Del Linz-P-2014-PhD-Thesis

Blast Resistance of Laminated
Glass Facades
Paolo Del Linz

Mechanics of Materials Division
Department of Mechanical Engineering
Imperial College London
A thesis submitted for the degree of Doctor of Philosophy of
Imperial College London and the Diploma of Imperial College
April 2014
1
Acknowledgments
There are many people I need to thank for being able to complete the three year journey this
project represents. Of course, I could not have done it without the support of all my
supervisors, John Dear and Bamber Blackman at Imperial College and David Smith, Luke
Pascoe and Ryan Sukhram at ARUP Security and Resilience Consulting. I am grateful for
having been given the opportunity to undertake this project and for the continued support and
encouragement through all the difficulties that inevitably cropped up over time. The financial
support for this project was given by EPSRC and Arup Consulting and is gratefully
acknowledged. I also wish to thank CPNI and GL Group for providing access and help during
my blast test. I would especially like to thank Ian, Darren and Willie for all the help with
setting up on site.
I need to thank all my colleagues who helped me out over the years. Paul gave me a very
good point to start from and provided very useful advice on the direction of my work. Hari
also helped me very significantly, with lots of advice throughout the years. I am grateful to
Mark, Alex, Yi and Aditya for all the conversations and help for my laboratory and blast
tests. Suresh, Amit and Tony were always ready to help with test preparations when needed,
for which I am grateful. Alex Crosse was also available to discuss topics when I needed it,
and helped getting me unstuck in many occasions. I also need to thank Michelle Hoo Fatt for
the very helpful discussion on material models. My officemates deserve thanks for
suggestions and, especially, for putting up with my fallings out with my computer. There are
almost too many people to mention for this. Richard, Tom and Charles, definitely all deserve
a big thank you though.
I’d also like to thank my colleagues at Hyder Consulting who gave me the opportunity to
work on some amazing projects. They planted the worm in me which grew into my wish to
switch to research, and were very encouraging throughout. Leigh and Mark deserve a special
mention due to their help on current industry practice and codes.
My Italian friends deserve a mention. They helped to distract me and keep me sane during
these years, providing both advice and de-stressing occasions. Matteo, Stefano, Federico,
Vidja, Giulia and Alessandro all were close to me throughout this process. Grazie a tutti voi!
2
I would like to thank my family in Italy. My parents and brother were close to me and
supported me in all my choices. A special mention definitely needs to go to my father for not
letting his English stopping him from proofreading my thesis! Grazie, mi siete stati sempre
vicini, e lo apprezzo tantissimo.
Finally, I owe too much gratitude for words to my wife Samantha, who was close to me
throughout this journey. She knew exactly what taking care of a PhD student would involve,
and she still encouraged me to apply and helped me to overcome all the difficulties I faced in
the past few years. Thank you so much Sam, I dedicate this to you.
3
Declaration of Originality
The work presented in this thesis was produced by me during the course of my PhD studies.
Where contributions by others are used, they have been appropriately referenced in the text
and in the bibliography.
Copyright Declaration
The copyright of this thesis rests with the author and is made available under a Creative
Commons Attribution Non-Commercial No Derivatives licence. Researchers are free to copy,
distribute or transmit the thesis on the condition that they attribute it, that they do not use it
for commercial purposes and that they do not alter, transform or build upon it. For any reuse
or redistribution, researchers must make clear to others the licence terms of this work.
4
Abstract
The aim of this thesis is to improve the understanding of the behaviour of Polyvinyl Butyral
(PVB) laminated annealed glass façade panels subjected to blast loading.
A full scale blast test was performed. During this, deflection and strain data were collected
employing digital image correlation techniques (DIC). Local reaction forces were measured
using several pairs of strain gauges on the support. The full field deflection and strain data
obtained were in line with those observed in historical tests. The strain gauge data available
showed that the reaction forces varied along the edge, with higher values being reached at the
quarter length gauge locations. The results from this test and from other historical
experiments were used to calculate the reaction forces along the entire perimeter of the glass
pane. The results showed that the forces reach an early peak before the glass failure, and then
rise gradually approaching a plateau at high central deflections.
To explain the specific form of this force time history, the detailed behaviour of the laminated
material after the glass skins failed was studied. Existing experimental data was employed to
fit a material model to the PVB material. Two Prony series models with different hyperelastic
springs and a model employing a full finite deformation viscoelastic law derivation were
employed. It was found that the finite deformation viscoelastic model could represent the
material’s behaviour more accurately and fully include its rate dependency.
One of the PVB models was employed to study the delamination between the glass and the
membrane. Delamination energies were found for different speeds of deformation, and these
parameters were employed to study the delamination of samples presenting different crack
arrangements. The results showed that these had only a limited influence on the behaviour of
the composite.
5
Table of Contents
Acknowledgments...................................................................................................................... 2
Declaration of Originality .......................................................................................................... 4
Copyright Declaration ................................................................................................................ 4
Abstract ...................................................................................................................................... 5
List of Figures .......................................................................................................................... 10
List of Tables ........................................................................................................................... 16
Nomenclature ........................................................................................................................... 17
Acronyms ................................................................................................................................. 21
1. Introduction ...................................................................................................................... 22
1.1. Project aims ............................................................................................................... 24
2. Literature Review ............................................................................................................. 25
2.1. Blast Pressure Theory................................................................................................ 25
2.1.1. Blast wave characteristic and interactions with structures ................................ 25
2.1.2. Loading estimate methods ................................................................................. 27
2.2. Behaviour of Laminated Glass .................................................................................. 28
2.2.1. PVB material model ........................................................................................... 28
2.2.2. Glass behaviour .................................................................................................. 29
2.2.3. PVB – glass delamination .................................................................................. 30
2.3. Overall response of laminated glass .......................................................................... 31
2.3.1. Low rate loading ................................................................................................ 31
2.3.2. Impact loading ................................................................................................... 32
2.3.3. Blast loading studies .......................................................................................... 32
2.4. Restraint mechanism ................................................................................................. 36
2.5. Conclusion................................................................................................................. 38
3. Full Scale Blast Test ......................................................................................................... 39
6
3.1. Introduction ............................................................................................................... 39
3.2. Method ...................................................................................................................... 39
3.2.1. Experimental set up and data collection ............................................................ 39
3.2.2. Data analysis ...................................................................................................... 48
3.3. Results ....................................................................................................................... 55
3.3.1. Blast Pressures ................................................................................................... 55
3.3.2. Window general observations ............................................................................ 58
3.3.3. Digital Image Correlation calibration comparison ............................................ 61
3.3.4. DIC Analysis Results ......................................................................................... 66
3.3.5. Reaction data ...................................................................................................... 73
3.4. Discussion ................................................................................................................. 78
3.5. Conclusion................................................................................................................. 80
4. Blast Reaction Forces ....................................................................................................... 82
4.1. Introduction ............................................................................................................... 82
4.2. Method ...................................................................................................................... 82
4.2.1. Pre-cracked reactions ......................................................................................... 84
4.2.2. Post-cracked reactions ....................................................................................... 89
4.3. Results ....................................................................................................................... 91
4.3.1. DIC Results - Extrapolation of deflection data .................................................. 91
4.3.2. Pre-cracked reactions ......................................................................................... 93
4.3.3. Post crack reactions............................................................................................ 98
4.4. Discussion ............................................................................................................... 107
4.5. Conclusion............................................................................................................... 110
5. PVB Material Properties ................................................................................................. 111
5.1. Introduction ............................................................................................................. 111
5.2. Methods ................................................................................................................... 112
5.2.1. Experimental data ............................................................................................ 112
7
5.2.2. Experimental Results ....................................................................................... 113
5.2.3. Material model fit ............................................................................................ 116
5.2.4. Finite deformation viscoelasticity .................................................................... 120
Data fit method ............................................................................................................... 123
5.3. Results ..................................................................................................................... 126
5.3.1. Prony series - Hoo Fatt spring ......................................................................... 126
5.3.2. Prony series – Reduced polynomial spring...................................................... 130
5.3.3. Finite deformation viscoelasticity .................................................................... 134
5.4. Discussion ............................................................................................................... 136
5.5. Conclusion............................................................................................................... 139
6. Glass – PVB Delamination ............................................................................................. 141
6.1. Introduction ............................................................................................................. 141
6.2. Method .................................................................................................................... 142
6.2.1. Experimental data ............................................................................................ 142
6.2.2. Delamination energies ..................................................................................... 148
6.2.3. Parametric studies ............................................................................................ 157
6.3. Results ..................................................................................................................... 160
6.3.1. Experimental data analysis .............................................................................. 160
6.3.2. Single crack modelling .................................................................................... 162
6.3.3. Effect of crack spacing study ........................................................................... 171
6.3.4. Inclined Cracks study....................................................................................... 173
6.3.5. Staggered crack study ...................................................................................... 174
6.4. Discussion ............................................................................................................... 177
6.4.1. Delamination energies ..................................................................................... 177
6.4.2. Parametric studies ............................................................................................ 179
6.5. Conclusion............................................................................................................... 183
7. Conclusion ...................................................................................................................... 186
8
7.1. Blast Test Results and Analysis .............................................................................. 186
7.2. PVB Characterisation .............................................................................................. 188
7.3. Delamination Modelling ......................................................................................... 188
7.4. Further Work ........................................................................................................... 190
7.4.1. Blast testing ...................................................................................................... 190
7.4.2. Bending capacity of cracked laminated glass .................................................. 191
7.4.3. PVB material properties ................................................................................... 191
7.4.4. Glass – PVB delamination ............................................................................... 192
7.4.5. Alternative interlayer materials........................................................................ 192
7.4.6. Alternative holding mechanisms...................................................................... 192
7.5. Conclusion............................................................................................................... 193
References .............................................................................................................................. 194
Appendix A – Finite Deformation Viscoelasticity Law Derivation ...................................... 202
Nomenclature ..................................................................................................................... 202
Derivation ........................................................................................................................... 202
9
List of Figures
Figure 1-1: Effect of an explosion on office buildings in London’s Docklands area. ............. 23
Figure 2-1: Typical shape of blast pressure wave time history. .............................................. 26
Figure 3-1: Section sketch of the glazing metal supports. All dimensions are in mm. ........... 41
Figure 3-2: Cubicle elevation with dimensions. ...................................................................... 41
Figure 3-3: Steel structure of the front cubicle panel. ............................................................. 42
Figure 3-4: Section through the test pad, showing the charge and the cubicle containing the
glazing sample. ................................................................................................... 43
Figure 3-5: Plan view of the test pad, showing some of the other test cubicles present.......... 44
Figure 3-6: Plan view of the DIC set up inside the cubicle. .................................................... 45
Figure 3-7: Speckle pattern on the glazing. ............................................................................. 46
Figure 3-8: Cubicle front face elevation showing the location and numbering of the strain
gauges used. ........................................................................................................ 47
Figure 3-9: One of the sets of calibration pictures used for the procedure. ............................. 50
Figure 3-10: Flow chart summarising the calibration and DIC analysis process used for the
experimental data. ............................................................................................... 51
Figure 3-11: Sketch showing the forces acting on the strain gauges and the main dimensions
used in the calculation. ....................................................................................... 53
Figure 3-12: Strain distribution in the supporting plate’s cross section. ................................. 53
Figure 3-13: Pressure time history at 25 m stand-off. The experimental and CFD incident data
are shown. The charge was set of at time 0 ms. ................................................. 56
Figure 3-14: Pressure time history at 23 m stand off calculated with Air3D. The charge was
set of at time 0 ms. .............................................................................................. 57
Figure 3-15: Comparison of Air3D results for two mesh sizes. The charge was set of at time 0
ms........................................................................................................................ 58
Figure 3-16: Existing diagonal crack location. ........................................................................ 59
Figure 3-17: Front view of the tested panel, showing main areas of damage. ........................ 60
Figure 3-18: Detail of central diagonal failure. ....................................................................... 60
Figure 3-19: Detail of the pane failure at the bottom support.................................................. 61
10
Figure 3-20: 3D plot of the deflection data. The shaded data set is the data obtained with the
new calibration method, whilst the white facets curve shows the old calibration
data...................................................................................................................... 62
Figure 3-21: A plot of the relative and absolute deflection errors. .......................................... 62
Figure 3-22: Relative and absolute differences in the velocity data obtained with the two
calibrations.......................................................................................................... 63
Figure 3-23: Comparison of strain data obtained with both calibrations and with two methods
for the new calibration data. Both strains in the x and in the y direction are
shown. ................................................................................................................. 64
Figure 3-24: Comparison of angle measurements time history along the left hand vertical
window side. ....................................................................................................... 65
Figure 3-25: Comparison of angle measurements time history along the top window side. ... 65
Figure 3-26: Relative error for the two data sets displayed above. ......................................... 66
Figure 3-27: Out-of-plane deflections of the glass panel measured with the DIC method. The
deflection is in mm. Time 0 ms is at the blast wave arrival. .............................. 67
Figure 3-28: A plot of the curvature and deflections along a horizontal cut taken across the
centre of the window. ......................................................................................... 68
Figure 3-29: Central deflection time history from DIC data. Time 0 ms is at the blast wave
arrival. ................................................................................................................. 69
Figure 3-30: Central velocity time history from DIC data. Time 0 ms is at the blast wave
arrival. ................................................................................................................. 69
Figure 3-31: Out of plane velocities measured with the DIC method. The velocity is given in
m/s. Time 0 ms is at the blast wave arrival. ....................................................... 70
Figure 3-32: Strain in x direction calculated from the DIC deflection data. The strain is given
as a percentage. Time 0 ms is at the blast wave arrival. ..................................... 71
Figure 3-33: Strain in y direction calculated from the DIC deflection data. The strain is given
as a percentage. Time 0 ms is at the blast wave arrival. ..................................... 72
Figure 3-34: Cubicle elevation showing location of broken gauges........................................ 74
Figure 3-35: Angle and reaction force at gauge location 1 as shown in Figure 3-35. Time 0
ms is at the blast wave arrival. ............................................................................ 75
11
Figure 3-38: Reaction force data obtained on the top edge of the window. The gauge
locations are as indicated in Figure 3-35. Time 0 ms is at the blast wave arrival.
............................................................................................................................ 77
Figure 3-39: Reaction force data obtained on the left hand side edge of the window. The
gauge locations are as indicated in Figure 3-35. Time 0 ms is at the blast wave
arrival. ................................................................................................................. 77
Figure 4-1: DIC set up used in Hooper et al.’s blast tests [37]. ............................................... 83
Figure 4-2: An image of the FEA model geometry employed for tests 1 to 3. The detail shows
the mesh in the single sided joint area. ............................................................... 87
Figure 4-3: An image of the FEA model geometry employed for test 4. The detail shows the
mesh in the double sided joint area. ................................................................... 87
Figure 4-4: Representations of the reactions assumed in the pre glass cracking force
calculation. .......................................................................................................... 89
Figure 4-5: Original out of plane deflection data (top) and the same data with missing and
unrealistic points substituted (bottom). .............................................................. 92
Figure 4-6: Comparison of central deflection obtained through DIC and FEA modelling for
Test 2. Time = 0 ms is the blast wave arrival time. ............................................ 93
Figure 4-7: Comparison of central deflection obtained through DIC and FEA modelling for
Test 4. Time = 0 ms is the blast wave arrival time. ............................................ 94
Figure 4-8: Proportion of the stresses due to bending moments and membrane forces found
with the FEA model for Test 2. .......................................................................... 94
Figure 4-9: Samples of bending stress data. Both the original DIC data and the filtered curve
are shown. ........................................................................................................... 95
Figure 4-10: Out-of-plane reaction force along one edge. The bending and membrane
components are shown separately. ..................................................................... 96
Figure 4-11: Bending moment at the gauges location on the bottom side of the panel in Test
1calculated from the gauges results and from the analysis results. .................... 97
Figure 4-12: Bending moment at the gauges location on the top side of the panel in Test 1
calculated from the gauges results and from the analysis results. ...................... 97
Figure 4-13: Stress strain curves for the estimate of cracked glass material model. ............... 98
Figure 4-14: Stress strain curve and model fit for Test 1 the top of the window frame. ......... 98
Figure 4-15: PVB reaction stress along the bottom edge of Test 1 at time steps t = 26 ms and t
= 28 ms. ............................................................................................................ 100
Figure 4-16: Legend of edge locations for reaction data plots. ............................................. 100
12
Figure 4-17: Out of plane reactions on individual panel sides for Test 3. The edge locations
are shown in Figure 4-16. ................................................................................. 101
Figure 4-18: Total out of plane reactions forces for Test 1. The pre crack peak at low
deflection is apparent. ....................................................................................... 102
Figure 4-22: Total in plane edge reaction force along each edge for Test 3. See Figure 4-16
for the location of the edges. ............................................................................ 104
Figure 4-23: Total in plane reactions for Test 1. ................................................................... 105
Figure 4-27: DIC strains along a cut in test 2. ....................................................................... 108
Figure 5-1: Typical chemical structure of the PVB polymer chain [62]. .............................. 111
Figure 5-2: Dimensions of Dumb-bell PVB specimen. ......................................................... 113
Figure 5-3: Low rate true Stress - stretch curves for PVB under uniaxial tensile loading. Only
Wang’s data for the 0.2 s-1 case are presented for clarity. ................................ 114
Figure 5-4: High rate true stress – stretch curves for PVB under uniaxial tensile loading.... 115
Figure 5-5: A comparison of the results from Hooper [65] and Wang at a strain rate of 0.2 s-1.
.......................................................................................................................... 115
Figure 5-6: A summary of the material model assumed to model PVB. ............................... 121
Figure 5-7: The lowest and highest available strain rate data were used to fit the two spring
models. The figure shows the raw data and the model fits............................... 124
Figure 5-8: Low strain rate Prony series with Hoo Fatt spring fit for the tests at rates 0.01 s-1
and 0.02 s-1. ....................................................................................................... 127
Figure 5-9: Low strain rate Prony series with Hoo Fatt spring fit for the tests at rates 0.1 s-1
and 0.2 s-1. Hooper’s data are shown for the 0.2 s-1 case.................................. 127
Figure 5-10: Low strain rate Prony series with Hoo Fatt spring fit for the tests at rates 2 s-1
and 8 s-1. ............................................................................................................ 128
13
Figure 5-11: High strain rate Prony series with Hoo Fatt spring fit for the tests at rates 20 s-1
and 60 s-1. .......................................................................................................... 128
Figure 5-12: High strain rate Prony series with Hoo Fatt spring fit for the tests at rates 200 s-1
and 400 s-1. ........................................................................................................ 129
Figure 5-13: Low strain rate Prony series with the reduced polynomial spring fit for the tests
at rates 0.01 s-1 and 0.02 s-1. ............................................................................. 132
at rates 0.1 s-1 and 0.2 s-1. Hooper’s data are shown for the 0.2 s-1 case. ......... 133
at rates 2 s-1 and 8 s-1. ....................................................................................... 133
Figure 5-16: High strain rate Prony series with the reduced polynomial spring fit for the tests
at rates 20 s-1 and 60 s-1. ................................................................................... 134
Figure 5-17: High strain rate Prony series with the reduced polynomial spring fit for the tests
at rates 200 s-1 and 400 s-1. ............................................................................... 134
Figure 5-18: Finite deformation viscoelasticity material model fit for the rates 0.02 s-1, 2 s-
1
and 8 s-1. .......................................................................................................... 135
Figure 5-19: Finite deformation viscoelasticity material model fit for the rates 20 s-1 and 200
s-1. ..................................................................................................................... 136
Figure 5-20: Relative fit error for the 0.02 s-1 strain rate case. .............................................. 138
Figure 5-21: Relative fit error for the 8 s-1 strain rate case. ................................................... 138
Figure 5-22: Relative fit error for the 200 s-1 strain rate case. ............................................... 139
Figure 6-1: Diagram of a typical single crack experimental sample. .................................... 143
Figure 6-2: Section showing the expected PVB behaviour during the tests. ......................... 144
Figure 6-3: Image of a single crack sample, obtained using a high speed camera. ............... 145
Figure 6-4: Experimental stress-displacement data for a 1 m/s loading speed on a 10 mm
crack spacing specimen. ................................................................................... 146
Figure 6-5: Modified experimental picture with white lines inserted at delamination and glass
edges locations. ................................................................................................. 147
Figure 6-6: Main dimensions and features of the single crack delamination models. ........... 150
Figure 6-7: Detail of the mesh used in the model. The view shows a corner of the model. .. 151
Figure 6-8: Cohesive zone model stress-displacement law used in this study. ..................... 152
Figure 6-9: Peak stress against the strain rate from Jagota [76]. ........................................... 154
Figure 6-10: Boundary conditions applied in the model. ...................................................... 155
Figure 6-11: Steps undertaken to perform the delamination energy analysis. ....................... 156
14
Figure 6-12: Front view of a multiple cracks, 20 mm spacing model. .................................. 158
Figure 6-13: Front view with dimensions of the three inclined cracks models. .................... 159
Figure 6-14: Main dimensions and components of the staggered cracks models.................. 160
Figure 6-15: Delamination progress time history for 0.1 m/s test. ........................................ 161
Figure 6-16: Typical stress distribution in a single crack sample during deformation. The
stresses are given in Pa. .................................................................................... 163
Figure 6-17: Comparison of the models with and without PVB ends. .................................. 164
Figure 6-18: Stress-Displacement plot of the 0.1 m/s model with high rate material properties
results when run with and without mass scaling. ............................................. 165
Figure 6-19: Stress-Displacement plot of the 10 m/s model results when run with different
values of cohesive elements stiffness k. ........................................................... 166
Figure 6-20: Stress-displacement results of the model run at 0.1 m/s with low rate material
properties. ......................................................................................................... 168
Figure 6-21: Plots of the stress results for a 1 m/s model at several time points. .................. 168
Figure 6-22: A detail of the side elevation of the model taken at 25 ms. .............................. 169
Figure 6-23: Plots of the stresses of a 20mm crack spacing model at several time points. ... 171
Figure 6-24: Stress plots at several time points during the analysis of a 33.7° inclination
model. ............................................................................................................... 173
Figure 6-25: Plots of the stress results of a 10 mm stagger model at several time points. .... 175
Figure 6-26: Section view of stress contours taken at different time points highlighting the
movement of the delamination front. ............................................................... 176
Figure 6-27: Stress-displacement plot for the FEA model with cracks 10mm out of alignment.
.......................................................................................................................... 176
Figure 6-28: PVB strain rate in single crack models. ............................................................ 178
Figure 6-29: PVB strain rate in two multiple crack models. ................................................. 180
Figure 6-30: Plot of the plateau stresses versus the crack concentration. .............................. 181
Figure 6-31: Plot of the plateau stresses versus the inclination of the cracks........................ 181
Figure 6-32: Plot of the plateau stresses versus the amount of crack misalignment. ............ 182
Figure 6-33: Model detail showing stresses in the glass and PVB for a low rate PVB material
model with 5 mm crack misalignment. ............................................................ 183
15
List of Tables
Table 3-1: Incident pressure results. ........................................................................................ 56

Table 3-2: Reflected pressure results. ...................................................................................... 57
Table 4-1: Summary of main blast parameters of the tests considered in this work. .............. 84
Table 4-2: Summary of linear elastic material properties applied in the FEA models. ........... 86
Table 4-3: Dimensions used for reaction calculations. ............................................................ 89
Table 4-4: Johnson Cook fit coefficients for Test 2. The results for each window side are
shown. Refer to Figure 4-17 for a legend of the sides’ locations. ........................ 99
Table 4-5: Summary of the pre-cracked out-of-plane peaks and out-of-plane and in-plane
plateaux for the blasts considered. ...................................................................... 107
Table 5-1: Material parameters for the Hoo Fatt’s spring. .................................................... 129
Table 5-2: Prony series parameters for the Hoo Fatt’s spring viscoelastic model................. 130
Table 5-3: Material parameters for the reduced polynomial hyperelastic spring. ................. 131
Table 5-4: Prony series parameters for the reduced polynomial viscoelastic model. ............ 131
Table 5-5: Hyperelastic spring parameters for finite deformation viscoelasticity. ................ 136
Table 5-6: Damping function parameters for finite deformation viscoelasticity. .................. 136
Table 6-1: Experimental plateau stresses used in the analysis [3]. ........................................ 146
Table 6-2: Summary of the material models employed in FEA models................................ 153
Table 6-3: Peak stresses used in the models. ......................................................................... 154
Table 6-4: Experimental delamination speeds for different crack configurations and
displacement speed. ............................................................................................ 162
Table 6-5: Results of models where the delamination peak stress was varied by 25%. The
original cases are included for comparison. ........................................................ 167
Table 6-6: Summary of energy and delamination speed results for single crack models...... 170
Table 6-7: Results of the models with different crack spacings. The experimental results are
included where relevant for comparison. ............................................................ 172
Table 6-8: Results of the models with different crack inclination. ........................................ 174
Table 6-9: Results of models with not aligned cracks on the two laminate faces. ................ 177
16
Nomenclature
αo Ogden spring function constant
αi Hoo Fatt’s hyperelastic springs constants
ε Strain
εɺ Strain rate
εɺ0 Reference strain rate for Johnson-Cook
ε1 Strain on top face of window support steel plate
ε2 Strain on bottom face of window support steel plate
εa Axial force strain of window support steel plate
εb Bending strain of window support steel plate
εp Plastic strain for Johnson-Cook
η Damping function
θ Angle of force F from the in plane position
λ Material’s stretch
λɺ Rate of stretch
λi Inelastic stretch
λe Elastic stretch
µo Ogden spring function constant
µi Hoo Fatt’s and hyperelastic springs constants
ρr Material’s density
σ Stress
17
σ o (ε ) Hyperelastic stress function dependant on strain
τi Prony series’ individual relaxation time constants
ψ Hyperelastic spring’s work function
ψɺ Rate of work
ψeq Equilibrium spring’s work function
ψoe Overstress spring’s work function
A Johnson-Cook constant
B Johnson-Cook constant
B Left Cauchy-Green strain tensor
Bi Inelastic Left Cauchy-Green strain tensor
Be Elastic Left Cauchy-Green strain tensor
C Johnson-Cook constant
C1, C2, … Reduced polynomial’s hyperelastic springs constants
Cη1, Cη2, … Damper’s function constants
D Horizontal distance between window’s reaction centroid and strain gauges

pair
d Window support steel plate’s thickness
d0 Lever arm between inclined PVB membrane and centre of gauge’s pair.
E Young’s modulus
F Membrane force applied by PVB membrane on supports
F Deformation tensor
Fa Axial force on support plate
Fe Elastic portion of deformation tensor
18
Fi Inelastic portion of deformation tensor
Fij Deformation tensor’s component
I Second moment of area
I Identity matrix
I1B , I 2 B First and second invariants of the total Left Cauchy-Green strain tensor
I1B e , I 2 Be First and second invariants of the elastic portion of the Left Cauchy-Green
strain tensor
g∞ Prony series’ long term stiffness constant
gi Prony series’ individual stiffness constants
g (t ) Time dependent Prony series
H Vertical distance between window’s membrane centroid and strain gauges

pair
k Stiffness parameter for Cohesive Zone Model
L Material’s velocity gradient
M Moment on support plate
N Number of series’ elements used
n Johnson-Cook constant
P Blast pressure
p Hydrostatic pressure
P0 Atmospheric pressure
Pr Reflected pressure
Ps Incident pressure
Pp Peak blast overpressure
19
R Stand off distance
S Peak delamination stress for Cohesive Zone Model
S n , S s , St Delamination stresses in the normal and the two shear directions
S no , S so , Sto Peak delamination stresses in the normal and the two shear directions
T Cauchy true stress tensor
t Time
Toe Overstress spring Cauchy true stress tensor
tp Duration of the ositive phase of blast pressures
X Material’s elements’ original position coordinates’ vector
x Material’s element new coordinates’ vector
Xj Material’s elements’ original position coordinates’ vector component
xi Material’s element new coordinates’ vector component
y Maximum distance of the material’s fibers from section’s centroid
Z Blast scaled distance
W Equivalent TNT charge weight
20
Acronyms
CFD Computational fluid dynamics
CZM Cohesive zone modelling
DIC Digital image correlation
FEA Finite element analysis
FPS Frames per second
LVDT Linear variable differential transformer
PVB Polyvinyl butyral
SDOF Single degree of freedom
TNT Trinitrotoluene
21
1. Introduction
In today’s society the security of buildings needs to be the object of careful considerations
due to the risk of both terrorist attacks and accidents. Large structures whose use causes a
significant amount of people to congregate in their interior or vicinity, such as airports and
office buildings, represent ideal targets for malicious attacks. Therefore the protection of
these buildings is of high importance to guarantee the safety of bystanders during wilful or
accidental explosions.
These structures are generally the object of careful architectural design, which often includes
the use of large glazed facades for both aesthetic and natural lighting reasons. The failure of
such glazing elements, and especially the glass shards produced in those situations, represent
one of the main sources of injuries during blasts [1]. When the glazing elements break, glass
parts can be propelled both inside and outside the building, injuring and potentially killing
people in both areas. Additionally, once the building envelope is pierced, blast pressure
waves will be able to penetrate the internal space. These will cause further injuries and, even
where this is not the case, will cause significant damage to the internal fittings and
equipment. This will add considerably to the overall cost and time required to restore the
structure to its original function. Therefore, the design and detailing of the glazing elements
is a very important part of the strategy to guarantee the resilience of buildings.
Several materials have been used for this purpose. Typical building glazing is built with
monolithic glass panes. Either annealed glass or toughened glass can be used for these. Both
of these materials will fail catastrophically once their limiting stress is reached. In the case of
annealed glass, this maximum is determined by the presence of flaws in its matrix. Once the
stress capacity is exceeded at one of these flaws locations, the cracks propagate quickly
throughout the pane. Generally, relatively large shards with sharp edges are produced, their
pattern depending on the exact stress and defects distribution. In the case of a blast, these
fragments can be propelled in the building space, causing significant injury. Toughened glass
panes, instead, underwent a tempering process to improve their capacity. A layer of material
in a compression state is created on all surfaces, increasing the overall tensile stress limit of
the structure. However, once this is exceeded, the whole panel fragments suddenly in small,
22
dull pieces. Whilst these in a normal application would not present the same level of risk as
sharp shards, during a blast they can be propelled at high velocity, becoming a significant
hazard. Therefore, both types of glass are not ideal for use in application where blast
resistance is a priority. Figure 1-1 shows the effect of a blast on a monolithic glass façade,
highlighting the amount of damage caused by in these situations. The picture was provided
by David Smith at Arup Resilience Security and Risk.
Figure 1-1: Effect of an explosion on office buildings in London’s Docklands area.
Laminated glass is a composite material which has been used for applications where resisting
high levels of load is a key requirement. The material is composed of layers of glass,
annealed or toughened, interposed with polymer membranes to which they are bonded.
Several types of polymers are used for the internal membrane, the choice depending on the
details of the glazing’s application. For blast resistance the most common material used is
Polyvynil Butyral (PVB). This is bonded to the glass using a heating process at high pressure
carried out in an autoclave. Whilst the designer is free to choose the thickness of the glass,
PVB is produced in minimum thicknesses of 0.38 mm, and supplied only as multiples of this
measure. A thickness of 1.52 mm is generally recommended for blast resistant designs [2].
During a blast, once their capacity is reached, the glass panes fragment. The splinters will
however remain attached to the layers of PVB, which will continue to transfer the loads to the
glazing’s supports. The membrane will also prevent the pressure waves entering the building
23
space. Whilst the glazing units will need to be substituted after the event, this still represent a
significantly more positive outcome than those typical of other kinds of glazing.
1.1. Project aims

The main challenge for designers is to decide on appropriate approaches to design all the
elements of the glazing which will affect its performance. These include, between others, the
thickness and number of the constituent layers and the restraint mechanism. The connection
of the glazing panel to the supporting structure is especially important in this respect. In this
project, some of these aspects have been considered, with a special attention to the reaction
forces produced in a blast test. An understanding of their absolute level and of their evolution
during a blast is essential to decide on specific laminate and connection designs.
The results of a new blast test have been examined to assess whether the reactions change
along a window’s perimeter. The test was performed on a standard size window using a
realistic explosive quantity and stand-off distance. Optical measuring techniques and more
traditional strain gauges were employed to collect the necessary data. The results from the
blast test, together with data from blast tests performed by others [3], have been used to
calculate reaction forces along the full length of the edges. The optically collected strain data
were related to the gauge’s information to produce a cracked laminate overall material
property law, which was employed to estimate the reactions.
Following this, the characteristic of the cracked laminate material was studied to explain the
observed features of the reaction forces. Firstly, the PVB material was characterised. Existing
experimental data were used to fit material properties laws covering the high strain rate
behaviour relevant to the situations of interest. Viscoelasticity was employed to achieve this.
Several formulations were utilised, including Prony series and a full finite deformation non-
linear viscoelastic model. The results of all the approaches were finally be compared.
Finally, the delamination of the glass fragments from the PVB membrane was considered.
The PVB material properties previously derived were used in finite element analysis (FEA)
models to determine both the physical properties of the process and the effect of different,
realistic glass configurations on its macroscopic characteristics, specifically on the material’s
overall stresses. The aim of this work was to not only identify some likely levels of blast
loading reaction forces, but also to clarify their time history behaviour and the likely effect of
different glass fragmentation patterns.
24
2. Literature Review
Several aspects of the blast loading of windows are of interest for this research. These cover
both the blast loading and the properties of the composite, in term of its constitutive
materials, their interaction and the overall behaviour. Knowledge of these topics is necessary
to then attempt to interpret and predict the performance of the system in a realistic situation.
2.1. Blast Pressure Theory

The pressures originating from explosions have been studied by several authors in the past,
attempting to provide methods to describe both the general behaviour of the blast waves and
how they interact with structures in their path. Some analytical solutions have been proposed
for a selection of these issues, whilst empirical or numerical methods are commonly
employed to describe the more complex cases. These will be examined below.
2.1.1. Blast wave characteristic and interactions with structures
The pressure’s time history is initially characterised by a sharp rise of its level at the arrival
of the shock wave. This is followed by an exponential decrease and eventually a negative
phase, where suction is imposed on the supporting structure. Its general shape is shown in
Figure 2-1. The area under the positive portion of the curve is the impulse which, together
with the maximum pressure, represents a useful measure of the total energies imposed on
structures.
25
Overpressure
Time
Figure 2-1: Typical shape of blast pressure wave time history.
The general time history of pressures can be described by the Friedlander equation [4]:
t
−
tp  t 
P = PS e 1 − p  Eq. 2-1
 t 
Where P is the blast overpressure at time t, Ps is the peak overpressure and tp is the positive
phase duration. The equation is able to reproduce accurately the time history of the majority
of the blast waves it is fitted to. The different pressures produced by distinct charges can be
compared using scaling laws. One of these is described by Hopkinson [5] and Cranz et al. [6],
and is given by:
1
Z = R /W 3 Eq. 2-2
Where Z is a scaled distance, R is the stand off distance of interest and W is the equivalent
TNT charge mass. Blasts with the same scaled distance should in theory produce comparable
effects. Should different types of explosives be used, their mass needs to be converted to an
equivalent TNT one. This can be done using similitude parameters, which are also used to
produce more complex predictions of blast effects. For example, Geers and Hunter [7]
26
present several similitude constants for different types of explosives to be used for blast wave
predictions.
The interaction of blast waves and structures is also the object of several studies. The
phenomenon of wave reflection is of special importance, as it can amplify substantially the
loads imposed on targets when compared to the free field overpressures. Kingery and Coulter
[8] studied the reflection of blast waves on finite targets. They performed 15 tests using eight
targets at different distances to obtain data for different overpressure levels. After each test
the targets were rotated by a set number of degrees to collect data about reflections at
different angles. The collected data highlight the influence of the angle of incidence on the
level of reflected pressures. Additionally, for 0º cases, the classic equation derived by
Rankine and Hugoinot can be employed [9]:
 7 P + 4 Ps 
Pr = 2 Ps  0  Eq. 2-3
 7 P0 + Ps 
where P0 is the atmospheric pressure, Ps is the incident pressure and Pr is the reflected
pressure. It should be noted that this is only valid for the limiting case of an infinite target,
and hence cannot be used to convert the entirety of the blast pressure time history for a real
structure. The initial peak though can be estimated, as size effects will be less significant at
the early stages of loading.
2.1.2. Loading estimate methods
Whilst the methods presented above can provide guidance on the general behaviour of the
blast waves, it is challenging to predict blast loading on specific target configurations
analytically. Several empirical and FEA methods have been proposed to obviate this limit.
Firstly, empirical solutions based on results from blast tests can be used [9]. These have been
implemented in CONWEP [10], a software tool utilised to predict realistic loading
considering explosive characteristics and target geometries. The issue with this method is
that, where the analysed situation differs significantly from the tests used to derive the
implemented functions, the results produced might not be realistic.
Alternatively, FEA software packages which can simulate the whole blast event exist. Air3D
[11] is a relatively easy to use tool which implements computational fluid dynamics (CFD)
27
principles to simulate both the explosion event and the propagation of the pressure waves in
potentially complex geometries of targets. The limitation of this tool is that fluid structures
interactions are not included, and hence all the obstacles are treated as perfectly rigid. The
only possibility offered consists of including a failure pressure for specific panels, which will
then be suddenly removed if the results reach this limiting value. Other software, such as
ANSYS AUTODYN, can couple the CFD analysis with a FEA model of the targets,
including therefore the effects of structural movements on the pressure wave estimation.
However, the use of such complex tools requires significant user skill and computer time,
precluding their utilisation in at least some design situations.
2.2. Behaviour of Laminated Glass

Laminated glass is a composite material; therefore to explain its behaviour fully the material
properties and interaction of the various constituent parts need to be considered carefully.
Whilst no Eurocode building standard is currently available to guide its design, the ASTM
E1300 standard [12] can be used to design the glazing elements to resist wind and other
horizontal forces. Blast loading can be included in the analysis using codes such as ASTM
F2248 [13], which provide guidance on the derivation of equivalent 3 s loads to represent the
blast pressures. These codes can therefore be used to achieve an approximate design for the
panels. However, as the capacity design principles are based on a definition of failure
coinciding with the cracking of the glass, the engineering solutions reached using them would
likely be significantly overdesigned. More complex design methods, based on either single
degree of freedom (SDOF) approaches or FEA analysis require the estimate of more precise
parameters for the constituent materials and their interaction. The literature concerning these
will be considered in turn.
2.2.1. PVB material model
The PVB material presents a non-linear, time dependant behaviour when subject to impulsive
loads of the magnitudes of interest. Several studies have been performed in the past few years
attempting to characterise it at a range of different strain rates. Some research concentrated
on achieving estimates of the material modulus for use in design. Vallabhan et al. performed
28
shear tests on laminated glass samples and obtained estimated of the modulus at different
strain levels [14]. Others, like Du Bois et al and Timmel et al, attempted to apply a
hyperelastic law to simulate the material’s characteristics. They employed experimental
results to fit a hyperelastic model of the PVB material. Several different spring functions
were considered, using a Mooney Rivlin representation for their final impact model [15],
[16].
Several authors then attempted to fit viscoelasticity laws to model the time dependant part of
the deformations. Xu and Li have used elastic models which were used for windscreen
impact simulations [17]. Subsequently, they performed tests at strain rates of up to 4500 s-1
using a split Hopkinson bar apparatus to obtain stress strain curves. They then fitted a
viscoelastic law to this, assuming a Mooney Rivlin hyperelastic spring to account for the non-
linearity [18].
Iwasaki and Sato identified a shift in the PVB behaviour at high strain rates of 113 s-1 [19]
and they included a viscoelastic material model fit for the lower strain rates cases (below 1
s-1) [20]. Liu et al. also performed several tests at different strain rates, from 4x10-4 s-1 to 118
s-1) both in tension and in compression. Instead of trying to obtain a single material property
law covering the full range of their tests, they the fitted the data with several viscoelastic
material laws to cover the different rates and stress regimes [21].
In general, except for the work of Liu et al., these attempts did not include a model for the
high strain rate behaviour of the material. This was generally included in research concerned
with modelling the full composite using linear elastic – linear plastic laws with a strain rate
dependency. For example, both Larcher [22] and Zhang et al. [23] applied this approach in
their work. However, none of the proposed models can cover all ranges of strain rates. This
could be important when considering blast loads, as it is likely that different areas of PVB
will be deformed at different speeds.
2.2.2. Glass behaviour
The capacity of annealed glass is one of the parameters necessary to model the laminate’s
overall response. Griffith [24] suggested that material flaws have a strong influence on
failures in brittle materials. These are present in significant quantities in annealed glass, their
presence caused both by the production methods and by the subsequent use of the panels. A
large study with a high number of glass samples was performed to estimate breaking
strengths. A ring on ring test was used for these to avoid asymmetries in the stress
29
distribution in the samples [25]. This showed that samples failed between 30 and 120 MPa,
with a 95% probability that the capacity would be greater than 45 MPa. This was then
extrapolated to account for rate effects, and a failure stress of 80 MPa was calculated for
impulsive loadings [3]. Recently, some authors started including more complex, strain rate
dependent failure models in FEA models of laminated glass [26]. The Johnson Holmquist
Ceramic constitutive model [27] was implemented in the software and used for this.
2.2.3. PVB – glass delamination
The failure of the bond between PVB and glass is of great importance to explain the overall
behaviour of laminated panes. After the glass layers reach their capacity and fracture, the
PVB membrane will tend to expand. The higher strains will occur at the crack locations,
creating significant stress concentrations at the layer’s connections in the vicinity of the edges
of the glass fragments. Delamination is likely to take place at these positions and it is possible
that this could absorb a significant part of the total energy imposed in the system.
Some studies have been performed to analyse this behaviour. Rahul-Kumar et al. [28]
performed compressive shear tests at different rates to obtain estimates of the cohesive
strength of the bond. The results show the significance of strain rate for the delamination
characteristics, with different peak stresses being reached during the different tests.
Muralidhar et al. [29] performed tensile tests of laminated glass specimens with a central pre
crack. The force-displacement curves obtained reached a delamination plateau stress, which
was maintained for much of the deformation. The author then used the results to determine
delamination energies. The applied external work was calculated, and the strain energy of the
PVB was deducted from this. It was then assumed that the remaining portion of the energy
was absorbed by the failure progression. A hyperelastic material law was used for the PVB,
ignoring strain rate effects. Butchart and Overend [30] used a similar approach. In her study a
very similar experimental set up was used and two different delamination speeds considered,
albeit both were low compared to those observed during blasts. A linear viscoelastic model
was used for the membrane. The energy estimates were then produced with a closed form
analytical method and compared. The results indicated rate dependency in the absorbed
energies. However further investigations were deemed necessary to establish whether this
was a real effect or an artefact of uncertainties in the material models used.
30
Iwasaki et al. [20] also employed a similar test set up to measure delamination energies. In
this occasion, only one layer of glass was laminated to the PVB. Additionally, an initial area
of PVB was already detached before the commencement of the loading phase. Higher speeds
were achieved, up to 2.5 m/s. As mentioned above, a viscoelastic material law was used for
the PVB. Whilst the method used for the calculation is not discussed in depth, energies were
extracted, showing that a similar amount was absorbed at the speeds considered.
Hooper [3] also performed the same type of tests, applying though a greater range of speeds,
up to 10 m/s. Additionally, cases with more complex crack patterns were considered to
represent more closely the behaviour of a cracked laminate panel.
2.3. Overall response of laminated glass

Several studies, both experimental and analytical, have been conducted on the overall
response of laminated glass to different loading situations, ranging from relatively slowly
varying loads to impulsive impact and blast forces.
2.3.1. Low rate loading
The low rate behaviour of the material is of interest for the design of both glazing subject to
regular wind loads and structural elements, such as beam and columns.
Norville et al. [31] and Behr et al. [32] performed studies on the strength of laminated glass,
comparing it with annealed glass of similar thicknesses. They analysed static and transient
wind loading, reaching the conclusion that the interlayer does seem to have a role in
transferring shear forces, creating a composite action between the two glass plates, rather
than simply acting as a spacer element.
Other quasi static studies were conducted in the context of the use of laminated glass for
structural elements, such as beams or columns. As an example, Galuppi and Royer-Carafagni
[33] concentrated on laminated glass beams. They formulated an analytical, time dependent
solution based on linear glass properties and a linear viscoelastic PVB material law. They
then compared the deflection results with a fully linear solution and with numerical FEA
31
results. The authors’ conclusion was that the linear solution would significantly
underestimate the deflections under highly impulsive loads, such as blast wave pressures.
2.3.2. Impact loading
Several studies were performed concerning impact loading of the material. This is an
important situation for the use of this composite system in the automotive and aeronautical
industries. Also, whilst some of the load characteristics are different from what seen during
blasts, there are several similarities, such as the magnitude of the rate of loading and the
expectation that the glass plies will fracture. Several impact tests have been performed on
laboratory samples. Bennison et al. [34] conducted impact tests on circular plates of
laminated glass. The data were compared with FEA simulations. The authors applied a linear
elastic – viscoelastic material law to represent the PVB membrane and did not include a glass
failure mechanism. They modelled instead the glass as pre fractured by reproducing the
observed laboratory crack distribution in the model’s initial geometry. The results were used
to study the stress distribution through the glass thickness. The work was developed in a
following paper by the same group [35], where force displacement graphs were produced and
delamination was included in the finite element analysis. Pyttel et al. [36] and Du Bois et al.
[15], [16] also performed impact tests and modelling. In this case the fracture of the glass was
included through the use of an erosion mechanism, where the stiffness of glass elements was
reduced to 0 when a limiting stress was reached. To achieve an improved agreement with
empirical results, Pyttel et al. introduced an additional total strain energy factor was included
to determine the sudden onset of cracking.
2.3.3. Blast loading studies
The blast loading case has been studied extensively, both with numerical and experimental
studies. Some of these results have then been used to produce simplified design approaches
which can practically be used in real projects.
32
Experimental studies
Several experimental programmes have been carried out to study the blast capacity of this
material. Hooper et al. [37] and Stephens [38] performed blast tests measuring the deflections
of laminated glass panes. Hooper’s data is especially interesting as he used digital image
correlation (DIC) techniques to obtain the full field 3D deflection and strain data, which will
be used for further analysis. The same measurement technique was also employed by Kumar
and Shukla [39]. This study used shock tube experiments on smaller samples comparing the
performance of several different types of glazing materials. Wired glass, sandwich glass
panels without interlayer membrane and laminated glass with the addition of protective films
on both exterior surfaces were tested. DIC was employed to obtain full field deformation
information and the tested samples were also analysed to compare their damage levels. The
author concludes that laminated glass provided the best option to limit the effect of blast
pressures, outperforming the other options on all the measured quantities.
Analytical solutions
Attempts have been made to analyse the observed responses using analytical solutions. Wei
and Dharani [40] took such an approach and carried out studies using von Karman large
deflection theories. They used this theory to estimate the deflection of non cracked glass
panes and the probability of their failure. In another paper [41] the same authors combined
these results with Griffith fracture energy balance theory to estimate the likely locations and
density of cracks occurring at failure. Whilst this approach produced some realistic results for
the pre-crack phase of deformation, the complexity of the post cracking behaviour caused this
approach to not be as conclusive.
FEA studies
To overcome the issues with analytical solutions, FEA approaches were employed by several
authors. Wei and Dharani [40] used FEA models to validate their analytical solutions.
However their models were generally limited to the pre-crack phase. A possible technique to
model the post glass cracking phase of the deformation consists in using a smeared material
property for the composite, without considering its constitutive elements separately. Such an
33
approach was used by Hooper [3]. In his work he produced two separate models for the pre
crack and post crack phases. He used both linear elastic and viscoelastic models for the PVB
formulations before the glass failure. When a limiting stress was reached in the glass pane,
the analysis was stopped and restarted with post-cracking material properties. The stiffness of
the glass was reduced to 0 throughout and a plastic material model derived from laboratory
experiments was applied to the PVB. The model deflection predictions are similar to the
experimental data at early stages of the analysis, however the deflected shape and peak
deflections diverge from the available data at later times in the analysis.
Recently more works have been produced using a single model attempting to accurately
represent all phases of the system’s behaviour. These models use a variety of 2D and 3D
element formulations, including differing detail levels on material’s properties and failure
conditions. Larcher [22] performed a FEA analysis of this type. He used a layered model,
employing an erosion technique to model failure. The author then applied a linear elastic
PVB model for low strains. A limit was set, transforming the behaviour of this material to
perfectly plastic beyond this point. The results of the model were then compared to
experimental information and to the results of a smeared property model. The data showed
that the authors’ formulation matched experimental deflection to a greater extent than simpler
models. Zhang et al. [23] also developed FEA models of windows subject to blast loading. In
this case the aim was to produce pressure-impulse curves for use in design. In the study,
dynamic properties were used for both the glass and PVB failure. Delamination was also
included, albeit the authors lamented the lack of published dynamic properties for this
element.
FEA of the system was also used to assess the individual contributions of its elements to
energy absorption. Hidallana-Gamage et al. [42] produced such a study. In this work the
authors used dynamic material models for the main components, including also the structural
sealant. Failure parameters were also included for all individual components, however glass –
PVB delamination was not included. The deflection, stresses time history and the energy
absorption of each element were then compared, providing guidance on their relative
importance. Amadio and Bedon [43] also performed numerical analyses of the blast events,
similarly to the aforementioned authors. In this case though the focus was to verify whether
viscoelastic absorption devices used as glazing support units could prevent some of the
damage to the panels. The results showed that these could absorb significantly more energy
than the standard options, improving the outcomes of the pressure loading.
34
SDOF approaches and design aids
The studies highlighted above attempted to consider several aspects of the composite’s
behaviour in detail. The results produced, though they agreed with experimental observations
to a varying degree, seemed to generally be able to consider their key characteristics and
could therefore be used for practical design work. However, their computational and
expertise requirements are likely to be high and ill suited to the constraints of design offices.
Historically, single degree of freedom (SDOF) simplifications have been used as a relatively
accurate tool to produce designs within acceptable time scales. In this approach, the complex
system of the laminated glass and its connections to the supports is approximated with an un-
damped spring model. Failure is considered to occur when a limiting deflection is reached.
This limit is generally estimated from the results of blast trials performed in the past. The
application of this approximation necessitates the introduction of mass and stiffness constants
to produce estimates of the effective mass, stiffness and external force in the system. The
values for these parameters have generally been calculated assuming large deflection bending
to represent the deformation of the glass plate.
To calculate the support reactions, a key element of the design, a resistance function is
assumed. This represents the internal force resisting the deflection. If the system behaved in a
static manner, the function would be equal to the externally applied force which would cause
the same magnitude of deflection [44]. Besides this limiting case, for simple system, such as
a beam undergoing small amplitude vibrations, the resistance can also be calculated directly
from structural principles. However, in the case of a blast loaded window’s deflections,
especially after cracking, obtaining such an estimate is not as straightforward. At present, a
shape for the force function is assumed and calculations are carried out with reference to
previous tests to verify its magnitude [1]. Factors can then be applied to this resistance and
the externally applied force to produce estimates of the reactions on each of the four window
sides [44].
This approach has been used by several authors in the past. Smith [1] suggests a linear
resistance function’s shape after the glass cracks. He then states that the curve presented has
been used to produce estimates of the window capacity. Fischer and Häring [45] also
conducted analysis of blast experimental data attempting to fit linear resistance functions.
They then compared the deflections obtained from single degree of freedom models with
35
those obtained during the experiments. They conclude that a tri-linear response function is
reasonable at representing the deflections observed during the tests.
The SDOF approach was used by Stewart and Netherton [46], [47]. They produced estimates
of the risk probability of damage from explosions, including both parameters such as
likelihood of blasts taking place in the target’s vicinity and, using the SDOF approximation,
the structural capacity of the glazing. Morrison [48] concentrated his work on improving the
parameters estimation for SDOF methods. He studied the properties of the constitutive
elements of the composite to produce closer approximations for use in industry.
The results from tests, these SDOF calculations and FEA models can be also employed to
produce pressure-impulse diagrams with iso-damage curves, as done by Zhang et al. [23].
These can then be used to produce quick designs for specific windows. Their main limitation
though is that the validity of the results will be restricted to the specific sizes considered
when the diagrams were generated.
2.4. Restraint mechanism

One of the key aspects of window blast design is ensuring that the glass pane will be
contained in the frame during the event. The restraint mechanisms employed for laminated
glazing are important to guarantee this and need to be able to resist the reaction forces
produced.
Structural silicone
One of the most common systems used for this purpose, especially with laminated glass, is
structural silicone. The behaviour of this component has been the object of several studies.
Ramesh et al. [49] studied the effect of aging and specimen size on the elastic properties of
the silicone material, concluding that, whilst aging has a significant impact on the Young
modulus of the material, the length of the specimen did not seem to. Meunier et al. [50]
carried out tests on silicone material to characterise its properties. Tensile, pure shear,
compression, plane strain compression and bulge tests were carried out, using digital image
correlation techniques to obtain the full strain distribution across the samples. The authors
then used the data to fit five hyperelastic material models, namely the Neo-Hookean,
36
Mooney, Gent, Haines and Wilson and Ogden models. These models were used to simulate a
tensile experiment on a silicone membrane with several holes. They then compared the FEA
results with experimental results for the membrane, finding that the models seemed to
represent the material satisfactorily. Their material laws are therefore relevant for attempting
to model the whole window system subject to blast.
Hooper [3] performed tensile experiments aimed at assessing the sealant’s capacity under
realistic blast loading. Structural silicone beads were applied to glass and aluminium
substrates. The tests consisted of pulling the silicone at different angles using a high speed
Instron servo hydraulic machine. This was thought to represent the load applied during a
blast, as the window will deflect during the event and the force on the silicone will be applied
at a varying angle. Hooper managed to use his results to derive a limiting capacity for the
silicone being examined. This was then applied together with PVB strength data to obtain
estimates of the maximum useful width of silicone to be used to fix blast resistant windows.
Glazing tape
Glazing tapes are increasingly being considered a possible substitute for silicone. New 3M
[51] and Tremco [52] products are being marketed for blast resistant uses and are therefore
attracting a rising amount of research to determine their properties. Most of the studies
concentrates on low strain rates events, covering, for example, the effect of wind loading.
Townsend et al. [53] produced a study of cumulative damage due to wind loading. The
authors found that glazing tape designed according to industry standard for the peak 3 s gust
will accumulate damage, however it will not generally be at risk of failure for the duration of
its design life. The same authors [54], [55] also carried out work characterising the material
properties of 3M VHB tape. They subjected the tapes to ramp to fail and creep tests. The tests
were carried out at different temperatures and strain rates, the speed of the machine ranging
from 5 mm/min to 500 mm/min. They produced a set of storage and loss moduli for the
material and an estimate of the glass transition temperature of the material. The study though
did not cover the higher strain rates commonly observed during blast events.
A 3M study conducted at the university of Waterloo [56] analysed VHB tape samples tested
at higher strain rates. During the study both a high speed Instron and a Split Hopkinson bar
37
were used to produce the data, reaching rates of approximately 7000 s-1. The produced
diagrams could be used to evaluate the material properties at high strain rates.
2.5. Conclusion
The research described above shows that a significant amount of research has been performed
in the past concerning the blast phenomena and the laminated materials of interest to this
study.
The behaviour of the blast pressure waves has been studied with numerous field tests and
analytical studies from an early age. Therefore several methods are available to predict the
pressure intensities and the imposed loading on structures in their path. Whilst some of these
are of complex application, relatively quick estimates can be obtained with empirical
solutions and CFD applications.
The constituent materials of the composite were also considered by several authors. However,
whilst simplified rate dependant solutions do exist to model their response to loading at rates
above quasi static, there are few studies considering this aspect thoroughly. This is an
important factor, especially for the PVB membrane, as it is likely that its behaviour will be
complex, varying in the area of the glazing panels. Modelling it will therefore require
formulations of wide applicability. There also seems to be a lack of delamination parameters
derived for higher speed deformations, which might decrease the accuracy of the more
complex simulations proposed by authors.
The overall blast event was the object of both practical tests and simulations. The
experiments performed and reported provide insight on the deflections and strains which can
be observed, together with information on likely failure mechanisms and limits. The results
could though be analysed further and combined with the available and new knowledge
relating to the material properties to provide more detailed information on the forces
developed during the loading.
Experimental results are available to achieve this and it is proposed that this study will
develop these points, using both blast test data and laboratory information to this effect. The
material properties at high strain rates will be studied further and the results employed to
interpret the results from realistic pressure waves loading.
38
3. Full Scale Blast Test
3.1. Introduction
The performance of laminated glazing subject to blast loads is affected by many aspects of
the material behaviour and overall system design. To achieve a safe and efficient solution all
the factors need to be considered to optimise the final product. However the influence of the
properties of individual elements, such as material properties and restraint mechanisms, are
often difficult to differentiate between and quantify. This can lead to significant uncertainties
on how a specific design will perform during an explosion. The most conclusive method to
assess the overall behaviour of a system is to carry out a full scale blast test on a complete
assembly. Within this trial, much data on the performance of specific components can be
obtained using relevant instrumentation. The information can then be used to glean more
understanding on the general effects of blasts and the influence of specific component factors
on the overall performance. In this chapter one such blast test, performed in May 2013, is
described. Whilst several blast tests had been previously carried out by others, such as
Hooper et al. [37], it was realised that additional information could be gathered with a test
aimed at collecting extra data on the reaction forces. The results of this test will be described
in this chapter, together with a novel method for analysing the data.
3.2. Method
The data collection and analysis method included both on site experimental techniques and
post processing analytical techniques. All these were necessary to obtain and interpret the
results of the experiment.
3.2.1. Experimental set up and data collection
The data collection methods used had to take into consideration both the blast set up and the
exact data necessary to achieve the experimental objectives. The test geometry and the
different data collection methods used will be described below.
39
Sample geometry and test pad disposition
The blast test was performed in a dedicated testing facility run by DNV GL Group at the
RAF Spadeadam facility. A 100 kg TNT equivalent charge was used, the sample being
located at a 23 m stand-off distance. The charge weight was dictated by the requirements of
the other tests being performed at the same time. The stand-off distance was adjusted to
maximise the displacement of the window whilst preventing catastrophic failure. To achieve
this a SDOF model was employed to estimate the behaviour of the system before the test took
place.
The test panel consisted of an approximately 1.5 x 1.2 m laminated glass window held in a
steel frame. The laminate make up was two 3 mm thick annealed glass outer layers with a
1.52 mm PVB interlayer, producing a total thickness of 7.52 mm. It was sourced and installed
in its frame by Structural Glazing Limited. A double sided silicone connection to the frame
was provided as the restraint mechanism. Figure 3-1 shows a sketch of the support with the
relevant dimensions. The test assembly was located in a steel cubicle provided by the test
facility. The glazing was mounted into this with the longer sides horizontal. The window
opening was in the centre of the front wall, 0.6 m above the ground. Figure 3-2 shows a
sketch of the front of the cubicle with the relevant dimensions and Figure 3-3 shows the steel
structure composing the front walls. The assembly ensured that the support conditions would
be symmetric on the two vertical sides of the glazing, avoiding differences in the response at
the two locations. The response on the horizontal sides would differ due to the effects of
ground and the free edge.
40
Figure 3-1: Section sketch of the glazing metal supports. All dimensions are in mm.
Figure 3-2: Cubicle elevation with dimensions.
41
Figure 3-3: Steel structure of the front cubicle panel.
In this test, 100 kg of TNT equivalent explosive was employed. The liquid explosive was
held in a spherical container to reduce irregularities in the blast pressure distribution due to
non-symmetric explosive shapes. The explosion was initiated with a detonator placed in the
middle of the charge. The explosive was positioned on polystyrene blocks 1 m high to
minimise blast energy lost in ground shock and crater formation. Figure 3-4 shows a section
of the test set up, showing both the charge and the cubicle containing the glazing sample. The
test pad was shared with other cubicles at different stand offs. Their relative position were
arranged to avoid the interactions of reflected waves.
42
Figure 3-4: Section through the test pad, showing the charge and the cubicle containing
the glazing sample.
The incident, side-on, pressures were collected employing a piezoelectric pressure gauge.
PCB Piezotronics gauges, model 102-A06, were employed to collect the data. These were
mounted on a steel disc aligned on the radius from the charge to avoid diffraction of the
pressure wave around the sensors. The data was collected 1.4 m above the ground, 0.2 m
higher than the centre line of the sample window. Figure 3-5 shows the disposition of the
different elements on the test pad. The pressure gauge available for this analysis was located
25 m away from the charge, a distance 2 m greater than the tested panel, as data collection
requirements for the other targets took precedence.
The test facility provided the recorded data already converted into pressures for use in this
analysis.
43
Figure 3-5: Plan view of the test pad, showing some of the other test cubicles present.
DIC data collection
Digital image correlation techniques (DIC) were employed to collect 3D deflection and strain
data throughout the window surface. This measurement method requires two cameras
orientated towards the sample at different angles taking pictures simultaneously throughout
the test. The pairs of images are then used by a DIC software to track each dot of a speckle
pattern on the sample and hence calculate the displacements and strains occurring across the
area of interest in all three directions.
High speed cameras were employed to capture the transient deformations as these took place
in a time period of only a few milliseconds during the blast. The instruments used were
supplied by the test centre, consisting in a pair of Fastcam MC2 by Photron. This is a system
of two satellite cameras controlled by a single separate processor unit. The equipment is
capable of capturing grayscale images at a maximum resolution of 512 x 512 pixels at a rate
44
of 2000 frames per second (fps), or at a maximum rate of 10,000 fps when reducing the
resolution to 512 x 96 pixels. In this test a resolution of 512 x 352 was used, as this provided
enough field of view to cover the window area. The rate could then be raised to 3000 fps.
The pixel size of the cameras is 10 µm, giving a total sensor size of 5.12 x 5.12 mm, of which
5.12 x 3.52 mm was used with the chosen resolution. Lenses of 8 mm focal length were used.
The cameras were synchronised by the controlling processor unit. Two 1 kW halogen lamps
lit the speckled surface. These were chosen since a high intensity of light was required to
capture the images with short exposure times.
The two cameras were located on an extruded aluminium post on a heavy steel stand. As
ground vibrations were expected to be transferred to the whole cubicle, the stand was isolated
from the ground with rubber pads. Sand ballast was added around the supports to further
minimise its movements. The cameras were positioned 2.5 m away from the window and 1.2
m apart. The height of the camera plane was 1.2 m to coincide with the centre line of the
window pane. Figure 3-6 shows the layout. The angle subtended by the camera system was
27° as shown in the figure. This guaranteed that enough information could be obtained from
the analysis of the separate pictures to calculate the out of plane displacement data [57].
Figure 3-6: Plan view of the DIC set up inside the cubicle.
45
The software used calculated the movements of the sample by tracking each element of an
irregular speckle pattern which was applied to the surface of the specimen. The size of each
individual dot in this pattern needed to be sufficient for the algorithm to recognise it in the
images. This minimum speckle area is generally considered to be 3 x 3 pixels. Therefore, the
size of the window was divided by the resolution of the cameras to provide a minimum
speckle size. In this case this was found to be approximately 9 mm, considering the window
maximum dimension to be 1.5 m and the resolution to be 512 pixels. Before applying the
speckling, the window was prepared by painting the internal surface with white matt paint
and the outer surface with black paint. This was done to minimise the amount of light
penetrating the window from the flash of the explosion, which could have overexposed the
images. The black dots forming the speckle pattern was then applied randomly with matt
paint. An approximate 50% ratio between black speckle area and white background was
aimed at and a final cover of 32% was achieved. Figure 3-7 show an image of the final
window surface.
Figure 3-7: Speckle pattern on the glazing.
Before the test, pictures were taken of a standard calibration object. These pictures were to be
used by the algorithm to locate the cameras and correct for picture distortions.
46
The DIC system was triggered by the test site staff before detonation to ensure that the whole
pane excursion was captured. The pictures were then analysed to obtain deformation data
using the procedure described below.
Strain gauges set up
Frame reaction data were collected using pairs of strain gauges attached to the window frame
at several locations along each edge. The pairs were located on the supporting steel near the
window restraint positions, as shown in Figure 3-1. Each gauge was located on one side of
the plate. To obtain meaningful information it was necessary to align each gauge with
minimal relative lateral displacement. A total of 24 strain gauges in 12 pairs were used. Three
pairs were located at quarter points along each side. Figure 3-8 shows a sketch of the gauge
locations along the perimeter.
Figure 3-8: Cubicle front face elevation showing the location and numbering of the
strain gauges used.
Model C2A-06-062LW-350 gauges produced by Vishay PG were used, with a resistance of

350 Ω. A 20 m long RJ45 cable was used to connect each gauge pair to the recording
47
equipment, controlled by the test site personnel. Prior to the positioning of the window frame
each gauge pair was tested for failures and flatness. The recording equipment was triggered
before the detonation as per the cameras. Data were collected at a frequency of 200 kHz.
3.2.2. Data analysis
The data collected during the test were analysed to produce the final results. The techniques
used to analyse the pressure, optically collected deformation and the strain gauges data are
presented here.
Pressure distribution
The side on pressure readings from the pressure gauge were converted into a reflected
pressure using the method presented by Kingery and Coulter [8]. Equation 3-1 below was
applied:
 7 P + 4 Ps 
Pr = 2 Ps  0  Eq. 3-1
 7 P0 + Ps 
where P0 is the atmospheric pressure, Ps is the incident pressure and Pr is the reflected
pressure. This reflected value assumes an infinite surface with no pressure leakage around the
edges and no enhancement due to ground proximity. Whilst these conditions were not
respected at the cubicle location, it was decided that the errors due to the different standoff
distance of the gauge were likely to be more significant for the peak pressure datum. The
formula however was not used to obtain a full time history of reflected pressures, as the error
for this information would have been very significant. For these reasons, a computational
model was employed to estimate the pressures at 23 m stand-off. The software used was the
Air3D package [58]. The 9th version of the program was used [11]. This package requires the
input of blast parameters, namely the TNT equivalent explosive mass and the diameter of the
charge. The layout of the test needs to be then modelled, specifying the position of the charge
with respect to the ground level and the shape of any obstruction, in this case the test cubicle.
The software can allow for a failure pressure to be specified for window elements, however
this option was not employed in this case. The analysis is performed firstly in one dimension
48
until the blast wave reaches the ground, then in 2 dimensions until the wave reaches the target
proximity and finally in 3 dimensions to compute the pressures on the area of interest. A half
model of the cubicle was produced, setting a reflective symmetry boundary condition along
the vertical window centre line. Output was requested at 50 mm intervals throughout the
window area. Two models were created, one locating the cubicle at the same stand-off as the
gauge, 25 m, and one with the correct 23 m distance. Additionally, an extra model was run
without a cubicle to calculate the incident pressures at 25 m as this information could be
directly compared with the experimental data.
The data obtained was analysed with Matlab to calculate the distribution of the pressures on
the window area and estimates of the total pressure and impulse on the window. The 25 m
peak pressure results could then be compared with the recorded pressures to evaluate
eventual differences with the computational model.
DIC calibration
The DIC analysis package used was ARAMIS by GOM Optical Measuring Techniques. For
the software to perform the calculations, it requires the input of calibration images. These are
pictures of a proprietary object in which the software can recognise specific points. The
algorithm can then compute the exact camera positions and account for picture deformations,
such as lens barrelling. In this case however, due to problems with the cameras used, these
pictures were not available. A new procedure had therefore to be applied to produce
calibration parameters.
Whilst it is theoretically possible to produce a calibration using alternative algorithms without

calibration objects [59], this technique was not implemented in the available software. It was
decided that writing a proprietary algorithm would not be feasible, hence a method
compatible with ARAMIS had to be found. Finally, it was decided to use virtual 3D images
for the calibration. As a first step, the cameras and the vector of the direction they were aimed
at had to be located with respect to the window. To do this, a 3D rendering software (PovRay)
was employed. This software allows the use of script files to define images to be rendered. A
generic file describing an image of a black rectangle of the same size as the window pane was
prepared. Using Matlab, a series of 1296 rendering files were produced varying the camera
position and its direction in a grid with steps of 40 mm in each direction. The x and y position
49
of the cameras and of the target point were being changed. An algorithm was then employed
to compare the proportion of the lengths of the rectangle’s sides to those measured in a pre-
blast picture of the window. The image which presented the closest approximation to a
geometrical similar shape to the window was then chosen. A further iteration with 1296
images was performed, this time cutting the steps to 8mm. The same analysis was done, and
the position of the camera and its direction were assumed to be as indicated in the image
generation file. This procedure was applied to both cameras. The camera positions found
were then used to render images of the calibration object. A rectangular object with
dimensions of 500 x 400 mm was modelled graphically and a series of 18 images
representing different item rotations and positions was generated. The window frame was
included in these for reference and was employed to cut the images to the size and proportion
they would have had if they had been taken with the test cameras. Figure 3-9 shows an
example of the sets of pictures used for this procedure. The images were then used in
ARAMIS to produce a calibration. The software requires the input of the camera’s sensor
dimensions to produce the required data. As the produced calibration images were of a higher
resolution, the pixel size used for the calibration was adjusted accordingly.
Figure 3-9: One of the sets of calibration pictures used for the procedure.
The software could then be used to analyse the pictures taken in the experiment. The only
further difference from the standard procedure concerned the setting of analysis start points.
These are employed by the software as reference locations on the sample’s surface. Due to
the non-standard calibration technique used, these had to be created manually by selecting the
correct area of the panel in each picture at each time step.
50
A DIC calculation is performed based on facets, i.e. elements of the sample area which are
tracked individually. In the case of this test, square facets of 10 x 10 pixels were employed.
Determine approximate position of camera from experimental information.
Create images of the window shape as seen from a grid of possible positions.
Compare the window shape iterations to a real image. Select the image which
is most geometrically similar.
Repeat the comparison process using a smaller grid of camera positions.
Generate 3D images of a calibration object for the calibration process.
Input the images into ARAMIS adjusting sensor information to suit the image
resolution.
Set the DIC analysis starting points manually and proceed with analysis.
Import DIC results in Matlab and correct for lens aberration effects.
Figure 3-10: Flow chart summarising the calibration and DIC analysis process used for
the experimental data.
The calibration technique used could not account for lens aberrations. Specifically, barrelling
was very evident due to the very short focal length of the lenses used. The results produced
therefore showed the surface of the specimen to be curved. This problem was corrected
numerically during post processing. All the data were imported in Matlab. The curvature in
the original coordinates was eliminated for further use of the data. The data points were also
interpolated to a regular 20 mm grid. These were then used to produce sections and specific
51
plots to analyse the behaviour of the window at different time frames. A summary of the
process described here is given in Figure 3-10 for clarity.
As this technique had not been used before, its results could not be assumed to be exact. To
validate it, the method was applied to the third of Hooper et al.’s previous blast tests [37].
This experiment was chosen as it represented a typical test with respect to the window size
and its calibration images allowed a calibration of higher precision when compared with
other tests. As full calibration information was available, it was possible to directly compare
the results of the technique with the data obtained through a normal calibration. The out of
plane deflections, velocities and strains in the two main directions and the deflection angles at
four locations were compared to produce error estimates. The deflection comparison was
made between the DIC results obtained with the two techniques. The angles and velocities
were also calculated from these results for this exercise. The strains were compared using
both the alternative calibration DIC output and values recalculated using the out of plane
deflection results, as explained below. The x and y direction strains were used.
DIC data analysis
The technique described above was applied to the available test images. Deflection data were
obtained from the DIC analysis software and adjusted to correct for aberrations as discussed.
It was decided to recalculate the strain data from the obtained movements instead of
employing the data directly produced by the DIC algorithm. To do this, the deformed
distance of nearby interpolated points was calculated using the DIC deflected coordinates
information. This was then used to calculate the strain in the x and the y directions, as these
would be more useful for later analysis.
The velocity was also calculated from the out of plane deflections. A centre point
differentiation technique was used at each point of the grid, producing full field velocity
information.
The displacement data were also used to calculate the springing angles at the location of the
strain gauge pairs. The deflection profiles at the different positions along the edge were
considered and angles were calculated using simple trigonometry.
52
Reaction force
The data from the strain gauges were analysed using a method similar to that described by
Hooper et al. [37]. The strain gauges provide strain data on both side of the supporting steel
plate, which will deform elastically under the reactions from the pane. These strains can
therefore be easily converted into stresses, and these used to calculate the bending moments
and axial forces acting on the angle plate. Figure 3-11 shows a diagram of the system with the
main force and reactions acting on the steel plate, whilst Figure 3-12 shows a schematic of
the main forces and plate strains.
Figure 3-11: Sketch showing the forces acting on the strain gauges and the main
dimensions used in the calculation.
ε1
ε2
Figure 3-12: Strain distribution in the supporting plate’s cross section.
53
The bending strain at the gauge location can be calculated from the recorded data as:
ε1 − ε 2
εb = Eq. 3-2
2
Where ε1 and ε2 are the strains on opposite sides of the steel plate and will tend to be of
opposite signs.
The strain due to the axial force is equal to:
ε1 + ε 2
εa = Eq. 3-3
2
The bending moment M, which is measured in Newton-metres per unit width (Nm/m),
applied to the plate is therefore equal to:
ε b EI ε b Ed 2 ε b × 2 ×1011 × 0.007 2
M (Nm/ m) = = = Eq. 3-4
y 6 6
Where I is the second moment of area, y is half the plate’s thickness, equal to d, and E is the
Young’s modulus of steel. The axial force per meter width will be equal to:
Fa (N/m) = ε a Ed = ε b × 2 × 1011 × 0.007 Eq. 3-5
These moments and forces will be caused by the window reactions. The bending moment at
the strain gauges is:
M (Nm/ m) = F × cos(θ ) × 0.0135 + F × sin(θ ) × 0.070 Eq. 3-6
And the axial force at the same positions is:
Fa (N/m) = F × cos(θ ) Eq. 3-7
In theory therefore a system of equations could be set up using the four formulae listed above
equating the forces measured by the gauges with the reaction forces. The reaction force and
its angle θ could then be found. However, Hooper et al. found that even relatively small
errors in the alignment of the gauges cause significant errors in this estimate. This was also
found in this case, with the strain readings producing non realistic values for the reactions.
The solution was to use the DIC data to estimate the angle at which the reaction force is
applied. The force could be then calculated directly from either the bending moment or the
54
axial force information. The bending moment measurement was used, as Hooper had found
the axial force strains to be the most affected by the alignment errors.
The reaction force therefore was given by:
ε b × 2 ×1011 × 0.007 2
F × cos(θ ) × 0.0135 + F × sin(θ ) × 0.07 = Eq. 3-8
6
Which can be rearranged to give:
ε b × 2 ×1011 × 0.007 2
F (N/m) = Eq. 3-9
6 × (cos(θ ) × 0.0135 + sin(θ ) × 0.070)
The force so obtained will be in Newton per meter width. Once the forces were calculated at
the locations of the gauge pairs using this formula, the results were presented and compared
to examine the changes in reaction force along a window edge.
3.3. Results
The data collected during the test were analysed employing the techniques described above.
The results are illustrated below.
3.3.1. Blast Pressures
The temperature on site at the time of the test was 18°C. The experimentally recorded peak
pressure was converted to a reflected pressure with the formula stated above in paragraph
3.2.2. The peak pressure recorded by the site pressure gauge was 83.2 kPa and the positive
phase lasted approximately 20 ms. The pressure wave reached the gauge 42.4 ms after
detonation. The experimental pressure time history is shown in Figure 3-13 together with the
Air3D CFD estimate. A comparison between the values obtained is shown in Table 3-1.
55
50
40
Air3D Pressures (at 25 m)
Average Pressure (kPa)
Experimental Pressures (at 25 m)

30
20
10
-10
-20
0 20 40 60 80 100 120
Time (ms)
Figure 3-13: Pressure time history at 25 m stand-off. The experimental and CFD
incident data are shown. The charge was set of at time 0 ms.
Table 3-1: Incident pressure results.
Peak Incident Positive Phase Arrival Time Incident Impulse

Case
Pressure (kPa) Duration (ms) (ms) (kPa ms)
Experimental 40.3 15.6 42.4 214.4
Air3D 41.6 14.8 40.5 242.4
The 25 m Air3D analysis with a cubicle produced a peak reflected pressure of 91.2 kPa, with
a positive phase duration of 10 ms and an arrival time of 40.5 ms. The 23 m stand-off Air3D
estimate is shown in Figure 3-14. The peak pressure in this case was 107.0 kPa at a time of
arrival of 35.4 ms. Impulses were calculated for the computational data. The results for the
reflected data are summarised in Table 3-2.
56
120
100
80
60
40
20
-20
30 32 34 36 38 40 42 44 46 48 50
Time (ms)
Figure 3-14: Pressure time history at 23 m stand off calculated with Air3D. The charge
was set of at time 0 ms.
Table 3-2: Reflected pressure results.
Case Peak Reflected Reflected Impulse Arrival Time

Pressure (kPa) (kPa ms) (ms)
25 m Stand off experimental 93.8 N/A 42.4
25 m Stand off Air3D 91.2 359 40.5
23 m Stand off Air3D 107.0 406 35.4
The incident data shows a difference between the recorded peak pressures of 3.3%, whilst the
impulse error is 13%. The reflected data presents similar errors. The calculated reflected peak
pressure is 3% lower and the time of arrival 2 ms shorter. Some of the differences could be
due to limitations in the CFD software. A mesh sensitivity study was run, and the mesh used
(75 mm size) did not show significant differences when compared to a 50 mm mesh data, as
shown in Figure 3-15. However, limitations in the modelling of the shock wave remain and
are inherent in the software employed. Whilst these numerical imprecisions can explain some
57
of the minor difference in the blast peak and the more relevant difference in the arrival time,
it is also possible that the limitations of the conversion equation used (Eq. 3-1) contributed to
this difference. For example, ground effects are already significant in amplifying the pressure
waves before this reached the target, potentially increasing its velocity.
100
80
60
40
75mm mesh
20 50mm mesh
-20
35 40 45 50 55 60 65 70 75 80
Time (ms)
Figure 3-15: Comparison of Air3D results for two mesh sizes. The charge was set of at
time 0 ms.
3.3.2. Window general observations
Unfortunately the outer glass ply fractured diagonally before the test when the pane was
being mounted on the cubicle. The location of the crack is highlighted in Figure 3-16.
58
Figure 3-16: Existing diagonal crack location.
The glass plies fractured early during the blast, as expected. The paint adhered well to the
surface, with only limited peeling. This prevented an accurate observation of the crack
distribution throughout the pane. Where visible though, generally near the edges, it appeared
that the glass fragments were small, making the glass panes completely opaque. Diagonal
cracks appeared in both directions, with an increased density near the corners. The majority
of the peeled paint is concentrated in these regions. Circumferential cracks are also evidenced
by damage to the paint. These locations are shown in Figure 3-17.
59
Figure 3-17: Front view of the tested panel, showing main areas of damage.
The PVB tore in two main locations, along one diagonal and along the bottom edge. The
diagonal tear, shown in detail in Figure 3-18, took place in the central window area. This is
most likely due to the pre-existing diagonal crack. The PVB failed towards the end of the
inbound deflection, with an evident oscillation of the two sides in the outbound stroke.
Figure 3-18: Detail of central diagonal failure.
Another tear was present on the bottom edge. This is similar to what observed by Hooper,
who reported PVB failures occurring along the edges preferentially. Figure 3-19 shows a
60
detail of this failure. The PVB became torn near the window support at the edge of the
silicone. Glass was still attached to the silicone connections. This, together with longitudinal
fracture of the glass, delamination of PVB and silicone failure is one of the typical edge
failure mechanisms observed in previous tests. The specific failure encountered here could
have been favoured by the double sided restraints, which offered more resistance to the
alternative mechanisms. Additionally, the sharper angle caused by the increased stiffness of
the support system would have brought the PVB layer in closer contact with the sharp glass
edges, causing it to be cut.
Figure 3-19: Detail of the pane failure at the bottom support.
3.3.3. Digital Image Correlation calibration comparison
The calibration technique was applied to a test whose real calibration data were available for
comparison. Once the cameras were located with the technique explained above, the
calibration pictures were generated.
The DIC analysis was performed using the generated calibration pictures and displacement
and strain data. These were imported into Matlab, and interpolated to a regular grid for easy
61
comparison with the existing data. The initial coordinates were flattened out to account for
barrelling distortions previously described.
Figure 3-20 shows a 3 dimensional plot of both sets of deflection data at 15 ms, whilst Figure
3-21 shows a comparison of these results, presenting both relative and absolute errors.
Figure 3-20: 3D plot of the deflection data. The shaded data set is the data obtained with
the new calibration method, whilst the white facets curve shows the old calibration data.
Figure 3-21: A plot of the relative and absolute deflection errors.
The deflections in most of the panel area are similar between the two sets, with errors within
20% of the fully calibrated DIC results. The bigger differences are observable along the
edges. The 3D plot shows that the behaviour of the panel compares well between the two
analyses, however there is some error in the starting position, which is different in the two
data sets. This could justify the higher errors in these areas. A comparison of the velocities
62
obtained is shown in Figure 3-22. The situation is similar to that described for the deflections,
with higher errors present along the edges. This could be expected, as this information was
calculated from the out of plane displacement information.
Figure 3-22: Relative and absolute differences in the velocity data obtained with the two
calibrations.
The strain data are shown in Figure 3-23. As mentioned, two strain estimates are shown for
the alternative calibration method, the DIC output and strains recalculated from the deflection
information. The differences between the two calibrations’ results are higher than for the
deflection information. The general behaviour of the strains is similar, with the highest
deformations in both directions concentrated along the window edges. However the
magnitudes are different. For example, the strains in the x direction are 14% or 7%
(depending on the calculation method) in the alternative calibration data, whilst the fully
calibrated output shows around 3-4% strains. Therefore, whilst the order of magnitude of the
strain values and information on their distribution of the pane can be obtained it is not
possible to obtain results with greater accuracy. These uncertainties will need to be
considered when utilising the data of the latest blast for further analyses in subsequent
chapters.
63
Figure 3-23: Comparison of strain data obtained with both calibrations and with two
methods for the new calibration data. Both strains in the x and in the y direction are
shown.
As the edge angles will be needed to calculate reaction forces, the likely uncertainty of these
data was also estimated. The angles were calculated at the four central locations of each side.
Figure 3-24 and Figure 3-25 show the results at the left hand and top window sides, with the
relative errors shown in Figure 3-26. The maximum inaccuracy is 20% on the left hand side.
This will be considered when assessing the results of further analyses.
64
30
25
20
Degrees
15
10
New Calibration
5 Traditional Calibration
20 22 24 26 28 30 32 34 36 38
Time (ms)
Figure 3-24: Comparison of angle measurements time history along the left hand
vertical window side.
30
25
20
Degrees
15
10
New Calibration
5 Traditional Calibration
20 22 24 26 28 30 32 34 36 38
Time (ms)
Figure 3-25: Comparison of angle measurements time history along the top window
side.
65
New Calibration / Traditional Calibration 1.4
1.3
1.2
1.1
0.9
0.8 Top angle

Left hand side angle
0.7
26 28 30 32 34 36 38
Time (ms)
Figure 3-26: Relative error for the two data sets displayed above.
3.3.4. DIC Analysis Results
The high speed images from the test were processed using DIC software. The DIC algorithm
could not produce results for the central area of the window when peak deflections were
reached, as the central diagonal tear ruined the speckle pattern available at the latter stages of
the deformation. The relevant information was obtained through interpolation of the available
data. A total of 64 images was utilised for the analysis, producing data for a period of 21.3 ms.
Figure 3-27 shows the out of plane deflection results. The low deflection area around the
window edge represents results for the steel support, which was also covered with a speckle
pattern to estimate its deflections. As can be seen in the series, these deflections are
significantly lower than for the glass, peaking at 50 mm. The deflected shape of the window
changes significantly during the event. The panel initially shows a relatively flat area in the
middle. Most of the deformation at low deflections is concentrated around the edges, which
could be an explanation for the higher cracking densities which are traditionally observed in
these areas.
66
Figure 3-27: Out-of-plane deflections of the glass panel measured with the DIC method.
The deflection is in mm. Time 0 ms is at the blast wave arrival.
67
To visualise this effect more accurately, Figure 3-28 shows a plot of the curvature of the
window and the out of plane deflection 4 ms after the arrival of the blast wave. These were
calculated along a horizontal cut taken at the centre of the window. The plot shows clearly
the flat central area both the curvature and in the deflections. Later, as the deflections grows,
the curved area increases, moving towards the centre of the pane. Ultimately the whole
window assumes a deformed shape similar to what could be expected for static or harmonic
deflections. The deflection of the mid point of the glass panel peaked at 297 mm. Figure 3-29
shows a time history of this datum.
-3
x 10
1 100
0.9 90
Deflection
0.8 80
Curvature
Displacement (mm)
0.7 70
Curvature (1/mm)
0.6 60
0.5 50
0.4 40
0.3 30
0.2 20
0.1 10
0 0
0 200 400 600 800 1000 1200 1400 1600 1800
Section position (mm)
Figure 3-28: A plot of the curvature and deflections along a horizontal cut taken across
the centre of the window.
Figure 3-30 shows a plot of the central velocity, whilst Figure 3-31 shows the calculated
normal velocity across the area of the pane. As expected from the deflection results and
literature analyses [37], the area of high velocity decreases as the deflection progresses. The
maximum velocity recorded was 34 m/s.
68
300
250
200
Deflection (mm)
150
100
50
-50
-5 0 5 10 15 20
Time (ms)
Figure 3-29: Central deflection time history from DIC data. Time 0 ms is at the blast
wave arrival.
40
30
20
Velocity (m/s)
10
-10
-20
-30
-5 0 5 10 15 20
Time (ms)
Figure 3-30: Central velocity time history from DIC data. Time 0 ms is at the blast wave
arrival.
69
Figure 3-31: Out of plane velocities measured with the DIC method. The velocity is
given in m/s. Time 0 ms is at the blast wave arrival.
70
Figure 3-32: Strain in x direction calculated from the DIC deflection data. The strain is
given as a percentage. Time 0 ms is at the blast wave arrival.
71
Figure 3-33: Strain in y direction calculated from the DIC deflection data. The strain is
given as a percentage. Time 0 ms is at the blast wave arrival.
72
Figure 3-32 shows DIC results for the strain in the horizontal direction, whilst Figure 3-33
shows the vertical strain. In both cases, larger strains seemed to be concentrated along the
edges. This is consistent with the observed deformed shape and curvatures described above.
As can be seen, the vertical strains were larger. The average values peaked at 13%, with only
limited locations displaying higher values. The edge horizontal strains showed an average of
12% strain. However it should be considered that, as observed in the comparison described in
paragraph 3.3.3, the calibration method errors where bigger for this type of results. Therefore
care should be taken when interpreting these exact values. Unfortunately the resolution of the
picture and focus was not sufficient in this experiment to observe isolated crack locations
through the strain data.
3.3.5. Reaction data
Several gauges failed during the blast test. This was due to both to atmospheric conditions
during the night the window spent on site and to some cables being damaged during the
positioning of the sample in the cubicle. These issues severely curtailed the amount of
information available to assess the variation of reaction forces along the window edges.
However, all the pairs along the top edge of the window and two on the left hand edge
survived. Figure 3-35 shows the location of the failed and working gauges.
73
Figure 3-34: Cubicle elevation showing location of broken gauges.
The deflection angle was calculated using the DIC data. Errors within 20% are expected, in
line with the comparison results illustrated above in section 3.3.3. Typical angles observed
ranged between 20 and 30 degrees at the various gauge positions. The formulae described
above were applied to find the reaction forces. Figure 3-35 to Figure 3-37 show some of the
results, specifically those collected along the top window edge. It can be seen that the noise
present in the data is high, especially during the advanced phases of the deformation. Whilst
this prevents accurate reaction measurements, the relative magnitude of the forces can still be
assessed. In general, all the reaction forces exhibited a plateau at high deflections, indicative
of a maximum force being reached.
74
40 40
35 35
Glass failure Angle
Reaction Force
30 30
Angle (Degrees)
25 25
Force (kN)
20 20
15 15
10 10
5 5
0 2 4 6 8 10 12 14 16 18 20
Time (ms)
Figure 3-35: Angle and reaction force at gauge location 1 as shown in Figure 3-35. Time
0 ms is at the blast wave arrival.
40 40
35 35
Glass failure Angle
Reaction Force
30 30
Angle (Degrees)
25 25
Force (kN)
20 20
15 15
10 10
5 5
0 2 4 6 8 10 12 14 16 18 20
Time (ms)
75
40 40
35
Glass failure 35
Angle
Reaction Force
30 30
Angle (Degrees)
25 25
Force (kN)
20 20
15 15
10 10
5 5
0 2 4 6 8 10 12 14 16 18 20
Time (ms)
Figure 3-38 shows the three forces calculated along the top edge. The plateau of the central
force is lower than that measured at the quarter points. This is related to the shape of the
deflection observed in previous tests [37], which would justify higher forces away from the
centre of the panel. This effect will be studied in more detail. Figure 3-39 instead shows the
two forces calculated along the left hand edge. The variation between these is smaller,
indicating a more constant loading on the shorter edge.
76
40
35
Location 1
Location 5
30
Location 6
25
Force (kN)
20
15
10
0 2 4 6 8 10 12 14 16 18 20
Time (ms)
Figure 3-38: Reaction force data obtained on the top edge of the window. The gauge
locations are as indicated in Figure 3-35. Time 0 ms is at the blast wave arrival.
40
35
Location 3
30 Location 9
25
Force (kN)
20
15
10
0 2 4 6 8 10 12 14 16 18 20
Time (ms)
Figure 3-39: Reaction force data obtained on the left hand side edge of the window. The
gauge locations are as indicated in Figure 3-35. Time 0 ms is at the blast wave arrival.
77
3.4. Discussion
The test pane failed both in the centre and along its bottom edge. Edge failure is typical, as
that is the area where high curvature and tension forces will cause a greater amount of
damage. Central tearing in the PVB is less common. In this case it was probably caused by
the pre-existing crack which crossed the window diagonally. This would have weakened the
local area and a tear initiated in this location at approximately the time when maximum
deflections, and hence the maximum curvature in the centre, was reached.
The edge failure was due to a tear in the PVB. In the past, other failure mechanisms have
been observed, including glass or silicone failures and debonding of the PVB membrane from
the glass plies. However, in this test the double sided silicone joints strengthened the edge
connection capacity, therefore causing a PVB tear to become more likely as a failure
mechanism.
The DIC data were obtained through a non standard calibration procedure due to the failure
of obtaining a calibration using the available images. When the results obtained where
compared with fully calibrated results from a previous test, it could be seen that the deflection
data, especially in the internal area of the window, were accurate, with errors well within
20%. These inaccuracies though increased towards the edges, highlighting uncertainties in
the deflection in these areas. The velocity results showed a similar trend. However, the
calculated angles showed a maximum error of about 15% at higher deflections. This, together
with the 3D plots of deflection shown in Figure 3-20, might indicate that the calculated
deflected shape is accurate, displaying however errors on the exact window dimensions’
estimate.
The calculated strains showed a much larger error, with the estimated data in some cases
being in cases twice the magnitude of the calibrated data. This is most likely due to the
compounding of smaller deflection errors when the strains are calculated, together with the
fact that the higher strains are recorded along the window edges where the deflection
uncertainties are higher. Therefore, care must be taken when using these data for further
analyses, especially if very accurate strain values are required. It is though useful to note that
the trend of the data is realistic, showing similar patterns near the window edges. This will
78
help when using these data to calculate and compare material properties later, as the level of
errors will be similar throughout the area of interest.
The deformed shape-time history shown in Figure 3-29 corresponds to those observed by
Hooper in previous tests. The deformation is initially concentrated along the window edge,
with a largely flat central section. As time progresses, the deformation fronts moves towards
the centre, until a fully curved surface is reached. As Hooper explained, this phenomenon is
influenced by the deformation speed and wave speed in the panel. Interestingly, the initial
diagonal crack in the window did not affect significantly the deflection time history. This is
probably due to the crack being on one side only of the window pane. Therefore, a hinge was
not formed until after the glass failure. Additionally, the crack was not located in areas of
high deformation, hence it can be assumed that it would not affect the overall blast capacity
of the pane significantly until ultimate failure was reached.
Maximum strain locations were related to the deflected shape. Whilst, as discussed, the exact
magnitude of the strains are uncertain due to the calibration technique used, the greater
deformations seem to be consistently located near the window edges where the highest
curvatures were also observed. This is consistent with the observed deflected shape and had
been seen in previous tests as well.
As the pressure gauges were at a greater distance from the charge than the test window, it was
necessary to utilise Air3D to produce loading pressure estimates for further analysis. Some
differences between the experimental and computational data were noted. When the CFD
analysis was run for a stand-off of 25 m, the same as that of the pressure gauges, its arrival
time and peak pressures were different from the recorded ones. As mentioned, this is most
likely due to limitation in the analysis software as well as in the conversion formula used to
obtain experimental reflected pressures. However, as no experimental data was available at
the correct distance, the Air3D data was employed for further stages of the analysis.
The reaction forces at the locations of the strain gauges were found to be similar to results
previously reported by Hooper [3]. The springing angles of 20° to 30° are similar to literature
data, as are the reaction magnitudes of 20-30 kN/m. However, a high level of noise was
present in the data at higher deflections. Additionally, as discussed by Hooper, the reaction
force calculation is sensitive to misplacement of the gauges and changes in the supports’
geometry due to the blast loads effects. Also, in this case, the angles calculated from DIC
included a relevant error, estimated to be up to 20%. The force calculation though proved to
79
be less sensitive to this uncertainty, producing an error of ± 15% for angles of relevant
magnitudes. Whilst these problems need to be considered, the data is still useful for
comparison purposes. As shown in Figure 3-38, the reaction forces are higher at the quarter
positions of the span. This relates well to deflected shapes described in literature, which
included the presence of the two deflection peaks at some distance away from the centre of
the pane. Whilst this was not detected clearly here, the lower camera resolution and non-
standard calibration might have caused this effect to be hidden. The more constant force
along the vertical edge also can be interpreted as evidence for this phenomenon, as in this
direction the peaks would be aligned along the window centreline, removing higher order
effects and producing a more uniform distribution of reaction forces.
3.5. Conclusion
A blast test was performed to assess the behaviour of a laminated glass panel under blast
loading. The distribution of reaction forces along each edge was required; therefore several
strain gauge pairs were applied along each edge to measure this datum at several locations.
The panel was loaded with an estimated pressure of 107.0 kPa and an impulse of 405.7 kPa
ms. Due to problems with the calibration images taken during the test set up, an alternative
calibration method was used. Its implications were assessed employing historical data from a
previous blast test. It was concluded that the deflection data obtained is accurate to within 20%
of values measured using a calibration pictures in most of the window area, however the
strain information is not reliable beyond order of magnitude comparisons.
The failure mechanisms observed were similar to those seen in previous tests. However, as
the support condition was changed from single sided silicone to double sided, tearing of the
PVB was encouraged rather than other possible mechanisms observed by Hooper. An
existing glass crack crossing the window diagonally also affected the failure, initiating a PVB
tear in the central area of the window.
The deflected shape and maximum deflection of the window were also comparable with other
historical results. A maximum displacement of 297 mm and a maximum velocity of 34 m/s
were measured. This indicates that the double sided joint and the existing crack did not affect
these parameters significantly before failure.
80
One of the main objectives of this test was to measure reaction forces at several points on the
window’s perimeter. These were calculated at the strain gauge pairs’ locations. Although
many of these sets had failed due to damage either before or during the test, a full set of
readings were available from three gauge pairs along the top horizontal edge and from two
gauge pairs’ along the left hand edge. Significant noise precludes a detailed analysis of the
absolute magnitudes, however it could be seen that the forces along the longer edge were not
constant, having higher values at the quarter points locations than at the centre. The values
were more uniform along the shorter edge. This can be related to previously noted deflected
shapes. These showed more than one peak, caused by higher order modes of vibration and
interaction of stress waves. Generally, forces of 20-30 kN/m were recorded, which is
comparable with literature data [3]. The reaction force distribution will be studied in more
detail in Chapter 4, where the full reaction forces at pre and post cracking will be estimated.
81
4. Blast Reaction Forces
4.1. Introduction
An important consideration in designing a blast resistant glazing system is the connection

between panes and its supports. During a blast, significant reaction forces are applied to the
restraint mechanism, both out of the window plane and in the window plane, especially after
the glass cracks. Calculating these forces is therefore a very important step in improving and
optimising the design of blast resilient windows. Whilst in previous work forces at local
points were found directly from strain gauge readings, ideally the whole reaction along each
edge is needed to inform future design.
The objective of this part of the work was to produce such an estimate using available blast
experimental data. Information from both the test described in Chapter 3 and from previous
tests performed by Hooper et al. [37] have been employed. The aim was to calculate the
overall reaction forces both prior to and after cracking of the glass pane. Different techniques
have been used for the two situations, as in them the panel behaves in significantly different
manners. Results from different tests data sets have then been compared and their implication
for the window design and the behaviour of the laminated material considered.
4.2. Method
In addition to the blast test described in the previous chapter, Hooper et al. [37] performed
several blast tests on laminated glass. His 2 nd, 3rd and 4th tests are employed in this study. The
specimens and techniques used for these were similar to those described in the previous
chapter. The test panes consisted of two 3 mm thick plies of annealed glass with a 1.52 mm
PVB interlayer. The explosive charges used ranged from 30 kg to 100 kg and the target was
located between 8 m and 30 m away from the explosives. Data collection was achieved
employing the digital image correlation techniques (DIC) to provide 3D full field information
on deflections and strains in all directions. Two Photron S3 cameras were used to take
images of the window at 1 ms intervals. In Hooper et al.’s tests, two window panes were
tested side by side in the same cubicle, whilst in the further test a single window was placed
82
in a smaller cubicle. Therefore, slightly different boundary conditions applied on the two
vertical sides of the windows, with the central edge showing an increased flexibility in
Hooper et al.’s tests due to the cubicle construction. Figure 4-1 shows the typical DIC set up
used in the historical tests. Differently from what described in Chapter 3, the cameras were
placed vertically.
Figure 4-1: DIC set up used in Hooper et al.’s blast tests [37].
All the tests included at least four pairs of strain gauges located at the centre of each window
side on the supporting steel angles of the frame. These were used to measure the local
reaction force. The disposition of the gauges was similar to that used in the most recent test,
shown in Figure 3-1. Therefore the only difference in this respect between Hooper’s tests and
the most recent one was the number of gauge pairs, four in Hooper’s tests instead of the 12
used in the latest experiment reported in Chapter 3. The general characteristics of the blasts
considered is summarised in Table 4-1 below. For tests 1 and 4 the pressure and impulse data
were obtained from CFD simulations using Air3D, whilst experimental values were available
for the other two cases.
For all the four tests considered the available 3D deflection and strain data were interpolated
at regular grid points. As described in the previous chapter, it was found that as the
deflections increased the DIC technique proved unable to provide data for a growing band of
83
facets along the edges. An extrapolation technique was therefore used to obtain the out of
plane deflections. This involved firstly fitting a 4th order polynomial surface to the available
interpolated information. Afterwards, the experimental data were substituted by the fitted
results either where the deflected values were not found or where they presented
unrealistically high positive or negative data points along the edges.
Table 4-1: Summary of main blast parameters of the tests considered in this work.
Charge Stand-off Peak Reflected Impulse

Test Author
Weight (kg) (m) Pressure (kPa) (kPa x ms)
1 Hooper 15 13 31.2 284
2 Hooper 30 16 132 413
3 Hooper 30 14 152 461
4 Del Linz 100 23 107 406
Whilst it would have been ideal to employ the same technique to the strain data, their more
noisy nature prevented an acceptable fit to be achieved. It was therefore considered that
extrapolated points would not be realistic.
All the blast test data obtained was employed to calculate reactions of both the pre-cracked
and the post-cracked pane. Due to the different behaviour of the glass in the two situations,
separate approaches where employed.
4.2.1. Pre-cracked reactions
In the pre-cracked phase, the applied blast loading will generate in the panes a combination of
out of plane bending and in plane membrane actions. Soon after the blast wave reaches the
target, as the deflections will be small, membrane forces will be negligible in accordance with
traditional small deflection theory. However, the window can deflect significantly compared
with its thickness before cracking takes place. For example, in the tests considered here, the
central deflection generally reached up to 40 mm before the glass shattered. In these later
stages therefore the small deflection theory assumptions would not be valid and membrane
forces will be significant.
84
If the strains on both window faces were available, the bending and direct stresses could have
been separated and accounted for. However, the DIC data were only collected for one face,
preventing this analysis to be performed directly. Finite element modelling was therefore
used to estimate the proportion of surface stresses attributable to bending and to membrane
actions at each point in the window. It was decided that, whilst the FEA values of stress and
strain might not accurately reflect real deformations taking place during a test, it would be
acceptable to use only the proportion between the different actions. This consideration was
also strengthened by the fact that no other kind of direct measurements could be used. A pre-
cracked FEA model similar to that developed by Hooper was employed. However, in this
situation the simulation was developed further by including additional elements. These
included the steel support structure and a layer of silicone between the glass and the supports.
The model was developed using ABAQUS 6.10. The laminated glass was represented with
2D shell stress elements using reduced integration (S4R). The section was defined as a
layered composite, using the real thicknesses of the 3 layers (3 mm for the glass and 1.52 mm
for the PVB). The elements were of square shape, with a side length of 2 mm at the silicone
interface and 4 mm elsewhere. A mesh size sensitivity analysis was run and elements of this
dimension were found to be small enough to avoid size effects. Steel support angles with the
same dimensions as in the experiment were added. Again, 2D reduced integration shell
elements were employed for these. The element dimension varied between 2 mm in the areas
of contact with the silicone and 4 mm in other areas. In this case a uniform material section
was used, setting the thickness to the real steel dimensions. The layer of silicone was
included using 3 dimensional stress elements with reduced integration (C3D8R). This was
considered necessary due to the high through thickness deformation expected in this material.
Cubic elements were used, with a side length of 2 mm. The overall silicone thickness
assumed was 6 mm, with a bite depth of 20 mm. Three elements where therefore used across
the 6 mm thickness to achieve the necessary quality of through thickness results. Tie
conditions were set between the different parts, without any allowance for sliding. A similar
model was used for test 4. A different geometry was used to reflect the actual dimensions of
the glass support as well as of the double sided silicone. Additionally, approximately 4 mm
elements were generally used in the steel and glass parts, whilst 4 x 2 x 2 mm elements were
applied to the silicone. The dimensions were changed only in the contact areas to ensure that
the mesh at the interfaces between the parts corresponded exactly. Specifically, the element
dimensions in the glass and the steel were reduced to 4 x 2 mm.
85
The glass, PVB and steel were all given linear elastic material properties. Both the steel and
the glass will behave linearly in the range of deformation of interest. Whilst the PVB would
be likely to show some non linearity, its much lower stiffness implies that its contribution to
load resistance can be considered negligible before the glass layers fail. Hence, for simplicity,
it was also considered acceptable to employ a linear model for this element as well. Table 4-2
summarises the constants used for these three materials. The silicone, due to the expected
large strains acting on it, was instead represented using a hyperelastic material model.
Table 4-2: Summary of linear elastic material properties applied in the FEA models.
Material Density Young’s modulus

Material Poisson’s ratio
(kg/m3) (GPa)
Glass 2530 72 0.22
PVB 1100 0.5 0.485
Steel 7800 200 0.3
A Mooney-Rivlin model was used. This is a hyperelastic model, able to represent the non-
linear deformations of rubber materials. Its work function is given by [60]:
W = Cmr1 ( I1 − 3) + Cmr 2 ( I 2 − 3) Eq. 4-1
where W is the work done by the deformation, I1 and I2 are the invariants of the deformation
tensor and Cmr1 and Cmr2 are materials fitting constants determined with experimental data.
The details of hyperelastic formulations will be described in more detail in Chapter 5. The
constants used were derived by Meunier et al. [50], with values of Cmr1 = 0.14 MPa and Cmr2
= 0.023 MPa. Figure 4-2 and Figure 4-3 show a view of the overall geometry and a detail of
the mesh in the area of the connection between the different parts.
86
Figure 4-2: An image of the FEA model geometry employed for tests 1 to 3. The detail
shows the mesh in the single sided joint area.
Figure 4-3: An image of the FEA model geometry employed for test 4. The detail shows
the mesh in the double sided joint area.
87
The load was applied as a uniformly distributed pressure over the whole glass pane. Its
pressure amplitude time history was estimated from computation fluid dynamics (CFD)
models of the experiments. The Air3D [58] analysis code was used for this. This produced
pressure estimates at 0.0001 s intervals at points on a 50 x 50 mm grid on a quarter of the
window. It was observed that the pressure on the glass varied only by 10% over the whole
area. Considering this and the fact that the FEA model would be used only to find the ratio of
bending and axial stresses, it was decided that an average pressure would be appropriate for
this application. This was therefore calculated from the Air3D output and inputted in the FEA
model.
The model was set to run until a limiting stress of 80 MPa, was reached anywhere in the glass
elements. The strain data were then extracted and the proportion of bending and axial strains
was calculated. These proportions were then applied to the experimental DIC strain data. As a
checking procedure, the central deflection was also obtained and compared with the recorded
experimental data.
Once the bending strains were found, the bending stresses and moments were calculated
along sections at 20 mm intervals in both directions. The shear force could then be found by
numerically differentiating the moments with respect of the distance along the window’s
span. It was assumed that the out of plane reaction due to the bending moment at the edges
would be equal to the magnitude of the shear force. As the strain data were found to be quite
noisy, they were filtered to obtain a smoother output. It was assumed that the glass could not
resist stresses higher than 80 MPa. Therefore, if a stress higher than this was found in the
data, the stress was reduced to 80 MPa. Fourier transforms were then employed to eliminate
some of the noise, eliminating higher order frequencies from the data set. Additionally, an
average of the last three points of the bending moment diagram slope was employed when
calculating the reaction force. The results at each facet were then numerically integrated to
calculate the total reaction force along the edges of the window.
The membrane strains were converted to stresses using the same linear elastic model for the
glass which was used in the FEA model. Again, Fourier transforms were employed to reduce
the noise in the data and an average of the last three data points near each edge was taken to
obtain edge stress values. To calculate the total membrane force at each point, the stress was
multiplied by the glass plies thickness, as it was considered that the modulus of the PVB was
much smaller and would provide only a very small contribution to the overall result. This
contribution would therefore be negligible, especially considering the amount of noise, and
88
hence uncertainty, present in the data. The calculated force would be acting in the direction of
the plane of the glass. To estimate the in-plane and out-of-plane components, the glass angle
at each time step was calculated using the out of plane deflection data at each facet. The
membrane reaction was therefore split and the out of plane component could be added to the
bending reaction calculated previously.
Figure 4-4: Representations of the reactions assumed in the pre glass cracking force
calculation.
As a checking procedure, the moment at the strain gauges location was calculated using the
recorded membrane and bending force. Figure 4-4 below shows a schematic of the forces
assumed for this calculation. The shear force and out of plane membrane force were
multiplied by D, whilst the in plane component of the membrane force was multiplied by H.
The dimensions used for the four cases considered are listed in Table 4-3.
Table 4-3: Dimensions used for reaction calculations.
Test D (mm) H (mm)
1 to 3 31 9.75
4 57.5 13.5
4.2.2. Post-cracked reactions
After glass cracking takes place, it is assumed that no bending stiffness will be left in the
laminated glazing. Therefore, the blast forces will be resisted entirely through membrane
89
action of the PVB. Hooper et al. [37] developed a method to calculate the reaction force at
the gauge location, as described in detail in the previous chapter. Briefly, this employs the
data from the strain gauges pairs and the glazing angle calculated with the out-of-plane
deflection DIC data. Whilst in theory data from the gauges would provide sufficient
information for a determinate solution of the axial and bending forces, in practice the
system’s sensitivity to misalignment of the gauges proved too high to reliably estimate both
the bending moment and the axial force acting using solely their data. Hooper et al. showed
that errors have a more significant effect on the axial force component estimate than on the
bending moment. Using the DIC records to provide the angle at which the reaction force is
applied allowed the membrane force in the pane to be calculated by considering only the
equation for the bending moment component.
To calculate reaction forces at points along the edge away from the location of the strain
gauges, a material model linking the recorded DIC strains and stresses in the laminate was
required. In this research such a material model is fitted at each strain gauge pair location and
then applied along the edge on which the pair is located.
To obtain a stress strain curve for the cracked laminate, an analysis similar to Hooper et al.’s
was performed at the gauges locations. The resulting force, divided by the PVB thickness,
provided the stresses in the glass at each time point. The DIC strains at the locations nearest
to the gauges were then extracted, averaging the results over an area of 3 x 3 facets to reduce
the effects of noise. This enabled a stress strain curve to be plotted for each edge, and a
chosen material model to be fitted using a least squares difference approach. The calculated
strain was also differentiated with respect to time to provide strain rate estimates for use in
the fit. In test 4, this analysis could not be performed on one window edge due to the failure
of the pair of strain gauges. The data obtained on the top edge was applied to both locations
to obtain useful results.
As the stress-strain data showed that the stress reached a plateau after an initial stiff linear
increase, a plastic material model was used to represent the composite material. It is realised
that this model does not reflect a real physical characteristic of the PVB material, which in
itself does not show plastic behaviour in the same way as metals do. However, it was
considered that for the purpose of this analysis such a model would be useful for capturing
the shape of the observed data. Similarly to the method applied by Hooper et al. for their FEA
models [37], a Johnson-Cook [61] model was used, whose constitutive equation is:
90
 εɺ 
σ = ( A + Bε Pn ) 1 + C ln  Eq. 4-2
 εɺ0 
Where A, B, C and n are constants to be determined, εp is the plastic strain, εɺ is the strain rate
and εɺ0 is a reference strain rate. In this case a reference rate of 1 s-1 was assumed.
This model provided enough flexibility to account for rate effects and was at the same time
straightforward to apply. Since, as mentioned above, plasticity models are only used to
represent the observed shape of the data, it was decided that it would not be appropriate to
employ more sophisticated, physically based models in this occasion.
Once the model was calibrated, the stress levels at each point could easily be found, allowing
the calculation of the membrane force throughout the perimeter. Again, the glass springing
angle was calculated at each point along the edge to split the total reaction force in the glass
into out of plane and in plane reactions in the frame.
4.3. Results
4.3.1. DIC Results - Extrapolation of deflection data
The original out of plane deflection data were extrapolated to cover areas where the DIC
analysis did not produce results, as described above. Figure 4-5 below shows a plot of the
original data and a plot of the interpolated deflections for one of the blasts. Sets of data
obtained in this manner were employed for further steps of the analysis.
91
Figure 4-5: Original out of plane deflection data (top) and the same data with missing
and unrealistic points substituted (bottom).
92
4.3.2. Pre-cracked reactions
As discussed above, the proportions of surface strains due to out of plane bending and in
plane membrane forces were calculated using a FEA model of the glass pre-cracking phase.
Figure 4-6 and Figure 4-7 show sample plots of the central deflection time history obtained
from the FEA model and the DIC data for Tests 2 and 4. In general, the FEA deflections are
lower than the recorded ones, however the trends are similar. The results for the latest test
show a larger discrepancy, although the trend seems to follow the experimental data. In
Hooper’s test data shown in Figure 4-6, the central deflection reached at the glass cracking
condition was 22.2 mm, which is compatible with recorded test values, in which failure took
place at 12 mm to 35 mm as shown in Figure 4-18 to Figure 4-21 below. This comparison of
the measured and estimated failure deflections produces the same result for all of Hooper’s
cases, whilst the latest blast test shows a larger discrepancy. This will be discussed below.
25
20
Central Deflection (mm)
15
10
Experimental Data
5 FEA Data
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (ms)
Figure 4-6: Comparison of central deflection obtained through DIC and FEA modelling
for Test 2. Time = 0 ms is the blast wave arrival time.
93
12
10
2
Experimental Data
FEA Data
0
0 0.5 1 1.5 2 2.5
Time (ms)
Figure 4-7: Comparison of central deflection obtained through DIC and FEA modelling
for Test 4. Time = 0 ms is the blast wave arrival time.
Figure 4-8 presents the proportion of stresses due to bending moments and membrane forces
found through the FEA model for Test 2 as an example. The proportion of bending strain was
found to be higher towards the edges of the panel, whilst it decreased to almost zero in the
central area. The membrane proportion shows the opposite behaviour, rising significantly in
the centre of the window.
0.9
0.8
0.7 Cut Location

Stress Proportion
0.6
Membrane stresses proportion
0.5
Bending stresses proportion
0.4
0.3
X
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Distance along cut (m)
Figure 4-8: Proportion of the stresses due to bending moments and membrane forces
found with the FEA model for Test 2.
94
These proportions were applied to the recorded DIC strains, and the strain along each cut in
both directions considered in turn. As described above, the data so obtained were filtered to
improve the shear force estimate. Figure 4-9 shows a typical plot of the bending stresses
across a cut, showing the original data and the filtered result.
7
x 10
10
6
Bending stress (Pa)
-2
-4 Experimental data
Filtered data
-6
-8
0 0.2 0.4 0.6 0.8 1 1.2
Distance along short dimension cut (m)
Figure 4-9: Samples of bending stress data. Both the original DIC data and the filtered
curve are shown.
The plot shows that the maximum bending strains are reached near the supports, tapering to
the lowest values towards the centre of the panel. This behaviour is typical and can be seen
throughout the specimens, although it becomes more pronounced as the time increases. The
membrane stresses were calculated using the same filtering techniques.
The overall out-of-plane reactions calculated along one of the edges are presented in Figure
4-10. It can be seen that the membrane force component is significantly lower, even though it
increases as the loading progresses.
95
Out of plane reaction force (kN/m) 12
10
-2
Membrane force reaction
-4 Bending reaction
-6
0 0.5 1 1.5
Position along long window edge (m)
Figure 4-10: Out-of-plane reaction force along one edge. The bending and membrane
components are shown separately.
As a checking procedure, the in-plane and out-of-plane reactions were employed to back
calculate the bending moments at the strain gauge locations. Figure 4-11 and Figure 4-12
show examples of the results obtained, including both the moment calculated from gauge data
and from the analytical results. The two sample graphs show that the quality of the results is
variable, with the fit showing a good agreement in some locations, whilst it differs
significantly from the measured reactions at others.
96
180
160
Moment from reactions results
140
Moment from strain gauges data
Moment (Nmm/mm)
120
100
80
60
40
20
0
20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6 21.8 22
Time (ms)
Figure 4-11: Bending moment at the gauges location on the bottom side of the panel in
Test 1calculated from the gauges results and from the analysis results.
200
180
160
Moment (Nmm/mm)
140
120
100
80
60
40
Moment from reactions results
20 Moment from strain gauges data
0
20 20.2 20.4 20.6 20.8 21 21.2 21.4 21.6 21.8 22
Time (ms)
Figure 4-12: Bending moment at the gauges location on the top side of the panel in Test
1 calculated from the gauges results and from the analysis results.
The overall reaction results will be shown below, together with the post crack reaction
results.
97
4.3.3. Post crack reactions
The material model was fitted to stress train curves produced as detailed above. A typical
stress strain curve together with the data fit is shown below in Figure 4-13.
14
Experimental data
12 Model fit
10
Stress (MPa)
2
0 1 2 3 4 5 6 7
-3
Plastic strain x 10
Figure 4-13: Stress strain curves for the estimate of cracked glass material model.
10
7
Stress (N/mm2)
3 Experimental data
Model fit
2
0
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01
Plastic strain
Figure 4-14: Stress strain curve and model fit for Test 1 the top of the window frame.
98
As can be seen, the data were noisy. Additionally, a few data sets did not display the typical
expected behaviour, as can be seen in Figure 4-14. As a result of this, in some instances the
best fit found for the material model was a straight line. Table 4-4 below shows the Johnson-
Cook material model coefficients obtained for Test 2 as a typical example. Figure 4-17 shows
a legend for the location of the sides. In general, it can be observed that the rate dependency
was not strong, probably due to the lack of data at well defined strain rates. Also, in some
cases a straight line provided the best fit for the available data, as is indicated by the constant
n being equal to 0.
Table 4-4: Johnson Cook fit coefficients for Test 2. The results for each window side are
shown. Refer to Figure 4-17 for a legend of the sides’ locations.
Side A (MPa) B (MPa) C n
Top 9.38 2.32 0.0 0
Bottom 7.37 6.01 0.0 0.11
Wall 7.01 0.84 0.0 0
Central 3.6 7.76 0.11 0.05
These models were employed on their respective sides to calculate the reactions at all points.
Figure 4-15 shows a plot of the reactions along an edge of one of the windows at two time
steps. The reactions seem to reach two maxima near the ends of the window side.
99
12
11.5
11
10.5
Stress (MPa)
10
9.5
9
Stress at t=28ms
Stress at t=26ms
8.5
7.5
7
0 0.5 1 1.5
Position along window edge (m)
Figure 4-15: PVB reaction stress along the bottom edge of Test 1 at time steps t = 26 ms
and t = 28 ms.
Figure 4-16: Legend of edge locations for reaction data plots.
100
12
10
Reaction Force (kN)
Top reactions
2 Bottom reactions
Wall reactions
0 Central reactions
-2
0 50 100 150 200 250
Figure 4-17: Out of plane reactions on individual panel sides for Test 3. The edge
locations are shown in Figure 4-16.
Figure 4-17 shows the out of plane reactions on each side of the glazing panel during Test 2.
The legend of the edge locations is shown in Figure 4-16. It can be seen that the forces reach
a first peak during the pre-cracked phase, up to a central displacement of approximately 30
mm. After dropping they increase again, reaching a plateau level at higher displacements.
The reactions of the wall and central sides reach higher values as these sides were
significantly longer. The total out of plane reaction forces are shown in Figure 4-18 to Figure
4-21. As seen for the individual sides, in all four cases the edge reactions reduce significantly
when the glass shatters. The pre-cracked reactions in tests 1 to 3 peaked between 35 and 50
kN. This could be due to the maximum capacity of the window before the failure of the glass
is reached. In test 4 this peak was slightly higher at approximately 55 kN, possibly due to
differences in the test set up. The post-cracked reaction magnitude instead varied between the
tests, ranging between 21 kN in the test 4 and almost 55 kN in test 3. In all cases the reaction
force approached a plateau level. This was similar to the observed behaviour of the laminated
glass material. The results for test 4 showed a significantly greater level of noise, due to the
different recording equipment used. The general shape of the results and magnitude of the
loads was however similar to what was measured in the previous tests.
101
deflection is apparent.
102
103
The in plane forces were also considered. For these, pre-cracked reactions have been ignored.
They would be dependant only on the small membrane forces present, and noise would have
prevented achieving a realistic estimate. The results showed a somewhat different behaviour.
A plateau level was reached as in the out of plane case, however the rise time was very short
and the forces were significantly more consistent as the deflection increased.
50
40
Reaction Force (kN)
30
20
Top reactions
Bottom reactions
10 Wall reactions
Central reactions
0
0 50 100 150 200 250 300
Figure 4-22: Total in plane edge reaction force along each edge for Test 3. See Figure
4-16 for the location of the edges.
Figure 4-22 shows a typical split between the reactions on the four window sides during Test
3. It can be observed that the shorter sides, the top and the bottom, exhibited a lower total
reaction force. The highest reaction force acted on the cubicle’s wall side, due to its higher
stiffness. This distribution, whilst likely, is not always realised, as other factors, such as the
blast pressure distributions can affect the reaction forces. As an example, the results shown in
Figure 4-17 display a different distribution of forces.
Figure 4-23 to Figure 4-26 present the total in plane force results for all the blasts considered.
As mentioned, the rise time of the reactions was lower than the time difference between the
images, 0.3 ms in the latest test, after the failure of the glass for most of the blasts. Test 2, in
Figure 4-24, showed reactions which increased as the deflection increases. However, this
gradual increase was still small compared with the initial magnitude. The out-of-plane pre-
104
cracked peak forces and plateau levels and in plane plateau levels are summarised in Table
4-5.
120
100
Reaction Force (kN)
80
60
40
20
0
-20 0 20 40 60 80 100 120 140
Figure 4-23: Total in plane reactions for Test 1.
120
100
Reaction Force (kN)
80
60
40
20
0
0 50 100 150 200 250
105
160
140
120
Reaction Force (kN)
100
80
60
40
20
0
0 50 100 150 200 250 300
80
70
60
Reaction Force (kN)
50
40
30
20
10
0
0 50 100 150 200 250 300
106
Table 4-5: Summary of the pre-cracked out-of-plane peaks and out-of-plane and in-
plane plateaux for the blasts considered.
Out of plane pre crack Out of plane post In plane post crack
Test
reactions peak (kN) crack plateau (kN) plateau (kN)
1 39 35.2 90.8
2 32 31.7 78.3
3 34 54.5 90.4
4 56 21.0 73.6
4.4. Discussion
The analysis described above relies on several assumptions. However, it is possible to assess
its reliability considering some of the results shown.
When considering the pre-cracked results, it was noted that the proportions of bending and
axial force plots shown above correlate with the deflected shape observed during the
experiments. The window exhibits most deformation near the edges, being relatively flat in
the central area, especially at earlier time points. This would indicate that most of the bending
moment would also be concentrated along the edges, which is confirmed in Figure 4-8.
Figure 4-9 again confirms this result, showing far higher stresses near the edges. This is not
just an effect of applying the FEA proportions to the raw data. Figure 4-27 below shows a
total strain plot in the same location and at the same time step used for Figure 4-9, where
again the increases in strains at the edges can be seen.
107
-3
x 10
5
3
DIC strain
-1
0 0.2 0.4 0.6 0.8 1 1.2
Position along the cut (m)
Figure 4-27: DIC strains along a cut in test 2.
The noise in the data in this phase however is fairly high. The error in the DIC strain
measurements is in the order of 100 microstrains, which is a significant percentage of the
strains observed before the failure of the glass. The filtering shown in Figure 4-9 goes
towards reducing this effect, however it cannot eliminate it completely at all points. Given the
vast quantity of data to be processed, an automated system had to be employed. The filtering
and the averaging over the last few data points does seem to reduce the variability of the data
along each edge, as can be seen in Figure 4-10, however this cannot be completely
eliminated. It is most likely that the total reaction forces produced will be of the right
magnitude; however it can be difficult to decide whether each individual datum can be
considered accurate. As an example, the curves shown in Figure 4-11 and Figure 4-12 show
that some calculated reactions do follow a similar trend to the strain gauges data, however
others depart from it, sometimes significantly.
The issue of small strains recorded was less significant for the post-cracked analysis. In this
case however one of the major issues was the material model employed to calculate the
reactions. Hooper [37] employed the Johnson-Cook model for his FEA model. Whilst this
does not represent the theoretical physical behaviour of the material, it does show a similar
trend as seen in the recorded data. It is hypothesised that the plateaux in the cracked
laminated glass stresses are caused by the delamination of the glass fragments from the PVB
membrane. This complex behaviour is of course very different from plasticity in metals,
108
however the final stress strain curves seem to be similar, hence justifying the adoption of the
model equation.
The plateau stresses calculated here are between 10-15 MPa for a strain rate of 3 s-1. This is
similar to the results obtained through Hooper et al.’s laboratory tests, giving credence to the
results obtained above. The laboratory results were not directly employed as the material
properties might be heavily dependent on crack spacing and other cracked laminate
characteristics. The fit therefore had to be performed again for the available blast data. In test
4 no data from strain gauges on the bottom of the pane were available and results from
another side were used. More importantly, the material model fits for this test cannot be
compared directly to the results obtained for other cases. As explained in Chapter 3, the DIC
calibration technique used does not guarantee the accuracy of the strain data. However, the
distribution of the strains was found to be more reliable. Therefore, in this case it was decided
that the reactions could be calculated, provided that the Johnson-Cook fit was performed with
the same DIC strains which would be used for the reaction calculation. As the reactions
obtained proved to be of the same general magnitude and presented the same time history
behaviour seen in tests 1 to 3, it is likely that the method managed to produce reliable
estimates.
The analysis performed here produced an estimate of the reactions along each edge. These
could then be compared with the results obtained in the previous chapter. In this case as well,
as shown in Figure 4-15, two peaks are present away from the side’s central point. This ties
in well with previous results, including those shown in Chapter 3.
The pre-cracked peak reactions show some consistency between the tests analysed. This is
due to the inherent capacity of the glass plies. Supporting this is the fact that the reaction peak
for Test 4 is higher than for Tests 1 to 3. In this case the window size was slightly reduced to
allow the use of a larger frame, which included double sided rebates. Smaller pane
dimensions imply a shorter span in all directions, which will cause a reduction of the bending
moments generated by the applied loading. Therefore, the glass layers could resist higher
pressures before the bending and membrane force stresses caused its fracture.
The post-cracked reactions clearly approach a plateau level at high deflections. This
behaviour is similar to what was observed when considering the cracked glass stress strain
behaviour, linking these two aspects. In further chapters the behaviour of the cracked
109
laminate will be studied to justify this behaviour and assess the effect of parameters such as
the density of cracks on the plateau forces observed.
The in-plane reactions showed a different behaviour, increasing very quickly to their plateau
levels. Their increased steadiness compared to the out-of-plane reactions was probably an
effect of the geometric variations of the glass angles θ. Whilst the out-of-plane reactions are
proportional to sin θ, the in plane ones are proportional to cos θ. This tends to vary much
more slowly at the angles recorded, which were in the region of 30°, as shown in the previous
chapter. This would cause the increase in stability of the recorded forces.
4.5. Conclusion
In this chapter the available experimental data was employed to calculate reaction forces
along the edges of the glazing panels. Pre-cracked reactions were estimated using a finite
element model to split bending moments and membrane forces strains from the total
deformations measured with DIC. These components were then calculated, and reaction
forces found. The post-cracked reactions were found using the DIC strains, assuming that
only membrane forces would be acting in this phase. A material model for the cracked
laminate was fitted using the strain gauge data and DIC strains in the region at the centre of
each panel. This was then used to convert DIC strains along all the edges into stresses.
The out-of-plane reactions show a distinct early peak before glass failure. In the first three
tests this seems to be of a similar magnitude, whilst in the fourth it is higher. The size of this
force is related to the geometric properties of the panel and hence to the maximum forces
which can be resisted by the glass before shattering. Hence, as the panel used for Test 4 was
smaller, higher pressures could be applied before the bending moment stresses reached the
material’s capacity.
Both in-plane and out-of-plane post-cracked reactions reached a plateau as the central
deflections increased. This is compatible with the observed behaviour of the cracked laminate
composite material, which shows a distinct plateau in its stress strain graph. One possible
cause for this is the delamination between the PVB interlayer and the outer glass plies, which
will tend to detach as the strains increase. This aspect of the material will be examined in the
next two chapters in an attempt to explain the observed behaviour. The PVB material
properties will be studied first, as they will influence the delamination behaviour. Afterwards,
the delamination will be investigated and its relation to the reaction results presented.
110
5. PVB Material Properties
5.1. Introduction
Previous chapters analysed the behaviour of laminated glass plates subjected to real blast
loads. The observed reaction forces followed a specific pattern, reaching a plateau which was
subsequently maintained until either the window deflection started reducing or the failure of
the laminated plate was reached. Explaining the presence of such a plateau requires a deeper
understanding of the constituent materials of the cracked laminated plate and of their relation.
Material models are therefore required to explain the behaviour of the composite at different
strain rates. The PVB membrane was the specific focus of this chapter. Several constitutive
equations have been fitted to PVB data, attempting to explain its behaviour over a range of
strain rates.
Figure 5-1 shows the typical chemical structure of Polyvinyl Butyral. In common with other
polymer materials, it exhibits a high degree of non-linearity in its strain response, which
complicates its modelling. Additionally, it presents a high strain rate dependency, showing a
significant shift in its behaviour at higher strain rates compared to lower rate deformations.
To build a complete model of the composite material it is important to be able to represent
the PVB behaviour at all rates of interest. Specifically, Hooper et al. [37] measured strain
rates as high as 30 s-1 during his blast tests, hence a model being able to account for the PVB
behaviour from quasi static levels to this rate would provide a strong basis for future finite
element analyses.
Figure 5-1: Typical chemical structure of the PVB polymer chain [62].
In this study, experimental data from uniaxial tests at different strain rates were used to model
the behaviour at the rates of interest. Both the hyperelastic characteristics of the material and
its time dependence were taken into account using visco-hyperelastic formulations. Two
111
general approaches were attempted. Firstly, Prony series were combined with two different
hyperelastic models. Both a reduced polynomial model and a Hoo Fatt model as proposed by
Hoo Fatt and Ouyang [63] were used. Following this, a different approach based on a full
mathematical solution of finite deformation viscoelasticity was applied, as proposed and
developed by several authors including Huber and Tsakmakis [64] and Hoo Fatt and Ouyang
[63]. The results from the various approaches were then compared considering their fit to the
experimental results and their ease of application.
5.2. Methods
5.2.1. Experimental data
Tensile tests at different strain rates were performed on the same PVB material to obtain the
data necessary for the model calibration. The material tested in all cases was Saflex PVB
produced by Solutia Inc. with product number RB-41. The equipment used was different for
the high and the low strain rate experiments, as no single machine could produce the range of
desired speeds.
All the tests at a rate above 0.2 s-1 and some of the tests at rates below this were performed by
Hooper et al. [65]. For all these experiments the samples were cut in a dog bone shape. The
height of the samples was 75 mm, with a 20 mm gauge length. The gauge length width was 4
mm and the material thickness was 0.76 mm throughout. The tests were performed on a
servo-hydraulic Instron tensile testing machine, employing a loss motion device to ensure a
constant strain rate at the higher speeds. Lightweight titanium alloy grips were used to
minimise inertial effects due to their mass. Strain data were collected optically using a high
speed camera. An image analysis was performed on each picture to obtain the deformations.
Force data were collected using a pizoeletric load cell for speeds above 0.1 ms-1 and a strain
gauge load cell for lower speeds. Tests were run at speeds of 0.005 ms-1, 0.1 ms-1, 0.32 ms-1,
1 ms-1, 2 ms-1, 5 ms-1 and 10 ms-1, providing results for average rates from 0.2 s-1 to 400 1 s-1.
Lower rate tests were performed by Wang. The specimens in this case were cut following the
dimensions specified in British Standard 37-2005 [66]. The size of the specimens is detailed
in Figure 5-2. Wang employed a low load and low speed single column 1kN Zwick tensile
testing machine. The strain data were collected optically in a manner similar to that used by
112
Hooper, with the difference that high speed cameras were not needed in this case. A Matlab
routine was employed to identify black lines in the images recorded during the tests. These
lines had been drawn at regular intervals on the samples. The distance between these could
then be used to calculate the strain at each time step. The inbuilt force sensor of the machine
was used to capture loading data. Tests were run at strain rates of 0.01 s-1, 0.02 s-1, 0.1 s-1 and
0.2 s-1.
Data from the two sets of experiments have been employed to produce material models for
the PVB. As both sets of experiments produced data for the rate of 0.2 s-1, these curves will
be compared.
35 (mm)
6.17(mm)
115 (mm)
1.56 (mm)
Figure 5-2: Dimensions of Dumb-bell PVB specimen.
5.2.2. Experimental Results
The experimental results are presented here before the material models will be described in
detail.
The data is shown as true stress versus stretch ratio λ, which is equal to the length of the
sample at time t ( lt ) over the original length ( l0 ) :

λ(t)= Eq. 5-1

The results of the low rate tests are shown in Figure 5-3. As expected the curves present
nonlinear behaviour together with significant strain rate sensitivity. At these rates however,
the shape of the curves does not change significantly, mostly showing an increase in the
overall stiffness as the deformation speed increases.
113
The tests performed at strain rates above and including 2 s-1are shown in Figure 5-4. The
grips in the tests at a rate of 200 s-1 appear to have slipped at a stretch of roughly 2.5. The
material non linearity is still very evident, as is the rate sensitivity. Additionally the shape of
the curves at the higher strain rates, starting from a rate of 8 s-1, changes significantly,
exhibiting a much higher stiffness at small strains. This is consistent with past observations
by Iwasaki and Sato [20]. The effect becomes marked for rates of 20 s-1 and above. Ideally a
PVB material model used for finite element application should be able to represent this
switch in behaviour to ensure that all situations can be modelled accurately.
80
0.01 /s data
70 0.02 /s data
0.1 /s data
60 0.2 /s data
(MPa)
50
(MPa)
Stress
40
TrueStress
30
20
10
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Stretch
Stretch Ratio
Figure 5-3: Low rate true Stress - stretch curves for PVB under uniaxial tensile loading.
Only Wang’s data for the 0.2 s-1 case are presented for clarity.
The failure point of the material did not change significantly with the rate, the final stretch
varying between ≈ 2.7 and ≈ 3.2 depending on the specific test.
Figure 5-5 shows a plot of the tests at 0.2 s-1 performed by Hooper et al. and Wang. The
curves are similar. Whilst there is a difference at stretches above 1.5, their divergence is
limited and will not affect the accuracy of the material model significantly. Therefore the data
from the two sets of experiments were used directly without applying corrections for shape
effects.
114
120
2 /s data
8 /s data
100 20 /s data
60 /s data
200 /s data
80
(MPa)
400 /s data
Stress(MPa)
60
True Stress
40
20
0
1 1.5 2 2.5 3 3.5
Stretch
Stretch Ratio
Figure 5-4: High rate true stress – stretch curves for PVB under uniaxial tensile
loading.
100
80 Wang's data
True Stress (MPa)
Hooper's data
60
40
20
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-5: A comparison of the results from Hooper [65] and Wang at a strain rate of
0.2 s-1.
115
5.2.3. Material model fit
The experimental results presented above highlight the high degree of rate dependence of the
PVB material. Material models need to take this into account and ideally include a
mechanism to represent this change in behaviour. Hooper et al. [65] argued that a viscoelastic
model employing a Prony series would not be able to cover all the strain rates of interest.
Specifically, it is challenging to ensure that a single Prony series model will guarantee the
changes in the small strain behaviour observed in the experiments. This issue could be
overcome by using different models for different rate conditions. As an example, the models
proposed by Liu et al. [21] represented different strain rate ranges and loading type and hence
ensure all the material behaviour was covered.
However, work by several authors, for example that by Hoo Fatt and Ouyang [63] shows that
models which exhibit the change in behaviour can be derived using finite deformation
viscoelasticity laws. Through these, different hyperelastic models can be combined, ensuring
that the whole range of deformation speeds is covered. The technique has been described
thoroughly by Huber and Tsakmakis [64] and has been used by Hoo Fatt and Ouyang [63],
Arruda and Boyce [67] , Amin et al. [68], [69] and others.
Both the Prony series and the finite deformation viscoelasticity approaches have been used in
this study. As mentioned above, two Prony series models have been developed using
different hyperelastic models, specifically a Hoo Fatt model and a reduced polynomial model.
As it was not possible to model the entire PVB behaviour with a single set of parameters, two
parameters fits have been performed for each hyperelastic formulation, one for the lower
strain rate and one for the higher strain rate data sets. In this case, the “low rate” model
covers tests up to and including an average rate of 8 s-1, whist the “high rate” models cover
rates higher than this. The rigorous finite deformation viscoelasticity derivation has then been
employed to formulate a single model covering the entire range of rates. This required the use
of different hyperelastic springs and a nonlinear viscosity function. The different material
models are described in more detail below.
Basic Assumptions and definitions
When deriving the material models described above, certain assumptions have been made
with regards to the PVB behaviour. The main assumption is that the material has been
116
considered incompressible. This implies that no change of volume takes place during the
tests.
The coordinates of a point in the material before deformations take place (t = 0) are defined
by the vector X. If this element of material is moved by the deformation, its new coordinates
∂xi
will be given by a new vector, x. The deformation tensor F is given by Fij = . The
∂X j
product of the diagonal members of the deformation tensor should be 1 for an incompressible
1
material. Therefore, if F11 is equal to the stretch ratio (λ), F22 and F33 will be equal to .
λ
For uniaxial tension, F will therefore be equal to:
 
λ 0 0 
 
 1 
F = 0 0  Eq. 5-2
 λ 
 1 
0 0
λ 

The left Cauchy-Green strain tensor B is given by B = FFT . Therefore, in this case:
 
λ 2 0 0
 
1
B= 0 0 Eq. 5-3
 λ 
 1
0 0 
 λ
The two invariants of B are defined by I1 = tr ( B ) = λ 2 + 2λ −1 and
1
I2 =
2
{(tr (B)) 2 − tr (BB )} = λ −2 + 2λ .
These quantities, especially B and its invariants, will be used below to derive all the material
models.
117
Prony series derivation
These viscoelastic material models include a non-linear hyperelastic function σ 0 (ε ) ,
dependent on strain, combined with a Prony series, g ( t ) , to include the time dependency.
The overall material model for a step strain relaxation test is shown in Eq. 5-4:
σ = σ 0 (ε ) × g (t ) Eq. 5-4
The hyperelastic function has been derived from the basic equation [69]:
∂ψ ∂ψ −1
T = − pI + 2 B−2 B Eq. 5-5
∂I1 ∂I 2
where T is the Cauchy true stress tensor. The function implies that a function representing the
work of the spring (ψ) needs to be differentiated with respect to the first two strain invariants
of the strain tensor B. Lamber-Diani and Rey [70] argued that to determine the portion of the
equation relating to I2, a biaxial test would be required. As this was not available here, work
functions related solely to I1 were chosen for this analysis. The specific hyperelastic functions
will be derived below.
The time dependent function g(t) was assumed to take the form of a Prony series:
N
 t 
g ( t ) = g∞ + ∑gi exp  −  Eq. 5-6
1  τi 
The model parameter fit was performed assuming six terms of this series, using either the τi
terms derived by Hooper or others at more regular logarithmic intervals to facilitate the fitting
process. As stated above, Eq. 5-4 is valid for stress relaxation test data. As only data from
ramp tests were available for the model calibrations, the related convolution integral would
have needed to be solved. As a solution of this would not be straightforward to obtain in this
case, the method described by Goh et al. [71] was employed to perform the material model
fit. A square minimisation technique was applied to determine the coefficients. The Solver
function of Microsoft Excel 2012 was used to solve the optimisation problem.
118
Hoo Fatt formulation
Hoo Fatt and Ouyang [63] used a model similar to an Ogden spring when deriving a material
law for a polymer material. The rationale for this formulation was to be able to represent the
sharp change in stiffness which took place at small strains. As the behaviour of the material
under consideration here is similar, the same hyperelastic law has been employed. However,
in this case the model was expanded into a summation in a similar manner to the classic
Ogden model. This allowed the sharp stiffness change to be reproduced with a single
hyperelastic spring in the Prony series approach. The work function used is:
ψ = ∑µi ( I1 − 3)
αi
Eq. 5-7
i
Two terms were used. If this function is substituted into Eq. 5-5, the hyperelastic strain is
given by:
T = − pI + 2α1µ1 ( I1 − 3)(α1 −1) B + 2α 2 µ 2 ( I1 − 3)(α 2 −1) B Eq. 5-8
The term p in the equation above can be found considering the boundary conditions σ22 = σ33
=0. Substituting into Eq. 5-8, a formula for p is derived:
(α1 −1) 1 (α 2 −1) 1

p = 2α1µ1 ( I1 − 3) + 2α 2 µ 2 ( I1 − 3) Eq. 5-9
λ λ
and therefore the final hyperelastic expression for uniaxial tension is given by:
 1
(
σ 11 = 2  λ 2 −  α1µ1 ( I1 − 3)
 λ
(α1 −1)
+ α 2 µ 2 ( I1 − 3 )
(α 2 −1)
) Eq. 5-10
Reduced Polynomial formulation
Various authors in the past have used a polynomial formulation to represent the PVB
material, e.g. [29]. As discussed above, a model independent of I2 was necessary in this case.
Therefore a reduced polynomial formulation was used in this work. Whilst this model does
not guarantee an equally accurate representation of the change in stiffness at higher rates, it
was considered that this procedure would guarantee the straightforward application of the
model in finite element software for further analysis of the composite material.
The work function in this case is given by:
119
ψ = ∑Ci ( I1 − 3)
i
Eq. 5-11
i
A third order model was employed. Using the function above with Eq. 5-5 the stress tensor is
given by:
(
T = − pI + 2 C1 + 2C2 ( I1 − 3) + 3C3 ( I1 − 3) B
2
) Eq. 5-12
The term p can be found using the same method described above, giving:
(
p = 2 C1 + 2C2 ( I1 − 3) + 3C3 ( I1 − 3)
2
) λ1 Eq. 5-13
Inserting this in Eq. 5-12 and deriving an equation for σ11 gives:
 1
(
σ 11 = 2  λ 2 −  C1 + 2C2 ( I1 − 3) + 3C3 ( I1 − 3)
 λ
2
) Eq. 5-14
5.2.4. Finite deformation viscoelasticity
Model Derivation
The finite deformation viscoelasticity models are derived by solving the mathematical
equations representing a “springs and dampers” system, similar to those assumed in
traditional viscoelasticity theory. However, in this case all small strain assumptions are
avoided to produce a solution valid for any level of deformation. For the situation of interest
here a system of two nonlinear springs and a damper is assumed, as shown in Figure 5-6.
This is the same system as model A described by Huber and Tsakmakys [64], and was also
used by several other authors previously mentioned [63], [68].
120
Figure 5-6: A summary of the material model assumed to model PVB.
The total observed stress will be equal to the sum of the stresses affecting each spring. The
single spring will be referred to as the equilibrium spring (“eq” in symbols), whilst the spring
in series with the damper will be referred to as the overstress spring (“oe” in symbols). The
deformation of the equilibrium spring and the summation of that of the damper and overstress
spring series will be the same and equal to the overall measured deformation. Therefore, as
shown in Figure 5-6, the equilibrium deformation of the spring will be equal to B. The
deformation of the overstress spring is instead dependant on the viscous behaviour of the
damper, introducing a time dependency in the model.
Huber and Tsakmakys [64] and Hoo Fatt and Ouyang [63] provide a detailed derivation of
the general equations governing this model. The first step of the method, introduced by
Lubliner [72] in the context of finite deformation plasticity, is to decompose the overall
system deformation into an equilibrium and an instantaneous part, F = Fe Fi . Fi represents the
deformation state which would be reached if the load was instantaneously removed from the
deformed sample. It could therefore be associated with the deformation of the damper. Fe is
instead assumed to be the deformation of the overstress spring, hence the strain responsible
for the stress acting in the damper and spring series.
121
Additionally, it is assumed that the system will obey the thermodynamic condition [69]:
− ρ rψɺ + S E iL ≥ 0 Eq. 5-15
ɺ −1 , measuring the rate

where SE is Cauchy (true) stress and L is the velocity gradient, L = FF
of change of the velocity in the volume of the material at a given moment in time.
Huber and Tsakmakys show the derivation of a system of equations from these assumptions,
and conclude with a material model given by:
∂ψ eq ∂ψ eq ∂ψ oe ∂ψ
T = − pI + 2 B−2 B −1 + 2 Be − 2 oe Be −1 Eq. 5-16
∂I1B ∂I 2B ∂I1Be ∂I 2Be
2
Bɺ e = LB e + B e LT − B e Toe Eq. 5-17
η
∂ψ oe ∂ψ oe −1
Where Toe = 2 Be − 2 B e , Be is the left Cauchy-Green strain tensor for the
∂I1Be ∂I 2 Be
overstress spring and η represents the viscosity function of the damper. Spring functions
similar to those derived above for the Prony series models can then be substituted into the
equations. This derivation method has been used here, assuming a one term Ogden [73]
function for the equilibrium spring and a three terms Hoo Fatt’s spring function for the
overstress response. Substituting these springs into the system above, the following equations
are obtained for uniaxial tension:
1
− αo  1
σ 11 = µo λ α − µo λ
o 2
+ 2 λ2 − ×
 λ
Eq. 5-18
(
× α1µ1 ( I1 − 3)
(α1 −1)
+ α 2 µ 2 ( I1 − 3 )
(α 2 −1)
+ α 3 µ 3 ( I1 − 3 )
(α 3 −1)
)
2  1
Bɺ e = LB e + B e LT − B e 2  λ 2 −  ×
η  λ
Eq. 5-19
(
× α1µ1 ( I1 − 3)
(α1 −1)
+ α 2 µ 2 ( I1 − 3 )
(α 2 −1)
+ α 3 µ 3 ( I1 − 3 )
(α 3 −1)
)
The constants of these equations were fitted using the method explained below. Additional
detail of the derivation is provided in Appendix A.
122
Data fit method
Hoo Fatt and Ouyang presented a method to fit the model constants for this kind of
mathematical representation. In her paper, the constants for the two springs were obtained
first. At low, quasi static, rates it is assumed that the damper will offer no resistance to the
deformation. Therefore, the overstress spring will not be stretched, giving Fi = F and Fe = I .
The measured stress will hence be caused by the deformation of the equilibrium spring.
Therefore, data sets in this range of rates can be used to fit the parameters of the equilibrium
spring directly. As the two lowest strain rates data available were similar (see Figure 5-3), it
was assumed that the quasi static condition was being approached. The Ogden model spring
could then be fitted with the data from the lowest rate experiment (0.01 s-1).
Again, the data for the higher rates, at 200 s-1 and 400 s-1, showed less variation than data at
lower speeds. It was therefore assumed that these sets would approximate an instantaneous
response of the material. In this condition, it can be assumed that the damper will not move,
and hence that the overstress spring deformation will be equal to the total deformation of the
material, specifically Fi = I and Fe = F . As the total measured stress is equal to the sum of
the stresses acting on the two springs and the equilibrium spring stress can be calculated
using the material constants obtained in the previous (quasi static) step, the stress acting on
the overstress spring could be found. This allowed a stress strain curve to be derived for the
overstress spring, whose material constants could then be fitted. This procedure was followed
using the 400 s-1 data.
Figure 5-7 shows the two extreme stress-stretch curves together with the fitted models.
123
Figure 5-7: The lowest and highest available strain rate data were used to fit the two
spring models. The figure shows the raw data and the model fits.
Once the parameters of the springs had been fitted, the viscous function had to be considered.
The data showed that a constant would not be sufficient to model the different material
behaviour. A function as used by Hoo Fatt and Ouyang was assumed and the parameters were
fitted to the data. To achieve this, the method described in the same paper was followed. This
assumed a matrix Fe of the form:
 
 
 λe 0 0 
 1 
Fe =  0 0  Eq. 5-20
 λe 
 
0 1 
0
 λe 
 
Therefore Be is given by:
124
 
 2 
 λe 0 0
 1 
Be =  0 0 Eq. 5-21
 λe 
 1
 0 0 
 λe 
Using this assumption, λ and λe could be substituted in the overstress strain rate differential
equation, which becomes:
ɺ
λλ 2T
λɺe = e − oe λe Eq. 5-22
λ 3η
2λλe Toe
Therefore, η = . The values of η found in this way were fitted to the proposed
(ɺ − λɺ λ
3 λλe e )
equation:
{ ( ) }
η = Cη1 1 − exp ( Cη 2 ( I1B − 3) ) + Cη 3 ( Cη 4 I13B + Cη 5 I12B + Cη 6 I1B + Cη 7 )
e e e
Eq. 5-23
This function was substituted into the system of equations of the model. Values for η and the
required deformation invariants were obtained from the available data. Toe was calculated by
subtracting the equilibrium spring stress from the experimentally measured stress. This data
was used to calculate λe from the known overstress spring equation. As this equation is
nonlinear and difficult to invert, the stretch was found using a numerical solver in Matlab
[74]. The overstress rate was then found through numerical differentiation. A polynomial was
fitted to the λe data to achieve this, since otherwise the noise present in the experimental
curve would have prevented realistic estimates being obtained. The values of η obtained were
then used to fit the constants required.
Once the parameters were obtained, the resulting system of equations was solved using
Matlab. An inbuilt numerical solver was employed for the differential equation.
125
5.3. Results
5.3.1. Prony series - Hoo Fatt spring
Figure 5-8, Figure 5-9 and Figure 5-10 show the experimental data and the fitted material
models for the low strain rate (up to 8 s-1) cases. The plots show that this function does not
represent the material behaviour at the lowest strain rates accurately (Figure 5-8). The fit
improves for the curves in Figure 5-9. However, the quality decreases again for the two
higher rates (Figure 5-10), especially for the 8 s-1 data set, where the initial behaviour is not
captured by the model. Whilst the shapes of the stress-stretch curves change slightly
throughout the range of rates, it is at 8 s-1 that the stress curve starts to show the increase in
initial stiffness that cannot be modelled by the basic hyperelastic spring. A plot of the relative
error of these fits will be shown later.
Figure 5-11 and Figure 5-12 show the data for the high strain rate (from 20 s-1) cases. The
model captures the initial stiffness quite accurately. Additionally, the model follows the
experimental results at higher stretches more accurately than the low rate cases. The fit again
seems to be of lower quality for the highest rate cases, where the initial stiffness does not
match the experimental stiffness. However, the model again follows the experimental data
closely after this stage. Figure 5-20, Figure 5-21 and Figure 5-22 show plots of the relative
errors for selected strain rates and will be discussed below. Table 5-1 and Table 5-2
summarise the parameters which were used in both the low and high rate cases.
126
70
60
0.01 /s Experimental Data
50 0.01 /s Model Fit
True Stress (MPa)

0.02 /s Model Fit
40
30
20
10
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Stretch Ratio
Figure 5-8: Low strain rate Prony series with Hoo Fatt spring fit for the tests at rates
0.01 s-1 and 0.02 s-1.
120
100
True Stress (MPa)
80
60
40 0.1 /s Experimental Data

0.1 /s Model Fit
20
0.2 /s Model Fit
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-9: Low strain rate Prony series with Hoo Fatt spring fit for the tests at rates 0.1
s-1 and 0.2 s-1. Hooper’s data are shown for the 0.2 s-1 case.
127
120
100
True Stress (MPa)
80
60
40 2 /s Experimental Data
2 /s Model Fit
8 /s Experimental Data
20
8 /s Model Fit
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-10: Low strain rate Prony series with Hoo Fatt spring fit for the tests at rates 2
s-1 and 8 s-1.
120
100
True Stress (MPa)
80
60
20 /s Model Fit
20
60 /s Model Fit
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-11: High strain rate Prony series with Hoo Fatt spring fit for the tests at rates
20 s-1 and 60 s-1.
128
120
100
True Stress (MPa)
80
60

200 /s Model Fit
20
400 /s Model Fit
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Stretch Ratio
Figure 5-12: High strain rate Prony series with Hoo Fatt spring fit for the tests at rates
200 s-1 and 400 s-1.
Table 5-1: Material parameters for the Hoo Fatt’s spring.
Parameter Low rate High rate

(up to 8 s-1) (above to 20 s-1)
µ1 (MPa) 4.37 13.6
α1 1.38 1.37
µ2 (MPa) 5.68 80.2
α2 0.618 0.645
129
Table 5-2: Prony series parameters for the Hoo Fatt’s spring viscoelastic model.

τ1 (s) 2.45x10-5 2.45x10-5
g1 0.1 0.759
τ2 (s) 2.21 x 10-3 2.21 x 10-3
g2 0.1004 0.0624
τ3 (s) 4.98 x 10-2 4.98 x 10-2
g3 0 0
τ4 (s) 0.624 0.624
g4 0.139 0
τ5 (s) 24.9 24.9
g5 0.213 0
τ6 (s) 1000 1000
g6 0.224 0
ginf 0.224 0.178
5.3.2. Prony series – Reduced polynomial spring
The fitting of the reduced polynomial spring model was again performed for the low strain
rate and the high strain rate cases separately. The low strain rate (up to 8 s-1) results are
shown in Figure 5-13, Figure 5-14 and Figure 5-15. The results are similar to those for the
Hoo Fatt spring model. The model follows the experimental data for the 0.1 s-1 and 0.2 s-1
data sets in Figure 5-14. However, the slower rates (Figure 5-13) and the higher rates (Figure
130
5-15) do not show a good agreement. The lower rates stiffness is over estimated, producing
higher stresses than those produced experimentally. The 8 s-1 case again shows that the model
could not capture the initial stiffness, producing lower stresses than the experiment.
Table 5-3: Material parameters for the reduced polynomial hyperelastic spring.

C1 (MPa) 40.7 311
C2 (MPa) 75.7 0.0112
C3 (MPa) 0 1.3
Table 5-4: Prony series parameters for the reduced polynomial viscoelastic model.

τ1 (s) 1x10-4 1x10-6
g1 0 0
τ2 (s) 1 x 10-3 1x10-5
g2 0.898 0.948
τ3 (s) 1 x 10-2 1x10-3
g3 0 0.0151
τ4 (s) 0.1 0.1
g4 0 0.00112
τ5 (s) 1 10
g5 0 0
τ6 (s) 100 1000
g6 0.102 0
ginf 0 0.0354
131
The results of the fits for the higher strain rates (above 20 s-1) are shown in Figure 5-16 and
Figure 5-17. In this case the model follows the experimental results fairly closely once the
material softens, though without modelling the initial deformation accurately. This is
probably due to limitations in the spring model used, specifically that the mathematical
formulation of the reduced polynomial model could not capture the change in stiffness
without employing a much greater number of constants. Table 5-3 and Table 5-4 show the
parameters which were used for these fits.
70
60
0.01 /s Model Fit
True Stress (MPa)
50
40 0.02 /s Model Fit
30
20
10
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Stretch Ratio
Figure 5-13: Low strain rate Prony series with the reduced polynomial spring fit for the
tests at rates 0.01 s-1 and 0.02 s-1.
132
120
100
True Stress (MPa)
80
60
40 0.1 /s Experimental Data

0.1 /s Model Fit
20
0.2 /s Model Fit
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
tests at rates 0.1 s-1 and 0.2 s-1. Hooper’s data are shown for the 0.2 s-1 case.
120
100
True Stress (MPa)
80
60
2 /s Model Fit
20
8 /s Model Fit
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
tests at rates 2 s-1 and 8 s-1.
133
120
100
True Stress (MPa)
80
60
20 /s Model Fit
20
60 /s Model Fit
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-16: High strain rate Prony series with the reduced polynomial spring fit for the
120
100
True Stress (MPa)
80
60

200 /s Model Fit
20
400 /s Model Fit
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Stretch Ratio
Figure 5-17: High strain rate Prony series with the reduced polynomial spring fit for the
5.3.3. Finite deformation viscoelasticity
Figure 5-18 and Figure 5-19 show selected plots of the experimental data with their
respective model fits obtained using finite deformation viscoelasticity. The quality of the fit is
134
again lower for the 8 s-1. However, in this case the model overestimates the initial stiffness
rather than underestimating it as in the Prony series results. The lower rate cases show more
accurate results than for the Prony series fittings, especially in the 0.02 s-1 case. At the lower
rates the model shows some noise at low stretches, up to about λ = 1.1. This is due to the
difficulty of reaching a stable numerical solution for the nonlinear differential equation in
these cases. Also, the model does not capture the sudden jump in the 200 s-1 curve above λ =
2.5. However, it is thought that this was due to grip slippage during the test. The model is
able to switch between the high rate behaviour and the low rate behaviour, as can be seen by
the difference between the 200 s-1 and the 0.02 s-1 curves. Table 5-5 and Table 5-6 shows the
parameters which were fitted for the two springs.
140
0.02 /s data
120 0.02 /s model
2 /s data
2 /s model
True Stress (MPa)
100
8 /s data
8 /s model
80
60
40
20
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-18: Finite deformation viscoelasticity material model fit for the rates 0.02 s-1, 2
s-1and 8 s-1.
135
True Stress (MPa) 150
100
50 20 /s data
20 /s model
200 /s data
200 /s model
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-19: Finite deformation viscoelasticity material model fit for the rates 20 s-1 and
200 s-1.
Table 5-5: Hyperelastic spring parameters for finite deformation viscoelasticity.
µo (MPa) αo µ1 (MPa) α1 µ2 (MPa) α2 µ3 (MPa) α3
0.632 4.141 117.373 0.746 16.035 0.806 -121.468 0.768
Table 5-6: Damping function parameters for finite deformation viscoelasticity.
Cη1 (MPa s) Cη2 Cη3 (MPa s) Cη4 Cη5 Cη6 Cη7
1.131 -0.248 0.026 -1.527 6.626 4.876 -5.893
5.4. Discussion
All models can capture aspects of the PVB behaviour at the various strain rates, though
showing different limitations. The Prony series models for low rates struggle to represent the
data at both the bottom and the top end of their rate range. They produce stresses greater than
136
measured for the slower tests. The 8 s-1 curve fit instead does not show the initial stiffness,
which starts being evident at this stage. The high rate (above 20 s-1) fits show a good
agreement beyond the initial stiffness region. However, both models struggle to account for
the 200 s-1 and 400 s-1 cases. This error is much smaller in the Hoo Fatt spring model, which
shows the correct behaviour, albeit without reaching the correct stress, whilst the reduced
polynomial model struggles to show the initial stiffness in any of the high rate cases.
Employing more terms might have improved this. However, the results might then show
significant oscillation, precluding their use in finite element models. After the initial stiffness
phase though the high rate models produce a better fit than the low rate models. The greater
accuracy of the high strain rate fits might be due to the fact that the shape of the stress stretch
curves for these cases shows relatively less variation, allowing a single model to capture their
behaviour. It could therefore be argued that at least a third model could be fitted for
intermediate rates, including the 8 s-1 rate. More experiments would be needed at speeds
above and below this rate to achieve this.
The rigorous finite deformation viscoelasticity is able to fit most of the data sets considered.
Again, the fit quality is lower at lower stretches, however it tends to improve significantly as
the deformation increases. Additionally, the model can represent the higher stiffness region at
high strain rate employing just one set of constants, thereby opening the possibility of using a
single model to represent the PVB material for a range of situations. This however, comes at
the cost of greater model complexity, both in terms of the model derivation and of the
procedure needed to fit the material constants.
Figure 5-20, Figure 5-21 and Figure 5-22 show plots of the relative errors for strain rates of
0.02 s-1, 8 s-1 and 200 s-1 . These data were calculated by dividing experimental stresses by
modelled stresses at each point. It can be seen that the relative error for the finite deformation
model, shown in black lines for each rate, is generally lower, i.e. the curves are closer to 1.
The greater errors of all the models tend to be at very low strains. This is most likely due to
the difficulty of mathematically representing the sharp change in stiffness observed in the
available data. Also, confirming the observation made before, the Prony series models show
greater relative errors for the lower strain rate cases, again especially at low strains. The finite
deformation model performs better in this region, even though it does show high relative
errors at very low stretches. The quality of the fits discussed above is therefore highlighted in
these results. However, it should also be considered that a relatively small absolute error will
translate to a more significant relative error if the stress magnitude is smaller, as is the case at
137
stretch ratios approaching 1. This should also be borne in mind when considering the
seemingly constantly improving fit quality as the stretches increase. In fact, in this case the
larger absolute value of the stress will tend to minimise relative errors. Whilst the limitations
of this error estimate need to be considered, the plots shown below are still a useful tool to
compare the different models at different strain rates, showing in general that more accurate
results are produced by the finite deformation viscoelasticity model.
2
Experimental stress / Model stress
0.02 /s Prony Hoo Fatt Spring error

0.02 /s Prony Reduced Polynomial Spring error
1.5 0.02 /s Finite Deformation Viscoelasticity error
0.5
0
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8
Stretch Ratio
Figure 5-20: Relative fit error for the 0.02 s-1 strain rate case.
2
Experimental stress / Model stress
1.5
8 /s Prony Hoo Fatt Spring error

0.5
8 /s Prony Reduced Polynomial Spring error
8 /s Finite Deformation Viscoelasticity error
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-21: Relative fit error for the 8 s-1 strain rate case.
138
Experimental stress / Model stress 2
1.5
200 /s Prony Hoo Fatt Spring error

0.5 200 /s Prony Reduced Polynomial Spring error
200 /s Finite Deformation Viscoelasticity error
0
1 1.5 2 2.5 3 3.5
Stretch Ratio
Figure 5-22: Relative fit error for the 200 s-1 strain rate case.
5.5. Conclusion
The PVB tensile experimental data demonstrated the high level of non-linearity and strain
rate dependency exhibited by this material. Further analysis of the overall laminated glass
composite material will require reliable material models for this important component, which
will need to be modelled to account for various loading condition and strain rates. Two
methods have been used to produce three material models which account for this shift in
behaviour. In the first approach, two hyperelastic functions were used to produce four
separate Prony series viscoelastic models. These were fitted to the high rate experimental
data, including 20 s-1 and above, and to the low rate experimental data, including 8 s-1 and
below.
The Hoo Fatt hyperelastic spring model produced more accurate results for the high rate fit,
even though the quality deteriorated somewhat at higher rates. The low rate fit did not
manage to capture the change in behaviour of the material in the range considered. This
caused it not to represent closely the small strain behaviour of the fastest case, as the
experimental curve started to show an initial higher stiffness. Additionally the behaviour of
the lower rate data sets was not captured accurately. The same limitations were observed in
the reduced polynomial model, which produced results of similar quality. This representation
modelled more accurately the material behaviour at intermediate stretches whilst it produced
worse results at lower deformations. However, the results followed a similar pattern to that
139
observed previously. These results suggest that, ideally, a third Prony series model might be
needed for intermediate strain rates, thereby avoiding the need to produce a single model
which has to account for these changes in behaviour. Both Prony series models performed
significantly better at rates above 20 s-1, especially at larger stretch levels. The Hoo Fatt
spring performed better at lower stretches, as it was better able to represent the rapid change
in stiffness. The reduced polynomial model could not manage this, showing higher errors at
this stage.
To overcome the Prony series limitations, a finite viscoelasticity model was also derived
following the method reported by Huber and Tsakmakys [64]. This was shown to account for
the entire range of rates considered, generally with fit qualities which at least matched the
results of the Prony series. However, both the model derivation and fitting procedure are
significantly more complex. Additionally, the model in the form presented requires a large
number of constants to be determined (eight for the springs and seven for the damper). Also,
as the model is not currently included in commercially available finite element software, a
user material subroutine would need to be developed in order to apply it.
If the phenomena of interest can be limited to a specific small range of strain rates, it would
probably be advantageous to employ one of the Prony series viscoelastic models, as these
require fewer constants to be included and they are commonly supported by FE software.
However, if the rates of interest are not known or are spread over a larger range, the finite
viscoelasticity model might well prove advantageous. In this case the large number of
constants to be fitted would still be smaller than the total number of parameters for both the
Prony series models and the overall accuracy would be improved.
In the next chapter the delamination of glass layers from the PVB membrane was studied
employing FEA models. To achieve this, the PVB membrane was modelled employing the
reduced polynomial Prony series model. Whilst the finite viscoelasticity model could provide
a better approximation of the behaviour, it was decided that the complications relating to its
implementation would make this not feasible for this project. The chosen model is instead
already implemented in ABAQUS, making its use straightforward.
140
6. Glass – PVB Delamination
6.1. Introduction
The cracked laminate behaviour will be affected by two main elements, the PVB membrane
material properties and the behaviour of its interface with the glass fragments. In the previous
chapter the properties of the membrane were studied in detail. Some of the results obtained
will now be used to consider the delamination behaviour of the composite material.
During a blast, the pressure wave causes the glass panes to fragment at relatively low
displacements. As the PVB membrane is subsequently stretched, the glass fragments will
tend to detach from its surface through a delamination process. This absorbs considerable
energy, and tends to cause a plateau in the forces recorded as a steady-state equilibrium is
reached and the delamination progresses at a constant rate. As a first step to estimate the
delamination effect on a realistic window system, the basic energy absorption information
needs to be determined. Hooper [3] performed several tensile tests at different speeds on
laminated glass samples with varying numbers of cracks. In the present study some of these
tests are employed to calculate delamination energies using one of the PVB models derived in
the previous chapter. Previous authors, for example Muralidhar et al. [29], developed
analytical solutions to calculated the dissipated bond energy directly from measured the
external work. However, the PVB model used in this work has a more complex algebraic
formulation, which prevented the derivation of a closed form solution for the delamination
energy. Therefore, FEA models were used to fit the delamination energy parameters instead
of explicit analytical solutions. Both the plateau stresses obtained experimentally and the
speed of the delamination visible in the available high speed pictures were used to compare
the experimental and modelling results.
Once the energy values had been determined, a parametric study was performed on the effect
on the plateau stresses of different crack configurations. Whilst in a real blast situation the
cracks are concentrated in several typical areas, as described in Chapter 3, their exact density
and orientation varies. Additionally, some cracks might not be aligned in the two panes,
introducing significant complexities in the material’s behaviour. The effect of these factors
therefore needs to be considered to inform possible designs. Experimental data were available
141
for 20 mm and 10 mm crack spacing. This information was extended to 5 mm distances
through FEA analysis to produce estimates of wider application. Additionally, diagonal
cracks and cracks in different positions on the two sides of the PVB have been explored to
assess their effect on the overall behaviour.
6.2. Method
The delamination energies and effect of crack distribution were the subject of separate
investigations. Additionally, the delamination speed, one of the parameters used to verify the
accuracy of the results, was extracted from the experimental images. The procedure used for
each of these steps is described below.
6.2.1. Experimental data
The tests were performed by Hooper [3] and consisted in laboratory tensile experiments on
pre-cracked laminated glass rectangular samples, of size 150 x 60 mm. The glass layers were
3 mm thick. Whilst the PVB membrane thickness was varied, only the results from samples
with a 1.52 mm thick interlayer were used in this analysis. The cracks were generated by
scoring the glass with a dedicated tool and applying a sharp shock to propagate full depth
cracks.
142
Figure 6-1: Diagram of a typical single crack experimental sample.
Specimens were prepared with three crack patterns: a single crack at mid length and multiple
cracks at 20 mm and 10 mm spacing. Aluminium tabs were bonded to each end of the
specimen to connect them to the actuator arm of the testing machine. The tests were
performed using an Instron high speed servo-hydraulic testing machine at several speeds. The
0.1 m/s, 1 m/s and 10 m/s results are analysed here. Figure 6-1 shows a diagram of a single
crack specimen. The loads were measured using a piezoelectric load cell and the
displacements were recorded by an LVDT sensor measuring the position of the actuator.
Photoelasticity techniques were used to highlight the position of the delamination front either
side of the cracks as the test progressed. A Photron FASTCAM SA1 high speed camera was
used to record colour pictures at different frame rates depending on the speed of the test.
143
Figure 6-2: Section showing the expected PVB behaviour during the tests.
Figure 6-2 shows the expected behaviour of the PVB membrane during the test. Delamination
is expected to take place on both sides and both above and below the crack locations. Figure
6-3 shows a typical test image. A delamination front and a glass edge are identified in the
figure.
144
Figure 6-3: Image of a single crack sample, obtained using a high speed camera.
The recorded force and positions of the delamination fronts were employed in the present
study to assess the delamination stress plateaux and the propagation speeds of the failure
events. Figure 6-4 shows an example plot of the experimental data. The load had been
converted to stress dividing the force data by the unloaded PVB cross-sectional area. The
plateau stress values employed in later analyses are given in Table 6-1.
145
20
18
16
14
Stress (MPa)
12
10
0
0 50 100 150 200 250
Displacement (mm)
Figure 6-4: Experimental stress-displacement data for a 1 m/s loading speed on a 10

mm crack spacing specimen.
Table 6-1: Experimental plateau stresses used in the analysis [3].
Crack Configuration Test Speed (m/s) Repetitions Plateau Stress (MPa)
Single Crack 0.1 1 13.2
Single Crack 1 1 21.3
Single Crack 10 1 26.3
10 mm Spacing 1 1 17.5
20 mm Spacing 1 1 16.5
Whilst the force readings were already available, the speed of the delamination propagation
was calculated for use in this study. The approximate positions of the four main elements, the
top and bottom delamination fronts and the glass edges, were determined in the images. Their
distance could then be found at several time steps and the delamination velocity calculated.
146
Two methods were employed for this, depending on the test information available. Where the
images were clear enough, measurements could be performed with an automatic procedure
using Matlab. An algorithm was prepared to locate areas where the gradient of the intensity
of the colour was highest. The software then assumed that these areas indicated the location
of either a delamination front or of a glass edge and recorded its position. The distance
between the uppermost two locations and the lower two was then calculated, thus providing
the length of the delaminated area in the picture. The procedure was repeated on all the
available images and the speed of delamination was obtained. As a checking procedure, the
same algorithm modified the original pictures, including in them a horizontal white line at
each of the four locations together with a vertical line to indicate the section along which the
analysis was performed. A sample of these modified pictures is shown in Figure 6-5 below.
Figure 6-5: Modified experimental picture with white lines inserted at delamination and
glass edges locations.
147
Where the images were complicated by the presence of unwanted glass fragments or in the
case of “multiple cracks” specimens when more delamination fronts were present, the
automated algorithm did not produce acceptable results. In these situations some of the
pictures, taken at regular intervals, were modified by hand. White horizontal lines one pixel
wide were included at the relevant locations. The delaminated areas and speeds were then
determined from these. Four pictures were employed for each test.
6.2.2. Delamination energies
The procedure used to determine the PVB to glass delamination energies from these tests is
similar to that applied to peel tests. It involves calculating the external work done from the
force displacement information, and then deducting the dissipated energy components, in this
case solely the internal strain energy of the PVB material. The remaining energy can be
assumed to be absorbed by the delamination process.
In previous work using similar test set ups ( [29], [30], [20] ), the delamination energies of
PVB-to-glass interfaces was determined using results from predominantly lower speed tests,
below and assuming different PVB material properties. Muralidhar et al. derived closed form
equations considering several hyperelastic material formulations. A polynomial formulation
was chosen to calculate the delamination energy in the experimental cases considered.
Butchart and Overend instead employed a linear-elastic viscoelastic model. Iwasaki did
include some higher strain rate tests. He then used an elastic PVB model to calculate the
delamination energies. In the present work a more realistic material model, including both
non-linear hyperelasticity and viscoelastic effects has been formulated. However, the
complexity of the material model prevented a closed form analytical solution for the
delamination energy from being formulated. Instead, a finite element model employing
cohesive elements was used.
FEA Modelling
The analysis was performed with ABAQUS release 6.10. Three main elements were included
in all the models employed: (i) the internal PVB membrane, (ii) the glass fragments and (iii)
two layers of cohesive elements connecting the glass and the PVB parts.
148
The geometry of the models was slightly modified from the experimental geometry due to
numerical issues. A 5 mm long PVB part was included on the top and bottom of the main
body of the materials. The overall dimension was kept constant at 150 mm. This part was
added to dampen vibrations, which would otherwise have prevented significant information
from being extracted from the FEA data. However, as potentially this addition could have
significantly modified the results, such as the plateau stresses, the data obtained from one
model without these end parts were compared with those obtained when they were included.
All other material and boundary condition parameters were kept constant. The results of this
comparison are shown below. Additionally, a half model with a boundary condition on the
vertical symmetry plane was employed to increase the computational efficiency of the
calculations.
Once a basic model geometry was fixed, it was used for all the speeds and material properties
considered. This final geometry consisted of a model of 150 x 30 mm overall dimensions.
The glass layers were 3 mm thick and of 140 mm overall height. They were split into separate
parts at mid-lenght to simulate the pre-existing crack. The PVB interlayer was modelled as
1.52 mm thick. Separate PVB components were used for the top and bottom composite areas.
The cohesive zone was represented with two 0.01 mm thick element layers added to each
face of the membrane. Tie conditions were set between these and the glass faces touching
them. The PVB parts were also connected with a tie condition. Only the PVB material
elements were included in the constraint, excluding the cohesive layers. This was done to
avoid numerical issues which emerged when the cohesive elements were included in the tie.
Figure 6-6 shows the geometry of the models with the main dimensions.
149
Figure 6-6: Main dimensions and features of the single crack delamination models.
The glass parts were meshed with 1 mm cubic stress elements (C4S4). The PVB membrane
was meshed with the same type of elements. However, their thickness was reduced to 0.76
mm to ensure at least two elements were present in the thickness direction of the layer. It was
considered that, for this analysis, through thickness effects would not be dominant and hence
the use of two elements would be acceptable in this direction.
The cohesive elements were also set to have 1 mm sides. However, their thickness was 0.01
mm. They were generated from the PVB mesh, and hence were a feature of the same parts of
the model. The typical mesh at one of the corners of the model is shown in Figure 6-7.
150
Figure 6-7: Detail of the mesh used in the model. The view shows a corner of the model.
The properties of the main components are summarised in Table 6-2. Glass was represented
using linear elastic material properties, as the stresses it would be subject to would generally
not be enough to justify non-linear approaches. Its Young’s modulus was 70 GPa, with a
density of 2500 kg/m3 and a Poisson’s ratio of 0.33. The PVB was modelled with the reduced
polynomial viscoelastic models derived in the previous chapter. As it had been decided that
the details of the material model employed for this material would affect the energy
parameters, each FEA model was run with both the low rate and the high rate coefficients
derived. The cohesive elements were modelled using a cohesive zone modelling (CZM)
technique. This, as described by Chandra et al. [75], requires the user to specify of the
behaviour of the material before a peak stress is reached, the peak stress itself and a
degradation law which governs the loss of capacity past the peak. A bilinear formulation was
employed, assuming that the behaviour both before and after the peak capacity could be
modelled with linear laws. Figure 6-8 shows a stress-displacement plot of such a law. The
failure during the tests will take place through a combination of mode I and mode II failure
deformations. Muralidhar et al. [29] state that their results indicate that for this situation
mode-mixity can be ignored. This approach was implemented here defining the same
parameters for the failure modes. A quadratic damage initiation model available in the
software represented by Eq. 6-1 was used to define the onset of failure [76].
151
2 2 2
 S n   S s   St 
 o + o  + o  Eq. 6-1
 S n   S s   St 
where Sn , S s and St are the stresses in the normal and the two shear directions and S no , S so
and Sto are the capacities for the same modes.
Figure 6-8: Cohesive zone model stress-displacement law used in this study.
A traction elasticity formulation was used for the pre-peak phase. The traction stiffness ( k )
parameters were determined assuming the cohesive material would behave in a similar way
to the PVB bulk material. Employing the Young’s modulus derived by Hooper [3], a stiffness
parameter was calculated assuming a mesh size of 1 mm square and of 0.01 mm thickness.
Therefore, the tensile stiffness of the cohesive elements was assumed to be equal to 5300
GN/m. A shear modulus was estimated from this assuming isotropic behaviour and a
Poisson’s ratio of 0.5. The onset of damage therefore took place in tension at displacements
of 0.002 mm for the 10 m/s case, representing a strain of 0.2 considering the element
thickness. Typical ultimate failure displacements were instead in the range of 0.35 mm to 0.7
mm. As no direct measurements were available for these quantities, a sensitivity analysis was
performed. The assumed value was considered to be the minimum for the stiffness. The
model was also analysed with stiffness values of 105 GN/m and 106 GN/m. No higher value
152
was employed due to convergence issues of the FEA model. The results obtained with higher
k values were then compared with those obtained with the original stiffness.
Table 6-2: Summary of the material models employed in FEA models.
Material Model type Summary of parameters
Glass Linear elastic E = 70 GPa, ρ = 2500 kg/m3, υ = 0.33
PVB Viscoelastic – Prony series with a See Table 5-3 and Table 5-4 for fit
reduced polynomial spring parameters
Cohesive Traction separation k = 53 MN/m

elements
A quadratic stress traction separation failure condition was applied to govern the degradation
of the bond. The peak stress ( S ) was estimated from information provided by Rahul-Kumar
et al. [28]. In his work the bond capacity between PVB and glass was measured using
compressive shear tests at different strain rates. The peak stresses from his tests were
employed to determine the failure stress parameter. Firstly, his results were extracted and
plotted from data available in the paper. These data, plotted with a logarithmic horizontal axis,
are shown in Figure 6-9. An exponential law was then fitted to the available points and used
to extrapolate the values to higher rates relevant to those used in this study. The fit to the data
is given by:
S = 0.782 ln(εɺ ) + 8.22 Eq. 6-2
where S is the peak stress and εɺ is the strain rate.
153
11
10
9
Peak Stress (MPa)
5
Experimental Data
4 Logarithm Fit
3
2 -3 -2 -1 0 1
10 10 10 10 10
Strain Rate (/s)
Figure 6-9: Peak stress against the strain rate from Jagota [77].
As extracting strain rate information directly from the experimental results proved unfeasible,
an approximate value was determined by dividing the overall displacement speed by the
sample length. A sensitivity analysis was then performed to check the effect of the peak stress.
Several 10 m/s speed models were run with the low and high rate material properties. In each
case, the peak stress was varied by 25% both above and below the value found with Rahul-
Kumar et al.’s data. The total cohesive energy was firstly kept constant when changing the
peak stress. Where differences in the plateau results were observed, a further model was run
refitting the delamination energy. The difference in the energies could then be assessed. The
final cohesive peak stresses used are shown in Table 6-3 for the three rates modelled.
Table 6-3: Peak stresses used in the models.
Speed (m/s) Peak stress (MPa)
0.1 7.68
1 9.48
10 11.28
154
The last parameter of the cohesive zone modelling, representing the final displacement at
failure, was used to fit the experimental data.
A symmetry boundary condition was set on one vertical side. A pinned condition was set at
the bottom PVB surface, whilst a velocity condition was set on the top surface. The location
of these is shown in Figure 6-10.
The 0.1 m/s model was run with artificially increased mass for some elements. This was done
to increase the stable analysis increment to a level that avoided using an excessive number of
steps to cover the required analysis time. A minimum step time of 2.5 x 10-7 s was enforced
through an ABAQUS option. As this can affect the quality of the model results a model was
run without the condition for comparison. The full analysis time could not be achieved
without this modification, as the maximum advised number of steps, two millions [76], would
have been reached. A 0.08 s period, instead of the full 0.5 s period, was run and the results up
to that point were compared.
Figure 6-10: Boundary conditions applied in the model.
155
The models were run to fit the experimental data for the 0.1, 1 and 10 m/s tests. Therefore six
models were run in total, two at each speed. All the models were run for a total deformation
of 50 mm, therefore analysis times of 0.5 s, 0.05 s and 0.005 s were set respectively. Once the
stress plateau levels seen in the experimental results were obtained, the delamination energy
was calculated from the peak stress and failure displacement parameters used in the cohesive
formulation. The delamination speed was also extracted for comparison with experimental
results. Element failure data at four time steps were used for this.
The main steps of this analysis are summarised in Figure 6-11 below.
Calculate experimental delamination speed.
Create single crack FEA models.
Perform analysis to check influence of PVB end tabs.
Perform analysis to check influence of mass scaling.
Perform sensitivity analysis on influence of k .
Perform sensitivity analysis on influence of peak stress S.
Obtain delamination energies and speeds. Compare with experimental data.
Perform studies on the effect of crack distributions.
Figure 6-11: Steps undertaken to perform the delamination energy analysis.
156
6.2.3. Parametric studies
Crack spacing
The influence of crack density was considered an important parameter which might affect the
overall material behaviour. Its effect was therefore studied with several FEA models. Crack
spacings of 20 mm, 10 mm, and 5 mm were considered. The two lower crack densities had
already been the subject of laboratory tests, to which the numerical results were compared.
The models employed were very similar to those used for the delamination energy studies.
Separate glass and PVB parts were used for each section between cracks. These parts were
then connected to each other in the same manner described above in section 6.2.2. The same
mesh types and sizes were used. The only exception was the 5 mm crack spacing model,
where the mesh dimension was reduced to 0.5 x 0.5 mm along the main surfaces. However,
the through thickness dimension was kept constant at 1 mm to reduce the computational time.
The same material properties summarised in Table 6-2 were used for the main components,
applying the appropriate delamination failure constants determined in the previous study.
Specifically, as 1 m/s models were run, the constants pertaining to this speed were employed.
Figure 6-12 shows a typical model with 20 mm crack spacing.
157
Figure 6-12: Front view of a multiple cracks, 20 mm spacing model.
All the models were run at a displacement rate of 1 m/s, and both PVB material models were
used. Therefore, a total of six models were run for this section. Again, the plateau stresses
and delamination speeds were extracted for evaluation. For the 10 mm and 20 mm cases these
were also compared with experimental data. However, this was not available for the
remaining crack spacing of 5 mm.
Crack inclination
Models were also run to assess the effect of the inclination of the cracks on the material
properties. A 20 mm crack spacing were used for all cases. Three models were run with
different inclinations. In one model, one side of the cracks was shifted by 20 mm, producing
an angle of 18.43°. In a second model a shift of 40 mm was used, creating an angle of 33.69°.
A third model was run using a 4.76° angle; however, in this case the direction of the
inclination was alternated creating non parallel crack positions. Figure 6-13 shows the three
models with the cracks locations and main dimensions. The three models were run using the
same material properties as before, including both PVB models. Therefore six models were
run in total. The entire model width was included without symmetry conditions due to the
158
non-symmetric crack patterns. Elements of the same type as stated above were employed;
however, the size was changed to account for the inclination. 1 mm sides were used as a
general guideline, and the size was modified as needed by the mesh generation algorithm to
fit the shape of the part. All the models were run at 1 m/s.
Figure 6-13: Front view with dimensions of the three inclined cracks models.
The plateau stresses were extracted for comparison between the models.
Staggered cracks
The effect of cracks not being aligned on the two sides of the PVB was considered next. A 20
mm crack spacing was again used and two cases were considered, one where the crack lines
were displaced by 10 mm, and one where they were displaced by 5 mm. Figure 6-14 presents
the main model dimensions and components. The model with 10 mm misalignment is shown.
The only difference in the 5 mm misalignment case is the length of the top and bottom glass
parts.
To avoid numerical instabilities, the geometry of the parts of the model had to be modified
for this task. The PVB membrane was divided in two separate layers along the centre of the
thickness. The two halves were then constrained to each other with a tie condition. The PVB
parts on the two sides were staggered to follow the glass sections. This ensured that each
glass fragment was connected to just one PVB part. Additionally, to ensure that the failure
would occur at the staggered crack locations rather than at the top or bottom of the specimen,
159
the first and last glass fragments on each side were tied without a failure condition. The same
material properties and mesh sizes were used as described above. Again, the plateau stresses
were extracted for evaluation. The results of the previously described 20 mm spacing model
without staggering were also employed for comparison.
Figure 6-14: Main dimensions and components of the staggered cracks models.
6.3. Results
The results for the models described above will be considered here in turn. The delamination
speeds extracted from the experimental pictures will be described first, as these will be used
to assess the FEA results.
6.3.1. Experimental data analysis
The pictures available were analysed as described above. The automated algorithm was
employed for the 0.1 m/s experimental images to identify the delamination fronts and glass
edges in every picture. This allowed the progress of the phenomenon to be monitored
throughout the deformation to estimate its general behaviour.
160
0.035
0.03
Delaminated distance (m)
0.025
0.02
0.015
Experimental data
Speed fit
0.01
0.005
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time (s)
Figure 6-15: Delamination progress time history for 0.1 m/s test.
The delaminated distance data are shown in Figure 6-15, together with a plot produced using
the estimated overall speed. As can be seen, the speed varies moderately during the course of
the experiment. However, as the speed variations were small compared to the uncertainties
and would be difficult to quantify in cases where only a limited number of pictures were
analysed, it was considered that a constant value would represent an acceptable
approximation.
Whilst manually preparing the images for other experimental speeds, it was noticed that
frequently the exact location of the key positions was difficult to estimate exactly. Often an
area of roughly three pixels could have been used for each point. Therefore, an error of ±1
pixel for the position of each element was accounted for. In single cracks four such elements
were identified, two delamination fronts and two glass edges, giving a total error range of 4
pixels. In a multiple crack specimen this had to be multiplied by the total number of cracks,
five cracks in 20 mm spacing specimens and ten in 10 mm spacing specimens. Considering
this and the total delaminated distance to be measured, the overall error was estimated to be
5% in single crack and 20% in multiple cracks specimens. This was added to the results. The
data obtained are shown below in Table 6-4 with the estimated errors.
161
Table 6-4: Experimental delamination speeds for different crack configurations and
displacement speed.
Crack Configuration Speed Test (m/s) Delamination speed (m/s)
Single Crack 0.1 0.079±0.004
Single Crack 1 1.40±0.07
Single Crack 10 9.05±0.50
10 mm Spacing 1 0.80±0.16
20 mm Spacing 1 0.95±0.20
The results show that the delamination speed is related to the deformation speed. Generally it
seems to be lower than the test velocity, except for the 1 m/s test with a single crack where
the results indicate a higher speed. The data also show a decrease in the delamination speed
as the number of cracks increase, with the 10 mm spacing test producing the lowest
delamination speed for the 1 m/s set of experiments. However, it should be noted that the
multiple cracks results do overlap when the uncertainties are considered, hence this result
might not be significant. This information was used to compare the model results presented
below.
6.3.2. Single crack modelling
The models described above were run to fit the delamination energy parameters. As discussed,
the values of the cohesive elements stiffness, peak stress, the effects of mass scaling and the
presence of PVB end tabs were considered in order to measure their significance on the
model results.
162
Figure 6-16: Typical stress distribution in a single crack sample during deformation.
The stresses are given in Pa.
Figure 6-16 shows typical stress contour plot for a single crack specimen. The stress variation
in the glass parts between the laminated and delaminated areas is visible, as well as the
deformation of the detached PVB membrane. The delamination progresses from the top and
the bottom of the specimen as well as from the central crack due to the presence of the PVB
end tabs.
The effect of the PVB end tabs was considered first. Figure 6-17 shows a plot of the PVB
reaction stresses versus overall displacement for a 1 m/s model with high rate material
properties both with and without free PVB ends. The model without the end parts presented
greater oscillations in the early part of the deformation. However, the steady-state plateau
reached is very similar, giving confidence in the validity of the results of the models used.
163
60
50
Engineering Stress (MPa)
40
30
20
With PVB end parts
Without PVB end parts
10
0
0 5 10 15 20 25 30 35 40 45 50
Displacement (mm)
Figure 6-17: Comparison of the models with and without PVB ends.
The effects of mass scaling were considered next. The analysis was run on a 0.1 m/s model,
as this was the case when mass scaling was used to limit the total number of analysis steps.
The 80 ms model was run with the high strain rate PVB material properties. The increase in
mass was equal to a factor of 73, however the ratio between kinetic energy to internal energy
remained below 0.1%. This indicated a quasi static condition which minimised the
importance of the extra mass [78]. The stress-displacement results are shown in Figure 6-18
together with data from the same model run for the full 0.5 s period with mass scaling. It can
be seen that, whilst the initial peak is marginally lower without the scaling, the curves mostly
superimpose. Mass scaling was therefore used in these cases.
164
18
16
14
12
10
6
No Mass Scaling
4 Mass Scaling
2
0
0 5 10 15 20 25 30 35 40 45 50
Displacement (mm)
Figure 6-18: Stress-Displacement plot of the 0.1 m/s model with high rate material
properties results when run with and without mass scaling.
The effect of varying the initial stiffness ( k ) of the cohesive elements was considered next.
Three models were run, with k varying between 5.2 x 107 N/m and 1 x 109 N/m. The low rate
material properties and a test speed of 10 m/s were used. The results are plotted in Figure
6-19. Again, no significant difference between the plateau results can be seen, with the three
curves being in good agreement.
165
35
30
25
20
15
7
k = 5.3 x 10 N/m
10
8
k = 1 x 10 N/m
9
5 k = 1 x 10 N/m
0
0 5 10 15 20 25 30 35 40 45 50
Displacement (mm)
Figure 6-19: Stress-Displacement plot of the 10 m/s model results when run with
different values of cohesive elements stiffness k.
The peak delamination stress ( S ) was then varied by 25% above and below the assumed
value. The study was performed on both material model types to highlight potential
differences in their behaviour. In all cases the plateau levels did change by a relatively small
amount. As an example, the stresses of the low rate model changed by 2.7% when the peak
stress was increased by 25%. The effect of this on the estimated energy was calculated, and
the possible uncertainty estimated at 7%.
166
Table 6-5: Results of models where the delamination peak stress was varied by 25%.
The original cases are included for comparison.
Material Peak Stress Modelling plateau Experimental plateau

Energy (J)
model (MPa) stress (MPa) stress (MPa)
Low rate 11.28 * 1630 26.3 26.3
Low rate 14.1 1630 27.0 26.3
Low rate 14.1 1550 26.3 26.3
Low rate 9.02 1630 25.6 26.3
Low rate 9.02 1740 26.3 26.3
High Rate 11.28 * 2765 26.3 26.3
High Rate 14.1 2765 26.7 26.3
High Rate 14.1 2670 26.3 26.3
High Rate 9.02 2765 26.0 26.3
High Rate 9.02 2890 26.3 26.3
* The original cases are included for comparison
Having completed this parametric study, the energy parameters were then fitted to the
experimental data. Figure 6-20 shows a typical graph of results, in this case for the 0.1 m/s
model with low rate material properties.
167
Engineering Stress (MPa) 15
10
0
0 5 10 15 20 25 30 35 40 45 50
Displacement (mm)
Figure 6-20: Stress-displacement results of the model run at 0.1 m/s with low rate
material properties.
Figure 6-21: Plots of the stress results for a 1 m/s model at several time points.
168
The delamination front could be seen moving along the PVB layer as the analysis progressed.
Figure 6-21 shows the stress distribution in the glass and PVB through a selection of frames
obtained from the model. The stress distribution in the glass parts indicates the position of the
delamination fronts, with high level of stress indicative of an area where the bond is still
effective. The detail of a delamination location is shown in Figure 6-22. This presents an
elevation of a glass fragment, indicating the position of the failure fronts. The PVB stresses
increase rapidly beyond the failures, as the membrane has to resist the loading in the detached
area.
Figure 6-22: A detail of the side elevation of the model taken at 25 ms.
169
The results of the energy fit are shown in Table 6-6 below. The delamination energies varied
between 3230 J and 1170 J depending on the deformation speed and the PVB material
properties used. It is significantly that the delamination speeds are often different from the
experimentally measured ones. The largest difference is shown by the 0.1 m/s model with
high rate material properties, which has a 390% error. In general the 0.1 m/s case shows the
worst agreement, with a 120% error in the low rate properties case as well. The 1 m/s model
has the lowest errors, especially for the low rate material properties. In most cases the speed
of the model exceeds the experimental one, signifying an excessive stiffness in the material
models employed. Possible explanations for these results will be discussed in section 6.4.
Table 6-6: Summary of energy and delamination speed results for single crack models.
Model
Model Experimental
Deformation Material Plateau Energy %
Delamination Delamination
Speed (m/s) Model Stress (J) Difference
Speed (m/s) Speed (m/s)
(MPa)
0.1 Low rate 13.2 2690 0.17 0.079 120
High
0.1 13.2 1170 0.39 0.079 390
Rate
1 Low rate 21.3 3225 1.12 1.43 28
High
1 21.3 2890 2.24 1.43 57
Rate
10 Low rate 26.3 1630 14.67 9.05 62
High
10 26.3 2765 18.13 9.05 100
Rate
The energies found here were used in the following parametric studies. As it proved difficult
to choose the correct material model for the 1 m/s case, both were used for subsequent
analyses.
170
6.3.3. Effect of crack spacing study
The models with different crack spacings were run as described, utilising the delamination
parameters obtained with the single crack models. All the models were run at 1 m/s speed,
using both the low rate and high rate PVB material properties. In all cases the delamination
spread both above and below each crack location and at the top and bottom edges of the
model. Figure 6-23 shows the stress distribution in the glass and PVB of a 20 mm crack
spacing model at five time steps. In the image, the stress variations near the cracks at the top
and bottom edges of the glass pieces highlight the presence of delamination fronts.
It can be seen that the area of glass subject to stress decreases as the delaminated areas spread
away from the initial positions.
Figure 6-23: Plots of the stresses of a 20mm crack spacing model at several time points.
Both the peak stresses and the delamination speeds were extracted from the models. The
results are shown in Table 6-7 below, together with a comparison with the experimental data
where this is available. The plateau stresses show a smaller error when compared with the
laboratory data. The largest difference is 30%, with the smallest being 10%. This indicates
that the models can predict this parameter within acceptable error. The delamination speeds
show similar differences when compared with experimental results to those highlighted
above for the single crack cases.
171
Model Experimental Model Experimental
Crack Material % %
Plateau Plateau Stress Delamination Delamination
Spacing (mm) Model Difference Difference
Stress (MPa) (MPa) Speed (m/s) Speed (m/s)
20 Low rate 18.5 16.5 15 1.2 0.95±0.2 30
High
20 21.1 16.5 30 1.7 0.95±0.2 80
Rate
10 Low rate 16.1 17.5 10 1 0.80±0.16 25
High
10 21.7 17.5 25 1.1 0.80±0.16 40
are included where relevant for comparison.
Rate
172
5 Low rate 15.4 N/A N/A 0.95 N/A N/A
High
5 21.2 N/A N/A 1.1 N/A N/A
Rate
The experimental results are included where relevant for comparison.

Table 6-7: Results of the models with different crack spacings. The experimental results
The effect of crack density on the stress results was small, with the biggest difference being
20% between the models using the low rate material model. This highlights the relative
insensitivity of the phenomenon to the density of initial failures.
6.3.4. Inclined Cracks study
The effects of crack inclination were considered next. The models described above were run
using the parameters found in previous parts of the study. Figure 6-24 shows stress
distributions at different time steps in one of the models. Again, the spread of the
delamination fronts can be deduced from the stress distribution in the glass, with the
variations near the cracks and at the top and bottom edges indicating the presence of
delamination. The failures seem to spread away from the cracks, with the failure front mostly
parallel to the crack direction.
Figure 6-24: Stress plots at several time points during the analysis of a 33.7° inclination
model.
The plateau stress results were collected from each of the models and are presented in Table
6-8. Again, their variation is small, with a maximum difference of 5% for the low rate models.
Additionally, the model plateau stresses are similar to those observed in the cases with no
crack inclination, supporting the view that inclination does not have a major effect on the
cracked material behaviour.
173
Table 6-8: Results of the models with different crack inclination.
Material Model Plateau

Crack Configuration
Model Stress (MPa)
20 mm spacing, inclined at
Low rate 16.7
18.4°
High Rate 20.9
18.4°
Low rate 17.6
33.7°
High Rate 20.9
33.7°
20 mm spacing, inclined at 4.8°,

Low rate 16.4
alternating direction
20 mm spacing, inclined at 4.8°,

High Rate 21.5
alternating direction
6.3.5. Staggered crack study
The two “staggered cracks” models were then considered, similarly to the cases described
above. Three dimensional views of the stress distribution are shown in Figure 6-25. In this
case the position of delamination fronts is less clear when considering only the glass stresses,
probably due to the more complex situation created by the asymmetric geometry. However,
the PVB deformed shape shows that two failure fronts move through the sample, starting
from the top and the bottom of the model.
174
Figure 6-25: Plots of the stress results of a 10 mm stagger model at several time points.
This can be also seen in Figure 6-26, where a detail of the model is shown. In these frames,
the delamination progresses downward, crossing a crack location in its travel. This behaviour
can be related to the overall stress data. Figure 6-27 shows the stress displacement
information for one of the 10 mm misalignment models. The stress reaches a peak at a
displacement of about 15 mm, before dropping to a plateau level which is maintained until
the end of the simulation. This indicates that a higher level of work needs to be done on the
material before a stable delamination process is initiated. However, after this point, the fronts
will move through the sample in a manner similar to that observed for the aligned cracks
models.
175
Figure 6-26: Section view of stress contours taken at different time points highlighting
the movement of the delamination front.
45
40
35
30
25
20
15
10
0
0 50 100 150 200 250 300 350 400 450 500
Displacement (mm)
Figure 6-27: Stress-displacement plot for the FEA model with cracks 10mm out of
alignment.
176
Table 6-9: Results of models with not aligned cracks on the two laminate faces.
Crack Configuration Material Model Model Plateau Stress (MPa)
20 mm spacing, 10 mm out
Low rate 23.5
of phase
20 mm spacing, 10 mm out
High Rate 22.0
of phase
20 mm spacing, 5 mm out of
Low rate 18.6
phase
20 mm spacing, 5 mm out of
High Rate 21.6
phase
The plateau stress information for all the models are shown in Table 6-9. The high rate PVB
models again show very similar data, which is also analogous to that seen in previous cases.
The low rate model with 5 mm stagger also conforms to this observation. The 10 mm
misalignment, low rate PVB model instead shows a higher plateau stress, similar to that
observed for a single crack.
6.4. Discussion
The model runs presented above underline several aspects of both the delamination energies
fitting process and of the behaviour of the cracked composite materials with different failure
configurations. These will be considered in turn.
6.4.1. Delamination energies
Two values of delamination energy were obtained for each deformation speed, producing
energies ranging from 1170 J and 3225 J. The exact magnitudes proved to be very dependent
on the material model utilised. This could be expected, as the strain energy in the material is
one of the variables deducted from the total external work to determine the delamination
177
energy. The effect shown here though is large, with the two energies being often different by
a factor of two. It is therefore very important to use a realistic material model for all these
studies. This is complicated in the situations which were considered here as the strain rates
were consistently high and varied significantly between the different cases. This implied that
a simple quasi static approximation would not be appropriate and the more complex models
derived in the previous chapter had to be used. The delamination speeds could provide a
measure of the accuracy of the material models used in these cases. When these were
compared to the experimental data, large difference could be seen between the different
options. The worst fit was obtained in the 0.1 m/s case, with both models showing too high
delamination progression. A stiffer model will produce higher delamination speed, as more
PVB will have to be freed to stretch to allow for the imposed overall movement. Therefore it
can be argued that for the slowest test both material models proved too rigid. This can also be
related to the material model fit results shown in the Chapter 5, where the lower rate model
showed a higher stiffness than seen in the experimental curves.
Figure 6-28: PVB strain rate in single crack models.
178
Additionally, the strain rate will vary significantly along the length of the delaminating
sample. Figure 6-28 shows a plot of the strain rate distribution in three single crack models
run at different speeds. Whilst the rate in the 0.1 m/s model peaks at 7.5 /s and was therefore
in the range used for the derivation of the low rate model, in the other models these maxima
are significantly higher, with rates of 42 /s for 1 m/s deformation speed and 280 /s for 10 m/s.
Additionally, the deformation speed drops considerably away from the failure front. This is
especially important in the 1 m/s case, where the rate is below 3 /s in delaminated areas. This
implies that, ideally, a single material model will be required to model both ranges of rates to
obtain better results. Otherwise, as seen in this study, it will be challenging to model the
phenomenon accurately.
The peak rates observed lend credibility to the peak delamination stresses used. Whilst they
are significantly higher than those used to estimate the employed stresses, alternative values
derived using these higher rates are within the 25% variations considered in the sensitivity
analysis. For example, for the 1 m/s case the estimated rate was 6.66 /s, whilst the measured
rate was 42 /s. These would produce peak stresses of 9.48 MPa and 11.14 MPa respectively, a
difference of 18%. Additionally, the uncertainties present due to the difference between the
material models are much larger than those seen with the peak stresses variations,
diminishing the importance of this parameter at this stage.
Given these uncertainties, it is difficult to assess the trend of the energy in relation with the
deformation speed. The data though does seem to show that its order of magnitude will not
change in the range considered here.
Whilst the values obtained cannot be trusted absolutely, it was considered that this did not
preclude performing the subsequent parametric study on the effects of different crack
configurations, as the data produced would still be useful for comparison purposes.
6.4.2. Parametric studies
The results for different crack configurations have been described and compared to the
available experimental data above. These were all run at the same deformation speed of 1 m/s
to allow for easy comparison of the results.
179
Figure 6-29: PVB strain rate in two multiple crack models.
In the cases with 10 mm and 20 mm horizontal cracks spacings, the delamination speeds were
compared to experimental data. It was evident that neither material model captures the full
behaviour of the material. Again, the need for a unique material model covering both ranges
was highlighted, as shown by the strain rate data shown in Figure 6-29. The plateau levels
obtained through modelling were much more similar to the experimental data available than
the delamination speeds, with the largest error being 30%. The plateau levels shown by the
various models were also similar, with only a few exceptions. Figure 6-30 shows a plot of the
plateau stress levels for different crack densities (calculated in terms of crack / mm) and
Figure 6-31 shows a plot of the same values, this time compared to the crack angles. In the
case of the high strain rate models the data appear to lie around a line of constant stress. In
the case of lower rate models there is a higher variation, especially between the single crack
specimen and the multiple crack specimens. However, when the multiple crack cases are
compared the differences were not high, with a maximum of 20%, and the plot suggests a
trend of diminishing variations should the density be increased further. These results are in
agreement with the observations by Hooper [3], as his results showed only minor differences
180
between the different crack spacings and also between samples with regularly spaced cracks
and a randomly cracked samples.
25
20
Plateau Stress (MPa)
15
10
Low Rate PVB Model

High Rate PVB Model
5 Experimental Data
0
0 0.05 0.1 0.15 0.2 0.25
Crack Concentration (crack/mm)
Figure 6-30: Plot of the plateau stresses versus the crack concentration.
25
20
15
10
Low Rate PVB Model

5 High Rate PVB Model
0
-5 0 5 10 15 20 25 30 35
Crack Inclination (°)
Figure 6-31: Plot of the plateau stresses versus the inclination of the cracks.
The “staggered cracks” cases also show similar plateau stresses, except for the lower rate
model, which shows a discrepancy for the 10 mm misalignment case. This is shown in Figure
181
6-32. However, as could be seen in Figure 6-27 above, a higher peak stress was reached
before the forces stabilise. This indicates that where there are no areas with aligned cracks, a
higher force is required to initiate delamination. Once the process is started though, the
failure travels through the sample progressively, without changing its behaviour significantly
from previous cases. This is also shown in Figure 6-26, where the front is seen travelling
downwards through a glass crack location instead of propagating in both directions from
there.
25
20
15
10
Low Rate PVB Model

5 High Rate PVB Model
0
0 5 10 15
Crack Misallignment (mm)
Figure 6-32: Plot of the plateau stresses versus the amount of crack misalignment.
This can also explain the difference in the results of the lower rate PVB model. In the 5 mm
stagger configuration, whilst a higher activation stress is required, several delamination fronts
initiate in the small areas between cracks, as can be seen in Figure 6-33. After the
delamination begins, the system behaves similarly to the multiple cracks specimens. In
contrast, with 10 mm misalignment delamination fronts originate at only two locations, as
could be seen in Figure 6-25. This causes the model to behave similarly to the single crack
specimens. This did not affect the plateau stress of the higher rate PVB models, as little
change was caused by the crack density parameter. However, the lower rate PVB model is
affected more significantly, causing the jump in plateau stresses.
182
Figure 6-33: Model detail showing stresses in the glass and PVB for a low rate PVB
material model with 5 mm crack misalignment.
These are important considerations to decide the relevance of this case to engineering design.
The results imply that, unless only a small minority of cracks are aligned and the crack
spacing is quite large, the delamination will simply originate at the aligned locations and
travel through the rest of the glass without changing the overall behaviour significantly.
Whilst it is inevitable that, under blast, many cracks in the window will not be aligned, many
of them will be. Furthermore, the crack density is likely to be higher than the one used here,
causing the absolute amount of misalignment between failure lines to be small, therefore
making this parameter unlikely to affect the forces of the system.
6.5. Conclusion
The available experimental data was utilised to obtain values for the delamination energies
and to investigate the effect of different crack patterns which are likely to occur under blast
183
loading. The effects of different materials properties were considered, and their influence
estimated using a second experimental measure, the speed of delamination growth.
The results showed that the magnitudes of the energies of delamination are likely to be in the
order a few thousand Joules for the specimens tested, with recorded values between 1170 J
and 3225 J. Additionally, their variation in the speed range of interest is unlikely to exceed
this level of difference.
However, especially in the case of slower deformation speeds, neither of the material models
managed to capture the entire process accurately, therefore reducing the reliability of the
energy values obtained. In cases where the results were in closer agreement, as in the 1 m/s
tests, it was still challenging to decide which of the two models represented the phenomenon
more accurately. Therefore, for further use in this study, both models were used for each case
analysed.
Whilst the imprecisions in the material models used, especially for the low rate model,
contributed to this error, the strain rate plots presented above in paragraphs 6.4.1 and 6.4.2
also highlight the great variability of this parameter in the areas of interest. Therefore, a
single material model to represent both strain rate ranges is needed to obtain more accurate
energy measurements. One such model was derived in Chapter 5, not only covering the full
range of required speeds, but also achieving a lower level of errors at low rates.
Consequently, including this material law in the FEA software and using it to repeat the
energy fitting would constitute an important follow up to this work and would be very likely
to produce significantly more accurate results.
The two energy parameters obtained for the 1 m/s case were used for the parametric study on
crack configurations. The high rate results for all cases show very constant results, indicating
a lack of sensitivity of the plateau stresses to all the variables considered. The low rate PVB
model results showed a larger variation, with the plateau stress decreasing as the crack
concentration increased. This variation was still only 20% when comparing realistic crack
spacings of 20 mm and 5 mm. The results were much closer to each other when considering
inclined cracks. One of the staggered crack results, with a 10 mm misalignment, presents the
only significant change, bringing the results in line with those of the single crack specimens,
as explained in section 6.4.2.
184
These results might indicate that a softer material model will be more influenced by the rate
changes which are caused by having a different number of cracks and hence delamination
zones. This effect though decreased as the crack spacing approaches 5 mm, and it is likely to
be less important with further reductions of this distance. This is also borne out by Hooper’s
tests [3], where only small differences where observed between the different crack spacings,
even when the results obtained with a randomly finely cracked specimen were taken into
consideration.
The results suggest that the crack configuration is unlikely to have an important effect on the
overall composite behaviour, especially with the small crack spacings which are typically
observed in tested panels. Designers therefore should be able to rely on standard results in
this respect. However, the delamination energies and plateau stresses are affected by the
speed of the overall deformation, which will be dependent on the exact blast loading. This
could explain the different reaction levels observed in Chapter 4. Once more precise
estimates are obtained by using an improved material model, the data could be used to
produce a more accurate overall smeared material property law similar to that used in Chapter
4, without the disadvantage of high noise levels being present in the forces and strains
recorded during blast tests.
185
7. Conclusion
In this project, the response of laminated glazing panels to blast loading was studied, paying
particular attention to the reaction forces which were applied to the supporting structure. The
variation of the loading in the panels’ geometry and its time history were analysed, as well as
the properties of the constituent elements of the cracked laminate and their relationship.
To achieve this, a blast test was performed and its results investigated to obtain new reaction
distribution information. The information was then combined with existing data to produce a
full reaction estimate, both in terms of its geometrical distribution and its evolution in time.
The properties of the cracked laminate were then studied. Firstly, the behaviour of the PVB
membrane across a range of strain rates was characterised. Then the delamination of the glass
fragments from the membrane was considered, producing both energy absorption estimates
and an assessment of the effect of different crack distributions.
7.1. Blast Test Results and Analysis

A blast test was performed on a laminated glass pane using a 100 kg TNT equivalent
explosive charge at a 23 m stand-off distance. Data were collected using optical techniques
(DIC), strain gauges and a pressure gauge. The information was then analysed to provide
information on both the general behaviour of the glass plate and the distribution of reaction
forces along its edge. A novel calibration technique was employed to extract the full field
displacement and strain information of the glass pane. This allowed for results to be obtained
from the experimental information in the absence of calibration images, albeit at the cost of
higher uncertainties.
The results obtained showed that the pane movement and strains presented a similar
behaviour to those observed in literature tests. Maximum deflections of 297 mm and
velocities of 34 m/s were recorded. Whilst the uncertainties in the strain measurements are
higher and prevent an accurate assessment of their magnitudes, their geometrical distribution
confirmed previous observations, with higher strains present near the perimeter of the
window.
Although several of the strain gauges used failed during the test, enough data was obtained to
reach some conclusions on the variation of reaction forces along two of the edges. The
186
recorded loads were in the range of 20 – 30 kN/m and did not increase further once this level
was reached. Whilst significant noise was present, the data also showed a variation of
loading along the longer glazing sides, with higher forces observed away from the central
position. On the shorter sides instead the forces seemed more constant.
The information obtained was then combined with historical blast test information to
calculate the overall reaction forces acting on the supports. These were found using
conjunctly both the strain gauge information and the DIC strain output. The post-cracked
reaction information from the gauges at the supports was used to calculate the stresses acting
on the membrane, and this was combined with the DIC strains to produce stress-strain curves
for the cracked material. A plastic material property was fit to these and applied throughout
the panel’s perimeter to calculate the post-cracked reactions.
A different method was applied for pre-cracked reactions. The DIC output was employed,
together with a FEA analysis, to estimate the bending moments acting in the window. Their
variation were then employed to calculate shear forces and hence reaction forces.
Both the overall time history of the reactions and their distribution along the window edges
were considered. The distribution confirmed the observations from the data of the blast test.
The forces varied in magnitude along the longer edges, presenting two separate peaks at some
distance from the centre.
The time history of the reaction showed an initial peak before the window shattered, dropping
close to zero immediately after. The magnitude of these first maxima ranged between 32 and
56 kN depending on the blast and window construction characteristics. Afterwards, the force
rose more gradually and approached a plateau. This ranged between 21 kN and 55 kN for out
-of-plane forces and between 74 kN and 91 kN for in-plane reactions.
As the forces did not increase for the latter part of the deformation, it was decided that a
mechanism other than simply increasing the strain energy had to contribute to the absorption
of the blast pressures. The delamination of glass fragments from the central membrane could
significantly contribute to the energy absorption and therefore was studied next.
187
7.2. PVB Characterisation
To be able to assess the delamination phenomenon discussed above, it was considered
important to use a realistic material property law for the PVB membrane. This needed to be
able to represent the behaviour of this layer at the strain rates of interest, which were likely to
be significantly higher than the quasi static level.
To achieve this, existing experimental data from uniaxial tensile tests at different strain rates
were employed to fit viscoelastic material property laws. Prony series with different
hyperelastic springs were considered, and two main models were fitted. One of these
employed a reduced polynomial hyperelastic formulation, whilst the other employed a spring
function proposed by Hoo Fatt and Ouyang [63]. Each of these was split into low rate (below
and including 8 s-1) and high rate (from 20 s-1) models, separately covering a different range
of strain rates. The results for both spring types showed a closer agreement to experimental
data for the higher rate cases, whilst the stiffness tended to be overestimated for lower rate
data.
Afterwards, a single material law derived with non linear finite deformation viscoelasticity
was employed. Two different non linear springs were combined in parallel, with a non linear
damper located in series with one of the springs. This method produced a single material
model which could account for the observed behaviour at all rates. Additionally, the errors at
low strain rates were significantly smaller for this approach than those seen in the Prony
series models. However, these improvements came at a cost of an increase in the model’s
mathematical complexity and a greater number of fitting constants if compared to each
individual Prony series model. Additionally, this model is not included as a matter of course
in standard FEA software, requiring therefore the development of user material routines for
its use. It was decided that for the following delamination study the reduced polynomial
models would be used for its ease of application to the FEA solver used, ABAQUS.
7.3. Delamination Modelling

The data from existing laboratory tensile tests of glass specimens with pre cracks were
available for this analysis. The experiments consisting in single crack specimens loaded at
different speeds were used for the first phase of this study. The geometry of the sample and
loading conditions were reproduced using FEA models in ABAQUS. These were then
188
employed to estimate the delamination energy parameters using both the low rate and the
high rate material properties laws derived previously. To achieve this, one of the constants
determining the cohesive zone model (CZM) parameters were varied to match the FEA stress
plateau levels to the recorded ones. The absorbed energy was then calculated finding the area
underneath the CZM stress-deformation graph. The results showed that the exact material
definition employed affected the delamination energies significantly. Therefore a second
parameter, the speed of delamination, was used to assess which PVB material model
produced the better representation of the real system for each deformation speed considered.
Whilst this showed that one of them, the low strain rate model, produced a better
approximation of the experimental results, neither captured the behaviour exactly. This is
likely due to both the models’ intrinsic limitations, for example the excessive stiffness of the
low rate law mentioned above, and by the complexity of the PVB strain rate distribution
throughout the samples. However, all the estimated delamination energies varied between
1000 J and 3500 J, giving confidence that the final values would be of a similar magnitude.
The parameters found in this manner were then used to study the effect of different crack
distributions in the samples. Several densities, inclinations and misalignments were
considered to assess their impact on plateau delamination stresses. Two experimental cases
were available for comparison, with 10 mm and 20 mm distance between the cracks.
When considering the crack densities and inclinations, the results indicated that these factors
only have a limited impact on the final stresses. Whilst, again, the comparison with the
available experimental results showed significant differences in the delamination speed, the
observed stress plateaux all showed an error of 30% or less. For the low strain models this
was limited further to a maximum 15% difference. These also showed the largest difference
between the crack patterns considered, which was though contained to within 20% of the
absolute plateau levels for realistic crack concentrations. The high rate PVB models instead
showed more constant reactions throughout, with differences of less than 1 MPa on absolute
magnitudes of between 21 and 22 MPa.
The possibility of crack misalignments was also considered. This more complex situation
caused a larger deviation from previously observed plateaux levels. This was again more
evident in the low strain rate models. The difference could be explained by the different
delamination propagation conditions created in the cases studied. The largest misalignment
189
considered caused the delamination to spread in a manner similar to that seen with single
cracks, hence causing the low rate PVB stresses to be higher.
Whilst the exact results obtained cannot be considered definitive due to the issues with PVB
material property laws highlighted above, the models indicate that in general realistic crack
distributions on their own will not affect the delamination stresses importantly. Additionally,
the presence of plateau stresses is strong evidence for this to be one of the main causes of the
time history behaviour of the reactions observed in the blast test data.
The lack of importance of cracks distribution highlighted here leaves the differences in the
strain rates as the main element to differentiate between the reaction forces seen in separate
blasts. In this respect the results presented, if refined using an improved material law, could
be used to produce more accurate smeared material properties for the composite material
after the glass failure. This would provide a powerful tool for practical design of façade
elements.
7.4. Further Work

The results obtained with this research could be expanded in several ways with further
studies. Some further work ideas will be discussed here.
7.4.1. Blast testing
In Chapter 3 it was attempted to analyse the variation of reaction forces along the glazing’s
edges through experimental data. As discussed, a significant proportion of the strain gauges
used for this failed. Therefore the same study could be repeated to collect more information.
In the same test, a pre crack introduced into the sample accidentally did not seem to affect the
results. Further tests could be performed to explore whether this was due to specific factors
connected with the test or whether it is a common feature of the panes’ behaviour. There is
some uncertainty on whether the pre-cracked phase contributes significantly to the overall
response of the material, as the energy absorbed in this initial situation is relatively small
when compared with the total amount. A test with a pre cracked specimen would therefore
yield valuable data in this respect.
190
In the same context, the opposite situation could be also tested by employing glazing panels
composed of significantly thicker glass plies or of an increased number of layers. In this case
the maximum pre-cracked forces would increase, possibly increasing the contribution of this
portion of the behaviour.
7.4.2. Bending capacity of cracked laminated glass
The blasts tests with more glass layers would also introduce new complexities in the form of
a possible residual bending resistance of the cracked material. This effect could be studied
and quantified through laboratory experiments. Some unpublished preliminary laboratory
tests were performed to measure its contribution on thin laminates with two glass layers. The
results showed some potential minor contributions from bending at low deflections. It is
likely that additional layers would be constrained by PVB membranes to a much higher
extent, intensifying this effect and potentially making it an important factor in the behaviour
of the glazing panes. Both low and high rate tests could be performed, attempting to quantify
this contribution and helping to explain potential blast test measurements. DIC techniques
could be employed to obtain strain data throughout the sample thickness, providing definitive
evidence of the importance of the various mechanisms. The implementation of this technique
could though be complicated by the cracks, which would introduce areas of low quality in the
speckle pattern.
7.4.3. PVB material properties
In the present study, the effect of strain rates on the PVB membranes was analysed. A finite
viscoelasticity material law was derived which provided predictions of the behaviour of the
material in the cases considered. This material model could be included into FEA software
for further use in simulations of more complex cases.
The effect of temperature variations was not considered at this stage. It is likely that, in real
usage, the glazing elements will be at a range of temperatures when hit by a blast pressure
wave. Considering normal ambient conditions, realistic temperatures in the UK could vary
between 0ºC and 30ºC. Additionally, it is possible that the blast itself will cause a raise in the
temperature. This could take place either through the thermal radiation emitted or through
191
indirect heating due to the high strain energies imparted by the pressure waves. As the glass
transition temperature of the membrane is approximately 20ºC [3], the difference in response
based on this parameter could prove very significant. Laboratory tests at a range of
temperature and strain rates could be performed. The results could then be used to expand the
material models derived here.
7.4.4. Glass – PVB delamination
Once the available PVB material law is implemented in a FEA software, the delamination
study presented here could be repeated to obtain more precise values of delamination
energies. This would avoid the need to choose an appropriate material representation by
estimating the strain rates likely to occur during the event of interest. Additional laboratory
tests could also be performed on some of the crack patterns considered numerically to further
verify the model.
Both these steps would increase the confidence in the modelling results of the delamination
energies and the plateau stresses for different crack patterns. The numerical data could then
be used to produce more complex FEA models of the whole window system. Additionally,
the collected information could be employed to determine smeared material parameters for
use in simple analyses run for practical design purposes.
7.4.5. Alternative interlayer materials
A laboratory and numerical analysis project similar to that performed on PVB could be
carried out on alternative interlayers materials. Whist PVB is the more common option for
blast resistance, project specific factors and requirements could lead designers to different
choices, such as the DuPont SentryGlass system [79]. Both the material properties and the
delamination energies of these alternative options could be assessed using similar methods to
those presented in this work.
7.4.6. Alternative holding mechanisms
The results of this study were based on results obtained on silicone bonded windows.
Alternative options could be considered. Glazing tapes, for example, are proposed by
192
manufacturers as an alternative to the use of silicone joints [51], [52]. Laboratory tests similar
to those done by Hooper on the silicone [3] could be carried out to test their capacities. DIC
could be employed as a data collection device, providing therefore at the same time a realistic
strain distribution throughout the deformation process. Blast tests could then be performed to
measure the overall pane behaviour in a realistic situation.
7.5. Conclusion
In this research several aspects of the behaviour of laminated glass when subjected to blast
loading have been examined. The data produced regarding the reactions forces, PVB material
models and delamination phenomena will help to improve the design of glazing to resist this
type of loads. It will be possible therefore to optimise the designs further than achieved in
current practice, producing safer and more cost effective solutions. The possibilities for
further studies highlighted in section 7.4 could lead to further improvements in the
understanding of these structural elements and hence would again help to produce improved
designs.
Additionally, several elements developed here could be incorporated in the design for
alternative loading cases. Impact loading of laminated glass is another area of significant
interest due to its applications in the automotive and aviation industries. Elements such as the
PVB high rate behaviour and delamination will be relevant to such applications and it is
hoped that the results obtained with this project will be helpful to practitioners in these areas
as well.
193
References
[1] D. Smith, “Glazing for Injury Alleviation Under Blast Loading – United Kingdom
Practice,” Proc. 7th international conf. on architectural and automotive glass Glass
Processing Days. pp. 335–340, 2001.
[2] CPNI, “Guidance note: Glazing enhancement to improve blast resistance,” Home
Office Scientific Development Branch, Report CPNI EBP 02/13, 2013.
[3] P. Hooper, “Blast Performance of Silicone-Bonded Laminated Glass,” PhD Thesis,

Department of Mechanical Engineering, Imperial College, London, 2011.
[4] J. M. Dewey, “The shape of the blast wave: studies of the Friedlander equation,” in
21st International Symposium on Military Aspects of Blast and Shock, Israel, 2010. p.
22
[5] B. Hopkinson, “British ordnance board minutes 13565.” London, 1915.
[6] K. J. Cranz, O. von Eberhard, K. E. Becker, and O. Poppenberg, Lehrbuch der

ballistik, Berlin: BG Tuebner,1910.
[7] T. L. Geers and K. S. Hunter, “An integrated wave-effects model for an underwater
explosion bubble,” Journal of the Acoustical Society of America, vol. 111, no. 4, pp.
1584–1601, 2002.
[8] C. N. Kingery and G. A. Coulter, “Reflected Overpressure Impulse on a Finite

Structure,” US Army Armament Research and Development Center, Ballistic Research
Laboratory, Technical Report ARBRL-TR-02537, 1983.
[9] C. N. Kingery and G. Bulmash, "Air blast parameters from TNT spherical air burst and
hemispherical surface burst," US Army Armament Research and Development Center,
Ballistic Research Laboratories, Technical Report ARBRL-TR-02555, 1984.
[10] D. W. Hyde, “User’s Guide for Microcomputer Programs ConWep and FunPro,
Applications of TM 5-855-1, “Fundamentals of Protective Design for Conventional
Weapons,”” US Army Engineer, Waterways Experiment Station, Report AD-A195
867, 1988.
194
[11] T. A. Rose, Air3d version 9 users’ guide. Swindon, 2006.
[12] Standard Practice for Determining Load Resistance of Glass in Buildings, ASTM
International, Standard ASTM E1300 - 12ae1, West Conshohocken, PA, 2012.
[13] Standard Practice for Specifying an Equivalent 3-Second Duration Design Loading for
Blast Resistant Glazing Fabricated with Laminated Glass, ASTM International,
Standard ASTM F2248 - 12, West Conshohocken, PA, 2012.
[14] C. V. G. Vallabhan, Y. C. Das, and M. Ramasamudra, "Properties of PVB Interlayer

Used in Laminated Glass," vol. 4, no. 1. ASCE, 1992, pp. 71–76.
[15] P. A. Du Bois, S. Kolling, and W. Fassnacht, “Modelling of safety glass for crash
simulation,” Computational Materials Science, vol. 28, no. 3–4, pp. 675–683, 2003.
[16] M. Timmel, S. Kolling, P. Osterrieder, and P. A. Du Bois, “A finite element model for
impact simulation with laminated glass,” International Journal of Impact Engineering,
vol. 34, no. 8, pp. 1465–1478, 2007.
[17] J. Xu and Y. Li, “Crack analysis in PVB laminated windshield impacted by pedestrian
head in traffic accident,” International Journal of Crashworthiness, vol. 14, no. 1, pp.
63–71, Feb. 2009.
[18] J. Xu, Y. Li, D. Ge, B. Liu, and M. Zhu, “Experimental investigation on constitutive
behavior of PVB under impact loading,” International Journal of Impact Engineering,
vol. 38, no. 2–3, pp. 106–114, Feb. 2011.
[19] R. Iwasaki and C. Sato, “The influence of strain rate on the interfacial fracture
toughness between PVB and laminated glass,” J. Phys. IV France, vol. 134, pp. 1153–
1158, 2006.
[20] R. Iwasaki, C. Sato, J. L. Latailladeand, and P. Viot, “Experimental study on the

interface fracture toughness of PVB (polyvinyl butyral)/glass at high strain rates,”
International Journal of Crashworthiness, vol. 12, no. 3, pp. 293–298, 2007.
[21] B. Liu, Y. Sun, Y. Li, Y. Wang, D. Ge, and J. Xu, “Systematic experimental study on
mechanical behavior of PVB (polyvinyl butyral) material under various loading
conditions,” Polymer Engineering & Science, vol. 52, no. 5, pp. 1137–1147, May
2012.
195
[22] M. Larcher, “Simulation of laminated glass loaded by air blast waves,” in DYMAT
2009 - 9th International Conferences on the Mechanical and Physical Behaviour of
Materials under Dynamic Loading, EDP Science, 2009, pp. 1553–1559.
[23] X. Zhang, H. Hao, and G. Ma, “Parametric study of laminated glass window response
to blast loads,” Engineering Structures, vol. 56, no. 0, pp. 1707–1717, Nov. 2013.
[24] A. A. Griffith, “VI. The Phenomena of Rupture and Flow in Solids.,” Phil. Trans. Roy.
Soc.(Lon.) A, vol. 221, pp. 163–198, 1920.
[25] Glass in buildings - determination of the strength of glass panes. Part 3: General
method of calculation and determination of strength of glass by testing, British
Standards Institute, Standard prEN 13474-3 London, 2008.
[26] J. Zhang, N. Kikuchi, V. Li, A. Yee, and G. Nusholtz, “Constitutive modeling of

polymeric foam material subjected to dynamic crash loading,” International Journal of
Impact Engineering, vol. 21, no. 5, pp. 369–386, 1998.
[27] T. J. Holmquist, G. R. Johnson, C. M. Lopatin, D. E. Grady, and E. S. J. Hertel, “High

strain rate properties and constitutive modeling of glass,” in Proceedings of the 15th
International symposium on ballistics, Jerusalem, Israel, 1995.
[28] P. Rahul-Kumar, A. Jagota, S. J. Bennison, and S. Saigal, “Interfacial failures in a

compressive shear strength test of glass/polymer laminates,” International Journal of
Solids and Structures, vol. 37, no. 48–50, pp. 7281–7305, 2000.
[29] S. Muralidhar, A. Jagota, S. J. Bennison, and S. Saigal, “Mechanical behaviour in

tension of cracked glass bridged by an elastomeric ligament,” Acta Materialia, vol. 48,
no. 18–19, pp. 4577–4588, Dec. 2000.
[30] C. Butchart and M. Overend, “Delamination in fractured laminated glass,” in

Proceedings of the 3rd international conference on Engineering transparency.
International Conference at Glasstec, Dusseldorf, 2012, pp. 249-257.
[31] H. Norville, K. King, and J. Swofford, “Behavior and Strength of Laminated Glass,”
Journal of Engineering Mechanics, vol. 124, no. 1, pp. 46–53, Jan. 1998.
[32] R. Behr, J. Minor, and H. Norville, “Structural Behavior of Architectural Laminated

Glass,” Journal of Structural Engineering, vol. 119, no. 1, pp. 202–222, Jan. 1993.
196
[33] L. Galuppi and G. Royer-Carfagni, “The design of laminated glass under time-
dependent loading,” International Journal of Mechanical Sciences, vol. 68, no. 0, pp.
67–75, Mar. 2013.
[34] S. J. Bennison, A. Jagota, and C. A. Smith, “Fracture of Glass/Poly(vinyl butyral)

(Butacite®) Laminates in Biaxial Flexure,” Journal of the American Ceramic Society,
vol. 82, no. 7, pp. 1761–1770, Jul. 1999.
[35] M. Seshadri, S. J. Bennison, A. Jagota, and S. Saigal, “Mechanical response of cracked

laminated plates,” Acta Materialia, vol. 50, no. 18, pp. 4477–4490, 2002.
[36] T. Pyttel, H. Liebertz, and J. Cai, “Failure criterion for laminated glass under impact
loading and its application in finite element simulation,” International Journal of
Impact Engineering, vol. 38, no. 4, pp. 252–263, 2011.
[37] P. A. Hooper, R. A. M. Sukhram, B. R. K. Blackman, and J. P. Dear, “On the blast

resistance of laminated glass,” International Journal of Solids and Structures, vol. 49,
no. 6, pp. 899–918, 2012.
[38] R. A. C. Stephens, “Determination of the resistance of laminated glass subjected to

blast loading using high speed video,” Master thesis, Cranfield University, 2008.
[39] P. Kumar and A. Shukla, “Dynamic response of glass panels subjected to shock
loading,” Journal of Non-Crystalline Solids, vol. 357, no. 24, pp. 3917–3923, Dec.
2011.
[40] J. Wei and L. R. Dharani, “Response of laminated architectural glazing subjected to

blast loading,” International Journal of Impact Engineering, vol. 32, no. 12, pp. 2032–
2047, 2006.
[41] J. Wei and L. R. Dharani, “Fracture mechanics of laminated glass subjected to blast
loading,” Theoretical and Applied Fracture Mechanics, vol. 44, no. 2, pp. 157–167,
2005.
[42] H. D. Hidallana-Gamage, D. P. Thambiratnam, and N. J. Perera, “Failure analysis of

laminated glass panels subjected to blast loads,” Engineering Failure Analysis, vol. 36,
no. 0, pp. 14–29, Jan. 2014.
197
[43] C. Amadio and C. Bedon, “Blast Analysis of Laminated Glass Curtain Walls Equipped
by Viscoelastic Dissipative Devices,” Buildings, vol. 2, no. 3, pp. 359–383, 2012.
[44] J. M. Biggs, Introduction to Structural Dynamics. USA: McGraw-hill Publishing

Company, 1964.
[45] K. Fischer and I. Häring, “SDOF response model parameters from dynamic blast
loading experiments,” Engineering Structures, vol. 31, no. 8, pp. 1677–1686, 2009.
[46] M. G. Stewart and M. D. Netherton, “Security risks and probabilistic risk assessment
of glazing subject to explosive blast loading,” Reliability Engineering & System Safety,
vol. 93, no. 4, pp. 627–638, Apr. 2008.
[47] M. D. Netherton and M. G. Stewart, “The effects of explosive blast load variability on
safety hazard and damage risks for monolithic window glazing,” International Journal
of Impact Engineering, vol. 36, no. 12, pp. 1346–1354, Dec. 2009.
[48] C. Morison, “The Resistance Of Laminated Glass To Blast Pressure Loading And The
Coefficients For Single Degree Of Freedom Analysis Of Laminated Glass,” PhD
thesis, Engineering Systems Department, Cranfield University, 2007.
[49] K. Ramesh, R. Tock, R. Narayan, and C. Vallabhan, “Property evaluation of silicone

elastomers used in tension-adhesion joints,” Journal of Materials Science Letters, vol.
14, no. 13, pp. 964–967, 1995.
[50] L. Meunier, G. Chagnon, D. Favier, L. Orgéas, and P. Vacher, “Mechanical

experimental characterisation and numerical modelling of an unfilled silicone rubber,”
Polymer Testing, vol. 27, no. 6, pp. 765–777, 2008.
[51] 3M, 3MTM VHBTM Structural Glazing Tape, 2014 [Online]. Available:
http://solutions.3m.com.au/wps/portal/3M/en_AU/CMD_Markets/Home/Product_Solu
tions/Window/Windows/Blast/?PC_Z7_RJH9U5230GG1F0ICJNBIJM08U6000000_a
ssetType=MMM_Article&PC_Z7_RJH9U5230GG1F0ICJNBIJM08U6000000_assetI
d=1258562044229&PC_Z7_RJH9U5230GG1F0ICJNBIJM08U6000000_univid=1258
562044229. [Accessed: 06-Mar-2014].
[52] Tremco-Illbruck, “Technical Data Sheet, Product: Alfas Security Plus (Glazing
Tape),” Tyne & Wear, 2003.
198
[53] B. W. Townsend, D. C. Ohanehi, D. A. Dillard, S. R. Austin, F. Salmon, and D. R.
Gagnon, “Developing a Simple Damage Model for the Long Term Durability of
Acrylic Foam Structural Glazing Tape Subject to Sustained Wind Loading,” Journal of
Architectural Engineering, vol. 18, no. 3, pp. 214–222, 2012.

Gagnon, “Characterizing Acrylic Foam Pressure Sensitive Adhesive Tapes for
Structural Glazing Applications - Part II: Creep Rupture Results,” International
Journal of Adhesion and Adhesives, vol. In Press, , 2011.

Gagnon, “Characterizing Acrylic Foam Pressure Sensitive Adhesive Tapes for
Structural Glazing Applications - Part I: DMA and Ramp-to-Fail Results,”
International Journal of Adhesion and Adhesives, vol. In Press, , 2011.
[56] M. Salisbury, I. U. Garmendia, and C. Worswick, “High Rate Deformation of VHB

Tapes,” University of Waterloo, Waterloo, 2007.
[57] H. Schreier, J.-J. Orteu, and M. A. Sutton, Image Correlation for Shape, Motion and
Deformation Measurements. USA: Springer US, 2009.
[58] T. A. Rose, “An Approach to the evaluation of blast loads on finite and semi-infinite
structures,” PhD thesis, Engineering Systems Department , Cranfield University,
Shrivenham, 2001.
[59] P. M. Deseilligny and I. Clery, “Apero, An Open Source Bundle Adjusment Software
For Automatic Calibration And Orientation Of Set Of Images,” in ISPRS Trento 2011
Workshop,, 2011, Volume XXXVIII–5/W16 269–276.
[60] R. S. Rivlin, “Large Elastic Deformations of Isotropic Materials. I. Fundamental

Concepts,” Philosophical Transactions of the Royal Society of London. Series A,
Mathematical and Physical Sciences, vol. 240, no. 822, pp. 459–490, 1948.
[61] G. R. Johnson and W. H. Cook, “A constitutive model and data for metals subjected to
large strains, high strain rates and high,” in Proceedings of the 7th International
Symposium on Ballistics, The Hague, The Netherlands, 1983, pp. 541–547.
199
[62] A. K. Dhaliwal and J. N. Hay, “The characterization of polyvinyl butyral by thermal
analysis,” Thermochimica Acta, vol. 391, no. 1–2, pp. 245–255, Aug. 2002.
[63] M. S. Hoo Fatt and X. Ouyang, “Three-dimensional constitutive equations for Styrene
Butadiene Rubber at high strain rates,” Mechanics of Materials, vol. 40, no. 1–2, pp.
1–16, Jan. 2008.
[64] N. Huber and C. Tsakmakis, “Finite deformation viscoelasticity laws,” Mechanics of

Materials, vol. 32, no. 1, pp. 1–18, 2000.
[65] P. A. Hooper, B. R. K. Blackman, and J. P. Dear, “The mechanical behaviour of

poly(vinyl butyral) at different strain magnitudes and strain rates,” Journal of
Materials Science, vol. 47, no. 8, pp. 3564–3576, 2012.
[66] Rubber, vulcanized or thermoplastic - Determination of tensile stress-strain

properties, British Standard Institute, Standard BS ISO 37-2005, London, 2005.
[67] E. M. Arruda and M. C. Boyce, “A three-dimensional constitutive model for the large
stretch behavior of rubber elastic materials,” Journal of the Mechanics and Physics of
Solids, vol. 41, no. 2, pp. 389–412, 1993.
[68] A. Amin, M. S. Alam, and Y. Okui, “An improved hyperelasticity relation in modeling
viscoelasticity response of natural and high damping rubbers in compression:
experiments, parameter identification and numerical verification,” Mechanics of
Materials, vol. 34, no. 2, pp. 75–95, 2002.
[69] A. Amin, A. Lion, S. Sekita, and Y. Okui, “Nonlinear dependence of viscosity in

modeling the rate-dependent response of natural and high damping rubbers in
compression and shear: Experimental identification and numerical verification,”
International Journal of Plasticity, vol. 22, no. 9, pp. 1610–1657, 2006.
[70] J. Lambert-Diani and C. Rey, “New phenomenological behavior laws for rubbers and
thermoplastic elastomers,” European Journal of Mechanics - A/Solids, vol. 18, no. 6,
pp. 1027–1043, Nov. 1999.
[71] S. M. Goh, M. N. Charalambides, and J. G. Williams, “Determination of the

Constitutive Constants of Non-Linear Viscoelastic Materials,” Mechanics of Time-
Dependent Materials, vol. 8, no. 3, pp. 255–268, 2004.
200
[72] J. Lubliner, Plasticity Theory. New York: Macmillan Publishing, 1980.
[73] R. W. Ogden, “Large Deformation Isotropic Elasticity - On the Correlation of Theory

and Experiment for Incompressible Rubberlike Solids,” Proceedings of the Royal
Society of London. A. Mathematical and Physical Sciences , vol. 326 , no. 1567 , pp.
565–584, Feb. 1972.
[74] MathWorks, MATLAB - Documentation, [Online]. Available:

http://www.mathworks.co.uk/help/matlab/index.html. [Accessed: 22-May-2014].
[75] N. Chandra, H. Li, C. Shet, and H. Ghonem, “Some issues in the application of
cohesive zone models for metal–ceramic interfaces,” International Journal of Solids
and Structures, vol. 39, no. 10, pp. 2827–2855, May 2002.
[76] Dassault Systèmes Simulia Corp., Abaqus analysis manual v6.10. Providence, RI,
2009.
[77] A. Jagota, S. J. Bennison, and C. A. Smith, “Analysis of a compressive shear test for
adhesion between elastomeric polymers and rigid substrates,” International Journal of
Fracture, vol. 104, no. 2, pp. 105–130, 2000.
[78] L. Aktay, A. K. Toksoy, and M. Güden, “Quasi-static axial crushing of extruded

polystyrene foam-filled thin-walled aluminum tubes: Experimental and numerical
analysis,” Materials & Design, vol. 27, no. 7, pp. 556–565, 2006.
[79] S. J. Bennison, M. H. X. Qin, and P. S. Davies, “High-performance laminated glass for

structurally efficient glazing,” Innovative light-weight structures and sustainable
facades, pp. 1–12, 2008.
[80] B. D. Coleman and M. E. Gurtin, “Equipresence and constitutive equations for rigid
heat conductors,” Zeitschrift für angewandte Mathematik und Physik ZAMP, vol. 18,
no. 2, pp. 199–208, 1967.
201
Appendix A – Finite Deformation
Viscoelasticity Law Derivation
Nomenclature
Bɺ e Elastic deformation strain rate
D Symmetric part of overall deformation velocity gradient
Di Symmetric part of inelastic deformation velocity gradient
I 3B Third invariant of the total Left Cauchy-Green strain tensor
I 3Be Third invariant of the elastic Left Cauchy-Green strain tensor
S Weighted Cauchy stress tensor
Se Deviatoric portion of weighted Cauchy total stress tensor
S eq
E Deviatoric portion of weighted Cauchy equilibrium spring stress tensor
S oe
E Deviatoric portion of weighted Cauchy overstress spring stress tensor
W Antisymmetric part of overall deformation velocity gradient
Wi Antisymmetric part of inelastic deformation velocity gradient
Derivation
The method developed by Amin [69] is used in the following derivation. The final result
though are equivalent to those obtain by other authors. As discussed above in Chapter 5, the
first step of the derivation of finite deformation viscoelasticity laws consists in
multiplicatively decompose the deformation matrix F in its inelastic and elastic parts, Fe and
Fi.
As mentioned above, the left Cauchy Green strain tensor B is given by:
202
B = FFT Eq. A-1
And therefore:
B e = Fe FeT Eq. A-2
And
B i = Fi Fi T Eq. A-3
ɺ −1 ) can be split in its symmetric and anti-

The velocity gradient defined in Chapter 5 ( L = FF
symmetric parts:
ɺ −1 = D + W
L = FF Eq. A-4
1 1
Where D =
2
( L + LT ) and W = ( L − LT ) . The elastic and inelastic velocity gradients and
2
their components are formulated in the same manner.
The strain time rate of the elastic deformation is given by:
Bɺ e = Fɺ e FeT + Fe Fɺ eT Eq. A-5
Where Fe can be expressed as Fe = FFi and therefore Fɺ e = d dt ( FFi ) = LFe − Fe Li .

−1 −1
Inserting this result in Eq. A-5 and with some manipulation, the inelastic velocity gradient is
defined by:
Bɺ e = −2Fe Di FeT + B e LT + LB e Eq. A-6
As the material is assumed to be incompressible, the determinant of the deformation matrix

and of its components is equal to 1. The weighted Cauchy stress is given by S = ( det F ) T
and therefore in this case is equal to the Cauchy stress T. the hydrostatic stress component is
included in the stress tensor, giving:
S = − pI + S E Eq. A-7
The SE component of the stress can then be split in the equilibrium and overstress
components, giving:
203
S E = S eq oe
E + SE Eq. A-8
The Clausius Duhem inequality [80] is a thermodynamic inequality which is commonly used
to define the stress components. It takes the form:
− ρ rψɺ + S E iL ≥ 0 Eq. A-9
The rate of energy portion of the equation, − ρ rψɺ , can be split additively in the equilibrium
and overstress portions. This will give:
ρ rψ = ψ eq ( I B , I 2 B ) +ψ oe ( I B , I 2B )
e e
Eq. A-10
And
∂ψ eq ɺ ∂ψ eq ɺ ∂ψ oe ɺ ∂ψ oe ɺ
ρ rψ = IB + I2B + I Be + I 2 Be Eq. A-11
∂I B ∂I 2 B ∂I Be ∂I 2 Be
Where IɺB = I iBɺ , Iɺ2 B = ( I B I − B )iB

ɺ and similar equations are used for Iɺ and Iɺ .
Be 2 Be
The strain tensors B and Be can be expressed in terms of their invariants. As an example, B is
given by B = I B I − I 2 B B −1 + I 3B B −2 . The equation for Be is similar. I 3B is the third invariant
of the strain tensor, and it is given by:
I 3B = det ( B ) Eq. A-12
Therefore, for an incompressible material, I 3B will be equal to 1.
Substituting these expressions in Eq. A-11, the following is obtained:
 ∂ψ eq ∂ψ eq 
ρ Rψɺ =  I+ ( I 2B B −1 + B −2 ) iBɺ +
 ∂I B ∂I 2 B 
Eq. A-13
 ∂ψ 
+  oe I +
∂ψ oe
(
I 2 Be B e −1 + B e −2 )iBɺ e
 ∂I Be ∂I 2 Be 
The second part of the inequality is considered next.
The velocity gradient L can be expressed as:
204
L = Fɺ e Fe−1 + Fe Fɺ i Fi−1Fe−1 = L e + Fe L i Fe−1 Eq. A-14
Using this, the stress power can be split to highlight the contributions of the elastic and
inelastic deformations:
E iL + S E iL e + ( Fe S E Fe )i L i
−1 oe
S E iL = S eq oe T
Eq. A-15
−1 eq
Considering that S eq eq
E is an isotropic function of B, giving B S E B = S E and with some
algebraic manipulation, the following result can be obtained:
1 −1 eq ɺ 1
S E iL =
2
( B S E )iB + ( B e −1S oe
2
E )i B e + ( Fe S E Fe )i Di
ɺ −1 oe
Eq. A-16
The results shown in Eq. A-13 and Eq. A-16 can be inserted in the original inequality, giving
 1 −1 eq ∂ψ eq ∂ψ eq  ɺ
 (B SE ) − I+ ( I 2B B−1 + B−2 ) iB + ( Fe−1S EoeFe )i Di
 2 ∂I B ∂I 2B 
Eq. A-17
 1 
+  ( B e −1S oe
E )−
∂ψ oe
∂I Be
I+
∂ψ oe
∂I 2 Be
(
I 2 B e B e −1 + B e −2 )iBɺ e ≥0
 2 
One way to satisfy this relationship is to set the terms in brackets multiplied by the strain rate
tensors to 0 and then to ensure the ( Fe−1S oe
E Fe )i Di term is greater than 0. S E and S E can then
eq oe
be determined:
∂ψ eq ∂ψ eq
S eEq = 2
∂I B
B+2
∂I 2 B
(I 2B I + B −1 ) Eq. A-18
∂ψ oe ∂ψ oe
S oe
E = 2
∂I Be
Be + 2
∂I 2 Be
(
I 2 B e I + B e −1 ) Eq. A-19
As the hydrostatic portion of the stress ( -pI ) is undetermined ad will need to be defined with
respect of the boundary conditions, the terms which are proportional to I can be included in
its function and eliminated from the equations above, giving:
205
∂ψ eq ∂ψ eq
S eEq = 2 B+2 B −1 Eq. A-20
∂I B ∂I 2 B
∂ψ oe ∂ψ oe −1
S oe
E = 2 Be + 2 Be Eq. A-21
∂I Be ∂I 2 Be
These functions, considering the equivalence of T and S for uncompressible materials, give
the solution presented in Eq. 5-16:
∂ψ eq ∂ψ eq ∂ψ oe ∂ψ
T = − pI + 2 B−2 B −1 + 2 Be − 2 oe Be −1 Eq. A-22
∂I1B ∂I 2 B ∂I1Be ∂I2 Be
To satisfy the second element of the inequality, it is sufficient to set Di to be positive. A

function which has been used for this is:
1
Fe Di FeT = B eS oe
E Eq. A-23
η
Where η is a function representing the viscosity and is always above 0. This solution can be
2
combined with the relationship given in Eq. A-6, giving Bɺ e = B e LT + LB e − B eS oe
E , and if
η
the relationship between T and S is applied again, the final formula used in Chapter 5 is
obtained:
2
Bɺ e = B e LT + LB e − B e Toe Eq. A-24
η
206

Del Linz-P-2014-PhD-Thesis

Uploaded by

Copyright:

Available Formats

Del Linz-P-2014-PhD-Thesis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Del Linz-P-2014-PhD-Thesis

Uploaded by

Copyright:

Available Formats

Blast Resistance of Laminated

Paolo Del Linz

A thesis submitted for the degree of Doctor of Philosophy of

Imperial College London and the Diploma of Imperial College

Declaration of Originality .......................................................................................................... 4

Copyright Declaration ................................................................................................................ 4

List of Figures .......................................................................................................................... 10

List of Tables ........................................................................................................................... 16

1.1. Project aims ............................................................................................................... 24

2. Literature Review ............................................................................................................. 25

2.1. Blast Pressure Theory................................................................................................ 25

2.1.1. Blast wave characteristic and interactions with structures ................................ 25

2.1.2. Loading estimate methods ................................................................................. 27

2.2. Behaviour of Laminated Glass .................................................................................. 28

2.2.1. PVB material model ........................................................................................... 28

2.2.2. Glass behaviour .................................................................................................. 29

2.2.3. PVB – glass delamination .................................................................................. 30

2.3. Overall response of laminated glass .......................................................................... 31

2.3.1. Low rate loading ................................................................................................ 31

2.3.2. Impact loading ................................................................................................... 32

2.3.3. Blast loading studies .......................................................................................... 32

2.4. Restraint mechanism ................................................................................................. 36

3. Full Scale Blast Test ......................................................................................................... 39

3.2. Method ...................................................................................................................... 39

3.2.1. Experimental set up and data collection ............................................................ 39

3.2.2. Data analysis ...................................................................................................... 48

3.3. Results ....................................................................................................................... 55

3.3.1. Blast Pressures ................................................................................................... 55

3.3.2. Window general observations ............................................................................ 58

3.3.3. Digital Image Correlation calibration comparison ............................................ 61

3.3.4. DIC Analysis Results ......................................................................................... 66

3.3.5. Reaction data ...................................................................................................... 73

3.4. Discussion ................................................................................................................. 78

4. Blast Reaction Forces ....................................................................................................... 82

4.1. Introduction ............................................................................................................... 82

4.2. Method ...................................................................................................................... 82

4.2.1. Pre-cracked reactions ......................................................................................... 84

4.2.2. Post-cracked reactions ....................................................................................... 89

4.3. Results ....................................................................................................................... 91

4.3.1. DIC Results - Extrapolation of deflection data .................................................. 91

4.3.2. Pre-cracked reactions ......................................................................................... 93

4.3.3. Post crack reactions............................................................................................ 98

4.4. Discussion ............................................................................................................... 107

4.5. Conclusion............................................................................................................... 110

5. PVB Material Properties ................................................................................................. 111

5.1. Introduction ............................................................................................................. 111

5.2. Methods ................................................................................................................... 112

5.2.1. Experimental data ............................................................................................ 112

5.2.3. Material model fit ............................................................................................ 116

5.2.4. Finite deformation viscoelasticity .................................................................... 120

Data fit method ............................................................................................................... 123

5.3. Results ..................................................................................................................... 126

5.3.1. Prony series - Hoo Fatt spring ......................................................................... 126

5.3.2. Prony series – Reduced polynomial spring...................................................... 130

5.3.3. Finite deformation viscoelasticity .................................................................... 134

5.4. Discussion ............................................................................................................... 136

5.5. Conclusion............................................................................................................... 139

6. Glass – PVB Delamination ............................................................................................. 141

6.1. Introduction ............................................................................................................. 141

6.2. Method .................................................................................................................... 142