Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
In situ process analysis for defect control during composites manufacturing
(USC Thesis Other)
In situ process analysis for defect control during composites manufacturing
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
In Situ Process Analysis for Defect Control
During Composites Manufacturing
Mark Anders
A dissertation presented to the faculty of the
USC Graduate School
in partial fulfillment of the requirements for the degree
Doctor of Philosophy
(Mechanical Engineering)
May 2019
© Mark Anders, 2019
ii
Acknowledgements
First and foremost, thank you to my advisor, Prof. Steve Nutt. You have given me guidance
and opportunities that have launched me onto an incredibly exciting career trajectory, and you’ve
helped me acquire the knowledge, skills, and self-confidence that I’ll need to be successful. I have
you to thank for my entire professional future.
Tim Centea: I’m so fortunate to have had the privilege of working side by side with you
over these years. Your mentorship tremendously accelerated my growth as a researcher. Thank
you for your endless patience, your unfaltering positive attitude, and for being the best role model
anyone could hope for.
Thank you to my fellow PhD students at the M. C. Gill Composites Center, for always
making me feel welcome and for creating such a positive work environment. Special thanks to my
project teammates Jonathan Lo and Daniel Zebrine, it was a pleasure working with you. I also
want to thank the brilliant and talented undergraduate researchers that I worked with: Matthew
Thomas, Trisha Palit, and Emma Morrissey. Finally, I want to acknowledge our lab managers
Rohan Panikar and Yunpeng Zhang, for always keeping the lab running smoothly and relieving us
students from the burdens of equipment maintenance.
Funding for this work was provided by Henkel Corp., NASA Langley Research Center,
and the M. C. Gill Composites Center. My thanks in particular to our project supervisors, Ehsan
Barjasteh (Henkel) and Roberto Cano (NASA Langley). Material donations were provided by
Henkel, Airtech International, Hexcel, Solvay, and The Gill Corporation. I also want to thank Don
Wiggins and his team at the USC machine shop for turning my design ideas into reality.
iii
I’m grateful for the wonderful friends that I’ve lived with during this time: Joana Que,
Victoria Wolseley, Bill Edwards, and Ryan Jones. I cherish the memories of all the great times
we’ve had. I’m also grateful to my fellow sailors Duncan Cameron, Roland Vollman, and Scott
“Floyd” Barber. You kept me sane by getting me out of the lab and onto the water for thrilling
races and unforgettable adventures.
And of course, a heartfelt thank you to my parents, my siblings, and the rest of my family,
for always believing in me, supporting me, and pushing me to achieve my dreams.
Allison Smith: You pulled me up, helped me see the light at the end of the tunnel, and have
supported me through to the end. I can’t thank you enough. ♥
iv
Abstract
Polymer matrix composites are among the highest-performance structural materials
available today, but manufacturing processes for these materials have not yet been perfected. This
dissertation presents two investigations into manufacturing processes that are prone to particular
types of process-induced defects. In situ visualization was used as a diagnostic method in both
cases, first to analyze the evolution of process-induced defects in real time, and then to demonstrate
the effectiveness of proposed process modifications for avoiding these defects.
The first investigation considered resin transfer molding (RTM) with a benzoxazine resin,
which tended to result in composite parts with significant surface porosity. A lab-scale RTM tool
with in situ observation capabilities was used to identify the defect formation mechanism (Chapter
2). A modified cure temperature cycle was shown to effectively prevent surface porosity, and
guidelines for design of such cure cycles were developed (Chapter 3).
The second investigation considered the autoclave co-cure of honeycomb core sandwich
structures. This particularly challenging manufacturing process can result in multiple types of
process-induced defects, thus the project goal was to develop a physics-based process model to
enable prediction of defect formation. First, a lab-scale autoclave with in situ observation
capabilities was used to identify the possible problems that could occur during co-cure (Chapter
4). Process modifications were proposed to avoid the observed defect formation mechanisms, and
demonstrated to be effective. Finally, a study of prepreg gas permeability was conducted (Chapter
5). Prepreg permeability is a poorly-understood yet important topic for co-cure processing, because
it affects the gas pressure in the honeycomb core cells, which, in turn, affects the quality of the
v
skin/core bond-line. Two types of models for prepreg permeability were proposed, which can be
used as sub-models within the integrated co-cure process model.
Overall, this dissertation contributed to the knowledge on composite processing in several
ways. First, a previously unexplained type of defect in RTM composites was characterized, and a
practical mitigation strategy was developed. Second, a new visualization technique for observing
the skin/core bond-line during co-cure was developed. Third, an unconventional processing
strategy (in-bag pressurization) was applied to co-cure and shown to solve many of the problems
that can occur. Finally, a new approach to modeling the gas permeability of prepregs was
developed, which, unlike previous empirical models, accounts for the relevant process physics.
vi
Table of Contents
Acknowledgements ................................................................................................................... ii
Abstract ................................................................................................................................... iv
Table of Contents .................................................................................................................... vi
List of Tables ........................................................................................................................... ix
List of Figures............................................................................................................................x
CHAPTER 1. Introduction .......................................................................................................1
1.1 Motivation ....................................................................................................................1
1.2 Composite manufacturing methods ...............................................................................3
1.2.1 Liquid composite molding......................................................................................3
1.2.2 Prepreg processing .................................................................................................7
1.2.3 Co-cure of sandwich structures ............................................................................ 11
1.3 Process-induced defects in composite materials........................................................... 15
1.4 In situ process diagnostic methods .............................................................................. 17
1.5 Scope of dissertation ................................................................................................... 18
CHAPTER 2. Eliminating surface porosity during resin transfer molding (RTM) ............. 22
2.1. Introduction .................................................................................................................... 22
2.1.1. Literature review...................................................................................................... 23
2.1.2. Objectives and approach .......................................................................................... 29
2.2. Materials ........................................................................................................................ 30
2.3. Experimental methods .................................................................................................... 31
2.3.1. Thermal characterization .......................................................................................... 31
2.3.2. Molding ................................................................................................................... 32
2.4. Results and discussion .................................................................................................... 37
2.4.1. TGA and RDA data ................................................................................................. 37
2.4.2. Neat resin molded samples ....................................................................................... 39
2.4.3. Surface porosity ....................................................................................................... 42
2.5. Conclusions .................................................................................................................... 54
CHAPTER 3. Cure cycle design for eliminating surface porosity during RTM .................. 58
3.1. Introduction .................................................................................................................... 58
3.2. Experimental methods .................................................................................................... 61
3.2.1. Thermal characterization .......................................................................................... 61
3.2.2. RTM manufacturing trials ........................................................................................ 61
3.3. Cure kinetics modeling ................................................................................................... 65
3.4. Resin volatile release behavior ........................................................................................ 69
vii
3.5. Volatile-induced surface porosity ................................................................................... 71
3.6. Process map.................................................................................................................... 73
3.7. Cure cycle design ........................................................................................................... 76
3.8. Conclusions .................................................................................................................... 79
Interlude .................................................................................................................................. 81
CHAPTER 4. In situ process diagnostics for co-cure of sandwich structures ...................... 84
4.1. Introduction .................................................................................................................... 84
4.1.1. Literature review...................................................................................................... 87
4.1.2. Objectives and approach .......................................................................................... 91
4.2. Experimental facilities .................................................................................................... 92
4.3. Experimental methods .................................................................................................... 96
4.3.1. Materials and layup .................................................................................................. 96
4.3.2. Process parameters for case study ............................................................................ 98
4.4. Results.......................................................................................................................... 103
4.4.1. Case A ................................................................................................................... 103
4.4.2. Case B ................................................................................................................... 109
4.4.3. Case C ................................................................................................................... 110
4.4.4. Case D ................................................................................................................... 111
4.4.5. Case E ................................................................................................................... 112
4.5. Conclusions .................................................................................................................. 115
CHAPTER 5. Through-thickness gas permeability of prepreg facesheets during co-cure 117
5.1. Introduction .................................................................................................................. 117
5.2. Experiments ................................................................................................................. 121
5.2.1. Methods ................................................................................................................. 121
5.2.2. Results ................................................................................................................... 125
5.3. Modeling – constant-K approach .................................................................................. 136
5.4. Modeling – two-phase approach ................................................................................... 142
5.4.1. Model formulation ................................................................................................. 142
5.4.2. Coupled solution .................................................................................................... 150
5.5. Conclusions .................................................................................................................. 152
CHAPTER 6. Conclusions and future work ........................................................................ 155
6.1. RTM project ................................................................................................................. 155
6.1.1. Conclusions ........................................................................................................... 155
6.1.2. Recommendations for future work ......................................................................... 158
6.2. Co-cure project ............................................................................................................. 160
6.2.1. Conclusions ........................................................................................................... 160
6.2.2. Recommendations for future work ......................................................................... 163
6.3. Broader implications and final thoughts ........................................................................ 165
viii
References.............................................................................................................................. 167
Design details for lab-scale RTM system ..................................................... 175
A.1. Technical drawings (all dimension in inches) ............................................................... 175
A.2. Stress simulation .......................................................................................................... 179
A.3. Plumbing ..................................................................................................................... 180
A.4. Heating simulation ....................................................................................................... 182
Design details for the HD/RTM ................................................................... 184
B.1. Technical drawings ...................................................................................................... 184
B.2. System overview .......................................................................................................... 199
Preliminary results from the HD/RTM ....................................................... 204
C.1. Composite sample ........................................................................................................ 204
C.2. System calibration for solids ........................................................................................ 208
C.3. System calibration for liquids ....................................................................................... 212
Design details for the “mini autoclave” co-cure fixture .............................. 216
D.1. Technical drawings ...................................................................................................... 216
D.2. Stress simulation .......................................................................................................... 223
ix
List of Tables
Table 2.1: Thermal characterization tests ............................................................................................... 32
Table 2.2: Processing parameters for molded samples fabricated in the lab-scale RTM. ......................... 35
Table 2.3: Percent defective area for the cold sides of molded laminates. ............................................... 43
Table 3.1: List of tested cure cycles. ...................................................................................................... 62
Table 5.1: Parameters used in Eq. (5-7)................................................................................................ 137
Table 5.2: Parameters used in Eq. (5-10). ............................................................................................. 139
x
List of Figures
Figure 1.1: Composite materials usage in the 787 Dreamliner. Image source: The Boeing Company
(www.boeing.com) .................................................................................................................................. 2
Figure 1.2: Schematic of the resin infusion process. Image source: [2]. .................................................... 4
Figure 1.3: A boat hull being produced by resin infusion. Image source: Airtech International
(www.airtechintl.com). ............................................................................................................................ 5
Figure 1.4: Steps of the RTM process. ..................................................................................................... 6
Figure 1.5: Schematic of a prepreg laminate within an autoclave. Image source: [4]. .............................. 10
Figure 1.6: A very large (and expensive) autoclave. Image source: ASC Process Systems
(www.aschome.com). ............................................................................................................................ 10
Figure 1.7: Relative changes in stiffness, strength, and weight of a sandwich structure with increasing
core thickness. Image source: [8]. .......................................................................................................... 12
Figure 1.8: Elements of a honeycomb core sandwich structure. .............................................................. 12
Figure 1.9: Sandwich components on the Airbus A380. Image source: [9]. ............................................ 13
Figure 1.10: Sandwich components in a commercial aircraft engine nacelle. Image source: Hexcel
Corporation (www.hexcel.com). ............................................................................................................ 14
Figure 1.11: Mechanism of void formation inside and between fiber tows: (a) capillary flow lagging
and (b) capillary flow leading. Image source: [11]. ................................................................................ 16
Figure 1.12: Images of uncured OoA prepreg: (a,b) surface and (c) section view of two plies. Image
source: [12]. .......................................................................................................................................... 17
Figure 2.1: A laminate surface with extensive volatile-induced porosity. ................................................ 27
Figure 2.2: Exploded view of the lab-scale RTM tool. ........................................................................... 33
Figure 2.3: Mass ratio and viscosity for linear temperature ramps. ......................................................... 37
Figure 2.4: Temperature/pressure map of neat-resin molded samples. Red indicates conditions that lead
to void growth, while blue corresponds to conditions that suppress void growth. Points correspond to
individual tests and background shading is used to highlight the general trend. ...................................... 40
Figure 2.5: In situ observations of bubble nucleation and growth in resin cured at ambient pressure. ...... 41
xi
Figure 2.6: Some examples of surface porosity quantification by image binarization and measurement
of the percent defective area. The top row is formulation F1, Cycle A, middle row is F2, Cycle A, and
the bottom row is F2, Cycle B. The thermocouple used to measure the window-side temperature
history is visible in the first image. ........................................................................................................ 43
Figure 2.7: Temperature, viscosity, and pressure for formulation F1, Cycle A. ....................................... 45
Figure 2.8: Surface porosity formation recorded in situ for the composite sample with formulation F1,
Cycle A. Time labels correspond to the data in Figure 2.7. The thermocouple used to measure the
window-side temperature history is visible in the center of the frame. .................................................... 47
Figure 2.9: Schematic of the surface porosity formation mechanism. Liquid resin is initially under
hydrostatic pressure (top image). Then, a temperature gradient causes a “gelation boundary” to form
and move from the hotter side of the part to the colder side (middle image). Volatile release occurs at
the not-yet-vitrified colder side, as the cure shrinkage of the hotter side reverses the stress state from
compressive to tensile (bottom image). .................................................................................................. 48
Figure 2.10: Viscosity envelopes for laminates cured with Cycle A........................................................ 49
Figure 2.11: Viscosity envelopes for laminates cured with Cycle B. ....................................................... 50
Figure 2.12: Comparison of the cold-side surface porosity of molded samples with the through-
thickness viscosity mismatch at the time of the pressure drop................................................................. 52
Figure 2.13: Temperature, viscosity, and pressure for formulation F2, Cycle B. ..................................... 54
Figure 2.14: In-plane temperatures of an RTM with a 305 × 457 mm (12” × 18") cavity (left), and
surface porosity concentrated in the cooler edge regions of a molded sample (right). .............................. 56
Figure 3.1: Front view and exploded CAD render of the lab-scale RTM tool. ......................................... 62
Figure 3.2: Section view of lab-scale RTM tool (left) and a thermal simulation showing the through-
thickness temperature gradient (right). ................................................................................................... 64
Figure 3.3: A molded sample with porosity exclusively on the colder side. ............................................ 64
Figure 3.4: Isoconversional chart from dynamic DSC scans. Linear fits at some example α values are
shown in black. ..................................................................................................................................... 68
Figure 3.5: Experimental model parameters E a and A' over the full range of α. Inset shows the R
2
values of the linear fits used to obtain the model parameters at each iso-α value. .................................... 68
Figure 3.6: Comparison of measured and predicted rates of cure for the DSC scans from Figure 3.4 ...... 69
xii
Figure 3.7: Mass loss rate, complex viscosity, and degree of cure, as obtained by TGA, RDA, and the
cure kinetics model (respectively). Black arrows show the times when mass loss ceased, and red
horizontal bands show the corresponding viscosity and degree of cure values. The onset of gelation
and corresponding degree of cure are shown in green............................................................................. 70
Figure 3.8: Temperature, pressure, and model degree of cure data for the baseline cure Cycle A
(ramping directly to the high-temperature dwell). A binary map of the surface defects is shown, along
with a micrograph of a surface void. ...................................................................................................... 72
Figure 3.9: Process map showing times of critical degree of cure, pressure drop, and gelation, for a
range of intermediate dwell temperatures. The shaded green area is the process window for
intermediate dwells resulting in minimal surface porosity, and the shaded red area shows an
approximate temperature envelope for Cycle H. ..................................................................................... 74
Figure 3.10: Temperature, pressure, and model α for Cycle H (45 minute intermediate dwell at 150°C). 77
Figure 3.11: Degree of cure vs. temperature envelopes for cure cycles with 0, 45, and 120 minute
intermediate dwells at 150°C before the high-temperature dwell. Diagonal contour lines indicate
relative density, which varies with both temperature and degree of cure as described by Eq. (3-10). ....... 78
Figure Int-1: CAD model of the hybrid dilatometer/RTM. ..................................................................... 82
Figure 4.1: Polished-section micrographs of co-cured samples exhibiting a wide range of
morphologies. ........................................................................................................................................ 86
Figure 4.2: Section view of the co-cure fixture with sample and consumables. ....................................... 92
Figure 4.3: Photographs of the co-cure fixture. Top: frame with lower half of tool (left) and power
distribution enclosure (right). Bottom left: close-up of tool plate. Bottom right: tool plate with
laminate. ............................................................................................................................................... 95
Figure 4.4: Stages of the adhesive reticulation process. Left: film applied to honeycomb core and
perforated. Middle: film softening under applied heat and perforations opening due to surface tension.
Right: process completed when adhesive forms a bead along the edges of the cell walls. ........................ 97
Figure 4.5: The temperature cycle used for all tests (top), and the two cycles used for autoclave and
vacuum bag pressures (middle and bottom). Stages I, II, and III denote room-temperature,
intermediate-temperature, and high-temperature stages (respectively) of the process. ............................. 99
Figure 4.6: Diagram of Cases A through E. The three axes denote the test variables: (1) the adhesive
was either applied as a continuous film, or reticulated onto the honeycomb core; (2) the core cavity
was either sealed, or connected via the “core vent” to equilibrate the gas pressure in the core cavity
with that of the vacuum bag; and (3) autoclave and vacuum bag pressures were applied according to
either of two cycles – “simple” or “staged” – as shown in Figure 4.5. .................................................. 100
xiii
Figure 4.7: Temperature and pressure data for the three cases using the “simple” pressure cycle (A, B,
and C). Models for the glass transition temperature and viscosity (of both the prepreg resin and
adhesive) are shown, computed from the temperature history recorded at the outside of the vacuum
bag near the center of the laminate. ...................................................................................................... 104
Figure 4.8: In situ images of the bond-line during processing. Columns correspond to Cases A, B,
and C (left to right), and rows (top to bottom) correspond to the times of interest t 1 through t 4
indicated on Figure 4.7 ........................................................................................................................ 105
Figure 4.9: Temperature and pressure data for the two cases using the "staged” pressure cycle (D and
E). Models for the glass transition temperature and viscosity (of both the prepreg resin and adhesive)
are shown, computed from the temperature history recorded at the outside of the vacuum bag near the
center of the laminate. ......................................................................................................................... 106
Figure 4.10: In situ images of the bond-line during processing. Columns correspond to Cases D and E
(left to right), and rows (top to bottom) correspond to the times of interest t 1 through t 4 indicated on
Figure 4.9. ........................................................................................................................................... 107
Figure 5.1: Gas pressure in honeycomb core during pre-processing room-temperature vacuum holds.
Each trace represents one of eleven samples with identical configurations. Observed behavior consists
of (1) rapid initial evacuation prior to laminate compaction, (2) a delay of one to seven hours as flow
channels form, and (3) core evacuation once one or more flow channels have developed. Image
source: [85]. ........................................................................................................................................ 119
Figure 5.2: Evolution of facesheet air permeability during elevated temperature processing, showing
high variability and a non-monotonic trend. Image source: [85]. .......................................................... 119
Figure 5.3: Schematic of a sandwich structure (bottom image), with inset view of a single honeycomb
cell. ..................................................................................................................................................... 120
Figure 5.4: A schematic sectional view of the co-cure fixture (not to scale). ......................................... 121
Figure 5.5: Schematic representation of the two permeability test configurations. ................................ 124
Figure 5.6: Core pressure versus time for successive evacuation cycles, at room temperature and a
compaction pressure of 1 bar. .............................................................................................................. 126
Figure 5.7: Core pressure versus time for successive evacuation cycles, at room temperature and a
compaction pressure of 4 bar. .............................................................................................................. 126
Figure 5.8: Photographs of three prepregs with different fabric architectures. Images are back-lit to
highlight “pinhole” macro-pores in the woven fabrics (b) and (c). Image source: [84]. ......................... 127
Figure 5.9: Section view of 4 plies of prepreg prior to compaction and resin flow. ............................... 128
xiv
Figure 5.10: Schematic representation of a spherical meniscus with gas pressure P g on the left and
wetting fluid (liquid) pressure P w on the right. ..................................................................................... 128
Figure 5.11: Temperatures, pressures, and resin viscosity for a core evacuation test at 70 ºC and 300
kPa of compaction pressure. ................................................................................................................ 130
Figure 5.12: Core pressure versus time for core evacuation tests with 200 kPa of compaction pressure
at 50 ºC (blue), 70 ºC (orange), and 90 ºC (yellow). ............................................................................. 131
Figure 5.13: Core pressure versus time for evacuation tests at 50 ºC and 70 ºC ..................................... 132
Figure 5.14: Normalized bag/core pressure differences (top) and effective permeability values
(bottom) for steady-state flow tests at 50 °C (left) and 70 °C (right). .................................................... 133
Figure 5.15: Core pressure and effective permeability versus time for steady-state tests at three flow
rates. ................................................................................................................................................... 134
Figure 5.16: In situ images of bubbles moving through prepreg into the core cavity. Initial state (left),
during the temperature ramp (middle), and during the steady-state portion of the test (right). ............... 135
Figure 5.17: Simplified 1-D representation of a sandwich structure. ..................................................... 136
Figure 5.18: Measured pressure decay curve for a core evacuation test at 90 ºC and 300 kPa of
compaction pressure, and the corresponding best-fit constant-K model using Eq. (5-8)......................... 138
Figure 5.19: Best-fit K values for each core evacuation test (red markers), and the overall best-fit
polynomial surface of Eq. (5-10). ........................................................................................................ 139
Figure 5.20: Model/experiment comparisons for core evacuations with 200 kPa of compaction
pressure at 50 ºC (blue), 70 ºC (orange), and 90 ºC (yellow)................................................................. 140
Figure 5.21: Model/experiment comparisons for core evacuations at 70 ºC and 4 compaction
pressures. ............................................................................................................................................ 140
Figure 5.22: Capillary pressure head versus saturation for various porous materials, from [108]. .......... 145
Figure 5.23: Saturation versus capillary pressure head for various porous materials, from [108]. .......... 146
Figure 5.24: Graphical representation of Eq. (5-21) for P b = 10 kPa and various values of λ. ................ 147
Figure 5.25: An example of the model V1 output, over the spatial domain at t = 100 s (left), and over
time (right). ......................................................................................................................................... 151
Figure 6.1: Conceptual flowchart of the components of a process model for predicting volatile-
induced surface porosity. ..................................................................................................................... 159
Figure 6.2: Conceptual flowchart of the integrated co-cure process model. Image source: [112]. .......... 163
xv
Figure A-1: Dimensions for main tool body (sheet 1 of 2). ................................................................... 175
Figure A-2: Dimensions for main tool body (sheet 2 of 2). ................................................................... 176
Figure A-3: Dimensions for "picture frame" spacer plate. .................................................................... 177
Figure A-4: Dimensions for the window's retaining ring. ..................................................................... 178
Figure A-5: Stress simulation for the glass window with a resin pressure of 180 psi. ............................ 179
Figure A-6: Deflections in the window (color bar in units of µm) with a resin pressure of 180 psi. ....... 180
Figure A-7: Schematic of the fluid connections between components of the lab-scale RTM system...... 181
Figure A-8: Section view of the resin inlet/outlet design. ..................................................................... 181
Figure A-9: Predicted steady-state tool-face temperature distribution with heaters at 180 °C. ............... 182
Figure A-10: Predicted steady-state window-side temperature distribution with heaters at 180 °C. ....... 183
Figure B-1: Custom-built components of the HD/RTM. ....................................................................... 184
Figure B-2: Dimensions of the upper load frame connection. ............................................................... 185
Figure B-3: Dimensions of the upper ceramic insulation (sheet 1 of 3). ................................................ 186
Figure B-4: Dimensions of the upper ceramic insulation (sheet 2 of 3). ................................................ 187
Figure B-5: Dimensions of the upper ceramic insulation (sheet 3 of 3). ................................................ 188
Figure B-6: Dimensions of the upper plate (sheet 1 of 2). ..................................................................... 189
Figure B-7: Dimensions of the upper plate (sheet 2 of 2). ..................................................................... 190
Figure B-8: Dimensions of the piston. .................................................................................................. 191
Figure B-9: Dimensions of the ring (sheet 1 of 2). ................................................................................ 192
Figure B-10: Dimensions of the ring (sheet 2 of 2). .............................................................................. 193
Figure B-11: Dimensions of the base plate (sheet 1 of 2). ..................................................................... 194
Figure B-12: Dimensions of the base plate (sheet 2 of 2). ..................................................................... 195
Figure B-13: Dimensions of the lower plate. ........................................................................................ 196
Figure B-14: Dimensions of the bottom ceramic insulation. ................................................................. 197
xvi
Figure B-15: Dimensions of the stand. ................................................................................................. 198
Figure B-16: Picture of the HD/RTM fully assembled. ......................................................................... 199
Figure B-17: Annotated CAD model of the HD/RTM. ......................................................................... 200
Figure B-18: Annotated CAD model of the HD/RTM (section view). .................................................. 201
Figure B-19: Section view of the HD/RTM. The sample cavity is shown in yellow and arrows on the
right indicate the measured gap. ........................................................................................................... 202
Figure B-20: HD/RTM mounted in the load frame. Can you spot the elephant in the room? ................. 203
Figure C-1: Density changes in a thermoset resin. Image source: [115]. ............................................... 204
Figure C-2: Temperature and cavity thickness data from a composite sample, as well as models for
degree of cure and glass transition temperature. ................................................................................... 205
Figure C-3: Cavity thickness versus temperature, with cured and uncured coefficients of thermal
expansion (CTEs) in red. ..................................................................................................................... 206
Figure C-4: Comparison of modeled and measured sample thickness. .................................................. 207
Figure C-5: Sample thickness versus time for a cured sample in a second heating cycle. ...................... 208
Figure C-6: Comparison of thickness changes for an upward temperature ramp on a cured sample,
measured by TMA and in the HD/RTM (labelled “PVT” here, for “pressure-volume-temperature”
analyzer). ............................................................................................................................................ 209
Figure C-7: Comparison of thickness changes for a downward temperature ramp on a cured sample,
measured by TMA and in the HD/RTM. .............................................................................................. 209
Figure C-8: Measured thermal expansion versus temperature, determined from the slope of the TMA
data in Figure C-7. ............................................................................................................................... 210
Figure C-9: Measured and modeled advancement of the degree of cure for linear temperature ramps. .. 210
Figure C-10: Measured and modeled advancement of the degree of cure for isothermal dwells. ........... 211
Figure C-11: Measured DSC data (black lines), model predictions (mesh surface), and model/
experiment error (colored bands) for Eq. (C-4). ................................................................................... 212
Figure C-12: The density of glycerol as a function of temperature. ....................................................... 213
Figure C-13: An example of a test that leaked. ..................................................................................... 213
Figure C-14: Calibration test results using glycerol. ............................................................................. 214
xvii
Figure D-1: Dimensions for the main tool body (sheet 1 of 4). ............................................................. 216
Figure D-2: Dimensions for the main tool body (sheet 2 of 4). ............................................................. 217
Figure D-3: Dimensions for the main tool body (sheet 3 of 4). ............................................................. 218
Figure D-4: Dimensions for the main tool body (sheet 4 of 4). ............................................................. 219
Figure D-5: Dimensions for the window’s retaining ring. ..................................................................... 220
Figure D-6: Dimensions for the lid (sheet 1 of 2). ................................................................................ 221
Figure D-7: Dimensions for the lid (sheet 2 of 2). ................................................................................ 222
Figure D-8: Simulated stresses in the main tool body. .......................................................................... 223
Figure D-9: Simulated stresses (> 40 MPa) in the main tool body. ....................................................... 224
Figure D-10: Simulated stresses in the lid (deflections greatly exaggerated). ........................................ 224
Figure D-11: Simulated stresses (> 60 MPa) in the lid.......................................................................... 225
1
CHAPTER 1. Introduction
1.1 Motivation
Since the beginning of human history, ages of civilization have been defined by the
materials that we used. The Stone Age, the Bronze Age, the Iron Age; each advancement in
materials technology resulted in profound impacts to society. But what will historians of the future
label the present age? One could argue that today we live in “the Space Age” or “the Information
Age,” but in terms of the materials that we use for our tools, our vehicles, our weapons, and our
civil infrastructure, we have – until recently – still been living in an age of metals. Only in the last
few decades have we begun to fully embrace the transition to a new age, one of wholly synthetic
materials: polymers and their composites.
Although synthetic polymers were discovered more than a century ago, their utility as a
structural material – competitive with metals – was only fully realized upon the invention of
polymer-derived carbon fibers in the 1960s [1] and their incorporation into carbon fiber/polymer
matrix composites. Presently, carbon fiber reinforced polymers (CFRPs) are in the process of
replacing metals in almost all applications that demand maximum performance (i.e., specific
strength and stiffness), most notably in aerospace structures (see Figure 1.1). The ubiquitous
adoption of CFRPs, however, has been hindered, largely by difficulties and high costs associated
with composite manufacturing methods.
2
Figure 1.1: Composite materials usage in the 787 Dreamliner. Image source: The Boeing Company
(www.boeing.com)
Although many CFRP manufacturing methods exist, they all entail the arrangement of
carbon fibers into a prescribed shape and the embedding of those fibers within a polymer matrix.
Consequently, all CFRP manufacturing methods must meet similar demands: high fiber volume
fractions, correct fiber placement, and sufficiently low defect levels in the matrix. These demands
are often difficult to meet, however, due to the complexity of composite manufacturing processes
and incomplete understanding of the underlying physics. In industrial practice, process-induced
defects are a common occurrence, and their cause is not always known, often requiring costly trial-
and-error process troubleshooting. Furthermore, most composite manufacturing processes are
effectively opaque, i.e., the tooling and equipment used preclude the manufacturer from
identifying defects as they form during the process. Parts can be inspected after fabrication using
ex situ methods, but if defects exist, the manufacturer can only speculate as to how, when, and why
they formed.
3
Therefore, the underlying motivation for the work presented in this dissertation was to
address an ever-increasing industry need for more reliable, efficient, and robust composite
manufacturing processes, utilizing in situ diagnostics to identify and mitigate defect-formation
phenomena in realistic processing conditions.
1.2 Composite manufacturing methods
Composite manufacturing methods are numerous and diverse, and the preferred method
for producing a given composite part depends on many factors. The composite’s constituent
materials impose certain processing requirements (e.g., in terms of temperatures and pressures
required), as do the part complexity, the required tolerances, production speed/volume, and others.
This work focuses on manufacturing methods for continuous-fiber composites with a
thermoset resin matrix. The vast majority of such composites are produced by either a liquid
composite molding (LCM) process or by a prepreg process, although other specialized processes
exist for particular geometries, such as filament winding for cylindrical parts (e.g., composite
overwrapped pressure vessels) and pultrusion for beams and rods with constant cross-section. The
first half of this dissertation describes an investigation of an LCM process – resin transfer molding
(RTM) – and the latter half involves co-cure of sandwich structures, a particularly challenging type
of prepreg process. A brief overview of these processes is provided here.
1.2.1 Liquid composite molding
LCM processes generally include the following basic steps. First, layers of fibrous
reinforcement are placed onto a one-sided tool surface. Then the fibers are saturated with liquid
resin, and the resin is cured, creating a solid composite laminate. The simplest such method is wet
layup, in which each ply is laid onto the tool and then manually saturated with resin using a roller
4
or brush. This method, while inexpensive, is likely to result in parts with significant defect levels
unless performed by highly skilled technicians. Wet layup also makes it challenging to achieve
high fiber volume fractions, and risks exposing workers to harmful volatiles that can be released
by the resin.
The next step up in terms of process complexity is resin infusion, also known as vacuum
assisted resin transfer molding (VARTM
1
). In this process, dry fiber plies are laid onto a tool and
then covered by an impermeable polymer film called a vacuum bag. Vacuum is applied within the
bag, which compacts the fibers (due to the pressure difference between the evacuated bag and the
external environment) and draws resin into the fiber bed through an inlet port. The resin is then
allowed to cure, and the finished part can be de-molded.
Figure 1.2: Schematic of the resin infusion process. Image source: [2].
1
The term VARTM can be misleading, since RTM (resin transfer molding) typically involves vacuum, too. “Vacuum
only resin transfer molding” would be more appropriate.
5
Infusions can be performed rapidly in “one shot” by using multiple resin inlet gates and
vacuum ports, and by including a high-permeability distribution medium between the fibrous
reinforcement and the vacuum bag. Due to these process features, resin infusion is often the
preferred method for producing large structures such as wind turbine blades or boat hulls.
Figure 1.3: A boat hull being produced by resin infusion. Image source: Airtech International
(www.airtechintl.com).
Vacuum-only infusion processes have certain limitations: the rate of infusion is limited by
the available pressure difference to drive the flow (a maximum of 1 atmosphere), the achievable
fiber volume fraction is limited by the available compaction pressure (also a maximum of 1
atmosphere), and the bag-side of the cured laminate typically has a rough surface (due to the
absence of a rigid tool surface on the “B-side”). To overcome these three limitations, the resin
transfer molding (RTM) process can be used instead.
RTM is a closed-mold process, in which the vacuum bag is replaced by a second rigid tool
surface. The process steps are shown in Figure 1.4. First, the fibers are “preformed” in a separate
jig using a heated press. The dry fiber plies are typically pre-coated in a binder powder that holds
6
the preform together. The preform is then placed into the molding tool, which is held shut by bolts
or clamps. As with resin infusion processes, vacuum is applied to remove air initially present
within the mold cavity. The mold is then preheated and resin is injected under elevated pressure.
Finally, the mold is held at elevated temperature to allow the resin to cure, and afterwards the
finished composite part can be de-molded.
Figure 1.4: Steps of the RTM process.
RTM has several advantages over resin infusion: high-pressure resin injections take less
time than vacuum-only infusions, fiber volume fractions and dimensional tolerances can be tightly
7
controlled through careful design of the molding tool, and all surfaces of the molded part can have
a smooth finish. Furthermore, the process can be largely automated. The equipment costs for an
RTM system, however, are much higher than for the previously described processes, because the
molding tools must be extremely rigid to sustain the higher resin pressures without excessive
deflection (costs also increase substantially if robotics are used to automate the process). RTM is
therefore the preferred process for manufacturing of small to medium sized parts with complex
shapes and strict dimensional tolerances, particularly in high-volume production settings where
the equipment costs can be amortized over hundreds or thousands of parts.
The LCM process with the highest degree of complexity is high-pressure resin transfer
molding (HP-RTM). Whereas traditional RTM processes involve pressures ranging from 3 – 20
atmospheres, HP-RTM processes can have pressures up to 150 atmospheres. These processes have
gained traction in the automotive industry, where the high volume of production requires cycle
times to be much shorter than for aerospace composites. HP-RTM systems are often fully
automated, with pick-and-place robots for loading/unloading of parts, and hydraulic presses for
opening and closing the molding tools.
1.2.2 Prepreg processing
The majority of aerospace composites are fabricated using an intermediate product form
called a prepreg, which is a fiber ply (unidirectional or woven) that has been pre-impregnated
(hence the name) with a precisely controlled amount of partially cured (“B-staged”) resin. The
main advantages of prepreg processes over LCM processes are: (1) a controlled fiber volume
fraction, (2) ease of handling compared to dry fiber plies (the resin gives the prepreg “tack,” which
holds it in place once laid onto the tool), (3) the removal of the need to design an infusion/injection
strategy for each part, (4) the ability to use honeycomb core inserts (details are provided later in
8
this section), and (5) the shorter flow distances enable the use of higher viscosity resins (resins
with higher mechanical performance in the cured state are often more viscous in the uncured state,
particularly if additives such as toughening agents are incorporated into the formulation). The main
disadvantage of prepregs is that the resin must be pre-catalyzed before application to the fibers,
resulting in a perishable product that must be processed before the resin ages excessively. Most
prepregs must be stored in a freezer (a typical shelf life is 6-12 months), and the maximum
recommended room-temperature “out-time” is usually 30 days or less [3].
Early prepregs were produced by running dry fiber plies through a resin bath. A solvent
added to the resin was usually required to reduce the viscosity sufficiently for full saturation, and
after the resin bath, the prepreg was heated to allow the solvent to evaporate. However, complete
removal of solvent from resin – without curing the resin – is rarely feasible. Due to the tendency
of residual solvent to cause void growth during prepreg processing, alternative solvent-free
prepregging methods were developed. Most prepregs today are produced by a hot-melt process, in
which the resin is first made into films with precise thickness and then laminated onto dry fibers.
Processing of prepreg into finished parts consists of several steps: layup, bagging, and a
heated cure cycle. First, plies are cut from a roll into a desired shape, and placed with specified
fiber orientations onto a one-sided tool. This layup process is usually performed by hand, but large
structures with demanding tolerances and sufficiently simple geometries (e.g., wing skins of
commercial aircraft) can be laid up by machines, in processes such as ATL (automated tape layup)
and AFP (automated fiber placement).
Once prepreg plies have been laid onto the tool, a vacuum bag and related consumables are
placed over the uncured laminate. First, edge dams are placed around the perimeter of the laminate
to prevent excessive resin bleed, and often a dry fiberglass fabric is included to create a “breathing
9
edge dam” that allows gas evacuation from the edges of the laminate. Then a release film is placed
on top of the laminate to ease part removal after cure, which can be perforated to allow gas and
excess resin to escape through the upper surface of the laminate. Next, a nonwoven breather cloth
is placed above to provide continuity of gas escape pathways, and finally the vacuum bag film is
sealed to the molding tool, enveloping the entire assembly of prepreg and consumables.
Before heating the assembly to cure the resin, vacuum is applied under the bag to remove
air initially entrapped between and within the prepreg plies. This step, known as debulking, can
add significant time to the overall process, and for thicker laminates with many plies, intermittent
debulking is sometimes required after every few plies during the layup process.
The heated cure cycle often involves two stages. First the temperature is ramped to an
intermediate level that reduces the resin viscosity, causing the plies to consolidate as resin flows
into empty spaces, and (ideally) allowing any remaining gas bubbles to migrate to a low pressure
boundary of the laminate and escape. After a dwell period, the temperature is ramped to a higher
value and held in a second dwell period, which triggers the rapid advancement of the cure reaction.
As the resin cures, its viscosity increases until it transitions from a liquid into a gel, and further
advancement of cure causes the rubbery, gelled resin to vitrify into a rigid polymer.
The elevated temperatures of the cure process cause gas bubbles to increase in size (due to
ideal gas law behavior), and drive volatile species dissolved in the resin to evaporate (potentially
causing the formation of new gas bubbles). To address these problems, prepregs are often cured in
a heated pressure vessel called an autoclave. The elevated autoclave pressure is transferred to the
resin, suppressing the nucleation and growth of bubbles/voids.
10
Figure 1.5: Schematic of a prepreg laminate within an autoclave. Image source: [4].
Figure 1.6: A very large (and expensive) autoclave. Image source: ASC Process Systems (www.aschome.com).
11
Autoclave cure is often the preferred processing method for aerospace composites due to
its reliability; the elevated pressure “atones for many sins” that may occur in a production
environment, such as incomplete gas extraction due to poor vacuum quality during debulking.
However, autoclaves represent a bottleneck in the industrial production capacity of composites
because they are extremely expensive to acquire and operate (see Figure 1.6, showing an example
of a very large autoclave used to cure composite fuselage sections of the 787 Dreamliner).
Therefore, a growing interest in out-of-autoclave (OoA) prepreg processing has emerged in recent
years.
OoA prepregs are heated within an oven or using an integrally heated tool, and are more
sensitive to defects due to the absence of high pressures. Since gases trapped in the resin cannot
be driven into solution by elevated pressures, the extraction of gases prior to cure is critical. Thus,
OoA prepregs are designed with a partially saturated microstructure, in which dry regions at the
center of each ply act as gas evacuation pathways [5,6]. Since evacuation occurs in-plane, breath-
out distances for larger parts can become an issue, requiring long debulking times (on the order of
hours or even days). Recently, OoA prepregs with out-of-plane gas transport pathways have shown
potential to address the limitations of in-plane gas evacuation [7].
1.2.3 Co-cure of sandwich structures
Due to their planar geometry, composite laminates are generally not as stiff in out-of-plane
bending as they are under other loading conditions, unless additional elements are included in the
structural design (ribs, spars, hat-stiffeners, etc.). A common method for increasing bending
stiffness of a composite panel without adding separate stiffening elements is to include a hollow
honeycomb-celled core between plies. This “sandwich structure” design functions as the planar
analogue of an I-beam. Under bending loads, one side is in tension, the other side is in
12
compression, and the middle of the structure supports the resultant shear. By increasing the
distance of the load-bearing facesheets from the neutral bending axis, the stiffness of the structure
increases dramatically, with only a minimal weight penalty (see Figure 1.7).
Figure 1.7: Relative changes in stiffness, strength, and weight of a sandwich structure with increasing core
thickness. Image source: [8].
Figure 1.8: Elements of a honeycomb core sandwich structure.
13
Sandwich structures consist of three types of components, as shown in Figure 1.8: prepreg
facesheets, a honeycomb core insert (usually made from aluminum or a phenolic-coated aramid
paper), and a film adhesive that bonds the facesheets to the core. They appear in numerous
aerospace applications, including aircraft wing flaps, engine housings, and interior flooring panels
(see Figure 1.9). One method to manufacture such structures is called secondary bonding, in which
monolithic facesheets are cured individually (as described in the previous section) and then bonded
onto either side of a honeycomb core in a separate processing step. Although this approach is
relatively straightforward, it has two main drawbacks: (1) curing the facesheets and film adhesive
separately requires additional time, and (2) the facesheets must be perfectly formed for the
assembly to fit properly during bonding (any warping or distortion would prevent intimate contact
of the bonding surfaces).
Figure 1.9: Sandwich components on the Airbus A380. Image source: [9].
Both of these drawbacks can be eliminated by using a co-cure process, in which the
facesheets are cured and bonded to the core simultaneously using a single vacuum bagged
assembly. Although co-cure had the advantages of being a single-step process and guaranteeing
intimate contact of the bonding surfaces, it also introduces new complications, which arise as a
14
consequence of curing prepreg against the discontinuous surface of a honeycomb core. The upper
(bag-side) facesheet, rather than being supported by a solid tool surface as a monolithic laminate
would, is only supported by the edges of the core cell walls, resulting in “dimpling” of the plies
over the center of each cell. The lower (tool-side) facesheet also only experiences compaction at
the edges of the cells walls, and tends to “pillow” upwards at the center of each cell. Furthermore,
resin from both facesheets can bleed into the open core cells, leading to a loss of resin pressure
and potential resin starvation within the facesheets.
Some sandwich structures are made using a hybrid process, in which the tool-side facesheet
is cured as a monolithic laminate, and then the remaining elements (core and bag-side facesheet)
are co-cured onto the tool-side facesheet. For acoustic damping applications, these pre-cured tool-
side facesheets can be pre-drilled with an array of holes, which results in a sandwich structure
containing resonant cavities that can dissipate acoustic energy. Such structures are used for noise
reduction in the nacelles of commercial aircraft engines (see Figure 1.10).
Figure 1.10: Sandwich components in a commercial aircraft engine nacelle. Image source: Hexcel Corporation
(www.hexcel.com).
15
1.3 Process-induced defects in composite materials
Although composite materials can achieve exceptional mechanical properties, process-
induced defects can substantially decrease those properties, requiring designers to compensate by
using more generous safety factors, and sometimes requiring rejection of fabricated parts whose
defect levels exceed specifications. This section provides a brief overview of some common types
of process-induced defects. Later chapters consider particular defect types in more detail, in the
context of specific material/process combinations.
One category of process-induced defects pertains to incorrect fiber orientations. Fibrous
reinforcements can wrinkle, for example, due to errors in the layup process (hand layup or
automated processes), or when thick laminates with a high bulk factor (initial thickness divided by
final thickness) consolidate on tools with complex geometries. Another concern is geometric
deviations between the molding tool and fabricated parts, known as “spring-in” or warping, which
can occur due to the mismatch in thermal expansion coefficients of the fibers and the polymer
matrix. After the matrix gels at some super-ambient temperature during processing, shrinkage
upon cooling (and potentially chemical shrinkage due to polymerization) creates residual stresses
within the composite material. These stresses can cause the angles of molded parts, once de-
molded, to differ from the corresponding angles of the molding tool, and even flat laminates can
warp if the fiber ply orientations are not balanced and symmetrical across the mid-plane.
The other constituent of composites, the polymer matrix, can also contain process-induced
defects. For LCM processes, defects can form during the infusion/injection stage, for example,
premature resin gelation can lead to incomplete preform saturation and dry spots. The same
problems can occur if the locations of inlet gates and outlet vents are chosen poorly. Another
concern is the entrapment of air bubbles during infusion/injection. The dual-scale porosity of
16
woven fiber beds (i.e., micro-pores between individual filaments in a fiber tow, and macro-pores
between tows) results in dual-scale flow: capillary flow in the micro-pores and bulk flow in the
macro-pores [10]. A mismatch in these flow rates can create a partially-saturated flow front,
potentially leading to air entrapment in the fiber tows if the bulk flow leads the capillary flow, and
between the fiber tows if the capillary flow leads the bulk flow (shown in Figure 1.11).
Figure 1.11: Mechanism of void formation inside and between fiber tows: (a) capillary flow lagging and (b)
capillary flow leading. Image source: [11].
Air entrapment is a concern for prepregs as well, since air is inevitably trapped between
plies during layup. Room-temperature debulking evacuates air from any pore space that’s directly
connected to the vacuum bag environment, but not from gas pockets that are fully surrounded by
resin. Even out-of-autoclave prepregs, which have dry evacuation channels at the center of each
fiber tow, can trap air between layers, since the continuous resin films on each surface act as
barriers for gas transport. Figure 1.12 shows such a prepreg in the uncured state: the resin film on
the fabric surface is continuous except for occasional slits, so some inter-ply gas pockets could be
isolated from the connected network of evacuation channels. For OoA prepregs, the laminate is
maintained under vacuum during the heated cure cycle, which gives any remaining air bubbles
opportunity to escape once bubble mobility increases due to the drop in resin viscosity at higher
temperatures. Autoclave prepregs, on the other hand, are fully saturated and don’t contain
evacuation channels, but un-evacuated gas bubbles can be collapsed by the elevated autoclave
pressure.
17
Figure 1.12: Images of uncured OoA prepreg: (a,b) surface and (c) section view of two plies. Image source: [12].
For both LCM and prepreg processes, entrapped air bubbles that remain in the matrix upon
gelation become voids in the cured part, which are detrimental to mechanical performance.
However, entrapped air is not the only possible source of voids. Voids can also nucleate and grow
prior to gelation due to volatile species initially dissolved in the resin. One common volatile species
is water, which can be absorbed by the resin from ambient humidity [13]. Other possibilities
include solvents (which can be added to resin to reduce the viscosity), reaction byproducts (e.g.,
from phenolic resin’s condensation polymerization), and other impurities. In general, volatile-
induced voids can be prevented if the resin is cured under sufficient hydrostatic pressure, but
depending on the particular process used, maintaining resin pressure may not always be possible.
This issue is a recurring theme in the following chapters.
1.4 In situ process diagnostic methods
During industrial composite manufacturing processes, the only process variables that are
typically measured are temperature and pressure. Temperatures are commonly measured using one
or more thermocouples mounted onto the molding tools. For prepreg processes, pressure
transducers are used to monitor gas pressure within the vacuum bag and (if applicable) within the
autoclave. For LCM processes, tool-mounted transducers are often used to monitor resin pressure
during injection and cure. If a tool contains multiple pressure sensors, these can also be used to
track the progression of the flow front during injection.
18
Other process monitoring techniques exist, but are less common in production
environments (a comprehensive review of techniques can be found in [14]). One example is
dielectric sensors, which consist of interdigitated electrodes that emit a fringe field. When in
contact with a polymeric resin, the sensor can measure its dielectric properties (e.g., complex
impedance, complex permittivity), which can be correlated to the resin’s degree of cure and
viscosity [15]. This technique can be used for purposes such as assessing the remaining pot-life of
infusion resins or the elapsed out-time for prepregs. Other less-common process diagnostic
methods include non-contact laminate thickness measuring techniques (e.g., eddy current sensors
[16] and digital image correlation (DIC) [17]), and fiber optic sensors (fiber Bragg gratings), which
can be embedded within composite parts to monitor post-gelation strains during cure [18,19].
A simple yet powerful tool for diagnosis of void formation is visual observation. In general,
observing void growth in production environments is not feasible because the tools and equipment
used are opaque. In laboratory settings, however, dedicated tools for visual observation of
composites processing have been used. Examples include vacuum bag processing of prepregs
against glass tool plates [20,21] and miniature RTM tools with glass viewports [22]. Visual
observation of defect formation phenomena was the investigative approach chosen for the work in
this dissertation. Two tools that enable process visualization were custom-built for these studies:
a lab-scale RTM tool with a glass window, and a lab-scale autoclave that enables visual
observation of the core/facesheet bond-line of co-cured sandwich structures.
1.5 Scope of dissertation
This dissertation spans two research projects. The goal of the first project was to identify
the cause of – and potential solutions for – surface defects in composites produced by RTM using
a prototype benzoxazine/epoxy resin. The goal of the second project was to develop an integrated
19
physics-based process model for autoclave co-cure of honeycomb core sandwich structures. In
both projects, (1) a custom tool was designed and built to enable in situ observations of process-
induced defect formation, (2) relevant defect formation mechanisms were identified, and (3)
process modifications to prevent defect formation were proposed and demonstrated to be effective.
Chapters 2 and 3 of this dissertation pertain to the RTM project, while chapters 4 and 5 describe
contributions to the co-cure project. A summary of the scope of each chapter is provided here:
Chapter 2: We studied the mechanism of volatile-induced surface porosity formation
during the resin transfer molding (RTM) of aerospace composites using a blended
benzoxazine/epoxy resin, and identified reduction strategies based on material and processing
parameters. The influence of viscosity and pressure on resin volatilization were determined. Then,
in situ data was collected during molding using a lab-scale RTM system for different cure cycles
and catalyst concentrations. Finally, the surface quality of molded samples was evaluated. The
results show that surface porosity occurs when cure shrinkage causes a sufficient decrease in cavity
pressure prior to resin vitrification. The combination of thermal gradients and rapid gelation can
generate large spatial variations in viscosity, rendering the coldest regions of a mold susceptible
to porosity formation. However, material and cure cycle modifications can alter the resin cure
kinetics, making it possible to delay the pressure drop until higher viscosities are attained to
minimize porosity formation.
Chapter 3: We considered the design of two-stage cure temperature cycles for the resin
from the previous chapter, which were shown to prevent volatile-induced surface porosity.
Although the addition of an intermediate temperature dwell improved surface quality, it also added
significant cycle time to the processing, therefore a method to identify the fastest cure cycle that
would avoid surface porosity was desired. A process map was developed, which enabled the design
20
of cure cycles with optimal intermediate dwells. First, the resin cure kinetics were characterized
using a “model-free” isoconversional method. Next, thermogravimetric analysis (TGA) and
rheological dynamic analysis (RDA) were used to determine a threshold mechanical state, above
which the release of volatile species can no longer occur. Finally, molded composite samples were
fabricated in the lab-scale RTM tool, which featured temperature and pressure sensors, as well as
a transparent mold wall that enabled in situ observation of surface porosity formation. The results
were combined into a process map that shows, for any intermediate dwell temperature, the allowed
dwell time window to produce porosity-free parts. Furthermore, by considering an RTM tool with
a given tolerance for temperature control, the process window can be used to design the fastest
cure cycle that properly accommodates for the magnitude of thermal gradients present. Finally, to
validate the effectiveness of this approach, a molded sample was fabricated using an “optimized”
cure cycle designed with the process map and the surface quality was compared to the baseline
cases. Altogether, this work clarifies the complex mechanisms that can lead to surface porosity
formation during the RTM processing of this resin, and provides a practical method for identifying
process modifications that can reduce defect levels.
Chapter 4: We investigated the autoclave co-cure process for aerospace sandwich
structures, in which prepreg facesheets are cured and bonded to honeycomb core simultaneously.
Co-cure involves coupled physical phenomena, including gas migration, prepreg consolidation,
resin/adhesive flow and crosslinking, and potential mechanisms of defect formation. Due to the
“black box” nature of the process, however, scientific understanding of the complex and
interacting phenomena that cause defects remains incomplete. We addressed this challenge using
an in situ visualization method that enabled direct observation of the skin/core bond-line in realistic
autoclave conditions. Five cases are presented, which span a range of process conditions leading
21
to various defect-formation phenomena, and show the effectiveness of in-bag pressurization for
preventing bond-line defects. We demonstrated that in situ diagnostics can eliminate much of the
trial-and-error typically involved in process troubleshooting, by providing insights into the physics
of co-cure that would otherwise be difficult to obtain.
Chapter 5: The previous chapter showed that, during co-cure, the microstructural
characteristics of the facesheets and bond-line are influenced by the core pressure, which, in turn,
can depend on gas transport through the facesheets. Prior research has identified complex behavior,
namely evolving facesheet permeability with strong temperature and pressure dependence.
However, to date, a physics-based description of through-thickness gas permeability for prepregs
has remained elusive. In this work, we characterized the gas permeability of a resin-saturated
prepreg in co-cure-specific conditions using steady-state (constant flow-rate) and transient
(falling-pressure) experiments. The results show that gas transport is closely linked to the presence
and mobility of the prepreg resin. An empirical model for prepreg gas permeability is presented
and its limitations are discussed, motivating the need for a physics-based model. A multiphase
flow model based on partial desaturation of a fluid-filled porous medium is proposed to predict
the observed behavior. This study, for the first time, couples gas transport and evolving prepreg
saturation to predict evolving permeability and core pressure. The work applies directly to co-cure
of sandwich structures, as well as other processes that involve gas evacuation from prepregs.
The final chapter of this dissertation summarizes the knowledge gained from each project,
proposes possible future work on these topics, and discusses the overall significance and broader
implications of this body of work.
22
CHAPTER 2. Eliminating surface porosity during resin transfer molding
(RTM)
2.1. Introduction
Resin transfer molding (RTM) has been used to produce composite structures for aerospace
applications since the early 1980s [23]. RTM is preferred for geometrically complex small to
medium-sized parts that require low microstructural defect levels and excellent surface finish. The
process can also be largely automated to improve production rates and repeatability, allowing
medium to high volume production of high performance composites [24].
Resin transfer molding typically consists of three stages. The first, preforming, consists of
preparing the fiber reinforcement by cutting and stacking plies of dry fibers, pre-shaping them by
heated compaction, and placing them within the mold cavity. The second step involves injecting a
pre-catalyzed but uncured thermoset resin into the heated mold cavity and saturating the fibrous
preform. The final stage consists of imposing a temperature and pressure cycle that cures the resin
while suppressing the formation of microstructural defects.
Voids are the most common type of defect encountered in RTM parts. They are often a
result of air trapped within the mold cavity during injection, which can occur in the form of dry
spots from converging flow fronts [25], or incomplete preform saturation due to an imbalance
between the capillary and bulk flows that occur within dual-scale woven preforms [26,27]. Resin
infiltration can be optimized by adjusting the gate locations [28] and inlet pressure [26] to
minimize air entrapment. For a given injection scheme, air-induced voids can be further eliminated
by applying vacuum to the mold cavity prior to (and during) injection, flushing additional resin
23
through the system to evacuate bubbles, and increasing the applied hydrostatic pressure during
cure [29]. During typical RTM, the microstructure achieved at the end of the injection stage
remains largely stable during subsequent cure.
However, in some cases, voids can also arise from a second source: volatiles released by
the resin at elevated temperatures. The positive hydrostatic pressure used during RTM (in contrast
to vacuum-only resin infusion processes) was first developed to suppress the volatilization of water
during the condensation cure of phenolic resins [23,30]. The source of volatiles is not limited to
byproducts of polymerization – gas release can also occur due to residual solvents, vaporized
monomers, dissolved air and moisture, degradation byproducts, or other impurities/contaminants.
The detection, analysis, and control of volatile-induced porosity is particularly challenging because
voids can form after injection, at any point during the cure stage.
2.1.1. Literature review
Voids in thermoset matrix composites are known to be detrimental to both mechanical
properties [31] and cosmetic appearance [32]. However, while multiple studies have addressed the
formation of voids during resin injection, relatively few studies have been devoted to the topic of
void behavior during the curing stage of RTM.
The major void modeling approaches for thermoset composites derive from studies by
Kardos et al. [33] and Wood and Bader [34]. Kardos at al. [33] developed a model for void
nucleation and growth in the context of autoclave processing with an epoxy resin. They considered
a diffusion-based mechanism, and assumed water to be the primary diffusible species. The model
described both voids that exist initially at the onset of cure (and contain dry air or an air/water
mixture), and voids that nucleate spontaneously at supersaturated conditions and contain only
24
water vapor. They predicted a strong influence of initial dissolved moisture concentration on final
void size and pressure. Furthermore, they noted that void growth cannot occur if the void gas
pressure exceeds the saturated vapor pressure of the volatile species dissolved in the resin, and
constructed a “stability map” showing the resin pressure required to suppress void growth as
function of temperature and dissolved moisture content. Wood and Bader [34] described a similar
diffusion-based model for void growth in autoclaved epoxy laminates, which instead considered
nitrogen as the primary diffusible species. By experimentally determining surface tension,
dissolved gas concentration, and the gas diffusion coefficient, they were able to predict time-
dependent changes in bubble radii. Ledru et al. [35] developed a visco-mechanical void growth
model (also in the context of epoxy-matrix composites cured by autoclave), which described a
time-dependent void radius as a function of external pressure, temperature, viscosity, and surface
tension, for voids containing a fixed amount of gas (i.e. omitting diffusion effects). They
subsequently coupled the visco-mechanical model with a (water) diffusion-based model [36],
aiming to improve upon previous diffusion-based models by accounting for viscosity and polymer
crosslinking effects and by refining the predicted influence of hydrostatic pressure. They noted
that, compared to diffusion-only models, visco-mechanical phenomena reduce void size. Like
Kardos, they reported that initial dissolved volatile concentration and applied pressure were the
most significant factors for void size, but they encountered difficulties in confirming the diffusion
coefficient (which also strongly influences model predictions).
While the studies referenced in [33–36] consider autoclave processing, Lundström
confirmed that, in resin transfer molding processes, hydrostatic pressure can also cause gas
dissolution and can collapse voids entirely [37]. The concentration of volatile species initially
present in the resin also can be reduced prior to injection by vacuum-degassing [38], but the
25
effectiveness of this technique is limited by the resin pot life, since degassing requires low viscosity
and thus must be performed at high temperatures. Various time/temperature/vacuum pressure
combinations can be used to influence the amount of volatiles extracted, but excessive vacuum-
degassing increases the risk of pre-curing the resin, which can shorten the pot-life and complicate
resin injection.
The volumetric change of thermoset resins during cure has been associated with defect
formation. Eom et al. showed that, in three-dimensionally constrained thermoset resin, internal
tensile stresses can develop due to chemical cure shrinkage, leading to void formation [39].
Furthermore, they found a critical stress criterion and developed process windows, providing
guidelines to prevent void formation in autoclaved glass/epoxy laminates [40]. Similarly, Wisnom
et al. [41] studied the effects of thermal and chemical volumetric changes on residual stresses in
prepreg laminates, noting that it is possible for significant stresses to develop due to tool-part
interactions before vitrification and even before gelation, where the resin may have a very low
shear modulus but an appreciable bulk modulus. Merzlyakov et al. [42] also measured stresses in
constrained thermoset resin during cure, finding that cure-induced tensile stresses were lower than
expected, due to cohesive failure of the resin in the gelled (rubbery) state.
Cure shrinkage can be measured by monitoring sample thickness between parallel plates
on a rheometer [43], using a pressure-volume-temperature (PVT) analyzer [44], by a gravimetric
method [45], or by other methods [46]. The effects of resin volumetric changes on cavity pressure
in RTM have been described by Kendall et al. [47], who noted pressure increases due to thermal
expansion, as well as pressure drops attributed to cure shrinkage. Haider et al. [32] observed similar
pressure drops due to cure shrinkage in automotive RTM panels, and correlated these with an
increased surface roughness. Their unsaturated polyester resin exhibited much greater cure
26
shrinkage (7-10%) than epoxies, which was successfully compensated by including a
thermoplastic low profile additive (LPA). The expansion of LPA after gelation acted to reestablish
positive mold pressure, resulting in high-gloss “class A” surface finishes. Boyard et al. [48] also
noted increased surface roughness due to shrinkage-induced pressure drops in unsaturated
polyester bulk molding compound (BMC), and used a dilatometer to develop a model for
predicting cavity pressure. Recently, Landry and Hubert [49] described similar surface roughness
due to shrinkage in thermoplastic short-fiber PEEK composites, developed a model to predict the
pressure distribution, and correlated surface defects in colder zones with local pressure drops due
to non-uniform shrinkage [50].
While the studies in the previous paragraph describe shrinkage-induced defects similar to
those in our study, they differ in that none of the matrix materials exhibited significant volatility.
For resins that can off-gas under deficient pressure, shrinkage-induced surface defects manifest as
bubbles instead of increased roughness. This distinctive type of defect is particularly challenging
to diagnose, since it can easily be mistaken for porosity due to air (trapped during injection) unless
in situ observations are used to identify the nature of the porosity formation. Lab-scale RTM tools
have been used previously to observe in-mold volatile release, for example by Pupin et al. [51].
Their “Nano RTM” contained a glass window, allowing in situ observations of volatile release in
a phenolic resin. The primary volatile species was water – a byproduct of the phenolic
condensation reaction – which was produced in such quantities that pressure alone could not
suppress void nucleation. Void-free parts were obtained by removing the water via vacuum-
degassing within the mold until gelation. The volatility of the resin in this study, in contrast, is due
primarily to residual solvent, which can be forced to remain dissolved in solution using only
modest pressures. However, due to resin cure shrinkage, maintaining mold cavity pressure is not
27
always possible. The combined effects of shrinkage-induced pressure drops and high resin
volatility led to the topic of this work: volatile-induced surface porosity.
Figure 2.1: A laminate surface with extensive volatile-induced porosity.
In this study, we considered the RTM processing of a blended thermoset resin consisting
of benzoxazine and epoxy components. This combination was being investigated by Henkel, a
commercial resin supplier, for structural aerospace applications due to expected improvements in
high-temperature performance. Blending benzoxazine with epoxy has been shown to increase
cross-link density compared to pure benzoxazine, increasing both the Tg and toughness [52,53].
This copolymer offers a compromise, in terms of both cost and maximum service temperature,
between standard aerospace epoxies and ultra-high-temperature resins such as bismaleimides
(BMIs) and polyimides. However, the complex polymer chemistry results in comparatively
complex in-process behavior, particularly in terms of increased volatile release during cure. If not
properly controlled, this volatility can result in significant surface porosity on molded parts (shown
28
in Figure 2.1), preventing the production of laminates with high quality surface finishes. During
this study, several resin formulations were used to relate changes in resin properties to in-process
behavior and final part quality. These experimental variants were produced in batches in lab-scale
conditions. As a result, while the formulations may not have been as optimized as a commercial
product, they formed a representative set of model materials that made it possible to study volatile-
induced porosity by demonstrating key phenomena. The study in turn enables the development of
guidelines applicable to similar resin systems.
This study was part of a larger research effort [54,55] that included a companion study,
which focused on several aspects of the thermochemical evolution and in-process behavior of a
baseline formulation of this resin [56]. The primary volatile species was identified by Fourier
transform infrared spectroscopy (FTIR) as ethyl acetate, a residual solvent used in the synthesis of
the resin. While benzoxazines are traditionally known to exhibit low volatility, the experimental
formulations supplied for our work were prepared using an alternative method that avoided the
need for more hazardous solvents. The drawback of this approach was that the ethyl acetate solvent
could not easily be removed prior to cure. FTIR analysis of the volatiles also indicated the presence
of water, but a vacuum-degassing pretreatment was shown to reduce the amount of volatilized
water below the detectability limit in the FTIR signal. At higher temperatures (>170°C), monomers
and monomer fragments (reaction intermediates and/or degradation products) were also observed
to volatilize to some extent, although the precise relative abundances of the various species were
not ascertainable by FTIR. The cure shrinkage of the resin was measured by a modified
pycnometry approach [55], chosen because volatile release during cure at ambient pressure
precluded the use of most traditional methods. The volumetric shrinkage due to polymerization,
with negligible mass loss, was shown to be about 1%.
29
2.1.2. Objectives and approach
Despite the prior studies cited above, the causes of volatile-induced surface porosity during
liquid molding of such resins remain unclear, and effective strategies for avoiding defect formation
must be identified. Consequently, our primary objective in this study was to clarify the connections
between in-process resin behavior, temperature and pressure processing conditions, mold
characteristics, cure phenomena, and surface porosity.
The porosity described herein is more complex than the more commonly studied porosity
due to air and water, since there is an additional causative species: a residual solvent. The behavior
differs in that the ethyl acetate is entirely soluble in the resin – in contrast to the limited solubility
of air – and bubbles can very rapidly nucleate and grow, or shrink and collapse, due to changes in
applied pressure. Thus, in this study, our approach consisted of experimentally investigating
process phenomena, identifying a minimum “safe pressure” to prevent void nucleation and growth,
and exploring process conditions that complicate pressure control for the RTM of such materials.
First, we investigated the mass stability of the resin and its relationship to viscosity by
thermogravimetric analysis (TGA) and rheological dynamic analysis (RDA). Second, the in-mold
behavior of the resin at various cure pressures was analyzed by fabricating neat-resin molded
samples in a highly-instrumented lab-scale RTM tool, which included a transparent tool plate to
enable visual observation of the mold cavity throughout processing. Finally, the formation of
surface porosity during the processing of carbon fiber-reinforced molded samples was investigated
by fabricating samples within the same RTM tool.
Combining temperature, pressure, and viscosity data with direct in situ observation
capabilities provided a powerful set of tools to understand the mechanism behind the formation of
30
surface porosity. The results showed that there exists a critical viscosity above which volatilization
ceases, and a critical hydrostatic pressure above which volatile release and void formation are
suppressed. If the resin pressure falls below the critical value during cure, before the viscosity
attains its critical value, void growth occurs. Such a condition can occur due to the complex
combined action of chemical cure shrinkage and thermal gradients, which renders the coldest
surface of the part susceptible to void growth. By describing this distinctive void formation process
in detail, we clarify the processing conditions that cause volatile-induced surface porosity and
identify potential material and process modifications that can reduce or eliminate such defects
during RTM.
2.2. Materials
The resin selected for this study consisted of benzoxazine and epoxy constituents as well
as a catalyst used to accelerate the cure reaction. The resin was designed for vacuum degassing
and injection at 110°C followed by cure at a nominal temperature of 185°C, though modified
thermal cure cycles are possible. The nominal recommended cure pressure is 450 kPa. A free-
standing post-cure for 30 minutes at 220°C is recommended to maximize mechanical properties,
but was not considered in this study because the porosity formation phenomena of interest occurred
earlier, during in-mold processing.
Three formulations were prepared, with varying catalyst concentrations to modify the in-
process behavior of the resin. The catalyst was used to accelerate the reaction, but did not affect
the chemical composition of the matrix constituents. The baseline formulation contained 0.1% by
weight, while the subsequent formulations contained 1% and 2% catalyst by weight, respectively.
These three formulations are henceforth designated F1, F2, and F3, in order of increasing catalyst
concentration. The baseline formulation, F1, was used throughout the study to identify the effects
31
of pressure and the mechanical state of the resin on volatile release, whereas the F2 and F3
formulations were used to modify the cure kinetics during composite manufacturing tests, as
outlined in section 2.3.2.
The fiber bed selected to mold composite samples consisted of a five-harness satin (5HS)
carbon fiber fabric (Sigmatex Ltd.) with an areal weight of 364 g/m
2
and a 3000 fiber/tow count.
The fabric included a thermoplastic binder to facilitate preforming and placement into the mold
cavity.
2.3. Experimental methods
2.3.1. Thermal characterization
Thermogravimetric tests were performed with resin formulation F1 (using a TA
Instruments Q5000IR) to determine the volatile-induced mass loss. Samples weighing 40 ± 0.5 mg
were cured under atmospheric pressure, by increasing the temperature from 35°C to 250°C at rates
of 1, 2, and 3°C/minute.
The resin viscosity was measured with a rheometer (TA Instruments AR2000ex), with
disposable aluminum parallel-plate fixtures and an environmental test chamber (furnace)
accessory. All tests were performed in oscillatory mode at 5 Hz and with a 500µm gap. The test
procedure was designed to capture the material behavior before and after gelation by measuring
the absolute magnitude of the complex viscosity, which is dominated by viscous behavior (i.e. the
loss modulus G”) pre-gelation and becomes dominated by elastic behavior (i.e. the storage
modulus G’) as the resin approaches the later stages of cure. Prior to gelation, 1% strain was chosen
as the controlled variable. To remain within the torque limit of the instrument upon resin gelation,
32
a cross-over condition corresponding to a viscosity greater than 1000 Pa·s was set, at which point
the program control mode switched to torque-control at 500 µNm.
Rheological dynamic analysis (RDA) tests were performed in three sets, as summarized in
Table 2.1. The first set of RDA tests consisted of linear temperature ramps with formulation F1
identical to those carried out by TGA. A second set of tests consisted of cure cycles designed to
approximate the measured molding conditions described below. Finally, the viscosity of each
formulation was measured for 8 hour isothermal holds at the manufacturer’s recommended
injection temperature of 110°C to compare the expected pot life. The pot life of the F1 formulation
was >8 hours at the injection temperature, while the pot lives of F2 and F3 were ~5 and 4 hours,
respectively.
Table 2.1: Thermal characterization tests
Test set Test type Temperature cycles Formulation
Linear ramps RDA, TGA 35 - 250°C at 1, 2, 3°C/min F1
Realistic cycles RDA Cycle A
a
, Cycle B
b
F1, F2, F3
Pot life RDA 110°C for 8 hours F1, F2, F3
a. Cycle A = 2°C/min to 185°C, 90 min dwell
b. Cycle B = 2°C/min to 130°C, 180 min dwell, 2°C/min to 185°C, 90 min dwell
2.3.2. Molding
Lab-scale RTM system
The lab-scale RTM tool built for this study is shown in Figure 2.2. The main body of the
mold was an anodized aluminum block containing inlet and outlet ports with “line injection”
grooves used to generate a one-dimensional flow front. A “picture frame” plate with a rectangular
cavity was placed against the main body to define the thickness (3.2 mm) and in-plane dimensions
(76 127 mm) of the part. The second tool face was a 20 mm thick tempered glass plate, rated
33
for pressures up to 1200 kPa. This glass plate formed a rigid window that allowed direct in situ
observation of in-mold phenomena, including air evacuation and preform saturation during
injection, and the time and location of any subsequent void formation.
Figure 2.2: Exploded view of the lab-scale RTM tool.
The temperature control system of the RTM tool consisted of a K-type thermocouple
(OMEGA Engineering Inc.), a PID controller (Watlow EZ-ZONE
®
model PM6R1CA), and two
300 W heating rods embedded through the length of the main tool body. A pressure transducer
(GP:50 NY Ltd., model 131) was mounted at the center of the aluminum tool face. Temperature
34
and pressure data were recorded with a cRIO-9076 data acquisition system (running LabVIEW
2012, National Instruments). Additional ports were included in the main tool body to allow for
alternative thermocouple placement and the addition of other transducers (such as dielectric cure
monitoring sensors). However, such sensors were not used in this study. Visual data through the
window was recorded during both injection and cure. At the macro-scale, a DSLR camera (Canon
EOS Rebel T1i) and the intervalometer feature of Magic Lantern (v2.3), a third-party open source
software add-on, were used to capture full-field views of the molded part. At the micro-scale, a
low-magnification portable USB microscope (Dino-Lite model AD4113T) was used to obtain
high-resolution images.
The placement of the heating rods was designed to minimize in-plane temperature
gradients, but the nature of the overall geometry – with a glass tool-plate on one side – created an
inevitable through-thickness temperature gradient. During the molding of composite samples, the
temperature histories were measured directly at both the hot and cold sides of the tool, for at least
one sample with each cure cycle, by placing additional thermocouples into each side of the mold
cavity between the preform and the mold walls.
Neat resin samples
Molded neat-resin samples were fabricated using the F1 formulation under various
temperature and pressure conditions to analyze the pressure dependence of volatile release and
void formation. No reinforcing preform was used in these tests to allow observation of bubble
growth through the entire sample thickness.
The tool surfaces were coated with a liquid mold release agent (Frekote 770-NC, Henkel
Inc.) prior to sealing the mold. To minimize spatial temperature gradients, two layers of fiberglass
insulating fabric were wrapped around the RTM. A hole was cut in the insulation in front of the
35
window, and a second glass plate was clamped over it to act as a heat shield while preserving
transparency into the mold cavity.
Prior to injection, the resin was vacuum-degassed for 45 minutes at a reduced pressure of
6 kPa and at the nominal injection temperature of 110°C. While this procedure removed entrapped
bubbles and some of the dissolved volatiles from the resin, it was not possible to remove the
residual solvent entirely. Increased degassing times and temperatures were found to provide
diminishing returns and to increase the risk of pre-curing the resin.
Injections were performed at 300 kPa and 110°C with a pneumatic injector (Radius
2100cc). Vacuum was applied to the cavity both before and during injection using an external
vacuum pump. Excess resin was flushed through the outlet port until no bubbles remained in the
outlet tubing or in the cavity, as observed through the window.
After injection, a constant hydrostatic pressure was applied to the system by closing the
outlet valve, leaving the inlet valve open, and setting the pressure to a desired level on the injector.
Pressures of 101 (ambient), 160, 200, and 450 kPa were applied. The cure temperature cycles
consisted of 2°C/minute ramps to isothermal holds at 170, 185, or 200°C for 120 minutes. All
molding tests are summarized in Table 2.2.
Table 2.2: Processing parameters for molded samples fabricated in the lab-scale RTM.
Injection parameters Cure parameters
Type Press. (kPa) Temp. (°C) Press. (kPa) Temp. (°C) Formulation
Neat resin 300 110 101, 160, 200, 450 170, 185, 200 F1
Composite 300 110 450 Cycle A
a
, Cycle B
b
F1, F2, F3
a. Cycle A = 2°C/min to 185°C, 90 min dwell
b. Cycle B = 2°C/min to 130°C, 180 min dwell, 2°C/min to 185°C, 90 min dwell
36
Composite samples
Composite panels were manufactured with all three formulations using 8 layers of carbon
fabric set in a quasi-isotropic layup. Injection procedures were identical to those of the neat resin
samples, and the post-fill packing pressure was set to 450 kPa for all tests. Two cure temperature
cycles were used. The baseline, Cycle A, consisted of a 2°C/min ramp to 185°C followed by a 90
min dwell. A modified cure cycle, Cycle B, included an intermediate 180 min hold at 130°C before
the high-temperature dwell. The measured temperature profiles during these tests were accurately
replicated using combinations of ramps and holds within the rheometer to determine the evolution
of the resin viscosity.
Porosity assessments
The time of porosity formation was determined through time-lapse videos recorded at 30 s
intervals, at the macro-scale by DSLR camera and at the micro-scale by USB microscope. Micro-
scale visual data allowed precise timing to be determined for the formation of voids that were too
small to be observed otherwise, while macro-scale data was used to confirm that the microscope
observations were representative of the entire sample.
After de-molding, high-resolution photographs of the surfaces of each sample were used
to quantify surface porosity. An open-source image processing software (ImageJ v1.48) was used
to create binary maps of high and low porosity zones for quantifying the percent defective area on
the surface of each sample. Due to the variations in appearance of surface voids (which can
manifest as pits on the surface or as subsurface bubbles) and the high reflectivity of subsurface
carbon fibers, a semi-automated method was used, which combined automatic selection and
manual confirmation of defects.
37
2.4. Results and discussion
2.4.1. TGA and RDA data
In Figure 2.3, we compare the TGA and RDA data for linear temperature ramps with resin
formulation F1. The temperature range is truncated to the region of interest for clarity. The TGA
data is expressed as a unit-less ratio of the instantaneous sample mass to the initial mass. The RDA
data shows the absolute magnitude of the complex viscosity.
Figure 2.3: Mass ratio and viscosity for linear temperature ramps.
38
The rate of mass loss due to volatilization increased with temperature, until it abruptly
halted due to resin solidification. The faster temperature ramps exhibited lower total mass loss,
because there was less time for volatiles to evolve. However, all samples exhibited between 11%
and 13% mass loss, a large amount compared to traditional RTM resins [24]. Note that as these
tests were performed at ambient pressure, they represent a “worst case scenario” for RTM
processing in which the mold is not pressurized. Furthermore, the TGA tests differ from in-mold
processing conditions in that the samples have a free surface (and a large surface area to volume
ratio) exposed to a purge gas over which evaporation could occur, whereas within an RTM, volatile
voids must first nucleate and are thus also influenced by constraining fibers, surface tension, and
other factors. It is therefore not the total amount of weight loss by TGA that is comparable to RTM
processing (and relevant for our purposes), but rather the state of cure at which volatile release can
no longer occur. The volatilization behavior of the baseline formulation was studied and is
described in greater detail in a companion paper [56].
The resin viscosity remained low (<0.2 Pa·s, the lower detectability limit using this
geometry and set of test parameters) up to temperatures of 185 - 205°C (depending on ramp rate).
In that range, the viscosity exhibited a rapid six order-of-magnitude increase, until finally
stabilizing at ~1.2 10
5
Pa·s. The approximate gel points, estimated using the common G’-G”
crossover condition, are indicated by square markers. A full frequency-sweep analysis was not
performed to determine the “critical gel” time at which tan delta (tan δ = G”/G’) becomes
frequency independent [57]. However, additional tests at 1 Hz and 30 Hz showed similar behavior
and resulted in gel times that varied by no more than 3 minutes. A rapidly increasing storage
modulus (compared to the more slowly increasing loss modulus) caused the crossover to occur at
relatively low absolute values of complex viscosity. Note that, after gelation, since the resin is no
39
longer a liquid, the complex viscosity values are primarily a measure of the increasing elastic
(storage) modulus of the resin. The evolution of this curve provides a convenient method for
estimating the effect of cure on the mechanical state of the resin. The 1°C/minute test exhibited a
peak in tan delta centered at 204°C (indicated by the diamond marker), which can be used to
estimate the time of vitrification, as described by Bilyeu et al. [58]. No such peak was observed at
higher ramp rates, likely because the temperature was increased too quickly for the glass transition
temperature Tg to ever exceed the imposed temperature. For the RDA tests corresponding to
molded samples (section 2.4.3), vitrification occurred at or near isothermal conditions, and was
clearly detectable by a peak in tan delta.
The combined TGA and RDA datasets can be used to identify the resin state at which mass
loss ceased. In Figure 2.3, the cessation of mass loss was related to a corresponding viscosity.
Interestingly, these “critical viscosities” reside within a narrow band, irrespective of ramp rate,
suggesting that there exists a threshold mechanical state above which volatilization can no longer
occur. This state was reached just before vitrification (certainly no later, since the mobility of
potentially volatile molecules becomes greatly reduced when the polymer network transitions to a
glassy phase). Note, however, that mass loss can continue past the gel point, well into the rubbery
phase. A value of 7 10
4
Pa·s is used in section 2.4.3 as a conservative estimate for the critical
viscosity, corresponding to the 2°C/min ramp rate used in the temperature cycles for all molded
samples.
2.4.2. Neat resin molded samples
The neat resin panels were manufactured without entrapping air during injection. The test
results are summarized in Figure 2.4 as a process map relating cure temperature, imposed mold
pressure, and defect growth [33]. Each point represents one sample. Temperature and pressure
40
combinations that resulted in the generation of voids are shown in red, while samples that did not
grow voids are shown in blue. In cases when voids did develop, they nucleated on the hottest side
of the mold cavity (the tool side that contains the heating elements), grew rapidly, sometimes
detached and floated upwards, and eventually were fixed in place when the resin gelled. Figure
2.5 shows an example of void growth as observed in situ for a sample cured at ambient pressure
(the circular object visible in the background is a plug in the tool plate, which provides the option
to install additional sensors).
Figure 2.4: Temperature/pressure map of neat-resin molded samples. Red indicates conditions that lead to void
growth, while blue corresponds to conditions that suppress void growth. Points correspond to individual tests and
background shading is used to highlight the general trend.
41
Figure 2.5: In situ observations of bubble nucleation and growth in resin cured at ambient pressure.
All samples cured at ambient pressure exhibited significant void growth and averaged
~10% bulk porosity by weight (compared to a void-free reference sample). Implying, not that 10%
of the resin in the cavity vaporized, but rather that enough gas was released to occupy 10% of the
cavity volume. The small sample cavity and lack of constraining fiber preform allowed bubbles to
grow by displacing still-liquid resin (which was pushed out of the open vent). However, with a
closed vent and cavity pressure provided through the inlet, pressures above approximately 200 kPa
effectively suppressed void nucleation. The exact critical pressure for void suppression exhibited
a mild temperature dependence, since tests at 160 kPa developed bubbles at 200°C but not 170°C.
Overall, the threshold pressures required for the temperature range of interest were well within the
42
limits of almost all standard RTM systems. A conservative value of 200 kPa was chosen as a
“critical pressure” criterion, below which void nucleation and growth can be expected during cure.
Furthermore, these results were used as guidelines for the processing of the carbon fiber reinforced
samples described in the following section. Specifically, an imposed pressure of 450 kPa, which
is more than twice that required to suppress volatilization, was chosen to ensure that no bulk voids
formed during cure. Although not described here in detail, fiber-reinforced samples cured at
ambient pressure did exhibit void growth similar to neat-resin samples, with significant porosity
appearing primarily in the resin-rich zones between fiber tows. However, elevated pressure was
equally effective in preventing volatilization through the bulk of reinforced samples as in neat resin
samples.
2.4.3. Surface porosity
Despite sufficient applied pressure, surface porosity developed exclusively on the colder
side of all manufactured samples. The window in the mold afforded visual confirmation that,
initially after injection, the surfaces of all samples were bubble-free. However, porosity developed
on the window side during high-temperature dwells late in the cure cycle. The aluminum tool-side
(which was slightly hotter, since it contained the heating elements) had effectively no defects for
all samples, and micrographs of polished cross-sections confirmed that internal porosity was also
negligibly low. For this reason, only the quality of the cold sides is considered in the subsequent
description.
Table 2.3 summarizes the percent defective area of the cold side of each sample. Increased
catalyst concentrations resulted in reduced surface porosity, and cure Cycle B (which included an
intermediate 3-hour dwell at 130°C) produced superior surface quality compared to the baseline,
Cycle A. Figure 2.6 consists of representative images of cold-side laminate surfaces, in both full
43
color and binarized forms. The thermocouple used to directly measure the window-side
temperature history is visible in the first image.
Table 2.3: Percent defective area for the cold sides of molded laminates.
Formulation
F1 F2 F3
Cycle A 26.0% 3.32% 3.48%
Cycle B 36.8% 0.44% 0.17%
Figure 2.6: Some examples of surface porosity quantification by image binarization and measurement of the percent
defective area. The top row is formulation F1, Cycle A, middle row is F2, Cycle A, and the bottom row is F2, Cycle
B. The thermocouple used to measure the window-side temperature history is visible in the first image.
44
In situ data and porosity formation mechanism
Figure 2.7 shows the measured sensor data for a composite sample with significant porosity
(formulation F1 cured under Cycle A, 26% defective area). The temperature graph consists of the
controller temperature (black) and the temperatures of the hot and cold sides of the mold cavity
(in red and blue, respectively). The hot side, being closest to the heating elements, followed the
control temperature to within 1°C, while the cold side lagged by roughly 4 minutes (or,
equivalently, 8°C) during ramps and stabilized at 4°C below the hot side at steady-state conditions
during the high-temperature dwell.
The RDA data corresponding to the tool (hot) and window (cold) sides of the sample is
shown in Figure 2.7b. The viscosity remained low during the temperature ramp, but rose abruptly
after the gel point (indicated by square markers). At this point, the mismatch in viscosity between
the hot and cold sides became significant. Due to a 13 minute lag in the onset of gelation on the
cold side, the maximum viscosity difference was almost five orders of magnitude. This indicates
that the sample did not cure uniformly, but rather experienced a “gelation boundary”, which began
on the hot side of the part and moved through the thickness toward the colder side.
Figure 2.7c shows the mold cavity pressure (black), the pressure in the injector air supply
line (green), and the “critical pressure” above which volatile release is suppressed (dashed line).
The moments at which vacuum and positive pressure (300 kPa) were applied to the cavity and
injector, respectively, are visible at 45-50 minutes. They correspond to the injection phase. Once
injection was complete, the supply pressure was increased to 450 kPa during the temperature ramp
and dwell at 185°C. The cavity pressure after injection was initially below the supply level due to
frictional losses within the injector, but slowly increased during the temperature ramp due to
thermal expansion of the resin [47]. Once the final high-temperature dwell was reached, the cavity
45
pressure began to decrease, and rapidly dropped below the critical pressure, as indicated by the
dotted vertical line.
Figure 2.7: Temperature, viscosity, and pressure for formulation F1, Cycle A.
This pressure drop was caused by cure shrinkage – a decrease in specific volume due to
polymer cross-linking – and is an unavoidable consequence of curing resin in a rigid, fixed-volume
46
molding tool [47]. Because pressure is proportional to volumetric changes (by the bulk modulus),
even a small amount of cure shrinkage can reverse the stress state within the curing resin from a
state of hydrostatic compression to tension (assuming sufficient adhesion between the resin and
mold surfaces). Initially after injection, cavity pressure was provided hydrostatically through the
inlet. Once the resin near the inlet gelled, the pressure in the rest of the mold cavity became isolated
from the externally applied pressure, and instead became entirely governed by volumetric changes.
The cumulative cure shrinkage of the resin, once the inlet had gelled, determined the timing of the
pressure drop.
In the case of Figure 2.7, surface porosity occurred because of the relative timing of the
pressure drop and the increasing viscosity at the coldest location of the mold cavity. The pressure
drop occurred when the cold-side viscosity was in the vicinity of 100 Pa·s, which is past the gel
point, but far below the critical value of 710
4
Pa·s that has been shown to suppress volatile
release. The hot-side, meanwhile, had already reached a high viscosity level, and therefore did not
off-gas.
Once the mold cavity pressure decreased below the critical value, voids began to form at
the cold surface. The period of void formation was clearly observable through the window in the
mold, and is indicated on Figure 2.7. Figure 2.8 shows a series of images recorded in situ for the
same sample, beginning with an initially defect-free surface just before the pressure drop. Almost
immediately after the pressure fell below the critical value, voids began to form and continued to
grow for 15 minutes. All void growth finally ceased as the cold-side viscosity reached the “safe
zone”, where the resin was nearly vitrified. In cases of extreme porosity such as in Figure 2.8,
much of the surface separated from the tool face, clearly manifest through the window because of
the change in refractive index (compare Figure 2.8c and d), and accompanied by a concurrent
47
spike/drop in measured cavity pressure (see Figure 2.7 at 136 minutes). The volatile gasses were
less dense than the liquid resin, so their release at the cold surface relieved the tensile stresses
generated by cure shrinkage. This phenomenon can be seen in Figure 2.7 by the gradual recovery
in pressure during the time of porosity formation. Note that, since the pressure drop occurred when
most of the resin was already approaching high viscosities (and due to the presence of the fibrous
preform), the total volume of volatiles released was much lower than for the ambient-pressure neat
resin samples (which began to grow bubbles earlier, during the ramp to the high-temperature
dwell).
Figure 2.8: Surface porosity formation recorded in situ for the composite sample with formulation F1, Cycle A.
Time labels correspond to the data in Figure 2.7. The thermocouple used to measure the window-side temperature
history is visible in the center of the frame.
48
In summary, the tendency of this resin to exhibit surface porosity in RTM processing
stemmed from a combination of two factors: significant potential for volatile release during cure,
and rapid, highly temperature-dependent gelation, which increased the material sensitivity to in-
mold temperature gradients. Additionally, the distribution of the porosity – concentrated on the
cold surface – was determined by the direction of the temperature gradient.
Figure 2.9: Schematic of the surface porosity formation mechanism. Liquid resin is initially under hydrostatic
pressure (top image). Then, a temperature gradient causes a “gelation boundary” to form and move from the hotter
side of the part to the colder side (middle image). Volatile release occurs at the not-yet-vitrified colder side, as the
cure shrinkage of the hotter side reverses the stress state from compressive to tensile (bottom image).
49
The effect of formulation and cure cycle
The following results illustrate the influence of cure kinetics on surface porosity, and reveal
a strategy that can be used to minimize this type of defect. Figure 2.10 shows a comparison of the
viscosity envelopes for the three samples cured using Cycle A, with the times corresponding to the
pressure drops indicated by vertical dotted lines. Figure 2.11 shows the same type of data for
samples with Cycle B.
Figure 2.10: Viscosity envelopes for laminates cured with Cycle A.
The viscosity data shown for F1 in Figure 2.10 (Cycle A) is the same as from Figure 2.7.
During the intermediate dwell of Cycle B at 130°C (Figure 2.11), formulation F1 remained inert
and the viscosity did not increase. When the temperature was increased to 185°C, the F1 gelation
profile was nearly identical to that of Cycle A, indicating that no significant cure progress occurred
50
during the intermediate dwell. The cold-side viscosity at the time of the pressure drop was low in
both cases, and accordingly, the porosity was most severe for these samples.
Figure 2.11: Viscosity envelopes for laminates cured with Cycle B.
The viscosity envelopes for formulations F2 and F3 shown in Figure 2.10 display an onset
of gelation at lower temperatures than for F1. Viscosity increased earlier with increasing catalyst
concentration, but more importantly, the widths of the envelopes were reduced. Consequently, at
the moment of the pressure drop, the cold-side viscosities were greater than for the sample with
minimal catalyst (F1). As a result, the Cycle A-F2 and -F3 samples displayed almost an order of
magnitude reduction in surface porosity compared to the baseline F1 formulation.
51
The influence of the catalyst was even more pronounced for Cycle B (Figure 2.11), in
which the F2 and F3 viscosities exhibited a progressive but limited increase during the
intermediate temperature dwell. This prevented the through-thickness viscosity mismatch from
becoming as large as for Cycle A (or as for F1 under both cure cycles), once the temperature was
increased towards 185°C. As expected, the cold-side viscosities during the pressure drops for F2
and F3 were greater in Cycle B than in Cycle A. Accordingly, these samples exhibited the lowest
levels of surface porosity.
While the cold-side viscosity directly affected the release of resin volatiles at that surface,
this parameter alone was not sufficient to fully explain the resulting surface porosity. For example,
formulation F1 had a greater cold-side viscosity for Cycle B than for Cycle A at the time of the
pressure drop, but the total defective surface area was greater for Cycle B. The explanation for this
apparent discrepancy lies in the magnitude of the viscosity difference between the hot and cold
sides. After injection, uncured resin in the mold is initially under hydrostatic pressure. As the
curing reaction progresses, cure shrinkage (generally assumed to progress linearly with degree-of-
cure [46]) causes a decrease in specific volume and, in extremis, reverses the stress state from
compressive to tensile. If a temperature gradient exists, then the shrinkage contribution of the
hotter side to the total shrinkage is greater than that of the colder side, which has not progressed
as far. If the imbalance in the progression of cure is sufficiently large, then the critical amount of
total shrinkage to cause a pressure drop occurs when the cold-side viscosity remains low enough
to allow volatile release. The greater the difference in viscosity between the hot and cold sides, the
earlier the pressure drop occurs relative to the evolution of the cold-side viscosity, allowing more
time for volatile release to occur before the critical viscosity is attained.
52
The relationship between the magnitude of the viscosity difference at the moment of the
pressure drop and the resulting surface porosity roughly followed an exponential trend, as shown
in Figure 2.12. Formulation F1 exhibited the largest viscosity difference and extensive surface
porosity for both cure cycles. In comparison, F2 and F3 showed a significant reduction in defective
surface area with Cycle A, and an even greater reduction with Cycle B.
Figure 2.12: Comparison of the cold-side surface porosity of molded samples with the through-thickness viscosity
mismatch at the time of the pressure drop.
The combination of increased catalyst loading and an intermediate temperature dwell in
the cure cycle effectively prevented surface porosity, by allowing the resin to reach higher
viscosities before the inevitable pressure drop. Cure shrinkage progressed only enough to cause
pressure drops near the end of conversion, and the extent of conversion possible at 130°C was
limited (exemplified by the decreasing slope of the viscosity envelope for F3 in Figure 2.12). The
mold could therefore be held at 130°C with little risk of pressure loss, reducing the amount by
53
which the cold-side viscosity lagged when the RTM tool was ramped to the final dwell
temperature.
An additional effect of inducing gelation at a lower temperature is revealed by comparing
the pressure histories of Figure 2.7 and Figure 2.13. In the former case, when the temperature was
ramped to 185°C, the pressure increased only slightly (about 3 kPa/min). As the resin in the cavity
underwent thermal expansion, it could flow out through the constant-pressure inlet port to maintain
equilibrium with the supply pressure. However, in Figure 2.13, during the final temperature ramp,
the resin had already gelled and thus could not flow. Consequently, thermal expansion caused a
sharper pressure increase (about 16 kPa/min), an effect that can have multiple consequences.
Firstly, the magnitude of the pressure spike is not easily predicted. This can cause dangerous stress
levels in the molding tool, so care must be exercised not to exceed the maximum pressure allowed
by the design of the tool. This is especially important when using brittle tool plates (i.e. glass),
which can fracture catastrophically rather than simply yielding. Despite these concerns, thermal
expansion is beneficial in that it counteracts the volumetric shrinkage induced by cure, making it
possible to maintain cavity pressure after gelation (when externally applied pressure becomes
ineffective). The thermally-induced pressure increase delays the cure-shrinkage-induced pressure
drop, affording additional time for the resin to progress through the rubbery phase and near
vitrification, where the potential for volatile release diminishes.
54
Figure 2.13: Temperature, viscosity, and pressure for formulation F2, Cycle B.
2.5. Conclusions
This work was part of a comprehensive effort to understand the relationships between resin
properties, in-mold phenomena, and part quality for a next-generation blended resin system. In
55
this study, we sought to use a practical, experimental approach to reducing defect formation in this
complex resin. A versatile tool was developed, allowing in situ process diagnostics to provide
insights into what is generally a “black-box” process. By combining in situ data with volatile
release criteria developed through thermal analysis, we described the causes and mechanisms
underlying volatile-induced surface porosity. In short, volatile-induced surface porosity can occur
when curing resins with high volatility in a rigid mold cavity in which thermal gradients exist,
even if the temperature differences are relatively small. After the inlet gels, external hydrostatic
pressure cannot be used to maintain cavity pressure, and cure shrinkage inevitably causes a
pressure drop. Volatile release can then occur in colder zones within the cavity, where the resin
has not yet vitrified.
The results demonstrated how several parameters strongly influenced surface porosity,
although there is room for further process refinement. The modified cure Cycle B improved on the
baseline case of ramping directly to the final high-temperature dwell, but the cycle was by no
means optimized. A process that minimizes both surface porosity and total cycle time would be
desirable from a manufacturing perspective. The ultimate goal is to find a robust solution, one that
is insensitive to small deviations from the prescribed processing parameters, so that even in a larger
RTM tool, which may have larger thermal gradients, surface porosity can still be prevented.
Note that in the molding tool used for this study, the thermal gradients were primarily
through-thickness, acting as a simple 1-D “model case” for the phenomenon of surface porosity.
In general, thermal gradients potentially include in-plane as well as through-thickness components.
Although not described in detail here, preliminary tests using the same resins in a second, larger
RTM (which contained substantial in-plane temperature gradients) showed that the distribution of
56
surface porosity is consistently correlated with the coldest zones in the mold cavity (see Figure
2.14).
Figure 2.14: In-plane temperatures of an RTM with a 305 × 457 mm (12” × 18") cavity (left), and surface porosity
concentrated in the cooler edge regions of a molded sample (right).
The effect of in-plane temperature gradients on pressure loss behavior, however, requires
a more sophisticated description. For the one-dimensional case considered in the present study, we
assume that the pressure detected by the sensor is transmitted through the thickness, i.e., that the
pressure measured on the hot side is equal to that on the cooler side. For large RTM parts with in-
plane temperature gradients, in-plane stresses can vary with location, so the “critical pressure”
criterion for void formation would be useful only if applied locally. As a further complication,
while the specific volume of the resin is an intrinsic property, the total amount of shrinkage
resulting in a pressure drop will vary between tools (depending part thickness, mold rigidity, etc.),
potentially making the timing difficult to predict.
Despite the simplifications used in this study, lessons can be learned from the data to guide
the processing for other resins and more complex geometries. First, the lag in the viscosity increase
of the coldest region is directly related to porosity formation, so care should be exercised to
57
minimize or avoid cold zones in all cases. Secondly, a two-dwell cure cycle has been shown to be
effective at separating the cure reaction into multiple stages - first gelling the part while providing
pressure externally, then maintaining pressure while in the rubbery phase (between the gel point
and vitrification) by increasing the temperature, and lastly, reaching the final degree of cure after
the pressure has dropped but volatile release is no longer possible. Because the improvement in
surface quality comes at the expense of additional cycle time, increased catalyst concentrations
can be employed to accelerate the first stage of cure. The limiting factor to this approach is the
reduction in pot life, which may become dangerously short if the resin is highly catalyzed.
A generalized model to predict volatile-induced surface porosity for arbitrarily shaped parts
would allow defect prediction and cure cycle optimization for parts manufactured under spatially-
varying pressure and temperature fields. Such a model would require a description of heat transfer,
thermochemical phenomena (degree of cure, viscosity), volumetric changes (thermal and
chemical), and mechanical properties (at least bulk modulus, to predict pressure behavior based
on volumetric changes). In this work, we experimentally determined and demonstrated the “critical
pressure” and “critical viscosity” criteria, which could be applied to such a model to predict the
timing, location, and severity of volatile-induced surface porosity.
58
CHAPTER 3. Cure cycle design for eliminating surface porosity during RTM
3.1. Introduction
Resin transfer molding (RTM) is a popular composite manufacturing technique for
producing small to medium sized parts with low microstructural defect levels, excellent surface
finishes, and potentially complex geometries [24]. Voids are the most common type of defect
encountered in RTM, and are often a result of incomplete preform saturation during injection. Air
can become trapped within the mold cavity due to improper injection pressure [26] or gate/vent
placement [25], and considerable efforts have been made to develop models and protocols that
ensure successful injection for arbitrary mold geometries. However, voids can also arise from
another source, namely volatiles released by the resin after injection, during the curing phase of
the RTM process.
In this chapter, we investigated volatile-induced porosity in the context of a prototype resin
that was developed for RTM of primary aerospace structures. This blended formulation
(corresponding to “F3” from the previous chapter) contains both benzoxazine and epoxy
constituents, as well as a proprietary catalyst. The benzoxazine component imparts excellent
resistance to moisture uptake and chemically aggressive environments, as well as favorable
flammability, smoke, and toxicity (FST) properties. The resin’s epoxy component acts to toughen
the otherwise brittle benzoxazine matrix, and also increases the Tg by raising the maximum
theoretical cross-link density [52]. While this resin has many potential benefits compared to
standard aerospace epoxies, it also presents some additional challenges. One concern is an
increased resin volatility, which must be suppressed by the positive hydrostatic pressure applied
59
during RTM processing. The previous chapter (published as [59]) showed that the minimum
“critical pressure” required to suppress volatile release for this resin is ~200 kPa (absolute).
While the minimum required pressure is easily within the limits of most standard RTM
systems, the challenge lies in maintaining mold cavity pressure during chemical cure shrinkage
caused by polymerization. Despite the total cure shrinkage of this resin being relatively low
(around 1% [60]) compared to other thermoset resins, the density increase is sufficient to create a
loss of cavity pressure during cure.
The previous chapter described how the combined effects of increased resin volatility and
shrinkage-induced pressure drops led to volatile-induced surface porosity. The severity of this type
of defect is determined by the timing of the inevitable pressure drop relative to the gelation of the
resin. In the presence of thermal gradients (which exist to varying extents in almost all molding
tools), the progression of cure in the coldest regions of the mold cavity lags behind that of the
hotter regions. When the thermal gradient is in the through-thickness direction (as is the case for
the molding tool used in this study), the cold side of the part is susceptible to surface porosity
formation because the viscosity at that location can still be low (facilitating void nucleation and
growth) when the cumulative shrinkage of the entire part becomes enough to cause a pressure
drop.
During RTM processing, if the mold is ramped directly to the high-temperature dwell
(185°C) after resin injection (at 110°C), even modest thermal gradients (5-10°C) cause the cure
gradient to become sufficiently large, such that the cold side is still at low viscosity when the
pressure drop occurs, resulting in volatile release at that surface. The previous study also showed
that, conversely, if a 3 hour intermediate dwell at 130°C is included in the cure cycle, surface
porosity is almost entirely avoided. The intermediate dwell acts to progress the cure reaction at a
60
reduced rate, limiting the amount by which the cold side lags when the pressure drop finally occurs.
By raising the viscosity above a threshold value, the tendency for volatile release greatly
diminishes and the formation of voids is inhibited.
This modified cure cycle, while effective at suppressing surface porosity, has the drawback
of an almost tripled cycle time compared to the baseline cure cycle. From a manufacturing
standpoint, this translates to reduced output and increased costs. Therefore, a method to minimize
both surface porosity and cycle time is desired. The goals of this study are, first, to describe the
criteria that suppress the formation of surface porosity, and second, to develop a process map,
which enables the design of cure cycles that prevent surface porosity using the minimum cycle
time, given the magnitude of thermal gradients present in the tool.
The first step of our approach was to model the kinetics of the resin curing reaction to
enable prediction of the reaction rate for arbitrary temperature cycles. A purely empirical “model-
free” isoconversional method was applied, using differential scanning calorimetry (DSC) scans of
the cure exotherm. Next, the cure kinetics model was applied to rheological dynamic analysis
(RDA) and thermogravimetric analysis (TGA) data, to determine the degree of cure values
corresponding to gelation and to the cessation of volatile release. Then, molded samples were
fabricated with extended dwells at a range of intermediate temperatures, to determine the
maximum dwell time before a pressure drop occurs. A lab-scale RTM tool was used, which
included temperature and pressure sensors, as well as a glass tool-plate on one side that enabled
direct visual observation of surface porosity formation and other in-mold phenomena. Finally, the
results were combined into a process map, an “optimized” cycle was designed, and a molded
sample was fabricated with the new cure cycle to validate the utility of the process map. The results
61
show that this method can be used to develop faster cure cycles, with negligible reduction in
surface quality.
3.2. Experimental methods
3.2.1. Thermal characterization
The heat of reaction was measured via DSC (Q2000, TA Instruments) using 5 – 10 mg
samples of resin in hermetically sealed aluminum pans. Temperature cycles consisted of linear
ramps from 35 to 315°C at rates of 5, 7, and 10°C/min. Rates above those normally seen in RTM
processing were used for the purpose of maximizing signal quality. TGA (Q5000IR, TA
Instruments) tests were performed at ambient pressure on samples weighing 40 ± 0.5 mg to
determine the mass loss due to volatilization. Temperature cycles consisted of linear ramps from
35 to 315°C at rates of 1, 2, and 3°C/min. The viscosity was measured using a rheometer
(AR2000ex, TA Instruments) with disposable aluminum parallel plate fixtures and the same
temperature cycles as the TGA tests. All data processing and cure modeling was performed in
MATLAB (R2014b, The Mathworks, Inc.).
3.2.2. RTM manufacturing trials
Molded composite samples were fabricated under a variety of cure temperature cycles. All
cure cycles started at 110°C, used 2°C/min ramps, and ended with a 60 minute dwell at 185°C.
The cycles differed in the time and temperature of an intermediate dwell, summarized in Table
3.1. Cycle A is considered the baseline case, with no intermediate dwell, and is included to
demonstrate the need for a 2-stage cycle and to show how surface porosity forms when rapid
gelation is induced. Cycle B corresponds to the previous chapter, and represents a cure cycle that
has been shown to minimize surface porosity. Cycles C through G, with “extended” dwells, denote
62
intermediate dwells that were held until the pressure drop occurred, and were used to determine
the maximum possible hold time at each intermediate temperature as described in Section 3.6.
Finally, Cycle H was designed using the methods described in this chapter, and represents an
attempt at reducing the cycle time compared to Cycle B while achieving similarly low surface
porosity levels.
Table 3.1: List of tested cure cycles.
Cure cycle Intermediate dwell temperature Intermediate dwell time
A None None
B 130°C 180 minutes
C 130°C Extended
D 140°C Extended
E 150°C Extended
F 160°C Extended
G 170°C Extended
H 150°C 45 minutes
Figure 3.1: Front view and exploded CAD render of the lab-scale RTM tool.
The lab-scale RTM used for fabricating molded samples is shown in Figure 3.1. The main
tool body is an aluminum block containing heating elements, inlet/outlet ports, thermocouples, and
a pressure sensor. A “picture frame” style spacer plate determines the thickness (3.2 mm) and in-
63
plane dimensions (76 127 mm) of the molded part, and a 20 mm thick tempered glass plate acts
as the second tool-face.
All molded samples were reinforced with 8 layers of five-harness satin (5HS) carbon fiber
fabric (364 g/m
2
areal weight, 3000 fiber/tow count, Sigmatex Ltd.) stacked in a quasi-isotropic
layup. Resin was vacuum-degassed for 45 minutes at 80°C before being loaded into a pneumatic
injector (Radius 2100cc). Injections were performed at 110°C and 300 kPa, with vacuum being
applied to the mold cavity both before and during injection. Excess resin was flushed through the
outlet port until no bubbles remained in the mold cavity, as observed through the glass tool plate.
Injection was completed by closing the outlet valve, leaving the inlet valve open, and applying a
post-fill hydrostatic pressure of 450 kPa using the regulator on the injector’s compressed air
supply. Finally, a portable low-magnification microscope was aimed at the window and used to
record time-lapse videos, allowing the exact timing of porosity formation to be determined.
After de-molding, surface porosity was quantified by a measure of percent defective area.
High-resolution photographs were converted to binary maps of void distribution using an open-
source image processing software (ImageJ v1.48). Due to the configuration of the molding tool
(with heating elements on only one side of the sample), the window-side was always cooler than
the tool-side, and consequently, surface porosity appeared exclusively on the window-side of
molded samples (see Figure 3.2 and Figure 3.3). Thus, only the window-side was considered when
evaluating the surface quality. Additionally, to directly measure the magnitude of the through-
thickness temperature gradient, thermocouple wires were embedded into each side of the samples
cured under cycles A, E, and H, by running the wires in through the resin outlet port and placing
the tips between the preform and the tool faces prior to sealing the mold.
64
Figure 3.2: Section view of lab-scale RTM tool (left) and a thermal simulation showing the through-thickness
temperature gradient (right).
Figure 3.3: A molded sample with porosity exclusively on the colder side.
65
3.3. Cure kinetics modeling
The heat of the exothermic cure reaction was measured by DSC. The degree of cure (α),
which ranges from 0 for resin in the uncured state to 1 for fully cured resin, was computed using
a normalized running integral of the heat release. H(t) represents the heat flow over time, and the
average measured total heat HT = 473±4 J/g.
𝛼 ( 𝑡 ) =
1
𝐻 𝑇 ∫ 𝐻 ( 𝑡 ) 𝑑𝑡 𝑡 0
(3-1)
The instantaneous cure rate, therefore, is equal to the instantaneous heat flow normalized by the
total heat of reaction.
𝑑𝛼 𝑑𝑡 ( 𝑡 ) =
1
𝐻 𝑇 𝐻 ( 𝑡 ) (3-2)
Cure kinetics of thermoset resins are generally modeled using a rate equation that contains
an Arrhenius temperature dependence multiplied by a function f(α), which describes the behavior
of the particular type of reaction under consideration [61].
𝑑𝛼 𝑑𝑡 = 𝐴 exp [
−𝐸 𝑎 𝑅𝑇 ( 𝑡 )
] 𝑓 ( 𝛼 ) (3-3)
Common choices for f(α) include n
th
-order (Eq. (3-4)) and autocatalytic (Eq. (3-5))
expressions, but many variations have been used to capture more complex phenomena, for
example the inclusion of a diffusion factor to account for the reduced reaction rate when in the
diffusion-limited regime.
𝑓 ( 𝛼 ) = ( 1 − 𝛼 )
𝑛 (3-4)
𝑓 ( 𝛼 ) = 𝛼 𝑚 ( 1 − 𝛼 )
𝑛 (3-5)
66
However, for resins that undergo multiple reactions, a single-term expression as shown in Eq. (3-3)
may not suffice to fully describe the cure kinetics behavior.
The blended resin in this study exhibits a complex set of overlapping and interdependent
reactions, whose exothermic profile cannot be adequately described using Eq. (3-3). Modeling the
cure kinetics of a variant of this resin has been previously attempted using a multi-term approach
[60], but the model was unable to accurately capture dynamic behavior at varying ramp rates.
To circumvent the difficulties encountered when applying the traditional kinetics modeling
approach to a complex resin, an alternative method was employed, which is purely
phenomenological and does not require any specific knowledge of the chemistry. As first described
by Friedman in 1964 [62], this “model-free isoconversional method” assumes only that, at every
degree of cure, the reaction rate obeys some Arrhenius-type temperature dependence. As long as
this one requirement holds true, the method should theoretically be applicable to resins with any
form of α-dependency (which can be unknown).
Starting from Eq. (3-3), assume that Ea and A are no longer constant. Let them be functions
of α, and re-write the expression using a modified pre-exponential factor A’ that is the product of
A(α) and f(α).
𝑑𝛼 𝑑𝑡 = 𝐴 ( 𝛼 ) exp [
−𝐸 𝑎 ( 𝛼 )
𝑅𝑇 ( 𝑡 )
] 𝑓 ( 𝛼 ) (3-6)
𝐴 ′
( 𝛼 ) = 𝐴 ( 𝛼 ) 𝑓 ( 𝛼 ) (3-7)
𝑑𝛼 𝑑𝑡 = 𝐴 ′( 𝛼 )exp [
−𝐸 𝑎 ( 𝛼 )
𝑅𝑇 ( 𝑡 )
] (3-8)
67
Taking the natural logarithm of Eq. (3-8) gives Eq. (3-9). If we consider 1/T to be an
independent variable x, and ln(dα/dt) to be a dependent variable y, then Eq. (3-9) is of the linear
“slope-intercept” form.
ln (
𝑑𝛼 𝑑𝑡 ) = ln( 𝐴 ′( 𝛼 ) )−
𝐸 𝑎 ( 𝛼 )
𝑅𝑇 ( 𝑡 )
↔ "𝑦 = 𝑚𝑥 + 𝑏 " form (3-9)
slope =
−𝐸 𝑎 ( 𝛼 )
𝑅
𝑦 − intercept = ln( 𝐴 ′( 𝛼 ) )
The task of modeling the cure kinetics now consists of finding the apparent Arrhenius
parameters – modified pre-exponential factor A’(α) and activation energy Ea(α) – for every value
of α. For the resin in this study, first the degree of cure and cure rate were computed by Eq. (3-1)
and Eq. (3-2) from DSC scans, using linear temperature ramps at 5, 7, and 10°C/minute. Figure
3.4 shows the results in the form of Eq. (3-9). For 1000 evenly spaced α values ranging from 0 to
1, the corresponding points on each DSC trace were selected and a linear fit was applied. The
activation energy was extracted from the slope of the resulting line, and the modified pre-
exponential factor was extracted from the y-intercept.
Figure 3.5 shows the resulting model parameters, as well as the R
2
values of the linear fits
used to obtain the model parameters for each α. The fits were generally very close, with the
exception of some error in the α > 0.95 range. The measured end of cure is sensitive to the
particular choice of baseline used, hence the increased error in that regime, but fortunately the
exact end of cure is unimportant for our purposes.
68
Figure 3.4: Isoconversional chart from dynamic DSC scans. Linear fits at some example α values are shown in
black.
Figure 3.5: Experimental model parameters E a and A' over the full range of α. Inset shows the R
2
values of the linear
fits used to obtain the model parameters at each iso-α value.
69
To apply the model, simply prescribe an arbitrary time/temperature cycle and use the rate
equation (3-8) with a numerical integration technique (e.g., forward Euler method) to stepwise
compute the evolution of the degree of cure. For each time step, use the apparent Arrhenius
parameters Ea(α) and A’(α) from Figure 3.5 corresponding to the current degree of cure. Figure
3.6 shows a comparison of the measured DSC data from Figure 3.4 and the model predictions for
the same temperature cycles, indicating near-perfect agreement. Model/experiment comparisons
for other temperature cycles suggest satisfactory agreement, but uncertainties arise in the measured
data due to “lost heat” during temperature jumps to isothermal conditions and due to signal-to-
noise issues during low-temperature isotherms or low-rate temperature ramps.
Figure 3.6: Comparison of measured and predicted rates of cure for the DSC scans from Figure 3.4
3.4. Resin volatile release behavior
Figure 3.7 presents a comparison of data obtained by TGA, RDA, and the cure kinetics
model for linear temperature ramp cure cycles at 1, 2, and 3°C/minute. The top graph shows the
rate of mass loss, which initially displays a roughly exponential temperature dependence but
eventually drops to zero as the resin solidifies. This mass loss is due to the evaporation of residual
solvent present in the resin, and leads to void formation in closed-mold processing conditions when
insufficient pressure is applied. Dots on each mass loss curve indicate the moments when
70
evaporation had effectively stopped (defined using a threshold rate of 10
-6
% mass per °C), and
black vertical arrows point to the corresponding times on the graphs below.
Figure 3.7: Mass loss rate, complex viscosity, and degree of cure, as obtained by TGA, RDA, and the cure kinetics
model (respectively). Black arrows show the times when mass loss ceased, and red horizontal bands show the
corresponding viscosity and degree of cure values. The onset of gelation and corresponding degree of cure are
shown in green.
71
The absolute magnitude of the complex viscosity, |η*|, is shown on the middle graph. The
viscosity values corresponding to the end of mass loss are found to reside in the highest order-of-
magnitude and within a narrow range, suggesting that there exists a threshold mechanical state,
beyond which mass loss is no longer possible due to the dissolved molecules becoming “locked
in” to the growing polymer network. The threshold value of 1.7-2.0 10
4
Pa·s is well above the
gel-point, meaning that the complex viscosity at that state is dominated by the increasing elastic
modulus. Markers on each curve indicate the initial onset of gelation, defined by the moment |η*|
> 1 Pa·s, and green arrows indicate the corresponding times on the bottom graph.
The degree of cure, as predicted by the kinetics model, shows that the two phenomena of
interest both occur within narrow ranges of α. The green arrows corresponding to the onset of
gelation fall almost exactly at α = 0.15, which will be referred to as αgel in subsequent discussion.
Similarly, the black arrows – corresponding to the end of weight loss due to resin solidification –
fall in the range α = 0.30-0.34. In subsequent discussion, a value of 0.35 is used as a conservative
estimate for the critical degree of cure (designated αcrit) above which volatilization is expected to
be minimal.
3.5. Volatile-induced surface porosity
By applying the criteria from the previous section to data collected from molded samples,
we demonstrate the need for a two-stage cure cycle. Cycle A (Table 3.1) represents the baseline
case, without an intermediate temperature dwell, which results in high porosity levels. Conversely,
Cycle B results in minimal surface porosity.
Figure 3.8 shows the measured temperature and pressure data for the sample cured using
the baseline Cycle A, and the degree of cure (α) envelope as predicted by the kinetics model. The
72
binary map used to determine the fractional defective surface area is shown, along with a
micrograph of a representative surface void. The pressure graph shows the injector’s supply
pressure, the mold cavity pressure, and the “critical pressure” required to suppress volatilization.
The α envelope was computed using the temperature histories of the thermocouples embedded in
the part, with red indicating the hotter tool-side and blue indicating the cooler window-side.
Figure 3.8: Temperature, pressure, and model degree of cure data for the baseline cure Cycle A (ramping directly to
the high-temperature dwell). A binary map of the surface defects is shown, along with a micrograph of a surface
void.
The cavity pressure, despite the application of sufficient supply pressure, dropped below
the critical value at t = 95 minutes. From t = 96 – 99 minutes, surface porosity formation was
observed. The degree of cure envelope shows that the pressure drop occurred when the cold-side
α was below αcrit, thus the resin at that surface was still able to release volatiles. The formation of
73
surface porosity ceased when the cold-side α > αcrit. In addition to using in situ observations,
porosity formation can, at least qualitatively, be identified by the post-drop cavity pressure
behavior. Decreasing pressure corresponds to increasing cure shrinkage, and jaggedness in the P
< Pcrit regime can be attributed to void growth. Because the volatile gases are less dense than the
resin, their release counteracts the density increase due to cure shrinkage. The cavity pressure in
Figure 3.8 shows this behavior between t = 96 – 99 minutes, followed by a continued drop in
pressure that eventually reaches negative values, indicating a tensile stress state in the mold cavity.
The pressure jump back up to ~200 kPa at t = 128 minutes is caused by the laminate ultimately
detaching from the sensor tip.
In contrast to the baseline case, cure Cycle B has been demonstrated (in the previous
chapter) to result in minimal porosity (0.17% defective area), by using an intermediate temperature
dwell to slowly progress the cure reaction and reduce the amount by which the cold-side α lags
when the pressure drop occurs. Cycle B, however, requires an additional 3 hours of processing
time. The following section describes the procedure used to build a process map, which shows the
range of intermediate dwell times and temperatures that produce behavior similar to Cycle B. The
characteristics of a two-stage cycle that prevent surface porosity formation are described in more
detail in the context of the “optimized” Cycle H in Section 3.7.
3.6. Process map
The purpose of the process map (Figure 3.9) is to enable the design of cure cycles that
avoid surface porosity while minimizing cycle time, based on quantitative criteria for volatile
release and pressure drops. First, the baseline Cycle A and the “extended dwell” cure Cycles C
though G are shown as dashed black lines. The times corresponding to each cycle when α = αcrit
are shown in blue. If cavity pressure can be maintained until the entire cure envelope has surpassed
74
αcrit, minimal surface porosity can be expected. To find the maximum time that each intermediate
temperature can be held, the time of pressure drop was measured experimentally for samples cured
under each of the “extended dwell” cycles (C – G), given by the red line on Figure 3.9.
Figure 3.9: Process map showing times of critical degree of cure, pressure drop, and gelation, for a range of
intermediate dwell temperatures. The shaded green area is the process window for intermediate dwells resulting in
minimal surface porosity, and the shaded red area shows an approximate temperature envelope for Cycle H.
The location of the red points indicating the timing of pressure drops – to the left of the
corresponding αcrit points – presents an apparent discrepancy. For all intermediate dwell
temperatures, the pressure drop occurred before the critical degree of cure was reached.
Accordingly, the samples for Cycles C – G all exhibited 2-5% surface porosity. However, for
Cycle B, which started the final temperature ramp 8 minutes before the pressure drop would have
occurred at 130°C (as in Cycle C), the pressure drop did not occur until 27 minutes into the ramp
toward 185°C, and the surface only developed porosity over 0.17% of the area.
75
The explanation for this delayed pressure drop lies in the thermal expansion of the resin
during temperature ramps. In the case of Figure 3.8 (Cycle A), the cavity pressure increased during
the temperature ramp, but only slightly. Because the resin is initially liquid and connected to a
constant-pressure supply, as it undergoes thermal expansion in the mold cavity, the resin has the
ability to back-flow through the inlet port to maintain equilibrium hydrostatic pressure. However,
once the resin has gelled, it loses the ability to flow. In the case of Cycle B, the resin gelled during
the intermediate dwell at 130°C and the cavity pressure began to decrease below the supply
pressure due to cure shrinkage. However, when the temperature was ramped upwards, thermal
expansion temporarily caused the cavity pressure to increase, delaying the pressure drop long
enough for the cold-side α to surpass αcrit.
The phenomenon of thermal expansion is critical to maintaining mold pressure, because
after the resin has gelled, the supply pressure loses all ability to influence cavity pressure (as clearly
evidenced by the pressure drops that occur in every test). To exploit the effect of thermal expansion
on cavity pressure, we must first gel the resin at an intermediate temperature, and only then ramp
to the final cure temperature. We can therefore define the minimum times for intermediate dwells
by αgel, which is indicated by the green points on Figure 3.9.
The green shaded region – between the gel times and the pressure drop times – represents
the allowable “process window” for intermediate temperature dwells. For any of the temperatures
in the range shown, a dwell shorter than the gel time will be ineffective (resulting in behavior
similar to that of Cycle A), and a dwell that’s too long will also result in a pressure drop before α
> αcrit. Dwell times between the two limits, however, will have delayed pressure drops, which gives
the resin additional time to reach or exceed αcrit.
76
3.7. Cure cycle design
To use the process map for cure cycle design, consider the temperature envelope bounded
by the hottest and coldest regions in the mold cavity. The intermediate dwell must be held at least
until the cold side reaches αgel, but the hotter side must be used to conservatively predict the
maximum dwell time, since pressure drops are determined by the cumulative contribution of cure
shrinkage through the entire laminate. Because the process window gets narrower at higher
temperatures, RTM tools with wider temperature envelopes will require longer, lower-temperature
holds to meet this requirement. The red shaded region on Figure 3.9 shows the “optimized” cure
Cycle H (from Table 3.1), which uses a 45 minute dwell at 150°C. The width of the temperature
envelope is approximated using a cold-side temperature lag of 3% (relative to room temperature)
and a time lag of 4 minutes. The 45 minute dwell ends before the pressure drop would occur, and
the cold side is well past αgel. Figure 3.10 shows the results of the RTM sample fabricated using
Cycle H.
At the time of the pressure drop, the entire degree of cure envelope for Cycle H was above
αcrit, and consequently, the sample exhibited 95% less surface porosity than the case of Cycle A
(Figure 3.8). The few surface voids that did appear in Cycle H were in the form of barely visible,
shallow dimples, as opposed to the large sub-surface bubbles that formed under Cycle A.
The beneficial effect of the post-gelation temperature ramp can be understood by
considering the following simple approximation for relative density [63].
𝜌 𝜌 0
( 𝛼 , 𝑇 ) = 1 + 𝑐 1
( 𝑇 − 𝑇 0
)+ 𝑐 2
( 𝛼 − 𝛼 0
)
(3-10)
77
Figure 3.10: Temperature, pressure, and model α for Cycle H (45 minute intermediate dwell at 150°C).
Eq. (3-10) assumes linear changes in relative density due to thermal expansion and cure
shrinkage (which is assumed to progress linearly with α). Using T0 = 25°C and α0 = 0 for the
reference state ρ0, a total shrinkage value of 1% (from [60]), and a thermal expansion coefficient
for a similar resin from [64], Figure 3.11 was generated for the cure cycles with intermediate
temperature dwells at 150°C (Cycles A, E, and H).
Because resin can no longer flow after the gel point, it is the density increase beyond that
point which causes the pressure drop. The resin gels at ~170°C for Cycle A, and the pressure drop
occurs soon after, when the cold-side α is still < αcrit. For an extended dwell at 150°C, gelation
occurs during the dwell, and then continued cure shrinkage causes a pressure drop. For Cycle H,
however, gelation is induced at 150°C, but then the temperature ramp causes the cure envelope to
78
move parallel to the iso-ρ contours. This acts to increase α without increasing density, pushing the
entire α envelope above αcrit before the pressure drop occurs. The “hump” in cavity pressure during
the final temperature ramp in Figure 3.10 can be directly attributed to the temporarily decreasing
density seen during the corresponding ramp in Figure 3.11.
Figure 3.11: Degree of cure vs. temperature envelopes for cure cycles with 0, 45, and 120 minute intermediate
dwells at 150°C before the high-temperature dwell. Diagonal contour lines indicate relative density, which varies
with both temperature and degree of cure as described by Eq. (3-10).
79
3.8. Conclusions
The value of the process map is to enable improvements in cure cycle design by describing
the phenomena relevant to surface porosity formation over a wide range of temperatures. A cure
cycle was designed that reduced the required intermediate dwell time by a factor of four (45 min.
for Cycle H vs. 180 min. for Cycle B), and was demonstrated to be almost as effective at preventing
surface porosity (0.19% defective area for cycle H vs. 0.17% for cycle B).
By considering the entire temperature envelope, use of the process map can take into
account the limitations of any particular RTM tool’s heating scheme. For larger temperature
gradients, simply dwell at lower temperatures for longer times, to ensure that the entire part falls
within the required process window.
There are, however, some limitations to the approach described here. First, only a through-
thickness temperature gradient is considered. This one-dimensional simplification ignores in-plane
gradients, and thus complex part geometries and temperature distributions may require a more
sophisticated description. Furthermore, the time of pressure drop has only been experimentally
determined for the particular molding tool used in this study. Since the timing of this event may
vary between tools (due to geometry, rigidity, part thickness, mold pressure, etc.), a more
comprehensive description of density changes and related pressure behavior is desired. Finally, we
do not consider varying ramp rates, which could also influence the pressure behavior. The 2°C/min
ramp used coincidentally causes thermal expansion to occur at roughly the same rate as cure
shrinkage (see the slope of the “optimal dwell” envelope in Figure 3.11 compared to the iso-ρ
contours), but the ramp rate could certainly be changed to influence the timing of the pressure
drop.
80
Overall, the methods used in this study have potential relevance to a wide range of
composite molding applications. First, the “model-free” kinetics characterization technique is
advantageous because it does not require any knowledge of the reaction mechanism. It can, in
theory, be used for any resin that obeys an Arrhenius temperature dependence, including complex
formulations that may undergo multiple overlapping reactions. Second, the use of an RTM with a
transparent tool-plate can provide insight into in-mold phenomena, which are generally not
observable in situ due to the “black box” nature of standard metal RTM tools. Finally, by
combining a cure kinetics model with defect-formation criteria developed through thermal
characterization (TGA, RDA) and from in situ observations, porosity formation can be predicted
and guidelines can be developed for cure cycle modifications that reduce defect levels.
81
Interlude
One important type of material characterization that’s conspicuously absent from the
previous chapters is measurement of the resin volumetric changes, which govern the mold cavity
pressure after gelation of the inlet gate. We know the resin density at room temperature in the
uncured and fully cured states from pycnometry [55], but to predict cavity pressure throughout
processing, a full description of the resin’s thermal and chemical volumetric changes would be
needed, as well as a description of the resin bulk modulus.
Measurement of these material properties is not trivial. The resin must be measured in the
uncured state, the cured state, and throughout the transition between these states. The measurement
device must be capable of reaching all the relevant process temperatures, and must maintain
sufficient resin pressure to prevent volatile release. Many of the commonly-used density
measurement techniques fail to meet at least one of these requirements. No suitable device was
available in our lab at the time, so I resorted to building my own.
This tool was a piston-type dilatometer, inspired by the similar device from the work of
Boyard et al. [48] and Nawab et al. [65]. The tool was designed to replicate an RTM process and
thus included inlet and outlet ports for resin injection, as well as integrated heaters and a cavity
pressure transducer. As opposed to the lab-scale RTM, which had a fixed-size mold cavity and
variable mold pressures due to resin density changes, this hybrid dilatometer/RTM (HD/RTM)
featured a variable-thickness mold cavity and used load-control mode on a mechanical test frame
(Instron 5800 series) to maintain constant cavity pressure. A noncontact capacitive displacement
sensor was used to monitor changes in mold cavity thickness, from which resin volumetric changes
82
could be computed. The figure below shows an oblique view (left) and section view (right) of a
CAD model of the HD/RTM (made in SolidWorks 2015).
Figure Int-1: CAD model of the hybrid dilatometer/RTM.
The device was used to fabricate composite samples, and calibrated by measuring solid
and liquid materials with known properties. Details for the design of the system are provided in
Appendix B, and some preliminary results are in Appendix C. Due to factors beyond my control,
unfortunately, the work remains unpublished.
83
Despite the demonstrated efficacy of the process modifications that were prescribed in the
previous chapters for avoiding surface porosity, the “real world” solution to all the issues with this
resin turned out to be: “use a different resin.” The material supplier ultimately decided to focus on
other product lines, and ceased production of this type of resin.
The lack of access to material for additional experiments was the first roadblock for further
publications. Additionally, at this time, a new NASA-funded project began that demanded my full
attention. I began design of yet another experimental system, the “mini autoclave” that is featured
in the next two chapters. The HD/RTM remains available for use in future projects at the M. C.
Gill Composites Center.
84
CHAPTER 4. In situ process diagnostics for co-cure of sandwich structures
4.1. Introduction
Composite honeycomb sandwich structures consist of two prepreg facesheets (or “skins”),
bonded with a film adhesive to either side of a low-density, hexagonal-cell Nomex or aluminum
core [2]. The skins and film adhesive are often co-cured in a single process cycle under elevated
temperature and pressure within an autoclave, using a one-sided tool and enveloped in a vacuum
bag.
The co-cure process consists of the following steps. First, under ambient external
temperature and pressure, vacuum is pulled within the vacuum bag to consolidate the prepreg plies
and to remove air initially entrapped between them. After some time – which varies depending on
part size and other factors – autoclave pressure is applied, further compacting the sandwich
structure. Often, the vacuum bag is either brought to partial vacuum or fully vented to atmospheric
pressure. Next, the temperature is increased to reduce the prepreg resin and adhesive viscosities.
An isothermal dwell provides time for flow to occur, as surface tension causes the adhesive to
form fillets along the edges of the cell walls against the prepreg skin [66]. Finally, a second dwell
at a higher temperature causes the resin and adhesive to gel and vitrify, after which the autoclave
can be cooled and depressurized, and the finished part can be removed.
Manufacturing defects associated with any of the three components – skin, adhesive, and
core – can occur during processing [2]. The discontinuous honeycomb substrate can lead to
compaction variations and dimpling within the skins, as well as resin bleed into the open core cells,
causing skin porosity due to resin starvation. Additionally, the gas in the core cells can be driven
85
through the prepreg – which, if it occurs during resin gelation, results in skin porosity – and low
gas pressure in the core can lead to a reduction in the hydrostatic pressure of the prepreg resin,
potentially allowing bubble nucleation/growth from dissolved volatile species and further
contributing to porosity in the skins [67]. Adhesive fillets can also be disrupted by analogous
phenomena: low pressures can lead to bubble nucleation/growth from volatile release, gas
migrating through the adhesive can get locked in place during gelation, and excessive flow can
cause dripping/bleeding into undesired locations. Finally, defects relating to deformation of the
honeycomb core can arise if excess autoclave pressure is applied [68], although these issues are
beyond the scope of the present work.
The traditional means of quality assurance and process refinement typically rely on ex situ
inspection, i.e., parts are inspected outside of the processing environment after cure. While process
variables such as temperatures and pressures are measured during autoclave processing, the
autoclave and vacuum bag serve as barriers that prevent real-time monitoring of microstructural
changes (e.g., void formation). Figure 4.1 shows polished sections of co-cured samples to
emphasize the limitations of post-mortem analysis and motivate the need for in situ process
diagnostics. These samples consist of the same materials and were cured under the same
temperature cycle, but other process modifications (namely the initial adhesive format, the
breathability of the core, and the pressure cycle) caused significant variations in part quality (a
detailed description of the process parameters for these samples is given in section 4.3.2). The
images in Figure 4.1 provide information about the final state of the parts, but supply little insight
into the co-cure process. How, when, and why did these bond-lines evolve into their final states?
What phenomena are responsible for causing such variability in skin and bond-line quality?
86
Figure 4.1: Polished-section micrographs of co-cured samples exhibiting a wide range of morphologies.
To reliably fabricate defect-free sandwich structures, we must identify the relevant process
phenomena, characterize the time-dependent material behaviors, and develop appropriate
processes based on scientific understanding. To achieve these goals, we implemented an
experimental method that enabled in situ observations of bond-line evolution during autoclave co-
87
cure. The work was inspired by – and builds upon – previous research described in the following
section.
4.1.1. Literature review
While co-cure is inherently a complex process involving coupled phenomena, much of the
relevant material behavior has been investigated in simpler contexts. For example, Pearce et al.
[69] studied void formation in film adhesives during secondary bonding of honeycomb core to
glass and metal substrates. They found volatile release at sub-ambient core pressures to be the
primary cause of voids, and using FTIR spectroscopy, they identified the volatiles as water
(absorbed from environmental humidity) and residual solvent (from the filming process). Hayes
et al. [70] studied another simplified system: honeycomb sandwich structures without film
adhesive, using “self-adhesive” prepregs for which excess prepreg resin formed the skin/core
bond-line. They compared prepregs fabricated by both a solvent-based process and a hot-melt
process, and found that residual solvent in the former case caused significantly higher void content
in both the skins and fillets. Also using a self-adhesive prepreg/honeycomb system, Yuan et al.
[71] studied the influence of process parameters (cure temperature and pressure cycle) on part
quality. They measured internal core pressure by inserting a capillary tube into the honeycomb,
and found that higher-rate temperature ramps increased the buildup of core pressure, as well as the
final void content. They also reported that full autoclave pressure throughout processing resulted
in resin bleed and skin porosity from resin starvation. However, delaying the application of
autoclave pressure until just before gelation prevented these problems.
Researchers have attempted to correlate mechanical properties of cured sandwich
structures (e.g., flatwise tensile strength, peel strength, lap shear strength, etc.) with material and
process inputs [72–77]. Due to the complexity of the co-cure process, however, these studies often
88
resort to statistical techniques such as design of experiments (DoE) and analysis of variance
(ANOVA) [72]. Such approaches can be useful in identifying statistically-significant trends, but
do not offer a complete understanding of the underlying physics. Thus, while they can be used to
determine “optimal” process parameters for a given set of materials, they provide limited insight
into why a particular set of parameters results in a particular outcome. Furthermore, the conclusions
from such studies cannot necessarily be applied to other materials. Due to the coupling of various
process phenomena, even seemingly minor changes to the materials or process can have far-
reaching and potentially counterintuitive consequences. For example, Hou et al. [73] observed that
certain adhesives foamed under vacuum at cure temperatures, while others did not, but that
foaminess was not a reliable indicator of flatwise tensile strength. They showed that a thermal
aging pretreatment of one adhesive increased the minimum viscosity during cure, reducing the
severity of foaming and leading to an increase in flatwise tensile strength, while another adhesive
with an even higher minimum viscosity did not foam at all, yet exhibited the lowest flatwise tensile
strength (a different problem – poor wetting and inferior fillet formation – occurred in that case).
Nagarajan et al. [74] noted similar complications in interpreting results of a DoE/ANOVA study,
due to strong interactions between factors. In particular, the factor levels that resulted in higher
mechanical properties for samples with standard aluminum honeycomb generally had the opposite
effect for samples with a “vented” core structure. Overall, samples with larger fillets and fewer
voids had superior properties, but how and why a given set of process parameters led to a specific
bond-line morphology remained unclear.
A recurring theme in the literature on sandwich panel manufacturing is the importance of
the gas pressure within the core cells: it directly affects the hydrostatic pressure of the
resin/adhesive at the skin/core interface, and is thus a key parameter influencing void formation.
89
The core pressure is influenced by multiple competing factors, making it difficult to predict, and
measuring core pressure in a real sandwich panel during autoclave cure is highly impractical. Two
bodies of work – those of Tavares et al. [78–82] and Kratz & Hubert [83–86] – address these
challenges and describe the behavior of core pressure in the context of vacuum bag-only (VBO)
co-cure (i.e., using only atmospheric pressure, instead of elevated autoclave pressure, to provide
compaction). Both groups utilized lab-scale experimental fixtures, which consisted of tool plates
containing recessed pockets. Honeycomb core was placed into the pocket and covered with
prepreg, forming a “half-sandwich” assembly, and a gas pressure sensor connected to the interior
of the now-sealed pocket was used to measure the simulated “core” pressure.
Tavares et al. described a falling-pressure method to characterize the air permeability K of
prepreg skins, and showed a dependence on resin viscosity [78]. They considered the influences
of the film adhesive and the initial impregnation state of the prepreg on both gas migration through
the skin, and on resin bleeding into the core cells [79]. They also explored the effects of skin
modifications (perforations in each ply, perforations through the entire skin, etc.) on air
permeability [80]. The various skin-modification schemes resulted in varying ranges of core
pressures, and the authors noted that different issues arose at opposite extremes of core pressure:
low core pressures caused volatile-induced void formation within the adhesive, while high core
pressures resulted in small fillets due to insufficient compressive force between the skin and core.
An intermediate core pressure range of 40 – 70 kPa limited the severity of both of these issues and
resulted in samples with the greatest peel fracture energy [81].
Using a similar experimental setup for single-skinned "half-sandwich" assemblies, Kratz
& Hubert also measured core pressure during room-temperature vacuum holds and during cure
[83]. Unidirectional fabrics were effectively impermeable over a 24-hour vacuum hold [84],
90
because the pressure difference across the skin was insufficient to overcome the capillary pressure
of the viscous resin between the closely-spaced fibers. Conversely, woven fabrics allowed
through-thickness air evacuation because the fabric structure included larger macro-pores between
the overlapping fiber tows (where gas must overcome a lower capillary pressure to displace liquid
resin). However, a delay was observed between the application of vacuum and the onset of core
evacuation, attributed to the transient process of air displacing resin to form continuous open
channels through the skin. Also, the evacuation process halted before the core pressure fell to
match the vacuum bag pressure, resulting in a residual core pressure. At this point, the gas pressure
in the air channels became insufficient to balance the surrounding resin pressure, causing the
channels to close and preventing further gas flow through the skin.
During cure, the core pressure depended on competing factors [85]. Continual evacuation
through the skin – governed by the prepreg air permeability – tended to reduce the core pressure,
while pressure increased upon heating due to ideal gas law behavior and the vaporization of
moisture (initially absorbed by the organic Nomex honeycomb core from ambient humidity).
Based on these phenomena, a model for core pressure was developed and compared to measured
core pressures in full-scale sandwich structures using embedded MEMS sensors [86]. The
model/experiment comparisons affirmed the validity of the assumption that a single-skinned co-
cured structure could be used to characterize the processing of realistic sandwich structures.
For all of the aforementioned studies, visual inspection of sandwich structures was limited
to ex situ methods (e.g., optical microscopy of polished sections and computed X-ray tomography
[83,84,87]), which cannot adequately characterize the transient process phenomena that occur
during co-cure. We have published preliminary results for an in situ visualization concept: a
simplified method [88] using a windowed tool that was originally designed for resin transfer
91
molding [89], and initial exploratory tests [90,91] using the tool described here. This chapter
presents an application of the in situ diagnostic method: first, the causes of observed bond-line
defects are identified and explained, and subsequently, a solution strategy is proposed and
demonstrated to be effective at preventing bond-line defects.
4.1.2. Objectives and approach
The objectives of this chapter are (1) to reveal the dynamic behavior of bond-line evolution
and the aforementioned defect formation phenomena during co-cure in realistic autoclave
conditions, and (2) to demonstrate the utility of in situ visualization as a diagnostic tool for the co-
cure process. We present a case study consisting of five co-cure experiments. The test parameters
were selected to showcase a variety of process conditions and phenomena, and to exemplify
insight-driven process design. In lieu of using a DoE approach to correlate process parameters with
manufacturing defects, we identified sources of defects through direct visual observations, and
implemented informed process modifications to prevent the observed defect formation
mechanisms. The order in which the cases are presented follows a logical narrative: each of the
first three cases (A, B, & C) exhibited a distinct type of defect formation, which was addressed by
modifying the subsequent case. After three iterations of identifying and addressing process-
induced defect mechanisms, the fourth experiment (Case D) resulted in an “ideal” sample. Finally,
Case E tested the feasibility of utilizing the strategy from Case D (which included an “equilibrated
core” condition) in the most generic case of co-cure (i.e., where the core pressure can differ from
the vacuum bag pressure).
92
4.2. Experimental facilities
Figure 4.2: Section view of the co-cure fixture with sample and consumables.
The design of our system is based on the concept used by Tavares [78–82] and Kratz [83–
86], but augmented with additional capabilities. Figure 4.2 shows a CAD rendering of the tool,
sectioned at the mid-plane to reveal interior features. The main lower component is a 280 mm
square anodized aluminum plate containing a centrally-located 76 mm square pocket (19 mm
depth). A 12.7 mm thick glass disk with conically-tapered sides is mounted in a matching hole at
the center of the tool plate pocket, with silicone sealant at the tapered interface providing an airtight
seal. Square glass spacers of varying thickness can be placed into the pocket, to accommodate
varying thicknesses of honeycomb core that are placed above, such that the top of the honeycomb
is flush with the plane of the tool plate. A prepreg skin is shown extending beyond the edges of
the pocket to form an “edge band” that seals the interior of the core pocket. Consumables (edge
93
dams, release film, and breather cloth) surround the sample, and a vacuum bag film is located
above, sealed around the perimeter with tacky-tape. A hole near the corner of the tool plate (not
visible in Figure 4.2) is located under the breather cloth inside the sealed perimeter of the vacuum
bag, enabling the application of vacuum within the bagging assembly through a port on the
underside of the tool plate. Another port (the “core vent”) is located in a corner of the core pocket,
allowing the gas pressure in the core cavity to be controlled, if desired. A self-sealing fitting
prevents gas flow through this port when not in use.
The upper large component shown in Figure 4.2 is an aluminum block with a hollowed-
out underside that is bolted onto the tool plate through a set of holes around the perimeter. An O-
ring groove at the interface between the lower and upper aluminum parts (the tool plate and the
“lid”) creates a sealed environment, and a port through the tool plate (located outside the perimeter
of the vacuum bag) allows the internal “autoclave cavity” to be pressurized. Traditionally, the
molding tool for an autoclaved part is a stand-alone structure that can be moved in and out of the
autoclave. In this case, however, the lower aluminum piece acts as both a tool plate and as part of
the “autoclave” pressure vessel. This design enables the observation of the interior of a co-cured
part from outside the autoclave.
The tool plate contains a pass-through for thermocouple wires to measure temperature at
desired locations in and around the vacuum bag assembly. The upper and lower aluminum blocks
contain resistive heating elements, which can be controlled independently through a pair of PID
controllers (Watlow model PM6R1CA). The system was designed to safely support temperatures
up to 200°C and autoclave pressures up to 800 kPa.
Pressure transducers (Omega model PX32B1) measure gas pressure in the hoses providing
vacuum and autoclave pressures (Pbag and Pauto, respectively). A port in the corner of the core
94
cavity pocket connects to another transducer that measures the gas pressure within the core (Pcore).
The interface between the glass spacer and the honeycomb core is intentionally not airtight,
allowing gas pressure to equalize everywhere within the core cavity (i.e., the pressure in the core
cells visible through the window equals the pressure measured by the core pressure transducer). A
fourth pressure transducer (Gefran ME2 series), mounted in the tool plate, measures the prepreg
resin pressure in the edge band. However, the design of this feature renders the measurements
susceptible to artefacts that complicate interpretation. The data from this transducer does not
contribute meaningfully to the results of this study, and is thus omitted from subsequent discussion
to avoid potential confusion.
The tool plate and lid assembly is mounted on a frame (shown in Figure 4.3), allowing a
portable digital microscope (Dino-Lite Edge AM7815MZTL) to be located under the tool, facing
upwards at the window. The frame also contains an enclosure that houses the power supplies for
the heaters and sensors.
Autoclave pressure is supplied by an air compressor and controlled using a regulator and
overpressure relief valve. A vacuum pump connects to the vacuum bag port, and for tests with an
“equilibrated core” condition (explained in Section 4.3.2), another hose from the (otherwise
sealed) core vent is connected to the main vacuum line using a T-junction. Furthermore, using a
three-way valve, the vacuum line can be switched such that a compressed nitrogen tank, rather
than the vacuum pump, is connected to the vacuum bag. This configuration enables venting of the
bag using dry N2 (versus potentially humid air from the environment), but more importantly, it
allows super-ambient pressures to be applied within the vacuum bag. This “in-bag pressurization”
concept is explained in greater detail in Section 4.3.2.
95
Figure 4.3: Photographs of the co-cure fixture. Top: frame with lower half of tool (left) and power distribution
enclosure (right). Bottom left: close-up of tool plate. Bottom right: tool plate with laminate.
96
Temperature and pressure signals are acquired using a digital data acquisition system
(cDAQ-9174, National Instruments) and recorded using a LabVIEW interface on a desktop PC.
The same PC receives images from the digital microscope, and timestamps on both types of data
enable synchronization.
Prior to sample fabrication, all pressure transducers were calibrated over the relevant range
of process pressures and temperatures. Errors in reported pressure values due to thermal drift were
< 3 kPa. Leak testing of the core cavity was performed by sealing a metal plate over the core
pocket, evacuating the core cavity using the core vent, and monitoring changes in core pressure
after removal of the vacuum hose. Threads on the fittings (which included a slow-curing liquid
thread sealant) were tightened until, at both room temperature and the maximum process
temperature, no detectable changes in core pressure occurred over a minimum duration of 1 hour.
4.3. Experimental methods
4.3.1. Materials and layup
The prepreg skins consisted of a widely-used carbon/epoxy autoclave prepreg (HexPly
AGP193PW/8552S, Hexcel Corp., AS4 fibers, 193 g/m
2
fiber areal weight, 3000 fibers/tow, plain
weave fabric, fully-saturated with 38% by weight 8552 toughened epoxy resin). The trailing “S”
indicates the prepreg was manufactured by a solvated tower process [92].
The adhesive was a toughened epoxy film (Loctite EA 9658 AERO 060UNS, Henkel
Corp.) with 290 g/m
2
areal weight and no supporting carrier, designed for metal, composite, and
honeycomb bonding in aircraft engine nacelles [93]. The viscosity of this adhesive is tailored for
“controlled flow” [94], to prevent issues such as hole blockage from excessive bleed, and to
provide a “reticulation” capability. Although many film adhesives are supported by a lightweight
97
carrier film (typically nonwoven glass or polymer fibers), EA 9658 AERO is offered in the
unsupported “UNS” form so that it can be placed onto honeycomb core and then reticulated. The
reticulation process – which can be performed in several ways [75,95] – results in adhesive located
only along the edges of the honeycomb cell walls, with an opening in the film at the center of each
cell. Figure 4.4 shows the stages of adhesive reticulation. First, the adhesive film was placed onto
the top side of the honeycomb core and manually perforated in the center of each honeycomb cell
using a ~1 mm diameter needle. Then, heated air was blown over the surface of the film using a
handheld hot-air gun. As the film softened under the applied heat, surface tension caused the holes
to widen, reshaping the film into an adhesive bead along the edges of the cell walls. Temperatures
were monitored using a thermocouple and did not exceed 70°C. Reticulation was completed in <
3 minutes, ensuring that the process did not advance the cure of the adhesive.
Figure 4.4: Stages of the adhesive reticulation process. Left: film applied to honeycomb core and perforated. Middle:
film softening under applied heat and perforations opening due to surface tension. Right: process completed when
adhesive forms a bead along the edges of the cell walls.
The honeycomb core consisted of a phenolic-coated aramid (Nomex) paper (Gillcore
HD433, The Gill Corp., 6.4 mm cell diameter, 12.7 mm thickness, 48 kg/m
3
density).
Identical material kits (consisting of core, one adhesive layer, four prepreg plies, and
vacuum bag consumables) were used for all samples. First, a liquid mold-release agent (Frekote
770-NC, Henkel Corp.) was applied to the tool surfaces and to a glass spacer in the bottom of the
core pocket. A 76 mm square of honeycomb core was placed into the pocket, its top surface flush
98
with the plane of the tool plate. Four plies of prepreg were cut into 126 mm squares, providing a
25 mm edge band to extend beyond the edges of the core pocket. The plies were stacked in a (0°)2S
orientation, one layer of film adhesive (also 126 mm square) was applied to the underside, and the
completed skin was placed onto the tool, centered over the core pocket. In cases where the adhesive
was reticulated, a 76 mm square was cut from the center of the 126 mm square of adhesive film
and applied onto the core (as described above) prior to placing it into the core pocket. The
remaining “frame” of film adhesive was applied to the prepreg skin, so in all cases the edge band
contained a continuous adhesive layer between the tool plate and the first prepreg ply. The primary
reason for extending the film adhesive beyond the edges of the core and into the edge band was to
ensure that no leak path existed between the core cavity and the vacuum bag. If the textured surface
of the prepreg were directly against the tool plate, gas pathways could exist at the interface.
Conversely, the smooth surface of the film adhesive against the tool plate reliably formed an
airtight seal.
The prepreg edges were sealed with tacky tape (GS 213, Airtech Intl.) to enforce out-of-
plane gas transport only. This configuration approximates the center of a large sandwich structure,
far from any breathing edges. A perforated release film (A4000, Airtech Intl.) was placed over the
prepreg skin, then a breather cloth (Airweave N10, Airtech Intl.), and finally a vacuum bag film
(Wrightlon 7400, Airtech Intl.) was placed over the entire assembly and sealed to the edges of the
tool plate with tacky tape.
4.3.2. Process parameters for case study
The thermal cycle used for all cases is shown in Figure 4.5, and closely resembles the
manufacturer-recommended cure cycle for the prepreg [92]. After a one-hour room-temperature
dwell, the temperature was raised to 110°C and held for one hour, then raised to 177°C and held
99
for two hours (all temperature ramps were 2°C/minute). The room-temperature, intermediate-
temperature, and high-temperature stages will be referred to as Stages I, II, and III, respectively.
Figure 4.5: The temperature cycle used for all tests (top), and the two cycles used for autoclave and vacuum bag
pressures (middle and bottom). Stages I, II, and III denote room-temperature, intermediate-temperature, and high-
temperature stages (respectively) of the process.
Three parameters were varied between cases: the initial adhesive format, the breathability
of the core, and the imposed Pauto and Pbag pressure cycles. The five test cases are represented as a
100
“cube diagram” in Figure 4.6. Each of the three axes corresponds to one of the varied parameters
as described below.
Figure 4.6: Diagram of Cases A through E. The three axes denote the test variables: (1) the adhesive was either
applied as a continuous film, or reticulated onto the honeycomb core; (2) the core cavity was either sealed, or
connected via the “core vent” to equilibrate the gas pressure in the core cavity with that of the vacuum bag; and (3)
autoclave and vacuum bag pressures were applied according to either of two cycles – “simple” or “staged” – as
shown in Figure 4.5.
Adhesive format
A continuous layer of film adhesive was used for Case A, while for Cases B – E, the portion
of the film over the honeycomb core was reticulated (see Section 4.3.1 for a detailed description
of the reticulation process).
101
Core breathability
Depending on the configuration of a given sandwich structure, the “breathability” of a
honeycomb core insert can vary from being completely sealed to being completely equilibrated
with the external environment. If the core cell walls are impermeable to gas (e.g., aluminum
honeycomb), then gas transport can only occur through the skins, and the relationship between
Pcore and Pbag depends on the skin permeability.
Some sandwich structures, however, have a “vented” construction that allows the internal
core pressure to equilibrate with external gas pressure. This can be achieved by using a honeycomb
with perforations/openings in the cell walls, and by including some pathway for gas to travel
between the interior and exterior of the sandwich (e.g., panels designed for acoustic damping can
be fabricated using a pre-cured tool-side skin that contains pre-drilled holes [96]). In addition to
acoustic applications, vented sandwich structures are commonly used to prevent moisture
accumulation during service, and to relieve internal pressure buildup for space launch vehicles
[97]. During processing of vented structures, Pcore is equal to Pbag, and can thus be directly
controlled. Furthermore, because a core/bag pressure difference cannot develop, equilibrated cores
avoid the possibility of driving gasses through the skin.
Cases A, B, and E were fabricated using a sealed core cavity, whereas Cases C and D
simulated a vented structure with an equilibrated core, by connecting the core vent to the vacuum
bag line.
Pressure cycle
The two pressure cycles used – “simple” and “staged” – are shown in Figure 4.5. Cases A,
B, and C used the simple cycle while Cases D and E used the staged cycle. The simple pressure
cycle included a one-hour room-temperature vacuum hold (Pbag) with ambient pressure in the
102
autoclave cavity (Pauto) for Stage I. Then Pauto was increased to 377 kPa (40 psig), and both Pbag
and Pauto were held constant during the heated portion of the cure cycle (Stages II and III). During
typical autoclave processing, the vacuum bag is vented after autoclave pressure is applied, whereas
for VBO processing, the bag is generally left under full vacuum throughout cure (because venting
the bag would eliminate the only source of compaction pressure). The “simple” pressure cycle
used here is thus atypical of autoclave cure, but showcases particular defect formation phenomena.
The staged pressure cycle was identical to the simple cycle during Stage I. However, before
ramping upwards in temperature, Pauto was increased to 205 kPa (15 psig) and then the vacuum
bag was vented with atmospheric pressure N 2. These pressures were maintained until the end of
Stage II, at which time Pauto was increased to 377 kPa (40 psig) and Pbag was increased to 239 kPa
(20 psig).
The concept of imposing super-ambient bag and core pressures during sandwich structure
manufacturing has been previously considered in the context of preventing core crush. Alteneder
et al. [98] pressurized the bag and core prior to heating, and then sealed the bag such that ideal gas
law behavior would further increase the pressure upon heating. This approach proved effective at
preventing core crush under autoclave pressures, but the authors did not discuss the effect of in-
bag pressurization on bond-line defects.
In-bag pressurization has been used to suppress void formation in monolithic laminates.
This technique was developed for the Air Force in the early 1980s [99] and summarized by
Campbell et al. [67]. When curing laminates with high-bleed prepregs (as opposed to low-bleed
“net resin content” prepregs) by traditional autoclave methods, they measured resin pressures that
initially matched the applied autoclave pressure. However, as resin bled out of the laminate,
progressively more of the autoclave pressure was carried by the fiber bed and the resin pressure
103
dropped, leading to void formation. Conversely, applying positive pressure within the bag limited
the extent to which the resin pressure could drop and resulted in void-free parts. Noting the
similarities between autoclave cure of high-bleed monolithic laminates and the co-cure of
sandwich structures (in which the resin pressure can drop – even in the absence of bleeding – due
to low gas pressure in the core), we implemented the in-bag pressurization concept in the staged
pressure cycle. Further discussion of the rationale behind the staged cycle is provided in Section
4.4.4.
4.4. Results
Measured data for Cases A, B and C is shown in Figure 4.7, and in situ images from these
tests are shown in Figure 4.8. For Cases D and E, measured data and corresponding in situ images
are shown in Figure 4.9 and Figure 4.10, respectively. To fully appreciate the dynamic behavior
captured in the visual data, the reader is strongly encouraged to view the time-lapse videos (in the
“supplemental materials” of the online version of this article [100]) from which the images in
Figure 4.8 and Figure 4.10 were taken.
4.4.1. Case A
Figure 4.7 shows the measured temperatures (top) and pressures (middle) for Case A,
which featured a continuous adhesive film, a sealed core, and the simple pressure cycle. Previously
developed models [88,101] were used to compute the glass transition temperature Tg of the prepreg
and adhesive, as well as the viscosity 𝜇 profiles (bottom graph). The indicated times of interest t1
through t4 correspond to the images in the left column of Figure 4.8.
104
Figure 4.7: Temperature and pressure data for the three cases using the “simple” pressure cycle (A, B, and C).
Models for the glass transition temperature and viscosity (of both the prepreg resin and adhesive) are shown,
computed from the temperature history recorded at the outside of the vacuum bag near the center of the laminate.
105
Figure 4.8: In situ images of the bond-line during processing. Columns correspond to Cases A, B, and C (left to
right), and rows (top to bottom) correspond to the times of interest t 1 through t 4 indicated on Figure 4.7
106
Figure 4.9: Temperature and pressure data for the two cases using the "staged” pressure cycle (D and E). Models for
the glass transition temperature and viscosity (of both the prepreg resin and adhesive) are shown, computed from the
temperature history recorded at the outside of the vacuum bag near the center of the laminate.
107
Figure 4.10: In situ images of the bond-line during processing. Columns correspond to Cases D and E (left to right),
and rows (top to bottom) correspond to the times of interest t 1 through t 4 indicated on Figure 4.9.
108
Upon application of vacuum within the bag, Pcore remained unchanged, since the
continuous adhesive film acted as a barrier for gas transport. Little change was observed visually
during Stage I, although significant flow occurred as the temperature was raised between t1 and t2.
First, prepreg resin (which is clear, while the aluminum-powder-filled adhesive is gray) pushed
through the adhesive film in a grid-like pattern, clearly emanating from the “pinholes” between
the fiber tows in the fabric weave. Additionally, the mixture of adhesive and prepreg resin migrated
toward the cell walls, forming fillets. During this period, Pcore rose to ~200 kPa due to ideal gas
law expansion (and likely some volatile generation). During the intermediate temperature dwell,
little change was observed visually, and Pcore decayed due to continual application of vacuum to
the bag.
During the second temperature ramp (between t3 and t4), the adhesive fillets appeared to
“inflate” and bubbles appeared in the locations where the prepreg resin had created discontinuities
in the film adhesive. After t4, all motion abruptly ceased, as the adhesive had gelled. Note that
during the second temperature ramp, the adhesive viscosity was already rapidly increasing,
whereas the prepreg resin was just reaching its minimum viscosity. Thus, during this ramp, bubbles
could still nucleate and grow within the prepreg resin. However, they became trapped by the
adhesive film. The section view of the sample in Figure 4.1A confirms that the fillet “inflation”
was due to the growth and entrapment of bubbles between the adhesive film and the first ply of
prepreg.
Pcore followed a similar trend during Stage III as in Stage II, first increasing during the
temperature ramp (although to a lesser extent), and decaying during the dwell.
109
4.4.2. Case B
Case A showed that the continuous adhesive film was disrupted by the behavior of the
prepreg resin. Specifically, the mismatch in gel times was such that, during the second temperature
ramp, the prepreg resin formed bubbles that distorted – and became trapped by – the rapidly gelling
adhesive layer. To prevent the adhesive from acting as a barrier for gas transport, the film was
reticulated onto the core in Case B.
Figure 4.7 shows Pcore for Case B (all other temperatures and pressures are identical to
Case A and thus omitted for clarity). The core evacuation during Stage I, in the absence of a
continuous adhesive film, depended on the prepreg permeability. Initial evacuation was rapid, but
halted as the skin compacted (analogous to the behavior observed by Kratz [84]). A new pathway
for gas transport formed after ~45 minutes (the delay caused by gradual displacement of the highly
viscous prepreg resin), allowing core evacuation to continue. During the remainder of the cycle,
Pcore followed a similar trend as in Case A, although shifted downward due to the lower initial
pressure upon heating.
In Figure 4.8 (Case B, t1) the reticulated adhesive is visible along the edges of the cell
walls, and the first prepreg ply is visible in the center of each cell. During the room-temperature
vacuum hold, minor bubble formation was observed in the adhesive and prepreg resin. As the
temperature was increased between t1 and t2, the adhesive fillets formed, and a mixture of prepreg
resin and bubbles could be seen pushing out from the pinholes in the fabric weave. By t2, most of
the bubbles had disappeared, leaving a layer of resin at the skin/core interface and creating a
continuous meniscus with the adhesive. During the dwell between t2 and t3, bubbles gradually
reappeared in the prepreg resin. During the final temperature ramp, the bubbles grew, yet none
burst before resin gelation, resulting in a large cluster of bubbles at the center of each cell.
110
Contrasting this behavior to Case A reveals that the bubbles in the bond-line in Case A
were not solely due to the adhesive acting as a barrier, but also a consequence of the core pressure.
Even without the barrier (i.e., Case B), Pcore was low enough to allow bubble growth within the
resin at the skin/core interface, yet not so low that the bubbles burst. Considering this insight, the
modification for Case C was chosen to afford improved control over Pcore.
4.4.3. Case C
Case C simulated the conditions of a vented sandwich structure – in which Pcore can
equilibrate with Pbag – by connecting the core vent to the vacuum line. During the room
temperature vacuum hold, thin-walled bubbles grew from the pinholes in the fabric and burst,
indicating that air initially trapped between the prepreg plies was evacuating into the core.
Previously, in Cases A and B, vacuum was only applied on the upper surface of the skin, so air in
the core had to travel upwards through the skin to escape. Conversely, in Case C, the core was
evacuated directly and simultaneously with the vacuum bag, so air was not driven through the skin,
and air initially entrapped within the skin could evacuate in either direction (up through the
perforations in the release film, or down into the core), along the path of least resistance.
Upon initial heating, the vacuum in the core pulled a foamy mixture of gas and resin out
from the fabric pinholes. Towards the end of the first ramp (see Figure 4.8, t2), vigorous frothing
was observed, with large bubbles growing and bursting repeatedly. Bubble formation slowed
during the intermediate-temperature dwell, but, by that point, much of the adhesive had been
spattered onto the cell walls, leaving little at the skin/core interface to form fillets (see Figure 4.8,
t3). Small bubbles remained along the edges of the cell walls against the skin, which grew during
the second temperature ramp and became fixed at gelation (t4). The section view of this sample
111
(Figure 4.1C) provides a complementary perspective of the small, void-filled fillets and the
adhesive coating the cell walls.
4.4.4. Case D
The staged pressure cycle used for Case D was designed to avoid the previously-observed
defect formation phenomena, by applying positive pressure within the vacuum bag to suppress the
release of gases that could not be evacuated prior to gelation. Stage I was identical to Case C:
reticulated adhesive and full vacuum in the bag and core, to maximize air removal from the skin
while the resin and adhesive were still at high viscosity (the “equilibrated” Pcore in Case D matched
Pbag exactly and is omitted from Figure 4.9 for clarity). During Stage II, to avoid the fillet
disruption that occurred in Case C and duplicate the undisrupted flow seen in Case B, the vacuum
bag was vented with N2 to ~100 kPa. To prevent excess resin exuding toward the skin/core
interface, Pauto was increased to only 205 kPa (15 psig), maintaining the same compaction pressure
(Pauto – Pbag) as in Stage I. Finally, in Stage III, recall that during the second temperature ramp in
Cases A and B, a Pcore range of ~50 – 150 kPa caused bubbles to grow yet remain sessile until
gelation. In Case C, vacuum pressure in the core caused bubbles to burst, but this also caused
severe fillet disruption. Therefore, in Case D, the opposite approach was used to prevent defects
in Stage III: the application of positive pressure to suppress bubble formation and growth. First,
Pauto was increased to 377 kPa (40 psig) to maintain compaction on the sample, and then Pbag was
increased to 239 kPa (20 psig).
The images in the left column of Figure 4.10 correspond to the times indicated on Figure
4.9. During the initial vacuum hold, thin-walled bubbles similar to Case C appeared and burst,
indicating air evacuation from the skin. During the temperature ramp and dwell of Stage II, fillet
formation occurred, the flow undisrupted by bubbles (similar to Case B t2 but without the bubble
112
formation seen at ~50 kPa in Case B t3). The image captured at t3 for Case D (Figure 4.10) was
taken in the brief time after Pauto was increased, but before Pcore was increased. Momentarily, resin
containing small bubbles exuded from the fabric pinholes due to the increased compaction
pressure. However, once the bag/core pressure was set to the final value, the excess resin receded
into the skin, and the bubbles disappeared. The equal gas pressures on the upper and lower
boundaries of the skin (i.e. Pbag and Pcore, respectively) resulted in a hydrostatic condition in which
no gas was driven through the skin. Furthermore, the gas pressure in the core/bag imposed a lower
bound for the resin pressure in the skin, which prevented the formation of bubbles. The resin and
adhesive remained effectively motionless during the second temperature ramp and until gelation
(t4).
4.4.5. Case E
While the previous Case D effectively prevented porosity within the skin and fillets (see
Figure 4.1D), in some situations it may not be possible to use a vented sandwich structure with an
equilibrated core (e.g., in marine applications where water ingress could be catastrophic). Thus,
Case E was included to determine if the in-bag pressurization strategy from Case D could be
effectively applied in the case of a sealed core, where gas transport must occur through the bag-
side skin.
During Stage I, Pcore rapidly approached Pbag, but air evacuation halted as the skin
compacted. Evacuation continued after a delay, similar to the behavior of Case B, although with a
shorter delay time (the difference attributed to inherent variability of the materials). When the bag
was vented to atmospheric pressure, Pcore slowly started to rise, but again the rate of change of
Pcore varied spontaneously as discreet channels for airflow opened and closed. The pressure
gradient across the skin reversed direction as Pcore exceeded Pbag due to ideal gas law behavior, but
113
eventually a gas pathway opened, and the two pressures equilibrated. When Pbag was raised at the
beginning of Stage III, Pcore rose to match it within ~90 s (>15 times faster than the equilibration
at the beginning of Stage II), suggesting that the formation of gas transport pathways through the
skin was facilitated by lower resin viscosity.
The in situ observations from Case E were effectively identical to Case D. Fillet formation
occurred during the first temperature ramp, and the resin/adhesive remained almost motionless
from t2 until gelation, with the exception of the same momentary bleeding-then-receding of resin
during the spike in compaction pressure at t3 (when Pauto had already been increased but Pbag had
not).
A section view of the sample from Case E (Figure 4.1E) reveals slightly more porosity
within the fillets than Case D. In Case D, air was first extracted from the sample in both directions
during Stage I, and when Pcore and Pbag were increased during Stages II and III, Pcore and Pbag were
equal, so no gas was driven through the skin. Whenever a core/bag pressure difference exists,
however, gas tends to tunnel through the skin by displacing liquid resin in the macro-pores of the
fabric. In Case E, gas was driven through the skin, and the direction of gas flow reversed several
times. Anytime the core pressure was nearing equilibrium with the bag pressure, gas traveling
through the skin at that moment would have ceased to experience a pressure gradient and become
stationary. Gas bubbles remaining in the bond-line upon gelation contributed to the porosity of the
cured sample.
Case E illustrates how fundamentally conflicting goals render the co-cure process
particularly challenging. Gas removal from the skin can prevent porosity due to entrapped air. In
addition, maintaining resin pressure at the skin/core bond-line by exerting sufficient core gas
pressure at the resin/gas interface can prevent void formation in the bond-line. However,
114
evacuating the bag tends to pull gas from the core into the skin, and reduces the core pressure.
Furthermore, re-pressurizing the core requires driving gas back through the skin. For this reason,
co-cure with an equilibrated core is much simpler: the gas pressure in the core used to maintain
resin pressure can be decoupled from the gas migration through the skin.
Lastly, the repeatability of the strategy from Case E has been tested by fabricating
additional samples with identical parameters (although the full results are omitted here for brevity).
Significant variability was observed in the intermittent delays in core evacuation at room
temperature (e.g., compare Pcore during Stage I for Cases B and E), which is attributed to
irregularity in the size of pinholes in the fabric weave, and to variations in the alignment of pinholes
between neighboring plies. However, irrespective of the extent of core evacuation during Stage I,
in all tests, Pcore reliably equilibrated with Pbag upon the application of elevated bag pressure at t3
(i.e., when the resin viscosity was sufficiently low that gas could displace resin rapidly). The
repeatability of the core pressurization demonstrates that the in-bag pressurization strategy can be
used reliably, provided that the fiber bed has sufficient permeability for the core to equilibrate
before resin/adhesive gelation (for reference, plain weave prepregs similar to the one used here
have been reported to have permeabilities ranging from 10
-19
m
2
to 10
-16
m
2
, depending on the
compaction level, resin viscosity, and saturation [85]). A facesheet consisting of unidirectional
plies, for example, may preclude core pressurization – even at low resin viscosities – due to the
absence of macro-pores between fiber tows. Conversely, a woven fabric with a tow width smaller
than the honeycomb cell size will guarantee that each cell is connected to at least one macro-pore,
increasing the likelihood that every cell can equilibrate with the externally applied gas pressure.
115
4.5. Conclusions
We have demonstrated a tool and method to directly observe bond-line formation for co-
cure in realistic autoclave conditions, and presented experiments that exhibited a variety of process
phenomena. Visual data acquired with this tool revealed the occurrence and timing of defect-
formation mechanisms, including (a) gas entrapment by early gelation of the adhesive, (b) bond-
line bubble growth at intermediate core pressures, and (c) fillet disruption from the bursting of
bubbles at low core pressures. Previous studies have speculated on the occurrence of these
phenomena but, to our knowledge, this method is the first to provide in situ visual data as a function
of time, temperature, and pressure. A strategy to eliminate defects via in-bag pressurization was
demonstrated, and the complications associated with driving gas through the skin were explained.
The case study demonstrates the advantages of in situ visualization for composites
manufacturing (specifically for autoclave co-cure, in this case). Through visual observations, the
process conditions corresponding to defect formation can be identified, informed process
modifications can be made, and theories about the relevant physical phenomena can be validated.
Altogether, this approach can provide understanding and insight into the co-cure process that could
not be attained by traditional inspection of the sandwich structures after manufacturing. Most
importantly, much of the guesswork can be eliminated from process troubleshooting, and the
interactions between materials and phenomena can be analyzed in realistic conditions.
Note that, while we focused on bond-line porosity as the primary quality metric, this is not
the only relevant structural parameter governing mechanical performance. Sandwich components
are complex structures with multiple potential failure modes, and the "weakest link" limits the
overall performance (e.g., even a void-filled bond-line may be sufficiently strong such that the
sandwich structure undergoes core damage before failing at the bond-line).
116
The tool described herein enables additional experiments worthy of investigation.
Possibilities include a more detailed investigation of prepreg permeability, extension to other
prepreg/adhesive/core material combinations, and process optimization for challenging cases such
as thick-skinned sandwich structures. The insights gained can also inform the development of
physics-based models for co-cure processes.
117
CHAPTER 5. Through-thickness gas permeability of prepreg facesheets
during co-cure
5.1. Introduction
Gas transport through prepregs is critically important during the manufacturing of
honeycomb core sandwich structures, although scientific understanding of this topic is limited.
Here we describe experimental observations of behavior that cannot be adequately described using
simple single-phase models for flow in porous media. We also outline a multi-phase modeling
approach that accounts for the presence of liquid resin and the influence on through-thickness gas
transport. The multi-phase flow model is a component of a research effort to develop a physics-
based process model for autoclave co-cure of honeycomb core sandwich structures, the goal of
which is to enable prediction of process-induced defect formation based on material and process
input parameters.
Sandwich structures with a honeycomb core can provide much greater specific bending
stiffness and strength than monolithic laminates by distancing the load-bearing plies from the
neutral bending axis (analogous to an I-beam) [2]. Sandwich structures can be produced by pre-
curing laminate facesheets and then bonding them to the core using film adhesive, or by co-curing
the facesheets and film adhesive in a single step. The co-cure approach offers advantages,
particularly faster manufacturing and reduced infrastructure requirements. However, preventing
defect formation during co-cure is challenging. For typical autoclave cure of monolithic laminates,
autoclave pressure can impart hydrostatic pressure to the prepreg resin, suppressing the nucleation
and growth of voids. Conversely, the cells of honeycomb core are hollow, providing a non-uniform
118
substrate that can transfer autoclave pressure to the prepreg only at the cell walls. Note that the
core cells are also filled with gas (air), and the gas pressure directly affects the prepreg resin
pressure of the innermost prepreg plies (adjacent to the honeycomb core) through the surface of
the bond-line. Depending on the relative magnitudes of the autoclave, bag, core, and resin
pressures, facesheet resin can sometimes bleed into the cells, resulting in a loss of prepreg resin
pressure and an increased likelihood of void growth within the facesheet. To further compound
the issue, the core pressure also affects defect formation within the adhesive bond-line, with lower
pressure levels resulting in void nucleation and growth within the film adhesive, and potential
disruption of the load-bearing fillets that join the core to the laminates. The core gas pressure can
be affected by multiple factors, most notably (1) ideal gas law behavior upon heating/cooling; (2)
volatile release from the core material, the prepreg resin, and/or the film adhesive; and (3) gas
transport through the facesheets.
To predict the formation of process-induced defects, a process model must incorporate
simulation of resin pressure, which depends on the core pressure and, in turn, on the gas
permeability of the prepreg facesheets. Although the permeability of dry fiber beds to a single fluid
phase (liquid or gas) has been reported previously (e.g., in the context of liquid composite molding)
[102], the gas permeability of prepregs has been characterized only to a limited extent. Previous
research has shown that prepreg gas permeability can exhibit high sample-to-sample variability
(particularly at room temperature, see Figure 5.1), and that permeability can vary by several orders
of magnitude during processing [78,79,84] depending on the architecture and compaction state of
the fiber bed, the degree of resin impregnation, and the resin viscosity (see Figure 5.2). Modeling
efforts have relied on empirical correlations for specific materials and process conditions [85], but
these cannot be easily applied if changes are made to the materials or the process.
119
Figure 5.1: Gas pressure in honeycomb core during pre-processing room-temperature vacuum holds. Each trace
represents one of eleven samples with identical configurations. Observed behavior consists of (1) rapid initial
evacuation prior to laminate compaction, (2) a delay of one to seven hours as flow channels form, and (3) core
evacuation once one or more flow channels have developed. Image source: [85].
Figure 5.2: Evolution of facesheet air permeability during elevated temperature processing, showing high variability
and a non-monotonic trend. Image source: [85].
To date, physics-based models for prepreg permeability, which can accurately predict gas
transport for varying material and process conditions, have remained elusive. Here we propose an
120
approach to address this need, which is adapted from a mathematical framework initially
developed by hydrologists to describe groundwater and gas flow through porous stone and soil.
First, experimental results to justify and motivate our approach are presented. Then, terminology
and equations for two-phase flow in porous media are described and applied to gas transport
through prepregs. A non-dimensional solution method is provided, and finally, the model
implementation and comparisons with experimental data are discussed.
Figure 5.3: Schematic of a sandwich structure (bottom image), with inset view of a single honeycomb cell.
121
5.2. Experiments
5.2.1. Methods
Experimental apparatus
To measure gas flow through a prepreg facesheet, a custom-built lab-scale co-cure fixture
was used. Figure 5.4 shows a simplified schematic sectional view. The system included an
integrally-heated flat tool plate featuring a recessed cavity, into which a honeycomb core insert
was placed, such that the top of the core was flush with the upper surface of the tool plate. A
prepreg facesheet was placed over the core, with an edge band extending onto the tool plate to
isolate the “core cavity” from the external environment. The edges of the prepreg were sealed with
tacky-tape to ensure that the only pathway for gas transport into or out of the cavity was through
the prepreg. Consumables placed over the sample included a perforated release film, breather
cloth, and a vacuum bag film.
Figure 5.4: A schematic sectional view of the co-cure fixture (not to scale).
The tool plate contained a through-hole located inside the perimeter of the vacuum bag,
which was connected on the underside to a vacuum pump and a pressure transducer (for measuring
the bag pressure Pbag). The core cavity contained two additional ports: one for a permanently
attached gas pressure transducer (for measuring the core pressure Pcore), and also a “core vent”
122
port. This second port could be used to control Pcore if desired, and a self-sealing quick-release
fitting preserved the airtightness of the core cavity in cases where the port was unused. Note that
the underside of the honeycomb core insert was in contact with a rigid surface that did not form
an airtight interface, allowing gas pressure to equilibrate everywhere within the core cavity
(between neighboring honeycomb cells, and with the core pressure transducer). The cavity was
also equipped with a sealed glass window, allowing real-time imaging of the facesheet/core
interface using a digital microscope. Additional details regarding the design of the experimental
apparatus can be found in the previous chapter.
Materials
Facesheets consisted of plain weave (PW) fully-saturated carbon/epoxy prepreg designed
for autoclave cure (HexPly AGP193PW/8552S, Hexcel Corp., AS4 fibers, 193 g/m
2
fiber areal
weight, 3000 fibers/tow, 38% by weight 8552 toughened epoxy resin), with a four-ply (0º)2s layup.
Plies were cut into 127 mm squares and located centrally over the core, providing a 25 mm edge
band extending beyond the edges of the core pocket on each side.
The honeycomb core material consisted of a phenolic-coated aramid (Nomex) paper
(Gillcore HD132, The Gill Corporation) with 3.2 mm cell diameter, 12.7 mm thickness, and 48
kg/m
3
density. Core pieces were cut into 76 mm squares to fit the core pocket of the test fixture.
Although sandwich structures often include a film adhesive at facesheet/core interfaces,
the goal of this work was to characterize gas transport through prepreg specifically, and thus, no
adhesive was used in the experiments described here.
123
Core evacuation tests
The most common permeability test method for low-permeability media (such as prepreg)
entails imposing a pressure gradient across a porous medium and measuring the rate of pressure
changes in a connected reservoir [78,80,85]. For this study, such tests consisted of applying
vacuum within the vacuum bag and measuring the resulting decay in Pcore over time. The mass
flow rate of gas passing through the facesheet can be determined from the decay in Pcore if the total
volume of the core cavity is known, and permeability K can be computed via Darcy’s law (i.e., by
relating the mass flow rate of the gas with viscosity µ to the pressure gradient ∇𝑃 across the
laminate). Eq. (5-1) shows the vector form of Darcy’s law (the mass flow rate can be obtained
from the flow velocity 𝑣 ̅ by multiplying by the gas density ρ and the area A over which the flow
occurs):
𝑣 ̅ = −
𝐾 ̿
𝜇 ∇𝑃
(5-1)
K is direction-dependent, and is thus a second order tensor. The experimental setup
described here, with edges of the prepreg sealed, gives a measurement of transverse permeability
Kzz
only.
𝐾 ̿
= [
𝐾 𝑥𝑥
𝐾 𝑥𝑦
𝐾 𝑥𝑧
𝐾 𝑦𝑥
𝐾 𝑦𝑦
𝐾 𝑦𝑧
𝐾 𝑧𝑥
𝐾 𝑧𝑦
𝐾 𝑧𝑧
] (5-2)
Tests were performed at ambient temperature (~22 ºC), 50 ºC, 70 ºC and at 90 ºC. These
values were chosen to span a range of resin viscosities, while limiting resin polymerization/cross-
linking over the test duration, so that the resin viscosity could be assumed constant. Autoclave
pressures during testing ranged from ~100 kPa abs (atmospheric pressure, VBO conditions) to
~400 kPa abs (45 psig).
124
Because core evacuation tests rely on gas flow due to a pressure difference between the
core and bag sides of the laminate, data collected after the core has fully evacuated is not
meaningful. Therefore, during these tests, the sealed core vent was intermittently opened to
introduce new air into the core cavity, enabling successive measurements on a single sample. Some
hysteresis may arise during such tests due to repeated cycling (further discussion is provided in
section 5.2.2.1).
Figure 5.5: Schematic representation of the two permeability test configurations.
Steady-state flow tests
To provide additional insight into the nature of gas transport through a prepreg facesheet,
a second, complementary test method was developed. “Steady-state” tests were conducted with an
identical sample configuration and range of process temperatures as the core evacuation tests. First,
the core and bag were evacuated simultaneously, using the same vacuum pump connected to both
ports. Then, the core vent was switched to a gas mass flow controller (MC Series, Alicat
Scientific), while the bag was retained under vacuum. The flow controller was used to introduce
air into the core cavity at known, controlled rates ranging from 1 to 9 standard cm
3
/minute
(equivalently, ~21 to 192 µg/s).
125
This air inflow increased Pcore, resulting in a pressure difference between Pcore and Pbag,
and subsequent gas flow through the facesheet. If/when Pcore stabilized at a constant value, the rate
of flow exiting the core cavity through the facesheet was equal to the (known) rate of gas flow into
the core cavity through the flow controller. Thus, the permeability could be determined at each
moment in time by relating the flow rate to the corresponding pressure gradient across the
facesheet.
5.2.2. Results
Core evacuation tests
The measured Pcore versus time is shown in Figure 5.6 for a test conducted with ambient
pressure in the autoclave (VBO conditions) at ~22 ºC. Sudden increases in Pcore correspond to
manual opening of the core vent to introduce new air into the core cavity. The data shows evidence
of complex material behavior. First, the rate of decay was not constant between successive
evacuations. The core evacuated most rapidly during the first cycle, with Pcore decaying more
slowly in subsequent cycles. Furthermore, the decay in Pcore sometimes halted altogether (most
apparent in the last cycle, during which Pcore only dropped to ~36 kPa before gas stopped flowing
through the facesheet). Finally, the maximum rate of decay did not always coincide with the
maximum core/bag pressure difference (as Darcy's law would predict for a medium with constant
permeability). This behavior is most apparent in the last two decay cycles. During the second-to-
last evacuation, the rate of decay of Pcore increased after ~4 minutes, despite the fact that the core
pressure was greater at the onset of flow. Similarly, during the last cycle, the evacuation was
initially slow, gradually accelerated to reach a maximum rate, then slowed again.
Figure 5.7 shows the measured Pcore versus time for a test at room temperature, in which
400 kPa of compaction pressure was applied just after the start of the first evacuation cycle. The
126
times corresponding to core refilling events, although less obvious in this data, are, as before,
identifiable by abrupt increases in Pcore. The reduction in facesheet permeability over time was
accelerated by the greater compaction pressure; under a compaction pressure of 100 kPa the
permeability was continually decreasing for >2 hours, whereas a compaction pressure of 400 kPa
rendered the facesheet effectively impermeable in <1 hour.
Figure 5.6: Core pressure versus time for successive evacuation cycles, at room temperature and a compaction
pressure of 1 bar.
Figure 5.7: Core pressure versus time for successive evacuation cycles, at room temperature and a compaction
pressure of 4 bar.
The observed deviations from constant-permeability behavior can be attributed to
compaction of the fiber bed and gradual flow of viscous resin within the prepreg. First, because
the prepreg plies were not pre-compacted before testing, compaction began upon the initial
application of vacuum to the bag (due to compaction pressure stemming from the difference
between Pbag and the autoclave environment). Prior to compaction (i.e., at the start of the first
evacuation cycle), gaps between the prepreg plies allowed in-plane gas flow to occur, connecting
127
high-transverse-permeability locations within each ply to those of adjacent plies. The plain weave
fabric of the prepreg contained "pinholes" where gaps between warp-direction and weft-direction
tows crossed, and these openings acted as the path of least resistance for gas transport through a
single ply (see Figure 5.8b). For gas to travel through a series of plies it required a pathway from
one pinhole to the next between plies, and these inter-ply spaces became constricted as the
laminates gradually compacted under applied pressure. This process is consistent with the overall
observed trend of slower evacuations (i.e., decreasing permeability) over time.
Figure 5.8: Photographs of three prepregs with different fabric architectures. Images are back-lit to highlight
“pinhole” macro-pores in the woven fabrics (b) and (c). Image source: [84].
Although compaction affects the gas permeability of dry fiber beds in a similar manner as
prepregs, to explain the remaining notable features in the data (i.e., accelerating flow, and residual
pressure after a cessation of flow), one must consider the presence of the liquid phase. Figure 5.9
shows a section micrograph of 4 prepreg plies in the as-received state, prior to compaction and
resin flow (this sample was "cold-cured" to allow sectioning and polishing without changing the
initial microstructure, using the method in [103]). The large dark areas are inter-ply gaps, initially
present after layup, but likely to disappear due to intimate ply contact during cure.
128
Figure 5.9: Section view of 4 plies of prepreg prior to compaction and resin flow.
The fiber tows are discernable in Figure 5.9 by the slightly yellow/brown tint, while the
resin is lighter gray. The fiber tows are fully saturated with resin (unlike vacuum bag-only
prepregs, which feature a partially impregnated microstructure with dry channels for air evacuation
at the center of each fiber tow). Furthermore, each resin/gas interface is a curved meniscus.
According to the Young-Laplace relationship (Eq. (5-3)), the difference between the gas pressure
Pg on one side of a meniscus with radius R and the pressure of the wetting fluid Pw on the other
side (at equilibrium) is equal to the capillary pressure Pc due to surface tension 𝜎 .
𝑃 𝑔 − 𝑃 𝑤 = 𝑃 𝑐 =
𝜎 𝑅 (5-3)
Figure 5.10: Schematic representation of a spherical meniscus with gas pressure P g on the left and wetting fluid
(liquid) pressure P w on the right.
To migrate, gas trapped within resin-rich regions of the prepreg must displace liquid resin
by overcoming the capillary pressure, which is inversely proportional to the size of pore spaces
129
between fiber tows. This relationship explains why gas transport can halt despite the existence of
a core/bag pressure difference: unless Pg > Pw + Pc, gas cannot displace resin to advance toward a
low-pressure boundary. For example, in Figure 5.6, during the last evacuation cycle, the laminate
had compacted to the extent that a core pressure of ~36 kPa was insufficient to overcome the
capillary pressure of the meniscus blocking the largest available pathway through the fiber bed.
The concept of capillary pressure reinforces the assertion that gas bubbles preferentially
travel in the macro-pore spaces between fiber tows, rather than in the micro-pores within fiber
tows; gas will displace resin in the largest available channels first, due to the lower capillary
pressure that must be overcome. This behavior has also been observed using in situ observations
[104].
Another notable feature of the data in Figure 5.6 – spontaneously or gradually accelerating
gas transport – can be explained by considering the viscosity of the resin. Even if the gas pressure
is sufficient to displace liquid resin, resin flow does not occur instantaneously. Rather, the flow
rate is limited by viscosity, potentially leading to non-negligible pathway desaturation times at
high viscosities (e.g., at room temperature). This transience can result in entirely new pathways
opening spontaneously (as in the second-to-last evacuation cycle), and existing pathways gradually
expanding to allow less restricted gas flow (as in the last evacuation cycle).
While the behavior of the samples at room temperature was dominated by changes in
compaction, high-temperature tests revealed differences in behavior due to the resin viscosity. To
isolate this effect, these samples were pre-compacted by imposing full vacuum in both the core
and bag, then applying autoclave pressure and holding for 1 hour at room temperature. Next, the
core and bag were vented to atmospheric pressure (to prevent resin foaming/redistribution upon
heating), the autoclave pressure was stepped up to maintain the same compaction pressure (Pauto –
130
Pbag) as prior to the venting of the bag, and the sample was heated to the test temperature of 50 ºC,
70 ºC or 90 ºC. Finally, Pauto was brought back down to the initial pressure level, the core vent was
sealed, and vacuum was reapplied to the bag, starting the core evacuation test. Although laminate
compaction may not have reached its final state prior to beginning these tests, care was taken to
ensure that all samples were exposed to approximately equal conditions (i.e., equal time under
compaction pressure). Figure 5.11 shows an example of an elevated-temperature core evacuation
test.
Figure 5.11: Temperatures, pressures, and resin viscosity for a core evacuation test at 70 ºC and 300 kPa of
compaction pressure.
131
Figure 5.12 shows the core pressure versus time for core evacuation tests at 50 ºC, 70 ºC,
and 90 ºC, under 200 kPa of compaction pressure. At 70 ºC and 90 ºC, the core evacuated rapidly
and without delay, although the higher temperature (lower resin viscosity) sample evacuated faster.
Conversely, at 50 ºC, flow halted and restarted intermittently because of higher resin viscosity. As
the core pressure decayed, the driving force for bubble motion decreased and caused gas flow to
temporarily cease. Simultaneously, resin gradually flowed out of the fiber bed (through the
perforated release film into the breather cloth) due to the vacuum in the bag and due to squeeze
flow from compaction. As the resin flowed and its pressure decreased, bubbles regained mobility
and gas flow resumed. These competing effects (decreasing driving force for gas bubbles and
decreasing resin pressure), caused the intermittent evacuation behavior observed at 50 ºC. At 70
ºC and 90 ºC, however, where the resin viscosity was lower, the time required for resin to
desaturate pores in the fiber bed did not result in noticeable delays.
Figure 5.12: Core pressure versus time for core evacuation tests with 200 kPa of compaction pressure at 50 ºC
(blue), 70 ºC (orange), and 90 ºC (yellow).
132
Steady-state flow tests
Figure 5.13 shows the core pressure versus time for steady-state tests at 50 ºC and 70 ºC,
with 100 kPa of compaction pressure (VBO conditions). At the higher temperature (i.e., lower
resin viscosity), Pcore rapidly stabilized, indicating that the rate of gas flow through the prepreg
equalled the rate at which gas was introduced into the core cavity. At the lower test temperature,
however, the higher resin viscosity markedly impeded gas flow. The ranges of the data with a
constant upward slope correspond to zero gas flow through the prepreg (computed based on the
rate of gas flow into the core cavity through the flow controller, and the volume of the core
reservoir). As Pcore increased, the driving force for bubble motion eventually exceeded the
resistance to flow (i.e., the sum of the resin pressure and capillary pressure), and gas began to flow
through the prepreg. Pcore then started to drop, but, just as in the core evacuation tests, the flow
ceased once Pcore became insufficient to overcome the capillary pressure. The same competing
effects that caused the start/stop behavior in the core evacuation test at 50 ºC (Figure 5.12) caused
the saw-tooth behavior observed in the corresponding steady-state flow test (Figure 5.13).
Figure 5.13: Core pressure versus time for evacuation tests at 50 ºC and 70 ºC
Figure 5.14 shows normalized core pressures versus time for the same tests as Figure 5.13
(top row) and the corresponding instantaneous permeability values (bottom row) computed using
the following form of Darcy’s law:
133
𝑚 ̇ 𝑜𝑢𝑡 =
−𝐾𝐴
2𝜇𝐿
( 𝑃 𝑏𝑎𝑔 2
− 𝑃 𝑐𝑜𝑟𝑒 2
)(
𝑀 𝑎𝑖𝑟 𝑅𝑇
) (5-4)
where L is the thickness of the facesheet, Mair is the molar mass of air, R is the ideal gas constant,
T is the temperature in Kelvin, and 𝑚 ̇ 𝑜𝑢𝑡 is the mass outflow rate from the core (computed from
the mass inflow, set by the flow controller, minus the mass accumulation in the core).
𝑚 ̇ 𝑜𝑢𝑡 = 𝑚 ̇ 𝑖𝑛
−
𝑃 ̇ 𝑐𝑜𝑟𝑒 𝑉 𝑐𝑜𝑟𝑒 𝑀 𝑎𝑖𝑟 𝑅𝑇
(5-5)
Figure 5.14: Normalized bag/core pressure differences (top) and effective permeability values (bottom) for steady-
state flow tests at 50 °C (left) and 70 °C (right).
The portions with a constant upward slope in Pcore (at 50 °C, left column) correspond to
zero gas flow through the facesheet. When gas was flowing at 50 °C, the permeability K was on
134
the order of 10
-17
m
2
, whereas the lower resin viscosity at 70 °C resulted in greater permeability
values by more than an order of magnitude.
Another effect that is not present in single-phase flow though porous media can be seen in
Figure 5.15, which shows core pressures and corresponding permeability values for steady-state
flow tests at 70 °C and three mass flow rates. For a porous medium with constant permeability,
the steady-state buildup in Pcore would increase proportionally with the gas flowrate, however in
these tests, the relative increases in measured Pcore values were less than the corresponding relative
increases in flowrate (i.e., higher gas pressures led to an increase in facesheet permeability). This
behavior, too, was due to the presence of the second fluid phase.
Figure 5.15: Core pressure and effective permeability versus time for steady-state tests at three flow rates.
135
Visual observations performed using the test fixture provide evidence that gas transport
through fully saturated prepreg is dominated by bubble motion in the inter-tow spaces. A separate
steady-state test at 50 ºC was conducted, in which the direction of gas flow was reversed. In this
case, the mass flow controller was connected to the vacuum bag port, and vacuum was pulled
through the core vent. This configuration allowed imaging (through the core cavity window) of
the prepreg boundary at which outflow occurred. Figure 5.16 shows images recorded during three
stages of the test.
Figure 5.16: In situ images of bubbles moving through prepreg into the core cavity. Initial state (left), during the
temperature ramp (middle), and during the steady-state portion of the test (right).
The first (left) image in Figure 5.16 shows the initial state of the prepreg, prior to heating.
The center image was recorded during the ramp to 50 ºC, when vacuum was being applied to both
sides of the prepreg. During this time, a foamy mixture of bubbles and resin exuded from the
pinholes (indicated by red arrows) as the sample compacted and air initially entrapped between
the plies was evacuated. The right image was recorded during the “steady-state” portion of the test.
As air bled through the prepreg, bubbles appeared at the pinholes, grew, and then ruptured. This
behavior continued steadily over several hours.
136
Altogether, the experimental results support the assumptions that in fully-saturated woven
prepreg, (1) gas travels through the prepreg in the form of bubbles through the macro-pore spaces
between fiber tows, and (2) the mobility of these bubbles is limited by the resin viscosity and
capillary forces.
5.3. Modeling – constant-K approach
A first attempt to model the permeability of prepreg facesheets consisted of fitting a single,
constant K value to each core evacuation test. The sample configuration was represented by the
simplified 1-D model shown in Figure 5.17 (left side), where z = 0 corresponds to the inner
boundary of the bag-side facesheet, z = L corresponds to the external boundary of the bag-side
facesheet, Vcore is the volume of the gas reservoir (honeycomb core), and VP is the volume of the
facesheet (porous medium) through which gas flow occurs.
Figure 5.17: Simplified 1-D representation of a sandwich structure.
137
Using the ideal gas law, Eq. (5-4) can be rewritten as follows:
−𝐾𝐴
𝐿𝜇 𝑉 𝑐𝑜𝑟𝑒 =
𝑑 𝑃 𝑐𝑜𝑟𝑒 𝑑𝑡 2
𝑃 𝑏𝑎𝑔 2
− 𝑃 𝑐𝑜𝑟𝑒 2
(5-6)
Integrating Eq. (5-6) from 0 to t yields the following expression [84]:
−𝐾𝐴 𝑃 𝑏𝑎𝑔 𝐿𝜇 𝑉 𝑐𝑜𝑟𝑒 𝑡 = ln (
( 𝑃 𝑐𝑜𝑟𝑒 ( 𝑡 0
)+ 𝑃 𝑏𝑎𝑔 ) ( 𝑃 𝑐𝑜𝑟𝑒 ( 𝑡 )− 𝑃 𝑏𝑎𝑔 )
( 𝑃 𝑐𝑜𝑟𝑒 ( 𝑡 0
)− 𝑃 𝑏𝑎𝑔 ) ( 𝑃 𝑐𝑜𝑟𝑒 ( 𝑡 )+ 𝑃 𝑏𝑎𝑔 )
) (5-7)
The values of the constants used in Eq. (5-7) are provided in Table 5.1.
Table 5.1: Parameters used in Eq. (5-7).
Name Units V alue Description
K zz m
2
[computed] Transverse air permeability
t s [measured] Time
L m 1e-3 Travel length (skin thickness)
µ Pa·s 1.85e-5 Air viscosity
V core m
3
1.1061e-4 Core volume
A m
2
5.806e-3 In-plane area
P core Pa [measured] Core pressure
P bag Pa [measured] V acuum bag pressure
Eq. (5-7) can also be expressed as follows, solved for Pcore [81]:
𝑃 𝑐𝑜𝑟𝑒 ( 𝑡 )= 𝑃 𝑏𝑎𝑔 𝑐 + 𝑒 −𝑠𝑡
𝑐 − 𝑒 −𝑠𝑡
(5-8)
where c and s are:
𝑐 =
𝑃 𝑐𝑜𝑟𝑒 ( 𝑡 0
)+ 𝑃 𝑏𝑎𝑔 𝑃 𝑐𝑜𝑟𝑒 ( 𝑡 0
)− 𝑃 𝑏𝑎𝑔 𝑠 =
𝐾𝐴 𝑃 𝑏𝑎𝑔 𝐿𝜇 𝑉 𝑐𝑜𝑟𝑒 (5-9)
For each elevated-temperature core evacuation test, a least-squares curve fitting method
(in Matlab 2017b) was used to find the best-fit constant K value that minimized the error between
138
the data and the predicted core pressure using Eq. (5-8). An example for one such fit is shown in
Figure 5.18.
Figure 5.18: Measured pressure decay curve for a core evacuation test at 90 ºC and 300 kPa of compaction pressure,
and the corresponding best-fit constant-K model using Eq. (5-8).
Then, the resulting K values were mapped against the corresponding resin viscosity and
compaction pressure of each test, and a polynomial surface was fit (again using a least-squares
method) to the data. The simplest (i.e., lowest-order) polynomial expression that was found to give
satisfactory results was of the following form:
log
10
𝐾 = 𝐶 00
+ 𝐶 10
𝑋 + 𝐶 01
𝑌 + 𝐶 20
𝑋 2
+ 𝐶 11
𝑋𝑌 (5-10)
where K is the facesheet permeability (in m
2
), X is the base-10 logarithm of the resin viscosity (in
log(Pa·s)), and Y is the compaction pressure (in units of bar). The values of the Cxx fitting constants
are provided in Table 5.2. The surface described by Eq. (5-10) is shown in Figure 5.19, along with
the points used to find the values of the fitting constants.
139
Table 5.2: Parameters used in Eq. (5-10).
Name V alue
K Output: permeability [m
2
]
X Input: log 10(viscosity) [log 10(Pa·s)]
Y Input: compaction pressure (Pauto – Pbag) [bar]
C 00 -15.0954
C 10 0.3446
C 01 0.0358
C 20 -0.3069
C 11 -0.1013
Figure 5.19: Best-fit K values for each core evacuation test (red markers), and the overall best-fit polynomial surface
of Eq. (5-10).
Utilizing the “constant K” modeling approach consists of simply inputting the current resin
viscosity and compaction pressure values into Eq. (5-10) to obtain a permeability K, and then using
Eq. (5-8) to predict Pcore as a function of time. Examples of model/experiment comparisons are
shown in Figure 5.20 and Figure 5.21.
140
Figure 5.20: Model/experiment comparisons for core evacuations with 200 kPa of compaction pressure at 50 ºC
(blue), 70 ºC (orange), and 90 ºC (yellow).
Figure 5.21: Model/experiment comparisons for core evacuations at 70 ºC and 4 compaction pressures.
The model predictions show satisfactory agreement with the observed trends of (1)
decreasing K with increasing compaction, and (2) decreasing K with increasing resin viscosity.
The former trend holds true for dry fiber beds as well, and has been studied extensively by other
141
researchers. A common expression to capture the effect of fiber bed compaction on permeability
is the Kozeny-Carman equation [105]:
𝐾 𝑧𝑧
=
𝑟 𝑓 2
4𝑘 ( 1 − 𝑣 𝑓 )
3
𝑣 𝑓 2
(5-11)
where rf is the fiber radius, vf is the fiber volume fraction, and k is the Kozeny constant. Note that
the model used here, Eq. (5-10), includes compaction pressure instead of fiber volume fraction as
an input, but an equivalence could easily be determined from a fiber bed compaction/stress curve,
to cast Eq. (5-10) in terms of vf instead of compaction pressure.
The main limitation with existing permeability/compaction models such as Kozeny-
Carman is that they only capture the behavior of a single fluid phase within a porous medium. As
shown in the experiments earlier in this chapter, when considering the permeability of prepregs to
transverse gas flow, the presence of the second fluid phase – the liquid resin – plays a critical role.
Figure 5.19 shows that, for this particular prepreg, the influence of the liquid resin phase was in
fact more significant than the effects of fiber bed compaction. The permeability varied by less than
one order of magnitude over the range of tested pressures at any given resin viscosity, but at each
of the tested pressures, the variations in permeability with different resin viscosities exceeded two
orders of magnitude.
Although the modeling approach used here did capture the effects of both compaction and
resin viscosity, Eq. (5-10) is of limited utility because it is purely empirical, rather than physics-
based. Errors could arise if the equation is applied outside the range of tested conditions: at
extremely high resin viscosities (room temperature) for example, the model predicts a low but
nonzero permeability, whereas experiments have shown that gas flow can halt entirely (i.e.,
permeability can drop to zero) and unpredictably. Furthermore, the Eq. (5-10) model could result
142
in incorrect predictions if changes are made to the sample configuration and/or test procedures
(e.g., if a film adhesive were included, or if the plies were not fully pre-consolidated). These
limitations motivated the development of a physics-based model, described in the following
section.
5.4. Modeling – two-phase approach
The following model framework provides a one-dimensional description of gas transport
through a prepreg facesheet (through-thickness only), which accounts for the presence of the liquid
resin phase. The fiber bed is assumed to be pre-compacted and perfectly rigid, leading to an
embodiment of the model that does not capture compaction-related phenomena. Compaction of
prepregs has been modeled previously [106,107], and further refinement of the model could
include coupling with compaction behavior. The current form of the model is intended to describe
gas transport behavior depending on the resin viscosity and fiber bed saturation.
The mathematical framework was adapted from the work of Brooks & Corey [108], who
studied the partial desaturation of sands and soils during draining processes (in which air replaces
water within the porous medium). They described functional relationships between the saturation,
the liquid and gas pressures, and the relative permeabilities for both fluids. In this work, the flow
of gas through a prepreg facesheet is treated as an analogous scenario. We consider a woven fiber
bed (instead of sand/soil) and our wetting fluid is resin (instead of water), but the relevant physics
are the same.
5.4.1. Model formulation
The model consists of a pair of coupled partial differential equations (PDEs) that describe
liquid saturation S and gas pressure Pg over one spatial dimension z (through the facesheet
143
thickness) and over time t. The same schematic representation that was used for the empirical
“constant K” model (Figure 5.17) applies for the two-phase model.
Constitutive equations
First, the permeability K of a porous medium (as it appears in the typical Darcy’s law Eq.
(5-1)), when fully saturated by only a single fluid, is defined as the intrinsic permeability Ki. Ki is
a property of the porous medium, therefore it should, theoretically, have the same value
irrespective of the fluid used characterize the porous medium
2
. Next, we define the effective
permeability Ke as the observed permeability of the porous medium to one phase (liquid or gas)
given the presence of both phases. We assume that the fluids are immiscible, that gas can only
flow through desaturated pores, and that liquid can only flow through saturated pores, therefore
Ki ≥ Ke. The relative permeability Kr is a “knockdown factor” that relates the effective permeability
to the intrinsic permeability: Kr = Ke/Ki. Equivalently, using the subscripts g to denote the gas
phase and w to describe the wetting (liquid) phase:
𝐾 𝑒𝑤
= 𝐾 𝑖 𝐾 𝑟𝑤
(5-12)
𝐾 𝑒𝑔
= 𝐾 𝑖 𝐾 𝑟𝑔
(5-13)
By writing the effective permeability as a product of intrinsic and relative permeabilities,
we can separate the effects of the fiber bed from those of the resin (liquid phase) on the effective
gas permeability. While Ki is determined entirely by the fabric architecture, Krw and Krg are
functions of the saturation of the porous medium.
2
Strictly speaking, to account for “slip flow” of rarefied gases – a phenomenon known as the Klinkenberg effect [116]
– a correction factor must be applied, however, neglecting this effect was found to have negligible impact on the model
output.
144
The saturation S is defined as the fraction of pore space that is filled with liquid, and thus
0 ≤ S ≤ 1. In a drainage process, as saturation decreases over time, typically Kew reaches zero and
Keg reaches a maximum before S reaches zero. That is, there typically exists a residual saturation
Sr > 0. To keep the range of our saturation variable between 0 and 1, we use a change of variables
to define an effective saturation Se:
𝑆 𝑒 =
𝑆 − 𝑆 𝑟 1 − 𝑆 𝑟 (5-14)
Brooks & Corey used an analysis from Burdine [109] for equations relating effective
saturation to relative permeability:
𝐾 𝑟𝑤
= 𝑆 𝑒 2
∫
1
𝑃 𝑐 2
𝑑 𝑆 𝑒 𝑆 𝑒 0
∫
1
𝑃 𝑐 2
𝑑 𝑆 𝑒 1
0
(5-15)
𝐾 𝑟𝑔
= ( 1 − 𝑆 𝑒 )
2
∫
1
𝑃 𝑐 2
𝑑 𝑆 𝑒 1
𝑆 𝑒 ∫
1
𝑃 𝑐 2
𝑑 𝑆 𝑒 1
0
(5-16)
The above equations can be solved analytically if a suitable expression relating effective
saturation Se to capillary pressure Pc can be written. According to Eq. (5-3) (see section 5.2.2.1),
to desaturate a particular pore (that is initially saturated), Pg must exceed Pw + Pc, where 𝑃 𝑐 =
𝜎 𝑅 ⁄ . Consequently, pores of smaller radius R require greater gas pressures to desaturate. The
distribution of pore sizes depends on the porous medium (solid phase).
To determine the relationship between Pc and Se, Brooks & Corey measured the capillary
pressure head as a function of saturation for various porous materials (Figure 5.22). They plotted
the effective saturation against the capillary pressure head on a log-log chart (shown in Figure
5.23) and found that the data fit an expression of the following form:
145
𝑆 𝑒 = (
𝑃 𝑏 𝑃 𝑐 )
𝜆 for 𝑃 𝑐 ≥ 𝑃 𝑏
(5-17)
The “bubbling pressure” Pb represents the minimum threshold capillary pressure required to
desaturate the largest pore, and corresponds to the points in Figure 5.23 where the data first
diverges from Se = 1. The constant λ is the “pore size distribution index” and is found from the
slope of the lines connecting the data points in Figure 5.23. Large values of λ correspond to a
porous medium with uniformly sized pores, for which desaturation occurs rapidly once Pc > Pb
(e.g., dataset (2) in Figure 5.22). Conversely, a porous medium with small λ values has a wider
pore size distribution and thus desaturation occurs more gradually (e.g., dataset (4) in Figure 5.22).
Figure 5.22: Capillary pressure head versus saturation for various porous materials, from [108].
146
Figure 5.23: Saturation versus capillary pressure head for various porous materials, from [108].
Substituting Eq. (5-17) into Eq. (5-15) and Eq. (5-16) and simplifying yields:
𝐾 𝑟𝑤
= 𝑆 𝑒 2+3𝜆 𝜆
(5-18)
𝐾 𝑟𝑔
= ( 1 − 𝑆 𝑒 )
2
(1 − 𝑆 𝑒 2+𝜆 𝜆 ) (5-19)
or, equivalently:
𝐾 𝑟𝑤
= (
𝑃 𝑏 𝑃 𝑐 )
2+3𝜆
(5-20)
𝐾 𝑟𝑔
= [1 − (
𝑃 𝑏 𝑃 𝑐 )
𝜆 ]
2
[1 − (
𝑃 𝑏 𝑃 𝑐 )
2+𝜆 ]
(5-21)
147
Figure 5.24 shows the relationship between Pc and Krg for Pb = 10 kPa and various values of λ. For
Pc ≤ Pb the gas permeability is zero. As Pc increases, a fraction of the pores open, creating a
permeable network of gas flow channels. The desaturation occurs abruptly for larger values of λ,
and more gradually (over a wider range of capillary pressures) for smaller values of λ.
Figure 5.24: Graphical representation of Eq. (5-21) for P b = 10 kPa and various values of λ.
Conservation equations
The governing equation for the gas phase is formulated by combining the continuity
equation for a compressible fluid in a porous medium with Darcy’s law and the ideal gas law:
𝜕 𝜕𝑡
[( 1 − 𝑆 𝑒 ) 𝑃 𝑔 ] − (
𝐾 𝑖 𝜙 𝜇 𝑔 )
𝜕 𝜕𝑧
[𝐾 𝑟𝑔
𝑃 𝑔 𝜕 𝑃 𝑔 𝜕𝑧
] = 0 (5-22)
where μg is the viscosity of the gas, and ϕ is the fractional pore space of the porous medium.
148
The governing equation for the liquid phase is known as the Richards equation [110]. As
with Eq. (5-22), it is derived from the continuity equation (assuming incompressibility) and
Darcy’s law:
𝜕 𝑆 𝑒 𝜕𝑡
− (
𝐾 𝑖 𝜙 𝜇 𝑤 )
𝜕 𝜕𝑧
[𝐾 𝑟𝑤
(
𝜕 𝑃 𝑔 𝜕𝑧
−
𝜕 𝑃 𝑐 𝜕𝑧
)] = 0 (5-23)
where µ w is the viscosity of the wetting phase (resin). The substitution Pw = Pg – Pc (i.e., Eq. (5-3))
is made for the liquid phase pressure.
Boundary and initial conditions
The initial conditions for Se were set to 99% over the entire spatial domain, to represent a
fully saturated facesheet while ensuring stability at the beginning of the simulation. Initial
conditions for Pg were set to 101 kPa to represent an initially un-evacuated core and facesheet.
The bag-side (z = L) saturation boundary was set to 99 % (to approximate an excess of
resin at the upper prepreg surface) and the core side boundary (z = 0) used a zero-slope condition
(i.e., no resin flow into the core). The bag-side gas pressure boundary condition was set to 0 (i.e.,
full vacuum in the bag). For the core-side pressure boundary condition, the mass flux out of the
honeycomb core reservoir was set equal to the mass flux into the porous medium, resulting in the
following boundary condition at z = 0:
𝑉 𝑐𝑜𝑟𝑒 𝜕 𝑃 𝑔 𝜕𝑡
= ±
𝐾 𝑟𝑔
𝐾 𝑖 𝐴 𝜇 𝑔 [( 1 − 𝑆 𝑒 ) 𝑃 𝑔 ]
𝜕 𝑃 𝑔 𝜕𝑧
(5-24)
The nominal set of conditions used here represents a simplified implementation of the
model. Boundary and initial conditions can be modified to represent various scenarios, in terms of
sample configuration and process conditions.
149
Dimensionless analysis
To simplify the analysis, the problem is recast in terms of dimensionless variables.
Defining L as the length of the domain (i.e., the thickness of the facesheet), Pi as the initial gas
pressure, Pf as the final gas pressure (i.e., Pf = Pbag ≈ 0 for a core evacuation with full vacuum in
the bag), and ΔP = Pi – Pf, the expressions for dimensionless position, pressure, and time are as
follows:
𝑧 ̂=
𝑧 𝐿 0 ≤ 𝑧 ̂≤ 1 (5-25)
𝑃 ̂
=
𝑃 − 𝑃 𝑓 ∆𝑃 0 ≤ 𝑃 ̂
≤ 1
(5-26)
𝑡 ̂
=
𝑡 𝑡 𝑐 0 ≤ 𝑡 ̂
≤ ∞ (5-27)
The dimensionless form of Eq. (5-23) becomes:
𝜕 𝑆 𝑒 𝜕 𝑡 ̂
− (
𝑡 𝑐 ,𝑤 𝐾 𝑖 ∆𝑃 𝜙 𝜇 𝑤 𝐿 2
)
𝜕 𝜕 𝑧 ̂
[𝐾 𝑟𝑤
(
𝜕 𝑃 ̂
𝑔 𝜕 𝑧 ̂
−
𝜕 𝑃 ̂
𝑐 𝜕 𝑧 ̂
)] = 0 (5-28)
The characteristic time constant for the wetting phase tc,w is chosen to set the coefficient on the
spatial derivative term to unity:
𝑡 𝑐 ,𝑤 =
𝜙 𝜇 𝑤 𝐿 2
𝐾 𝑖 ∆𝑃 (5-29)
Eq. (5-29) then becomes:
𝜕 𝑆 𝑒 𝜕 𝑡 ̂
−
𝜕 𝜕 𝑧 ̂
[𝐾 𝑟𝑤
(
𝜕 𝑃 ̂
𝑔 𝜕 𝑧 ̂
−
𝜕 𝑃 ̂
𝑐 𝜕 𝑧 ̂
)] = 0 (5-30)
The dimensionless form for the gas phase equation contains leading coefficients and a
characteristic time, analogous to Eq. (5-28) and (5-29) for the liquid phase. If, instead, we use Eq.
150
(5-29) as the time constant in the gas phase equation, we obtain a new dimensionless parameter:
the ratio of fluid viscosities μw/μg.
𝜕 𝜕 𝑡 ̂
[( 1 − 𝑆 𝑒 ) 𝑃 ̂
𝑔 ] − (
𝜇 𝑤 𝜇 𝑔 )
𝜕 𝜕 𝑧 ̂
[𝐾 𝑟𝑔
(𝑃 ̂
𝑔 +
𝑃 𝑓 ∆𝑃 )
𝜕 𝑃 ̂
𝑔 𝜕 𝑧 ̂
] = 0 (5-31)
The reservoir boundary condition Eq. (5-24) becomes:
𝜕 𝑃 ̂
𝑔 𝜕 𝑡 ̂
= − (
𝜙𝐿
𝐿 𝑐𝑜𝑟𝑒 𝜇 𝑤 𝜇 𝑔 ) 𝐾 𝑟𝑔
[( 1 − 𝑆 𝑒 ) 𝑃 ̂
𝑔 ]
𝜕 𝑃 ̂
𝑔 𝜕 𝑧 ̂
(5-32)
From this nondimensionalization process, it is revealed that this problem has three
parameters that control the behavior. They are (1) the characteristic time Eq. (5-29), (2) the ratio
of fluid viscosities μw/μg, and (3) the ratio of the facesheet pore volume to the reservoir (core)
volume: 𝜙𝐿 /𝐿 𝑐𝑜𝑟𝑒 (where Lcore = Vcore/A).
5.4.2. Coupled solution
The problem was solved numerically in Matlab using a method known as IMPES (implicit
pressure, explicit saturation) [111]. An example of the output from the first version (V1) of the
code is shown in Figure 5.25. The left column shows the computed pressures (gas, resin, and
capillary), saturation (including the “gas saturation” (1 – S)), and relative permeabilities over the
thickness of the facesheet at one moment in time. These quantities were computed at each time
step, and, as shown in the right column, values for Pcore, average Keg, and average Se were plotted
over time.
A parametric study of the two-phase model (V1) was conducted to gain an understanding
of the various model parameters’ influences on the predicted behavior. The model exhibited trends
consistent with expectations (e.g., slower core evacuations with increasing resin viscosity).
151
Figure 5.25: An example of the model V1 output, over the spatial domain at t = 100 s (left), and over time (right).
Next, the model output (Pcore(t), or, equivalently, Pg(t) at z = 0) was compared to
experimental data to determine the values of the fitting constants: Ki, Pb, and λ. The code used to
find the best-fit model constants consisted of three Matlab scripts. The first was a function for
solving the non-dimensional model, which took as inputs the necessary set of model parameters
and a stop condition (i.e., the number of time steps to compute). This function output non-
dimensional time and core pressure vectors. A second function was used to (1) determine the
number of non-dimensional time steps needed to match the length of an experimental dataset for
a given set of model parameters, (2) call the non-dimensional solving function, (3) convert the
non-dimensional solution to practical units of pressure and time, and (4) interpolate the solution at
the times corresponding to the measured data points. The output of this function was the modeled
values of core pressure, which could be directly compared to measured core pressure data. This
152
function was used in the third script, which contained a nonlinear least-squares solver function to
find the values of Ki, Pb, and λ that minimized the error between measured and modeled values.
The results from the parametric study and model/experiment comparisons are not shown
here because they contain a known issue: after using the model for ~6 months, a sign error was
discovered in one of the second derivative terms of the IMPES finite difference method. Correcting
this error led to V2 of the code, which suffered from stability issues. Due to the extremely high
viscosity ratio μw/μg (which is on the order of 10
10
at room temperature), extremely small time
steps were necessary to ensure a stable solution, which resulted in impractically-long execution
times for the code.
Presently, the code is being reworked by the researcher who initially proposed this two-
phase modeling approach (Thomas Cender from the University of Delaware), in an effort to find
an alternate numerical method that can solve the problem within a reasonable amount of time.
Once a practical method has been selected, a “V3” of the non-dimensional model-solving function
will be implemented. The other two Matlab scripts, which only call the non-dimensional solver,
should be usable without further modifications.
5.5. Conclusions
In this chapter, we have shown experimental results that demonstrate the need for a
multiphase approach to model the air permeability of prepregs. Whereas the permeability of dry
fiber beds is purely a function of the porous medium, the gas permeability of prepregs is also
affected by a second fluid phase: the resin. The resin occupies some fraction of the pore space,
impeding gas flow, and the rate at which air can displace resin to create gas flow channels is limited
153
by the resin viscosity. Furthermore, surface tension effects (i.e., capillary pressure) can prevent
gas flow entirely despite the existence of a pressure gradient.
An empirical model for prepreg permeability was developed and its limitations discussed.
Although the model was capable of capturing trends in the experimental data, its range of
applicability is restricted to the particular materials and configuration used in those experiments.
Furthermore, this simple model cannot capture path-dependent phenomena (e.g., the transient
formation of flow channels, which causes delays in core evacuations at lower temperatures).
A 1-D model was presented for two-phase flow in (initially) fully-saturated woven prepreg.
However, further work is required to determine an appropriate numerical solution method and
appropriate model parameters for the relevant range of material configurations and process
conditions. The model should, in theory, capture the effect of resin viscosity on gas permeability
(i.e., less gas flow at higher resin viscosities due to slower desaturation), as well as path
dependencies. Although the model does not explicitly account for facesheet compaction (which
can strongly affect the facesheet permeability), changes in facesheet compaction can be described
using the existing mathematical framework, for example by defining the intrinsic permeability Ki
as a function of compaction, rather than a single constant. The parameters λ and Pb may also depend
on the compaction state of the fiber bed.
Additional modifications could be made to the model formulation, such as adding a second
or third spatial dimension, including an anisotropic description of the fiber bed permeability, and
including a dual-scale description of the fiber bed porosity (inter-tow macro-pores and intra-tow
micro-pores). However, there is often a trade-off between complexity and utility; adding
complexity may require extensive additional material characterization, while only providing
154
marginal improvements in the accuracy of predictions. The appropriate degree of complexity
depends on the desired model outputs, and remains to be determined.
Once completed, the gas transport model will be coupled with other sub-models to form an
integrated process model for autoclave co-cure of sandwich structures [112]. The integrated model
will account for additional factors that influence core pressure (e.g., heating/cooling-induced
pressure changes and moisture desorption from organic core material), and the computed core
pressure will feed into a diffusion-based sub-model for predicting void growth in the bond line
[113]. Predicting facesheet voids is more complicated: although this gas transport sub-model
includes a description of the desaturated space within the facesheet (i.e., the final value of (1 – Se)
after gelation corresponds to the cured facesheet porosity), the assumption of a rigid porous
medium does not allow a description of the transfer of pressure from the autoclave to the resin.
Therefore, a compaction sub-model is also needed, which describes load sharing between the resin
and fiber bed, and the resulting squeeze flow of resin (and gas).
155
CHAPTER 6. Conclusions and future work
This chapter summarizes the outcomes and knowledge gained from the work presented in
the previous chapters, and suggests additional studies that could be performed in continuation of
this work. The conclusions and recommendations are grouped by project: the RTM project first
(Chapters 2 and 3), and then the co-cure project (Chapters 4 and 5). Lastly, some comments on the
broader implications and significance of this work are provided.
6.1. RTM project
6.1.1. Conclusions
1. In situ process analysis equipment: A lab-scale RTM system was designed and built, which
enabled in situ visualization of in-mold phenomena in realistic processing conditions. This
tool was critical to the project because it enabled identification of the relevant surface
porosity formation mechanism and characterization of the time-dependent evolution of
surface porosity.
2. Defect mechanism: The mechanism causing the defects that tended to form in RTM
composites made using this resin was identified. This “volatile-induced surface porosity”
was similar in nature to shrinkage-induced defects that can occur with other resins (e.g.,
matrix micro-cracking, or surface roughness from premature tool/part separation), but due
to the resin volatility, manifested as bubble growth rather than cracks or surface roughness.
This defect type was particularly challenging to diagnose using ex situ methods, because
the porosity in cured parts was, at a glance, indistinguishable from porosity due to air
trapped during the injection process. It was shown that this volatile-induced surface
porosity occurred due to the interplay of multiple factors: (1) the resin reactivity and the
156
molding tool’s thermal gradients resulted in large local differences in the advancement of
cure, where gelation in hotter zones acted to isolate the still-liquid resin in colder zones
from the external supply pressure; (2) the resin cure shrinkage caused the mold cavity
pressure to drop and tensile stresses to eventually develop; and (3) the resin volatility
caused bubble nucleation and growth in the colder zones once the pressure fell below a
threshold value.
3. Process requirements: The process parameters required for fabrication of defect-free RTM
composites with this resin were quantified. The critical pressure required to prevent volatile
release was found to be ~200 kPa (absolute) over the range of processing temperatures. A
critical mechanical state was identified (in terms of degree of cure α and in terms of
complex viscosity |η*|), prior to which the critical pressure must be maintained if volatile
release is to be prevented. A two-stage cure temperature cycle was shown to be more
effective at meeting these process requirements than a single-stage cure cycle, by allowing
the cure of resin in colder zones to advance gradually, reducing the detrimental effect of
thermal gradients. An increased catalyst loading was shown to reduce surface porosity
formation further: whereas the onset of cure for low-catalyst formulations occurred only
once high temperatures were reached (leading to abrupt cure and large spatial gradients
soon after the onset of cure), the lower-temperature onset of cure for higher catalyst
formulations allowed the entire “cure envelope” to advance further during intermediate-
temperature dwells.
4. Thermal pressure control: A strategy was identified for maintaining mold cavity pressure
without having to change the materials or processing equipment. This “trick” of using
thermal expansion of the resin to counteract cure shrinkage enabled the fabrication of parts
157
with nearly perfect surface finish. The requirements for a cure temperature cycle that
leverages this “thermal pressure control” strategy are: (1) dwell at an intermediate
temperature until the cavity pressure starts to deviate from the external supply pressure,
indicating that the resin has gelled and cavity pressure can no longer be controlled via the
inlet line; and (2) heat the tool to the final cure temperature, using a ramp rate that balances
the rate of thermal expansion with the rate of cure shrinkage to maintain cavity pressure as
long as possible. This ramp rate, as well as the time and temperature of the intermediate
dwell, must have the correct value. An intermediate dwell that’s too long (for the given
temperature) will result in a pressure drop and porosity formation prior to the ramp, while
a dwell that’s too short will not create the necessary pressure increase during the ramp. A
ramp rate that’s too slow will not prevent the shrinkage-induced pressure drop, while a
ramp that’s too fast could create dangerous pressure spikes that can damage the tool. Some
acceptable cure cycles were found experimentally in this work, but a process model would
be needed to predict optimal cure cycles.
5. Dilatometer: A “hybrid dilatometer/RTM” device was designed and built to measure the
volumetric changes (cure shrinkage and thermal expansion) of this resin and its composites.
Although the results were never used due to external factors, the device functioned largely
as intended, and remains available at the M. C. Gill Composites Center for future use.
The final learning outcome of this project was more of a general life lesson: “Sometimes
the best way to solve your problem is to change the problem you’re trying to solve.” Despite our
valiant efforts at troubleshooting the processing of this resin, the manufacturer decided to halt
production and focus on alternative products. The requirements for defect-free part fabrication
were deemed too challenging, and a more forgiving material was desired. Indeed, most of the
158
issues considered here wouldn’t arise when using a resin without such high volatile content. For
this particular resin, however, the volatile species (ethyl acetate) was a necessary additive for
achieving a sufficiently low viscosity to enable resin injection/infusion. An alternate version of
this resin was formulated without ethyl acetate [114], but the viscosity was too high for practical
use in LCM processes.
6.1.2. Recommendations for future work
If this resin were still commercially available (or, for analysis of other resins with similarly
high volatility), the logical next step would be to characterize the thermal and chemical strains, as
well as the evolving elastic modulus during cure. These properties could be used in a process
model to predict mold cavity pressure, and, if combined with volatile release criteria, to predict
the formation of volatile-induced porosity. Figure 6.1 summarizes the various components this
model would consist of (assuming a 1-D model, with the temperature gradient primarily through-
thickness). The white boxes represent methods, properties, and models that were used in this work,
and yellow boxes represent those that would still be needed. Such a process model could be used
with an optimization algorithm, to determine cure temperature cycles that avoid porosity formation
by maximizing the degree of cure at the time of the pressure drop.
For complex-shaped parts and molding tools with 3-D temperature gradients, a 3-D process
simulation would be required. Commercially available software already exists for simulating
injection processes in 3-D, and for simulating the development of residual stresses after gelation.
To properly capture the defect formation behavior of this resin, an integrated simulation would be
ideal, which models both the injection and curing phases of the RTM process. The “gelation
boundary” discussed in Chapter 2 could be used to model the transition between phases (i.e.,
159
describe pressure and flow in the liquid regions, and viscoelastic stresses for the solid regions,
delineated using the moving boundary where α = αgel).
Figure 6.1: Conceptual flowchart of the components of a process model for predicting volatile-induced surface
porosity.
Of course, for a manufacturer who simply wants to produce high-quality parts, the “easy”
solution is to use either a different resin (with lower volatility) or a different process. Alternative
LCM processes exist that sacrifice dimensional tolerances for the ability to maintain resin pressure
by using flexible tooling materials. One example is resin infusion (VARTM), in which a
conformable vacuum bag film maintains consolidation pressure, but this pressure is limited to a
maximum of 1 atmosphere (which would be insufficient for the resin used in this work). Another
option is the “trapped rubber molding” process, in which an elastomer insert is included in the
160
cavity of a 2-part tool. The high CTE of the rubber creates pressure within the mold cavity, which
can be maintained despite resin cure shrinkage if the tool and insert are designed properly.
Lastly, the tools developed for this project could be used in future projects. The lab-scale
RTM enables visualization of processes such as void growth under controlled temperature and
pressure conditions, but isn’t strictly limited to resin transfer molding. The HD/RTM can be used
to measure volumetric changes for other materials as well, and is also not limited to RTM processes
(e.g., it could be used to study compaction of dry fibers and/or prepregs).
6.2. Co-cure project
6.2.1. Conclusions
1. Visualization tool: A lab-scale autoclave was designed and built, which enabled in situ
visualization of the skin/core bond-line in realistic co-cure processing conditions. This tool
is, to the best of my knowledge, the first of its type. By using a design in which the molding
tool and autoclave pressure vessel are a single component, we were able to shed light on
the previously unobservable evolution of the bond-line during autoclave co-cure. This tool
was critical to the overall goal of the project – to develop a physics-based process model
for co-cure – because it enabled identification of the relevant process phenomena and
characterization of the bond-line evolution over time.
2. Defect mechanisms: Visual evidence was obtained for the behavior of multiple co-cure-
specific defect formation mechanisms. It was shown that the gas pressure in the honeycomb
core cells has a significant effect on bond-line void growth. Furthermore, the bond-line
quality is affected not only by the film adhesive, but also by the prepreg resin, which can
mingle with the adhesive at the bond-line. For the materials used, core pressures of ~0.5 –
161
1.5 atmospheres resulted in volatile release and bubble growth from the prepreg resin at
the bond-line. It was also shown that very low core pressures can lead to rapid bubble
growth and subsequent bubble bursting, which may at first seem advantageous, in the sense
that gases can be evacuated (rather than remaining trapped as voids in the cured structure).
However, this bursting action disrupts the fillets by redistributing the adhesive, spattering
it onto the core cell walls where it makes no contribution to the bond-line strength.
3. Process requirements: The features of an “ideal” co-cure scenario were explained and
demonstrated. They are summarized here:
An “equilibrated core” condition (achieved by including cell-to-cell and core-to-
exterior pathways for gas in the configuration of the sandwich structure) is desirable
because it removes uncertainty from the process (normally the core pressure is
unknown and not directly controllable) and because it prevents the possibility of
driving gases through the bag-side facesheet (which can become trapped and contribute
to facesheet porosity).
Air initially entrapped between the facesheet plies must be evacuated. This can be
achieved by pulling full vacuum in the bag at room temperature (but at elevated
temperatures, full vacuum can cause bubble bursting and fillet disruption). An
equilibrated core and reticulated film adhesive can aid room-temperature gas extraction
by allowing gases to escape the facesheet via the core-side boundary.
The gas pressure within the vacuum bag and core must be sufficiently high to prevent
bubble growth at the highest processing temperature. Typically, the bag is vented to
atmospheric pressure during autoclave co-cure, but with the materials used in this work,
atmospheric pressure was insufficient for preventing the release of residual solvent
162
from the prepreg resin. The concept of super-ambient “in-bag pressurization” was
shown to effectively suppress void growth in these materials.
Lastly, the autoclave pressure must consolidate the sandwich structure to an
“appropriate” degree. Excessive pressure can lead to deformation of the honeycomb
core, resin starvation within the facesheets, and facesheet dimpling, while low
autoclave pressure can lead to poor facesheet consolidation and even separation of the
upper facesheet from the core (if the core pressure exceeds the autoclave pressure). The
compaction pressure that acts on the sandwich structure is the difference between
autoclave and bag pressures, therefore, if the bag pressure is adjusted, the autoclave
pressure can be adjusted accordingly to maintain the same level of compaction.
4. Characterization of prepreg permeability: The gas permeability of prepreg facesheets was
experimentally characterized and a physics-based modeling approach was proposed. The
experiments corroborated previously published findings: (1) prepreg gas permeability is
affected by the presence and mobility of the prepreg resin, (2) gas flow can cease entirely
despite the existence of a pressure gradient, (3) compaction acts to reduce permeability,
and (4) facesheet permeability can have a time/path dependence due to the transient
formation (and/or collapse) of gas transport pathways. The in situ data validated
assumptions about the nature of gas transport through prepregs: it was shown that, for
transverse flow in fully-saturated woven prepregs, gases move as discrete bubbles through
the macro-pore “pinholes” in the weave. Finally, a mathematical framework to describe
the relevant physics was proposed.
163
6.2.2. Recommendations for future work
The overall outcome of this project, a physics-based process model for autoclave co-cure,
is still under development, and will continue to be refined by my colleagues at USC and our
collaborators at the University of Delaware. The integrated co-cure model will ultimately include
several sub-models, as shown in Figure 6.2
Figure 6.2: Conceptual flowchart of the integrated co-cure process model. Image source: [112].
Inputs will consist of a process description: materials and part configuration, as well as the
imposed temperature, autoclave pressure, and bag pressure cycles. From these inputs, material
sub-models will compute the evolving material properties (degree of cure, viscosity, etc.). A
“controller” component of the software will collect and distribute data between various process
development sub-models.
The core pressure sub-model will track the changes in the mass of gas in the core cells due
to various sources (volatile release from the core, adhesive, and resin) and sinks (evacuation
through the facesheet), and compute pressure using the ideal gas law. The facesheet consolidation
164
sub-model will compute the fiber volume fraction of the facesheet(s) and the prepreg resin
pressure, based on pressure boundary conditions and compaction of the fiber bed [90].
For facesheet permeability, initially the empirical model (see section 5.3) could be used,
which is extremely simple to implement. The two-phase model could be used in a later release of
the software, but some of the variables overlap with the facesheet consolidation sub-model. For
example, the facesheet consolidation model includes a description of resin pressure due to prepreg
compaction, while the two-phase permeability model also describes resin pressure, but in the
context of capillary effects. A truly “integrated” model would couple these effects in a single
mathematical framework.
The fillet formation sub-model will describe the fillet height and shape, based on contact
angles and the amount of material present at the bond-line. The porosity formation sub-model will
use Fickian diffusion to predict the growth or shrinkage of bubbles within the adhesive and the
prepreg resin [113]. Altogether, the process model will output parameters that predict the “quality”
of the co-cured sandwich structure, such as average fillet geometry and bond-line porosity. The
exact form of these outputs has yet to be determined, and will be based on input from the industrial
partners that will be using this software, once released.
Whether or not the two-phase permeability model is ultimately implemented in the
integrated process model, further work could be done while treating it as a stand-alone model.
First, a computationally-efficient numerical solution method must be found. Then, model
parameters that make the predictions match experimental data could be determined. It may be
necessary to relax some of the simplifying assumptions and describe additional complexities (in
reality the porous medium is highly anisotropic, deformable, and contains dual-scale porosity, yet
none of these features were included in the first embodiment of the model). A model that describes
165
out-of-autoclave prepregs, which have a partially saturated initial microstructure, would require at
least a 2-D description (i.e., in-plane flow could no longer be ignored) and modification of the
initial conditions for the saturation.
Lastly, the lab-scale autoclave tool that was built for this project could be used for many
other studies. Possibilities include co-cure process characterization for other materials, and
additional permeability studies (for different materials, sample configurations, and process
conditions).
6.3. Broader implications and final thoughts
This dissertation includes two specific example of custom-built tools for in situ process
analysis, but, for composite manufacturing by almost any method, a similar approach could be
used to bridge the gap between lab-scale material characterization and industrial-scale part
fabrication. While typical material characterization methods (e.g., DSC, TGA, RDA) provide
critical information about material behavior in tightly-controlled conditions, industrial
manufacturing is oftentimes performed in less-controlled and non-ideal conditions (e.g., using
tools with temperature non-uniformities). In situ process analysis methods are particularly useful
for diagnosis of process-induced defects that occur as a consequence of these non-ideal conditions.
Without a complete understanding of the phenomena causing process-induced defects,
manufacturers often resort to trial-and-error methods, in which they test a large number of different
process conditions and then select the “recipe” with the most favorable result. This practice is time
consuming and costly, and does not guarantee an optimal result (or even an acceptable one).
Through the use of in situ process diagnostics, one can observe the defect formation as it occurs,
remove uncertainty about its cause, and make informed decisions about process modifications. By
166
eliminating the need for trial-and-error, it becomes feasible to troubleshoot realistic and
challenging processes using only a small number of tests to first assess the problem and then
demonstrate a solution.
Altogether, this work provides a practical and applied approach to improving composite
manufacturing techniques, which are inherently complex processes involving reactive multiphase
(solid, liquid, and gas) systems and many concurrent, coupled physical phenomena. Because
production capacity and manufacturing variability are still major challenges facing the composites
industry today, the improvement of manufacturing practices is – and will continue to be – a
critically important goal for engineers in this field.
167
References
[1] Newcomb BA. Processing, structure, and properties of carbon fibers. Compos Part A Appl Sci
Manuf 2016;91:262–82. doi:10.1016/j.compositesa.2016.10.018.
[2] Campbell FC. Manufacturing Processes for Advanced Composites. London: Elsevier Ltd.; 2003.
[3] Kim D, Centea T, Nutt SR. Out-time effects on cure kinetics and viscosity for an out-of-autoclave
(OOA) prepreg: Modelling and monitoring. Compos Sci Technol 2014;100:63–9.
doi:10.1016/j.compscitech.2014.05.027.
[4] Campbell FC. Manufacturing Technology for Aerospace Structural Materials. Elsevier; 2006.
doi:10.1016/B978-1-85617-495-4.X5000-8.
[5] Thorfinnson B, Biermann T. Production of Void Free Composite Parts Without Debulking. Proc.
31st Int. SAMPE Symp., 1986, p. 10.
[6] Thorfinnson B, Biermann T. Degree of Impregnation of Prepregs--Effects on Porosity. Proc. 32nd
SAMPE Int. Symp., 1987.
[7] Grunenfelder LK, Dills A, Centea T, Nutt S. Effect of prepreg format on defect control in out-of-
autoclave processing. Compos Part A Appl Sci Manuf 2017;93:88–99.
doi:10.1016/j.compositesa.2016.10.027.
[8] Hexcel Composites. Honeycomb sandwich design technology. 2000.
[9] Gladysz GM, Chawla KK. Chapter 7: Applications. Voids Mater., Elsevier; 2015, p. 131–56.
doi:10.1016/B978-0-444-56367-5.00007-5.
[10] Michaud V. A Review of Non-saturated Resin Flow in Liquid Composite Moulding processes.
Transp Porous Media 2016;115:581–601. doi:10.1007/s11242-016-0629-7.
[11] Lee DH, Lee W Il, Kang MK. Analysis and minimization of void formation during resin transfer
molding process. Compos Sci Technol 2006;66:3281–9. doi:10.1016/j.compscitech.2005.07.008.
[12] Hamill L, Centea T, Nutt S. Surface porosity during vacuum bag-only prepreg processing: Causes
and mitigation strategies. Compos Part A Appl Sci Manuf 2015;75:1–10.
doi:10.1016/j.compositesa.2015.04.009.
[13] Grunenfelder LK, Nutt SR. Void formation in composite prepregs – Effect of dissolved moisture.
Compos Sci Technol 2010;70:2304–9. doi:10.1016/j.compscitech.2010.09.009.
[14] Summerscales J. In-process monitoring for control of closed mould techniques for the manufacture
of thermosetting matrix composites. Adv. Polym. Mater. - Struct. Prop. Relationships, 2003, p. 1–
27.
[15] Kim D, Centea T, Nutt SR. In-situ cure monitoring of an out-of-autoclave prepreg: Effects of out-
time on viscosity, gelation and vitrification. Compos Sci Technol 2014;102:132–8.
doi:10.1016/j.compscitech.2014.07.027.
[16] Centea T, Hubert P. Out-of-autoclave prepreg consolidation under deficient pressure conditions. J
Compos Mater 2014;48:2033–45. doi:10.1177/0021998313494101.
[17] Govignon Q, Bickerton S, Morris J, Kelly PA. Full field monitoring of the resin flow and laminate
properties during the resin infusion process. Compos Part A Appl Sci Manuf 2008;39:1412–26.
168
doi:10.1016/j.compositesa.2008.05.005.
[18] Aktas A, Boyd S, Shenoi R. Cure and strain monitoring of novel unsaturated polyester/phenolic
resin blends in the vacuum infusion process using fibre Bragg gratings. J Compos Mater
2015;49:3599–608. doi:10.1177/0021998314568165.
[19] de Oliveira R, Lavanchy S, Chatton R, Costantini D, Michaud V, Salathé R, et al. Experimental
investigation of the effect of the mould thermal expansion on the development of internal stresses
during carbon fibre composite processing. Compos Part A Appl Sci Manuf 2008;39:1083–90.
doi:10.1016/j.compositesa.2008.04.011.
[20] Cender TA, Simacek P, Davis S, Advani SG. Gas Evacuation from Partially Saturated Woven Fiber
Laminates. Transp Porous Media 2016;115:541–62. doi:10.1007/s11242-016-0784-x.
[21] Hu W, Grunenfelder LK, Centea T, Nutt S. In situ monitoring and analysis of void evolution in
unidirectional prepreg. J Compos Mater 2018:002199831875918. doi:10.1177/0021998318759183.
[22] Pupin C, Ross A, Dubois C, Rietsch J-C, Vernet N, Ruiz E. Formation and suppression of volatile-
induced porosities in an RTM epoxy resin. Compos Part A Appl Sci Manuf 2016;94:146–57.
doi:10.1016/j.compositesa.2016.12.006.
[23] Potter KD. The early history of the resin transfer moulding process for aerospace applications.
Compos Part A Appl Sci Manuf 1999;30:619–21. doi:10.1016/S1359-835X(98)00179-1.
[24] Kruckenberg T, Paton R. Resin Transfer Moulding for Aerospace Structures. Springer Netherlands;
1998. doi:10.1007/978-94-011-4437-7.
[25] Pearce N, Guild F, Summerscales J. A study of the effects of convergent flow fronts on the
properties of fibre reinforced composites produced by RTM. Compos Part A Appl Sci Manuf
1998;29:141–52. doi:10.1016/S1359-835X(97)00041-9.
[26] Ruiz E, Achim V, Soukane S, Trochu F, Breard J. Optimization of injection flow rate to minimize
micro/macro-voids formation in resin transfer molded composites. Compos Sci Technol
2006;66:475–86. doi:10.1016/j.compscitech.2005.06.013.
[27] Leclerc JS, Ruiz E. Porosity reduction using optimized flow velocity in Resin Transfer Molding.
Compos Part A Appl Sci Manuf 2008;39:1859–68. doi:10.1016/j.compositesa.2008.09.008.
[28] Liu B, Bickerton S, Advani SG. Modelling and simulation of resin transfer moulding (RTM) - Gate
control, venting and dry spot prediction. Compos Part A Appl Sci Manuf 1996;27:135–41.
doi:10.1016/1359-835X(95)00012-Q.
[29] Lundström T, Gebart B. Influence from process parameters on void formation in resin transfer
molding. Polym Compos 1994;15:25–33. doi:10.1002/pc.750150105.
[30] Pollard HJ, Rees J. Improvements in or relating to the moulding of articles from thermosetting
resins and fibrous material. British Patent Specification 778,683, 1956.
[31] Varna J, Joffe R, Berglund LA, Lundström TS. Effect of voids on failure mechanisms in RTM
laminates. Compos Sci Technol 1995;53:241–9. doi:10.1016/0266-3538(95)00024-0.
[32] Haider M, Hubert P, Lessard L. An experimental investigation of class A surface finish of
composites made by the resin transfer molding process. Compos Sci Technol 2007;67:3176–86.
doi:10.1016/j.compscitech.2007.04.010.
[33] Kardos JL, Duduković MP, Dave R. Void growth and resin transport during processing of
thermosetting — Matrix composites. Adv. Polym. Sci., 1986, p. 101–23. doi:10.1007/3-540-16423-
169
5_13.
[34] Wood JR, Bader MG. Void control for polymer-matrix composites (1): Theoretical and
experimental methods for determining the growth and collapse of gas bubbles. Compos Manuf
1994;5:139–47. doi:10.1016/0956-7143(94)90023-X.
[35] Ledru Y, Piquet R, Schmidt F, Michel L, Bernhart G. Modeling of void growrth mechanisms during
the manufacturing of composite laminates. Proc. 9th Int. Conf. Flow Process. Compos. Mater.,
Montréal, Canada: 2008.
[36] Ledru Y, Bernhart G, Piquet R, Schmidt F, Michel L. Coupled visco-mechanical and diffusion void
growth modelling during composite curing. Compos Sci Technol 2010;70:2139–45.
doi:10.1016/j.compscitech.2010.08.013.
[37] Lundström TS. Measurement of void collapse during resin transfer moulding. Compos Part A Appl
Sci Manuf 1997;28:201–14. doi:10.1016/S1359-835X(96)00109-1.
[38] Afendi M, Banks WM, Kirkwood D. Bubble free resin for infusion process. Compos Part A Appl
Sci Manuf 2005;36:739–46. doi:10.1016/j.compositesa.2004.10.030.
[39] Eom Y, Boogh L, Michaud V, Sunderland P, Månson J-A. Stress-initiated void formation during
cure of a three-dimensionally constrained thermoset resin. Polym Eng Sci 2001;41:492–503.
doi:10.1002/pen.10746.
[40] Eom Y, Boogh L, Michaud V, Månson J. A structure and property based process window for void
free thermoset composites. Polym Compos 2001;22:22–31. doi:10.1002/pc.10512.
[41] Wisnom MR, Gigliotti M, Ersoy N, Campbell M, Potter KD. Mechanisms generating residual
stresses and distortion during manufacture of polymer-matrix composite structures. Compos Part A
Appl Sci Manuf 2006;37:522–9. doi:10.1016/j.compositesa.2005.05.019.
[42] Merzlyakov M, McKenna GB, Simon SL. Cure-induced and thermal stresses in a constrained epoxy
resin. Compos Part A Appl Sci Manuf 2006;37:585–91. doi:10.1016/j.compositesa.2005.05.013.
[43] Haider M, Hubert P, Lessard L. Cure shrinkage characterization and modeling of a polyester resin
containing low profile additives. Compos Part A Appl Sci Manuf 2007;38:994–1009.
doi:10.1016/j.compositesa.2006.06.020.
[44] Ramos J a., Pagani N, Riccardi CC, Borrajo J, Goyanes SN, Mondragon I. Cure kinetics and
shrinkage model for epoxy-amine systems. Polymer (Guildf) 2005;46:3323–8.
doi:10.1016/j.polymer.2005.02.069.
[45] Li C, Potter K, Wisnom MR, Stringer G. In-situ measurement of chemical shrinkage of MY750
epoxy resin by a novel gravimetric method. Compos Sci Technol 2004;64:55–64.
doi:10.1016/S0266-3538(03)00199-4.
[46] Nawab Y, Shahid S, Boyard N, Jacquemin F. Chemical shrinkage characterization techniques for
thermoset resins and associated composites. J Mater Sci 2013;48:5387–409. doi:10.1007/s10853-
013-7333-6.
[47] Kendall KN, Rudd CD, Owen MJ, Middleton V. Characterization of the resin transfer moulding
process. Compos Manuf 1992;3:235–49. doi:10.1016/0956-7143(92)90111-7.
[48] Boyard N, Millischer a., Sobotka V, Bailleul JL, Delaunay D. Behaviour of a moulded composite
part: Modelling of dilatometric curve (constant pressure) or pressure (constant volume) with
temperature and conversion degree gradients. Compos Sci Technol 2007;67:943–54.
170
doi:10.1016/j.compscitech.2006.07.004.
[49] Landry B, Hubert P. Experimental study of defect formation during processing of randomly-
oriented strand carbon/PEEK composites. Compos Part A Appl Sci Manuf 2015;77:301–9.
doi:10.1016/j.compositesa.2015.05.020.
[50] Landry B, Hubert P. Modelling Pressure Distribution During Cooling of Randomly-Oriented Strand
Carbon/PEEK Composites. Proc. 20th Int. Conf. Compos. Mater., Copenhagen, Denmark: 2015.
[51] Pupin C, Ross A, Ruiz E, Billotte C, Rietsch J-C, Bricourt P, et al. Porosity Control by Process
Parameters. Proc. 20th Int. Conf. Compos. Mater., Copenhagen, Denmark: 2015.
[52] Ishida H, Allen DJ. Mechanical characterization of copolymers based on benzoxazine and epoxy.
Polymer (Guildf) 1996;37:4487–95. doi:10.1016/0032-3861(96)00303-5.
[53] Rimdusit S, Tiptipakorn S, Jubsilp C, Takeichi T. Polybenzoxazine alloys and blends: Some unique
properties and applications. React Funct Polym 2013;73:369–80.
doi:10.1016/j.reactfunctpolym.2012.04.022.
[54] Anders M, Lo J, Centea T, Nutt SR. Minimizing Volatile-Induced Surface Porosity in RTM via
Material and Process Optimization. Proc. 20th Int. Conf. Compos. Mater., Copenhagen, Denmark:
2015.
[55] Lo J, Anders M, Centea T, Nutt S. Multi-Scale Material and Process Characterization for Resin
Transfer Molding: Case Study for a Blended Epoxy/Polybenzoxazine Resin. Proc. 20th Int. Conf.
Compos. Mater., Copenhagen, Denmark: 2015.
[56] Lo J, Anders M, Centea T, Nutt SR. The effect of process parameters on volatile release for a
benzoxazine–epoxy RTM resin. Compos Part A Appl Sci Manuf 2016;84:326–35.
doi:10.1016/j.compositesa.2016.01.024.
[57] Winter HH, Mours M. Rheology of Polymers Near Liquid-Solid Transitions. Neutron Spin Echo
Spectrosc. Viscoelasticity Rheol., vol. 134, Berlin, Heidelberg: Springer Berlin Heidelberg; 1997,
p. 165–234. doi:10.1007/3-540-68449-2_3.
[58] Bilyeu B, Brostow W, Menard KP. Separation of gelation from vitrification in curing of a fiber-
reinforced epoxy composite. Polym Compos 2002;23:1111–9. doi:10.1002/pc.10505.
[59] Anders M, Lo J, Centea T, Nutt S. Volatile-Induced Voids in Rtm Processing for Aerospace. 12th
Int. Conf. Flow Process. Compos. Mater., Enschede, NL: 2014.
[60] Centea T, Lo J, Anders M, Nutt SR. Process Modeling for Resin Transfer Molding of a Modified
Heterocyclic Phenolic/Epoxy Blended Resin. ASC 2014 Proc., 2014.
[61] Yousefi A, Lafleur PG, Gauvin R. Kinetic studies of thermoset cure reactions: A review. Polym
Compos 1997;18:157–68. doi:10.1002/pc.10270.
[62] Friedman HL. Kinetics of thermal degradation of char-forming plastics from thermogravimetry.
Application to a phenolic plastic. J Polym Sci Part C Polym Symp 1964;6:183–95.
doi:10.1002/polc.5070060121.
[63] Brauner C, Soprano P, Herrmann a. S, Meiners D. Cure-dependent thermo-chemical modelling
and analysis of the manufacturing process of an aircraft composite frame. J Compos Mater 2014.
doi:10.1177/0021998314527777.
[64] Yang L, Zhang C, Pilla S, Gong S. Polybenzoxazine-core shell rubber-carbon nanotube
nanocomposites. Compos Part A Appl Sci Manuf 2008;39:1653–9.
171
doi:10.1016/j.compositesa.2008.07.004.
[65] Nawab Y, Casari P, Boyard N, Jacquemin F. Characterization of the cure shrinkage, reaction
kinetics, bulk modulus and thermal conductivity of thermoset resin from a single experiment. J
Mater Sci 2013;48:2394–403. doi:10.1007/s10853-012-7026-6.
[66] Grimes GC. The Adhesive-Honeycomb Relationship. Appl Polym Symp 1966;3:157–90.
[67] Campbell FC, Mallow AR, Browning CE. Porosity in carbon fiber composites an overview of
causes. J Adv Mater 1995;26:18–33.
[68] Hsiao H-M, Lee S, Buyny R. Core Crush Problem in the Manufacturing of Composite Sandwich
Structures: Mechanisms and Solutions. AIAA J 2006;44:901–7. doi:10.2514/1.18067.
[69] Pearce PJ, Arnott DR, Camilleri A, Kindermann MR, Mathys GI, Wilson AR. Cause and effect of
void formation during vacuum bag curing of epoxy film adhesives. J Adhes Sci Technol
1998;12:567–84. doi:10.1163/156856198X00795.
[70] Hayes BS, Seferis JC, Edwards RR. Self-adhesive honeycomb prepreg systems for secondary
structural applications. Polym Compos 1998;19:54–64. doi:10.1002/pc.10075.
[71] Yuan C, Li M, Zhang Z, Gu Y. Experimental Investigation on the Co-Cure Processing of
Honeycomb Structure with Self-Adhesive Prepreg. Appl Compos Mater 2008;15:47–59.
doi:10.1007/s10443-008-9056-4.
[72] Grove SM, Popham E, Miles ME. An investigation of the skin/core bond in honeycomb sandwich
structures using statistical experimentation techniques. Compos Part A Appl Sci Manuf
2006;37:804–12. doi:10.1016/j.compositesa.2005.07.005.
[73] Hou T, Baughman JM, Zimmerman TJ, Sutter JK, Gardner JM. Evaluation of Sandwich Structure
Bonding in Out-Of-Autoclave (OOA) Processing. Sampe J 2011;47:32–9.
[74] Nagarajan S, Menta VGK, Chandrashekhara K, Berkel TR, Sha J, Wu P, et al. Out-of-Autoclave
Sandwich Structure: Processing Study. Sampe J 2012;48:24–31.
[75] Butukuri RR, Bheemreddy VP, Chandrashekhara K, Berkel TR, Rupel K. Evaluation of skin-core
adhesion bond of out-of-autoclave honeycomb sandwich structures. J Reinf Plast Compos
2012;31:331–9. doi:10.1177/0731684412437267.
[76] Hadsall KA, Storage KM, Storage TM, Smith JA. Evaluation and comparison of structural film
adhesives for out-of-autoclave (OOA) applications using alternate cure cycles. Proc. SAMPE 2013
Conf., 2013.
[77] Anandan S, Dhaliwal GS, Chandrashekhara K, Berkel TR, Pfitzinger D. Processing and
Performance of Out-of- Autoclave Bismaleimide Composite Sandwich Structures. Proc. Compos.
Adv. Mater. Expo, Dallas, TX: 2015, p. October 26-29.
[78] Tavares SS, Michaud V, Månson J-AE. Through thickness air permeability of prepregs during cure.
Compos Part A Appl Sci Manuf 2009;40:1587–96. doi:10.1016/j.compositesa.2009.07.004.
[79] Tavares SS, Michaud V, Månson J-AE. Assessment of semi-impregnated fabrics in honeycomb
sandwich structures. Compos Part A Appl Sci Manuf 2010;41:8–15.
doi:10.1016/j.compositesa.2009.09.005.
[80] Tavares SS, Caillet-Bois N, Michaud V, Månson J-AE. Non-autoclave processing of honeycomb
sandwich structures: Skin through thickness air permeability during cure. Compos Part A Appl Sci
Manuf 2010;41:646–52. doi:10.1016/j.compositesa.2010.01.013.
172
[81] Tavares SS, Caillet-Bois N, Michaud V, Månson J-AE. Vacuum-bag processing of sandwich
structures: Role of honeycomb pressure level on skin–core adhesion and skin quality. Compos Sci
Technol 2010;70:797–803. doi:10.1016/j.compscitech.2010.01.015.
[82] Sequeira Tavares S, Roulin Y, Michaud V, Månson J-AE. Hybrid processing of thick skins for
honeycomb sandwich structures. Compos Sci Technol 2011;71:183–9.
doi:10.1016/j.compscitech.2010.11.001.
[83] Kratz J, Hubert P. Processing out-of-autoclave honeycomb structures: Internal core pressure
measurements. Compos Part A Appl Sci Manuf 2011;42:1060–5.
doi:10.1016/j.compositesa.2011.04.009.
[84] Kratz J, Hubert P. Anisotropic air permeability in out-of-autoclave prepregs: Effect on honeycomb
panel evacuation prior to cure. Compos Part A Appl Sci Manuf 2013;49:179–91.
doi:10.1016/j.compositesa.2013.02.013.
[85] Kratz J, Hubert P. Vacuum bag only co-bonding prepreg skins to aramid honeycomb core. Part I.
Model and material properties for core pressure during processing. Compos Part A Appl Sci Manuf
2015;72:228–38. doi:10.1016/j.compositesa.2014.11.026.
[86] Kratz J, Hubert P. Vacuum-bag-only co-bonding prepreg skins to aramid honeycomb core. Part II.
In-situ core pressure response using embedded sensors. Compos Part A Appl Sci Manuf
2015;72:219–27. doi:10.1016/j.compositesa.2014.11.030.
[87] Davies P, Choqueuse D, Bourbouze G. Micro-Tomography to Study High-performance Sandwich
Structures. J Sandw Struct Mater 2011;13:7–21. doi:10.1177/1099636209344273.
[88] Centea T, Zebrine D, Anders M, Elkin C, Nutt SR. Manufacturing of Honeycomb Core Sandwich
Structures: Film Adhesive Behavior Versus Cure Pressure and Temperature. Proc Compos Adv
Mater Expo 2016.
[89] Anders M, Lo J, Centea T, Nutt SR. Eliminating volatile-induced surface porosity during resin
transfer molding of a benzoxazine/epoxy blend. Compos Part A Appl Sci Manuf 2016;84:442–54.
doi:10.1016/j.compositesa.2016.02.024.
[90] Centea T, Zebrine D, Anders M, Nutt SR. Facesheet Consolidation During Autoclave Co- Cure of
Honeycomb Core Sandwich Structures. Proc. SAMPE 2017 Conf., Seattle, WA: 2017.
[91] Anders M, Zebrine D, Centea T, Nutt SR. In Situ Observations and Pressure Measurements for
Autoclave Co-Cure of Honeycomb Core Sandwich Structures. J Manuf Sci Eng 2017;139:111012.
doi:10.1115/1.4037432.
[92] Hexcel Corporation. HexPly® 8552 Epoxy Matrix Properties 2013.
[93] Henkel Aerospace. LOCTITE EA 9658 AERO Epoxy Film Adhesive - Technical Process Bulletin
2013.
[94] Mckillen JM, Miao TY, Cornwell RE, Marsicano JM. High Strength and Durability Nacelle Film
Adhesive, Hysol EA9658. Proc SAMPE 2007 Conf 2007.
[95] Reller D, Oldroyd PK. Method and Apparatus for Reticulating an Adhesive to the Surface Network
of a Cellular Core Structure. US 2011/0151183, 2011.
[96] Welsh JM, Popp TD, Baca LL. Method for Forming a Composite Acoustic Panel. US Patent
6,179,943 B1, 2001.
[97] Epstein G, Ruth S. Honeycomb Sandwich Structures: Vented Versus Unvented Designs for Space
173
Systems. El Segundo, CA: 1993.
[98] Alteneder AW, Renn DJ, Seferis JC, Curran RN. Processing and Characterization Studies of
Honeycomb Composite Structures. Proc. 38th Int. SAMPE Symp., Anaheim, CA: SAMPE: Society
for the Advancement of Material and Process Engineers; 1993, p. 1034–47.
[99] Brand RA, Brown GG, McKague EL. Processing Science of Epoxy Resin Composites. GENERAL
DYNAMICS SAN DIEGO CA CONVAIR DIV: 1984.
[100] Anders M, Zebrine D, Centea T, Nutt SR. Process diagnostics for co-cure of sandwich structures
using in situ visualization. Compos Part A Appl Sci Manuf 2019;116:24–35.
doi:10.1016/j.compositesa.2018.09.029.
[101] Van Ee D, Poursartip A. NCAMP Hexply Material Properties Database for use with COMPRO
CCA and Raven. 2009.
[102] Vernet N, Ruiz E, Advani S, Alms JB, Aubert M, Barburski M, et al. Experimental determination
of the permeability of engineering textiles: Benchmark II. Compos Part A Appl Sci Manuf
2014;61:172–84. doi:10.1016/j.compositesa.2014.02.010.
[103] Howard S. Evaluating Consolidation Dynamics - Study of Key Consolidation Parameters in a
Vacuum Bag Only (VBO) Cure. Proc SAMPE 2016 Tech Conf 2016.
[104] Gangloff JJ, Cender TA, Eskizeybek V, Simacek P, Advani SG. Entrapment and venting of bubbles
during vacuum bag prepreg processing. J Compos Mater 2016. doi:10.1177/0021998316676325.
[105] Carman PC. Fluid flow through granular beds. Chem Eng Res Des 1997;75:S32–48.
doi:10.1016/S0263-8762(97)80003-2.
[106] Hubert P, Poursartip A. Method for the direct measurement of the fibre bed compaction curve of
composite prepregs. Compos Part A Appl Sci Manuf 2001;32:179–87. doi:10.1016/S1359-
835X(00)00143-3.
[107] Helmus R, Centea T, Hubert P, Hinterholzl R. Out-of-autoclave prepreg consolidation: Coupled air
evacuation and prepreg impregnation modeling. J Compos Mater 2015;0:1–11.
doi:10.1177/0021998315592005.
[108] Brooks RH, Corey AT. Hydraulic properties of porous media. Hydrol Pap 3 1964;Colorado S.
[109] Burdine NT. Relative Permeability Calculations From Pore Size Distribution Data. J Pet Technol
1953;5:71–8. doi:10.2118/225-G.
[110] Richards LA. CAPILLARY CONDUCTION OF LIQUIDS THROUGH POROUS MEDIUMS.
Physics (College Park Md) 1931;1:318–33. doi:10.1063/1.1745010.
[111] Chen Z, Huan G, Ma Y. Computational Methods for Multiphase Flows in Porous Media, Society
for Industrial and Applied Mathematics; 2006. doi:10.1137/1.9780898718942.
[112] Centea T, Simacek P, Anders M, Kermani NN, Zebrine D, Advani S, et al. UNDERSTANDING
AND MODELING THE CO-CURE OF HONEYCOMB CORE SANDWICH STRUCTURES.
Proc. SAMPE 2019 Conf., Long Beach, CA: 2019.
[113] Kermani NN, Simacek P, Advani SG. A Simple Tool to Simulate Honeycomb Sandwich Structure
Composites During Co-Cure Manufacturing Process. Proc. SAMPE 2019 Conf., Long Beach, CA:
2019.
[114] Lo J, Zhang X, Williams T, Nutt S. Eliminating porosity via reformulation of a benzoxazine–epoxy
174
resin transfer molding resin. J Compos Mater 2018;52:1481–93. doi:10.1177/0021998317727048.
[115] Ishida H, Allen D. Physical and Mechanical Characterization of Near-Zero Shrinkage Pol y
benzoxazines. J Polym Sci Part B Polym Phys n.d.:1019–30.
[116] Klinkenberg LJ. The Permeability Of Porous Media To Liquids And Gases. Drill Prod Pract
1941:14.
175
Design details for lab-scale RTM system
A.1. Technical drawings (all dimension in inches)
Figure A-1: Dimensions for main tool body (sheet 1 of 2).
176
Figure A-2: Dimensions for main tool body (sheet 2 of 2).
177
Figure A-3: Dimensions for "picture frame" spacer plate.
178
Figure A-4: Dimensions for the window's retaining ring.
179
A.2. Stress simulation
The window is a 20 mm thick piece of tempered, heat-resistant borosilicate glass (Pyrex).
It is 170 mm in diameter and rated for a maximum pressure of 1240 kPa (180 psi). An FEA model
was used to estimate stresses and deflections at the maximum rated pressure.
Figure A-5: Stress simulation for the glass window with a resin pressure of 180 psi.
Applying the maximum rated pressure to a 76 × 127 mm area on the underside of the window
generates a maximum stress of 20.5 MPa at the center of the top of the window. Using the
published fracture toughness value for Pyrex with a basic fracture mechanics calculation, this
maximum stress allows for a maximum flaw size of 0.5 mm before fracture. Thus, as long as there
are no visible flaws in the window, it should withstand the pressure safely. The deflection of the
180
window under max pressure is on the order of 100 µm (the nonzero starting coordinate in Figure
A-6 is due to the compression of the compliant washer around the perimeter).
Figure A-6: Deflections in the window (color bar in units of µm) with a resin pressure of 180 psi.
A.3. Plumbing
The fluid connections are shown in Figure A-7. The only “tricky” connection is for the
tubing into and out of the molding tool (see Figure A-8). Compression fittings are used, which
contain ferrules that compress around the PTFE tubing, creating a reliable seal. The fittings, which
normally contain an inner shoulder for the end of the tubing to rest against, are drilled-out to
remove the shoulder and allow the tubing to pass entirely through. The ferrule still seals the
connection, but by positioning the end of the tubing at the entrance of the mold cavity, we prevent
resin from curing in locations that would be difficult to clean (e.g., in interior channels of the
molding tool). Resin cures within the PTFE tubing during part fabrication, and afterwards, the
181
entire section of tubing can be removed and replaced. At the very end of the PTFE tubing, another
short section of an inner tube is used, which eases separation of the cured part from the mold.
Figure A-7: Schematic of the fluid connections between components of the lab-scale RTM system.
Figure A-8: Section view of the resin inlet/outlet design.
182
A.4. Heating simulation
Cylindrical 300 watt cartridge heaters are used, with thermocouples and closed-loop
temperature control. A solid-state relay provides power using a standard 110V wall outlet, and a
PID controller enables programmable temperature control. Features include: tuning capability for
all PID algorithm parameters, and savable temperature ramp/hold profiles with up to 40 steps.
A finite element heat transfer model was made to predict the temperature distributions
within the test cell prior to fabrication. Using natural external convection in room-temperature air
as a boundary condition and holding the heaters at 180 °C, the steady-state temperature distribution
at the tool face is shown in Figure A-9. This heater configuration shows temperature variations
within the sample area of less than 1 °C.
Figure A-9: Predicted steady-state tool-face temperature distribution with heaters at 180 °C.
183
Since the window on one side of the mold cavity does not contain heaters, there are
inevitably through-thickness temperature gradients. The same thermal model predicts a maximum
variation of 9 °C between the hottest and coldest regions of the mold cavity (spatial temperature
gradients can be brought below this level by insulating the exterior of the molding tool).
Figure A-10: Predicted steady-state window-side temperature distribution with heaters at 180 °C.
184
Design details for the HD/RTM
B.1. Technical drawings
The HD/RTM consists of 9 custom-machined pieces, listed here from top to bottom as
arranged in the completed assembly:
Figure B-1: Custom-built components of the HD/RTM.
Technical drawings of the parts are provided next, in the order listed above.
185
Figure B-2: Dimensions of the upper load frame connection.
186
Figure B-3: Dimensions of the upper ceramic insulation (sheet 1 of 3).
187
Figure B-4: Dimensions of the upper ceramic insulation (sheet 2 of 3).
188
Figure B-5: Dimensions of the upper ceramic insulation (sheet 3 of 3).
189
Figure B-6: Dimensions of the upper plate (sheet 1 of 2).
190
Figure B-7: Dimensions of the upper plate (sheet 2 of 2).
191
Figure B-8: Dimensions of the piston.
192
Figure B-9: Dimensions of the ring (sheet 1 of 2).
193
Figure B-10: Dimensions of the ring (sheet 2 of 2).
194
Figure B-11: Dimensions of the base plate (sheet 1 of 2).
195
Figure B-12: Dimensions of the base plate (sheet 2 of 2).
196
Figure B-13: Dimensions of the lower plate.
197
Figure B-14: Dimensions of the bottom ceramic insulation.
198
Figure B-15: Dimensions of the stand.
199
B.2. System overview
Figure B-16: Picture of the HD/RTM fully assembled.
200
Figure B-17: Annotated CAD model of the HD/RTM.
The HD/RTM consists of (1) a flat base plate with an embedded heater, thermocouple,
resin pressure sensor, resin inlet port, and three vertical alignment rods; (2) an annular ring piece
with a precise inner bore diameter, a circumferential resin “racetracking” groove, guide bearings,
and a location for a displacement sensor target; and (3) an upper assembly with a piston, guide
bearings, resin outlet port, heater, thermocouple, displacement sensor, and a load frame
connection. The device can be used to produce composite samples by RTM at temperatures up to
200 °C and pressures up to 10 bar (possibly more). The mold cavity consists of a 76 mm (3”)
diameter disk with variable thickness. The sensor target is designed for a 3 mm cavity, and the
201
capacitive displacement sensor can detect the nominal thickness ±1 mm with 1.5 nm resolution
and 1 µm linearity. Fiber preforms must be included in the sample cavity prior to inserting the
piston, and sufficiently thick to prevent collisions between the displacement sensor and its target.
The preforms should also contain a central hole ~5 mm in diameter to prevent fibers from
impinging on the pressure sensor tip (i.e., so the sensor reads resin pressure only).
Figure B-18: Annotated CAD model of the HD/RTM (section view).
The device is mounted in a mechanical test frame in static compression mode. The upper
assembly hangs from a universal joint that allows for angular alignment, while the lower plate and
202
ring assembly rests on a flat compression platen that allows for lateral alignment. The guide rods
ensure that the piston inserts into the lower ring piece without tilting and becoming stuck.
Figure B-19: Section view of the HD/RTM. The sample cavity is shown in yellow and arrows on the right indicate
the measured gap.
The load frame control software can operate in load-control mode to maintain a desired
piston pressure. Prior to injection, this pressure (minus frictional losses) acts on the fiber preform;
after injection, the piston pressure is equal to the sum of the resin pressure and the compressive
stress on the fiber bed. During testing the load frame also records data for crossbeam displacement
and compressive load.
The remaining external system components are almost the same as for the lab-scale RTM.
Temperature is controlled using a similar pair of PID controllers, and a Labview VI is used to
record system temperatures, resin pressure, and displacement (from the capacitive sensor). Resin
injections are also performed using the same resin injector as the lab-scale RTM.
203
Figure B-20: HD/RTM mounted in the load frame. Can you spot the elephant in the room?
204
Preliminary results from the HD/RTM
The volumetric changes of a thermoset resin are shown schematically in Figure C-1. The
resin begins in the uncured state at room temperature (lower circle on the left side), and decreases
in density due to thermal expansion upon heating to the cure temperature. Cure shrinkage causes
a density increase, and cooling to room temperature causes a further increase in density. While the
“apparent” shrinkage between uncured and cured states at room temperature was already known,
the goal of this work was to characterize the full path shown in Figure C-1 for this resin.
Figure C-1: Density changes in a thermoset resin. Image source: [115].
C.1. Composite sample
A composite sample was fabricated in the HD/RTM using the same type of fiber preform
as in Chapters 2 and 3 and the F2 formulation of the benzoxazine resin. First, the dry fiber plies
were inserted into the sample cavity and a controlled load of 0.9 kN (~225 kPa) was applied via
the piston. The HD/RTM was heated to 110 °C and held for 2 hours to pre-compact the fiber
205
preform. Vacuum was applied to the outlet vent, and the load-control was temporarily paused to
prevent piston movement during resin injection. Resin was injected, then the outlet and inlet were
sealed. Load control was resumed on the test frame, this time at a higher compaction force of 2.5
kN. The device was then allowed to cool to room temperature overnight while maintaining the
controlled load.
The device was heated again, this time slowly (1/2 °C per minute), and held for 1 hour at
the maximum safe operating temperature, in an effort to fully cure the resin. The HD/RTM was
finally cooled down again, completing the test. Figure C-2 shows temperature and cavity thickness
data for the curing phase of the test, as well as modeled values for degree of cure and Tg.
Figure C-2: Temperature and cavity thickness data from a composite sample, as well as models for degree of cure
and glass transition temperature.
206
The cavity thickness data is plotted against temperature on Figure C-3. If we assume that
the strains are purely thermal in the low temperature ranges (where curing has either not yet begun
or already completed), the coefficients of thermal expansion can be computed from the slopes of
the red lines, for the uncured composite (CTE0) and for the cured composite (CTE∞).
Figure C-3: Cavity thickness versus temperature, with cured and uncured coefficients of thermal expansion (CTEs)
in red.
The total volumetric strains εv are modeled as the sum of the thermal strains εT and chemical
strains εc. Thermal strains vary with temperature, and the CTE varies linearly between the values
corresponding to either extreme of the degree of cure α. The chemical cure shrinkage (CCS) is
purely a function of α.
𝜀 𝑣 = 𝜀 𝑇 ( 𝛼 , 𝑇 )+ 𝜀 𝑐 ( 𝛼 ) (C-1)
𝑑𝜀 𝑇 = [𝛼 ∙ 𝐶𝑇𝐸 ∞
+ ( 1 − 𝛼 )∙ 𝐶𝑇𝐸 0
]𝑑𝑇 (C-2)
𝑑 𝜀 𝑐 = 𝐶𝐶𝑆 ∙ 𝑑𝛼 (C-3)
207
Using a “model free” isoconversional cure kinetics model similar to the one described in Chapter
3, along with the above equations, the sample thickness was modeled and compared to the
measured data, shown here:
Figure C-4: Comparison of modeled and measured sample thickness.
The measured sample thickness follows the modeled surface with reasonable accuracy, but
the astute reader will notice an apparent inconsistency: even though the cured resin is known to be
~1% denser than the uncured resin at room temperature, the sample thickness was greater after
cure than before. The apparent CTE0 was so high that the sample thickness increased more during
the initial temperature ramp than it decreased due to cure shrinkage.
Was this behavior due to additional material entering the mold cavity upon heating? Or
was it part of the intrinsic response of the measurement device? To answer these questions, it was
necessary to calibrate the HD/RTM using materials with known properties, for both solid and
208
liquid samples. Since the displacement sensor measures a gap on one side of the device to infer
the thickness of the main sample cavity, any thermal expansion of the device itself can lead to
errors. To isolate the strains of just the sample, the baseline response of the HD/RTM must be
subtracted from the measured data.
C.2. System calibration for solids
To calibrate the HD/RTM thickness measurements for solid materials, the same composite
sample was measured in both the HD/RTM and a thermomechanical analyzer (TMA). First, a
second composite sample was fabricated using the same procedure described above, but prior to
removing the sample after curing, it was run through a second heating cycle. This cycle was
intended to measure the thermal strains of the fully cured composite. The sample thickness data
from this second heating cycle is shown below.
Figure C-5: Sample thickness versus time for a cured sample in a second heating cycle.
Additional cure shrinkage during this second heating cycle indicated that the sample did
not fully cure during the first heating cycle (despite the cure model predicting that full cure would
209
be reached). When the sample was heated a third time in the TMA, even more cure shrinkage
occurred. This residual shrinkage complicates interpretation of the results, since differences
between the data measured by the HD/RTM and by the TMA could stem from either the baseline
response of the HD/RTM or from actual variations in the sample thickness. The upward
temperature ramps showed some discrepancy between the two measurement techniques (which
could be due to either of the aforementioned causes), but the downward ramps showed nearly
perfect agreement.
Figure C-6: Comparison of thickness changes for an upward temperature ramp on a cured sample, measured by
TMA and in the HD/RTM (labelled “PVT” here, for “pressure-volume-temperature” analyzer).
Figure C-7: Comparison of thickness changes for a downward temperature ramp on a cured sample, measured by
TMA and in the HD/RTM.
210
Figure C-8: Measured thermal expansion versus temperature, determined from the slope of the TMA data in Figure
C-7.
The cure kinetics model did not predict that the temperature cycle shown in Figure C-2
would result in a less-than-fully-cured sample. Reviewing a comparison of the kinetics model with
DSC data, we see that the model can accurately predict the advancement of the curing reaction for
dynamic temperature ramps (see Figure C-9), but that the model over-predicts the cure rate during
isothermal dwells (in Figure C-10, compare the model to the last hour of the data at 160 °C).
Figure C-9: Measured and modeled advancement of the degree of cure for linear temperature ramps.
211
Figure C-10: Measured and modeled advancement of the degree of cure for isothermal dwells.
Since this isoconversional model could not capture the reduced cure rate in the diffusion-
limited regime, another modeling approach was attempted instead. The full set of DSC tests (4
dynamic ramps and 5 isothermal dwells) was fit to an equation of the following form:
𝑑𝛼 𝑑𝑡 = 𝐴 ∙ exp [
−𝐸 𝑎 𝑅𝑇
] ∙
𝛼 𝑚 ( 1 − 𝛼 )
𝑛 1 + exp[𝐶 ( 𝛼 − ( 𝑎 𝐶 0
+ 𝛼 𝐶𝑇
𝑇 ) ) ]
(C-4)
This model consists of an Arrhenius temperature dependence, an autocatalytic α dependence, and
a “diffusion factor” in the denominator that limits the reaction rate at higher degrees of cure. The
expression contains 7 fitting constants: A, Ea, m, n, C, αC0, and αCT. This rate equation can be
integrated over time to predict the evolution of α. Figure C-11 shows the modeled degree of cure
as a mesh surface, the measured DSC data (black lines), and the model/experiment error (colored
bands). As opposed to the isoconversional model, this cure kinetics model does capture the reduced
cure rate at high values of α and intermediate temperatures.
212
Figure C-11: Measured DSC data (black lines), model predictions (mesh surface), and model/experiment error
(colored bands) for Eq. (C-4).
To properly calibrate the HD/RTM for solid samples, another composite sample would
have to be fabricated, using a longer high-temperature dwell to ensure complete cure (the alternate
kinetics model would help for determining the required dwell time). The thermal strains for the
fully cured sample could then be isolated using a second heating cycle.
C.3. System calibration for liquids
The HD/RTM was calibrated for liquid samples using glycerol, an organic liquid that was
chosen due to its high boiling point, chemical inertness, non-toxicity, and well-known thermal
expansion characteristics. Samples were fabricated using the same procedure as outlined
previously, using glycerol instead of benzoxazine resin.
213
Figure C-12: The density of glycerol as a function of temperature.
The first test using glycerol failed, because the repeated heating and cooling of the inlet
tubing caused a leak to occur, evidenced by the change in sample thickness between the beginning
and end of the test (see Figure C-13). Applying silicone caulk sealant to the tubing prior to
assembly prevented this issue in subsequent tests.
Figure C-13: An example of a test that leaked.
Figure C-14 shows the cavity thickness versus temperature for a benzoxazine/carbon
sample (black) and a glycerol/carbon sample (solid blue). The closed loop for the glycerol data
confirms that the sample did not leak and that all strains were purely thermal. The hysteresis in
214
this curve shows that transient effects (i.e., thermal lag for heating versus cooling) were minimal.
The dashed blue line shows the expected thermal expansion of the sample using the known CTE
of glycerol (the expansion of the sample was computed using a rule of mixtures to account for the
fiber volume fraction, and assuming that the CTE of the fibers is negligible compared to that of
the resin). The measured increase in sample thickness, however, was roughly 2.5 times greater
than predicted. We can safely say that this additional expansion is not due to the baseline response
of the HD/RTM, since this behavior only occurs with liquid samples.
Figure C-14: Calibration test results using glycerol.
A likely explanation for this behavior is that additional material enters the sample cavity
upon heating. The sample cavity is not a perfectly closed system, since resin (or glycerol) also
resides in the inlet and outlet tubing, as well as the O-ring grooves that seal the interfaces between
the metal components. If, for example, the large O-ring between the base plate and ring pieces
expands as the device is heated, the expansion would displace resin (or glycerol) in that groove,
pushing it into the sample cavity. A term describing the additional volume of liquid in the cavity
215
(as a quadratic function of temperature) was added to the sample thickness calculations, resulting
in the red line shown in Figure C-14 (which matches the measured sample thickness). This
correction factor added ~1.1 ml of liquid at the maximum test temperature (corresponding to a
~14% increase in the total amount of liquid in the cavity).
To summarize: for a sample with resin and carbon fibers, two calibration steps should be
applied. First, the baseline response of the HD/RTM should be subtracted from the overall
measured response, to isolate the volumetric changes of the sample from that of the device itself
(this baseline could be obtained from the difference between the measured thickness of a fully
cured sample during a re-heat cycle, and the thickness of that same sample as measured by a pre-
calibrated instrument such as a TMA). Secondly, the temperature-dependent “extra volume”
correction should be applied, but only prior to the time of resin gelation.
Unfortunately, these calibrations were never applied because no more samples were made.
It was at this time that the resin supplier decided to abandon ship, and I began working on the co-
cure project instead. Chapter 6 discusses some possible future uses for the HD/RTM tool and
applications of the information that can be obtained from it.
216
Design details for the “mini autoclave” co-cure fixture
D.1. Technical drawings
Figure D-1: Dimensions for the main tool body (sheet 1 of 4).
217
Figure D-2: Dimensions for the main tool body (sheet 2 of 4).
218
Figure D-3: Dimensions for the main tool body (sheet 3 of 4).
219
Figure D-4: Dimensions for the main tool body (sheet 4 of 4).
220
Figure D-5: Dimensions for the window’s retaining ring.
221
Figure D-6: Dimensions for the lid (sheet 1 of 2).
222
Figure D-7: Dimensions for the lid (sheet 2 of 2).
223
D.2. Stress simulation
A finite element stress simulation was performed to ensure that the tool could be
pressurized safely. Figure D-8 shows von Mises stresses on the main tool body with 100 psi of
autoclave pressure applied to the appropriate surfaces (roughly double the maximum anticipated
pressure to be used with this device).
Figure D-8: Simulated stresses in the main tool body.
Figure D-9 shows only the stresses above 40 MPa in the main tool body. The washers used
to hold the upper and lower tool pieces together introduced artificially high stress concentrations
in the simulation. Elsewhere in the design, stresses remained well below the yield stress of the
material. Figure D-10 shows stresses in the lid, and Figure D-11 shows only the locations where
stresses exceeded 60 MPa. Again, the washers introduced artificial stress concentrations, but the
stresses stayed well below dangerous levels elsewhere.
224
Figure D-9: Simulated stresses (> 40 MPa) in the main tool body.
Figure D-10: Simulated stresses in the lid (deflections greatly exaggerated).
225
Figure D-11: Simulated stresses (> 60 MPa) in the lid.
Abstract (if available)
Abstract
Polymer matrix composites are among the highest-performance structural materials available today, but manufacturing processes for these materials have not yet been perfected. This dissertation presents two investigations into manufacturing processes that are prone to particular types of process-induced defects. In situ visualization was used as a diagnostic method in both cases, first to analyze the evolution of process-induced defects in real time, and then to demonstrate the effectiveness of proposed process modifications for avoiding these defects. ❧ The first investigation considered resin transfer molding (RTM) with a benzoxazine resin, which tended to result in composite parts with significant surface porosity. A lab-scale RTM tool with in situ observation capabilities was used to identify the defect formation mechanism (Chapter 2). A modified cure temperature cycle was shown to effectively prevent surface porosity, and guidelines for design of such cure cycles were developed (Chapter 3). ❧ The second investigation considered the autoclave co-cure of honeycomb core sandwich structures. This particularly challenging manufacturing process can result in multiple types of process-induced defects, thus the project goal was to develop a physics-based process model to enable prediction of defect formation. First, a lab-scale autoclave with in situ observation capabilities was used to identify the possible problems that could occur during co-cure (Chapter 4). Process modifications were proposed to avoid the observed defect formation mechanisms, and demonstrated to be effective. Finally, a study of prepreg gas permeability was conducted (Chapter 5). Prepreg permeability is a poorly-understood yet important topic for co-cure processing, because it affects the gas pressure in the honeycomb core cells, which, in turn, affects the quality of the skin/core bond-line. Two types of models for prepreg permeability were proposed, which can be used as sub-models within the integrated co-cure process model. ❧ Overall, this dissertation contributed to the knowledge on composite processing in several ways. First, a previously unexplained type of defect in RTM composites was characterized, and a practical mitigation strategy was developed. Second, a new visualization technique for observing the skin/core bond-line during co-cure was developed. Third, an unconventional processing strategy (in-bag pressurization) was applied to co-cure and shown to solve many of the problems that can occur. Finally, a new approach to modeling the gas permeability of prepregs was developed, which, unlike previous empirical models, accounts for the relevant process physics.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
In situ process monitoring for modeling and defect mitigation in composites processing
PDF
Defect control in vacuum bag only processing of composite prepregs
PDF
Sustainable manufacturing of out-of-autoclave (OoA) prepreg: processing challenges
PDF
Fabrication and analysis of prepregs with discontinuous resin patterning for robust out-of-autoclave manufacturing
PDF
Void evolution in vacuum bag-only prepregs
PDF
Material and process development and optimization for efficient manufacturing of polymer composites
PDF
Characterization, process analysis, and recycling of a benzoxazine-epoxy resin for structural composites
PDF
Sound transmission through acoustic metamaterials and prepreg processing science
PDF
Efficient manufacturing and repair of out-of-autoclave prepreg composites
PDF
Developing efficient methods for the manufacture and analysis of composites structures
PDF
Chemical recycling of amine/epoxy composites at atmospheric pressure
PDF
Comprehensive uncertainty quantification in composites manufacturing processes
PDF
Polymer flow for manufacturing fiber reinforced polymer composites
PDF
Real time surface analysis of complex oxide thin films during pulsed laser deposition
PDF
Processing and properties of phenylethynyl-terminated PMDA-type asymmetric polyimide and composites
PDF
Processing, thermal-oxidative stability and thermal cyclic fatigue of phenylethynyl-terminated polyimides
PDF
Vacuum-bag-only processing of composites
PDF
Development of composite oriented strand board and structures
PDF
Fatigue and fracture of pultruded composite rods developed for overhead conductor cables
PDF
Modeling and simulation of complex recovery processes
Asset Metadata
Creator
Anders, Mark
(author)
Core Title
In situ process analysis for defect control during composites manufacturing
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Mechanical Engineering
Publication Date
04/25/2019
Defense Date
03/18/2019
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
composites,defects,manufacturing,OAI-PMH Harvest
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Nutt, Steven (
committee chair
), Armani, Andrea (
committee member
), Spedding, Geoffrey (
committee member
)
Creator Email
anders@usc.edu,markanders1@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-145079
Unique identifier
UC11659972
Identifier
etd-AndersMark-7255.pdf (filename),usctheses-c89-145079 (legacy record id)
Legacy Identifier
etd-AndersMark-7255.pdf
Dmrecord
145079
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Anders, Mark
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
composites
defects