) Monomer Conversion (b) Initiator Concentration (mol/m
3
) Distributions
upon Two Stacked Scans


81
Figure 36 demonstrates the evolution of monomer conversion along the centerline in
depth of the two-layer-line part. It can be seen that the second line grows rapidly (a
noticeable increase in conversion is observed at the top of the first line) upon irradiation
(as discussed earlier, the laser starts to draw the second line at t = 40 sec). More monomer
is converted in the first layer region near the second layer due to the extra exposure from
the second line drawing. In about 20 ms, the reaction in all regions slows down in the
dark. Similar to the single line case, the dark reaction contributes to about 10 %
conversion of monomer at the locus where more irradiation is received. As shown in
Figure 36, the maximum +z = 1×10
-4
m (the top surface of the first layer is at z = 0),
verifying the second layer at the top of the first one is 4 mils (layer thickness) thick.



Figure 36 Monomer Conversion vs. Depth at the Centerline of Two-Layer-Line Part (Plot
Interval = 0.01 sec, except t = 1860 sec)

82
CHAPTER 6
MODEL VERIFICATION

To verify the process model, the single-line, overlapping-line, and stacked-line parts
have been fabricated in SLA and their dimensions measured to compare with the
simulation results. It’s found that a certain degree of cure (DOC) contour outlines the
built part within minimal error. For this reason, the SL cure process model established
and solved earlier can also be referred to as a “DOC threshold model” when used to
predict the fabricated part shape and dimensions. This DOC threshold model is valid not
only for single line parts, but for overlapping-line and stacked-line parts. While the
exposure threshold model predicts the cured part dimensions with up to 50% error, this
model has reduced the prediction error to 25 %.

6.1 DOC Threshold Model
The E4PETeA acrylate with 2 wt % DMPA initiator was used to grow single line
parts in SLA-250/50. The parts were elevated out of the resin vat 30 minutes after laser
scanning was finished, and drained on the platform for another 15 minutes. They were
cleaned for one minute in TPM (tri-propylene glycol monomethyl ether) and another
minute in water at room temperature using a Branson 5210 cleaning system in ultra-sonic
mode. The parts were then dried using compressed air and broken to expose their cross
sections, the image and dimensions of which were taken and measured by scanning
electron microscopy (SEM, Hitachi S800 FEG). The SEM measurements were calibrated

83
using a standard grid with known dimensions. Figure 37 shows a typical image of the
cross section of a single line part, which as expected is of parabolic shape.



Figure 37 SEM Image of Cross Section of a Single Line Part

The part building process has then been simulated and Figure 38 gives the DOC
contour of the built part which corresponds to half the image in Figure 37. For this single
line part, the outline is close to the 9% DOC contour. The DOC corresponding to this
contour line is defined as the critical DOC, above which the solid part can be formed
while below this DOC the resin has not been solidified enough and can be washed away
during the postprocessing step. In this sense, the SL cure process model developed in this
work can also be referred to as a DOC threshold model. Unlike the exposure threshold
model which only incorporates the exposure, the DOC threshold model takes the reaction
and transient intensity effects into account.

84


Figure 38 Degree of Cure Contour for Parts Built at V
s
= 1.071in/sec (with the measured
part contour shown in red)

About 20 single line parts were built and measured at the laser drawing speed V
s
=
1.071 in/sec (the scanning speed needs to be low enough for such a small part as a single
line to be formed as well as be strong enough to handle and measure). Their measured
outlines fall between the ~9% and ~10% DOC contours (Figure 38 shows a case where it
overlaps with the 9% DOC contour), which defines the range of the critical DOC. This
range can be applied to the simulated DOC profile to predict the cured part dimensions.
The critical DOC is lower than the gel point of the curing material (~14%, Appendix C),
which indicates that the cured polymer does not have to achieve an infinite molecular
weight to form a solid part.

85
The part building process has to be consistent in order for the fabricated parts to have
predictability. Obviously different part cleaning procedures could cause differences in the
part size or shape, therefore, it is important that parts be fabricated consistently, i.e.
following exactly the same building and postprocessing steps. On the other hand, it’s not
necessary to stick with the operating procedures described above. A different set of
postprocessing steps can be adopted, which could give a different value for the critical
DOC, but wouldn’t affect the prediction of the DOC threshold model. The part building
and cleaning procedure itself is not important; the consistency is important, i.e., the steps
have to be followed consistently once they are designed and adopted.

6.2 DOC Threshold Model Prediction
Three types of parts have been built and measured to verify the SL cure process
model or the prediction capability of the DOC threshold model: single line parts,
overlapping cured lines (nine lines), and stacked single lines (three layers).
6.2.1 Single Line Part Prediction
The single line parts have also been built at a laser scanning speed of 0.466 in/sec. To
verify the size of these parts is different from that of those built at V
s
= 1.071 in/sec, the
pairwise comparison (Neter et al., 1996) has been used:

0 :
0 :
'
' 0
≠ − =
= − =
i i a
i i
D H
D H
µ µ
µ µ
(54)


86
where
i
µ is the mean of the i th part dimension (here i = 1 for V
s
= 1.071 in/sec and i = 2
for V
s
= 0.466 in/sec), and D is the difference between the two means. The t test is
utilized to decide to conclude
0
H or
a
H , i.e., to conclude to support or to not support the
hypothesis that the part dimensions fabricated at these two scanning speeds are the same.
As shown in Table 7, for both part depth and width the magnitude of test statistics t* is
much greater than t (0.9995, 36) = 3.589. Therefore,
a
H is concluded with 99.9 % or
higher confidence. Parts built at V
s
= 0.466 in/sec have different (larger) size than those
built at V
s
= 1.071 in/sec.

Table 7 Dimensions of Single Line Parts Built at Two Laser-Scanning Speeds:
1. V
s
=1.071 in/sec; 2. V
s
=0.466 in/sec

Depth (µm) Full Width (µm)
scanning speed i i =1 i =2 i =1 i =2
Measurement 724 922 262 311
703 945 275 309
761 963 268 317
738 899 266 321
763 901 259 349
762 956 265 272
738 951 256 324
727 980 268 329
753 958 256 332
741 988 267 277
725 990 260 335
727 978 267 333
757 945 273 302
759 907 261 316
768 998 255 347
757 972 275 331
774 1008 283 334
737 964 260 345
938 328
982 334
average 745 957 265 322
Difference -212 -57
MSE 706.65 247.54
t* -24.56 -11.15

87
The building process at a laser moving speed V
s
= 0.466 in/sec has also been
simulated. The critical DOC range (9~10%) is applied to the DOC profile to predict the
cure depth and cure width. They are found to be in good agreement with the fabrication
results (Table 8). The prediction error is defined as follows:

% 100
value al experiment
value al experiment value simulated
error prediction ×

= (55)

Table 8 Single Line Part Prediction by DOC Threshold Model

depth (µm) full width (µm)
V
s

(in/sec)
experiment
(95% C.I.)
critical
DOC
Xc
model
prediction
error (%)
experiment
(95% C.I.)
critical
DOC
Xc
model
prediction
error (%)
1.071 745 ± 10 9% 660 -10 265 ±4 9% 260 -2
10% 560 -25 10% 236 -10
0.466 957 ± 15 9% 1050 10 322 ± 10 9% 322 0
10% 940 -2 10% 306 -5
*Xc model: DOC threshold model; C.I.: confidence interval

Part building at a higher scanning speed of 10.71 in/sec was attempted. It turns out,
however, that the resin is not cured enough to form a solid part. The simulation for this
faster writing speed process also demonstrates that the part cannot be formed. Figure 39
shows DOC contours for the 10.71 in/sec scanning speed. No DOC above 7% is observed
anywhere in the simulated vat, and thus since this is lower than the required 9% DOC no
solid polymer structure is predicted. This agreement also demonstrates and verifies the
model’s predictive ability.


88

Figure 39 Degree of Cure Contour for Parts Built at V
s
= 10.71 in/sec

6.2.2 Overlapping Line Part Prediction
The overlapping line parts have been built at V
s
= 17.967 in/sec and with a hatching
space h
s
= 0.5 mils. The shape of the cured lines is similar to that of the single cured line
due to the narrow line spacing and limited number (n = 9) of drawn lines. The size is
larger as shown in Table 9.





89
Table 9 Dimension Measurements of Overlapping Line Parts

Measurement Depth (µm) Full Width (µm)
1457 429
1406 435
1442 467
1449 463
1412 431
1370 446
1446 467
1004 494
1429 455
1454 519
1432 434
1042 563
1346 406
1431 513
Mean 1349 468
Standard Deviation 149.7 43.0
Coefficient of Variation 0.1 0.09
Half Length of 95% C.I. 86 25
Lower Limit 1278 448
Upper Limit 1420 489


This multiple-line part building process is simulated and the critical DOC is applied
to the DOC contour plot. The predicted dimensions are within 25 % of the experimental
measurements, as shown in Table 10.

Table 10 DOC Threshold Model Prediction for Overlapping Line Parts

depth (µm) full width (µm)
experiment
(95% C.I.)
critical
DOC
Xc
model
prediction
error (%)
experiment
(95% C.I.)
critical
DOC
Xc
model
prediction
error (%)
1349 ± 86 9% 1220 -10 468 ± 25 9% 364 -20
10% 1120 -20 10% 349 -25
*Xc model: DOC threshold model; C.I.: confidence interval



90
6.2.3 Stacked Line Part Prediction
The 3-layer stacked single line parts have been built at V
s
= 1.052 in/sec and with the
layer thickness L
T
= 4 mils. The cured lines have similar shape to the single cured line
with deeper and wider size. The dimension measurement results are shown in Table 11.

Table 11 Dimension Measurements of 3-Layer Stacked Line Parts

Measurement Depth (µm) Full Width (µm)
977 347
1146 341
951 356
1257 335
1178 319
1108 326
1073 376
995 362
904 338
941 361
932 353
1021 365
1018 358
1044 345
1040 326
890 342
945 367
Mean 1025 348
Standard Deviation 101.8 16.0
Coefficient of Variation 0.1 0.05
Half Length of 95% C. I. 52 8
Lower Limit 982 341
Upper Limit 1068 355


This layer-by-layer part building process is simulated and the critical DOC is applied
to the DOC contour plot. The predicted dimensions are within 25 % of the experimental
measurements, as shown in Table 12.


91
Table 12 DOC Threshold Model Prediction for 3-Layer Stacked Line Parts

depth (µm) full width (µm)
experiment
(95% C.I.)
critical
DOC
Xc
model
prediction
error (%)
experiment
(95% C.I.)
critical
DOC
Xc
model
prediction
error (%)
1025 ± 52 9% 1285 25 348 ± 8 9% 346 -1
10% 1150 10 10% 330 -5
*Xc model: DOC threshold model; C.I.: confidence interval

The prediction results of all three types of cured lines using DOC threshold model
have been demonstrated. The good agreement between experimental results and
simulation results (within 25 % error) validates the SL cure process model (i.e. DOC
threshold model). The critical DOC taken outside the 9~10% range leads to high
prediction error.

6.3 Exposure Threshold Model Prediction
The exposure threshold model has been widely used to guide the SL prototyping in
industry. As mentioned earlier, it assumes a critical exposure, E
c
, is necessary for a part
to be formed. The cure depth, C
d
, and linewidth, L
w
, (referring to the maximum depth and
width of the cured line, respectively) can be obtained from the following (Jacobs, 1992):

) ln(
max
c
P d
E
E
D C = (56)

P
d
w
D
C
w L
2
2
0
= (57)


92
where E
max
is the maximum exposure incident on the resin surface centerline during laser
scanning, w
0
is the laser beam radius, and D
P
is the penetration depth of the laser into the
resin at which the irradiance would be about 37 % of the surface irradiance. E
c
and D
P
are
regarded as the resin characteristic parameters and can be determined from
WINDOWPANE
TM
experiments (Jacobs, 1992).
6.3.1 Ec and Dp Determination
In SLA-250/50, the WINDOWPANE
TM
experiments were conducted to characterize
E
c
and D
P
of the model acrylate resin. The laser drawing speed was varied to achieve a
wide range of cure depth. According to Equation (56), the best linear fit was performed in
the semilog plot of cure depth, C
d
, versus maximum exposure, E
max
, as shown in Figure
40. E
c
and D
P
were thus found to be 7.22 mJ/cm
2
and 9.43 mils, respectively, for the
model resin: E4PETeA acrylate with 2 wt % DMPA.

SLA Working Curve for Model Acrylate Resin
y = 9.4271Ln(x) - 18.654
R
2
= 0.999
0
10
20
30
40
50
1 10 100 1000 10000
Exposure, Emax (mJ/cm2)
C
u
r
e

D
e
p
t
h
,

C
d

(
m
i
l
s
)
Cd(mils)
Low Cd
High Cd
Log. (Cd(mils))


Figure 40 Working Curve from WINDOWPANE
TM
Experimental Data

93
The “reverse WINDOWPANE
TM
” experiments (Jacobs, 1992) were performed to
verify the determined E
c
and D
P
values. The SL system uses these two values and sets the
desired cure depth for part building. The test parts were built, measured, and then
compared to the specified cure depth. It turned out that all cure depth values of the
produced windowpane parts are within 1 mil of the specified values, which guarantees
the correctness of E
c
and D
P
values determined above.
6.3.2 Single Line Part Prediction
Equations (56) and (57) have been used to calculate the cure depth and full width of
the cured line. For a single drawn line, the maximum exposure incident on the resin
surface can be expressed as follows (Jacobs, 1992):

s
L
V w
P
π
E
0
max
2
= (58)

Table 13 shows the calculated results for the case of single line parts. Obviously the
exposure threshold model has much larger prediction error than DOC threshold model
(within 25% error), with up to 50% for the depth and 27% for the width.

Table 13 Single Line Part Prediction Results Based on Exposure Threshold Model

depth (µm) full width (µm)
V
s

(in/sec)
experiment
(95% C.I.) Ec model
prediction
error (%)
experiment
(95% C.I.) Ec model
prediction
error (%)
1.071 745 ± 10 1118 50 265 ±4 336 27
0.466 957 ± 15 1310 37 322 ± 10 364 13
*Ec model: exposure threshold model; C.I.: confidence interval


94
The error of the exposure threshold model to predict the cure depth is quite large.
This is probably due to the inappropriate assumption of Gaussian laser beam intensity
distribution. Figure 41 is a picture of beam intensity profile taken by a digital camera. It
is apparent that the beam is not exactly a Gaussian distribution. The beam is not
symmetric, either, and time-varying. The picture below shows the beam width ratio in
two directions X/Y = 1.08.


Figure 41 Beam Intensity Profile of HeCd Laser in SLA-250/50

To obtain a better simulation of the beam profile, a top-hat distribution is introduced:

¹
´
¦
>
=
0
0 0
0
≤ ≤ 0
w r
w r I
I (59)

The Gaussian assumption is modified by equally combining with the top-hat
distribution as follows (i.e. taking the average of Gaussian and top-hat beam):


95

¦
¹
¦
´
¦
> −
≤ ≤ + −
=
0
2
0
2
0
0 0
2
0
2
0
) / 2 exp( 5 . 0
0 5 . 0 ) / 2 exp( 5 . 0
w r w r I
w r I w r I
I (60)

The integral of the surface irradiance over the exposed region must be equal to the
laser power P
L
incident on the resin surface. Therefore, the peak surface irradiance at
0 = r can be obtained as:

) /(
3
4
2
0 0
w P I
L
π = (61)

The integration of the surface irradiance over time gives the surface exposure:

¦
¦
¹
¦
¦
´
¦
> −
≤ − + −
=
0
2
0
2 0 0
0
2 2
0
0 2
0
2 0 0
) / 2 exp(
2
5 . 0
) / 2 exp(
2
5 . 0
) 0 , (
w y w y
V
I w
w y y w
V
I
w y
V
I w
y E
s
s s
π
π
(62)

Furthermore, the following equation gives the exposure into any depth z:

) / exp( ) 0 , ( ) , (
p
D z y E z y E − = (63)

Applying the critical exposure E
c
to the above equation, the cure depth can be
obtained by setting y = 0 and the linewidth obtained by setting z = 0.

96
As seen in Table 14, a better prediction can be achieved using the equally weighted
combination beam than using a pure Gaussian or a pure top-hat distribution. Varying the
weighting factors doesn’t improve the prediction accuracy because more weighted top-
hat distribution increases the width prediction error without reducing the depth error
significantly, and more weighted Gaussian distribution increases error in both depth and
width. It can also be observed that even the improved exposure threshold model
prediction results are not as good as those of the DOC threshold model using the raw
Gaussian beam. This also indicates a relatively low requirement of the DOC threshold
model for beam profile simulation accuracy comparing with the exposure threshold
model. On the other hand, a better simulation of beam profile might improve the
prediction of the DOC threshold model as well.

Table 14 Exposure Threshold Model Prediction Results using Modified Beam Profile

V
s
=1.071 in/sec depth(µm)
prediction
error (%)
full
width(µm)
prediction
error (%)
Experiment (95% C.I.) 745 ± 10 265 ±4
Ec model (G beam) 1118 50 336 27
Ec model (½G+½T beam) 1076 44 293 10
Ec model (T beam) 1064 43 220 -17
V
s
=0.466 in/sec depth(µm)
prediction
error (%)
full
width(µm)
prediction
error (%)
Experiment (95% C.I.) 957 ± 15 322 ± 10
Ec model (G beam) 1310 37 364 13
Ec model (½G+½T beam) 1276 33 325 1
Ec model (T beam) 1256 31 220 -32
* G: Gaussian distribution; T: top-hat distribution; ½G+ ½T: equally weighted
combination of Gaussian and top-hat distribution.


97
6.3.3 Overlapping Line Part Prediction
As mentioned earlier, the exposure threshold model basically derives the exposure
expression for a part building process, and then applies the critical exposure, which then
gives the equation that calculates the outline of the cured part. Suppose the laser starts
drawing at y = 0 (as shown in Figure 42), the exposure incident on any point of the resin
Q when the laser draws n overlapping lines with hatching space h
s
can be expressed as:

{ } ) / exp( ] / ) ) 1 ( ( 2 exp[ ] / ) ( 2 exp[ ) / 2 exp(
2
) , (
2
0
2 2
0
2 2
0
2
0
P s s
s
L
D z w h n y w h y w y
V w
P
z y E − − + − + + + − + − = "
π
(64)



Figure 42 Laser Movement when Drawing Overlapping Lines




98
Letting
c
E z y E = ) , ( , Equation (64) reduces to:

{ } ) / * exp( ] / ) ) 1 ( * ( 2 exp[ ] / ) * ( 2 exp[ ) / * 2 exp(
2
2
0
2 2
0
2 2
0
2
0
P s s
s
L
c
D z w h n y w h y w y
V w
P
E − − + − + + + − + − = "
π
(65)

where y*

and z* describe the outline of the cured part. When z*=0, two values will be
obtained for y*: +y
max
and -y
max
. The sum of their absolute values is the cured linewidth.
Setting y*=0, -h
s
, -2h
s
, …, -[rounded(n/2-1)] h
s
, the corresponding number of z*

values
can be obtained as local maxima, the global maximum of which is the maximum cured
depth of the part.
For the case of overlapping line parts with nine drawn lines n = 9 and hatching space
h
s
= 0.5 mils and laser scanning speed V
s
= 17.967 in/sec (i.e. the example tested in
section 6.2.2), + y
max
= 126 µm, - y
max
= - 227 µm, and thus the cured linewidth turns out
to be 353 µm, which is 25 % smaller than the measured result. The cure depth is found to
be 925 µm, which underestimates the experimental value by 30 %. As discussed in
section 6.2.2, the prediction errors for the part width and depth by the DOC threshold
model are within 25 % and 20 %, respectively.

99
6.3.4 Stacked Line Part Prediction
When the laser draws n stacked single lines with layer thickness L
T
, the exposure
incident on the resin at the bottom layer can be expressed as (let z = 0 at the top layer
surface):

| | ) / ) ) 1 ( ( exp( ) / ) ( ( exp ) / exp( ) / 2 exp(
2
) , (
2
0
2
0
P T P T P
s
L
D L n z D L z D z w y
V w
P
z y E − − − + + − − + − − = "
π
(66)

where z = C
d
when y = 0, letting
c
E z y E = ) , ( . For the case of 3-layer stacked single line
parts built at V
s
= 1.052 in/sec and with layer thickness L
T
= 4 mils (i.e. the example
tested in section 6.2.3), the cure depth is calculated to be 1,497 µm, which overestimates
the measured value by 46 %.
The exposure received at a point on the top of the nth layer is (Rosen, 2002):

| | ) / ) 1 ( exp( ) / 2 ( exp ) / exp( 1 ) / 2 exp(
2
) ) 1 ( , (
2
0
2
0
P T P T P T
s
L
T
D L n D L D L w y
V w
P
L n y E − − + + − + − + − = − "
π
(67)

Letting
c T
E L n y E = − ) ) 1 ( , ( , the maximum full width is calculated to be 336 µm at the
top of the first layer, 354 µm at the top of the second layer, and 362 µm at the top of the
third layer. This indicates that the width of the rib built by three stacked single vectors is
not uniform. Although this stair effect is not distinguishable in the experimental results,

100
the exposure threshold model gives a good prediction of the rib width. None of these
three values has deviated from the measured value by more than 5 %.
6.3.5 Comparison of DOC and Exposure Threshold Model
The predicted results by both the DOC threshold and exposure threshold models for
all three building types (single line, overlapping lines, and stacked lines) are summarized
in Table 15. The prediction error of the exposure threshold model is up to about 50 %,
while the DOC threshold model can predict within 25 % error. For the cases tested here,
the exposure threshold model appears to give better prediction of width than depth, which
is probably due to the inappropriate assumption of a Gaussian laser beam. It also appears
that the exposure threshold model is more sensitive to beam profile assumptions than is
the new DOC threshold model.

101
Table 15 Comparison of Prediction Results by Two Threshold Models
Single line part @ 1.071
in/sec
depth
(µm)
prediction error
(%)
full width
(µm)
prediction error
(%)
Experiment (95 % C.I.) 745 ± 10 265 ± 4
Xc model, Xc = 9 % 660 -10 260 -2
Xc = 10 % 560 -25 236 -10
Ec model 1118 50 336 30
Single line part @ 0.466
in/sec
depth
(µm)
prediction error
(%)
full width
(µm)
prediction error
(%)
Experiment (95 % C.I.) 957 ± 15 322 ± 10
Xc model, Xc = 9 % 1050 10 322 0
Xc = 10 % 940 -2 306 -5
Ec model 1310 40 364 15
Overlapping line part
depth
(µm)
prediction error
(%)
full width
(µm)
prediction error
(%)
Experiment (95 % C.I.) 1349 ± 86 468 ± 25
Xc model, Xc = 9 % 1220 -10 364 -20
Xc = 10 % 1120 -20 349 -25
Ec model 925 -30 353 -25
Stacked line part
depth
(µm)
prediction error
(%)
full width
(µm)
prediction error
(%)
Experiment (95 % C.I.) 1025 ± 52 348 ± 8
Xc model, Xc = 9 % 1285 25 346 -1
Xc = 10 % 1150 10 330 -5
Ec model 1497 50 336 -5

*Xc model: DOC threshold model; Ec model: exposure threshold model; C.I.: confidence interval


One might argue that the exposure threshold model is also very dependent on what
range of the working curve for the resin is used. Therefore, different regions of the
working curve were fit and these new Ec and Dp values were used for the exposure
threshold model prediction. For example, one option is to choose E
c
and D
P
in the higher
working range, as shown in Figure 43. E
c
and D
P
are fit to be 0.98 mJ/cm
2
and 6.08 mils,
respectively, in this range.


102
SLA Working Curve for Model Acrylate Resin
y = 6.0969Ln(x) + 0.0376
R
2
= 0.9955
0
10
20
30
40
50
1 10 100 1000 10000
Exposure, Emax (mJ/cm2)
C
u
r
e

D
e
p
t
h
,

C
d

(
m
i
l
s
)
Cd(mils)
Low Cd
High Cd
Log. (High Cd)

Figure 43 High Working Range for Model Acrylate Resin in SLA


However, the adoption of Ec and Dp from the higher working curve range of the resin
doesn’t improve the predictive ability of the exposure threshold model. As shown in
Table 16, the prediction error is up to 40% for the cured depth and 50% for the part
width.

Table 16 Exposure Threshold Model Prediction (high working range): 1. Single Line (1)
V
s
= 1.071 (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and 3. Stacked-line Parts

depth (µm) full width (µm)
Line Type
experiment
(95% C.I.) Ec model
prediction
error (%)
experiment
(95% C.I.) Ec model
prediction
error (%)
1 (1) 745 ± 10 1029 40 265 ±4 402 50
1 (2) 957 ± 15 1153 20 322 ± 10 425 30
2 1349 ± 86 905 -35 468 ± 25 430 -10
3 1025 ± 52 1322 30 348 ± 8 402 15


103
It should be mentioned that for the parts built and tested, the exposure doses fall in
the higher range of the working curve. This indicates that in order to predict these part
dimensions, the higher range of the curve should be used to determine Ec and Dp for the
exposure threshold model. However, as we can see above, the adoption of the higher
range of data gives similarly poor predictions.
6.3.6 Model Prediction using Ec and Dp Evaluated by a Different Protocol
In the previous sections where the exposure threshold model was used for part
dimension prediction, the 3D Systems WINDOWPANE procedure
10
was used to
fabricate and post-process the windowpane parts to evaluate Ec and Dp. However, the
washing procedure used in building parts or determining the critical DOC for our new
DOC threshold model is different from the 3D systems WINDOWPANE procedure.
Therefore, one might argue that the exposure threshold model would perform as well or
better than the DOC threshold model if identical washing procedure was used. Therefore,
in this study, an exposure threshold model was employed for part prediction in which Ec
and Dp were determined by fabricating the windowpane parts using the same post-
processing protocol as that for the DOC model development and regular part building
(see 6.1 “DOC Threshold Model”). E
c
and D
P
were found to be 5.24 mJ/cm
2
and 9.48
mils, respectively (Figure 44). Recall when the 3D System’s procedure was used, E
c
and
D
P
were found to be 7.22 mJ/cm
2
and 9.43 mils, respectively. As shown in Figure 44, two
different protocols generated curves with similar slopes. This indicates that the depth of
penetration, D
P
, of the resin (the slope of the curve) does not depend on the post-
processing procedure. This is expected because D
P
is a characteristic property of the

10
AccuMax
TM
ToolKit User Guide for use with SLA-190, 250, 350, 500, 3D Systems.

104
resin, which is related to the molar concentration and absorptivity of the initiator in the
resin, ) 3 . 2 /( 1 S ε D
P
= = 9.59 mils. Both D
P
values obtained by these two different
protocols are within 2% of this calculated value. The critical exposure, E
c
(the natural log
of E
c
is proportional to the intercept of the curve), however, is found to be affected by the
post-processing procedure significantly. This indicates that E
c
is not an inherent property
of the resin. For the same resin, it varies with the part processing procedure varying. E
c
is
an ambiguous concept, which leads to the poor predictive ability of the exposure
threshold model which takes both E
c
and D
P
as the resin characteristics.

Model Acrylate Resin Working
Curve_comparison of two protocols
y = 9.4806Ln(x) - 15.71
R
2
= 0.9659
y = 9.4271Ln(x) - 18.654
R
2
= 0.999
0
10
20
30
40
50
60
1 10 100 1000 10000
Exposure, Emax (mJ/cm2)
C
u
r
e

D
e
p
t
h
,

C
d

(
m
i
l
s
)
high Cd
Cd (mils)
low Cd
Cd (mils) (SOP)
low Cd (SOP)
high Cd (SOP)

Figure 44 Comparison of Working Curves Obtained by the 3D Systems WINDOWPANE
Procedure (labeled “SOP” in the figure) and by the Part Building Protocol


As shown in Figure 44, the correlation coefficient for the fitting is about 97%, while
it is more than 99% when the 3D Systems’ procedure is used. The reverse windowpane
parts were built using this set of Ec and Dp and it was found that the cured depth values

105
could be over 7 mils out of the nominal values (relatively, 15% different from the
specified values). Recall when the 3D Systems WINDOWPANE procedure was used to
determine Ec and Dp, all cure depth values of the produced windowpane parts were
found to be within 1 mil of the specified values (within 5% of nominal values). These
facts demonstrate that using the post-processing protocol is not an effective way to
characterize the resin working curve due probably to the non-uniform effect of this
protocol (draining, solvent washing, etc) on the part.
As shown in Table 17, the Ec and Dp characterization using the same post-processing
step as the regular part building worsens the prediction of the exposure threshold model
with the prediction error up to 60% (~10% less accurate than using the 3D Systems’
procedure). This is expected since the resin properties (Ec and Dp) were not characterized
properly by using this washing procedure as demonstrated by the poor predictive
performance of the reverse windowpane parts. In the 3D Systems’ procedure, rather than
draining or washing using solvent and ultrasonic equipment or compressed air drying,
after building the windowpane parts are placed with wet side (bottom side) down inside
the SLA chamber (the heat inside the chamber is nearly optimum for the draining
10
) with
the paper towel underneath. The paper towel strips are also placed with equal weights on
top of parts (ideal weight for drainage is 10-13g
10
) to drain the excess resin. The parts are
then put in the post-cure apparatus, PCA-250 (3D Systems), with dry side (top side)
down on a clean glass plate for further cure. It can be seen that the 3D Systems’
procedure is most likely much more effective at removing excess resin without damaging
the parts than the post-processing steps used for regular part building. The ultrasonic
vibration or air blowing used in our part building procedure may not clean the

106
windowpanes uniformly or they harm the cured dimensional uniformity. Such factors
could cause the correlation of working curve and model prediction to be worse than those
using Ec and Dp determined by the 3D Systems’ procedure.

Table 17 Exposure Threshold Model Prediction (protocol): 1. Single Line (1) V
s
=
1.071 in/sec (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and 3. Stacked-line Parts

depth (µm) full width (µm)
Line Type
experiment
(95% C.I.) Ec model
prediction
error (%)
experiment
(95% C.I.) Ec model
prediction
error (%)
1 (1) 745 ± 10 1201 60 265 ±4 347 30
1 (2) 957 ± 15 1395 45 322 ± 10 374 15
2 1349 ± 86 1007 -25 468 ± 25 367 -20
3 1025 ± 52 1581 55 348 ± 8 347 0


Again, one can argue about what range of the working curve to use. The adoption of
the high range of the new working curve (Figure 45) produces an Ec = 1.29 mJ/cm
2
and
Dp = 7.25 mils, but again this does not improve the model performance (Table 18).

Model Acrylate Resin Working
Curve_developed protocol (high range)
y = 7.2486Ln(x) - 1.8405
R
2
= 0.9439
0
10
20
30
40
50
60
1 10 100 1000 10000
Exposure (mJ/cm2)
C
d

(
m
i
l
s
)
high Cd
Cd (mils)
low Cd
Log. (high Cd)

Figure 45 Ec and Dp Determined in the High Range using Part Building Protocol

107

Table 18 Exposure Threshold Model Prediction (protocol; high working range): 1.
Single Line (1) V
s
= 1.071 in/sec (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and 3.
Stacked-line Parts

depth (µm) full width (µm)
Line Type
experiment
(95% C.I.) Ec model
prediction
error (%)
experiment
(95% C.I.) Ec model
prediction
error (%)
1 (1) 745 ± 10 1177 60 265 ±4 393 50
1 (2) 957 ± 15 1325 40 322 ± 10 417 30
2 1349 ± 86 1028 -25 468 ± 25 420 -10
3 1025 ± 52 1499 50 348 ± 8 393 15


Likewise, the adoption of the low range of the curve (Figure 46) produces an Ec =
7.90 mJ/cm
2
and Dp = 10.51 mils, and this does not improve the model predictions and in
fact even makes them worse (Table 19).

Model Acrylate Resin Working
Curve_developed protocol (low range)
y = 10.506Ln(x) - 21.724
R
2
= 0.9977
0
10
20
30
40
50
60
1 10 100 1000 10000
Exposure (mJ/cm2)
C
d

(
m
i
l
s
)
high Cd
Cd (mils)
low Cd
Log. (low Cd)

Figure 46 Ec and Dp Determined in the Low Range using Part Building Protocol



108
Table 19 Exposure Threshold Model Prediction (protocol; low working range): 1. Single
Line (1) V
s
= 1.071 in/sec (2) V
s
= 0.466 in/sec, 2. Overlapping-line, and 3. Stacked-line
Parts

depth (µm) full width (µm)
Line Type
experiment
(95% C.I.) Ec model
prediction
error (%)
experiment
(95% C.I.) Ec model
prediction
error (%)
1 (1) 745 ± 10 1222 65 265 ±4 332 25
1 (2) 957 ± 15 1436 50 322 ± 10 361 10
2 1349 ± 86 1007 -25 468 ± 25 349 -25
3 1025 ± 52 1630 60 348 ± 8 333 -5

It should be mentioned that for the parts built and tested, the exposure doses fall in
the higher range of the protocol working curve. This indicates that in order to predict
these part dimensions, the higher range of the curve (Figure 45) should be used to
determine Ec and Dp for the exposure threshold model. However, as we can see above,
the adoption of the higher range of data also gives similarly poor predictions.
6.4 Summary
The DOC threshold model was found to be more accurate at predicting the
dimensions of single line and multiple line and stacked line parts than the current
exposure threshold model. It was also found that evaluating Ec and Dp with the same
post-processing as used in regular part building, or adopting different ranges of the resin
working curve does not improve the predictive ability of the exposure threshold model,
and in fact generally, makes it worse. A more accurate beam profile approximation does
improve the predictions of the exposure threshold model, but it is still not as good as the
DOC threshold model (about 20% less accurate, see Table 15). Furthermore, the new
beam approximation makes the exposure threshold model more complex to utilize in SL.

109
CHAPTER 7
MODEL APPLICATIONS

As discussed earlier, given any part building condition, the DOC profile can be
simulated and the critical DOC applied to predict the cured part dimensions. This
capability of the SL cure process model (or critical DOC model) is not only a good
verification but a good application of the model.
This chapter demonstrates that the process model can also be used to investigate the
effects of material and process parameters on the SL performance, and identify factors
that affect the fabrication results significantly. The material and process optimization can
be performed for best performance, which also provides guidelines for SL material
development and process or laser improvement.
The SL performance properties that are investigated and addressed in this chapter are
the following: resolution, speed, maximum temperature rise in the resin bath, and
maximum DOC of the green part (i.e. the part that is formed right after the SL building
and has not been put in the post-cure apparatus, PCA, for further cure yet). Here SL
resolution is defined as the dimensions of the smallest parts that can be obtained
providing certain equipment and materials. The full width and maximum depth of a
single cured line part hence are referred to as the width and depth resolution, respectively.
Note the resolution decreases when the part size increases. The speed refers to the part
curing speed only; the reduction of the speed by the building delay between layers, part
draining and cleaning, etc., is not taken into account. The speed defined in the width (or

110
depth) direction is characterized by the time that is taken to obtain a single line part with
certain width (or depth).
The software Minitab (Minitab Inc.) has been employed for parameter effect
investigation and Evolver (Palisade Corporation) for parameter optimization.

7.1 Parameter Effect Investigation
Any of the parameters involved in the process model (as listed in Tables 1 and 4 in
Chapter 3) could affect the SL fabrication results. Among these parameters, the effects of
kinetic parameters are not investigated due to the complexity, their strong correlation
with one another, and the variety of ways people have employed to describe the
photopolymerization kinetics. The effect of CTE of the monomer is not tested either since
it strongly affects the kinetic values
cp
f and
ct
f and thus its factorial effect cannot be
tested without kinetic parameters also under investigation. This leaves 24 parameters to
screen to identify the important ones. The resolution III Plackett-Burman design with 32
runs (corresponding to 2
III
24-19
fractional factorial design, Neter et al. 1996) has been
chosen for this purpose. Table 20 lists these 24 factors and their two level values which
are determined based on SLA systems specifications (Rosen, 2002), polymer handbook
(Brandrup and Immergut, 1989), acrylate monomer descriptions in Sartomer
11
,
photoinitiator descriptions in Ciba
12
, polymer properties (Van Krevelen, 1990), Yaws’
chemical handbook (Yaws, 2003), as well as experience and knowledge about SLA
operations.

11
www.sartomer.com
12
www.cibasc.com

111
Table 20 Potential Sensitive Parameters and Their Level Values

Parameters Symbols Low Level (-1) High Level (+1) Units
laser scanning speed Vs 0.02 0.1 m/s
bath temperature Tb 301.15 308.15 K
laser power PL 0.024 0.1 W
beam radius wo 1.00E-04 2.00E-04 m
heat of polymerization deltH 3.45E+04 2.85E+05 J/mol
absorption coefficient of initiator ebx 20 60 m
3
/mol-m
quantum efficiency of initiation phi 0.1 0.6
Initiator wt% loading wt 1 5 wt%
chamber temperature Ta 296.15 303.15 K
heat convection coefficient hfc 0 4.18 W/m
2
-K
laser wavelength wL 325 354.7 nm
thermal conductivity cond 0.1 0.25 W/m-K
heat capacity (monomer) CpM 1500 3300 J/Kg-K
heat capacity (polymer) CpP 585 2500 J/Kg-K
diffusion coefficient (monomer) Dm 1.00E-18 1.00E-10 m
2
/s
diffusion coefficient (macroradical) DR 1.00E-20 1.00E-12 m
2
/s
diffusion coefficient (initiator) Ds 1.00E-18 1.00E-10 m
2
/s
CTE (polymer) alphaP 7.50E-05 1.23E-04 1/K
glass transition temp. (monomer) Tgm 173.15 223.15 K
glass transition temp. (polymer) Tgp 373.15 497.6 K
density (monomer) rouM 980 1128 Kg/m
3

density (polymer) rouP 1200 1800 Kg/m
3

molecular weight (monomer) MWm 0.198 2.156 Kg/mol
molecular weight (initiator) MWs 0.164 0.418 Kg/mol


For each run of the Plackett-Burman experiment, six responses are recorded including
width and depth resolution, curing speed defined in width and depth direction, maximum
DOC of the cured part, and maximum temperature rise during the curing process. Each
response is fitted versus these 24 factors. The absolute size of effects, P-values, effects
plot, normal plot, and results from stepwise selection have been inspected to screen out
the unimportant ones. Particularly, 10 factors appear to affect the depth resolution, in
which case a resolution III 1/64 fractional factorial experiment (2
III
10-6
) is conducted for
further screening; 12 factors could affect the maximum temperature rise, for which a

112
resolution III 1/256 fractional factorial design (2
III
12-8
) is employed for further screening.
Table 21 lists for each response the parameters that are identified as significant from the
final screening experiment.

Table 21 Significant Factors Identified from Screening Experiment

Responses
Factors
width
resolution
depth
resolution
speed
(width)
max
DOC
max
T rise
beam radius (wo) X X
monomer diffusion coefficient (Dm) X X X X
monomer glass transition temperature
(Tgm) X X
monomer molecular weight (MWm) X X X
initiator loading wt% (wt) X
initiator molecular weight (MWs) X
initiator absorptivity (ebx) X
quantum efficiency of initiation (phi) X
heat of polymerization (deltH) X
laser scanning speed (Vs) X
monomer heat capacity (CpM) X


The interactions among significant factors are investigated in follow-up 2
3
full
factorial experiments for each of the responses: width resolution, speed in width
direction, and maximum DOC, and a follow-up 2
4
full factorial design for maximum
temperature rise. A resolution V half-fraction 2
5-1
factorial design is used to study the
factor effects for depth resolution. No parameter seems significant for the speed
evaluated in depth direction, as shown in Table 21.

113
7.1.1 Sensitive Parameters for Width Resolution
Table 22 shows the full factorial design for the three parameters that have been
identified as important for width resolution. The response is evaluated and recorded for
each run. The other 21 parameters are fixed at values of the original process model (as
used in Chapter 6 for single-line part building).

Table 22 Full Factorial Design and Response Values (Width Resolution)

Run
Beam
radius (wo)
Monomer diffusion
coefficient (Dm)
Monomer glass transition
temperature (Tgm)
Width Resolution
(µm)
1 -1 -1 -1 250
2 1 -1 -1 494
3 -1 1 -1 270
4 1 1 -1 500
5 -1 -1 1 462
6 1 -1 1 890
7 -1 1 1 480
8 1 1 1 892


The initial inspection of the absolute size of the effects (Table 23) and normal plot
(Figure 47) demonstrate that the active effects are the main effects for laser beam radius
and monomer glass transition temperature, and the interaction effect between them. This
is confirmed by a formal test Lenth’s method (Wu and Hamada, 2002) which provides
quantitative evidence for effect significance (Table 23): these three effects can be
declared significant at the 0.001 level or even smaller, i.e., with at least 99.9%
confidence. Besides, the monomer diffusion coefficient can be claimed important with
almost 90% confidence.


114
Table 23 Estimated Factorial Effects and Lenth’s Test for Width Resolution

Term Effect Coef abs(tPSE,i) P-value
Constant 529.75
Beam radius (wo) 328.5 164.25 48.67 <0.001
Monomer diffusion coefficient (Dm) 11.5 5.75 1.70 0.101
Monomer glass transition temperature (Tgm) 302.5 151.25 44.81 <0.001
wo*Dm -7.5 -3.75 1.11 0.233
wo*Tgm 91.5 45.75 13.56 <0.001
Dm*Tgm -1.5 -0.75 0.22 >0.4
wo*Dm*Tgm -0.5 -0.25 0.07 >>0.4

Effect
P
e
r
c
e
n
t
350 300 250 200 150 100 50 0
99
95
90
80
70
60
50
40
30
20
10
5
1
Factor Name
A wo
B Dm
C Tgm
Effect Type
Not Significant
Significant
AC
C
A
Normal Probability Plot of the Effects
(response is width resolution, Alpha = .15)
Lenth's PSE = 6.75

Figure 47 Normal Plot for Width Resolution

The regression model is then reduced to retain only the effects that are identified as
active. As shown in Appendix D, the fit of the reduced model appears to be good and it
has great predictive ability (R
2
pred
= 99.63%). The statistical significance of the two main
effects and one interaction is confirmed as well: almost no risk is taken when claiming
these three effects are significant (P-value = 0 for all three effects).

115
The stepwise selection result also demonstrates the importance of these three effects.
Furthermore, it also indicates the less importance of two additional effects: main effect
for monomer diffusion coefficient and interaction between beam radius and monomer
diffusion coefficient (Appendix E). This is demonstrated by Lenth’s test results (Table
23) as well: the confidence to claim them significant is about 10% and 20%, respectively.
The effect significance of the parameters can also be observed from the main effects
and interaction plots (Figure 48). From the main effects plot, it appears that monomer
diffusion coefficient doesn’t have a noticeable effect on the width resolution, while with
beam radius or monomer Tg increasing, the width resolution decreases (part width
increases) significantly. It makes sense that the cured part gets wider when the incident
beam gets wider. An increase in monomer Tg reduces the free volume of the curing
system. Suppose the critical free volume for propagation and termination are constant,
then a decrease in free volume decreases the termination effect relative to propagation.
Thus, a bigger part is built. Further investigation later demonstrates that effect of
monomer Tg on the width resolution is not a simple linear relationship.


116
M
e
a
n

o
f

w
i
d
t
h

r
e
s
o
l
u
t
i
o
n

(
µ
m
)
1 -1
700
600
500
400
1 -1
1 -1
700
600
500
400
wo Dm
Tgm
Main Effects Plot (data means) for width resolution (µm)

(a) Main Effects Plot
wo
800
600
400
Dm
Tgm
1 -1
1 -1
800
600
400
1 -1
800
600
400
wo
-1
1
Dm
-1
1
Tgm
-1
1
Interaction Plot (data means) for width resolution (µm)

(b) Two-Factor Interaction Plot
Figure 48 Factorial Effects Plot for Width Resolution: (a) main effect (b) interaction

117
From the interaction plot, it can be seen that at high or low level of beam radius, there
is little effect of monomer diffusion coefficient on the width resolution, which leads to
unimportant interaction between these two factors as well. This relation is also illustrated
by the Dm-against-wo plot in which the two joined lines are almost overlapped with each
other. Furthermore, the large vertical distance of the overlapped lines in Dm-against-wo
plot and correspondingly the large space between two separate lines in wo-against-Dm
plot reflect an active effect of beam radius on the width resolution. Both wo-against-Tgm
and Tgm-against-wo plot are synergistic, which indicates a simple relation: whether beam
radius (and respectively, monomer Tg) is high or low, the width resolution decreases with
monomer Tg (and respectively, beam radius) increasing. On the other hand, the degree to
which the width resolution decreases when beam radius (and monomer Tg, respectively)
increases depends on high or low monomer Tg (and beam radius, respectively) being
adopted. Beam radius (and respectively, monomer Tg) affects the width resolution more
significantly at high monomer Tg (and respectively, high beam radius) than at low
monomer Tg (and respectively, low beam radius). The relation between monomer
diffusion coefficient and monomer Tg won’t be detailed here since it is similar to the
relation between monomer diffusion coefficient and beam radius.
In summary, laser beam radius and monomer glass transition temperature are two
sensitive parameters that affect the width resolution significantly. The width resolution
decreases (part width increases) with either of these two factors increasing. The
sensitivity of beam radius (and respectively, monomer Tg) depends on the level of
monomer Tg (and respectively, beam radius). The beam radius (or monomer Tg) is more
sensitive at high monomer Tg (or high beam radius).

118
7.1.2 Sensitive Parameters for Speed (Width Direction)
Similarly, a 2
3
full factorial design is employed for speed defined in width direction
to investigate the interaction among beam radius, monomer diffusion coefficient, and
monomer molecular weight. The speed is characterized by the time taken to cure a 200
µm wide single line part.
The inspection of effects size, normal plot, and effects plot as well as the results of
Lenth’s test, stepwise selection, and reduced model regression reaches a consistent
agreement on active effects. The active effects are identified to be main effects for beam
radius and monomer molecular weight and interaction between them. Table 24 shows the
Lenth’s test results, from which we can see that these three effects can be declared
significant with 99.5% confidence.

Table 24 Estimated Factorial Effects and Lenth’s Test for Speed (Width)

Term Effect Coef abs(tPSE,i) P-value
Constant 0.3994
Beam radius (wo) -0.7276 -0.3638 97.01 <<0.001
Monomer diffusion coefficient (Dm) 0.0061 0.0031 0.81 0.361
Monomer molecular weight (MWm) 0.0504 0.0252 6.72 0.005
wo*Dm -0.0064 -0.0032 0.85 0.340
wo*MWm -0.0511 -0.0256 6.81 0.005
Dm*MWm 0.0036 0.0018 0.48 >0.4
wo*Dm*MWm -0.0039 -0.0019 0.52 >0.4


Although the size and sign of the regression coefficients can reveal some information
on how the sensitive parameters affect the response, the effects plot provides more
information regarding to interactions between parameters in a graphical view. The main
effects plot below clearly demonstrates the speed increases significantly when beam

119
radius increases and increases slightly when monomer molecular weight decreases. Little
time is needed when the beam size is large. Further investigation later (Section 7.2.3)
demonstrates that the effect of beam radius on curing speed is quite nonlinear. The
decrease in monomer molecular weight leads to the increase in monomer molar
concentration, and thus the increase in curing speed.
From the interactions plot, beam radius is found to be a more sensitive parameter than
monomer molecular weight for width speed. The effect of beam radius is slightly more
significant at higher monomer molecular weight than at lower molecular weight. On the
other hand, monomer molecular weight seems more important at low beam radius than at
high beam radius.





120
M
e
a
n

o
f

t
i
m
e

(
s
e
c
)

(
w
i
d
t
h

d
i
r
e
c
t
i
o
n
)
1 -1
0.8
0.6
0.4
0.2
0.0
1 -1
1 -1
0.8
0.6
0.4
0.2
0.0
wo Dm
MWm
Main Effects Plot (data means) for time (sec) (width direction)

(a) Main Effects Plot
wo
0.8
0.4
0.0
1 -1
Dm
0.8
0.4
0.0
1 -1
0.8
0.4
0.0
MWm
1 -1
wo
-1
1
Dm
-1
1
MWm
-1
1
Interaction Plot (data means) for time (sec) (width direction)

(b) Two-Factor Interaction Plot
Figure 49 Factorial Effects Plot for Speed (Width): (a) main effect (b) interaction

121
7.1.3 Sensitive Parameters for DOC
Similarly, a 2
3
design full factorial experiment has been conducted for maximum
DOC to investigate the interaction among monomer diffusion coefficient, monomer
molecular weight, and monomer glass transition temperature.
The active effects are identified to be main effects for all three parameters mentioned
above as well as interaction between monomer molecular weight and monomer Tg. The
Lenth’s test (Table 25) shows the main effects for monomer Tg and molecular weight and
interaction between them can be declared significant with at least 99.9% confidence,
while the main effect for monomer diffusion coefficient is identified as active with 95.2%
confidence. Both Lenth’s test and stepwise selection also show that the interaction effect
between monomer diffusion coefficient and monomer Tg could be important as well.

Table 25 Estimated Factorial Effects and Lenth’s Test for Maximum DOC

Term Effect Coef abs(tPSE,i) P-value
Constant 0.23125
Monomer diffusion coefficient (Dm) -0.015 -0.0075 2.35 0.048
Monomer glass transition temperature (Tgm) 0.1455 0.07275 22.82 <0.001
Monomer molecular weight (MWm) 0.093 0.0465 14.59 <0.001
Dm*Tgm 0.008 0.004 1.25 0.189
Dm*MWm -0.0005 -0.00025 0.078 >>0.4
Tgm*MWm 0.094 0.047 14.75 <0.001
Dm*Tgm*MWm 0.0005 0.00025 0.078 >>0.4

The main effects plot below demonstrates that the DOC of the part increases with
monomer Tg or molecular weight increasing. As mentioned earlier, an increase in
monomer Tg decreases termination effect, thus increases DOC. High monomer molecular
weight (Low monomer concentration) decreases the curing speed, but could eventually
contribute to DOC considering diffusion limitation. Further investigation later shows a

122
slight curvature in monomer molecular weight and significant nonlinear behavior in
monomer Tg. High monomer diffusion increases monomer composition at the centerline
of the part, and leads to a slight decrease in DOC.
As shown in interaction plot above, the interaction between monomer Tg and
monomer molecular weight appears significant. Furthermore, molecular weight is much
more sensitive at high than at low monomer Tg, which again demonstrates characteristics
of diffusion-limited reaction. Similarly, monomer Tg is more sensitive at high that at low
monomer molecular weight.







123

M
e
a
n

o
f

m
a
x

D
O
C
1 -1
0.30
0.25
0.20
0.15
1 -1
1 -1
0.30
0.25
0.20
0.15
Dm Tgm
MWm
Main Effects Plot (data means) for max DOC

(a) Main Effects Plot

Dm
0.4
0.3
0.2
1 -1
Tgm
0.4
0.3
0.2
1 -1
0.4
0.3
0.2
MWm
1 -1
Dm
-1
1
Tgm
-1
1
MWm
-1
1
Interaction Plot (data means) for max DOC

(b) Two-Factor Interaction Plot
Figure 50 Factorial Effects Plot for Max DOC: (a) main effect (b) interaction

124
7.1.4 Sensitive Parameters for Temperature Rise
A full factorial (2
4
) design is adopted to investigate interactions between the four
parameters that are identified as important for temperature rise in the resin vat.
The Lenth’s test (Table 26) shows heat of polymerization and monomer molecular
weight are sensitive parameters. Main effects for these two factors and their interaction
effect can be declared important at a risk of <10%. The normal plot also illustrates these
three effects are significant. The main effect for scanning speed is claimed to be active
with >85% confidence. The significance of these four effects has been confirmed by the
regression of the reduced model (only four effects are included in the model) which turns
out to have a good fit quality. The stepwise regression selects nine additional factorial
effects such as main effect for monomer heat capacity, interactions between scanning
speed and monomer molecular weight, etc. These additional effects are identified by
Lenth’s method as less likely to be important (about 70% or even less).

Table 26 Estimated Effects and Lenth’s Test for Maximum Temperature Rise

Term Effect Coef abs(tPSE,i) P-value
Constant 19.2
Laser scanning speed (Vs) -18.12 -9.06 1.53 0.138
Heat of polymerization (deltH) 28.59 14.3 2.41 0.035
Monomer molecular weight (MWm) -27.77 -13.89 2.34 0.039
Monomer heat capacity (CpM) -8.95 -4.48 0.75 0.418
Vs*deltH -12.6 -6.3 1.06 0.27
Vs*MWm 12.12 6.06 1.02 0.29
Vs*CpM 2.01 1.01 0.17 >0.4
deltH*MWm -20.26 -10.13 1.71 0.099
deltH*CpM -5.83 -2.91 0.49 >0.4
MWm*CpM 5.58 2.79 0.47 >0.4
Vs*deltH*MWm 7.91 3.95 0.67 >0.4
Vs*deltH*CpM 0.34 0.17 0.029 >>0.4
Vs*MWm*CpM -0.22 -0.11 0.019 >>0.4
deltH*MWm*CpM 3.18 1.59 0.27 >0.4
Vs*deltH*MWm*CpM 1.07 0.54 0.09 >>0.4

125
The effects plot (Figure 51) demonstrates how the significant parameters affect the
temperature rise.
An increase in heat of polymerization (“deltH”) leads to more heat generated for the
same amount of reaction and thus larger temperature rise in the resin, as shown in the
main effects plot in Figure 51. The temperature rise decreases as laser scanning speed
(“Vs”), monomer molecular weight (“MWm”), or monomer heat capacity (“CpM”)
increases. This makes physical sense since higher laser moving speed deposits less
energy to the resin, and higher monomer molecular weight causes lower monomer
concentration. Both of them lead to less amount of reaction and thus generate less heat.
For the same amount of heat generated, higher monomer heat capacity leads to smaller
temperature rise. The effect of monomer heat capacity is less significant as observed.
The interactions plot shows that the temperature rise is more sensitive to the heat of
polymerization at low monomer molecular weight. Monomer molecular weight is a more
sensitive parameter at high heat of polymerization.
Other interaction effects shown in Figure 51 (b) are less significant and not discussed
here.



126
M
e
a
n

o
f

m
a
x

T

r
i
s
e

(
K
)
1 -1
30
20
10
1 -1
1 -1
30
20
10
1 -1
Vs deltH
MWm CpM
Main Effects Plot (data means) for max T rise (K)

(a) Main Effects Plot
Vs
1 -1 1 -1
50
25
0
50
25
0
deltH
MWm
50
25
0
1 -1
50
25
0
1 -1
CpM
Vs
-1
1
deltH
-1
1
MWm
-1
1
CpM
-1
1
Interaction Plot (data means) for max T rise (K)

(b) Two-Factor Interaction Plot
Figure 51 Factorial Effects Plot for Temperature Rise: (a) main effect (b) interaction

127
7.1.5 Sensitive Parameters for Depth Resolution
A resolution V five-factor half-fraction (2
V
5-1
) factorial design is adopted to
investigate interactions between the five parameters that are identified as important. For a
design of this resolution, all main effects are clear of other main effects and two-factor as
well as three-factor interactions, and confounded with four-factor interactions. All two-
factor interactions are clear of other two-factor interactions, and aliased with three-factor
interactions. Ignoring high order interactions, all main effects and two-factor interactions
can be clearly evaluated.

Table 27 Estimated Effects and Lenth’s Test for Depth Resolution

Term Effect Coef abs(tPSE,i) P-value
Constant 386
Initiator absorptivity (ebx) -204.2 -102.1 1.32 0.184
Quantum efficiency of initiation (phi) 434.5 217.2 2.81 0.023
Initiator wt% (wt) -297.7 -148.9 1.93 0.070
Monomer diffusion coefficient (Dm) -36.8 -18.4 0.24 >0.4
Initiator molecular weight (MWs) 136.3 68.1 0.88 0.354
ebx*phi -179.3 -89.6 1.16 0.234
ebx*wt 41 20.5 0.27 >0.4
ebx*Dm -108 -54 0.70 0.445
ebx*MWs -10 -5 0.065 >>0.4
phi*wt -303.2 -151.6 1.96 0.067
phi*Dm -37.8 -18.9 0.25 >0.4
phi*MWs 186.8 93.4 1.21 0.216
wt*Dm -7 -3.5 0.045 >>0.4
wt*MWs 3.5 1.8 0.023 >>0.4
Dm*MWs 98 49 0.63 0.478

As shown in Table 27, the Lenth’s test illustrates that the quantum efficiency of
initiation and initiator loading are sensitive parameters. Main effects for these two factors
and their interaction effect can be declared important at a risk of less than 10%. The
normal plot also illustrates these three effects are significant. The stepwise selection

128
demonstrates six additional important factorial effects such as main effects for initiator
absorptivity and initiator molecular weight and interactions between quantum efficiency
and initiator molecular weight, etc. These additional effects are identified by Lenth’s
method as less likely to be important (about 80% or even less, see items in red and not
bold in Table 27).
Either an increase in initiator wt% loading (“wt”) or initiator absorption coefficient
(“ebx”) leads to a smaller beam penetration depth and thus an increase in depth
resolution, as shown in Figure 52 (a). Higher quantum efficiency of initiation (“phi”)
generates more radicals to grow a bigger part. The effects of monomer diffusion
coefficient (“Dm”) and initiator molecular weight (“MWs”) are much less significant.
The interactions plot in Figure 52 (b) shows that quantum efficiency (“phi”) plays a
bigger role at low absorption coefficient (“ebx”) and low initiator loading, while
absorption coefficient and initiator loading are more sensitive at high quantum efficiency.
The interaction between “ebx” (or “MWs”) and “Dm” is antagonistic. At low absorption
coefficient (or high initiator molecular weight), the cure depth increases slightly with
monomer diffusion increasing. This reveals another side of monomer diffusion effect:
faster diffusion facilitates the movement and reaction of the species thus to form a bigger
part. The phi-against-MWs plot is antagonistic as well. At low quantum efficiency, the
cure depth decreases when the initiator molecular weight increases (beam penetration
depth increases). This indicates that an increase in penetration depth does not necessarily
lead to an increase in the cure depth. There might not be enough radicals (in this case,
quantum efficiency is low) to cure the part tip enough. Other interaction effects shown in
Figure 52 (b) are minor and are not discussed here.

129
M
e
a
n

o
f

d
e
p
t
h

r
e
s
o
l
u
t
i
o
n

(
µ
m
)
1 -1
600
500
400
300
200
1 -1 1 -1
1 -1
600
500
400
300
200
1 -1
ebx phi wt
Dm MWs
Main Effects Plot (data means) for depth resolution (µm)

(a) Main Effects Plot
ebx
900
600
300
1 -1 1 -1
phi
900
600
300
900
600
300
wt
Dm
900
600
300
1 -1
900
600
300
1 -1
MWs
1 -1
ebx
-1
1
phi
-1
1
wt
-1
1
Dm
-1
1
MWs
-1
1
Interaction Plot (data means) for depth resolution (µm)

(b) Two-Factor Interaction Plot
Figure 52 Factorial Effects Plot for Depth Resolution: (a) main effect (b) interaction

130
Table 21 from the screening experiment is revised according to the follow-up effect
significance investigation above for each response. In Table 28, “XX” denotes more
significant parameters, while “X” denotes relatively less important ones.

Table 28 Significant Factors for Investigated Responses

Responses
Factors
width
resolution
depth
resolution
speed
(width)
max
DOC
max
T rise
beam radius (wo) XX XX
monomer diffusion coefficient (Dm) X XX
monomer glass transition temperature (Tgm) XX XX
monomer molecular weight (MWm) XX XX XX
initiator loading wt% (wt) XX
initiator molecular weight (MWs) X
initiator absorptivity (ebx) X
quantum efficiency of initiation (phi) XX
heat of polymerization (deltH) XX
laser scanning speed (Vs) X
monomer heat capacity (CpM)


131
7.2 Resolution and Speed Prediction by Regression Model
As is well known, SL curing resolution and speed can be influenced by a lot of
factors such as material properties, reaction kinetics, and process or laser parameters. To
obtain an explicit function of resolution or speed in terms of these properties is almost
impossible. A useful assumption is made that the function is bilinear in which the linear
terms model the effect of influential factors and bilinear terms model the important
interaction effects. In case of the existence of curvature effect, the square of the factor
actual value is used to represent the quadratic effect. The response is then fitted versus
the involved parameters and combinations. The predictive ability of regression models is
verified.
7.2.1 Regression Prediction Model for Depth Resolution
Recall that the active effects for depth resolution are main effects for quantum
efficiency of initiation and initiator wt% loading and their interaction effect. The
regression model based on these three effects, however, doesn’t have a good fit or good
predictive ability. The other important effects identified by stepwise selection are then
included into the regression model as well. The regression equation of this revised model
as well as its good fit quality and predictive ability is demonstrated in Appendix F.
To verify this regression model, three simulations have been conducted using the SL
cure process model. The conditions for the simulations are shown in Table 29, and the
results and comparison with regression model predicted results are shown in Table 30.
The prediction error of the regression model is found to be within 15 %.


132
Table 29 Simulation Conditions to Test Predictive Ability of Regression Models

Condition I II III units
beam radius (wo) 1.10E-04 1.50E-04 2.00E-04 m
monomer diffusion coefficient (Dm) 1.00E-14 1.00E-14 1.00E-14 m
2
/s
monomer glass transition
temperature (Tgm) 208.74 180 210 K
monomer molecular weight (MWm) 0.53 1 1.5 Kg/mol
initiator loading wt% (wt) 2 1 5 wt%
initiator molecular weight (MWs) 0.26 0.2 0.4 Kg/mol
initiator absorptivity (ebx) 19.87 40 60 m
3
/mol-m
quantum efficiency of initiation (phi) 0.6 0.3 0.1

Table 30 Depth Resolution Predicted by Regression Model

Condition I II III
simulation results (µm) 810 525 155
predicted results (µm) 872 474 175
prediction error (%) 7.7 -9.8 12.7

7.2.2 Regression Prediction Model for Width Resolution
The prediction error of the fitted bilinear model versus the important effects (main
effects for beam radius and monomer Tg and interaction between them) is up to 60%.
The addition of the two less active effects (main effect for monomer diffusion coefficient
and its interaction with beam radius) makes the residual plot versus the predictor
monomer diffusion coefficient appear like a parallel band centered about zero, as it
should be, but doesn’t improve the model predictive ability significantly. Further
investigation reveals the possibility of curvature presence in monomer diffusion
coefficient and monomer Tg predictors (Figure 53). A three-factor-mixed-level (two
levels for beam radius; three levels for monomer diffusion coefficient and monomer Tg)
full factorial design is then adopted in order to capture the curvature effect.

133

200
250
300
350
400
450
500
1 2 3
factor level
w
i
d
t
h

r
e
s
o
l
u
t
i
o
n

(
µ
m
)
wo
Dm
Tgm

Figure 53 Curvature in Factors for Width Resolution

The effects plot below demonstrates significant linear effect in beam radius, less
significant and almost linear effect in monomer diffusion coefficient, and significant
curvature effect in monomer glass transition temperature.


134
M
e
a
n

o
f

w
i
d
t
h

r
e
s
o
l
u
t
i
o
n

(
µ
m
)
2 1
700
600
500
400
300
3 2 1
3 2 1
700
600
500
400
300
wo Dm
Tgm
Main Effects Plot (data means) for width resolution (µm)


(a) Main Effects Plot

wo
800
600
400
Dm
Tgm
3 2 1
3 2 1
800
600
400
2 1
800
600
400
wo
1
2
Dm
3
1
2
Tgm
3
1
2
Interaction Plot (data means) for width resolution (µm)


(b) Interaction Plot

Figure 54 Curvature Effect for Width Resolution: (a) main effect (b) interaction

135
As shown in Appendix G, regression analysis has been performed based on stepwise
selection results, using Tgm to represent the linear effect of monomer Tg, and Tgm^2 to
represent the quadratic effect of monomer Tg, and wo*Tgm to represent the interaction
between beam radius and monomer Tg. The regression model appears to have a good fit
and good predictive ability. The verification is shown in Table 31.

Table 31 Width Resolution Predicted by Regression Model

Condition I II III
simulation results (µm) 281 342 488
predicted results (µm) 332 329 613
prediction error (%) 18.3 -4.0 25.6

7.2.3 Regression Prediction Model for Speed (Width Direction)
Curvature has been found in beam radius and in monomer molecular weight (Figure
55) for this response. Further investigation demonstrates that the effect of beam radius on
the curing speed is quite nonlinear (Figure 56).

0
0.2
0.4
0.6
0.8
1
1 2 3
factor level
t
i
m
e

(
s
e
c
)

(
w
i
d
t
h
)
wo
MWm

Figure 55 Curvature in Factors for Speed (Width)

136

0
0.2
0.4
0.6
0.8
1
50 100 150 200 250
beam radius (µm)
t
i
m
e

(
s
e
c
)

(
w
i
d
t
h
)

Figure 56 Nonlinear Behavior of Beam Radius for Speed (Width)

The speed increases dramatically when spot size (beam diameter) increases slightly
above 200 µm and then almost keeps constant when the spot size increases further. This
makes sense recalling the speed is characterized by the time taken to cure 200 µm wide
part, 200 µm taken as the low level of spot size, and the laser assumed to be Gaussian.
This, however, brings complexity to the regression analysis. It’s inappropriate to find a
single model to capture all the behavior. To go around this issue, the time consumed to
cure 100 µm wide part is used to characterize the curing speed (200 µm wide part is
critical here because it overlaps the low level value of spot size). A two-factor-three-level
3
2
design is performed to capture the curvature effect. The effects plot is shown in Figure
57.
As shown in Figure 57, the curing time increases as laser beam radius increases
because for the same power and bigger beam size, a lower intensity irradiates the resin
inside the 100µm diameter spot.


137
M
e
a
n

o
f

t
i
m
e

(
s
)

(
w
i
d
t
h
)
3 2 1
0.036
0.032
0.028
0.024
0.020
3 2 1
wo MWm
Main Effects Plot (data means) for time (s) (width)

(a) Main Effects Plot
wo
MWm
3 2 1
0.036
0.032
0.028
0.024
0.020
3 2 1
0.036
0.032
0.028
0.024
0.020
wo
3
1
2
MWm
3
1
2
Interaction Plot (data means) for time (s) (width)

(b) Interaction Plot
Figure 57 Factors Curvature for Speed (Width): (a) main effect (b) interaction
(Level “2” value of beam radius = 110µm)


138
A regression analysis has been performed (as shown in Appendix H) based on
stepwise selection results. The regression model appears to have a good fit and good
predictive ability. The verification is shown in Table 32.

Table 32 Curing Time (Width Direction) Predicted by Regression Model

Condition I II III
simulation results (sec) 0.0185 0.0294 0.0368
predicted results (sec) 0.0185 0.0235 0.0336
prediction error (%) 0 -20.1 -8.7

7.2.4 Regression Prediction Model for Maximum DOC
Similarly, curvature has been found in monomer glass transition temperature and
slightly in monomer molecular weight (Figure 58). A three-factor-three-level 3
3
design
has been performed to detect and capture the curvature effect. The effects plot is shown
in Figure 59.

0.14
0.16
0.18
0.2
0.22
1 2 3
factor level
m
a
x
i
m
u
m

D
O
C
Dm
Tgm
MWm

Figure 58 Curvature in Factors for Maximum DOC

139
M
e
a
n

o
f

m
a
x

D
O
C
3 2 1
0.35
0.30
0.25
0.20
0.15
3 2 1
3 2 1
0.35
0.30
0.25
0.20
0.15
Dm Tgm
MWm
Main Effects Plot (data means) for max DOC

(a) Main Effects Plot
Dm
0.4
0.3
0.2
Tgm
MWm
3 2 1
3 2 1
0.4
0.3
0.2
3 2 1
0.4
0.3
0.2
Dm
3
1
2
Tgm
3
1
2
MWm
3
1
2
Interaction Plot (data means) for max DOC

(b) Interaction Plot
Figure 59 Factors Curvature for Maximum DOC (a) main effect (b) interaction

140
As shown in Appendix I, regression analysis has been performed based on stepwise
selection results. The regression model appears to have a good fit and good predictive
ability. The verification is shown in Table 33.

Table 33 Maximum DOC Predicted by Regression Model

Condition I II III
simulation results 0.247 0.137 0.193
predicted results 0.180 0.125 0.198
prediction error (%) -27.1 -8.8 2.6

7.2.5 Regression Prediction Model for Maximum Temperature Rise
Curvature has been found more significant in monomer molecular weight, laser
scanning speed, and heat of polymerization, and negligible in monomer heat capacity
(Figure 60). A three-factor-three-level 3
3
design has been conducted to detect and capture
the curvature effect. The effects plot is shown in Figure 61.

0
20
40
60
80
100
120
1 2 3
factor level
m
a
x

T

r
i
s
e

(
K
)
Vs
deltH
MWm
CpM

Figure 60 Curvature in Factors for Maximum Temperature Rise

141
M
e
a
n

o
f

m
a
x

T

r
i
s
e

(
K
)
3 2 1
40
30
20
10
0
3 2 1
3 2 1
40
30
20
10
0
Vs deltH
MWm
Main Effects Plot (data means) for max T rise (K)

(a) Main Effects Plot
Vs
50
25
0
3 2 1
deltH
50
25
0
3 2 1
50
25
0
MWm
3 2 1
Vs
3
1
2
deltH
3
1
2
MWm
3
1
2
Interaction Plot (data means) for max T rise (K)

(b) Interaction Plot
Figure 61 Factors Curvature for Max Temp Rise (a) main effect (b) interaction

142
Regression analysis is performed referring to stepwise selection results. It turns out
that monomer heat capacity needs to be considered in order for the model to have a good
fit. As shown in Appendix J, the regression model appears to have a good fit and good
predictive ability. The verification is shown in Table 34. Table 35 lists the conditions
used to test the regression model.

Table 34 Maximum Temperature Rise Predicted by Regression Model

Condition I II III
experimental results 43.7 1.7 0.9
predicted results 55.8 2.0 1.0
prediction error (%) 27.7 15.9 23.3


Table 35 Conditions used for Test of Temperature Rise Regression Model

Condition I II III units
MWm 0.528 1.5 1 Kg/mol
deltH 2.85E+05 8.00E+04 3.45E+04 J/mol
Vs 0.0272 0.06 0.1 m/s
CpM 1500 2000 1850 J/Kg-K


143
7.3 Parameter Optimization
For a given stereolithography apparatus and given photosensitive material,
Chockalingam and coworkers (2003) performed optimization on operation parameters
such as layer thickness, hatch spacing, hatch style, hatch over cure, and fill cure depth in
order to obtain parts with desired dimensions. In this work, the optimization performed
for part dimensions is more focused on the smallest feature that SL can build given any
material or apparatus available.
Based on the regression prediction models established in the previous section,
response optimization can be performed to find the highest resolution or speed or
maximum DOC as well as their corresponding parameter settings. Equation (68) shows
the objective function to be minimized (e.g. for resolution and speed) or maximized (e.g.
for DOC) subject to the constraints Equation (69). The variation bounds for design
variables (i.e. significant factors) in Equation (69) are listed in Table 36. Evolver
(Palisade) has been used to solve the optimization problem. It is an add-in for Microsoft
Excel that uses genetic algorithms to perform optimization.
The optimization problem is formulated as follows:

min or max ) , , , (
3 2 1
" x x x f y = (68)

s.t.
i i i
b x a ≤ ≤ (69)




144
Table 36 Parameter Range Used for Response Optimization

Factors Low Level High Level Units
beam radius (wo) 1.00E-04 2.00E-04 m
monomer diffusion coefficient (Dm) 1.00E-14 1.00E-14 m
2
/s
monomer glass transition temperature (Tgm) 1.00E-12 1.00E-12 K
monomer molecular weight (MWm) 0.198 2.156 Kg/mol
initiator loading wt% (wt) 1 5 wt%
initiator molecular weight (MWs) 0.164 0.418 Kg/mol
initiator absorptivity (ebx) 20 60 m
3
/mol-m
quantum efficiency of initiation (phi) 0.1 0.6 N/A
heat of polymerization (deltH) 3.45E+04 2.85E+05 J/mol
laser scanning speed (Vs) 0.02 0.1 m/s
monomer heat capacity (CpM) 1500 3300 J/Kg-K


Table 37 lists the optimization results for each response. It appears that a small beam
contributes to improvement in both width resolution and curing speed. Relatively lower
monomer glass transition temperature would be preferred for the concern of resolution,
which however would also decrease the DOC. Low monomer diffusion coefficient
benefits width resolution and DOC, while for depth resolution easier diffusion of
monomer is preferable. The quantum efficiency of initiation, initiator wt% loading,
initiator absorption coefficient, and initiator molecular weight are parameters that only
affect depth resolution. It can be seen that high quantum efficiency and initiator loading
are not necessarily good for SL. High absorptivity and low molecular weight of initiator
are preferred for the depth resolution due to the correspondingly smaller penetration
depth of the laser beam. Monomer molecular weight is another parameter to adjust in
need of higher DOC or curing speed. The temperature rise in the resin due to the heat
generated by curing can be reduced to negligible value by adjusting the laser scanning
speed and adopting material with smaller heat of polymerization and higher heat
capacity. For the above responses of interest, most of the important factors are material

145
property parameters. Process parameters such as laser power, bath temperature, and heat
convection at resin surface are comparatively insignificant.
There might be other criteria to evaluate the SL process. For example, if we have a 60
µm depth resolution, the layer thickness must be half or less of standard layer thickness 4
mils (100 µm) in order for neighboring layers to attach together. This would eventually
decrease the part fabrication speed (not curing speed).

Table 37 Evolver Optimization Results for Investigated Responses

Responses
Optimal Conditions
Factors
Width
Resolution
Depth
Resolution
Curing
Time
(width)
Max
DOC
Max T
rise Units
beam radius (wo) 1.00E-04 1.00E-04 m
monomer diffusion
coefficient (Dm) 1.00E-14 1.00E-12 1.00E-14 m
2
/s
monomer glass transition
temperature (Tgm) 186.17 223.15 K
monomer molecular
weight (MWm) 1.05 1.50 1.22 Kg/mol
initiator loading wt% (wt) 1 wt%
initiator molecular weight
(MWs) 0.164 Kg/mol
initiator absorptivity (ebx) 60 m
3
/mol-m
quantum efficiency of
initiation (phi) 0.1 N/A
heat of polymerization
(deltH) 3.45E4 J/mol
laser scanning speed (Vs) 0.1 m/s
monomer heat capacity
(CpM) 3300 J/Kg-K
Optimal Responses 215 µm 60 µm 18 ms 34% 0 K


Table 37 demonstrates the smallest part that can be obtained is about 215 µm wide
(slightly larger than the beam diameter) and 60 µm deep. This is evaluated based on the
kinetic parameters of the model tetraacrylate material; these values could be different for
resins with different kinetics. Recall in the effect significance investigation, kinetic

146
parameters were not tested due to the complexity and the variety of ways people have
employed to describe the photopolymerization kinetics.
The effect analysis of parameters has provided guidelines to perform cost-effective
trials to improve SL fabrication performance.

7.4 Parameter Analysis using Exposure Threshold Model
As described earlier, the exposure threshold model only predicts the cured line depth
(depth resolution) and width (width resolution). It doesn’t provide information for part
DOC or curing speed as the DOC threshold model. In this section, the effect of
parameters on cure depth and line width will be investigated and sensitive parameters be
identified using the exposure threshold model. Regression equations will be established
to predict the cured part dimension and parameter optimization be performed to obtain
the best resolution.
7.4.1 Parameter Significance Investigation
The parameters involved in the exposure threshold model are shown in Table 38 and
a 2
5
full factorial design with 32 runs has been conducted to study the factor effects. For
each run, the cure depth and line width of a single line part are calculated using Equations
(55) and (56) and recorded.





147
Table 38 Parameters in Exposure Threshold Model and Their Level Values

PARAMETER Symbol low level (-1) High level (+1) Units
critical exposure Ec 5 20 mJ/cm
2

depth of penetration Dp 5 10 mils
laser power PL 24 100 mW
beam radius wo 0.1 0.2 mm
laser scanning speed Vs 20 100 mm/s

The analysis of the design illustrates that all five parameters listed in Table 38 are
sensitive parameters for cure depth, while four of the parameters (depth of penetration
excluded) are significant factors for cured line width. The main effects plots for cure
depth and line width are shown in Figures 62 and 63, respectively.
From Figure 62, it can be seen that the cure depth increases as laser power and depth
of penetration increase and decreases as critical exposure, laser scanning speed, and beam
radius increase. From Figure 63, it can be seen that the line width increases as laser
power and beam radius increase and decreases as critical exposure and laser scanning
speed decrease. The line width is not subject to the change of depth of penetration. These
observations make physical sense and are consistent with Equations (56) and (57), only
expressing the relations in a graphical way. Note that Equation (57) can be rewritten as:

) ln( 2
max
0
c
w
E
E
w L = (70)

Equation (70) obviously shows that the line width is not dependent on the depth of
penetration.


148
M
e
a
n

o
f

C
d

(
m
i
l
s
)
1 -1
40
35
30
25
20
1 -1 1 -1
1 -1
40
35
30
25
20
1 -1
Ec Dp PL
wo Vs
Main Effects Plot (data means) for Cd (mils)

Figure 62 Main Effects Plot for Cure Depth


M
e
a
n

o
f

L
w

(
m
m
)
1 -1
0.54
0.48
0.42
0.36
0.30
1 -1 1 -1
1 -1
0.54
0.48
0.42
0.36
0.30
1 -1
Ec Dp PL
wo Vs
Main Effects Plot (data means) for Lw (mm)

Figure 63 Main Effects Plot for Line Width

149
Recall the parameter effects on depth resolution (cure depth) investigated using DOC
threshold model. The significant factors are initiator wt% loading and quantum efficiency
of initiation and less significant ones are initiator molecular weight and initiator
absorptivity. The beam radius, laser power, and laser scanning speed are insignificant
comparing to these parameters. As discussed earlier, the depth of penetration is
dependent on absorptivity, initiator molar concentration (molecular weight and wt%
loading of the initiator), while the critical exposure by definition appears dependent on
quantum efficiency of initiation. From this point of view, the parameter significance is
consistent by either exposure or DOC threshold model.
Recall the parameter effects on width resolution (line width) investigated using DOC
threshold model. The significant factors are beam radius and monomer glass transition
and a less significant one is monomer diffusion coefficient. Both models tell that the
beam radius is a significant factor for line width and depth of penetration is an
insignificant one. By the DOC threshold model, the material parameters seem more
important than process parameters such as laser power and scanning speed. However, not
much information regarding to material property effect can be extracted from exposure
threshold model.

150
7.4.2 Parameter Optimization
The optimization is performed based on the exposure threshold model Equations (56)
and (57). The optimization problem is formulated similarly to Equations (68) and (69).
The objective here is to minimize the cure depth and line width. The variation bounds of
the parameters are as shown in Table 38. The optimization results by Evolver are
demonstrated in Table 39.
Table 39 Evolver Optimization Results using Exposure Threshold Model

Factors
Conditions
Responses
critical
exposure
(Ec)
depth of
penetration
(Dp)
laser
power
(PL)
beam
radius
(wo)
scanning
speed
(Vs)
Optimal
Response units
cure depth (Cd) 20 5 24 0.2 100 7.8 mils
line width (Lw) 20 24 0.1 100 0.2 mm
units mJ/cm
2
mils mW mm mm/s

From Table 39, we can see that higher critical exposure, smaller penetration depth,
lower laser power, larger beam size, and faster drawing speed contribute to smaller cure
depth; higher critical exposure, lower laser power, smaller beam size, and faster drawing
speed lead to smaller line width. This conclusion confirms the observations in the
previous section and can also be drawn from the expressions of Equations (56) and (57).
Therefore, to obtain a smaller size of part, resin with higher critical exposure and
smaller penetration depth as well as laser with lower power and faster drawing speed is
preferred. However, the depth of the cured line has to compromise with the width to
decide desirable beam radius. From the effect analysis using DOC threshold model, we
can see that when the variation of material properties also becomes an option, smaller
beam size is favorable for a small part since the depth is no more significantly dependent
on the beam radius. Furthermore, as shown in Table 39, the smallest part size is the same

151
as the beam size; while according to the analysis using DOC threshold model, it’s slightly
larger than the beam size due to diffusion and reaction in the vicinity of irradiation.
In summary, there are limited parameters to adjust when using the exposure threshold
model to guide the SL fabrication process. Only two parameters are used to characterize
material properties: critical exposure and depth of penetration. As discussed in the
previous chapter, unlike other material properties such as absorptivity or molecular
weight, the critical exposure is not an inherent property of the material. It is very specific
to a certain SL apparatus and certain fabrication as well as post-processing conditions,
and therefore has to be experimentally re-evaluated each time these conditions vary.
Therefore, the parameter analysis based on the exposure threshold model cannot provide
a useful guide for material development. Furthermore, what parameters affect the
temperature rise during building, the SL curing speed, as well as the DOC of the part and
how they affect these properties can not be investigated based on the exposure threshold.

152
CHAPTER 8
CONCLUSIONS & RECOMMENDATIONS

It can be concluded from this research that:
1. Due to the rapid radical photopolymerization, transient intensity rather than
the exposure incident on the resin should be adopted in stereolithography
process simulation. In other words, the irradiation period, although very short
(1~10ms scale), cannot be ignored and it’s not appropriate to consider the
dark reaction only.
2. By taking diffusion limitation into account, the photopolymerization kinetic
model can be extended to use in stereolithography with high laser intensity.
(also see Tang, Y., 2002)
3. The stereolithography cure process model established here can be used not
only to simulate and predict the cured size of a single line part, but parts with
overlapping lines and stacked lines. The simulation and prediction of parts
with more complex laser drawing patterns can also be expected.
4. The cure process model can be employed to investigate transient profiles of
temperature, monomer, initiator, and radical concentrations, as well as their
related properties such as propagation and termination rate. It simulates the
cure behavior during SL fabrication process, and provides insight into the
part building mechanisms.

153
5. The concept of critical DOC renders the cure process model a DOC threshold
model, corresponding to the concept of critical exposure and exposure
threshold model.
6. The laser beam radius, monomer glass transition temperature as well as
monomer diffusion coefficient are sensitive parameters for width resolution;
the initiator wt% loading, molecular weight and absorption coefficient as well
as quantum efficiency of initiation are parameters that affect the depth
resolution significantly; the laser beam radius and monomer molecular weight
are two factors that affect the curing time needed to form a certain width part;
the monomer diffusion coefficient, glass transition temperature, and
molecular weight are sensitive parameters for the maximum degree of cure a
part can reach.
7. Based on knowledge of the effective parameters, material or laser properties
can be modified to improve stereolithography speed or resolution.
The following is recommended for further study:
1. If a significant shrinkage occurs upon the laser scanning (i.e. there is a big
difference between the density of monomer and that of polymer), the
convection phenomenon occurs. In this case, both diffusion and convection
should be considered in the mass balance equations. Flach and Chartoff (1994)
developed a simple polymer shrinkage model based on the degree of
conversion from monomer to polymer and estimated the part shrinkage after
the cure reaction. They took the effect of conversion on the shrinkage into
account. In effect, the shrinkage also causes the occurrence of convection and

154
thus affects the monomer conversion. However, this was not considered in
their work. In order to incorporate the interactive effects between shrinkage
and conversion into the SL process model, a convection term needs to be
addressed and incorporated.
2. Since most of SL resins are comprised of not only acrylate but epoxy
(Melisaris, et al., 2000; Pang, et al., 2000; Steinmann, et al., 1995, 1999), a
model epoxy curing system (monomer and photoinitiator) should be
established and its cationic photopolymerization kinetics characterized in order
to investigate the epoxy cure behavior and its interaction with the acrylate cure.
This knowledge would enrich the material development guidelines prepared
here. (The modeling approach presented in this work can be extended to epoxy
resin or acrylate-epoxy blend.)
Finally, it should be mentioned that the possibility of extending the cure process
model without changing material properties or kinetics to SL commercial resin of
confidential compositions has been tested. It turns out that when extending the
application of the process model from the model acrylate studied here to commercial SL
resin SM 7110, the prediction error is found to be up to 30 %. This indicates that the
material properties or the kinetics do need to change accordingly when the resin changes.
Cationic polymerization plays a significant role in SM 7110 curing since it’s composed
of both acrylate and epoxy. That the exposure threshold model does not need to take
material change into account is a defect and also an advantage. One big benefit of the SL
cure process model established here is to investigate the effects of process, laser, or
material properties and to prepare material development guidelines for SL improvement.

155
APPENDIX A
NOVECURE OUTPUT WITH 365NM FILTER (EXFO)



156

APPENDIX B
NEGLIGIBLE HEATING EFFECT OF LIGHT IN DPC EXPEIRMENTS
shutter on
1.00min
16.00min
shutter off
50C (continuous irradiation)
negligible heating effect of light
-2
0
2
4
6
8
10
H
e
a
t

F
l
o
w

(
W
/
g
)
0 5 10 15 20 25
Time (min)

Exo Up



157
APPENDIX C
A BRIEF LITERATURE REVIEW ON GEL POINT ESTIMATION

158
Gel point is a critical value for the extent of reaction above which the produced
polymer becomes of infinite molecular weight. Flory (1953) developed a theory to
estimate the gel point for the nonlinear condensation polymerization systems. The
following assumptions were used: (i) all functional groups of each kind of structural unit,
A and B, are equally reactive, i.e. the reactivity of an A or B group is independent of the
size or structure of the molecule to which it is attached; (ii) the condensation between A
and B on the same molecule is negligible, i.e. cyclization can be ignored. If A and B
groups are initially present in equivalent quantities and only one of them has a
functionality greater than two, the branching probability (i.e. the probability that a given
functional group of a branch unit leads via a chain of bifunctional units to another branch
unit), α, can be reduced to be the square of the conversion of the functional units, D.


2
D α = (71)

Equation (72) states the critical condition for formation of infinite networks (Flory,
1953):


1
1
c
f
α =

(72)

where α
c
is the critical value of α, and f is the functionality of the branching unit, A or B,
whichever has a functionality greater than two.
Then the gel point, D
c
, for the condensation of the two functional units, A and B, can
be obtained from the following equation:

159

1
1
c
D
f
=

(73)

Miller et al. (1979), Macosko and Miller (1976), and Valles and Macosko (1979)
derived the weight average molecular weight for the nonlinear stepwise polymerization
system, and obtained the gel point as the conversion where the molecular weight
diverges. Their results are in agreement with Flory’s theory. Miller and Macosko (1976)
obtained results that agree with Flory’s theory, too, by deriving the probability of a finite
chain in a polymer network (from which the gel point was estimated). Miller et al. (1979)
and Macosko and Miller (1976) also extended the use of Flory’s model to include the
case where more than one type of branching unit is present in the condensation system.
According to Flory’s theory (1953), Miller and Macosko (1976) and Macosko and
Miller (1976) applied the same equations they developed for stepwise reactions to cross-
linking reactions of polymer chains. The cross-linking occurs by reacting side groups on
long, linear polymer chains or through unsaturation in the chain backbone. They assumed
all the reactive groups are of the same type and derived the gel point as a function of the
weight average degree of polymerization of the initial mixture of long chains (Macosko
and Miller, 1976).
Macosko and Miller (1976) also derived an expression to estimate the gel point for
networks formed by chain polymerization. The chainwise reaction involves initiation,
propagation, and termination. The gel point was found to be related to the monomer
functionality, probability that a growing chain adds one more unit, probability that a
chain terminates by combination, and the mole fraction of functional groups. Landin and
Macosko (1983, 1988) utilized this relation to express the gel point for the chainwise co-

160
polymerization of mono- and di-functional monomers. Miller and Macosko (1987, 1988)
employed the chain length and site distribution to demonstrate the gel point for the
network formed by chain crosslinking as a function of the expected number of sties on a
chain randomly chosen by site. Dotson and co-workers (1988) derived the weight-
average molecular weight for the crosslinking free radical polymerization and obtained
the gel point as the conversion where the molecular weight goes to infinity. Okay (1994)
also derived an equation to approximate the gel point for the free radical chain
copolymerization and cross-linking system. The gel point was expressed as a function of
the accumulated average chain lengths of the primary molecules, the initial mole fraction
of the functional groups in the monomer mixture, and the reactivity of the functional
groups. All these theoretical derivations for gel point of a chain polymerization system,
however, were difficult to parameterize and not validated in practice.
González-Romero and Macosko (1985) analyzed the kinetics of the free radical
crosslinking polymerization and derived a relation between the gel point and gel time for
a radical polymerization system that involves inhibition. They measured the viscosity of
the system during the reaction and took the time at which the viscosity goes to infinity as
gel time. The gel point thus can be calculated. This relation involving the rate constants
for inhibition and propagation was verified experimentally. In this relation, the inhibitor
was assumed to be ideal, i.e. the monomer doesn’t react with the radicals until the
inhibitor is consumed.
Suematsu and Kohno (2000) separated the critical point of branched polymers into
two terms: intermolecular reaction and cyclization, i.e. D
c
= D (inter) + D (ring). They
took this idea as a starting point and deduced an analytic expression for the gel point of a

161
polymerizing system consisting of same type functional units (Equation 74). Each
functional unit is assumed to have an equal chance to undergo cyclization.


1 2[ ]
1
c
D
f fC
Γ
= +

(74)

where D
c
is the conversion at which gelation occurs, f is the functionality, ] [Γ is the
molar concentration of rings formed by cyclization, and C is the initial monomer
concentration of the system. Assuming that the gel lattice has high dimensions, Suematsu
and Kohno (2000) used a percolation model and expressed the molar concentration of
cyclics in a form of the solution of the ring distribution function for the site-bond
percolation problem.
If the cyclization in the polymerizing system is negligible, Equation (74) can be
reduced to the following:


1
1
c
D
f
=

(75)

This is in agreement with the Flory’s theory if the branched polymerization here is
considered as a condensation. In Flory’s theory, if only one type of functional group is
present and these groups are capable of condensing with one another, then the branching
probability α equals the extent of reaction D, not D
2
. Incorporating this relation into the
critical condition Equation (72), the gel point expression is the same as Equation (75).

162
Miller and Macosko (1976), Miller et al. (1979), and Macosko and Miller (1976) also
obtained the same gel point expression for the stepwise homo-polymerization.
If considering the investigated tetraacrylate crosslinking system in our work as a
condensation (model of Suematsu and Kohno (2000) is for general branched polymers,
chainwise or stepwise, and does not need this assumption) and assuming the cyclization
to be negligible, the gel point of the system can be estimated to be ~14%. The critical
DOC determined (9~10%) for the DOC threshold model (Chapter 6) is thus found to be
lower than the gel point, which indicates that the produced polymer doesn’t have to
achieve an infinite molecular weight to form a solid part.

163
APPENDIX D
MINITAB REGRESSION OUTPUT OF REDUCED MODEL FOR WIDTH
RESOLUTION





164
APPENDIX E
MINITAB STEPWISE REGRESSION OUTPUT FOR WIDTH RESOLUTION




165
APPENDIX F
REGRESSION PREDICTION MODEL FOR DEPTH RESOLUTION




166
APPENDIX G
REGRESSION PREDICTION MODEL FOR WIDTH RESOLUTION



167

APPENDIX H
REGRESSION PREDICTION MODEL FOR CURING SPEED (WIDTH)




168
APPENDIX I
REGRESSION PREDICTION MODEL FOR MAXIMUM DOC



169


170
APPENDIX J
REGRESSION PREDICTION MODEL FOR MAXIMUM TEMPERATURE RISE



171


172
REFERENCES

Anseth, K.S. and Bowman, C.N. (1993) ‘Reaction Diffusion Enhanced Termination in
Polymerizations of Multifunctional Monomers’, Polymer Reaction Engineering,
Vol.1, pp.499-520.

Anseth, K.S., Bowman, C.N., and Peppas, N.A. (1994a) ‘Polymerizatioin Kinetics and
Volume Relaxation Behavior of Photopolymerized Multifunctional Monomers
Producing Highly Crosslinked Networks’, Journal of Polymer Science: Part A:
Polymer Chemistry, Vol.32, pp.139-147.

Anseth, K.S., Wang, C.M., and Bowman, C.N. (1994b) ‘Reaction Behaviour and Kinetic
Constants for Photopolymerizations of Multi(meth)acrylate Monomers’, Polymer,
Vol.35, pp.3243-3250.

Bass, J., Dugdale, J.S., Foiles, C.L., and Myers, A (1985) in Hellwege, K.H. and
Madelung, O. (Eds.) Numerical Data and Functional Relationships in Science and
Technology, New York: Springer-Verlag, pp.222.

Brandrup, J. and Immergut E.H. (Ed.) (1989) Polymer Handbook, 3rd Edition, New
York: John Wiley & Sons, V/77.

Bowman, C.N. and Peppas, N.A. (1991) ‘Coupling of Kinetics and Volume Relaxation
during Polymerizations of Multiacrylates and Multimethacrylates’, Macromolecules,
Vol.24, pp.1914-1920.

Buback, M., Degener, B., and Huckestein, B. (1989) ‘Conversion Dependence of Free-
Radical Polymerization Rate Coefficients from Laser-Induced Experiments, 1’,
Makromol. Chem., Rapid Commun., Vol.10, pp.311-316.

Buback, M. (1990) ‘Free-Radical Polymerization up to High Conversion. A General
Kinetic Treatment’, Makromol. Chem., Vol.191, pp.1575-1587.

Bueche, F. (1962) Physical Properties of Polymers, New York: Interscience.

Burel, F., Lecamp, L., Youssef, B., Bunel, C., and Saiter, J-M. (1999) ‘Synthesis and
Photoinitiated Polymerization of a New Urethane Acrylate Monomer: Influence of
Polymerization Temperature’, Thermochimica Acta, Vol.326, pp.133-141.

Cho, H.S., Park, W.S., Choi, B.W., and Leu, M.C. (2000) ‘Determining Optimal
Parameters for Stereolithography Processes via Genetic Algorithm’, Journal of
Manufacturing Systems, Vol.19, pp.18-27.


173
Chockalingam, K., Jawahar, N., and Vijaybabu, E. R. (2003) ‘Optimization of Process
Parameters in Stereolithography using Genetic Algorithm’, Proceedings of SPIE-The
International Society for Optical Engineering, 5062 (Pt. 1, Smart Materials,
Structures, and Systems), pp.417-424.

Cook, W.D. (1992) ‘Thermal Aspects of the Kinetics of Dimethacrylate
Photopolymerization’, Polymer, Vol.33, pp.2152-2161.

Cook, W.D. (1993) ‘Photopolymerization Kinetics of Oligo(ethylene oxide) and
Oligo(methylene) Oxide Dimethacrylates’, J. Polym. Sci. Part A: Polym. Chem.,
Vol.31, pp.1053-1067.

Crivello, J.V. and Dietliker, K. (1998) Photoinitiators for Free Radical, Cationic &
Anionic Photopolymerisation, 2nd Edition, Vol. III in Bradley, G. (Ed.) Chemistry &
Technology of UV & EB Formulation for Coatings, Inks & Paints, New York: John
Wiley & Sons.

DiGuilio, R.M. and Teja, A.S. (1990) ‘Thermal Conductivity of Poly(ethylene glycols)
and Their Binary Mixtures’, Journal of Chemical and Engineering Data, Vol.35,
pp.117-121.

Dotson, N.A, Galván, R., and Macosko, C.W. (1988) ‘Structural Development during
Nonlinear Free-Radical Polymerizations’, Macromolecules, Vol.21, pp.2560-8.

Eschl, J., Blumenstock, T., and Eyerer, P. (1999) ‘Comparison of the Curing Process of
Epoxy and Acrylate Resins for Stereolithography by Means of Experimental
Investigations and FEM – Simulation’, Solid Freeform Fabrication Symposium
Proceedings, pp.453-460.

Esposito Corcione, C., Greco, A., and Maffezzoli, A. (2003) ‘Photopolymerization
Kinetics of an Epoxy Based Resin for Stereolithography’, Journal of Thermal
Analysis and Calorimetry, Vol.72 (2), pp.687-693.

Esposito Corcione, C., Greco, A., and Maffezzoli, A. (2004) ‘Photopolymerization
Kinetics of an Epoxy-Based Resin for Stereolithography’, Journal of Applied
Polymer Science, Vol.92 (6), pp. 3484-3491.

Flach, L. and Chartoff, R.P. (1994) ‘A Simple Polymer Shrinkage Model Applied to
Stereolithography’, Solid Freeform Fabrication Symposium Proceedings, pp.225-
233.

Flach, L. and Chartoff, R.P. (1995a) ‘A Process Model for Nonisothermal
Photopolymerization with a Laser Light Source. I: Basic Model Development’,
Polymer Engineering and Science, Vol. 35, pp.483-492.

Flach, L. and Chartoff, R.P. (1995b) ‘A Process Model for Nonisothermal

174
Photopolymerization with a Laser Light Source. II: Behavior in the Vicinity of a
Moving Exposed Region”, Polymer Engineering and Science, Vol. 35, pp.493-498.

Flory, P.J. (1953) Principles of Polymer Chemistry, Ithaca, NY: Cornell University Press.

Fouassier, J.-P. (1995) Photoinitiation, Photopolymerization, and Photocuring –
Fundamentals and Applications, Cincinnati, OH: Hanser/Gardner Publications.

González-Romero, V.M. and Macosko, C.W. (1985) ‘Viscosity Rise during Free Radical
Crosslinking Polymerization with Inhibition’, Journal of Rheology, Vol.29, pp.259-
272.

Goodner, M.D., Lee, H.R., and Bowman, C.N. (1997) ‘Method for Determining the
Kinetic Parameters in Diffusion-controlled Free-Radical Homopolymerizations’, Ind.
Eng. Chem. Res., Vol.36, pp.1247-1252.

Goodner, M.D. and Bowman, C.N. (1998) ‘Modeling and Experimental Investigation of
Light Intensity and Initiator Effects on Solvent-Free Photopolymerization’, in Long,
T.E. and Hunt M.O. (Eds.), Solvent-Free Polymerizations and Processes:
Minimization of Conventional Organic Solvents, Washington, DC: American
Chemical Society, Vol.713, pp.220-231.

Goodner, M.D. and Bowman, C.N. (1999) ‘Modeling Primary Radical Termination and
Its Effects on Autoacceleration in Photopolymerization Kinetics’, Macromolecules,
Vol.32, pp.6552-6559.

Goodner, M.D. and Bowman, C.N. (2002) ‘Development of Comprehensive Free Radical
Photopolymerization Model Incorporating Heat and Mass Transfer Effects in Thick
Films’, Chemical Engineering Science, Vol.57, pp.887-900.

Hur, S. S., Lee, J. H., and Youn, J. R. (1997) ‘A Study on Simulation of the Deformation
of 3-D Stereolithography Products’, Han'guk Somyu Konghakhoechi, Vol.34(6),
pp.374-385.

Hur, S. S. and Youn, J. R. (2000) ‘Thermal Deformation of a Photo-cured Polymer for
the Analysis of Stereolithography’, Polymer-Plastics Technology and Engineering,
Vol.39 (4), pp.651-666.

Jacobs, P.F. (1992) Rapid Prototyping & Manufacturing: Fundamentals of
Stereolithography, Dearborn, MI: Society of Manufacturing Engineers.

Jayanthi, S., Keefe, M., and Gargiulo, E. P. (1994) ‘Studies in Stereolithography:
Influence of Process Parameters on Curl Distortion in Photopolymer Models’, Solid
Freeform Fabrication Symposium Proceedings, pp.250-258.


175
Kahle, O., Wielsch, U., Metzner, H., Bauer, J., Uhlig, C., and Zawatzki, C. (1998) ‘Glass
Transition Temperature and Thermal Expansion Behaviour of Polymer Films
Investigated by Variable Temperature Spectroscopic Ellipsometry’, Thin Solid
Films, Vol.313-314, pp.803-807.

Landin, D.T. and Macosko, C.W. (1983) ‘Rheological Changes during the
Copolymerization of Vinyl and Divinyl Monomers’, Organic Coatings and Applied
Polymer Science Proceedings: Preprints of Papers, Vol.48, pp.433-439.

Landin, D.T. and Macosko, C.W. (1988) ‘Cyclization and Reduced Reactivity of Pendant
Vinyls during the Copolymerization of Methyl Methacrylate and Ethylene Glycol
Dimethacryalte’, Macromolecules, Vol.21, pp.846-851.

Lecamp, L., Youssef, B., and Bunel, C. (1997) ‘Photoinitiated Polymerization of a
Dimethacrylate Oligomer: 1.Influence of Photoinitiator Concentration, Temperature
and Light Intensity’, Polymer, Vol.38, pp.6089-6096.

Lecamp, L., Youssef, B., Bunel, C., and Lebaudy, P. (1999) ‘Photoinitiated
Polymerization of a Dimethacrylate Oligomer: Part 3. Postpolymerization Study’,
Polymer, Vol.40, pp.6313-6320.

Mack, C.A. (1985) ‘Prolith: a Comprehensive Optical Lithography Model’, SPIE Optical
Microlithography IV, Vol.538, pp.207-220.

Mack, C.A. (1986) ‘Analytical Expression fro the Standing Wave Intensity in
Photoresist’, Applied Optics, Vol.25, pp.1958-1961.

Mack, C.A. (1994) ‘Standing Waves in Photoresist’, Microlithography World, pp.22-24.

Macosko, C.W. and Miller, D.R. (1976) ‘A New Derivation of Average Molecular
Weight of Nonlinear Polymers’, Macromolecules, Vol.9, pp.199-206.

Maffezzoli A., Micelli, F., Terzi R., and Luprano, V.A.M. (2001), ’Characterization of
the Kinetic Behavior of Resin Modified Glass-Ionomer Cements by DSC, TMA and
Ultrasonic Wave Propagation’, Journal of Materials Science: Materials in Medicine,
Vol.12, pp.151-156.

Maffezzoli A. and Terzi R. (1995), ‘Thermal Analysis of Visible-Light-Activated Dental
Composites’, Thermochimica Acta, Vol.269/270, pp.319-335.

Maffezzoli, A. and Terzi, R. (1998) ‘Effect of Irradiation Intensity on the Isothermal
Photopolymerization Kinetics of Acrylic Resins for Stereolithography’,
Thermochimica Acta, Vol.321, pp.111-121.

Marten, F.L. and Hamielec, A.E. (1979), ‘High Conversion Diffusion-Controlled
Polymerization’, American Chemical Society Symposium, Ser.104, pp.43-70.

176

Marten, F.L. and Hamielec, A.E. (1982), ‘High Conversion Diffusion-Controlled
Polymerization of Styrene I’, Journal of Applied Polymer Science, Vol.27, pp.489-
505.

Mateo, J.L., Serrano, J., and Bosch, P. (1997), ‘Photopolymerization of Di- and
Tetrafunctional Methacrylic Monomers in a Polymeric Medium: Kinetics and
Evidence of Reaction Diffusion throughout the Photopolymerization Reaction’,
Macromolecules, Vol.30, pp.1285-1288.

Melisaris, A.P. et al. (2000) ‘Liquid, Radiation-curable Composition, Especially for
Producing Flexible Cured Articles by Stereolithography’, United States Patent, Pat.
No. 6,136,497.

Miller, D.R., and Macosko, C.W. (1976) ‘A New Derivation of Post Gel Properties of
Network Polymers’, Macromolecules, Vol.9, pp.206-211.

Miller, D.R., Valles, E.M., and Macosko, C.W. (1979) ‘Calculation of Molecular
Parameters for Stepwise Polyfunctional Polymerization’, Polymer Engineering and
Science, Vol.19, pp.272-283.

Miller, D.R. and Macosko, C.W. (1987) ‘Molecular Weight Relations for Crosslinking of
Chains with Length and Site Distribution’, Journal of Polymer Science: Part B:
Polymer Physics, Vol.25, pp.2441-69.

Miller, D.R. and Macosko, C.W. (1988) ‘Network Parameters for Crosslinking of Chains
with Length and Site Distribution’, Journal of Polymer Science: Part B: Polymer
Physics, Vol.26, pp.1-54.

Nagamori, S. and Yoshizawa, T. (2001), ‘Research on Solidification of Resin in Stereo-
Lithography (Comparison between Measured and Estimated Shapes of Solidified
Resin, and Manufacturing Accuracy of Photo-cured Model)’, SPIE’s International
Symposium on Intelligent Systems and Advanced Manufacturing (Opto-Mechatronic
Systems).

Nagamori, S. and Yoshizawa, T. (2003), ‘Research on Shape of Solidified Resin in
Stereolithography’, Proceedings of SPIE-The International Society for Optical
Engineering, 5058 (Optical Technology and Image Processing for Fluids and Solids
Diagnostics), pp.447-456.

Neter, J., Kutner, M.H., Nachtsheim, C.J., and Wasserman, W. (1996) Applied Linear
Statistical Models, New York: McGraw-Hill.

Okay, O. (1994) ‘Kinetics of Gelation in Free Radical Crosslinking Copolymerization’,
Polymer, Vol.35, pp.2613-18.


177
Onuh, S.O. and Hon, K.K.(1998a) ‘Optimising Build Parameters for Improved Surface
Finish in Stereolithography’, International journal of Machine Tools Manufacture,
Vol.38, pp.329-342.

Onuh, S.O. and Hon, K.K.(1998b) ‘Application of the Taguchi Method and New Hatch
Styles for Quality Improvement in Stereolithography’, Proceedings of the Institution
of Mechanical Engineers, Part B, Vol.212, pp.461-472.

Pananakis, D. and Watts, D.C. (2000) ‘Incorporation of the Heating Effect of the Light
Source in a Non-isothermal Model of a Visible-light-cured Resin Composites’, J.
Mat. Sci., Vol.35, pp.4589-4600.

Pang, T.H. et al. (2000) ‘Liquid Radiation-curable Composition Especially for Producing
Cured Articles by Stereolithography having High Heat Deflection Temperatures’,
United States Patent, Pat. No. 6,100,007.

Rosen, D.W. (2002) ‘Stereolithography Technology ÷ Course Material for ME7227:
Rapid Prototyping in Engineering’.

Satio, A. (1993) ‘A Study on Development of 3D Model Stereo-Lithography’, Fiscal
1992, Ministry of Education, Japan, S No.03555022.

Schaub, D.A., Chu, K-R, and Montgomery, D.C. (1997) ‘Optimising Stereolithography
Throughput’, Journal of Manufacturing Systems, Vol.16, pp.290-303.

Schulz, G.V. (1956) ‘Über die Polymerisationskinetik in hochkonzentrierten Systemen –
Zur Kinetik des Trommsdorffeffektes an Methylmethacrylat’, Zeitschrift für
Physikalische Chemie, Vol. 8, pp.190-317.

Smith, B.C. (1996) Fundamentals of Fourier Transform Infrared Spectroscopy, New
York: CRC Press.

Sun, T. and Teja, A.S. (2003) ‘Density, Viscosity, and Thermal Conductivity of Aqueous
Ethylene, Diethylene, and Triethylene Glycol Mixture between 290K and 450K’,
Journal of Chemical and Engineering Data, Vol.48, pp.198-202.

Steinmann, B. et al. (1995) ‘Photosensitive Compositions’, United States Patent, Pat. No.
5,476,748.

Steinmann, B. et al. (1999) ‘Liquid, Radiation-curable Composition, Especially for
Stereolithography’, United States Patent, Pat. No. 5,972,563.

Suematsu, K. and Kohno, M. (2000) ‘Estimation of Critical Points of Branched
Polymers’, American Physical Society, Physical Review E, Vol.62, pp.3944-53.


178
Tang, Y. (2002) ‘Stereolithography Cure Modeling’, MS Thesis, Georgia Institute of
Technology.

Tryson, G.R. and Shultz A.R. (1979) ‘A Calorimetric Study of Acrylate
Photopolymerization’, J. of Poly. Sci.: Poly. Phys. Ed., Vol.17, pp.2059-2075.

Valles, E.M., and Macosko, C.W. (1979) ‘Structure and Viscosity of
Poly(dimethylsiloxanes) with Random Branches’, American Chemical Society,
Vol.12, pp.521-526.

Van Krevelen, D.W. (1990) Properties of Polymers – their correlation with chemical
structure; their numerical estimation and prediction from additive group
contributions, New York: Elsevier.

Wu, C.F.J. and Hamada, M. (2002), Experiments Planning, Analysis, and Parameter
Design Optimization, New York: John Wiley & Sons.

Yaws, C. L. (2003) Yaws' Handbook of Thermodynamic and Physical Properties of
Chemical Compounds, Norwich, N.Y.: Knovel.

STEREOLITHOGRAPHY CURE PROCESS MODELING

Approved by: Dr. John D. Muzzy, Advisor School of Chemical and Biomolecular Engineering Georgia Institute of Technology Dr. David W. Rosen School of Mechanical Engineering Georgia Institute of Technology Dr. Rigoberto Hernandez School of Chemistry & Biochemistry Georgia Institute of Technology Date Approved: July 18, 2005 Dr. Clifford L. Henderson, Co-advisor School of Chemical and Biomolecular Engineering Georgia Institute of Technology Dr. Peter J. Ludovice School of Chemical and Biomolecular Engineering Georgia Institute of Technology

ACKNOWLEDGEMENTS

I would like to take this opportunity to thank my advisor Dr. John Muzzy for the support and freedom he provided me to pursue this project as I envisioned it. I would also like to thank my co-advisor, Dr. Cliff Henderson, for his advice and insight. I appreciate being around Dr. David Rosen who pays attention to students’ academic growth, from which I benefit a lot. Thanks also go to Dr. Peter Ludovice and Dr. Rigoberto Hernandez for serving on my thesis committee. I’m thankful to Dr. Jonathan Colton for his welcome gesture and helpful advice when I borrowed his equipment. I’d also like to mention the help from Dr. C. P. Wong and his postdoc Zhuqing Zhang for use of photo-DSC, Dr. William Koros and his graduate student William Madden for use of the density gradient column, Dr. Laren Tolbert and Kendra McCoy for use of the UV/VIS spectrometer, and Dr. Tongfan Sun for measurement of liquid thermal conductivity. I appreciate the opportunity to work in both Henderson’s and Rosen’s research groups. I owe thanks to Augustin Jeyakumar for help with ellipsometer and optical microscope, Cody Berger for sharing his experience in standing wave issue, Lovejeet Singh for discussion regarding to CTE measurement, Benita Comeau and Mikkel Thomas (ECE) for use of surface profilometer, and Trevor Hoskins for being a wonderful officemate. Benay Sager (ME) has discussed with me regarding to the SLA minivat operation, Ec & Dp measurement, and other SLA problems.

iii

and that Jeff Andrews and Brad Parker in ChBE machine shop have kindly and carefully machined the light guide custom unit and user-designed DSC pans. It shouldn’t be forgotten. Yolande Berta has offered nice help regarding to SEM measurement and carbon coating unit operation. either.It’s worth mentioning the assistance from Safdar Ali and Marshall Sloane (ChBE undergraduates) as well as Andrew Mrasek (ME undergraduate) with part building in SLA 250 and dimension measurement by SEM. that Ms. iv . and I don’t take it for granted. I’m lucky to have had you around.

..... xiv SUMMARY...........................................................................................................1 Photopolymerization Kinetic Model..................................................................... 27 3....... 46 4.................................. 28 3.............................................. 59 4....................................2 Project Objective........................................................................ 17 3..........................5 Thermal Conductivity ...................................................................3 Project Strategy.........................................................8 Summary ................................2......................................................................................................................................................3 DPC Experiments.................................................... 61 4............... 74 5......... 46 4........................................................... 2 1...................1 Specific Heat Capacity........................................................................... 1 1...........4 Kinetic Model Validation ......................................................... iii LIST OF TABLES................ xviii CHAPTER 1 INTRODUCTION ..............................1 DPC Pan.......................................................................................................2 Kinetic Experiments............... 50 4.............................................................................................................4 Density ..................... 9 CHAPTER 2 MODEL DEVELOPMENT...............................2......................................................................... 65 5................................................................................................................................................ viii LIST OF FIGURES ......7 Absorption Coefficient..................... 64 CHAPTER 5 SIMULATIONS .............................................2 Overlapping Lines....... 17 3... 11 CHAPTER 3 KINETIC CHARACTERIZATION............................1 Introduction to Stereolithography ........................ x LIST OF SYMBOLS ..................................................... 35 3.3 Stacked Single Lines.........................1 Single Laser Drawn Line .........................................................................................................................3 Coefficient of Thermal Expansion...........6 Heat of Polymerization ................................................................................ 32 3..................................................................................................................3 Kinetic Data Analysis & Model Parameterization....................................................................................................................................... 58 4.......................................2.................... 44 CHAPTER 4 MATERIAL CHARACTERIZATION ...2 Glass Transition Temperature............................................................................................................................2 Standing Waves ............................................... 1 1................................................ 28 3............. 78 v .............. 62 4.............. 52 4......................................................TABLE OF CONTENTS ACKNOWLEDGEMENTS........ 67 5....................................................................................

...............................2 Single Line Part Prediction ................... 97 6...............4 Sensitive Parameters for Temperature Rise...................2 Resolution and Speed Prediction by Regression Model ...........1 Single Line Part Prediction ......................................................................2................................. 146 7...........................................2...............3....4 Parameter Analysis using Exposure Threshold Model......................................................................................................... 92 6.3............................. 109 7................................................ 88 6.....3....................................... 127 7.........2..............................................................................................................2................................... 163 vi ..................................................................CHAPTER 6 MODEL VERIFICATION ..........3 Parameter Optimization .......................................... 82 6................................................4 Summary ..............2 DOC Threshold Model Prediction .................................................2..............................................2 Overlapping Line Part Prediction .. 91 6.1 Parameter Effect Investigation... 146 7...... 124 7........................ 132 7........... 103 6............ 152 NOVECURE OUTPUT WITH 365NM FILTER (EXFO).....................................1........3.....................3 Regression Prediction Model for Speed (Width Direction)............... 85 6. 121 7..............2................................... 138 7........................... 143 7........................................................2 Sensitive Parameters for Speed (Width Direction)..............................5 Regression Prediction Model for Maximum Temperature Rise .......... 113 7.........1.............. 118 7......................................... 131 7.....1 Ec and Dp Determination..............................3 Overlapping Line Part Prediction ..........4 Regression Prediction Model for Maximum DOC .........................................2 Parameter Optimization .............2............................................3 Exposure Threshold Model Prediction .............................................. 157 APPENDIX D MINITAB REGRESSION OUTPUT OF REDUCED MODEL FOR WIDTH RESOLUTION... 100 6.. 131 7...........1............. 155 APPENDIX B NEGLIGIBLE HEATING EFFECT OF LIGHT IN DPC EXPEIRMENTS...............5 Comparison of DOC and Exposure Threshold Model....... 85 6..............3...........................................................3 Stacked Line Part Prediction..............................................................1 Parameter Significance Investigation ...........1 DOC Threshold Model ............................. 93 6..1 Sensitive Parameters for Width Resolution ....3 Sensitive Parameters for DOC ..... 110 7..................1.......................4 Stacked Line Part Prediction................4....... 108 CHAPTER 7 MODEL APPLICATIONS...... 150 CHAPTER 8 APPENDIX A CONCLUSIONS & RECOMMENDATIONS.4.............................................. 90 6........... 140 7........1 Regression Prediction Model for Depth Resolution ........2........................................... 99 6......2 Regression Prediction Model for Width Resolution........ 82 6...................... 156 APPENDIX C A BRIEF LITERATURE REVIEW ON GEL POINT ESTIMATION ..........6 Model Prediction using Ec and Dp Evaluated by a Different Protocol ....................5 Sensitive Parameters for Depth Resolution ........... 135 7............................1..................3.......

......................... 172 vii ..................................... 168 APPENDIX J REGRESSION PREDICTION MODEL FOR MAXIMUM TEMPERATURE RISE ............................ 167 APPENDIX I REGRESSION PREDICTION MODEL FOR MAXIMUM DOC........................................................................................................................................................................ 164 APPENDIX F REGRESSION PREDICTION MODEL FOR DEPTH RESOLUTION ................. 166 APPENDIX H REGRESSION PREDICTION MODEL FOR CURING SPEED (WIDTH) ............. 165 APPENDIX G REGRESSION PREDICTION MODEL FOR WIDTH RESOLUTION .................................................................................................................. 170 REFERENCES .................................APPENDIX E MINITAB STEPWISE REGRESSION OUTPUT FOR WIDTH RESOLUTION .....................................................................................................................................................

................ and 3...466 in/sec..... 2.......... 91 Table 13 Single Line Part Prediction Results Based on Exposure Threshold Model... Stacked-line Parts ....... 37 Determination of Rate Constants k t and k P .................... and 3.......................... 2................. 42 Characterized Material Properties . 102 Table 17 Exposure Threshold Model Prediction (protocol): 1................... 93 Table 14 Exposure Threshold Model Prediction Results using Modified Beam Profile..... 38 Kinetic Parameter Values ............... 89 Table 10 DOC Threshold Model Prediction for Overlapping Line Parts............................................. 26 Table 2 Table 3 Table 4 Table 5 Table 6 Table 7 Table 8 Table 9 k t / k P and k P / k t 1/ 2 Values Obtained from DPC Experiments........................................................................................................ Stacked-line Parts ......................................................................... 90 Table 12 DOC Threshold Model Prediction for 3-Layer Stacked Line Parts........ 87 Dimension Measurements of Overlapping Line Parts............... Single Line (1) Vs = 1.... Single Line (1) Vs = 1.............................. 86 Single Line Part Prediction by DOC Threshold Model..... 106 Table 18 Exposure Threshold Model Prediction (protocol.. 89 Table 11 Dimension Measurements of 3-Layer Stacked Line Parts..... Overlapping-line.... Stacked-line Parts ........... 107 viii ...... Overlapping-line..... 2.. 96 Table 15 Comparison of Prediction Results by Two Threshold Models...071 in/sec (2) Vs = 0................. high working range): 1........................................................................... and 3..........071 in/sec (2) Vs = 0....LIST OF TABLES Table 1 Material & Process Parameters Involved in the SL Cure Process Model ................. 101 Table 16 Exposure Threshold Model Prediction (high working range): 1....... 67 Dimensions of Single Line Parts Built at Two Laser-Scanning Speeds:....071 (2) Vs = 0..... Overlapping-line............................ Single Line (1) Vs = 1................... 64 Process and Laser Parameter Values Used for Simulations ....................466 in/sec....466 in/sec...

.. 113 Table 23 Estimated Factorial Effects and Lenth’s Test for Width Resolution ............... 142 Table 35 Conditions used for Test of Temperature Rise Regression Model.......... Single Line (1) Vs = 1........................................... low working range): 1...............................Table 19 Exposure Threshold Model Prediction (protocol........... 145 Table 38 Parameters in Exposure Threshold Model and Their Level Values .................... 144 Table 37 Evolver Optimization Results for Investigated Responses........ 138 Table 33 Maximum DOC Predicted by Regression Model ....................... 111 Table 21 Significant Factors Identified from Screening Experiment .. 108 Table 20 Potential Sensitive Parameters and Their Level Values .......... 118 Table 25 Estimated Factorial Effects and Lenth’s Test for Maximum DOC . 127 Table 28 Significant Factors for Investigated Responses .......................... 2................................................................. 121 Table 26 Estimated Effects and Lenth’s Test for Maximum Temperature Rise ...............................071 in/sec (2) Vs = 0...................................... 140 Table 34 Maximum Temperature Rise Predicted by Regression Model...... Overlapping-line... 124 Table 27 Estimated Effects and Lenth’s Test for Depth Resolution ......... 132 Table 30 Depth Resolution Predicted by Regression Model ............................. 147 Table 39 Evolver Optimization Results using Exposure Threshold Model ............................... and 3............. 135 Table 32 Curing Time (Width Direction) Predicted by Regression Model................ 114 Table 24 Estimated Factorial Effects and Lenth’s Test for Speed (Width)........ 132 Table 31 Width Resolution Predicted by Regression Model................. Stacked-line Parts . 112 Table 22 Full Factorial Design and Response Values (Width Resolution) ..466 in/sec.. 150 ix ............................................................. 142 Table 36 Parameter Range Used for Response Optimization....................... 130 Table 29 Simulation Conditions to Test Predictive Ability of Regression Models...................................................

.............................. 51 x ........ 28 Figure 8 Standing Wave Intensity at 365nm.. 27 Structure Formula of DMPA (Ciba)... 1/T ...... 41 Figure 16 Comparison of the Experimental and Simulated Polymerization Rate Curves (incident power = 0.............................. 12 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 2D Domain for Single Laser Drawn Line .... 5 Figure 2 Cured Shape of Single Laser Drawn Line.................................................................................................. 40 Figure 13 Nonlinear Fit of Termination Rate Constant k t vs....................... 19 Structure Formula of E4PETeA (Sartomer) ... (c) 70oC........................................ (b) 50oC...................LIST OF FIGURES Figure 1 Complex SL Process and Oversimplified Exposure Threshold Model......................... Conversion X (50oC)......... 12 Absorbed Intensity at Point Q(x....................................................................... 33 Figure 11 DPC Experimental Curves (Continuous and Flash Exposure at 50oC). 40 Figure 14 Semi-log Plot of True Kinetic Constants k P 0 and k t 0 vs...y.......... 50 Figure 20 Glass Transition of Cured Poly(E4PETeA) Detected by DSC ...... Conversion X (50oC) .............. 30 Figure 9 Standing Wave Intensity at 304-395nm ......................z)..................... 41 Figure 15 Linear Fit of 1 / f c (Critical Fractional Free Volume) vs............... 27 DPC Sample Pan ........... 48 Figure 18 Cp-T Plot of Cured E4PETeA Polymer Exported from MDSC Data.................................................................. 1/T...............................................................................1 mW): (a) 30oC...... 30 Figure 10 Isothermal DSC Runs to Detect the Onset Temperature of Thermal Cure ................... 34 Figure 12 Nonlinear Fit of Propagation Rate Constant k P vs............................... 45 Figure 17 Cp-T Plot of Liquid E4PETeA Monomer Exported from MDSC Data.................. 48 Figure 19 Glass Transition of Liquid E4PETeA Monomer Detected by DSC........

...................................... 81 Figure 37 SEM Image of Cross Section of a Single Line Part ...... Width at the Top Surface of the Single Line Part (Plot Interval = 0............................................. 56 Figure 26 Effect of Temperature on Heat Generated by Polymerization . Depth along the Centerline of the Single Line Part (Plot Interval = 0........... Case II.......... (c) Radical Concentration.. Single Laser Drawn Line....... 73 Figure 33 (a) Monomer Conversion (b) Initiator Concentration (mol/m3) Distributions upon Two Overlapping Scans..................... Depth at the Centerline of Two-Layer-Line Part (Plot Interval = 0...................................01 sec..... 66 Figure 29 Transients of (a) Intensity...... Overlapping Single-Layer Lines...................... (d) Monomer Conversion....... 61 Figure 27 Absorption Coefficient Spectrum of DMPA.............. 51 Figure 22 Effect of Heating Rate on Measured Tg Value of Cured Polymer ....... except for t = 1860 sec)............ except t = 1860 sec) .......... 54 Figure 25 CTEs of Poly(E4PETeA) below Tg Determined by Linear Regression of Curves Obtained by Fitting with Si Substrate Optical Data of 25oC (diamonds) and of Curves Obtained by Fitting with Temperature Dependent Si Substrate Data (triangles) ... and (e) Temperature at Point (x..................................................Figure 21 Effect of Heating Rate on Measured Tg Value of Liquid Monomer ....01 sec.......................... 70 Figure 30 Distribution of (a) Monomer Conversion and (b) Photoinitiator Concentration upon a Single Laser Scan... 76 Figure 34 Monomer Conversion vs................................................................ 72 Figure 31 Monomer Conversion vs................................................................ except for t = 1860 sec)......... 83 xi .......... (b) Initiator Concentration........................ 52 Figure 23 Temperature Dependence of Liquid E4PETeA Monomer Film Thickness .... 0)........................................01 sec........... 0....................... Width at the Top Surface of Two-Overlapping-Line Part (Plot Interval = 0...................................01 sec............................ except for t = 1860 sec) ............ 63 Figure 28 Three Basic Laser Drawing Patterns: Case I.............................. Case III.............. 80 Figure 36 Monomer Conversion vs....................................... 54 Figure 24 Temperature Dependence of Cured E4PETeA Polymer Film Thickness .. 73 Figure 32 Monomer Conversion vs............ Stacked Single Lines .... 77 Figure 35 (a) Monomer Conversion (b) Initiator Concentration (mol/m3) Distributions upon Two Stacked Scans..............................................

................................ 88 Figure 40 Working Curve from WINDOWPANETM Experimental Data ...... 94 Figure 42 Laser Movement when Drawing Overlapping Lines ..................... 84 Figure 39 Degree of Cure Contour for Parts Built at Vs = 10.............................................................................. 92 Figure 41 Beam Intensity Profile of HeCd Laser in SLA-250/50 ................... 114 Figure 48 Factorial Effects Plot for Width Resolution: (a) main effect (b) interaction.................... 104 Figure 45 Ec and Dp Determined in the High Range using Part Building Protocol ............................ 134 Figure 55 Curvature in Factors for Speed (Width) ..................... 116 Figure 49 Factorial Effects Plot for Speed (Width): (a) main effect (b) interaction........................... 137 Figure 58 Curvature in Factors for Maximum DOC .................................................................71 in/sec.............................................. 135 Figure 56 Nonlinear Behavior of Beam Radius for Speed (Width)........071in/sec (with the measured part contour shown in red). 129 Figure 53 Curvature in Factors for Width Resolution ................................................................................................... 97 Figure 43 High Working Range for Model Acrylate Resin in SLA .Figure 38 Degree of Cure Contour for Parts Built at Vs = 1............................. 102 Figure 44 Comparison of Working Curves Obtained by the 3D Systems WINDOWPANE Procedure (labeled “SOP” in the figure) and by the Part Building Protocol................... 133 Figure 54 Curvature Effect for Width Resolution: (a) main effect (b) interaction................................................. 120 Figure 50 Factorial Effects Plot for Max DOC: (a) main effect (b) interaction ....... 138 xii ......... 106 Figure 46 Ec and Dp Determined in the Low Range using Part Building Protocol .......... 107 Figure 47 Normal Plot for Width Resolution ................. 136 Figure 57 Factors Curvature for Speed (Width): (a) main effect (b) interaction. 123 Figure 51 Factorial Effects Plot for Temperature Rise: (a) main effect (b) interaction 126 Figure 52 Factorial Effects Plot for Depth Resolution: (a) main effect (b) interaction..................

...........................Figure 59 Factors Curvature for Maximum DOC (a) main effect (b) interaction . 141 Figure 62 Main Effects Plot for Cure Depth....................... 148 Figure 63 Main Effects Plot for Line Width.......................... 140 Figure 61 Factors Curvature for Max Temp Rise (a) main effect (b) interaction......................................... 148 xiii .................................................. 139 Figure 60 Curvature in Factors for Maximum Temperature Rise .................

LIST OF SYMBOLS AEp Pre-exponential Factor of Propagation Rate Constant Dependence on Temperature AEt Pre-exponential Factor of Termination Rate Constant Dependence on Temperature Parameter of Propagation Rate Constant Dependence on Fractional Free Volume Parameter of Termination Rate Constant Dependence on Fractional Free Volume Cure Depth of Resin Specific Heat Capacity Specific Heat Capacity of Monomer Specific Heat Capacity of Polymer Diffusion Coefficient of Monomer Penetration Depth of Laser into Resin Diffusion Coefficient of Polymeric Radical Diffusion Coefficient of Photoinitiator Exposure Critical Exposure Activation Energy for Propagation Activation Energy for Termination Fractional Free Volume Ap At Cd CP C P.M C P.P DM DP DP· DS E Ec Ep Et f f cp Critical Fractional Free Volume for Propagation xiv .

i..e. the Lateral Distance between Adjacent Laser Scan Centerlines Incident Laser Intensity Laser Peak Intensity Absorbed Light Intensity Thermal Conductivity Diffusion Limited Kinetic Constant Rate Constant of Propagation True Rate Constant of Propagation Reaction Limited (“true”) Kinetic Constant Rate Constant of Reaction Diffusion Rate Constant of Termination True Rate Constant of Termination Average Sample Thickness Over Temperature Range Average Free Path Length Linewidth of the Cured Line Monomer Concentration Initial Monomer Concentration h hs I Io Ia k kD kp k p0 kr k r* kt kt 0 L Lf Lw [M] [M]0 xv .f ct fM fP Critical Fractional Free Volume for Termination Fractional Free Volume of Pure Monomer Fractional Free Volume of Pure Polymer Heat Convection Coefficient Hatch Space.

[P·] Polymeric Radical Concentration Laser Power PL Q(t ) Qtot R Ri RN Rp Rrd Rt S Heat Integral in DPC Experiment Reference Heat of Reaction Gas Constant = 8.314J/mol-K Rate of Initiation Normalized Rate of Propagation Rate of Propagation Reaction Diffusion Parameter Rate of Termination Photoinitiator Concentration Characteristic Exposure Time SLA Resin Bath Temperature Glass Transition Temperature of Monomer Glass Transition Temperature of Polymer Ambient Temperature in SLA Chamber Rao Function Volume of Material Molar Volume per Structure Unit of Polymer Laser Scanning Speed Half Width of Laser Spot (@ 1/e2) Monomer Conversion te Tb TgM TgP Tinf UR V Vm Vs wo X xvi .

α Volumetric Coefficient of Thermal Expansion Coefficient of Thermal Expansion of Monomer Coefficient of Thermal Expansion of Polymer Linear Coefficient of Thermal Expansion Molar Absorptivity Quantum Yield of Initiation Volume Fraction of Monomer Laser Wavelength (325nm in SLA-250/50) Poisson’s Ratio Density Density of Pure Monomer Density of Pure Polymer Heat of Polymerization αM αP β ε φi φM λ ν ρ ρM ρP ∆H P xvii .

photopolymerizaion. The free radical photopolymerization kinetics is also characterized by differential photocalorimetry (DPC) and a comprehensive kinetic model parameterized for the model material. photoinitiation. The SL process model is then solved using the finite element method in the software package.SUMMARY Although stereolithography (SL) is a remarkable improvement over conventional prototyping production. are determined. mass and heat transfer. the thermal and physical properties of a model compound system. and maximum DOC of the xviii . The SL cure process model. and provides insight into the part building mechanisms. The DOC threshold model has been used to investigate the effects of material and process parameters on the SL performance properties. It predicts the cured part dimension within 25% error. simulates the cure behavior during the SL fabrication process. To parameterize the model. FEMLAB. it is being pushed aggressively for improvements in both speed and resolution. speed. In order to address this issue a quantitative SL cure process model is developed which takes into account all the sub-processes involved in SL: exposure. also referred to as the degree of cure (DOC) threshold model. However.2-dimethoxy-2-phenylacetophenone (DMPA) as initiator. and validated by the capability of predicting fabricated part dimensions. ethoxylated (4) pentaerythritol tetraacrylate (E4PETeA) with 2. it is not clear currently how these two features can be improved simultaneously and what the limits are for such optimization. such as resolution. maximum temperature rise in the resin bath. while the prediction error of the exposure threshold model currently utilized in SL industry is up to 50%.

The effective factors are identified and parameter optimization is performed. which also provides guidelines for SL material development as well as process and laser improvement.green part. xix .

htm. http://conceptual-reality.”2 ‘Stereolithography’. until the part is complete. ‘How Stereolithography (3-D Layering) Works’.. “It translates computer aided designs (CAD) into solid objects through a combination of laser. and the strategy to achieve the goal is demonstrated. with successive layers bonding to each other. 2001.g.L. Howstuffworks Inc. A basic printing process goes like this2: • • “A 3-D model of an object is created in a CAD program. 3D Systems) slices the 3-D CAD model into a series of very thin horizontal layers. 1998-2001.C. Conceptual Reality L. 2 1 1 . the stereolithography (SL) technology is introduced. The software (e. photochemistry and software technologies”1. 1.htm. www.com/stereolith.howstuffworks. • The newly built layer attached to the platform is lowered to just below the surface the distance of one layer. This process repeats layer by layer.com/stereo.CHAPTER 1 INTRODUCTION In this chapter. hardening it.. and a new layer of resin is then recoated and scanned on top of the previous one. the objective of this work is addressed.1 Introduction to Stereolithography Stereolithography is currently the most widely used process in the rapid prototyping and manufacturing (RP&M) field. • The sliced information is transferred to an ultraviolet laser that scans the top layer of the photosensitive resin. Lightyear.

“Traditional prototype production is a long, inefficient, expensive and fraught-withinaccuracy process that adds to the ultimate cost of a product, wastes manpower and materials, and slows the production cycle”3. SL technology provides a solution to these problems inherent in the traditional approach. “It is a technological breakthrough that allows solid physical parts to be made directly from computer data in a short time using an automated process.”3

1.2

Project Objective

Although SL is a remarkable improvement over the conventional prototyping production in many aspects, it still needs further improvement in speed and resolution to meet the demands of industry. Resolution is particularly important as it indicates the minimum feature sizes and surface finish achievable. One important factor that affects SL resolution is inherent in the nature of the laser. For example, for the case of a Gaussian laser and a resin obeying the Beer-Lambert law, the resin will cure in a shape of a parabolic cylinder upon a single laser scan vector (Jacobs, 1992). Using a smaller layer thickness can reduce this boundary effect, but it also increases the build time. Resolution can be improved by shrinking the laser beam size, but it also causes an increase in the building time. Increasing the laser intensity can improve SL speed since both the rate and degree of cure increase with the intensity (Maffezzoli et al., 1998). However, since the cure reaction is exothermic and SL resins have low thermal conductivities, the heat of reaction associated with the local photopolymerization cannot be easily dispersed. When the laser intensity is increased in order
3

‘Benefits of Stereolithography – Higher Quality, Lower Costs’, Pure Fluid Magic Inc., 1999, www.purefluidmagic.com/sl_bene.htm.

2

to increase the part building speed, it also unfortunately leads to faster heat generation. Consequently, some thermally initiated polymerization might occur in the vicinity of the exposed region, which would reduce the resolution of the prototype being constructed. Furthermore, the temperature gradients built within the resin might cause considerable thermal stresses and correspondingly thermal strains, which could deteriorate the mechanical/chemical properties of the part, or even manifest themselves as part distortion. Can these two features in the SL process, resolution and speed, be improved simultaneously or do they have to be compromised with each other? If there is an optimized solution, what are the limits for such optimization given a photosensitive material system? What are the most sensitive parameters that affect the resolution or speed? In order to answer these questions, being able to simulate and predict part shape, build time, and potential difficulties would be very beneficial. Current models of the SL process assume that the extent of resin cure is a function of only the amount of exposure to UV radiation (Jacobs, 1992). They utilize an exposure threshold model that assumes a dose E(x,y,z) that is greater than a minimum “critical exposure,” Ec, causes the resin to solidify at point (x,y,z). Basically it derives an exposure spatial distribution in the resin, e.g. Equation (1), for a single laser drawn line (Jacobs, 1992), and substitutes Ec for E(y,z), then y* and z* obtained (Equation 2) describe the cured shape of the part.

E ( y, z ) =

2 PL 2 exp(−2 y 2 / w0 ) exp(− z / D P ) π w 0V s

(1)

3

Ec =

2 PL 2 exp(−2 y * 2 / w0 ) exp(− z * / D P ) π w 0V s

(2)

where PL , w0 , and Vs are the laser power, beam radius, and scanning speed, respectively;
DP is the penetration depth of the laser into the resin, the depth where the laser intensity

decreases to 1/e (about 36.8%) of the intensity incident at the resin surface. DP can be expressed as DP = 1 /(2.3εS ) (Jacobs, 1992), where ε and S are the absorption coefficient and concentration of the photoinitiator, respectively. A Gaussian laser and a resin obeying Beer’s Law are assumed here. This exposure threshold model is an oversimplification of the SL process. As demonstrated in Figure 1, it directly connects the exposure to the resin and the final solid part shape. It ignores an important intermediate step: reaction. Therefore, how the reaction, the resin kinetic characteristics, as well as the diffusion and thermal effects influence the size, shape and properties of parts fabricated by SL cannot be investigated by using this model. Its ability to predict the cured part outline is challenged especially when part resolution is in demand.

4

e. laser beam diameter. chemical reaction. Saito (1993) conducted experiments varying laser power and scanning speed in SLA. i.Figure 1 Complex SL Process and Oversimplified Exposure Threshold Model Another deficiency of the exposure threshold model currently used in industry is that it also assumes the exposure is additive. The time delay between lines or layers is ignored as well as the chemical effects (e. when the laser draws multiple lines or layers to form a part. Nagamori and coworkers (2001. and laser scanning speed affect the cured depth and width. and claimed a relationship which is close to the power function between the cured depth and laser scanning speed on a semi-log plot. diffusion. it simply adds all the exposure deposited in the part building process to determine the part dimensions (see section 6. the exposure threshold model is just an oversimplification of the part building process. heat. 2003) performed SL curing tests to investigate how the laser power.3).) during the delay. etc) but the exposure. material change. All 5 . Again. They correlated the cured depth with energy density (exposure) and found a linear relation on the semi-log graph. etc.g. It ignores everything (reaction.

these studies were trying to directly connect the laser exposure to the part dimensions. Onuh and Hon (1998b) added two new hatch styles to their previous work (1998a) and studied the effects of these styles on the dimensional accuracy. and concluded that layer thickness and part orientation have more effects on the part dimensional accuracy. hatch spacing. and layer thickness. hatch spacing. hatch style. hatch overcure. Chockalingam and coworkers (2003) determined the part shrinkage (by comparing the SL finished part dimensions with the part dimensions on the CAD model) for an experimental set designed by genetic algorithm concerning the effects of layer thickness. For example. border overcure. hatch fill cure depth. Onuh and Hon (1998a) used the Taguchi method to design and conduct experiments concerning layer thickness. and hatch fill cure depth. Jayanthi and 6 . hatch style. and hatch fill cure depth. They then performed an optimization and identified an optimal value set of these parameters to obtain parts with the same shrinkage ratio in both depth and width directions. such as hatch spacing. as in the exposure threshold model introduced above. to minimize SL part building error. Schaub and coworkers (1997) identified four key variables that affect the part dimensional accuracy among various control variables in the SL process. A lot of work has been done to investigate the effect of process parameters and optimize the SL process. They analyzed the built results and optimized these building parameters to improve the surface finish of SL parts. Cho and coworkers (2000) also used a genetic algorithm based methodology to determine an optimal value set for the process parameters. They then used design of experiments and the ANOVA technique to analyze and compare the significance of these four parameters. hatch over cure. hatch overcure. but they are all based on the exposure threshold model currently used in industry.

) impact the SL fabrication results. which. resin properties. such as layer thickness. on the SL cure process. and to find the optimum combination of material and process parameters to improve SL resolution and speed. used it to control the SL part building. The ANOVA procedure was utilized to identify significant factors for each writing style. This is a different perspective. A more complete model is needed that accounts for reaction. Eschl and coworkers (1999) tested and simulated the transient post-fabrication shrinkage of SL parts to investigate the effect of the resin material type. Their methodology of studying material effects is based on an investigation of the built results rather than a direct study on the building process.coworkers (1994) performed a study on the influence of process parameters. Flach and Chartoff (1995a. cannot address the curing dynamics or the heating issue in SL building process. on curl distortion of the cured part. to investigate how the chemical effects (e. heat transfer and mass transfer in order to predict the cured shape and size more accurately. Their 7 . acrylate or epoxy.b) incorporated both reaction and heat transfer into an SL process model and simulated the cure process when the laser is stationary and when it moves along one line.g. was not taken into account. cure reaction. Mass transfer. and it was concluded that the hatch writing style yields better results than weave style. They found that the epoxy resin produces more accurate parts because the stress due to shrinkage is smaller and the final stiffness is higher. and fill cure depth. hatch overcure. however. This study was performed for two writing styles: hatch and weave. hatch spacing. etc. and analyzed the finished part property upon the variation of the process parameters. All these studies took the exposure threshold model for granted. however.

Its material properties and photopolymerization kinetics are characterized. Therefore. The SL cure process is simulated and the process modeling is verified experimentally. a tetraacrylate monomer is used for both simulation and part building in SLA. In this study. Furthermore. 2000) further studied the part deformation and the thermal stress formed in the built part when the laser is stationary and moves along one line. However. such as temperature rise in the SLA vat. significant factors which affect each of these responses are identified and optimized. The process model established incorporates both an energy balance and mass balances for multiple species. a systematic study of how the various SL parameters affect the SL process was not performed. Hur and coworkers (1997. However. For several responses that characterize the SL performance. the new model discards the additive exposure assumption used by the current exposure threshold model. They also presented the profiles for monomer conversion and photoinitiator consumption in the curing process. no experimental verification of the model was provided. their work did not directly provide guidance on how to improve the SL process. in addition to suffering from the deficiencies in Flach and Chartoff’s work (1995). and green part degree of cure. their work also ignored the dark polymerization reaction in the case of the laser moving. Furthermore.simulation results predicted that a substantial temperature increase (~90oC) occurs in the resin bath under certain conditions. 8 . HDDA) used in their work does not form well-made solid parts in SLA. the diacrylate monomer (hexanedioldiacrylate. part resolution. Since the chemical reactions are taken into account upon transient irradiation.

the material and process modifications can be made for SL technology improvement. In Chapter 2. and degree of cure (DOC). Therefore. Additionally. It incorporates laser exposure. a complex SL cure process model is established that captures effects that are ignored in the exposure threshold model. and the SL applications which are currently limited by poor prediction of the exposure threshold model can be activated. In Chapter 3. as well as a prediction of the cure behavior upon the variation of material or process parameters.1.3 Project Strategy In this work. polymer chain propagation and termination. species diffusion in the curing polymer network. which are necessary to characterize the cured part. photoinitiation. and heat transfer via conduction in the exposed region and its vicinity. incorporating exposure and dark reaction in one model. rate of polymerization. the sensitivity analysis of material parameters provides a guideline for developing new photosensitive SL resins. It gives a full description of the transient cure behavior of the resin in the SLA bath. the photopolymerization kinetics are characterized using differential photocalorimetry (DPC) and a comprehensive kinetic model is parameterized for a model acrylate resin system. This model investigates during the part building process the spatial and temporal distributions of temperature. Chapter 5 demonstrates the simulation results by solving the process model using the 9 . The thermal and physical properties of the model material are characterized in Chapter 4. the SL cure process model is formulated as a set of coupled partial differential equations describing mass and energy transport during the curing process. a fundamental understanding of the SL process that takes into account the detailed physics and chemistry of the underlying process can be expected.

Chapter 6 verifies the process model through part fabrication and measurement. In Chapter 7. Conclusions and recommendations are made in Chapter 8. significant material and process parameters are identified and optimized for SL resolution and curing speed.finite element method with the software package FEMLAB (Comsol Inc. 10 .).

This leads to a 2-dimensional rectangular domain in Cartesian coordinate (Figure 3) which is used to simulate the resin cure behavior during the single line drawing process. only a half section needs to be modeled. as shown in Figure 2. A 3-dimensional problem is thus reduced to a 2-dimensional one. 11 .CHAPTER 2 MODEL DEVELOPMENT The simplest case of complex laser drawing patterns in SL is that the laser moves along one direction and draws a single vector line. where the x axis is the laser moving direction. For a Gaussian laser and a resin obeying Beer’s law used in this work. Furthermore. 1992). Considering the repetitive cure behavior along the x-axis (the very ends of the line which may receive different amount of exposure are not of interest here). only the cross section of the parabolic cylinder needs to be modeled. The heat and mass transfer along x direction can be ignored due to infinitely small behavior difference between neighboring planes (cutting the parabolic cylinder into infinite number of parabolic planes) as well as low thermal conductivity and diffusion coefficients of the curing system. the cured shape upon a single laser drawn line is a parabolic cylinder (Jacobs. since the cross section is symmetric about z axis.

1995a).Figure 2 Cured Shape of Single Laser Drawn Line Figure 3 2D Domain for Single Laser Drawn Line The shaded region in Figure 3. which corresponds to the half cross section of the parabolic cylinder in Figure 2. The size of this region increases with time as heat conduction and/or molecular diffusion continues (Flach and Chartoff. is where most of the reaction occurs and the material properties vary significantly. 12 . The domain is chosen to be large enough to ensure ambient temperature and concentrations outside the rectangle at any time.

As shown in Equation (3) the heat generated by steps other than propagation is assumed to be negligible. and Rt are the rate of initiation. R P . respectively. polymeric radicals (including monomer radicals). The heating effect of the laser (325nm wavelength) is 13 . The photopolymerization mechanism and kinetics will be addressed in Chapter 3. ρC P  ∂ 2T ∂ 2T ∂ 2T  ∂T = k 2 + +  + ∆H P R P ∂t ∂ y2 ∂ z2  ∂ x (3)  ∂ 2 [M ] ∂ 2 [M ] ∂ 2 [M ]  ∂[ M ] + + = DM   + (− R P ) 2 ∂t ∂ z2  ∂ y2  ∂x (4)  ∂ 2 [ P•] ∂ 2 [ P•] ∂ 2 [ P•]  ∂[ P•] = DP•  + +  + ( Ri − Rt ) 2 ∂t ∂ y2 ∂ z2   ∂x (5)  ∂ 2S ∂ 2S ∂ 2S  ∂S = DS  2 + +  + (− Ri / φi ) ∂t ∂ y2 ∂ z2  ∂ x (6) These equations are coupled with one another through the reaction terms as source(s) or sink(s) and have to be solved simultaneously. Equation (3) is the energy balance of the curing system. propagation. Ri . and photoinitiator.Mass transfer by diffusion and heat transfer by conduction are the two transport phenomena occurring in the SL cure process. Equations (4)-(6) describe mass balances for monomer. φi is the quantum yield of initiation. and termination. respectively.

As the laser moves from -∞ to +∞ along x direction. it is beneficial to develop a time-varying description of I which integrates the two periods that any point would go through (irradiation and dark) into one equation  Equation (7). It can be safely ignored when compared with the large amount of heat generated by reaction (~105 J/mol). Attention should be paid when the assumption is made that the propagation and termination only occurs in the dark. Since Ri is proportional to the irradiance I. I increases to a maximum (the investigated point is directly irradiated) and then decreases till zero (the beam moves away from the investigated point). Although the exposure time for the resin is very short in SL (~20ms in this study). in Equation (5) the source/sink term can not be limited to Rt which only describes the radical reaction in the dark. A time parameter t 0 is introduced into the equation so that when time t goes from 0 to +∞.negligible (~101 J/mol or less) due to the low absorption of the curing resin (except the photoinitiator) and very short exposure time. the convection term should also be incorporated into Equations (3)-(6). The later simulation results demonstrate that significant reactions and material property variations occur during this 20ms. any point (x. Only diffusion and heat conduction phenomena are considered here due to the minor difference (within 6%) between the density of liquid monomer and cured polymer. The radical initiation rate Ri has to be incorporated in order to take the exposure reaction into account.0) only receives limited time of exposure. this assumption is not valid for the photosensitive material system studied in this work. To take shrinkage effect into account. which 14 . Therefore.y.

− 2Cd ≤ z ≤ 0. t ≥ 0 at y = 5wo . The domain size is initially set based on 15 . t ≥ 0 at z = −2Cd . The initial and boundary conditions corresponding to this 2D problem are established as follows: [Q] = [Q]i ∂Q =0 ∂y [Q] = [Q]0 ∂Q =0 ∂z [Q] = [Q]0 at t = 0. Neglecting the insignificant property variations along the laser scanning direction (x axis) (Figure 2). their initial values are equivalent to their boundary values [Q]i = [Q]0 . the terms containing x variations can be removed from Equations (3)(6). λ(nm) is the laser wavelength.in SL is defined as characteristic exposure time and expressed as t e = 4. t ≥ 0 at z = 0 . 0 ≤ y ≤ 5wo . 0 ≤ y ≤ 5wo . and the last quotient term is adopted to convert the unit of intensity from W/m2 to mol/m2-s. −2Cd ≤ z ≤ 0. I0 (W/m2) is the maximum intensity incident at the resin surface. 1992). −2Cd ≤ z ≤ 0 (a) at y = 0 . or photoinitiator concentration S.3wo / Vs (Jacobs. Any value greater than half t e can be used as t 0 . monomer concentration [M]. 1992). w0 is the laser beam radius.196 × 10 8 (7) where y and z axes are as shown in Figure 2. t ≥ 0 (b) (c) (d) (e) (8) where Q represents temperature T. polymeric radical concentration [P•]. Cd is cure depth. 0 ≤ y ≤ 5wo . I = I 0 exp − 2 (Vs (t − t 0 ) ) + y 2 / wo exp(− z / D p ) 2 2 { [ ] } λ (nm) 1. the maximum depth of the solidified area (Jacobs.

to accommodate the transient For the temperature condition at z=0 boundary. the reaction dependent source/sink terms need to be defined. 16 . The later simulation results show that the heat convection at the resin surface in the SLA chamber doesn’t have a noticeable effect on the part building results. 0 ≤ y ≤ 5wo. The kinetic model and its parameterization are detailed in Chapter 3. In order to solve the governing equations (3)-(6). respectively. heat transfer with the natural air environment in the SLA chamber can be incorporated by replacing the temperature condition (8d) with the following: k ∂T = h (Tinf − T ) at z = 0.the values of Cd and w0 and adjusted accordingly variations of the simulated properties. t ≥ 0 ∂z (9) where k is thermal conductivity of the curing resin system. h and Tinf are the air-resin heat transfer coefficient and ambient temperature in the SLA chamber.

and the kinetic model is parameterized and validated for a photosensitive acrylate system. 3. all source/sink terms in the balance equations are related to the resin cure kinetics. the kinetic experiments are designed. the kinetic coefficient data are extracted without using steady state assumption. Ignoring chain transfer reactions.1 Photopolymerization Kinetic Model As discussed in Chapter 2. the photocure mechanism for acrylate resin can be briefly described as follows:    → P1 • M + R•  ki hν PI → R • Initiation (10) Pn • + M kP Pn+1 • → k tc Pn • + Pm • → M n + m Propagation Termination by Combination Termination by Disproportionation Inhibition ktd Pn • + Pm • → M n + M m kin R • + In → Q 17 . a kinetic model of photoinitiated free radical polymerization is described. a curve fitting method is utilized to analyze the kinetic data.CHAPTER 3 KINETIC CHARACTERIZATION In this chapter.

where PI. Crivello. Pn• the polymeric radical with a chain length of n monomer units.z) (Figure 4) can be derived as in Equation (14). respectively.y. and In represent the photoinitiator. [ P •] and [ M ] are the polymeric radical concentration and monomer concentration. the expression of the absorbed intensity at any point Q(x. respectively. propagation and termination are expressed as Equations (11). and ε (m3/mol-m) and S (mol/m3) are the absorptivity and concentration of the initiator. y. z ) = I ( x.∆z → 0 ∆z → 0 (14) 4 (Fouassier. z )∆x∆y − I ′( x. I a ( x. y . R• is the primary radical. I is the intensity incident on the resin surface (mol/m2-s). z ) = lim lim ∆z ∆x∆y∆z ∆x. For a resin obeying Beer’s Law. and Mn the stable polymer molecule with a chain length of n monomer units. y.3εSI ( x. z ) ∆x∆y I ( x. monomer. (12). and I a is the absorbed light intensity or rate of absorption (mol/m3-s). y. M. 1995. the rates of initiation. and (13). y. and inhibitor.3εS∆z ) = 2. z )(1 − e −2. Correspondingly. Ri = φi I a (11)4 R p = k p [ P •][ M ] (12) Rt = k t [ P•] 2 (13) where k p and k t are the rate constants for propagation and termination. 1998) 18 . ∆y.

z ) .3φi εSI (16) The decay of the photoinitiator can be approximated as: 19 . for an irradiance at point ( x. y. y. z + 1cm) rather than I a ( x. Equation (14).Figure 4 Absorbed Intensity at Point Q(x. z ) : I ( x. eliminates this spatial inconsistency.e. i. y. y.z) Fouassier (1995) has claimed that the absorbed light intensity can be expressed as: I a = I (1 − e −2. however. The rate of initiation thus can be rewritten as: Ri = 2. z ) is evaluated.3ε S l ) (15) where l is taken as 1cm in order for I a to have the unit of photons/cm3-s.y. I a ( x. This always gives the absorbed intensity at a point 1cm lower than where the light is incident on.

During the polymerization. k p and k t . At the same temperature.3εSI dt (17) The temperature dependence of the kinetic constants. a rate decrease is observed in both propagation and termination reactions. and R is the gas constant. the values of k p and k t are expected to be larger in an environment with more free volume and less diffusion limitation. 1982) related the kinetic constants k p and k t directly to 20 . however. E p and Et are activation energies for propagation and termination. the curing system becomes more viscous. this is not what happens throughout the reaction. Marten and Hamielec (1979. Another reason for this is the decrease of the rate constants themselves. respectively. the free volume decreases.dS = − I a = −2. The rate constants are not only dependent on temperature but on the free volume of the reacting system. Due partially to the consumption of monomers and radicals. With the polymerization going on. the reaction is expected to accelerate due to the temperature rise caused by the heat of reaction. The reaction becomes diffusion controlled. The reaction is faster at higher temperature. is assumed to follow the Arrhenius form. and the mobility of the reacting species is reduced. k p = AEp e − E p / RT (18) k t = AEt e − Et / RT (19) where AEp and AEt are pre-exponential factors.

the diffusion coefficients of monomer and polymer radicals, respectively, with a temperature dependent proportionality constant. They assumed distinct regions exist for reaction and diffusion controlled polymerization, and divided the course of reaction into three conversion intervals to evaluate k p and k t . Bowman and Peppas (1991) adopted the same idea and coupled these intervals with volume relaxation during polymerization. These models don’t take any transition region into account and the parameters have to switch to different values in different conversion ranges, i.e. each stage in the polymerization has to be treated separately. This problem has been solved by combining the reaction-controlled rate constants for propagation and termination and the diffusioncontrolled mechanisms to incorporate the transition regions for both kp and kt. The rate constants kp and kt are expressed in terms of reaction resistances (Anseth and Bowman, 1993).

1 1 1 = + k p or k t k r k D

(20)

where k r is the reaction limited (“true”) kinetic constant, and k D is the diffusion limited kinetic constant. For propagation, the resistances to reaction simply come from the reaction itself and the monomer diffusion. For termination, the diffusion resistance is not only from the translational and segmental diffusion of polymer radicals (
1 , translational diffusion is kD

21

negligible for highly crosslinked chains), but from the reaction diffusion ( the segmental diffusion resistance).

1 , parallel to k r*

1 1 1 = + * kt kr k r + k D

(21)

According to Buback et al. (1989) and Buback (1990), the concept of reaction diffusion has been put forward by Schulz (1956) and has been refined and put into quantitative terms by several groups. The reaction diffusion is inherently a propagation step – the “frozen” polymer radical propagates via the reactive monomer matrix until encountering a second macroradical, which is also called “residual termination”. The rate coefficient of this process, k r* , is proportional to k p and to monomer concentration [M]. The proportionality constant Rrd (called reaction diffusion parameter) is independent of temperature, pressure, and conversion (Buback et al., 1989; Buback, 1990).

k r* = Rrd k p [ M ]

(22)

Anseth and Bowman (1993) assumed the diffusion limited kinetic constant k D to be proportional to the diffusion coefficient of the reacting species and modeled it using the Doolittle equation (Bueche, 1962). Equations (18) and (19), the expressions of k p and k t without diffusion consideration, define the true kinetic constant k r for propagation and the true kinetic constant k r for termination, respectively. Substituting all the above

22

information into Equations (20) and (21), the dependencies of the rate constant ( k p or k t ) on both temperature and fractional free volume are incorporated into one equation (Goodner et al., 1997, 1998, 1999, and 2002) which describes the rate of propagation or termination throughout the whole polymerization course without changing parameter values.

kp =

k p0 1+ e
A p (1 / f −1 / f cp )

(23)

kt =

kt 0 1 1+ Rrd k p [ M ] / k t 0 + e − At (1 / f −1 / f ct )

(24)

with

k p 0 = AEp e

− E p / RT

(25)

k t 0 = AEt e − Et / RT

(26)

where k p 0 and k t 0 are the true kinetic constants for propagation and termination, respectively, f is the fractional free volume of the curing system, f cp and f ct are critical fractional free volumes for propagation and termination, respectively, and A p and At are parameters that determine the rate at which the propagation and termination rate constants decrease in the diffusion-controlled region (Goodner et al., 1997, 2002). When the free volume of the polymerization system is much larger than the critical free volume,

23

TgP ) (29) φM = 1.025 + α M (T . glass transition temperatures..there is no diffusion limitation on propagation or termination. Tg’s. respectively.φ M ) (27) f M = 0. the fractional free volume f is related to monomer conversion X as follows (Goodner et al. and ρ ’s are the volumetric coefficients of expansion. 1997. of pure monomer and pure polymer. φ M is the volume fraction of monomer.TgM ) (28) f P = 0. f M and f P are the fractional free volumes of pure monomer and pure polymer. Equation (23) or (24) is reduced to Equation (18) or (19).025 + α P (T . The diffusion resistances have to be incorporated in the kinetic constants as in Equations (23) and (24). The free volume decreases with the curing reaction going on. the reaction (propagation or termination) becomes diffusion limited.X ρ 1. The free volume of the polymerization system is dependent on both temperature and composition (conversion). and densities. For a curing system comprised of pure monomer and pure polymer.X + M X ρP (30) In the above equations. 24 . 2002): f = f M φ M + f P (1 . When it decreases to be smaller than the critical free volume. and the α ’s.

3. The process & laser parameters can be recorded during the SL part building (as shown in Chapter 5). 25 . The determination of material properties will be addressed in Chapter 4. all the parameters (except the kinetic ones) involved are listed in Table 1.2 and 3. The kinetic experiment has been conducted and the kinetic model for a model material system parameterized in sections 3.Goodner and Bowman (2002) also described the critical fractional free volume for propagation or termination as a function of temperature: 1 1 E 1 1  = ref +  − ref  fc AR  T T  fc (31) From the kinetic model described above and the SL process model established in Chapter 2. respectively.

Table 1 Material & Process Parameters Involved in the SL Cure Process Model Parameters Process Parameters laser scanning velocity bath temperature thermal convection coefficient chamber temperature Laser Parameters laser power wavelength beam radius thermal conductivity heat of polymerization absorptivity (initiator) initiation quantum yield diffusion coefficient (monomer) diffusion coefficient (radical) diffusion coefficient (initiator) coefficient of thermal expansion (monomer) coefficient of thermal expansion (polymer) glass transition temperature (monomer) glass transition temperature (polymer) heat capacity (monomer) heat capacity (polymer) density (monomer) density (polymer) Resin Compositions monomer concentration initiator concentration Symbols Units Vs Tb h Ta PL λ m/s K W/m2K K W nm m W/m-K J/mol m3/molm m2/s m2/s m2/s 1/K 1/K K K J/kg-K J/kg-K kg/m3 kg/m3 mol/m3 mol/m3 Material Properties wo k ∆ΗP ε φi DM DP· DS αΜ αP TgM TgP CPM CPP ρΜ ρP [M] [S] 26 .

Pang et al. Sartomer) was chosen as the model compound. Melisaris et al.. 1999. Representative of the acrylate compounds commonly used in SL (Steinmann et al. 1995. The inhibitor was removed from the received E4PETeA by a prepacked inhibitor remover column (Aldrich).2 Kinetic Experiments In order to simulate the polymerization behavior in the SLA bath. 2000. Ethoxylated (4) PentaErythritol TetraAcrylate (E4PETeA..3. 2000).2-dimethoxy-2phenylacetophenone (DMPA. Irgacure®651. Ciba) was selected as its initiator. Figure 5 Structure Formula of E4PETeA (Sartomer) Figure 6 Structure Formula of DMPA (Ciba) 27 . The 2.. SR®494. a model material system is identified and its kinetics characterized. 0.2wt% DMPA was added in the acrylate.

1 DPC Pan The differential photocalorimetry (DPC) technique was used to monitor the photopolymerization kinetics. The depth of the depression was proved small enough for heat to dissipate quickly so that the temperature uniformity through the sample can be assured. When the light is incident on the sample surface. 1985). the reflectivity of which is 0. Figure 7 DPC Sample Pan 3.2 Standing Waves To ensure uniform reaction occurring in DPC so that diffusion and heat conduction can be ignored in the mass and energy balance. 1979). The aluminum pans were machined specially to have a 0. Therefore. The sample size was determined accordingly to fill the depression.. where the 28 . the uniformity of light intensity through the sample thickness also has to be assured.2. Considering the reflectivity of the aluminum pan. the transmitted light travels through the sample thickness and strikes the aluminum substrate. the light absorption of the photosensitive sample is not the only factor that affects the intensity uniformity.9642 (Bass et al.2. by which the thickness uniformity is ensured (Tryson and Shultz. most of light is reflected off the aluminum and travels through the sample again to the top surface.3.15mm depression to hold the sample.

According to the equation by Mack (1985. depending mainly on the absorptivity and thickness of the photosensitive material. 1986). This wavelength range is where DMPA absorbs (Chapter 4) and the light source irradiates (Appendix A). the electric field of the light in the photosensitive sample on a reflective substrate can be calculated. The intensity of the standing wave is the intensity that exposes the sample. and 1994) addressed in detail how the standing wave is formed in a thin film of absorbing material coated on a reflective substrate and described its intensity quantitatively. Figure 8 shows the variation of the intensity at 365nm wavelength through the sample thickness. Figure 9 sums the intensities at individual wavelengths from 304 to 395nm. 1986. and 1994). 1986. Figures 8 and 9 demonstrate the standing wave intensity through a 150µm thick sample (E4PETeA with 0.reflection by the air-sample interface makes the light travel down into the sample and then reflected by the aluminum.2wt% DMPA) contained in the aluminum pan. This process continues until the light wave dies out due primarily to absorption. Mack (1985. The intensity that exposes the sample can be obtained by squaring the magnitude of the electric field (Mack. which takes not only the absorption but reflection into account. The standing wave intensity outside the material absorption range is not of interest since it doesn’t contribute to the polymerization initiation. 29 . The waves traveling in opposite directions form a standing wave in the sample. The magnitude of the standing wave intensity can be different from material to material.

If the pan bottom doesn’t reflect. From Figure 9. according to the Beer’s Law. the intensity of 0. the light intensity varies with depth into the sample.15I0 through the sample thickness. Assuming reciprocity. In this case the top surface 30 . it can be seen that for the investigated curing system the average of the light intensities (304395nm) over depth is 1.15I0 (I0 is the intensity incident at the sample top surface).Figure 8 Standing Wave Intensity at 365nm Figure 9 Standing Wave Intensity at 304-395nm Obviously.93I0 would be expected at the bottom surface (150µm depth). the cure result under the standing wave intensity (Figure 9) should be the same as that under uniform intensity 1.

The average intensity the resin receives was found to be 15% (not 96. one dose from the incident beam and a second dose from the reflected beam by the aluminum. Most published work involving DPC experiments took this assumption. the intensity of which as discussed above is more than 10% different from the top surface intensity I0. causes standing wave formed in the sample. These two light waves interfere with each other. Furthermore.9642). the electric field being the sum of them. The intensity is the square of the magnitude of the electric field for plane waves (Mack.9642)I0 for the intensity that exposes the sample in the pan (Recall the aluminum reflectance is 0. ignoring the aluminum reflection and sample absorption. 31 . The aluminum reflection as well as the reflection from the sample-air interface. however. 1994).42%) higher than the top surface intensity. is not a simple sum. rather than one single reflected wave. but they assumed the sample only received two doses. as described earlier.intensity I0 can be taken as the intensity the sample receives through the thickness (negligible light attenuation). there are an indefinite number of reflected waves bouncing up and down in the sample and a standing wave is formed. The consideration of aluminum reflection by simply adding intensities of two doses together deviated even further away from the intensity the sample actually receives than the adoption of top surface intensity without reflection consideration. and came up with (1+0. The intensity. however. Anseth (1994a) and Mateo (1997) and their coworkers took the reflection of the aluminum pan into account.

06mW. The model material system developed here can also be thermally initiated and polymerized. used with laser power meter EPM 2000e. The incident power was adjusted and measured to be 0. illustrates that the polymerization can be thermally initiated at and above this temperature.15mm deep depression. Note that the heat flow oscillation in the initial stage represents the temperature overshooting and equilibration behaviour.16(±0. The differential scanning calorimeter DSC Q1000 with photo calorimetric accessory (TA Instruments) was adopted to monitor the photopolymerization of the model acrylate resin.3 DPC Experiments 1. a custom mount for the power probe (PM3.05)mg sample was put in the aluminum pan using a micropipette to cover the 0. A set of isothermal DSC experiments demonstrate that the material won’t be initiated thermally below 140oC (Figure 10). In order for the intensity measured to be exactly the intensity incident on the sample surface in the actual experiment. indicating the heat of polymerization. The light source Novacure 2100 (EXFO Photonic Solutions) was used with filtered wavelength at 365nm (Appendix A). 32 . DPC experiments should be performed below this temperature to avoid thermal polymerization. The big exothermal peak in 140oC DSC curve. Molectron) and twin light guides of the light source was designed and machined to simulate the DSC cell environment.2.3. The absence of this peak in 130oC DSC curve shows that the material system won’t polymerize at or below 130oC.

A typical set of experimental data is shown in Figure 11. limited by the liquid light guide requirement on temperature. the reaction is complete under the current temperature.Figure 10 Isothermal DSC Runs to Detect the Onset Temperature of Thermal Cure In this study. The light is on only for a very short time in flash exposure experiments. During the continuous irradiation experiment. For each temperature. the DPC experiments cannot be conducted above 70oC. 50. i. 33 . five or more different flash times were used in order to extract the kinetic constants at different conversions. Both continuous and flash exposure experiments were carried out isothermally at three different temperatures (30. the light is on until the heat flow curve drops to the baseline.e. 70oC).

no noticeable change in heat flow curve is observed when the light is turned off.001 50C(flash0. 1997).. A quantitative estimation also shows that the heating by light is insignificant (~100 J/mol for continuous irradiation and 10-1 J/mol for very short time exposure) comparing with the heat generated by reaction (~105 J/mol for continuous irradiation and 104 J/mol for very short time exposure). 34 .6m) 50C(flash0.4m) 50C(flash0.8m) Heat Flow (W/g) 6 4 2 0 -2 Exo Up 0 1 2 3 4 5 Time (min) Figure 11 DPC Experimental Curves (Continuous and Flash Exposure at 50oC) The heat flow curves are only contributed by the reaction generated heat.2m) 50C(flash0. after the reaction is completed.10 8 ––––––– –––– ––––– · ––– – – ––– ––– 50C(cont irrad). As shown in Appendix B. This demonstrates that the heating effect of the DPC light source is negligible (Lecamp et al.

1999. 1992. Qtot is in principle the total heat of reaction when dt all the monomers are converted.3 Kinetic Data Analysis & Model Parameterization Assuming that the heat produced by the reaction is proportional to the amount of monomer reacted (Burel et al. Lecamp et al.3. and Maffezzoli and Terzi. the degree of cure. X. 1999. the rate of propagation normalized by the initial monomer concentration is expressed as follows (the monomer consumption by initiation. is negligible compared with the consumption during propagation): RN = RP dX dQ 1 = = × dt dt Qtot [ M ]0 (33) where Q(t ) is the heat developed at any time t during a DPC measurement. 1995.. 35 .. which is the integration of the heat flow signal dQ . in DPC experiments can be defined as follows: X= Q(t ) Qtot (32) Correspondingly. ignored in Equation 33. 2001). Qtot is approximately the heat of polymerization measured at high enough temperature and light intensity. 1997. 1993. Cook.

At each temperature (e.g. 36 . 30. d [ P•] = −k t [ P•] 2 dt (34) Integrating the above equation and combining it with Equation (12). The ending time t2 is taken well before the reaction dies out. and 70oC). (1 − X ) / RN ) as a function of time t (Tryson and Shultz. 0.g. from a series of flash exposure experiments (e.2.6.For the dark reaction in the flash exposure experiment. 0.4. 50. RP = k P [ P•][ M ] . the ratio k t / k P can be determined from the slope of the plot of [ M ] / RP (i. 1979).e. leads to: k [M ] [M ] t 2 = t (t 2 − t1 ) + t1 RP kP RP (35) Therefore.8 min irradiation) the value of k t / k P at different conversions can be extracted. and 0. as shown in Table 2. The starting time t1 is a time point after the light is turned off. 0.

06 0.96 11.342 0.4 0.3 12.12 41.2 0.3φi εSI .37 0.6 0.6 0.18 0.150 0. Ri = 2.4 26.37 0. to obtain: 1 + y2 1 + y1 ln( ) − ln( ) = 2( Ri k t )1 / 2 (t 2 − t1 ) 1 − y2 1 − y1 (37) with y1.8 0.491 0.6 0.493 0.404 0.424 0.09 For the continuous irradiation.5 kp/kt^0.2 0.09 0.2 15. 2 (38) where Ri is evaluated using Equation (16). 2 = kt 1/ 2 1/ 2 k P Ri ( RN )t 1 .14 0.13 0.2 0.8 X 0.567 1/ 2 Values Obtained from DPC Experiments Continuous Irradiation kp/kt^0.48 7.163 0.43 10.34 0.5 13.2 12.16 0.3 50 70 dark reaction kt/kp 111.2 0.08 0.4 0.455 0.X 1.339 0.8 0.72 19.5 27.4 0.38 0.17 0.5 (QSSA) 0.94 182.25 21. d [ P•] = Ri − k t [ P•] 2 dt (36) Integrating and combining it with Equation (12).09 QSSA Effect Difference (%) 0.378 0.12 0.12 0.277 0. and Equation (17).08 0.9 2.07 0.2 21.14 0.39 0.12 0.78 18.5 21.66 20.92 8.46 14.18 0.19 0.2 10.15 0.34 0. RP = k P [ P•][ M ] . 37 .Table 2 k t / k P and k P / k t T(oC) 30 Flash Exposure Time (min) 0.

15 0.4 10.0 11. For each temperature. The values of k P and k t can thus be determined separately at several different conversions for each temperature (as shown in Table 3).38 0.15 0.2 0.17 0.57 kt/kp 111.16 0.07 kt(m3/mol-s) 1694 12 6 2 3860 11 3 0.6 0.49 0.3εSI .6 0.08 0.12 0.16 0.dS = −2.8 18. a trial and error analysis is performed using k t / k P data derived from the dark reaction to evaluate k P / k t 1/ 2 values at corresponding conversions.3 0.34 0. i.45 0. λ1 ∫ ε (λ ) I (λ )dλ is substituted for ε I in Equations (16) and (17).40 0. Table 3 Determination of Rate Constants k t and k P T(oC) 30 X 0.1 42.34 0.7 19.04 50 70 38 .7 20. Since both photoinitiator absorptivity and light intensity are a function of dt λ2 wavelength.3 21.3 kp(m3/mol-s) 15 0.5 7.1 21 0.49 0. from which the free volume and temperature dependence of k P and k t . Table 2 lists the ratio k t / k P obtained from flash exposure experiments and the ratio k P / kt 1/ 2 obtained from continuous irradiation experiments. Equation 1/ 2 (37) combines two kinetic constants together in terms of k P / k t . the parameters in Equations (23) and (24).9 182.2 0.38 0.9 8.5 14.34 0.e.28 0.1 kp/kt^0. can be determined.12 0.3 0.06 0.42 0.5 0.2 254 3 2 0.37 0.12 0.03 6 0.

For both HEMA and E4PETeA. the rate constants k t and k P drop dramatically with the conversion increasing. The drop is more significant for tetraacrylate due to the crosslinkage. For both HEMA and E4PETeA. Goodner and coworkers (1997) found that at similar conversion. 39 . the magnitude of k t is three orders higher than k P for 2-hydroxyethyl methacrylate (HEMA). k t drops at a lower conversion than k P . The scientific plotting and data analysis software Origin (OriginLab Corporation) is used for the nonlinear curve fitting of k P and k t data to determine the free volume dependence. It’s possible that the steric resistance in tetra-functional material studied here affects more significantly the true kinetic constants for termination than for propagation. as shown in Figures 12 and 13. Figure 15 illustrates the temperature dependence of critical fractional free volume (Equation 31) for propagation and termination. as demonstrated in Figure 14. the magnitude of k t is two order higher than k P . Linear curve fitting in Excel (Microsoft Corporation) is performed to determine the temperature dependence of true kinetic constants k P 0 and k t 0 (Equations 25-26).As shown in Table 3. at low conversion.

Conversion X (50oC) 40 . Conversion X (50oC) 4500 4000 3500 3000 kt (m /mol-s) 2500 2000 1500 1000 500 0 -500 0.6 3 Conversion X Figure 13 Nonlinear Fit of Termination Rate Constant k t vs.6 3 Conversion X Figure 12 Nonlinear Fit of Propagation Rate Constant k P vs.0 0.0 0.16 14 12 10 kp (m /mol-s) 8 6 4 2 0 -2 0.2 0.2 0.4 0.4 0.

1/T 41 . 1/T Figure 15 Linear Fit of 1 / f c (Critical Fractional Free Volume) vs.Figure 14 Semi-log Plot of True Kinetic Constants k P 0 and k t 0 vs.

The values of the kinetic parameters obtained are listed in Table 4. 1997). Accordingly. et al. however... The linear CTE of the tetraacrylate system (see Chapter 4). the critical fractional free volume for propagation and termination of E4PETeA (Figure 15) turns out to be one order of magnitude higher than that of HEMA (Goodner.4 0. is within the same range of the CTE used for the HEMA system (Goodner.The fractional free volume of the tetraacrylate system is found to be one order of magnitude higher than that of HEMA (Goodner.013 28. people usually apply the quasi steady state assumption (QSSA) and Equation (36) is then reduced to: Ri = k t [ P•]2 42 (39) .1 6.. et al. 1997). Table 4 Kinetic Parameter Values Parameters free volume parameter for propagation (see Equation 23) free volume parameter for termination (see Equation 24) reaction diffusion parameter (see Equation 24) pre-exponential factor for propagation (see Equation 25) pre-exponential factor for termination (see Equation (26) activation energy for propagation (see Equation 25) activation energy for termination (see Equation 26) Symbols Values Units N/A N/A m3/mol m3/mol-s m3/mol-s J/mol J/mol Ap At Rrd AEp AEt Ep Et 6. et al. 1997) because the CTE of the system is one order of magnitude higher than that of HEMA.4 8916 1627 2103 It should be mentioned that instead of integrating Equation (36) to obtain the relationship of k P and k t .

it has more than 10% deviation when the conversion is greater than 30%. a unique parameter set can be expected. The advantages of this regional analysis method are that only continuous irradiation experiments need to be conducted (flash exposure experiments are not required) and unlike the nonlinear curve fitting. the ratio k P / kt 1/ 2 can be directly calculated as follows: kP kt 1/ 2 = RN (1 − X ) Ri (40) Table 2 also lists the values of k P / k t 1/ 2 determined with steady state assumption and their comparison with the values obtained without this assumption. The second region.Substituting the above equation into Equation (12). in this specific case. However. Goodner and coworkers (1997) have divided the polymerization process under continuous light irradiation into four regions based on the free volume and treated the four regions individually to determine the kinetic parameters. for a highly crosslinked polymerization system as investigated here. To analyze the kinetic data and parameterize the kinetic model. This limits the application of the regional analysis method which requires 43 . autodeceleration. is defined well. It turns out that QSSA is valid only at low conversions. the first region where there are no diffusional limitations on either propagation or termination and the third region where there are no diffusional limitations on propagation but termination is reaction-diffusion controlled are often found to be illdefined. RP = k P [ P•][ M ] . autoacceleration. Often only the fourth region. could also be ill-defined or just have not enough data to determine parameters.

the regional analysis has lost one of its attractive characteristics mentioned above. 3. however. The agreement between the predicted and experimental results validates the adopted kinetic model and the determined kinetic parameters. the flash exposure experiment (so-called unsteady state analysis) has been proposed to find k t 0 (or k P 0 ) for systems that don’t have a well-defined first (or third) region (Goodner et al. Other restrictions should be applied for the fitted parameters in order to obtain a reasonable and unique parameter set. but should be physically reasonable as well as capable of predicting kinetic behavior. 1997). the regional analysis method is established based on the QSSA assumption throughout the reaction. flash exposure experiments also have to be conducted and analyzed to find k P 0 or k t 0 to complement the regional analysis.1 mW. Furthermore. In addition to the continuous irradiation experiments.the system have four distinct kinetic regions. The fitted parameters should not only achieve a best fit for the kinetic data. Figure 16 demonstrates the kinetic model simulations for a series of continuous irradiation experiments conducted at different temperatures with the light power of 0.4 Kinetic Model Validation It’s not easy to find a unique solution for the nonlinear curve fitting. 44 . which is not valid as shown in Table 2. In this case. To broaden its application..

0.025 0.02

dX/dt (1/s)

0.015 0.01 0.005 0 0 0.2 0.4 0.6 0.8 1 conversion X

predicted experimental

(a)
0.025 0.02 dt/dX (1/s) 0.015 0.01 0.005 0 0 0.2 0.4 0.6 0.8 1 conversion X predicted experimental

(b)
0.025 0.02
dX/dt (1/s)

0.015 0.01 0.005 0 0 0.2 0.4 0.6 0.8 1 conversion X

predicted experimental

(c) Figure 16 Comparison of the Experimental and Simulated Polymerization Rate Curves (incident power = 0.1 mW): (a) 30oC, (b) 50oC, (c) 70oC

45

CHAPTER 4 MATERIAL CHARACTERIZATION

As listed in Table 1, some process and material parameters need to be determined in order to simulate the SL cure process. The process and laser parameters (such as laser scanning speed Vs, bath temperature Tb, laser power PL, and laser beam radius wo) are obtained from the actual part building process in SLA-250/50 (3D systems, laser wavelength λ = 325nm5). h = 4.18 W/m2-K is taken as the value of heat convection coefficient at the interface of the natural air flow and the resin (Pananakis & Watts, 2000). φi = 0.6 is taken as the quantum efficiency of initiation for DMPA (Goodner et al., 2002). The thermal and physical properties of the resin are evaluated for the model compound system comprised of E4PETeA tetraacrylate and 2 wt% photoinitiator DMPA. The initiator concentration is higher than in Chapter 3 to facilitate the SL cure process. They are obtained from literature, theoretical approximation, or experimental determination.

4.1

Specific Heat Capacity

The modulated differential scanning calorimeter (MDSC) option for DSC 2920 (TA Instruments) was used to measure the specific heat capacity of the pure monomer and pure polymer. MDSC provides unique capabilities besides those of standard DSC such as separation of complex transitions, detection of weak transitions, accurate measurement of

5

SLA Systems Specifications, 3D Systems, http://www.3dsystems.com

46

polymer crystallinity, and direct determination of heat capacity and thermal conductivity.6 In addition to the standard DSC cell calibration (performed in calibration mode) for cell constant, baseline and temperature, a heat capacity calibration (performed in modulated mode) was also performed in order to obtain accurate heat capacity measurement. The sapphire standard (provided by TA Instruments) was used as the calibrant. The calibration procedure should be as close as possible to that of the following measurements. The weights of two pans with lids were matched to within ±0.1mg. One pair of pan and lid were sealed and used as reference; the other pair was used to hold the weighed calibrant or sample and then sealed and placed in the sample position in the DSC cell. The method which tells the machine what to execute was formulated by combining the recommendations on heat capacity calibration and on heat capacity measurements as well as general MDSC operating parameters.6 The liquid nitrogen cooling accessory (LNCA) was used for the optimum performance of the MDSC measurement. The nitrogen gas was used to purge and circulate in the DSC cell at a rate of 40ml/min before and during the experiment. A data sampling interval of 1.0 seconds/point was used. The MDSC directly measures the heat capacity and stores the signal. Figures 17 and 18 are exported plots of heat capacity signal for liquid E4PETeA monomer and its cured polymer, respectively. Samples were weighed 12.09±0.03 mg using an analytical balance (AG 245, Mettler Toledo).

‘Modulated DSCTM Option’, DSC 2920 Differential Scanning Calorimeter Operator’s Manual, Thermal Analysis & Rheology, TA Instruments, 1995.

6

47

Figure 17 Cp-T Plot of Liquid E4PETeA Monomer Exported from MDSC Data Figure 18 Cp-T Plot of Cured E4PETeA Polymer Exported from MDSC Data 48 .

The heat capacities were found to be functions of temperature as follows: C P . P = 9.6 × T ( K ) + 218. which justifies the experimental measurement. Furthermore. M = 5.6 (41) CP ..8 J/g-K for specific heat capacity) at 25 oC by the addition of group contributions (Van Krevelen. 49 . This calculated result is within 5% of the experimental value at the same temperature. P are the heat capacities (J/Kg-K) of monomer and cured polymer. i.. the heat capacity value of E4PETeA is close to those of other acrylates such as methyl and butyl acrylates.1× T ( K ) − 1535. 1990). M (1 − X ) + C P . P X (43) where X is monomer conversion. The molar heat capacity of liquid E4PETeA monomer was also calculated to be 947 J/mol-K (i. M and C P . etc (Yaws. 1.5 (42) where C P . A weight-averaged heat capacity was used for the curing material. mixture of monomer and cured polymer: C P = C P . 2003).e. respectively.e.

respectively. Figures 21 and 22 illustrate the effect of heating rate on the Tg measurement. A lower heating rate leads to gradual material change and thus less obvious glass transition.4.2 Glass Transition Temperature The glass transition temperatures of liquid E4PETeA monomer and its cured polymer are determined using a standard differential scanning calorimeter (DSC 2920. In the range of heating rates tested. Figures 19 and 20 are the heat flow curves at 10 oC/min heating rate and demonstrate the glass transition of liquid monomer and cured polymer. 10. The samples were weighed ~16mg for several heating rates: 5. TA Instruments). 20 o C/min. the measured Tg value increases linearly with the heating rate. a heating rate within 10-20 oC/min is recommended. 15. For Tg measurement. Figure 19 Glass Transition of Liquid E4PETeA Monomer Detected by DSC 50 .

28267x R = 0.Figure 20 Glass Transition of Cured Poly(E4PETeA) Detected by DSC -61 y = -67.98663 -62 -63 Tg ( C) -64 -65 -66 -67 0 5 10 o 2 o 15 20 25 Heating Rate ( C/min) Figure 21 Effect of Heating Rate on Measured Tg Value of Liquid Monomer 51 .547 + 0.

1990): β= 1 ∂L L ∂T (44) 52 . 4.236 y = 215.3 Coefficient of Thermal Expansion The coefficients of thermal expansion (CTE) of monomer and cured polymer were determined by using an ellipsometry technique to measure film thickness at different temperatures. are adopted for liquid monomer and cured polymer. the Tg values obtained by extrapolating linear curves in Figures 21 and 22 to 0 oC/min.99617x R = 0.99246 234 232 230 Tg ( C) 228 226 224 222 220 0 5 10 o o 2 15 20 25 Heating Rate ( C/min) Figure 22 Effect of Heating Rate on Measured Tg Value of Cured Polymer To eliminate the heating rate effect. The linear CTE is defined as (Van Krevelen.5 oC and 215. respectively. -67.23 + 0.2 oC.

2 oC. The film was spincoated on silicon substrate (with native oxide layer) from a 10 wt% propylene glycol methyl ether acetate (PGMEA) solution. the ellipsometer scan starts after the film reaches thermal equilibrium. No phenomenon such as discoloration or brittleness was observed. The temperature controller (OMEGA CN 76000) can control temperature within ±0. respectively.9×10-4 1/K and β (polymer) = 0. A hot plate is installed on the commercial ellipsometer. β (monomer) = 5. OMEGA) was used for temperature calibration. 53 . At each set temperature. The solid polymer film was obtained by baking the liquid film containing monomer and photoinitiator and solvent at 180 oC in vacuum for 60 hrs. hence no apparent degradation occurred Figures 23 and 24 demonstrate the temperature dependence of monomer film thickness above Tg and of polymer film thickness below Tg.where ∂L is the slope of the film thickness versus temperature plot. Thermocouple (HH 11.96×10-4 1/K are found from these two graphs. J. and L the average ∂T thickness over the temperature range investigated. Woollam) was used to determine the film thickness at elevated (heating) or lowered (cooling) temperatures. The variable angle spectroscopic ellipsometer (VASE VB 250. The monomer film was put in the vacuum oven and baked at 90 oC in vacuum for 1hr to remove the solvent without solidifying the monomer.A.

Figure 23 Temperature Dependence of Liquid E4PETeA Monomer Film Thickness Figure 24 Temperature Dependence of Cured E4PETeA Polymer Film Thickness 54 .

β (polymer) = 0.The heat treatment of the film before ellipsometric measurement reduced the entrapped solvent enough that no solvent effect was observed during the heating or cooling stages. The thickness variation (500-2400 Å) was achieved by varying the spin speed and time. as shown in Figure 25. This is probably due to the residual unconverted monomer which has greater CTE entrapped in the polymer matrix. The temperature dependence of Si substrate n & k spectra (complex refractive index: n(λ ) = n(λ ) + ik (λ ) ) was taken into account when fitting the ellipsometric data to determine the film thickness. In Figure 24. which might indicate that the trend discussed here depends on the material properties of the film such as molecular weight or cross-link density. For the films with thickness below 1000 Å. Two L-T curves are found to almost overlap with each other in Figure 23.96×10-4 1/K. Kahle and coworkers (1998) demonstrated that when temperature dependent substrate data were used for the fit. 1998). A slower increase in CTE was observed with film thickness decreasing.. 55 . compared with the result from the fit with only the optical properties of Si at 25 oC. This is different from what we observed here for the poly(E4PETeA) film. The films with thickness above 1000 Å were made for measurement. but remains approximately constant for greater thickness. The CTE increases drastically for thickness below 2000 Å. the fit for the first temperature scan (heating cycle) has a slightly higher slope than that for the second scan (cooling cycle). there was no pronounced thickness effect for the CTE within the thickness range of 500 to 105Å for the poly(methyl methacrylate) (PMMA) film they investigated. the thermal fluctuation of the air above the film could cause a big error in the thermal property quantification (Kahle et al. The CTE value at greater thickness (2400Å here) was taken as the bulk CTE.

.Figure 25 CTEs of Poly(E4PETeA) below Tg Determined by Linear Regression of Curves Obtained by Fitting with Si Substrate Optical Data of 25oC (diamonds) and of Curves Obtained by Fitting with Temperature Dependent Si Substrate Data (triangles) The polymer CTE value. however. 1998): β unconstrianed = β constrianed × 1 −ν 1 +ν (45) 56 . The true CTE is related to the constrained CTE by the following equation (Kahle et al. was measured under the constraint of the Si wafer and therefore it overestimates the true value.

Ed.4×10-4 1/K. The 1 −ν term converts expansion constrained by the Si 1 +ν substrate to a true unconstrained CTE value. 1990). therefore.40 (typical Poisson’s ratio value for polymers. The polymer CTE value is also in the same range as PMMA (Brandrup and Immergut.23×10-4 1/K.where ν is Poisson’s ratio. the true CTE is calculated to be 0. 1991). 1989). respectively. etc (Yaws.9×10-4 1/K. The volumetric CTEs of E4PETeA (above Tg) and its polymer (below Tg) are thus determined to be 1. the measured value is the true bulk CTE. 1990) for the poly(E4PETeA).77×10-3. These values are at the same magnitude as CTEs of ethylene glycol dimethacrylate (EGDMA) and its polymer (Bowman and Peppas. assuming the bulk material is isotropic. on the other hand. Taking ν below Tg as 0. β (monomer) = 5. The volumetric CTE can be obtained by the following equation (Van Krevelen. is not constrained by the substrate. The liquid film. The monomer CTE is at the same order of magnitude as other acrylates such as methyl and butyl acrylates. Van Krevelen. and 1. 57 .. α= 1 ∂V = 3β unconstrained V ∂T (46) where V is the volume of the material over the temperature range investigated. 2003).

Using a group contribution method (Van Krevelen. Two watercalcium nitrate solutions of different concentrations were used to fill the column and form a linear density gradient from top to bottom.42 cm3/mol at 298 K. Techne). which is within 10% of the value obtained from Equation (47) for the same temperature.4. the density of the cured polymer was found to be 1290 Kg/m3 at 25 oC.4 Density The density of the cured polymer was found to be 1200 Kg/m3 at 35 oC (column control temperature) by using density gradient column (DC-4. The temperature dependence of density can be described as follows using the volumetric CTE α : ρP = ρ P (308 K ) 1200 = 1 + α P (T − 308) 1 + α P (T − 308) (47) Similarly. we have the following for the monomer density: ρM = ρ M (298K ) 1128 = 1 + α M (T − 298) 1 + α M (T − 298) (48) where ρ M (298 K) =1128 Kg/m3 from the product technical data sheet. 1990). the molar volume per structural unit of the polymer was calculated to be 404. The density of cured polymer was also calculated at 298 K. This justifies the measurement and Equation (47) will be adopted. With the unit molecular weight of 528 g/mol. 58 .

molar volume per structural unit and Poisson’s ratio of the cured polymer. is calculated to be 22. L f .The density of the curing material system can be expressed as: ρ = ρ M φ M + ρ P (1 − φ M ) (49) where ρ P and ρ M are described in Equations (47) and (48). Vm . The factor   1 +ν   1/ 2 is nearly constant for 59 .3 that at 298 K. Rao function.42 cm3/mol. and Vm = 404. 4. It can be obtained from Sections 4.5 Thermal Conductivity The thermal conductivity of polymer can be calculated using the following equation (Van Krevelen. specific heat capacity. 1990): U k = ρC P L f  R V  m   3(1 − ν )      1 +ν    3 1/ 2 (50) where ρ .1 and 4. 1990). UR.25 g/ml.18 J/g-K. C P = 1. The Rao function. respectively. and φ M is the monomer volume fraction as described before. ρ = 1. and ν are density. UR. L f = 5×10-11 m for PMMA is taken. respectively.460 (cm3/mol)⋅(cm/s)1/3 using a group  3(1 − ν )  contribution method (Van Krevelen. C P . average free path length.

8 K) of other acrylates such as butyl acrylate and methyl acrylate. A U-shape Pyrex cell. The thermal conductivity of the liquid E4PETeA is found to be 0. with capillary as part of it. thermal conductivity of amorphous polymers can be evaluated at different temperatures. 1990).solid polymers (≈ 1.8 K by averaging the results of five experiments. The later modelling results demonstrate that thermal conductivity is not a sensitive parameter. The Pyrex capillary is employed as the wire. The temperature dependence is insignificant and thus ignored within the SL cure temperature range (refer to other acrylates. 2003). A Hewlett-Packard (Model 6213A) power supply is used to provide the voltage for heating. which is within 10 % of the value at 25 oC and therefore the temperature dependence can be ignored in the temperature range during the cure reaction.5% and close to the thermal conductivity values (~0. filled with liquid mercury is inserted into the liquid sample. 1990).13 W/m-K at 297. The thermal conductivity of the liquid acrylate monomer was measured using the relative transient hot-wire method (Sun and Teja.05) (Van Krevelen. 2003). Yaws. For approximation. The thermal conductivity of the cured polymer is thus calculated to be 0. etc (Yaws.135 W/m-K. The thermal conductivity of polymer is temperature dependent.161 W/m-K at 297.2) was thus found to be 0. the averaged value of the cured polymer and liquid 60 . A thermocouple is used to measure the sample temperatures. The thermal conductivity of the cured E4PETeA polymer at its glass transition temperature 230 oC (Section 4. From a generalized plot of k(T)/k(Tg) as a function of T/Tg based on available experimental data (Van Krevelen. The value is reproducible within 0. 2003).123 W/m-K at 298 K. Further details of the experimental apparatus and procedure as well as theory were described by DiGuilio and Teja (1990).

1995a).monomer (0. 0. This value is at the same order of magnitude as that used for hexanedioldiacrylate (HDDA) curing system. The heat generated due to polymerization was found to increase with temperature linearly (Figure 26).142 W/m-K) can be taken as that of the curing material system to be used in the process model. 520 y = 352. The DPC experiments were performed at constant light intensity (0.9928 500 Reaction Generated Heat (J/g) 480 460 440 420 400 380 20 40 60 80 o 2 100 120 140 Temperature ( C) Figure 26 Effect of Temperature on Heat Generated by Polymerization 61 .2 W/m-K (Flach and Chartoff.36 mW/cm2) for several different temperatures below 130 oC.1758x R = 0.6 Heat of Polymerization The isothermal standard DSC experiments performed on the model material show that the thermally initiated polymerization doesn’t occur below 130 oC.36 + 1. 4.

1. The absorption spectrum thus obtained is the direct subtraction of the absorption of pure 62 . 4. The DPC and subsequent DSC experiments were repeated for higher light intensities (30.36 mW/cm2. DMPA. To obtain the absorption spectrum of DMPA in its E4PETeA solution. was determined by using a UVVIS spectrometer (Lambda 19. 1996) was performed. 0. 540 J/g can be taken as the heat of polymerization of the model material used. 50..subtraction factor × A (monomer) (51) where A represents the absorption spectrum. Perkin Elmer) and Beer’s law.2 wt% DMPA in E4PETeA were used as sample and pure E4PETeA monomer as reference in the spectrometer. 0.05. The 0.7 Absorption Coefficient The absorption coefficient of photoinitiator. spectral subtraction (Smith. 1994b). and 60 mW/cm2) and no more heat due to reaction was detected. This value is within 20 % of the experimental result. Therefore.Additional standard DSC experiments were conducted at elevated temperatures (till 350 oC at a rate of 10 oC/min) for samples irradiated at 130 oC. A small amount of residual heat was detected and added to give the maximum total heat of 540 J/g generated at light intensity = 0. A (DMPA)= A (solution) .6 kcal/mol per acrylate double bond (Anseth et al. The heat of polymerization was also calculated to be 650 J/g from the theoretical enthalpy of 20. 40.

monomer from that of solution. The subtraction factor the reference absorption is multiplied by was taken as 1.0 due to the low concentrations investigated. The investigated system assumed to obey the Beer’s law, the extinction coefficient spectra of the three solutions of different concentrations overlap with one another (Figure 27).

30 25 20 15 10 5 0 300
Concentration of DMPA (wt% in E4PETeA) 0.05wt% 0.1wt% 0.2wt%

Extinction Coefficient (m /mol-m)

3

325

350

375

400

425

450

475

500

Wavelength (nm)

Figure 27 Absorption Coefficient Spectrum of DMPA

63

4.8

Summary

In this chapter, the material thermal and physical properties are measured experimentally and verified by the theoretical calculation and literature values for similar materials. These values are listed in Table 5 and used in the SL process model established in Chapter 2.

Table 5 Characterized Material Properties Material Parameters thermal conductivity heat of polymerization absorptivity (initiator) quantum yield of initiation coefficient of thermal expansion (monomer) coefficient of thermal expansion (polymer) glass transition temperature (monomer) glass transition temperature (polymer) heat capacity (monomer) heat capacity (polymer) heat capacity (curing system) density (monomer) density (polymer) density (curing system)
C P,M

Values 0.142 2.85e5 19.9 0.6 7 0.00177 0.00012 205.65 488.35 = 5.6 × T ( K ) + 218.6

Units W/m-K J/mol m3/mol-m

1/K 1/K K K J/kg-K J/kg-K J/kg-K kg/m3 kg/m3 kg/m3

CP , P = 9.1× T ( K ) − 1535.5 C P = C P , M (1 − X ) + C P , P X
1128 /(1 + α M (T − 298)) 1200 /(1 + α P (T − 308))

ρ = ρ M φ M + ρ P (1 − φ M )

7

Goodner et al., 2002

64

CHAPTER 5 SIMULATIONS

With the kinetic parameters determined in Chapter 3 (Table 4), material properties evaluated in Chapter 4 (Table 5), and laser and process parameters recorded in the part building process, the SL cure process model established in Chapter 2 is solved using the multiphysics modelling and simulation code FEMLAB. FEMLAB is a product of the COMSOL Group
8

and has many model types available for use (application models). It

also supports equation-based modelling, enabling users to enter their specific differential field equations. Application models were used in this research. The process model established earlier can be easily customized in the FEMLAB environment. Since SL curing is a coupled mass and energy balance problem, two application models, diffusion and heat transfer by conduction, have been employed to accomplish the description of the cure process model. The transient analysis mode is selected. The 2D geometry described in Chapter 2 is the domain in which the balance equations apply when the laser draws a single line. As mentioned earlier, a small domain size has been adopted initially, which has then increased until no significant deviation in the modeling results from different domain sizes is observed, i.e., the domain should be large enough to accommodate the phenomena occurring physically. The balance equations established in the process model are consistent with those described in FEMLAB application models. The initial conditions are applied to the domain and boundary conditions applied to each boundary of the domain. The numerical values or

8

COMSOL Group, http://www.comsol.com/

65

and Lagrange elements have been selected for domain discretization. COMSOL Group. Case I Case II Case III Figure 28 Three Basic Laser Drawing Patterns: Case I. time stepping convergency. and domain convergency (i. overlapping single-layer lines with certain spacing. the mesh convergency. quadratic. Case III. The area where the reaction occurs and the resin properties vary significantly has finer mesh.formula descriptions of the material. the 9 “User’s Guide – FEMLAB 3. Stacked Single Lines For each case. process. and kinetic parameters also enter the software. 66 . Triangular.e.0”. The initial and upper limit of the time step size can be set manually. the solution is converging to a stable value as the mesh is refined. Case II. Three basic cases of the laser drawing patterns in SL are simulated (Figure 28): a single laser drawn line (also see Figure 5). and stacked single lines with certain layer thickness. Overlapping Single-Layer Lines. The model is then solved using a timedependent nonlinear solver in the software. Single Laser Drawn Line. The absolute tolerance has been set for each individual dependent variable based on their initial values. The absolute and relative tolerances determine the limit for the error estimated in each integration step9.

radical concentration. initial and boundary conditions) for this case has been established in Chapter 2. temperature. The profile of the transient intensity exposed on the resin is also described in Chapter 2.1×10-4 Units m/s K W/m2-K K W nm m Process Parameters Laser Parameters 5. and then decreases as the 67 .0288 325 1.2-dimethoxy-2-phenylacetophenone as photoinitiator.48 0. The curing reaction occurs immediately upon the laser exposure. The graphs in Figure 29 demonstrate how the monomer conversion.0272 304.55 4. All the simulations presented here have used ethoxylated (4) pentaerythritol tetraacrylate loaded with 2 wt% 2. Table 6 Process and Laser Parameter Values Used for Simulations Parameters laser scanning velocity bath temperature thermal convection coefficient chamber temperature laser power wavelength beam radius Values 0.time step size is reduced. The values of the process and laser parameters used for the simulations are listed below. and initiator concentration at a particular spatial point (x.18 300. or the domain is enlarged) have been performed to ensure valid and accurate solution. domain.0) (any point on centerline of the cured line at the surface) vary with time. The temperature increases rapidly due to the rapid exothermic reaction (approximately 30oC increase during the first 0.1 Single Laser Drawn Line The process model (consisting of governing equations.1sec).0.

no more initiator is consumed to produce radicals. the laser directly exposes the investigated point at t = 16ms.reaction slows down and heat conduction plays a role. In Figure 29. The initial delay is due to the absence of irradiation. The transient intensity caused by laser movement (Figure 29a) has induced “Gaussian” radical concentration profile.1s as well. The initiator is consumed and the radicals are generated during the very short irradiation period. which gives the highest intensity (mol/m2-s) and leads to most consumption of the initiator and most generation of the radicals. In the subsequent dark period. the radicals are rapidly exhausted and the monomer is consumed significantly in the first 0. Due to the very fast reactions. (a) 68 .

(b) (c) 69 .

(d) Monomer Conversion. (b) Initiator Concentration. 0) 70 . 0. and (e) Temperature at Point (x. (c) Radical Concentration.(d) (e) Figure 29 Transients of (a) Intensity.

From the conversion contour in Figure 30 (a). Upon a single laser scan (with scan conditions listed in Table 6). (a) 71 . which occurs at the center of the single line where the resin receives the most exposure and the most initiator is consumed. A longer stay in the bath wouldn’t lead to significant change to the initiator distribution due to the dark environment and extremely low diffusion. Figure 30 (b) demonstrates the consumption status of the initiator after a single laser scan and half an hour of post-curing in the bath. it can be expected that the cross section of the cured line would be of bullet shape.Figures 30 shows when the radicals are used up and the temperature returns to the bath temperature. a maximum of 25% monomer can be converted. the sectional view of the monomer conversion and initiator concentration profiles.

At the center of the irradiation (y = 0) where more radicals are generated during exposure.025 sec). respectively.(b) Figure 30 Distribution of (a) Monomer Conversion and (b) Photoinitiator Concentration upon a Single Laser Scan Figures 31 and 32 demonstrate the evolution of monomer conversion along the width at the resin surface (y axis in Figure 2) and along the depth centerline (z axis in Figure 2) of the single line part. It can be seen that the reaction starts rapidly upon irradiation (as shown in Figure 29a. In about 20 ms. the reaction slows down in the dark. 72 . the dark reaction contributes to about 10% conversion of monomer. irradiation starts at t = 0.01 sec and ends at t = 0.

01 sec. Depth along the Centerline of the Single Line Part (Plot Interval = 0. except for t = 1860 sec) 73 . except for t = 1860 sec) Figure 32 Monomer Conversion vs.01 sec.Figure 31 Monomer Conversion vs. Width at the Top Surface of the Single Line Part (Plot Interval = 0.

5. The full cross section of the drawn overlapping lines (rather than half cross section investigated in single line case where symmetry can be easily determined) is considered here. I is the intensity incident on any point (y. I = I 0 exp − 2 (Vs (t − (n − 1)0. line spacing). The jump time of the laser from the end of one line to the start of a second line is negligible (within 1ms). while the domain and boundary conditions need to vary accordingly to accommodate multiple lines.e. As for the single line case.0025 − (n − 1)hs ) 2 / wo exp(− z / D p ) 2 2 { [ ] } λ (nm) 1.2 Overlapping Lines The governing equations established for single laser drawn line in Chapter 2 are also valid for the overlapping line case.94 − t 0 ) ) + ( y − 0. Equation (52) describes the transient intensity profile for the nth line. hs is the hatching space (i. The laser irradiation is imposed line by line. With the same laser moving speed as in single line case (see Table 6).0025 is used to position the first drawn line in a 0.005 m wide domain. The curing situation upon previous drawing(s) is employed as the initial condition for the next scanning. it takes about 940 ms to draw a single 1 inch line or pass the same x location for a second time. the 3D problem is reduced to a 2D one due to. ignoring the line ends. 74 .196 × 10 8 (52) where I0 is the maximum intensity incident on the resin surface (W/m2). and the time point at which the laser starts to draw the first line is taken as t = 0. z) in the resin (mol/m2–s). the repetitive cure behavior along the laser scanning direction (x axis). 0.

Figure 33 demonstrates for two overlapping lines with hatching space hs = 1. the two scans cannot be easily distinguished from each other because the reaction also occurs in the joint area of the two scans. which. the cured lines generated can be separated. The size of the same conversion outline (indicating part size) in Figure 33 (a) is obviously larger comparing with that in Figure 30 (a). with an uneven bottom surface. the distributions of monomer conversion and initiator consumption. 75 . As the hatching space varies from large to small. More discussions on superposition of adjacent scans and the effect of hatching space on the cured part shape can be found in Jacobs (1992). However. in Figure 33 (a). or with a flat bottom surface (as in this case). considering the limited diffusion in the crosslinked network. respectively.5w0 when the radicals generated are used up and the temperature returns to bath temperature. The monomer conversion and temperature rise are found to be up to 32 % and 40oC. is basically Figure 30 (b) added with a second laser drawing at a hs distance. The initiator consumption profile in Figure 33 (b) represents the two adjacent laser irradiations. while for the single line drawn with same speed and same laser power. as mentioned earlier. they are 25 % and 30oC.

(a) (b) Figure 33 (a) Monomer Conversion (b) Initiator Concentration (mol/m3) Distributions upon Two Overlapping Scans 76 .

Width at the Top Surface of Two-Overlapping-Line Part (Plot Interval = 0.01 sec. Figure 34 Monomer Conversion vs.Figure 34 demonstrates the evolution of monomer conversion along the width at the resin surface of the two-overlapping-line part. More monomer is converted in the first line region near the second line due to the extra exposure from the second line drawing. It can be seen that the second line grows rapidly (a noticeable increase in conversion is observed at the right of the first line) upon irradiation (as discussed earlier. the laser starts to draw the second line at t = 0. the reaction in all regions slows down in the dark. In about 20 ms. the dark reaction contributes to about 10 % conversion of monomer at the locus where more irradiation is received. except for t = 1860 sec) 77 . Similar to the single line case.94sec).

the repetitive cure behavior along the laser scanning direction (x-axis). Figure 35 demonstrates for two stacked lines with layer thickness LT = 4 mils when the radicals generated are used up (except the small amount of radicals trapped in the first layer due to high crosslinking) and the temperature returns to bath temperature. I = I 0 exp − 2 (Vs (t − (n − 1)40 − t 0 ) ) + y 2 / wo exp[−( z − (n − 1) LT ) / D p ] 2 2 { [ ] } λ (nm) 1. The time delay between drawing two neighboring layers is estimated to be 40sec.3 Stacked Single Lines In this case. The laser irradiation is imposed layer by layer. Additional sub-domains for subsequent layers (or lines) are added layer by layer at the top of the previous domain. 78 .5. Also as in the single line case. The monomer conversion and temperature rise are found to be up to about 40 % and 35 oC. the time point at which the laser starts to draw the first layer is taken as t = 0.196 × 10 8 (53) where LT is the layer thickness. the 3D problem is reduced to a 2D one due to. The curing situation upon previous drawn layer(s) is employed as the initial condition for the next layer. respectively. The depth of each sub-domain is the layer thickness. including resin recoating and laser beam analyzing time. The simulation of the first layer is exactly the same as of the single line case. ignoring the edge effect. single lines are built layer by layer with each line drawn on the top of the previous one(s). Equation (53) describes the transient intensity profile for the nth layer. the distributions of monomer conversion and initiator consumption.

79 . doesn’t necessarily mean that these two layers are separate physically. too small a layer thickness makes two layers well connected. however. The cure at the joint area could be enough to hold two layers together.comparing with 25 % and 30 oC for the single line drawn with same speed and same laser power. The same conversion outline (indicating part size) in Figure 35 (a) is obviously wider and deeper than that in Figure 30 (a) as well. Two layers can be easily distinguished from each other as shown in Figure 35. it increases the part building time. The maximum monomer conversion and initiator consumption obviously occur in the first layer near the joint boundary where the resin receives the maximum exposure during the first laser scan and still receives extra exposure during the second scan. which. The layer thickness affects how well the two adjacent layers are attached to each other. however. Too large of a specified layer thickness causes two layers to partially join together or even separate.

(a) (b) Figure 35 (a) Monomer Conversion (b) Initiator Concentration (mol/m3) Distributions upon Two Stacked Scans 80 .

Figure 36 Monomer Conversion vs. verifying the second layer at the top of the first one is 4 mils (layer thickness) thick. the maximum +z = 1×10-4 m (the top surface of the first layer is at z = 0). Similar to the single line case.01 sec. As shown in Figure 36. In about 20 ms. except t = 1860 sec) 81 .Figure 36 demonstrates the evolution of monomer conversion along the centerline in depth of the two-layer-line part. the dark reaction contributes to about 10 % conversion of monomer at the locus where more irradiation is received. Depth at the Centerline of Two-Layer-Line Part (Plot Interval = 0. the laser starts to draw the second line at t = 40 sec). the reaction in all regions slows down in the dark. More monomer is converted in the first layer region near the second layer due to the extra exposure from the second line drawing. It can be seen that the second line grows rapidly (a noticeable increase in conversion is observed at the top of the first line) upon irradiation (as discussed earlier.

but for overlapping-line and stacked-line parts. the single-line. It’s found that a certain degree of cure (DOC) contour outlines the built part within minimal error. The parts were then dried using compressed air and broken to expose their cross sections.CHAPTER 6 MODEL VERIFICATION To verify the process model. The SEM measurements were calibrated 82 . The parts were elevated out of the resin vat 30 minutes after laser scanning was finished. For this reason. overlapping-line. This DOC threshold model is valid not only for single line parts. They were cleaned for one minute in TPM (tri-propylene glycol monomethyl ether) and another minute in water at room temperature using a Branson 5210 cleaning system in ultra-sonic mode.1 DOC Threshold Model The E4PETeA acrylate with 2 wt % DMPA initiator was used to grow single line parts in SLA-250/50. and stacked-line parts have been fabricated in SLA and their dimensions measured to compare with the simulation results. 6. this model has reduced the prediction error to 25 %. While the exposure threshold model predicts the cured part dimensions with up to 50% error. Hitachi S800 FEG). and drained on the platform for another 15 minutes. the SL cure process model established and solved earlier can also be referred to as a “DOC threshold model” when used to predict the fabricated part shape and dimensions. the image and dimensions of which were taken and measured by scanning electron microscopy (SEM.

The DOC corresponding to this contour line is defined as the critical DOC. above which the solid part can be formed while below this DOC the resin has not been solidified enough and can be washed away during the postprocessing step. 83 . In this sense. which as expected is of parabolic shape. Figure 37 SEM Image of Cross Section of a Single Line Part The part building process has then been simulated and Figure 38 gives the DOC contour of the built part which corresponds to half the image in Figure 37. the SL cure process model developed in this work can also be referred to as a DOC threshold model.using a standard grid with known dimensions. the DOC threshold model takes the reaction and transient intensity effects into account. the outline is close to the 9% DOC contour. Unlike the exposure threshold model which only incorporates the exposure. Figure 37 shows a typical image of the cross section of a single line part. For this single line part.

This range can be applied to the simulated DOC profile to predict the cured part dimensions. Appendix C). 84 .Figure 38 Degree of Cure Contour for Parts Built at Vs = 1. which defines the range of the critical DOC. which indicates that the cured polymer does not have to achieve an infinite molecular weight to form a solid part. The critical DOC is lower than the gel point of the curing material (~14%.071 in/sec (the scanning speed needs to be low enough for such a small part as a single line to be formed as well as be strong enough to handle and measure). Their measured outlines fall between the ~9% and ~10% DOC contours (Figure 38 shows a case where it overlaps with the 9% DOC contour).071in/sec (with the measured part contour shown in red) About 20 single line parts were built and measured at the laser drawing speed Vs = 1.

. i. The part building and cleaning procedure itself is not important.The part building process has to be consistent in order for the fabricated parts to have predictability. A different set of postprocessing steps can be adopted. 6. it is important that parts be fabricated consistently.. the pairwise comparison (Neter et al. On the other hand.2 DOC Threshold Model Prediction Three types of parts have been built and measured to verify the SL cure process model or the prediction capability of the DOC threshold model: single line parts. it’s not necessary to stick with the operating procedures described above. but wouldn’t affect the prediction of the DOC threshold model.466 in/sec. which could give a different value for the critical DOC. the consistency is important. 6. following exactly the same building and postprocessing steps. the steps have to be followed consistently once they are designed and adopted. To verify the size of these parts is different from that of those built at Vs = 1. therefore.e.e.1 Single Line Part Prediction The single line parts have also been built at a laser scanning speed of 0. and stacked single lines (three layers). i.071 in/sec. 1996) has been used: H 0 : D = µi − µi' = 0 H a : D = µi − µi' ≠ 0 (54) 85 . Obviously different part cleaning procedures could cause differences in the part size or shape. overlapping cured lines (nine lines).2.

071 in/sec.466 in/sec have different (larger) size than those built at Vs = 1.54 -11. H a is concluded with 99.466 in/sec).466 in/sec Depth (µm) scanning speed i Measurement Full Width (µm) i =1 724 703 761 738 763 762 738 727 753 741 725 727 757 759 768 757 774 737 i =2 922 945 963 899 901 956 951 980 958 988 990 978 945 907 998 972 1008 964 938 982 957 -212 706.56 i =1 262 275 268 266 259 265 256 268 256 267 260 267 273 261 255 275 283 260 i =2 311 309 317 321 349 272 324 329 332 277 335 333 302 316 347 331 334 345 328 334 322 -57 247.e. Vs =0.where µi is the mean of the i th part dimension (here i = 1 for Vs = 1.071 in/sec. 36) = 3.15 average Difference MSE t* 745 265 86 . to conclude to support or to not support the hypothesis that the part dimensions fabricated at these two scanning speeds are the same. 2. Therefore. i. and D is the difference between the two means. The t test is utilized to decide to conclude H 0 or H a . Table 7 Dimensions of Single Line Parts Built at Two Laser-Scanning Speeds: 1. Parts built at Vs = 0. As shown in Table 7.589. Vs =1.9995.071 in/sec and i = 2 for Vs = 0.9 % or higher confidence. for both part depth and width the magnitude of test statistics t* is much greater than t (0.65 -24..

C.071 745 ± 10 957 ± 15 265 ±4 322 ± 10 9% 10% 9% 10% 260 236 322 306 *Xc model: DOC threshold model.: confidence interval Part building at a higher scanning speed of 10. The critical DOC range (9~10%) is applied to the DOC profile to predict the cure depth and cure width. No DOC above 7% is observed anywhere in the simulated vat.The building process at a laser moving speed Vs = 0. Figure 39 shows DOC contours for the 10.I.) depth (µm) critical Xc DOC model 9% 10% 0. This agreement also demonstrates and verifies the model’s predictive ability.71 in/sec was attempted. however.466 9% 10% 660 560 1050 940 prediction error (%) -10 -25 10 -2 full width (µm) experiment critical Xc (95% C. They are found to be in good agreement with the fabrication results (Table 8). The prediction error is defined as follows: prediction error = simulated value − experimental value × 100% experimental value (55) Table 8 Single Line Part Prediction by DOC Threshold Model Vs experiment (95% C. 87 . The simulation for this faster writing speed process also demonstrates that the part cannot be formed. It turns out.71 in/sec scanning speed.I.) DOC model prediction error (%) -2 -10 0 -5 (in/sec) 1.466 in/sec has also been simulated. and thus since this is lower than the required 9% DOC no solid polymer structure is predicted. that the resin is not cured enough to form a solid part.I.

Figure 39 Degree of Cure Contour for Parts Built at Vs = 10. The size is larger as shown in Table 9. The shape of the cured lines is similar to that of the single cured line due to the narrow line spacing and limited number (n = 9) of drawn lines.2 Overlapping Line Part Prediction The overlapping line parts have been built at Vs = 17.71 in/sec 6.967 in/sec and with a hatching space hs = 0.5 mils.2. 88 .

as shown in Table 10.: confidence interval 89 .1 86 1278 1420 Full Width (µm) 429 435 467 463 431 446 467 494 455 519 434 563 406 513 468 43. The predicted dimensions are within 25 % of the experimental measurements.) 468 ± 25 full width (µm) critical Xc DOC model 9% 364 10% 349 prediction error (%) -20 -25 *Xc model: DOC threshold model.I.I. Table 10 DOC Threshold Model Prediction for Overlapping Line Parts experiment (95% C.I.7 0.09 25 448 489 Mean Standard Deviation Coefficient of Variation Half Length of 95% C.Table 9 Dimension Measurements of Overlapping Line Parts Measurement Depth (µm) 1457 1406 1442 1449 1412 1370 1446 1004 1429 1454 1432 1042 1346 1431 1349 149.) 1349 ± 86 depth (µm) critical Xc DOC model 9% 1220 10% 1120 prediction error (%) -10 -20 experiment (95% C.0 0. Lower Limit Upper Limit This multiple-line part building process is simulated and the critical DOC is applied to the DOC contour plot. C.I.

05 8 341 355 Mean Standard Deviation Coefficient of Variation Half Length of 95% C.6. The dimension measurement results are shown in Table 11.0 0. 90 . The predicted dimensions are within 25 % of the experimental measurements. The cured lines have similar shape to the single cured line with deeper and wider size. Table 11 Dimension Measurements of 3-Layer Stacked Line Parts Measurement Depth (µm) 977 1146 951 1257 1178 1108 1073 995 904 941 932 1021 1018 1044 1040 890 945 1025 101.2. I.1 52 982 1068 Full Width (µm) 347 341 356 335 319 326 376 362 338 361 353 365 358 345 326 342 367 348 16.8 0.3 Stacked Line Part Prediction The 3-layer stacked single line parts have been built at Vs = 1.052 in/sec and with the layer thickness LT = 4 mils. Lower Limit Upper Limit This layer-by-layer part building process is simulated and the critical DOC is applied to the DOC contour plot. as shown in Table 12.

As mentioned earlier. DOC threshold model). and linewidth. 6.e.: confidence interval The prediction results of all three types of cured lines using DOC threshold model have been demonstrated. Cd. C. it assumes a critical exposure.I. 1992): C d = DP ln( E max ) Ec (56) Lw = 2 w0 Cd 2 DP (57) 91 . The cure depth.) 348 ± 8 full width (µm) critical Xc DOC model 9% 346 10% 330 prediction error (%) -1 -5 *Xc model: DOC threshold model.I. (referring to the maximum depth and width of the cured line.Table 12 DOC Threshold Model Prediction for 3-Layer Stacked Line Parts experiment (95% C. The critical DOC taken outside the 9~10% range leads to high prediction error. The good agreement between experimental results and simulation results (within 25 % error) validates the SL cure process model (i.3 Exposure Threshold Model Prediction The exposure threshold model has been widely used to guide the SL prototyping in industry.I. Ec.) 1025 ± 52 depth (µm) critical Xc DOC model 9% 1285 10% 1150 prediction error (%) 25 10 experiment (95% C. Lw. respectively) can be obtained from the following (Jacobs. is necessary for a part to be formed.

22 mJ/cm2 and 9. the WINDOWPANETM experiments were conducted to characterize Ec and DP of the model acrylate resin. (Cd(mils)) Figure 40 Working Curve from WINDOWPANETM Experimental Data 92 . The laser drawing speed was varied to achieve a wide range of cure depth.3. According to Equation (56). 1992).1 Ec and Dp Determination In SLA-250/50.43 mils.654 R2 = 0. as shown in Figure 40. and DP is the penetration depth of the laser into the resin at which the irradiance would be about 37 % of the surface irradiance. respectively. versus maximum exposure. 6.4271Ln(x) . SLA Working Curve for Model Acrylate Resin 50 Cure Depth.999 Cd(mils) Low Cd High Cd Log.18. Cd. Ec and DP were thus found to be 7. w0 is the laser beam radius. Emax (mJ/cm2) y = 9. Emax. the best linear fit was performed in the semilog plot of cure depth. Ec and DP are regarded as the resin characteristic parameters and can be determined from WINDOWPANETM experiments (Jacobs.where Emax is the maximum exposure incident on the resin surface centerline during laser scanning. for the model resin: E4PETeA acrylate with 2 wt % DMPA. Cd (mils) 40 30 20 10 0 1 10 100 1000 10000 Exposure.

which guarantees the correctness of Ec and DP values determined above. C.071 1310 37 957 ± 15 322 ± 10 0. 6.3. the maximum exposure incident on the resin surface can be expressed as follows (Jacobs.I. It turned out that all cure depth values of the produced windowpane parts are within 1 mil of the specified values. The SL system uses these two values and sets the desired cure depth for part building. measured. and then compared to the specified cure depth.) prediction error (%) 27 13 93 .) (in/sec) Ec model 1118 50 745 ± 10 265 ±4 1.The “reverse WINDOWPANETM” experiments (Jacobs.I.: confidence interval Vs experiment (95% C.I. with up to 50% for the depth and 27% for the width. For a single drawn line. Table 13 Single Line Part Prediction Results Based on Exposure Threshold Model depth (µm) full width (µm) Ec model 336 364 prediction experiment error (%) (95% C.2 Single Line Part Prediction Equations (56) and (57) have been used to calculate the cure depth and full width of the cured line. The test parts were built. Obviously the exposure threshold model has much larger prediction error than DOC threshold model (within 25% error). 1992): E max = 2 PL π w0Vs (58) Table 13 shows the calculated results for the case of single line parts. 1992) were performed to verify the determined Ec and DP values.466 *Ec model: exposure threshold model.

taking the average of Gaussian and top-hat beam): 94 . a top-hat distribution is introduced: I 0 I = 0 0 ≤r ≤w0 r > w0 (59) The Gaussian assumption is modified by equally combining with the top-hat distribution as follows (i. This is probably due to the inappropriate assumption of Gaussian laser beam intensity distribution. The beam is not symmetric.e. The picture below shows the beam width ratio in two directions X/Y = 1.08. and time-varying. Figure 41 Beam Intensity Profile of HeCd Laser in SLA-250/50 To obtain a better simulation of the beam profile. It is apparent that the beam is not exactly a Gaussian distribution.The error of the exposure threshold model to predict the cure depth is quite large. either. Figure 41 is a picture of beam intensity profile taken by a digital camera.

5 I 0  I = 2 0. the following equation gives the exposure into any depth z: E ( y.5  E ( y.0) exp(− z / D p ) (63) Applying the critical exposure Ec to the above equation.2 0.0) =  0.5   π w0 I 0 2 Vs 2 exp(−2 y 2 / w0 ) + I0 Vs 2 w0 − y 2 y ≤ w0 (62) y > w0 π w0 I 0 2 Vs exp(−2 y / w ) 2 2 0 Furthermore. 95 . the cure depth can be obtained by setting y = 0 and the linewidth obtained by setting z = 0.5 I 0 exp(−2r 2 / w0 )  0 ≤ r ≤ w0 r > w0 (60) The integral of the surface irradiance over the exposed region must be equal to the laser power PL incident on the resin surface. Therefore. the peak surface irradiance at r = 0 can be obtained as: I0 = 4 2 PL /(π w0 ) 3 (61) The integration of the surface irradiance over time gives the surface exposure:  0.5 I 0 exp(−2r 2 / w0 ) + 0. z ) = E ( y.

) 265 ±4 Ec model (G beam) 336 50 27 Ec model (½G+½T beam) 293 44 10 Ec model (T beam) 220 43 -17 prediction full prediction Vs =0. It can also be observed that even the improved exposure threshold model prediction results are not as good as those of the DOC threshold model using the raw Gaussian beam. T: top-hat distribution. Varying the weighting factors doesn’t improve the prediction accuracy because more weighted tophat distribution increases the width prediction error without reducing the depth error significantly.I. prediction error (%) Vs =1.466 in/sec depth(µm) error (%) width(µm) error (%) Experiment (95% C. This also indicates a relatively low requirement of the DOC threshold model for beam profile simulation accuracy comparing with the exposure threshold model. a better simulation of beam profile might improve the prediction of the DOC threshold model as well. On the other hand.As seen in Table 14. a better prediction can be achieved using the equally weighted combination beam than using a pure Gaussian or a pure top-hat distribution. and more weighted Gaussian distribution increases error in both depth and width. ½G+ ½T: equally weighted combination of Gaussian and top-hat distribution.071 in/sec depth(µm) 745 ± 10 1118 1076 1064 96 .I. Table 14 Exposure Threshold Model Prediction Results using Modified Beam Profile full prediction width(µm) error (%) Experiment (95% C.) 957 ± 15 322 ± 10 Ec model (G beam) 1310 364 37 13 Ec model (½G+½T beam) 1276 325 33 1 Ec model (T beam) 1256 220 31 -32 * G: Gaussian distribution.

6.3 Overlapping Line Part Prediction As mentioned earlier. the exposure incident on any point of the resin Q when the laser draws n overlapping lines with hatching space hs can be expressed as: E ( y. which then gives the equation that calculates the outline of the cured part. the exposure threshold model basically derives the exposure expression for a part building process.3. Suppose the laser starts drawing at y = 0 (as shown in Figure 42). z ) = 2 PL 2 2 exp(−2 y 2 / w0 ) + exp[−2( y + hs ) 2 / w0 ] + π w 0V s { 2 + exp[−2( y + (n − 1)hs ) 2 / w0 ] exp(− z / D P ) } (64) Figure 42 Laser Movement when Drawing Overlapping Lines 97 . and then applies the critical exposure.

which is 25 % smaller than the measured result. -2hs. + ymax = 126 µm. z ) = E c .2. respectively. the prediction errors for the part width and depth by the DOC threshold model are within 25 % and 20 %. -hs. the global maximum of which is the maximum cured depth of the part. which underestimates the experimental value by 30 %. The cure depth is found to be 925 µm. Equation (64) reduces to: Ec = 2 PL 2 2 exp(−2 y * 2 / w0 ) + exp[−2( y * + hs ) 2 / w0 ] + π w 0V s { 2 + exp[−2( y * +(n − 1)hs ) 2 / w0 ] exp(− z * / D P ) } (65) where y* and z* describe the outline of the cured part. …. For the case of overlapping line parts with nine drawn lines n = 9 and hatching space hs = 0. Setting y*=0.Letting E ( y. the example tested in section 6. and thus the cured linewidth turns out to be 353 µm. -[rounded(n/2-1)] hs.e. two values will be obtained for y*: +ymax and -ymax. . 98 . The sum of their absolute values is the cured linewidth.5 mils and laser scanning speed Vs = 17.967 in/sec (i.ymax = .227 µm.2.2). As discussed in section 6. the corresponding number of z* values can be obtained as local maxima.2. When z*=0.

2002): E ( y. the maximum full width is calculated to be 336 µm at the top of the first layer. the cure depth is calculated to be 1. For the case of 3-layer stacked single line parts built at Vs = 1. (n − 1) LT ) = 2 PL 2 exp(−2 y 2 / w0 )[1 + exp(− LT / D P ) + exp(−2 LT / D P ) + π w 0V s + exp(−(n − 1) LT / D P )] (67) Letting E ( y. z ) = 2 PL 2 exp(−2 y 2 / w0 )[exp(− z / D P ) + exp(−( z − LT ) / D P ) + π w 0V s + exp(−( z − (n − 1) LT ) / D P )] (66) where z = Cd when y = 0. the exposure incident on the resin at the bottom layer can be expressed as (let z = 0 at the top layer surface): E ( y. The exposure received at a point on the top of the nth layer is (Rosen. z ) = E c . letting E ( y.052 in/sec and with layer thickness LT = 4 mils (i.4 Stacked Line Part Prediction When the laser draws n stacked single lines with layer thickness LT.3. which overestimates the measured value by 46 %. (n − 1) LT ) = E c . This indicates that the width of the rib built by three stacked single vectors is not uniform. and 362 µm at the top of the third layer. 354 µm at the top of the second layer.2.e.497 µm.6. Although this stair effect is not distinguishable in the experimental results. 99 .3). the example tested in section 6.

6. which is probably due to the inappropriate assumption of a Gaussian laser beam. 100 .3.5 Comparison of DOC and Exposure Threshold Model The predicted results by both the DOC threshold and exposure threshold models for all three building types (single line.the exposure threshold model gives a good prediction of the rib width. overlapping lines. the exposure threshold model appears to give better prediction of width than depth. The prediction error of the exposure threshold model is up to about 50 %. For the cases tested here. while the DOC threshold model can predict within 25 % error. It also appears that the exposure threshold model is more sensitive to beam profile assumptions than is the new DOC threshold model. and stacked lines) are summarized in Table 15. None of these three values has deviated from the measured value by more than 5 %.

Therefore.) Xc model. 101 . Xc = 9 % Xc = 10 % Ec model depth (µm) 745 ± 10 660 560 1118 depth (µm) 957 ± 15 1050 940 1310 depth (µm) 1349 ± 86 1220 1120 925 depth (µm) 1025 ± 52 1285 1150 1497 prediction error (%) -10 -25 50 prediction error (%) 10 -2 40 prediction error (%) -10 -20 -30 prediction error (%) 25 10 50 full width (µm) 265 ± 4 260 236 336 full width (µm) 322 ± 10 322 306 364 full width (µm) 468 ± 25 364 349 353 full width (µm) 348 ± 8 346 330 336 prediction error (%) -2 -10 30 prediction error (%) 0 -5 15 prediction error (%) -20 -25 -25 prediction error (%) -1 -5 -5 *Xc model: DOC threshold model.) Xc model. one option is to choose Ec and DP in the higher working range.I.08 mils. Xc = 9 % Xc = 10 % Ec model Stacked line part Experiment (95 % C. Xc = 9 % Xc = 10 % Ec model Overlapping line part Experiment (95 % C.98 mJ/cm2 and 6.I.071 in/sec Experiment (95 % C. Ec and DP are fit to be 0.466 in/sec Experiment (95 % C. Ec model: exposure threshold model. C. Xc = 9 % Xc = 10 % Ec model Single line part @ 0.: confidence interval One might argue that the exposure threshold model is also very dependent on what range of the working curve for the resin is used. in this range. as shown in Figure 43. For example. respectively.Table 15 Comparison of Prediction Results by Two Threshold Models Single line part @ 1. different regions of the working curve were fit and these new Ec and Dp values were used for the exposure threshold model prediction.I.) Xc model.I.) Xc model.I.

Stacked-line Parts depth (µm) Line Type 1 (1) 1 (2) 2 3 experiment (95% C.0969Ln(x) + 0.) 265 ±4 322 ± 10 468 ± 25 348 ± 8 full width (µm) Ec model 402 425 430 402 prediction error (%) 50 30 -10 15 102 .071 (2) Vs = 0. Cd (mils) y = 6. 2.SLA Working Curve for Model Acrylate Resin 50 Cure Depth. Table 16 Exposure Threshold Model Prediction (high working range): 1. As shown in Table 16. the adoption of Ec and Dp from the higher working curve range of the resin doesn’t improve the predictive ability of the exposure threshold model. Overlapping-line. Emax (mJ/cm2) Figure 43 High Working Range for Model Acrylate Resin in SLA However. the prediction error is up to 40% for the cured depth and 50% for the part width.466 in/sec. Single Line (1) Vs = 1.) 745 ± 10 957 ± 15 1349 ± 86 1025 ± 52 Ec model 1029 1153 905 1322 prediction error (%) 40 20 -35 30 experiment (95% C.I.9955 Cd(mils) Low Cd High Cd Log. (High Cd) 40 30 20 10 0 1 10 100 1000 10000 Exposure.0376 R2 = 0. and 3.I.

This indicates that in order to predict these part dimensions. an exposure threshold model was employed for part prediction in which Ec and Dp were determined by fabricating the windowpane parts using the same postprocessing protocol as that for the DOC model development and regular part building (see 6. Recall when the 3D System’s procedure was used.It should be mentioned that for the parts built and tested. This is expected because DP is a characteristic property of the 10 AccuMaxTM ToolKit User Guide for use with SLA-190. respectively.6 Model Prediction using Ec and Dp Evaluated by a Different Protocol In the previous sections where the exposure threshold model was used for part dimension prediction. However. Ec and DP were found to be 5. 6.1 “DOC Threshold Model”). one might argue that the exposure threshold model would perform as well or better than the DOC threshold model if identical washing procedure was used.22 mJ/cm2 and 9. the washing procedure used in building parts or determining the critical DOC for our new DOC threshold model is different from the 3D systems WINDOWPANE procedure.3. Ec and DP were found to be 7. in this study. the exposure doses fall in the higher range of the working curve. the adoption of the higher range of data gives similarly poor predictions. This indicates that the depth of penetration. 3D Systems. the 3D Systems WINDOWPANE procedure10 was used to fabricate and post-process the windowpane parts to evaluate Ec and Dp. 500. two different protocols generated curves with similar slopes.43 mils. as we can see above. As shown in Figure 44. respectively (Figure 44). the higher range of the curve should be used to determine Ec and Dp for the exposure threshold model. DP.24 mJ/cm2 and 9. Therefore. 103 . of the resin (the slope of the curve) does not depend on the postprocessing procedure. Therefore. 350.48 mils. 250. However.

18.3εS ) = 9.999 10 100 1000 10000 low Cd (SOP) high Cd (SOP) Exposure. DP = 1 /(2. The reverse windowpane parts were built using this set of Ec and Dp and it was found that the cured depth values 104 . The critical exposure. Ec is an ambiguous concept. Emax (mJ/cm2) Figure 44 Comparison of Working Curves Obtained by the 3D Systems WINDOWPANE Procedure (labeled “SOP” in the figure) and by the Part Building Protocol As shown in Figure 44. however. which is related to the molar concentration and absorptivity of the initiator in the resin. which leads to the poor predictive ability of the exposure threshold model which takes both Ec and DP as the resin characteristics.654 R2 = 0.4806Ln(x) .resin. it varies with the part processing procedure varying.9659 high Cd Cd (mils) low Cd Cd (mils) (SOP) y = 9. For the same resin.59 mils. Ec (the natural log of Ec is proportional to the intercept of the curve). Model Acrylate Resin Working Curve_comparison of two protocols 60 Cure Depth. Both DP values obtained by these two different protocols are within 2% of this calculated value. while it is more than 99% when the 3D Systems’ procedure is used. Cd (mils) 50 40 30 20 10 0 1 y = 9.71 R 2 = 0. the correlation coefficient for the fitting is about 97%.4271Ln(x) . This indicates that Ec is not an inherent property of the resin. is found to be affected by the post-processing procedure significantly.15.

could be over 7 mils out of the nominal values (relatively. The paper towel strips are also placed with equal weights on top of parts (ideal weight for drainage is 10-13g 10) to drain the excess resin. all cure depth values of the produced windowpane parts were found to be within 1 mil of the specified values (within 5% of nominal values). The ultrasonic vibration or air blowing used in our part building procedure may not clean the 105 . etc) on the part. PCA-250 (3D Systems). solvent washing. It can be seen that the 3D Systems’ procedure is most likely much more effective at removing excess resin without damaging the parts than the post-processing steps used for regular part building. with dry side (top side) down on a clean glass plate for further cure. This is expected since the resin properties (Ec and Dp) were not characterized properly by using this washing procedure as demonstrated by the poor predictive performance of the reverse windowpane parts. These facts demonstrate that using the post-processing protocol is not an effective way to characterize the resin working curve due probably to the non-uniform effect of this protocol (draining. As shown in Table 17. rather than draining or washing using solvent and ultrasonic equipment or compressed air drying. The parts are then put in the post-cure apparatus. after building the windowpane parts are placed with wet side (bottom side) down inside the SLA chamber (the heat inside the chamber is nearly optimum for the draining 10) with the paper towel underneath. 15% different from the specified values). In the 3D Systems’ procedure. Recall when the 3D Systems WINDOWPANE procedure was used to determine Ec and Dp. the Ec and Dp characterization using the same post-processing step as the regular part building worsens the prediction of the exposure threshold model with the prediction error up to 60% (~10% less accurate than using the 3D Systems’ procedure).

29 mJ/cm2 and Dp = 7. Such factors could cause the correlation of working curve and model prediction to be worse than those using Ec and Dp determined by the 3D Systems’ procedure.) 745 ± 10 957 ± 15 1349 ± 86 1025 ± 52 Ec model 1201 1395 1007 1581 prediction error (%) 60 45 -25 55 experiment (95% C.) 265 ±4 322 ± 10 468 ± 25 348 ± 8 full width (µm) Ec model 347 374 367 347 prediction error (%) 30 15 -20 0 Again.071 in/sec (2) Vs = 0. (high Cd) 40 30 20 10 0 1 10 100 1000 10000 Exposure (mJ/cm2) Figure 45 Ec and Dp Determined in the High Range using Part Building Protocol 106 .I. Model Acrylate Resin Working Curve_developed protocol (high range) 60 50 Cd (mils) y = 7.9439 high Cd Cd (mils) low Cd Log.25 mils.2486Ln(x) .I. 2.windowpanes uniformly or they harm the cured dimensional uniformity.466 in/sec. Stacked-line Parts depth (µm) Line Type 1 (1) 1 (2) 2 3 experiment (95% C. Single Line (1) Vs = 1.1.8405 R 2 = 0. and 3. Overlapping-line. The adoption of the high range of the new working curve (Figure 45) produces an Ec = 1. Table 17 Exposure Threshold Model Prediction (protocol): 1. but again this does not improve the model performance (Table 18). one can argue about what range of the working curve to use.

506Ln(x) .466 in/sec. Overlapping-line.I. and this does not improve the model predictions and in fact even makes them worse (Table 19). 2.Table 18 Exposure Threshold Model Prediction (protocol.9977 1 10 100 1000 10000 2 Cd (mils) 40 30 20 10 0 Exposure (mJ/cm2) Log.) 745 ± 10 957 ± 15 1349 ± 86 1025 ± 52 Ec model 1177 1325 1028 1499 prediction error (%) 60 40 -25 50 experiment (95% C.724 R = 0. (low Cd) Figure 46 Ec and Dp Determined in the Low Range using Part Building Protocol 107 . high working range): 1.071 in/sec (2) Vs = 0.51 mils.) 265 ±4 322 ± 10 468 ± 25 348 ± 8 full width (µm) Ec model 393 417 420 393 prediction error (%) 50 30 -10 15 Likewise. the adoption of the low range of the curve (Figure 46) produces an Ec = 7.90 mJ/cm2 and Dp = 10. Stacked-line Parts depth (µm) Line Type 1 (1) 1 (2) 2 3 experiment (95% C.I. Single Line (1) Vs = 1. Model Acrylate Resin Working Curve_developed protocol (low range) 60 50 high Cd Cd (mils) low Cd y = 10. and 3.21.

This indicates that in order to predict these part dimensions. 2. as we can see above. Stacked-line Parts depth (µm) Line Type 1 (1) 1 (2) 2 3 experiment (95% C.466 in/sec.) 265 ±4 322 ± 10 468 ± 25 348 ± 8 full width (µm) Ec model 332 361 349 333 prediction error (%) 25 10 -25 -5 It should be mentioned that for the parts built and tested.071 in/sec (2) Vs = 0. the higher range of the curve (Figure 45) should be used to determine Ec and Dp for the exposure threshold model. and 3. see Table 15).I.I. It was also found that evaluating Ec and Dp with the same post-processing as used in regular part building. low working range): 1. 6.Table 19 Exposure Threshold Model Prediction (protocol. the new beam approximation makes the exposure threshold model more complex to utilize in SL. and in fact generally. A more accurate beam profile approximation does improve the predictions of the exposure threshold model. However.) 745 ± 10 957 ± 15 1349 ± 86 1025 ± 52 Ec model 1222 1436 1007 1630 prediction error (%) 65 50 -25 60 experiment (95% C. 108 . the exposure doses fall in the higher range of the protocol working curve. Overlapping-line. but it is still not as good as the DOC threshold model (about 20% less accurate. Single Line (1) Vs = 1. or adopting different ranges of the resin working curve does not improve the predictive ability of the exposure threshold model. the adoption of the higher range of data also gives similarly poor predictions. Furthermore. makes it worse.4 Summary The DOC threshold model was found to be more accurate at predicting the dimensions of single line and multiple line and stacked line parts than the current exposure threshold model.

The material and process optimization can be performed for best performance. The SL performance properties that are investigated and addressed in this chapter are the following: resolution. and identify factors that affect the fabrication results significantly. maximum temperature rise in the resin bath. etc. part draining and cleaning. The full width and maximum depth of a single cured line part hence are referred to as the width and depth resolution.e.CHAPTER 7 MODEL APPLICATIONS As discussed earlier. Here SL resolution is defined as the dimensions of the smallest parts that can be obtained providing certain equipment and materials. speed. and maximum DOC of the green part (i. Note the resolution decreases when the part size increases. the DOC profile can be simulated and the critical DOC applied to predict the cured part dimensions. which also provides guidelines for SL material development and process or laser improvement. given any part building condition. This chapter demonstrates that the process model can also be used to investigate the effects of material and process parameters on the SL performance. The speed defined in the width (or 109 . This capability of the SL cure process model (or critical DOC model) is not only a good verification but a good application of the model. is not taken into account. respectively. the reduction of the speed by the building delay between layers. The speed refers to the part curing speed only. PCA.. for further cure yet). the part that is formed right after the SL building and has not been put in the post-cure apparatus.

The software Minitab (Minitab Inc. The resolution III Plackett-Burman design with 32 runs (corresponding to 2III24-19 fractional factorial design. and the variety of ways people have employed to describe the photopolymerization kinetics. Among these parameters. photoinitiator descriptions in Ciba12. 7. 1996) has been chosen for this purpose. polymer properties (Van Krevelen. Table 20 lists these 24 factors and their two level values which are determined based on SLA systems specifications (Rosen. acrylate monomer descriptions in Sartomer11. polymer handbook (Brandrup and Immergut.com 110 .com www. the effects of kinetic parameters are not investigated due to the complexity.1 Parameter Effect Investigation Any of the parameters involved in the process model (as listed in Tables 1 and 4 in Chapter 3) could affect the SL fabrication results. Yaws’ chemical handbook (Yaws. their strong correlation with one another.sartomer. as well as experience and knowledge about SLA operations.cibasc.depth) direction is characterized by the time that is taken to obtain a single line part with certain width (or depth). 2002). The effect of CTE of the monomer is not tested either since it strongly affects the kinetic values f cp and f ct and thus its factorial effect cannot be tested without kinetic parameters also under investigation.) has been employed for parameter effect investigation and Evolver (Palisade Corporation) for parameter optimization. 2003). This leaves 24 parameters to screen to identify the important ones. 1989). Neter et al. 1990). 11 12 www.

418 Units m/s K W m J/mol 3 m /mol-m wt% K W/m2-K nm W/m-K J/Kg-K J/Kg-K m2/s m2/s m2/s 1/K K K Kg/m3 Kg/m3 Kg/mol Kg/mol For each run of the Plackett-Burman experiment.Table 20 Potential Sensitive Parameters and Their Level Values Parameters laser scanning speed bath temperature laser power beam radius heat of polymerization absorption coefficient of initiator quantum efficiency of initiation Initiator wt% loading chamber temperature heat convection coefficient laser wavelength thermal conductivity heat capacity (monomer) heat capacity (polymer) diffusion coefficient (monomer) diffusion coefficient (macroradical) diffusion coefficient (initiator) CTE (polymer) glass transition temp. for which a 111 .00E-10 1. and results from stepwise selection have been inspected to screen out the unimportant ones.85E+05 60 0.00E-04 3.15 0.00E-10 1. The absolute size of effects.00E-20 1.00E-18 7. and maximum temperature rise during the curing process. Particularly. curing speed defined in width and depth direction. 10 factors appear to affect the depth resolution.15 4.6 1128 1800 2.15 373.15 497.15 0 325 0. 12 factors could affect the maximum temperature rise. Each response is fitted versus these 24 factors. P-values.00E-04 2.1 1500 585 1.00E-18 1. maximum DOC of the cured part.7 0.164 High Level (+1) 0.1 2.00E-12 1. effects plot.156 0.15 0.6 5 303. in which case a resolution III 1/64 fractional factorial experiment (2III10-6) is conducted for further screening.198 0.25 3300 2500 1. normal plot.02 301.15 980 1200 0. six responses are recorded including width and depth resolution. (polymer) density (monomer) density (polymer) molecular weight (monomer) molecular weight (initiator) Symbols Vs Tb PL wo deltH ebx phi wt Ta hfc wL cond CpM CpP Dm DR Ds alphaP Tgm Tgp rouM rouP MWm MWs Low Level (-1) 0.50E-05 173.024 1.1 308.1 1 296.23E-04 223.45E+04 20 0. (monomer) glass transition temp.18 354.

and a follow-up 24 full factorial design for maximum temperature rise. 112 . Table 21 Significant Factors Identified from Screening Experiment Responses Factors beam radius (wo) monomer diffusion coefficient (Dm) monomer glass transition temperature (Tgm) monomer molecular weight (MWm) initiator loading wt% (wt) initiator molecular weight (MWs) initiator absorptivity (ebx) quantum efficiency of initiation (phi) heat of polymerization (deltH) laser scanning speed (Vs) monomer heat capacity (CpM) width resolution X X X X X X X X X X X depth resolution X speed (width) X X max DOC X X X max T rise X The interactions among significant factors are investigated in follow-up 23 full factorial experiments for each of the responses: width resolution. speed in width direction.resolution III 1/256 fractional factorial design (2III12-8) is employed for further screening. and maximum DOC. as shown in Table 21. Table 21 lists for each response the parameters that are identified as significant from the final screening experiment. No parameter seems significant for the speed evaluated in depth direction. A resolution V half-fraction 25-1 factorial design is used to study the factor effects for depth resolution.

with at least 99.e.. The response is evaluated and recorded for each run.001 level or even smaller.9% confidence. Table 22 Full Factorial Design and Response Values (Width Resolution) Run 1 2 3 4 5 6 7 8 Beam radius (wo) -1 1 -1 1 -1 1 -1 1 Monomer diffusion coefficient (Dm) -1 -1 1 1 -1 -1 1 1 Monomer glass transition temperature (Tgm) -1 -1 -1 -1 1 1 1 1 Width Resolution (µm) 250 494 270 500 462 890 480 892 The initial inspection of the absolute size of the effects (Table 23) and normal plot (Figure 47) demonstrate that the active effects are the main effects for laser beam radius and monomer glass transition temperature. 113 .7. Besides. i. The other 21 parameters are fixed at values of the original process model (as used in Chapter 6 for single-line part building).1. and the interaction effect between them. 2002) which provides quantitative evidence for effect significance (Table 23): these three effects can be declared significant at the 0. the monomer diffusion coefficient can be claimed important with almost 90% confidence.1 Sensitive Parameters for Width Resolution Table 22 shows the full factorial design for the three parameters that have been identified as important for width resolution. This is confirmed by a formal test Lenth’s method (Wu and Hamada.

75 164.15) 99 Effect Type Not Significant Significant A C AC F actor A B C N ame wo Dm Tgm 95 90 80 70 60 50 40 30 20 10 5 1 Percent 0 50 100 150 200 Effect 250 300 350 Lenth's PSE = 6.5 91.001 >0.i) 48.75 -0. the fit of the reduced model appears to be good and it has great predictive ability (R2pred = 99.5 11. The statistical significance of the two main effects and one interaction is confirmed as well: almost no risk is taken when claiming these three effects are significant (P-value = 0 for all three effects).101 <0.4 Normal Probability Plot of the Effects (response is width resolution.75 -0.63%).56 0.22 0.001 0.67 1.11 13.75 45. Alpha = .81 1.75 Figure 47 Normal Plot for Width Resolution The regression model is then reduced to retain only the effects that are identified as active.5 -7.07 P-value <0.001 0.70 44. As shown in Appendix D.4 >>0.25 abs(tPSE.5 -1.5 Coef 529.5 302.5 -0.25 5.233 <0. 114 .25 -3.75 151.Table 23 Estimated Factorial Effects and Lenth’s Test for Width Resolution Term Constant Beam radius (wo) Monomer diffusion coefficient (Dm) Monomer glass transition temperature (Tgm) wo*Dm wo*Tgm Dm*Tgm wo*Dm*Tgm Effect 328.

Thus. Furthermore. while with beam radius or monomer Tg increasing.The stepwise selection result also demonstrates the importance of these three effects. This is demonstrated by Lenth’s test results (Table 23) as well: the confidence to claim them significant is about 10% and 20%. 115 . The effect significance of the parameters can also be observed from the main effects and interaction plots (Figure 48). It makes sense that the cured part gets wider when the incident beam gets wider. it appears that monomer diffusion coefficient doesn’t have a noticeable effect on the width resolution. Further investigation later demonstrates that effect of monomer Tg on the width resolution is not a simple linear relationship. a bigger part is built. From the main effects plot. it also indicates the less importance of two additional effects: main effect for monomer diffusion coefficient and interaction between beam radius and monomer diffusion coefficient (Appendix E). respectively. then a decrease in free volume decreases the termination effect relative to propagation. the width resolution decreases (part width increases) significantly. Suppose the critical free volume for propagation and termination are constant. An increase in monomer Tg reduces the free volume of the curing system.

Main Effects Plot (data means) for width resolution (µm) 700 wo Dm Mean of width resolution (µm) 600 500 400 -1 700 600 500 400 -1 1 Tgm 1 -1 1 (a) Main Effects Plot Interaction Plot (data means) for width resolution (µm) -1 800 600 400 1 wo -1 1 wo 800 600 400 Dm -1 1 Dm 800 600 400 Tgm -1 1 T gm -1 1 -1 1 (b) Two-Factor Interaction Plot Figure 48 Factorial Effects Plot for Width Resolution: (a) main effect (b) interaction 116 .

Both wo-against-Tgm and Tgm-against-wo plot are synergistic. Beam radius (and respectively. beam radius) increasing. In summary. the degree to which the width resolution decreases when beam radius (and monomer Tg. the large vertical distance of the overlapped lines in Dm-against-wo plot and correspondingly the large space between two separate lines in wo-against-Dm plot reflect an active effect of beam radius on the width resolution. This relation is also illustrated by the Dm-against-wo plot in which the two joined lines are almost overlapped with each other. respectively) increases depends on high or low monomer Tg (and beam radius. laser beam radius and monomer glass transition temperature are two sensitive parameters that affect the width resolution significantly. low beam radius). which indicates a simple relation: whether beam radius (and respectively. monomer Tg) is high or low.From the interaction plot. monomer Tg) affects the width resolution more significantly at high monomer Tg (and respectively. which leads to unimportant interaction between these two factors as well. beam radius). Furthermore. The beam radius (or monomer Tg) is more sensitive at high monomer Tg (or high beam radius). there is little effect of monomer diffusion coefficient on the width resolution. respectively) being adopted. The sensitivity of beam radius (and respectively. 117 . monomer Tg) depends on the level of monomer Tg (and respectively. it can be seen that at high or low level of beam radius. On the other hand. The width resolution decreases (part width increases) with either of these two factors increasing. The relation between monomer diffusion coefficient and monomer Tg won’t be detailed here since it is similar to the relation between monomer diffusion coefficient and beam radius. high beam radius) than at low monomer Tg (and respectively. the width resolution decreases with monomer Tg (and respectively.

001 0. the effects plot provides more information regarding to interactions between parameters in a graphical view.0252 -0. and reduced model regression reaches a consistent agreement on active effects. a 23 full factorial design is employed for speed defined in width direction to investigate the interaction among beam radius.3994 -0. normal plot.0511 0. The main effects plot below clearly demonstrates the speed increases significantly when beam 118 .0031 0.i) 97.2 Sensitive Parameters for Speed (Width Direction) Similarly.0039 Coef 0.0032 -0. Table 24 Estimated Factorial Effects and Lenth’s Test for Speed (Width) Term Constant Beam radius (wo) Monomer diffusion coefficient (Dm) Monomer molecular weight (MWm) wo*Dm wo*MWm Dm*MWm wo*Dm*MWm Effect -0. stepwise selection.0061 0. Table 24 shows the Lenth’s test results.0256 0.7276 0.005 >0.005 0. The speed is characterized by the time taken to cure a 200 µm wide single line part.0064 -0. and effects plot as well as the results of Lenth’s test.81 6.52 P-value <<0.0019 abs(tPSE.3638 0. from which we can see that these three effects can be declared significant with 99. The inspection of effects size.340 0.0504 -0.361 0.81 0.5% confidence.4 Although the size and sign of the regression coefficients can reveal some information on how the sensitive parameters affect the response.0018 -0.48 0.4 >0.1.72 0.01 0. monomer diffusion coefficient.85 6.0036 -0. and monomer molecular weight. The active effects are identified to be main effects for beam radius and monomer molecular weight and interaction between them.7.

From the interactions plot. 119 . monomer molecular weight seems more important at low beam radius than at high beam radius. Little time is needed when the beam size is large.2. On the other hand.3) demonstrates that the effect of beam radius on curing speed is quite nonlinear. beam radius is found to be a more sensitive parameter than monomer molecular weight for width speed. The effect of beam radius is slightly more significant at higher monomer molecular weight than at lower molecular weight. and thus the increase in curing speed.radius increases and increases slightly when monomer molecular weight decreases. The decrease in monomer molecular weight leads to the increase in monomer molar concentration. Further investigation later (Section 7.

0 -1 1 -1 1 (b) Two-Factor Interaction Plot Figure 49 Factorial Effects Plot for Speed (Width): (a) main effect (b) interaction 120 .0 0.6 0.0 -1 0.8 0.8 wo Dm Mean of time (sec) (width direction) 0.2 0.0 MW m -1 1 0.6 0.4 0.Main Effects Plot (data means) for time (sec) (width direction) 0.8 Dm -1 1 Dm 0.4 0.8 1 wo -1 1 wo 0.2 0.4 0.8 0.4 0.4 M Wm 0.0 -1 1 MW m 1 -1 1 (a) Main Effects Plot Interaction Plot (data means) for time (sec) (width direction) -1 0.

004 -0.078 P-value 0.0005 Coef 0. Both Lenth’s test and stepwise selection also show that the interaction effect between monomer diffusion coefficient and monomer Tg could be important as well. but could eventually contribute to DOC considering diffusion limitation. thus increases DOC. an increase in monomer Tg decreases termination effect.0075 0.7.1.078 14.2% confidence.23125 -0. Table 25 Estimated Factorial Effects and Lenth’s Test for Maximum DOC Term Constant Monomer diffusion coefficient (Dm) Monomer glass transition temperature (Tgm) Monomer molecular weight (MWm) Dm*Tgm Dm*MWm Tgm*MWm Dm*Tgm*MWm Effect -0.00025 0.4 The main effects plot below demonstrates that the DOC of the part increases with monomer Tg or molecular weight increasing.82 14.59 1.047 0.75 0. High monomer molecular weight (Low monomer concentration) decreases the curing speed. and monomer glass transition temperature.4 <0.015 0.001 >>0. Further investigation later shows a 121 .1455 0.008 -0. while the main effect for monomer diffusion coefficient is identified as active with 95.048 <0.3 Sensitive Parameters for DOC Similarly. The Lenth’s test (Table 25) shows the main effects for monomer Tg and molecular weight and interaction between them can be declared significant with at least 99.094 0.093 0. a 23 design full factorial experiment has been conducted for maximum DOC to investigate the interaction among monomer diffusion coefficient. monomer molecular weight.35 22.25 0.9% confidence.i) 2.0005 0.07275 0.001 <0.001 0.189 >>0.0465 0.00025 abs(tPSE. The active effects are identified to be main effects for all three parameters mentioned above as well as interaction between monomer molecular weight and monomer Tg. As mentioned earlier.

molecular weight is much more sensitive at high than at low monomer Tg. Furthermore. High monomer diffusion increases monomer composition at the centerline of the part. Similarly. the interaction between monomer Tg and monomer molecular weight appears significant. which again demonstrates characteristics of diffusion-limited reaction. monomer Tg is more sensitive at high that at low monomer molecular weight. and leads to a slight decrease in DOC.slight curvature in monomer molecular weight and significant nonlinear behavior in monomer Tg. 122 . As shown in interaction plot above.

20 0.25 Tgm Mean of max DOC 0.3 MWm 0.4 Tgm -1 1 0.4 Dm -1 1 0.2 MWm -1 1 -1 1 -1 1 (b) Two-Factor Interaction Plot Figure 50 Factorial Effects Plot for Max DOC: (a) main effect (b) interaction 123 .15 -1 MWm 0.3 Dm 0.3 T gm 0.2 0.30 0.4 1 0.20 0.30 0.25 0.15 -1 1 1 -1 1 (a) Main Effects Plot Interaction Plot (data means) for max DOC -1 0.2 0.Main Effects Plot (data means) for max DOC Dm 0.

02 0.3 6.58 7.029 0.29 >0.41 2.12 28.91 0.418 0.4 >0.035 0.4 >0.27 0.019 0.4 Sensitive Parameters for Temperature Rise A full factorial (24) design is adopted to investigate interactions between the four parameters that are identified as important for temperature rise in the resin vat.4 >>0.06 1.13 -2.71 0.17 -0. etc.54 abs(tPSE.039 0. These additional effects are identified by Lenth’s method as less likely to be important (about 70% or even less).06 14.12 2.i) 1.17 1.01 -20.27 0.47 0.59 -27.1. The stepwise regression selects nine additional factorial effects such as main effect for monomer heat capacity.34 -0. Table 26 Estimated Effects and Lenth’s Test for Maximum Temperature Rise Term Constant Laser scanning speed (Vs) Heat of polymerization (deltH) Monomer molecular weight (MWm) Monomer heat capacity (CpM) Vs*deltH Vs*MWm Vs*CpM deltH*MWm deltH*CpM MWm*CpM Vs*deltH*MWm Vs*deltH*CpM Vs*MWm*CpM deltH*MWm*CpM Vs*deltH*MWm*CpM Effect -18.34 0.01 -10. The Lenth’s test (Table 26) shows heat of polymerization and monomer molecular weight are sensitive parameters.4 124 .75 1.59 0.4 >0.79 3.67 0. Main effects for these two factors and their interaction effect can be declared important at a risk of <10%.07 Coef 19. The main effect for scanning speed is claimed to be active with >85% confidence.48 -6.7.26 -5.89 -4.91 2.18 1.3 -13.4 >>0.95 0. The significance of these four effects has been confirmed by the regression of the reduced model (only four effects are included in the model) which turns out to have a good fit quality.4 0.77 -8.6 12.22 3.49 0.95 -12.53 2.2 -9.06 1. The normal plot also illustrates these three effects are significant.138 0.11 1.099 >0. interactions between scanning speed and monomer molecular weight.83 5.09 P-value 0.4 >>0.

and higher monomer molecular weight causes lower monomer concentration. monomer molecular weight (“MWm”). 125 . higher monomer heat capacity leads to smaller temperature rise. as shown in the main effects plot in Figure 51. Monomer molecular weight is a more sensitive parameter at high heat of polymerization. Both of them lead to less amount of reaction and thus generate less heat. For the same amount of heat generated. or monomer heat capacity (“CpM”) increases. The temperature rise decreases as laser scanning speed (“Vs”). The interactions plot shows that the temperature rise is more sensitive to the heat of polymerization at low monomer molecular weight. An increase in heat of polymerization (“deltH”) leads to more heat generated for the same amount of reaction and thus larger temperature rise in the resin. The effect of monomer heat capacity is less significant as observed.The effects plot (Figure 51) demonstrates how the significant parameters affect the temperature rise. This makes physical sense since higher laser moving speed deposits less energy to the resin. Other interaction effects shown in Figure 51 (b) are less significant and not discussed here.

Main Effects Plot (data means) for max T rise (K) Vs 30 20 10 -1 MW m 30 20 10 -1 1 -1 1 1 -1 CpM 1 deltH Mean of max T rise (K) (a) Main Effects Plot Interaction Plot (data means) for max T rise (K) -1 1 -1 1 50 Vs 25 0 50 25 0 50 MWm 25 0 50 25 0 -1 1 -1 1 CpM deltH Vs -1 1 deltH -1 1 MW m -1 1 CpM -1 1 (b) Two-Factor Interaction Plot Figure 51 Factorial Effects Plot for Temperature Rise: (a) main effect (b) interaction 126 .

1.27 0.4 -3.1 -89.7. All twofactor interactions are clear of other two-factor interactions.4 0.2 434.8 136.023 0.8 49 abs(tPSE.5 1. Table 27 Estimated Effects and Lenth’s Test for Depth Resolution Term Constant Initiator absorptivity (ebx) Quantum efficiency of initiation (phi) Initiator wt% (wt) Monomer diffusion coefficient (Dm) Initiator molecular weight (MWs) ebx*phi ebx*wt ebx*Dm ebx*MWs phi*wt phi*Dm phi*MWs wt*Dm wt*MWs Dm*MWs Effect -204. Ignoring high order interactions.9 -18. and aliased with three-factor interactions.4 0.9 93. The stepwise selection 127 . and confounded with four-factor interactions.25 1.24 0.023 0.3 41 -108 -10 -303.6 -18.4 >>0.4 0.4 0.5 -54 -5 -151. Main effects for these two factors and their interaction effect can be declared important at a risk of less than 10%. For a design of this resolution.5 98 Coef 386 -102.445 >>0.216 >>0.1 217.93 0.354 0.4 68.070 >0.234 >0.88 1.16 0.63 P-value 0.2 -37.96 0.067 >0.065 1. the Lenth’s test illustrates that the quantum efficiency of initiation and initiator loading are sensitive parameters.4 0.5 Sensitive Parameters for Depth Resolution A resolution V five-factor half-fraction (2V5-1) factorial design is adopted to investigate interactions between the five parameters that are identified as important.184 0.2 -148. all main effects are clear of other main effects and two-factor as well as three-factor interactions.7 -36.478 As shown in Table 27. all main effects and two-factor interactions can be clearly evaluated.8 -7 3.8 186.6 20.i) 1.81 1.70 0.045 0.3 -179.21 0. The normal plot also illustrates these three effects are significant.32 2.5 -297.

This indicates that an increase in penetration depth does not necessarily lead to an increase in the cure depth. etc. The interaction between “ebx” (or “MWs”) and “Dm” is antagonistic. This reveals another side of monomer diffusion effect: faster diffusion facilitates the movement and reaction of the species thus to form a bigger part.demonstrates six additional important factorial effects such as main effects for initiator absorptivity and initiator molecular weight and interactions between quantum efficiency and initiator molecular weight. the cure depth decreases when the initiator molecular weight increases (beam penetration depth increases). The interactions plot in Figure 52 (b) shows that quantum efficiency (“phi”) plays a bigger role at low absorption coefficient (“ebx”) and low initiator loading. Other interaction effects shown in Figure 52 (b) are minor and are not discussed here. These additional effects are identified by Lenth’s method as less likely to be important (about 80% or even less. Higher quantum efficiency of initiation (“phi”) generates more radicals to grow a bigger part. The effects of monomer diffusion coefficient (“Dm”) and initiator molecular weight (“MWs”) are much less significant. Either an increase in initiator wt% loading (“wt”) or initiator absorption coefficient (“ebx”) leads to a smaller beam penetration depth and thus an increase in depth resolution. quantum efficiency is low) to cure the part tip enough. see items in red and not bold in Table 27). The phi-against-MWs plot is antagonistic as well. the cure depth increases slightly with monomer diffusion increasing. At low absorption coefficient (or high initiator molecular weight). as shown in Figure 52 (a). There might not be enough radicals (in this case. At low quantum efficiency. 128 . while absorption coefficient and initiator loading are more sensitive at high quantum efficiency.

Main Effects Plot (data means) for depth resolution (µm) ebx 600 500 phi wt Mean of depth resolution (µm) 400 300 200 -1 Dm 600 500 400 300 200 -1 1 -1 1 1 -1 MWs 1 -1 1 (a) Main Effects Plot Interaction Plot (data means) for depth resolution (µm) -1 900 600 ebx 300 900 600 300 900 600 wt 300 900 600 300 900 600 MWs 300 -1 1 -1 1 -1 1 1 -1 1 ebx -1 1 phi -1 1 wt -1 1 Dm -1 1 MWs -1 1 phi Dm (b) Two-Factor Interaction Plot Figure 52 Factorial Effects Plot for Depth Resolution: (a) main effect (b) interaction 129 .

Table 28 Significant Factors for Investigated Responses Responses Factors beam radius (wo) monomer diffusion coefficient (Dm) monomer glass transition temperature (Tgm) monomer molecular weight (MWm) initiator loading wt% (wt) initiator molecular weight (MWs) initiator absorptivity (ebx) quantum efficiency of initiation (phi) heat of polymerization (deltH) laser scanning speed (Vs) monomer heat capacity (CpM) width resolution XX X XX depth resolution speed (width) XX max DOC XX XX XX max T rise XX XX X X XX XX XX X 130 . “XX” denotes more significant parameters.Table 21 from the screening experiment is revised according to the follow-up effect significance investigation above for each response. while “X” denotes relatively less important ones. In Table 28.

doesn’t have a good fit or good predictive ability. the square of the factor actual value is used to represent the quadratic effect. The regression model based on these three effects. however. A useful assumption is made that the function is bilinear in which the linear terms model the effect of influential factors and bilinear terms model the important interaction effects. In case of the existence of curvature effect.2 Resolution and Speed Prediction by Regression Model As is well known. and process or laser parameters.7. The predictive ability of regression models is verified. The conditions for the simulations are shown in Table 29. The regression equation of this revised model as well as its good fit quality and predictive ability is demonstrated in Appendix F. To obtain an explicit function of resolution or speed in terms of these properties is almost impossible. The other important effects identified by stepwise selection are then included into the regression model as well. reaction kinetics. and the results and comparison with regression model predicted results are shown in Table 30. 131 . three simulations have been conducted using the SL cure process model. The response is then fitted versus the involved parameters and combinations. To verify this regression model. SL curing resolution and speed can be influenced by a lot of factors such as material properties.1 Regression Prediction Model for Depth Resolution Recall that the active effects for depth resolution are main effects for quantum efficiency of initiation and initiator wt% loading and their interaction effect. The prediction error of the regression model is found to be within 15 %. 7.2.

Table 29 Simulation Conditions to Test Predictive Ability of Regression Models Condition beam radius (wo) monomer diffusion coefficient (Dm) monomer glass transition temperature (Tgm) monomer molecular weight (MWm) initiator loading wt% (wt) initiator molecular weight (MWs) initiator absorptivity (ebx) quantum efficiency of initiation (phi) I 1.74 0.5 5 0.00E-14 210 1.8 III 155 175 12.7 7.2.87 0.50E-04 1.4 60 0.10E-04 1. three levels for monomer diffusion coefficient and monomer Tg) full factorial design is then adopted in order to capture the curvature effect.6 II 1. A three-factor-mixed-level (two levels for beam radius.00E-14 180 1 1 0.3 III 2.2 40 0. The addition of the two less active effects (main effect for monomer diffusion coefficient and its interaction with beam radius) makes the residual plot versus the predictor monomer diffusion coefficient appear like a parallel band centered about zero.00E-14 208.7 II 525 474 -9.2 Regression Prediction Model for Width Resolution The prediction error of the fitted bilinear model versus the important effects (main effects for beam radius and monomer Tg and interaction between them) is up to 60%. 132 .00E-04 1. Further investigation reveals the possibility of curvature presence in monomer diffusion coefficient and monomer Tg predictors (Figure 53). but doesn’t improve the model predictive ability significantly.1 units m m2/s K Kg/mol wt% Kg/mol m3/mol-m Table 30 Depth Resolution Predicted by Regression Model Condition simulation results (µm) predicted results (µm) prediction error (%) I 810 872 7.53 2 0.26 19. as it should be.

less significant and almost linear effect in monomer diffusion coefficient. 133 .500 width resolution (µm) 450 400 350 300 250 200 1 2 factor level 3 wo Dm Tgm Figure 53 Curvature in Factors for Width Resolution The effects plot below demonstrates significant linear effect in beam radius. and significant curvature effect in monomer glass transition temperature.

Main Effects Plot (data means) for width resolution (µm) 700 wo Dm Mean of width resolution (µm) 600 500 400 300 1 700 600 500 400 300 1 2 3 Tgm 2 1 2 3 (a) Main Effects Plot Interaction Plot (data means) for width resolution (µm) 1 800 600 400 2 3 wo 1 2 wo 800 600 400 Dm Dm 1 2 3 800 600 400 T gm Tgm 1 2 3 1 2 1 2 3 (b) Interaction Plot Figure 54 Curvature Effect for Width Resolution: (a) main effect (b) interaction 134 .

8 time (sec) (width) 0. regression analysis has been performed based on stepwise selection results. and wo*Tgm to represent the interaction between beam radius and monomer Tg.4 0. The regression model appears to have a good fit and good predictive ability. The verification is shown in Table 31.6 0. 1 0.6 7. Further investigation demonstrates that the effect of beam radius on the curing speed is quite nonlinear (Figure 56). Table 31 Width Resolution Predicted by Regression Model Condition simulation results (µm) predicted results (µm) prediction error (%) I 281 332 18.0 III 488 613 25.2 0 1 2 factor level 3 wo MWm Figure 55 Curvature in Factors for Speed (Width) 135 . and Tgm^2 to represent the quadratic effect of monomer Tg.3 Regression Prediction Model for Speed (Width Direction) Curvature has been found in beam radius and in monomer molecular weight (Figure 55) for this response.3 II 342 329 -4. using Tgm to represent the linear effect of monomer Tg.As shown in Appendix G.2.

and the laser assumed to be Gaussian. This. 136 . It’s inappropriate to find a single model to capture all the behavior. A two-factor-three-level 32 design is performed to capture the curvature effect. a lower intensity irradiates the resin inside the 100µm diameter spot. however. This makes sense recalling the speed is characterized by the time taken to cure 200 µm wide part. 200 µm taken as the low level of spot size. To go around this issue.8 time (sec) (width) 0. As shown in Figure 57. The effects plot is shown in Figure 57.1 0.6 0. brings complexity to the regression analysis.2 0 50 100 150 beam radius (µm) 200 250 Figure 56 Nonlinear Behavior of Beam Radius for Speed (Width) The speed increases dramatically when spot size (beam diameter) increases slightly above 200 µm and then almost keeps constant when the spot size increases further. the curing time increases as laser beam radius increases because for the same power and bigger beam size. the time consumed to cure 100 µm wide part is used to characterize the curing speed (200 µm wide part is critical here because it overlaps the low level value of spot size).4 0.

024 0.036 wo MWm Mean of time (s) (width) 0.036 0.028 0.036 0.Main Effects Plot (data means) for time (s) (width) 0.032 0.020 1 2 3 1 2 3 (a) Main Effects Plot Interaction Plot (data means) for time (s) (width) 1 2 3 0.028 wo 0.024 0.032 0.020 1 2 3 MWm MW m 1 2 3 wo 1 2 3 (b) Interaction Plot Figure 57 Factors Curvature for Speed (Width): (a) main effect (b) interaction (Level “2” value of beam radius = 110µm) 137 .028 0.032 0.024 0.020 0.

The verification is shown in Table 32. The regression model appears to have a good fit and good predictive ability.22 maximum DOC 0.A regression analysis has been performed (as shown in Appendix H) based on stepwise selection results.0185 0 II 0.18 Tgm MWm 0.0185 0. curvature has been found in monomer glass transition temperature and slightly in monomer molecular weight (Figure 58). 0.14 1 2 factor level 3 Figure 58 Curvature in Factors for Maximum DOC 138 .16 0. The effects plot is shown in Figure 59.1 III 0.0336 -8. Table 32 Curing Time (Width Direction) Predicted by Regression Model Condition simulation results (sec) predicted results (sec) prediction error (%) I 0.7 7.2.0368 0. A three-factor-three-level 33 design has been performed to detect and capture the curvature effect.2 Dm 0.0294 0.0235 -20.4 Regression Prediction Model for Maximum DOC Similarly.

30 0.4 2 3 0.25 Dm Tgm Mean of max DOC 0.35 0.20 0.30 0.3 M Wm 0.25 0.2 0.15 1 0.2 0.15 1 2 3 2 MW m 3 1 2 3 (a) Main Effects Plot Interaction Plot (data means) for max DOC 1 0.3 T gm 0.Main Effects Plot (data means) for max DOC 0.4 Tgm 1 2 3 0.35 0.4 Dm 1 2 3 0.20 0.3 Dm 0.2 MW m 1 2 3 1 2 3 1 2 3 (b) Interaction Plot Figure 59 Factors Curvature for Maximum DOC (a) main effect (b) interaction 139 .

The regression model appears to have a good fit and good predictive ability. 120 100 80 60 40 20 0 1 2 factor level 3 max T rise (K) Vs deltH MWm CpM Figure 60 Curvature in Factors for Maximum Temperature Rise 140 .125 -8.5 Regression Prediction Model for Maximum Temperature Rise Curvature has been found more significant in monomer molecular weight.193 0. Table 33 Maximum DOC Predicted by Regression Model Condition simulation results predicted results prediction error (%) I 0.1 II 0.137 0. The effects plot is shown in Figure 61.2. and negligible in monomer heat capacity (Figure 60).198 2. and heat of polymerization.As shown in Appendix I. laser scanning speed. A three-factor-three-level 33 design has been conducted to detect and capture the curvature effect.180 -27.247 0. regression analysis has been performed based on stepwise selection results.6 7.8 III 0. The verification is shown in Table 33.

Main Effects Plot (data means) for max T rise (K) Vs 40 30 deltH Mean of max T rise (K) 20 10 0 1 40 30 20 10 0 1 2 3 2 MW m 3 1 2 3 (a) Main Effects Plot Interaction Plot (data means) for max T rise (K) 1 2 3 50 25 0 50 deltH 25 0 50 25 0 1 2 3 1 2 3 M Wm Vs Vs 1 2 3 deltH 1 2 3 MW m 1 2 3 (b) Interaction Plot Figure 61 Factors Curvature for Max Temp Rise (a) main effect (b) interaction 141 .

1 1850 units Kg/mol J/mol m/s J/Kg-K 142 .0 23.06 2000 III 1 3.7 2. Table 34 Maximum Temperature Rise Predicted by Regression Model Condition experimental results predicted results prediction error (%) I 43.Regression analysis is performed referring to stepwise selection results.528 2.00E+04 0.7 55. The verification is shown in Table 34. Table 35 lists the conditions used to test the regression model.45E+04 0.9 1. It turns out that monomer heat capacity needs to be considered in order for the model to have a good fit.85E+05 0. As shown in Appendix J. the regression model appears to have a good fit and good predictive ability.7 II 1.3 Table 35 Conditions used for Test of Temperature Rise Regression Model Condition MWm deltH Vs CpM I 0.5 8.0 15.9 III 0.0272 1500 II 1.8 27.

In this work. a i ≤ x i ≤ bi (69) 143 . hatch over cure.7.e.t. Evolver (Palisade) has been used to solve the optimization problem. response optimization can be performed to find the highest resolution or speed or maximum DOC as well as their corresponding parameter settings. the optimization performed for part dimensions is more focused on the smallest feature that SL can build given any material or apparatus available. The optimization problem is formulated as follows: min or max y = f ( x1 . The variation bounds for design variables (i. x 3 . It is an add-in for Microsoft Excel that uses genetic algorithms to perform optimization. significant factors) in Equation (69) are listed in Table 36. x 2 . for DOC) subject to the constraints Equation (69). hatch style. Based on the regression prediction models established in the previous section. and fill cure depth in order to obtain parts with desired dimensions. hatch spacing.g. Chockalingam and coworkers (2003) performed optimization on operation parameters such as layer thickness.3 Parameter Optimization For a given stereolithography apparatus and given photosensitive material. for resolution and speed) or maximized (e.g. Equation (68) shows the objective function to be minimized (e. ) (68) s.

00E-14 1. initiator wt% loading. Relatively lower monomer glass transition temperature would be preferred for the concern of resolution. which however would also decrease the DOC. while for depth resolution easier diffusion of monomer is preferable.156 5 0.1 3. It can be seen that high quantum efficiency and initiator loading are not necessarily good for SL.164 20 0. For the above responses of interest.00E-14 1.1 3300 Units m m2/s K Kg/mol wt% Kg/mol m3/mol-m N/A J/mol m/s J/Kg-K Table 37 lists the optimization results for each response. and initiator molecular weight are parameters that only affect depth resolution.6 2.85E+05 0. The temperature rise in the resin due to the heat generated by curing can be reduced to negligible value by adjusting the laser scanning speed and adopting material with smaller heat of polymerization and higher heat capacity.00E-12 2.00E-12 0. Monomer molecular weight is another parameter to adjust in need of higher DOC or curing speed.02 1500 High Level 2.198 1 0. It appears that a small beam contributes to improvement in both width resolution and curing speed. initiator absorption coefficient. Low monomer diffusion coefficient benefits width resolution and DOC. most of the important factors are material 144 .00E-04 1. The quantum efficiency of initiation.45E+04 0.Table 36 Parameter Range Used for Response Optimization Factors beam radius (wo) monomer diffusion coefficient (Dm) monomer glass transition temperature (Tgm) monomer molecular weight (MWm) initiator loading wt% (wt) initiator molecular weight (MWs) initiator absorptivity (ebx) quantum efficiency of initiation (phi) heat of polymerization (deltH) laser scanning speed (Vs) monomer heat capacity (CpM) Low Level 1.00E-04 1. High absorptivity and low molecular weight of initiator are preferred for the depth resolution due to the correspondingly smaller penetration depth of the laser beam.418 60 0.

00E-04 1. if we have a 60 µm depth resolution.164 60 0. kinetic 145 .1 215 µm 60 µm 3300 0K Kg/mol wt% Kg/mol m3/mol-m N/A J/mol m/s J/Kg-K 18 ms 34% Table 37 demonstrates the smallest part that can be obtained is about 215 µm wide (slightly larger than the beam diameter) and 60 µm deep. bath temperature. There might be other criteria to evaluate the SL process.17 Depth Resolution Max DOC Max T rise Units m m2/s K 1.00E-14 223. For example. This would eventually decrease the part fabrication speed (not curing speed). the layer thickness must be half or less of standard layer thickness 4 mils (100 µm) in order for neighboring layers to attach together.45E4 0.1 3.05 1.00E-12 1. and heat convection at resin surface are comparatively insignificant.property parameters.15 1.00E-14 186. Table 37 Evolver Optimization Results for Investigated Responses Responses Optimal Conditions Factors beam radius (wo) monomer diffusion coefficient (Dm) monomer glass transition temperature (Tgm) monomer molecular weight (MWm) initiator loading wt% (wt) initiator molecular weight (MWs) initiator absorptivity (ebx) quantum efficiency of initiation (phi) heat of polymerization (deltH) laser scanning speed (Vs) monomer heat capacity (CpM) Optimal Responses Curing Time (width) 1.22 1 0.50 1. Recall in the effect significance investigation. these values could be different for resins with different kinetics. This is evaluated based on the kinetic parameters of the model tetraacrylate material.00E-04 Width Resolution 1. Process parameters such as laser power.

146 .4. Regression equations will be established to predict the cured part dimension and parameter optimization be performed to obtain the best resolution. the effect of parameters on cure depth and line width will be investigated and sensitive parameters be identified using the exposure threshold model.parameters were not tested due to the complexity and the variety of ways people have employed to describe the photopolymerization kinetics.1 Parameter Significance Investigation The parameters involved in the exposure threshold model are shown in Table 38 and a 25 full factorial design with 32 runs has been conducted to study the factor effects. the exposure threshold model only predicts the cured line depth (depth resolution) and width (width resolution). 7. The effect analysis of parameters has provided guidelines to perform cost-effective trials to improve SL fabrication performance. In this section. For each run. the cure depth and line width of a single line part are calculated using Equations (55) and (56) and recorded. It doesn’t provide information for part DOC or curing speed as the DOC threshold model. 7.4 Parameter Analysis using Exposure Threshold Model As described earlier.

The line width is not subject to the change of depth of penetration. Note that Equation (57) can be rewritten as: Lw = w0 2 ln( E max ) Ec (70) Equation (70) obviously shows that the line width is not dependent on the depth of penetration.2 100 Units mJ/cm2 mils mW mm mm/s The analysis of the design illustrates that all five parameters listed in Table 38 are sensitive parameters for cure depth. respectively. and beam radius increase.1 20 High level (+1) 20 10 100 0.Table 38 Parameters in Exposure Threshold Model and Their Level Values PARAMETER critical exposure depth of penetration laser power beam radius laser scanning speed Symbol Ec Dp PL wo Vs low level (-1) 5 5 24 0. laser scanning speed. only expressing the relations in a graphical way. it can be seen that the cure depth increases as laser power and depth of penetration increase and decreases as critical exposure. The main effects plots for cure depth and line width are shown in Figures 62 and 63. From Figure 62. These observations make physical sense and are consistent with Equations (56) and (57). it can be seen that the line width increases as laser power and beam radius increase and decreases as critical exposure and laser scanning speed decrease. 147 . From Figure 63. while four of the parameters (depth of penetration excluded) are significant factors for cured line width.

36 0.30 -1 1 -1 1 1 -1 Vs 1 -1 1 Figure 63 Main Effects Plot for Line Width 148 .48 0.48 0.36 0.54 0.42 Dp PL Mean of Lw (mm) 0.54 0.Main Effects Plot (data means) for Cd (mils) Ec 40 35 30 Dp PL Mean of Cd (mils) 25 20 -1 wo 40 35 30 25 20 -1 1 -1 1 1 -1 Vs 1 -1 1 Figure 62 Main Effects Plot for Cure Depth Main Effects Plot (data means) for Lw (mm) Ec 0.42 0.30 -1 wo 0.

laser power. The significant factors are initiator wt% loading and quantum efficiency of initiation and less significant ones are initiator molecular weight and initiator absorptivity.Recall the parameter effects on depth resolution (cure depth) investigated using DOC threshold model. The significant factors are beam radius and monomer glass transition and a less significant one is monomer diffusion coefficient. The beam radius. not much information regarding to material property effect can be extracted from exposure threshold model. Both models tell that the beam radius is a significant factor for line width and depth of penetration is an insignificant one. As discussed earlier. and laser scanning speed are insignificant comparing to these parameters. By the DOC threshold model. the material parameters seem more important than process parameters such as laser power and scanning speed. However. the depth of penetration is dependent on absorptivity. From this point of view. initiator molar concentration (molecular weight and wt% loading of the initiator). while the critical exposure by definition appears dependent on quantum efficiency of initiation. the parameter significance is consistent by either exposure or DOC threshold model. Recall the parameter effects on width resolution (line width) investigated using DOC threshold model. 149 .

The objective here is to minimize the cure depth and line width. The optimization results by Evolver are demonstrated in Table 39. as shown in Table 39. Therefore. Table 39 Evolver Optimization Results using Exposure Threshold Model Factors Conditions Responses cure depth (Cd) line width (Lw) units critical exposure (Ec) 20 20 mJ/cm2 depth of penetration (Dp) 5 mils laser power (PL) 24 24 mW beam radius (wo) 0. lower laser power. From the effect analysis using DOC threshold model. Furthermore. and faster drawing speed contribute to smaller cure depth.1 mm scanning speed (Vs) 100 100 mm/s Optimal Response 7.2 0. we can see that when the variation of material properties also becomes an option.4. higher critical exposure. lower laser power. we can see that higher critical exposure. larger beam size. The variation bounds of the parameters are as shown in Table 38. smaller penetration depth.7. However.2 Parameter Optimization The optimization is performed based on the exposure threshold model Equations (56) and (57). the depth of the cured line has to compromise with the width to decide desirable beam radius. This conclusion confirms the observations in the previous section and can also be drawn from the expressions of Equations (56) and (57). to obtain a smaller size of part. smaller beam size. smaller beam size is favorable for a small part since the depth is no more significantly dependent on the beam radius.2 units mils mm From Table 39.8 0. the smallest part size is the same 150 . and faster drawing speed lead to smaller line width. resin with higher critical exposure and smaller penetration depth as well as laser with lower power and faster drawing speed is preferred. The optimization problem is formulated similarly to Equations (68) and (69).

it’s slightly larger than the beam size due to diffusion and reaction in the vicinity of irradiation. the critical exposure is not an inherent property of the material. 151 . the parameter analysis based on the exposure threshold model cannot provide a useful guide for material development. In summary. the SL curing speed. while according to the analysis using DOC threshold model. As discussed in the previous chapter. unlike other material properties such as absorptivity or molecular weight. there are limited parameters to adjust when using the exposure threshold model to guide the SL fabrication process. Furthermore. Therefore.as the beam size. what parameters affect the temperature rise during building. and therefore has to be experimentally re-evaluated each time these conditions vary. It is very specific to a certain SL apparatus and certain fabrication as well as post-processing conditions. Only two parameters are used to characterize material properties: critical exposure and depth of penetration. as well as the DOC of the part and how they affect these properties can not be investigated based on the exposure threshold.

transient intensity rather than the exposure incident on the resin should be adopted in stereolithography process simulation.CHAPTER 8 CONCLUSIONS & RECOMMENDATIONS It can be concluded from this research that: 1. but parts with overlapping lines and stacked lines. monomer. and provides insight into the part building mechanisms. initiator. In other words. cannot be ignored and it’s not appropriate to consider the dark reaction only. The cure process model can be employed to investigate transient profiles of temperature. (also see Tang. the irradiation period. The stereolithography cure process model established here can be used not only to simulate and predict the cured size of a single line part. Y. Due to the rapid radical photopolymerization. although very short (1~10ms scale). the photopolymerization kinetic model can be extended to use in stereolithography with high laser intensity. and radical concentrations. 152 . as well as their related properties such as propagation and termination rate. 2. 2002) 3. The simulation and prediction of parts with more complex laser drawing patterns can also be expected.. It simulates the cure behavior during SL fabrication process. 4. By taking diffusion limitation into account.

both diffusion and convection should be considered in the mass balance equations. monomer glass transition temperature as well as monomer diffusion coefficient are sensitive parameters for width resolution. In this case. material or laser properties can be modified to improve stereolithography speed or resolution. Flach and Chartoff (1994) developed a simple polymer shrinkage model based on the degree of conversion from monomer to polymer and estimated the part shrinkage after the cure reaction. Based on knowledge of the effective parameters. The concept of critical DOC renders the cure process model a DOC threshold model.5. the shrinkage also causes the occurrence of convection and 153 . there is a big difference between the density of monomer and that of polymer). They took the effect of conversion on the shrinkage into account. In effect. molecular weight and absorption coefficient as well as quantum efficiency of initiation are parameters that affect the depth resolution significantly. corresponding to the concept of critical exposure and exposure threshold model. the initiator wt% loading. the laser beam radius and monomer molecular weight are two factors that affect the curing time needed to form a certain width part. The laser beam radius.e. The following is recommended for further study: 1. the monomer diffusion coefficient. glass transition temperature. 6. If a significant shrinkage occurs upon the laser scanning (i. and molecular weight are sensitive parameters for the maximum degree of cure a part can reach. the convection phenomenon occurs. 7.

However.) Finally. or material properties and to prepare material development guidelines for SL improvement. 2000. This knowledge would enrich the material development guidelines prepared here. et al. it should be mentioned that the possibility of extending the cure process model without changing material properties or kinetics to SL commercial resin of confidential compositions has been tested. One big benefit of the SL cure process model established here is to investigate the effects of process. In order to incorporate the interactive effects between shrinkage and conversion into the SL process model. 1999). 154 . et al. the prediction error is found to be up to 30 %. Steinmann... This indicates that the material properties or the kinetics do need to change accordingly when the resin changes. (The modeling approach presented in this work can be extended to epoxy resin or acrylate-epoxy blend. Cationic polymerization plays a significant role in SM 7110 curing since it’s composed of both acrylate and epoxy. et al. 1995. this was not considered in their work. a model epoxy curing system (monomer and photoinitiator) should be established and its cationic photopolymerization kinetics characterized in order to investigate the epoxy cure behavior and its interaction with the acrylate cure. That the exposure threshold model does not need to take material change into account is a defect and also an advantage. It turns out that when extending the application of the process model from the model acrylate studied here to commercial SL resin SM 7110. laser. Pang.. Since most of SL resins are comprised of not only acrylate but epoxy (Melisaris. 2000.thus affects the monomer conversion. 2. a convection term needs to be addressed and incorporated.

APPENDIX A NOVECURE OUTPUT WITH 365NM FILTER (EXFO) 155 .

APPENDIX B NEGLIGIBLE HEATING EFFECT OF LIGHT IN DPC EXPEIRMENTS 10 50C (continuous irradiation) 8 Heat Flow (W/g) 6 4 2 16.00min shutter on 0 5 10 negligible heating effect of light 15 20 25 Time (min) 156 .00min shutter off 0 -2 Exo Up 1.

APPENDIX C A BRIEF LITERATURE REVIEW ON GEL POINT ESTIMATION 157 .

The following assumptions were used: (i) all functional groups of each kind of structural unit. A and B. the probability that a given functional group of a branch unit leads via a chain of bifunctional units to another branch unit). whichever has a functionality greater than two. A and B. Dc. (ii) the condensation between A and B on the same molecule is negligible. Then the gel point. A or B. cyclization can be ignored. for the condensation of the two functional units. D. and f is the functionality of the branching unit. 1953): αc = 1 f −1 (72) where αc is the critical value of α.e.Gel point is a critical value for the extent of reaction above which the produced polymer becomes of infinite molecular weight. α = D2 (71) Equation (72) states the critical condition for formation of infinite networks (Flory. can be reduced to be the square of the conversion of the functional units. If A and B groups are initially present in equivalent quantities and only one of them has a functionality greater than two. i. are equally reactive. can be obtained from the following equation: 158 . Flory (1953) developed a theory to estimate the gel point for the nonlinear condensation polymerization systems. i. α. the reactivity of an A or B group is independent of the size or structure of the molecule to which it is attached. the branching probability (i.e.e.

probability that a growing chain adds one more unit. (1979). and obtained the gel point as the conversion where the molecular weight diverges. 1976). and the mole fraction of functional groups. Their results are in agreement with Flory’s theory. The gel point was found to be related to the monomer functionality. 1988) utilized this relation to express the gel point for the chainwise co159 . Landin and Macosko (1983. propagation. by deriving the probability of a finite chain in a polymer network (from which the gel point was estimated). The cross-linking occurs by reacting side groups on long. According to Flory’s theory (1953). Miller et al. Macosko and Miller (1976) also derived an expression to estimate the gel point for networks formed by chain polymerization. Miller and Macosko (1976) obtained results that agree with Flory’s theory. The chainwise reaction involves initiation. probability that a chain terminates by combination. Miller and Macosko (1976) and Macosko and Miller (1976) applied the same equations they developed for stepwise reactions to crosslinking reactions of polymer chains. linear polymer chains or through unsaturation in the chain backbone. (1979) and Macosko and Miller (1976) also extended the use of Flory’s model to include the case where more than one type of branching unit is present in the condensation system.Dc = 1 f −1 (73) Miller et al. and Valles and Macosko (1979) derived the weight average molecular weight for the nonlinear stepwise polymerization system. and termination. Macosko and Miller (1976). too. They assumed all the reactive groups are of the same type and derived the gel point as a function of the weight average degree of polymerization of the initial mixture of long chains (Macosko and Miller.

the inhibitor was assumed to be ideal. however. González-Romero and Macosko (1985) analyzed the kinetics of the free radical crosslinking polymerization and derived a relation between the gel point and gel time for a radical polymerization system that involves inhibition. They measured the viscosity of the system during the reaction and took the time at which the viscosity goes to infinity as gel time. They took this idea as a starting point and deduced an analytic expression for the gel point of a 160 . the initial mole fraction of the functional groups in the monomer mixture. i. Miller and Macosko (1987. This relation involving the rate constants for inhibition and propagation was verified experimentally. the monomer doesn’t react with the radicals until the inhibitor is consumed. were difficult to parameterize and not validated in practice.e. Dotson and co-workers (1988) derived the weightaverage molecular weight for the crosslinking free radical polymerization and obtained the gel point as the conversion where the molecular weight goes to infinity. The gel point thus can be calculated.e. Okay (1994) also derived an equation to approximate the gel point for the free radical chain copolymerization and cross-linking system.and di-functional monomers. Dc = D (inter) + D (ring). 1988) employed the chain length and site distribution to demonstrate the gel point for the network formed by chain crosslinking as a function of the expected number of sties on a chain randomly chosen by site. All these theoretical derivations for gel point of a chain polymerization system. Suematsu and Kohno (2000) separated the critical point of branched polymers into two terms: intermolecular reaction and cyclization.polymerization of mono. and the reactivity of the functional groups. The gel point was expressed as a function of the accumulated average chain lengths of the primary molecules. In this relation. i.

161 . then the branching probability α equals the extent of reaction D. and C is the initial monomer concentration of the system.polymerizing system consisting of same type functional units (Equation 74). Assuming that the gel lattice has high dimensions. f is the functionality. Incorporating this relation into the critical condition Equation (72). If the cyclization in the polymerizing system is negligible. Dc = 1 2[Γ] + f − 1 fC (74) where Dc is the conversion at which gelation occurs. In Flory’s theory. the gel point expression is the same as Equation (75). not D2. if only one type of functional group is present and these groups are capable of condensing with one another. Each functional unit is assumed to have an equal chance to undergo cyclization. [Γ ] is the molar concentration of rings formed by cyclization. Equation (74) can be reduced to the following: Dc = 1 f −1 (75) This is in agreement with the Flory’s theory if the branched polymerization here is considered as a condensation. Suematsu and Kohno (2000) used a percolation model and expressed the molar concentration of cyclics in a form of the solution of the ring distribution function for the site-bond percolation problem.

which indicates that the produced polymer doesn’t have to achieve an infinite molecular weight to form a solid part. and Macosko and Miller (1976) also obtained the same gel point expression for the stepwise homo-polymerization. (1979). the gel point of the system can be estimated to be ~14%. 162 . chainwise or stepwise. and does not need this assumption) and assuming the cyclization to be negligible.Miller and Macosko (1976). If considering the investigated tetraacrylate crosslinking system in our work as a condensation (model of Suematsu and Kohno (2000) is for general branched polymers. The critical DOC determined (9~10%) for the DOC threshold model (Chapter 6) is thus found to be lower than the gel point. Miller et al.

APPENDIX D MINITAB REGRESSION OUTPUT OF REDUCED MODEL FOR WIDTH RESOLUTION 163 .

APPENDIX E MINITAB STEPWISE REGRESSION OUTPUT FOR WIDTH RESOLUTION 164 .

APPENDIX F REGRESSION PREDICTION MODEL FOR DEPTH RESOLUTION 165 .

APPENDIX G REGRESSION PREDICTION MODEL FOR WIDTH RESOLUTION 166 .

APPENDIX H REGRESSION PREDICTION MODEL FOR CURING SPEED (WIDTH) 167 .

APPENDIX I REGRESSION PREDICTION MODEL FOR MAXIMUM DOC 168 .

169 .

APPENDIX J REGRESSION PREDICTION MODEL FOR MAXIMUM TEMPERATURE RISE 170 .

171 .

pp.222. Makromol. Youssef. and Peppas.. Bass. New York: Interscience. Bueche. A (1985) in Hellwege. Bowman. Journal of Polymer Science: Part A: Polymer Chemistry. pp. 3rd Edition.A. K. M.M. (Eds. Choi. Buback. and Bowman.19. C.. Thermochimica Acta. M. Chem. Dugdale. F. pp. Macromolecules. and Bowman. and Myers.. Bowman. O.REFERENCES Anseth.H.139-147. Vol. L. pp. pp.32. (1994a) ‘Polymerizatioin Kinetics and Volume Relaxation Behavior of Photopolymerized Multifunctional Monomers Producing Highly Crosslinked Networks’.35. C.. Buback.S. M. Burel. Brandrup.) Numerical Data and Functional Relationships in Science and Technology. N. F.326.H. Foiles.L. New York: Springer-Verlag. and Saiter.) (1989) Polymer Handbook.. Bunel. A General Kinetic Treatment’. New York: John Wiley & Sons..N. Vol. pp. H.N.133-141. Degener. Polymer. B. W. Makromol. J. J. C..S.W. Vol. K. Wang. Vol. C.. pp.3243-3250.1914-1920. Vol. 172 .S.N. (1999) ‘Synthesis and Photoinitiated Polymerization of a New Urethane Acrylate Monomer: Influence of Polymerization Temperature’. (Ed.S.. C. Vol.191. (1991) ‘Coupling of Kinetics and Volume Relaxation during Polymerizations of Multiacrylates and Multimethacrylates’. (1962) Physical Properties of Polymers. and Leu. B. Polymer Reaction Engineering. (1990) ‘Free-Radical Polymerization up to High Conversion. Cho.18-27..S. K. and Madelung. C. and Huckestein.N.311-316.. V/77.. Vol. Anseth. B. N.A. J.1. and Peppas. (1994b) ‘Reaction Behaviour and Kinetic Constants for Photopolymerizations of Multi(meth)acrylate Monomers’. Chem.. Journal of Manufacturing Systems. Park.10.24.499-520. B. pp. pp. Anseth. Lecamp.C. and Immergut E.1575-1587. (2000) ‘Determining Optimal Parameters for Stereolithography Processes via Genetic Algorithm’.. C... (1989) ‘Conversion Dependence of FreeRadical Polymerization Rate Coefficients from Laser-Induced Experiments.. J-M. Rapid Commun. 1’..S.. Vol. (1993) ‘Reaction Diffusion Enhanced Termination in Polymerizations of Multifunctional Monomers’. K.

P.31. R. and Chartoff. K. Vol. G. Vol.453-460. and Vijaybabu. Structures. pp.P.117-121.2560-8. and Eyerer. (1998) Photoinitiators for Free Radical. C.35.687-693.V. 5062 (Pt.A. W. A. Vol. (1994) ‘A Simple Polymer Shrinkage Model Applied to Stereolithography’.W. pp. J. A. N. Cationic & Anionic Photopolymerisation. T.225233. pp.72 (2). (2003) ‘Optimization of Process Parameters in Stereolithography using Genetic Algorithm’. N.) Chemistry & Technology of UV & EB Formulation for Coatings. Solid Freeform Fabrication Symposium Proceedings. pp.. 3484-3491. (Ed. I: Basic Model Development’. Vol. Journal of Thermal Analysis and Calorimetry. (1990) ‘Thermal Conductivity of Poly(ethylene glycols) and Their Binary Mixtures’.483-492. pp. Part A: Polym. Flach.417-424. Inks & Paints. pp. and Macosko. DiGuilio. Dotson.. (1988) ‘Structural Development during Nonlinear Free-Radical Polymerizations’. W. (1992) ‘Thermal Aspects of the Kinetics of Dimethacrylate Photopolymerization’. Blumenstock.. E. R. (2003) ‘Photopolymerization Kinetics of an Epoxy Based Resin for Stereolithography’. R.. Greco. Journal of Chemical and Engineering Data.. P.S. Galván. A.. Sci. Eschl. L.D. Journal of Applied Polymer Science. Vol. Solid Freeform Fabrication Symposium Proceedings. Crivello. Flach. Cook. and Chartoff. 1. R. Esposito Corcione. 35. Vol. Chem.M. J. pp. A. (1995b) ‘A Process Model for Nonisothermal 173 .33.2152-2161. C. (1993) ‘Photopolymerization Kinetics of Oligo(ethylene oxide) and Oligo(methylene) Oxide Dimethacrylates’. Macromolecules.Chockalingam. Polymer Engineering and Science. A.P. pp. (1999) ‘Comparison of the Curing Process of Epoxy and Acrylate Resins for Stereolithography by Means of Experimental Investigations and FEM – Simulation’. R. K. (2004) ‘Photopolymerization Kinetics of an Epoxy-Based Resin for Stereolithography’. Greco. pp. J.21. and Teja. and Maffezzoli. L. Esposito Corcione. Smart Materials. 2nd Edition. and Dietliker. Jawahar. R. Flach.D. Vol. and Maffezzoli..92 (6). Cook..1053-1067.. and Systems). Proceedings of SPIE-The International Society for Optical Engineering.. Vol. C. III in Bradley. New York: John Wiley & Sons. Polymer. pp. (1995a) ‘A Process Model for Nonisothermal Photopolymerization with a Laser Light Source. Polym. L. and Chartoff.

(1985) ‘Viscosity Rise during Free Radical Crosslinking Polymerization with Inhibition’... González-Romero.D.34(6).259272.220-231. Vol.Photopolymerization with a Laser Light Source. (1997) ‘A Study on Simulation of the Deformation of 3-D Stereolithography Products’.D. pp. M. Chemical Engineering Science. S. Solid Freeform Fabrication Symposium Proceedings. OH: Hanser/Gardner Publications. R. (1992) Rapid Prototyping & Manufacturing: Fundamentals of Stereolithography. C. MI: Society of Manufacturing Engineers.J. 35. C. pp.57. Res. and Bowman. J. C. Photopolymerization. Lee. J. (1998) ‘Modeling and Experimental Investigation of Light Intensity and Initiator Effects on Solvent-Free Photopolymerization’.1247-1252. Vol.32. 174 .). and Youn. Ind. Han'guk Somyu Konghakhoechi. Eng.493-498. pp. (1997) ‘Method for Determining the Kinetic Parameters in Diffusion-controlled Free-Radical Homopolymerizations’... Lee. and Macosko.F. Goodner. Flory. and Bowman. S. Cincinnati. R. and Photocuring – Fundamentals and Applications.6552-6559.887-900. (Eds. M.M.R.N. Vol. Fouassier. Keefe.N. M. and Gargiulo. and Hunt M.N. C. DC: American Chemical Society. Vol. Chem. M. pp. V. and Youn. Jayanthi. Dearborn. pp.39 (4). Goodner. Hur. H. C. T. J. (2002) ‘Development of Comprehensive Free Radical Photopolymerization Model Incorporating Heat and Mass Transfer Effects in Thick Films’. P. Polymer-Plastics Technology and Engineering. Vol. (1999) ‘Modeling Primary Radical Termination and Its Effects on Autoacceleration in Photopolymerization Kinetics’. M. (1994) ‘Studies in Stereolithography: Influence of Process Parameters on Curl Distortion in Photopolymer Models’. pp. Goodner. Jacobs.. NY: Cornell University Press. E.-P.36.713.N. pp. (1953) Principles of Polymer Chemistry. S. pp. Hur. S. H.29. and Bowman.250-258. P. Polymer Engineering and Science.D.W.E.. in Long. P. and Bowman. (2000) ‘Thermal Deformation of a Photo-cured Polymer for the Analysis of Stereolithography’.. Ithaca. Macromolecules. (1995) Photoinitiation. Solvent-Free Polymerizations and Processes: Minimization of Conventional Organic Solvents. Journal of Rheology. pp. II: Behavior in the Vicinity of a Moving Exposed Region”.374-385.O. J. Washington. Vol.651-666. Vol.D. Goodner. Vol. S.

pp. C. and Macosko..M. F. (1986) ‘Analytical Expression fro the Standing Wave Intensity in Photoresist’. Vol.T. H. F. J. Postpolymerization Study’.A. pp. pp.W.. pp. A..L.6313-6320. and Bunel. C. Terzi R. C. pp. and Zawatzki.104. Mack. and Miller. Vol. (1979). ’Characterization of the Kinetic Behavior of Resin Modified Glass-Ionomer Cements by DSC. Vol. Journal of Materials Science: Materials in Medicine.R. Vol. Polymer. D. Marten. Lecamp. Macromolecules.151-156. Uhlig. Lecamp. C. Vol.W. pp. SPIE Optical Microlithography IV. pp. (1998) ‘Effect of Irradiation Intensity on the Isothermal Photopolymerization Kinetics of Acrylic Resins for Stereolithography’. B.313-314. TMA and Ultrasonic Wave Propagation’.269/270.538. C.38. Bunel. O. (1998) ‘Glass Transition Temperature and Thermal Expansion Behaviour of Polymer Films Investigated by Variable Temperature Spectroscopic Ellipsometry’..T. A. (1999) ‘Photoinitiated Polymerization of a Dimethacrylate Oligomer: Part 3. Metzner. pp. Macosko. Vol.. pp.A.43-70..319-335.. Organic Coatings and Applied Polymer Science Proceedings: Preprints of Papers. ‘High Conversion Diffusion-Controlled Polymerization’. Landin. (1997) ‘Photoinitiated Polymerization of a Dimethacrylate Oligomer: 1. D. Bauer.9..12.Kahle. L. C. (1995). Vol.207-220. Polymer. Vol...433-439. (1983) ‘Rheological Changes during the Copolymerization of Vinyl and Divinyl Monomers’. C. Micelli. Thin Solid Films. (1976) ‘A New Derivation of Average Molecular Weight of Nonlinear Polymers’. B. pp. Maffezzoli A. and Hamielec. (1994) ‘Standing Waves in Photoresist’. Maffezzoli A. and Macosko. (1985) ‘Prolith: a Comprehensive Optical Lithography Model’.. D. C. Landin. Applied Optics.22-24. and Terzi R. C. 175 . Mack. Vol. Macromolecules.E..803-807. (2001). L. Thermochimica Acta. V. Vol. ‘Thermal Analysis of Visible-Light-Activated Dental Composites’.48. and Lebaudy. and Terzi. pp. American Chemical Society Symposium. Vol.A. Temperature and Light Intensity’.25. C. Thermochimica Acta. Ser. Youssef. pp. U. Maffezzoli. Wielsch. Youssef. and Luprano.321. Mack. R.6089-6096.199-206. pp..21. (1988) ‘Cyclization and Reduced Reactivity of Pendant Vinyls during the Copolymerization of Methyl Methacrylate and Ethylene Glycol Dimethacryalte’. Microlithography World.Influence of Photoinitiator Concentration.846-851.111-121.40. P.A.1958-1961.W.

489505. Miller. Serrano. E. pp. J. D. (1976) ‘A New Derivation of Post Gel Properties of Network Polymers’. (1979) ‘Calculation of Molecular Parameters for Stepwise Polyfunctional Polymerization’. D. Vol.206-211. (1996) Applied Linear Statistical Models.26. M. and Macosko. C. Polymer. New York: McGraw-Hill.W.H. Journal of Polymer Science: Part B: Polymer Physics. United States Patent. C. Polymer Engineering and Science. D. (1997).L.. F. Mateo.2613-18. and Hamielec.M.W. Nagamori. Miller. Kutner.. and Wasserman. and Macosko. Vol.9. Radiation-curable Composition. No. D.and Tetrafunctional Methacrylic Monomers in a Polymeric Medium: Kinetics and Evidence of Reaction Diffusion throughout the Photopolymerization Reaction’.Marten. (2001). Proceedings of SPIE-The International Society for Optical Engineering. pp.19.W. and Macosko. Valles. Miller.. Macromolecules. and Bosch. (2003). W.R.2441-69. pp. ‘Research on Solidification of Resin in StereoLithography (Comparison between Measured and Estimated Shapes of Solidified Resin. O. (1982). S. Especially for Producing Flexible Cured Articles by Stereolithography’. Nachtsheim.30.E. Vol. T. 6.25. T. Neter..35. J.L. ‘Photopolymerization of Di. Pat. and Yoshizawa. J. ‘High Conversion Diffusion-Controlled Polymerization of Styrene I’. (2000) ‘Liquid. S. P.27.. ‘Research on Shape of Solidified Resin in Stereolithography’. C.R. and Manufacturing Accuracy of Photo-cured Model)’. Vol. Okay. Journal of Polymer Science: Part B: Polymer Physics. 176 .P. Macromolecules. Vol. SPIE’s International Symposium on Intelligent Systems and Advanced Manufacturing (Opto-Mechatronic Systems).136. C.447-456. Miller. Nagamori. and Yoshizawa. (1988) ‘Network Parameters for Crosslinking of Chains with Length and Site Distribution’.1-54. A.. Melisaris.R. Vol.272-283. and Macosko. pp.1285-1288.. Journal of Applied Polymer Science.R. (1987) ‘Molecular Weight Relations for Crosslinking of Chains with Length and Site Distribution’.497.J. pp. 5058 (Optical Technology and Image Processing for Fluids and Solids Diagnostics). C. Vol. pp.. pp. pp.W. et al. (1994) ‘Kinetics of Gelation in Free Radical Crosslinking Copolymerization’. A.

190-317.461-472. (1996) Fundamentals of Fourier Transform Infrared Spectroscopy. Journal of Manufacturing Systems. K. 8.476.290-303. Japan. D.Onuh. G. No. 5.4589-4600. A. (1995) ‘Photosensitive Compositions’. Chu. Mat.007. Sun. K. T. Satio. United States Patent. B. 6. Diethylene. J. et al. United States Patent. and Montgomery.212. Pat. Schaub. Vol. B. Vol. Vol.O. and Hon. S. No. Part B. and Watts. D.38. pp. pp. Pananakis.K. K.198-202. Viscosity. Especially for Stereolithography’.C. (2000) ‘Liquid Radiation-curable Composition Especially for Producing Cured Articles by Stereolithography having High Heat Deflection Temperatures’. Zeitschrift für Physikalische Chemie.. Pat. Onuh. and Triethylene Glycol Mixture between 290K and 450K’. Suematsu.A.C. Steinmann.C. Physical Review E. and Kohno. and Hon. Vol.. Smith. Sci. B.V. S No.H. and Teja. (2000) ‘Estimation of Critical Points of Branched Polymers’.48. (1956) ‘Über die Polymerisationskinetik in hochkonzentrierten Systemen – Zur Kinetik des Trommsdorffeffektes an Methylmethacrylat’.W.S.748. Vol.563. United States Patent.K. Ministry of Education. et al. (2003) ‘Density. (2000) ‘Incorporation of the Heating Effect of the Light Source in a Non-isothermal Model of a Visible-light-cured Resin Composites’. A. (1993) ‘A Study on Development of 3D Model Stereo-Lithography’.3944-53.03555022. Steinmann. Schulz. (1997) ‘Optimising Stereolithography Throughput’. K-R. et al.329-342. Rosen.972. S. pp.35. Vol.O. 5. (1999) ‘Liquid. pp. D. (2002) ‘Stereolithography Technology  Course Material for ME7227: Rapid Prototyping in Engineering’. Journal of Chemical and Engineering Data. Fiscal 1992.100. Pat. and Thermal Conductivity of Aqueous Ethylene. New York: CRC Press.(1998a) ‘Optimising Build Parameters for Improved Surface Finish in Stereolithography’. pp. No. pp. D. American Physical Society. Proceedings of the Institution of Mechanical Engineers. M. International journal of Machine Tools Manufacture. pp. Radiation-curable Composition. Vol. 177 . Pang. T. D.16.(1998b) ‘Application of the Taguchi Method and New Hatch Styles for Quality Improvement in Stereolithography’.62.

G. MS Thesis.2059-2075. Sci.F. (1990) Properties of Polymers – their correlation with chemical structure. Phys. (2003) Yaws' Handbook of Thermodynamic and Physical Properties of Chemical Compounds. Norwich. Wu.: Poly.Y. Y. C.Tang. and Hamada. Analysis. M. (1979) ‘A Calorimetric Study of Acrylate Photopolymerization’.W. D. (2002)..: Knovel.W. Experiments Planning. Ed. pp. pp.M. their numerical estimation and prediction from additive group contributions. (1979) ‘Structure and Viscosity of Poly(dimethylsiloxanes) with Random Branches’. Van Krevelen. and Parameter Design Optimization. C.R. New York: Elsevier. L.J. 178 . Tryson. and Shultz A. J. N.521-526. E. (2002) ‘Stereolithography Cure Modeling’. New York: John Wiley & Sons. Yaws.R. Georgia Institute of Technology. Valles.12. C. American Chemical Society. Vol.17. and Macosko. Vol.. of Poly.

Sign up to vote on this title
UsefulNot useful