Experimental and Numerical Study On The Low Velocity Impact Damage of A Shear Dominated Compsoite

Download as pdf or txt
Download as pdf or txt
You are on page 1of 32

AIAA SciTech Forum 10.2514/6.

2019-1269
7-11 January 2019, San Diego, California
AIAA Scitech 2019 Forum

Experimental and Numerical Study on the Low Velocity Impact


Damage of a Shear Dominated Composite Laminate

Shiyao Lin∗ and Anthony M. Waas †


University of Michigan, Ann Arbor, MI, 48109

In this paper, the low velocity impact (LVI) on a shear dominated composite laminate
with stacking sequence [45/0/-45/90]3s has been investigated experimentally and numerically.
The stacking sequence is specifically designed such that each interface would have a ply angle
difference of 45° between the upper and lower adjacent plies, providing a shear dominated
damage behavior with respect to delamination. Drop-tower tests are carried out with the
impact energy of 25J and an impactor mass of 7.5Kg. 3D digital image correlation (3D) is
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

done to measure the in-situ deformation of the specimen bottom surface. Nondestructive
inspections (NDIs) including ultrasound C-scanning and X-ray based micro-CT are performed
to characterize the internal damage after the LVI event. A continuum shell based finite
element (FE) model has been applied to predict the LVI event, in the sense of load-time and
load-displacement responses, and damage modes including fiber breakage, matrix splitting
and delamination. The intra-ply damage mechanisms are modeled using Enhanced Schapery
Theory (EST) and the inter-ply damage mechanism is based on defining cohesive contact
behavior between plies. FE predictions are compared against experimental findings and good
agreement, especially in terms of detailed ply-by-ply damage patterns, are obtained. Further
discussion on the interactions between damage modes in LVI events are provided with the help
of post-impact micro-CT images and numerical results.

I. Nomenclature

WT = Total work potential


WE = Elastic potential
WS = Dissipated potential
S = Dissipated potential due to matrix micro crackings
f
SI = Dissipated potential due to fiber macro damage
SIm = Dissipated potential due to matrix transverse macro damage
SImI = Dissipated potential due to matrix shear macro damage
Sr = Reduced damage variable
gs = Schapery degradation function in shear direction
es = Schapery degradation function in transverse direction
XT = Tensile critical strain in fiber direction
XC = Compressive critical strain in fiber direction
YT = Tensile critical strain in transverse direction
YC = Compressive critical strain in transverse direction
Z = crithical shear strain
f
DI = Damage variable of fiber
DI m = Damage variable of matrix in transverse direction
f
DI I = Damage variable of matrix in shear

∗ PhD student, Department of Aerospace Engineering, University of Michigan,


† Richard A. Auhll Department Chair, Felix Pawlowski Collegiate Professor, AIAA Fellow

Copyright © 2019 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
II. Introduction
iber reinforced polymer composites (FRPC) have shown appealing behavior in a wide variety of applications,
F especially in the aerospace industry, as a result of its excellent attributes such as high strength-to-weight ratio, ability
to be tailored, and tolerance to fatigue and corrosive environments. However, during the life of a composite structure,
LVI is often of concern during service and maintenance operations. An event such as dropping a tool can cause barely
visible impact damage (BVID) with internal failure and reduction in compressive strength, depending on the severity
of the LVI event. Therefore, the LVI problem has received considerable attention and standardized test methods [1]
together with numerous modeling techniques [2–13] have been developed to investigate LVI damage on laminated
composites. According to test results and computational predictions, damage mechanisms including fiber breakage,
matrix splitting and delamination are always found. And, most of the time, these mechanisms interact with each other.
It has been widely observed that, on the impacted side of the laminated panel, a local indentation would be caused
by the impact event, with carbon fiber kinking and matrix splitting. In the middle part of the material, matrix splits
happen within the plies and delaminations between plies can be induced by the matrix splits. The interaction between
delamination and matrix splitting is often referred to as ‘delamination migration’ by [8, 10]. In lower plies, extensive
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

matrix splitting is found due to tensile stress generated by the global bending of the panel. Sometimes, with impact
energies higher than the BVID limit, fiber tensile rupture may also happen in the bottom plies, which usually correlate
to a sudden drop of load in the load-time and load-displacement curves [3]. A great amount of numerical studies have
been done on simulating and predicting the LVI induced damage, and a brief literature review is provided as follows.

A. Typical Damage Modes and Interaction in an LVI Event


Understanding the process of impact damage initiation and growth and identifying the governing parameters is
important for developing modeling tools for LVI problems. On the impacted surface, indentation and cracks due to fiber
kinking are usually observed and on the back surface, signs of matrix splitting are often seen. The indentation and
fiber kinking on the top surface are shown in figure 1(a) and the matrix splitting on the back surface is shown in 1(b).
Sometimes, when the deflection is large enough, fiber rupture on the back surface occurs. With the help of ultrasonic
C-scanning, delamination inside the specimen can be detected. The delamination shapes are usually referred to as being
“peanut” shape and the longer axis of the “peanut” is usually aligned with the fiber direction of the lower ply. The
delamination footprint is shown in 1(c). A more detailed characterization of internal damage can be made using X-ray
based micro-CT. Internal matrix splits, and the interaction between intra-ply and inter-ply damage, which are difficult to
observe with C-scanning, can be seen in micro-CT scans. The intra-inter interaction of failure is usually observed as the
“pine tree” shape [14]. Conventionally, the first damage to appear is matrix cracking. When the matrix crack increases,
delamination will quickly occur [12]. The interactions between the matrix damage and delamination typically have two
types. When the laminate is thick, the matrix cracks are first induced in the top layers near the impacted side because of
the high localized shear stresses. Damage would progress top down, resulting a “pine tree” pattern. For thin laminates,
bending tensile stresses in the back side of the laminate will induce matrix cracks and then delamination. A reversed
“pine tree” pattern can be observed [14]. The two interaction types are shown in 2(a-b) and the corresponding “pine
tree” patterns are shown in 2(c-d).

B. Numerical Modeling
Several numerical models that have been implemented with the finite element method (FEM) are briefly reviewed in
this paper. It is by no means an exhaustive review, but such survey would be beneficial to understanding the failure
mechanisms as well as examining the developed numerical model in this article. Falzon et al. [6, 11] used 3D volume
elements with user-defined material to capture intra-ply failures and Abaqus in-built surface-based cohesive behavior to
capture inter-ply failure. For fiber failure, tensile and compressive damage were both simulated. For matrix damage,
tensile, compressive and shear failure were simulated. Nonlinear shear response of the matrix damage was also modeled
using a fitted cubic law. The shear strain was decomposed into an elastic and inelastic part to capture the post-impact
indentation. Matrix crushing and fragment interaction effects was modeled by setting the fiber-direction normal stress
residual value as comparable to transverse compressive strength. Soutis et al. [13] used 3D volume elements with
user-defined material to capture intra-ply failures. Delamination was simulated using a finite-thickness layer of Abaqus
in-built cohesive element. Nonlinear matrix shear response was captured based on a semi-empirical expression. Inelastic
shear strain was also defined to account for the indentation. Crushing and fragment interaction was considered by setting
a fiber-direction residual stress equal to a finite transverse strength at final failure. In the later work of Soutis et al. [15],
fiber failure was modeled using 3D volume elements, while matrix damage and delamination was modeled using Abaqus

2
Table 1 A summary of numerical models in literature

Bouvet et al. Falzon et al. Soutis et al. Gonzalez et al. Thorsson et al.
[12] [11] [13, 15] [10] [3–5, 16]
Tensile failure Yes Yes Yes Yes Yes
Fiber failure Compressive failure Yes Yes Yes Yes Yes
Fiber crushing Yes Yes Yes No No
Nonlinear shear No Yes Yes No Yes
Matrix damage Inelastic strain Yes Yes Yes No No
Matrix crushing No No No No Yes
Delamination Cohesive COH3D8 COH3D8 COH3D8 DCZM
contact
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

in-built cohesive elements. The reason for the newer approach is believed to be a reduction in the computation time.
Gonzalez et al. [10] implemented intra-ply failure model using 3D volume elements with user-defined material. For
delamination, two strategies including finite-thickness cohesive element and non-thickness cohesive contact were carried
out. Shear nonlinearity and matrix crushing effects were not considered. Thorsson et al. [3–5, 16] used 3D continuum
shell elements with user-defined material for intra-ply failures and user-defined element based on discrete cohesive
zone model (DCZM) for delamination. Nonlinear shear response of matrix was accounted using a thermodynamically
based work potential theory, namely the Schapery theory [17]. Matrix compression residual strength was modeled with
estimated values from calibration simulations. A summary of the literature review on damage models can be found in
table 1. In table 1, the COH3D8 is the 3D 8-noded cohesive element in-built in Abaqus, and the cohesive contact is the
cohesive behavior based contact definition in-built in Abaqus. The DCZM is the explicit user-defined element following
the DCZM model, as described in [18, 19].

C. Objectives of the Study and Structure of the Article


According to the literature review, the three major damage modes including fiber kinking, matrix splitting and
delamination are usually included in numerical models. However, the interaction between them is the key concern. For
example, in the middle plies, the delamination is often triggered by the upper and lower ply matrix splitting. In the
top most plies, where both fiber kinking and matrix splitting are present, the interaction can become more complex.
Therefore, the objective of this paper is to design and conduct tests on a specific stacking sequence, which can embody
the interaction phenomena, for post-impact observation and associated predictive modeling.
The specifically designed stacking sequence is [45/0/-45/90]3s , which is a shear-dominated layup. The shear-
dominated is in the sense of delamination, which means that, each interface has a difference in the angle of 45°. In this
way, the upper and lower plies of the interface should have the corresponding matrix splitting along the fiber direction
and a fan-shaped delamination would be formed due to the enclosing of the adjacent-ply matrix splits. In addition, as
the stacking sequence is repeating itself in a rotating manner, the delamination through the thickness should have a
rotating-fan like pattern when subjected to a face-on view. However, this phenomenon might be only significant in the
middle and bottom parts of a specimen, since in the top-most few plies, the fiber kinking will also play a role and cause
delamination.
The paper is arranged as follows. Starting from section III, the experimental results will be demonstrated and
analyzed. In section IV, the numerical model used in this study will be explained, theoretically and its numerical. Section
V will compare the numerical results with experimental findings and the results will be investigated and discussed.
Conclusions and future work will be provided in section VI.

III. Experimental Results


As mentioned in the introduction, the studied laminate stacking sequence is [45/0/-45/90]3s . The thickness measured
is 4.46 mm. Divided by 24 plies, the average ply thickness is 0.186 mm. The material type is T800S/3900-2B. Special
attention should be paid to the fact that the material is an interface toughened material manufactured by Toray Industries.

3
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 1 Typical failure morphologies: (a) indentation and fiber kinking on the top surface, (b) fiber rupture and
matrix splitting on the back surface and, (c) delamination footprint of the "peanut" shape [3]

Fig. 2 Typical types of failure interaction: (a) matrix shear crack induced delamination, (b) matrix tensile
crack induced delamination [20, 21], (c) the “pine tree” pattern, (d) the reverse “pine tree” pattern

4
A. Face-on Impact Tests

1. Drop Tower Test Setups


Following the ASTM standard for drop-weight impact tests [1], the specimen size is 150 mm × 100 mm. Face-on
impact tests are performed using an Instron CEAST Dynatup 9350 drop tower system. The impact energy is 25J,
which is below the BVID limit. The impactor mass is 7.5Kg, which is follows the study in [5]. The impactor shape is
hemispherical and the diameter is 20 mm. The impact fixture design draws from the ASTM standard [1], while some
changes have been made. In stead of clamping the specimen, the specimen is fixed by sandwiching between upper and
lower rollers. The radius of the roller is 6.35mm, and the center of the roller is placed 6.35mm from the edge of the
specimen. To validate the test results, three specimens are tested under the same impact energy.
The impact test results in terms of load-time curves and load-displacement curves are shown in figure 3 and 4. As
seen in the figure, the results are highly repetitive. No clear damage induced oscillation is seen, meaning that there is no
drastic damage growth in the material and the tested specimens are within the BVID limit. Typically, instantaneous
drops in both load-time and load-displacement curves denotes fiber breakages in the bottom plies. In figure 4, residual
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

displacements are noticed. The average residual displacement is 1.30mm. There are many reasons contributing to this
residual displacement. As in [5], a simplified 1D spring-mass system is used to explain the phenomenon. As in figure 5,
the spring is prevented from fully bouncing back to its original length. Therefore, the mass will separate from the spring
before returning to its original height. As the separation happens, the contacting force between the mass and spring
becomes zero. In this way, a residual displacement can be found when the load comes to zero in the load-displacement
curves, as in figure 4. In LVI events of laminated composites, there are several mechanisms leading to the residual
displacement/strain phenomenon. According to [12], in matrix splitting, fiber debris might stick out of matrix and into
splitting gaps. When the laminate bounces back, fiber debris would stop the matrix splits from closing up. In addition,
as delamination happens, friction forces exist between detached plies, and this would also contribute to the residual
strain effect. Microscopy would help to understand the enriched mechanisms for this phenomenon.
Critical parameters including the maximum load, time to reach maximum load, impact duration, maximum impactor
displacement and residual displacement of the three tests are measured and listed in table 2. From table 2, the critical
parameters are highly repeatable.

2. 3D Digital Image Correlation


During the tests, in-situ 3D digital image correlation (DIC) is performed with two Photron SA-X high-speed cameras
to capture the out-of-plane deformation of bottom surfaces of the specimens. The two cameras are synchronized and
triggered by the signal from the drop tower. The camera resolution is 684 × 512 pixels. 20,000 frames per second
recording rate is used. With the 3D DIC high-speed measuring system, the full displacement and strain fields of the
back surfaces can be obtained using the 3D DIC software ARAMIS.
The 3D DIC results of specimen 1 are shown in figure 6. In the 3D DIC analysis, a 10mm × 10mm area in the center
of the back surface is taken to calculate averaged center displacement. In addition, a section crossing the center is taken
to measure the displacement distribution over the section. The area and section are denoted as "Area A" and "Section
B" in the rightmost array of the figure 6. Average displacement values of the Area A, displacement distributions over
the Section B and the displacement contour results are plotted in figure 6. Five time points in the displacement history
are chosen. In the leftmost array, residual displacement of the back surface is found to be 0.34mm, as a comparison
to the top surface residual displacement, 1.24mm. In the middle array, at 1.4ms, 2.8ms and 4.2ms, curvatures of the
curves are higher near centers, meaning that the deformation of the back surface is a combination of global deflection
and localized deflection in the center of the specimen. As shown in the rightmost array of displacement contours, no
obvious damage on the back surface is found.
The average displacement of area A over time is plotted against the impactor displacement history in figure 7. As
shown, at first, before 0.5ms, the displacements of both the surfaces are almost identical, which correlates with the
Kirchhoff-Love plate theory, in which the out-of-plane displacement field is assumed uniform through the thickness
of the plate. However, after 0.5ms, the displacements start to deviate. In this sense, the deformation of the specimen
does not follow the plate theory any more. The major factor is the highly localized center deformation of the specimen,
which can be observed in figure 6. The displacement field of the specimen can be regarded as a combination of global
flexural deformation admissible to the Kirchhoff-Love plate theory and localized deformation. The two components are
shown in figure 8. As shown in figure 7, the residual center displacements of the top and bottom surface when the
impactor and specimen separates are 1.24mm and 0.34mm.

5
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 3 Load-time curves of three specimens with 25J impact energy and 7.5Kg impactor mass

Fig. 4 Load-displacement curves of three specimens with 25J impact energy and 7.5Kg impactor mass

Fig. 5 Simplified spring-mass impact model [5]

6
Area A Displacement Section B Displacement Displacement Field

0.0 ms
Area A

Section B

1.4 ms
Area A
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Section B

2.8 ms
Area A

Section B

4.2 ms
Area A

Section B

5.6 ms
Area A

Section B

Fig. 6 3D DIC measurement of out-of-plane displacement of the back surface of the impacted specimen

Table 2 Critical parameters of the three impact tests

Spec. 01 Spec. 02 Spec. 03


Max impact load (N) 10571.7 10499.8 10643.8
Time at the max load (ms) 2.57 2.60 2.58
Impact duration (ms) 5.51 5.49 5.46
Max impactor displacement (mm) 4.46 4.50 4.49
Residual displacement (mm) 1.24 1.36 1.31

7
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 7 Center displacement of the top and bottom surface

Local flexural deformation

Global flexural deformation

LVI deformation

Fig. 8 In-situ deformation components of the impacted specimen subjected to LVI

8
30.2 mm

23.1 mm
2

10 mm 1
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 9 C-scanning result of specimen 1

B. Post-impact NDIs

1. Ultrasound C-Scanning
Specimen 1 was subjected to ultrasound C-scanning and the damage footprint was measured. In addition, B-scanning
was performed to characterize the damage with a sectional view. The machine used was a Mistras UPK-T36 immersion
ultrasound system.
The C-scanning and B-scanning results are shown in figure 9 and 10. From figure 9, the impact damage footprint is
almost a circle, with the max spanning in 1 direction being 30.2mm and that in 2 direction being 23.1mm. The reason
for the circular damage footprint is that the laminate is quasi-isotropic. The damage footprint is colored with time of
flight (TOF) information, which represent the depth of the damage. C-scanning can give a good characterization of the
overall size and shape of the damage. However, due to the overshadowing effect, which means that the damage in the
lower plies and interfaces are overshadowed by the damage in the upper plies and interfaces, interface-by-interface and
ply-by-ply damage information cannot be obtained. Figure 10 shows the sectional view of the specimen. The upper
surface and lower surface are highlighted with red bold lines. As seen in the figure, different from that shown in figure 8,
only local deformation is observed. The reason is that the figure 8 shows the deformation during the LVI and figure
10 shows the deformation after the LVI. The material under global bending is mostly in the elastic region while the
local deformation is more involved with damage. According to [12], fiber debris and friction between delaminated
plies tend to prevent the local deformation from fully returning to zero. Therefore, after LVI, the global deformation
vanishes but the local deformation component is preserved. The local deformation on the top surface is usually referred
to as permanent indentation. The permanent indentation from the B-scanning is 1.24mm, which is the same as in-situ
measurements from drop tower tests and 3D DIC. Notice the NDIs are performed several days after the LVI tests, hence
there is no relaxation of the residual center displacement of the upper surface with time. The B-scanning result is not
fine enough to characterize the bottom surface center displacement.

2. Micro-CT
As mentioned in C-scanning analyses, the overshadowing effect makes the visualization of damage ply by ply and
interface by interface impossible. Micro-CT scanning can overcome this and give insight into enriched damage inside
the specimen. Specimen 1 is subjected to the micro-CT scanning. Dye penetrant containing Zinc Iodide is applied to
the specimen by saturating the specimen into the dye penetrant fluid for 12 hours. The recipe for the penetrant can be
found in [22]. The micro-CT machine used is an NSI X5000 CT system. The scanning voltage is 100KV, the current is
250 µA. The scanning voxel size is 21.6 × 21.6 × 21.6 µm3 .
After micro-CT, a 3D scanning model can be reconstructed. However, it is difficult to be visualized in the paper and
therefore only planar slices are presented. The top-to-bottom slices are shown in figure 11 and 12. However, it should
be noted that there is no strict criterion on which slice captures delamination or matrix splitting, and since the laminate

9
1.24 mm

2
1

Fig. 10 B-scanning result of specimen 1


Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

is deformed out of plane while the slices are planar, the damage captured from micro-CT slices is not perfectly revealing
the delamination and matrix splitting damage. A better way to fully visualize delamination is by deplying the laminate
but this technique will damage the specimen, which would be used further for compression after impact (CAI) tests.
Delaminations with deplying technology are shown in [23].
As shown in figure 11, fan-shaped delaminations are clearly observed in middle interfaces. However, in the top and
bottom interfaces, the delaminations do not have such a good pattern. As seen, from the 4th interface to 13th interface,
since the stacking sequence [45/0/-45/90]3s is self-repeating, the fan-shaped delamination is rotating ply-by-ply. Another
note is that, in the center of each slice in the middle interfaces, there is no damage happening. In addition, there is
no delaminatoin in interface 12, whose upper and lower ply angles are 90◦ . Since in the top and bottom plies, the
matrix splitting is complex and extensive, only in the middle plies they are noted with solid red lines in figure 12. As
shown, from the 5th to the 18th , the matrix splitting is also rotating ply by ply. In the upper plies (ply 1 to 4), obvious
fiber kinking is observed. In the lower plies, the matrix splitting is extensive. The reasons will be given after detailed
inspections.
As noticed in figure 11 and 12, in the upper, middle and lower layers and interfaces, the damage mechanisms are
different. Therefore, these three parts of the specimens are shown in figure 13, 15 and 17. The figure 13(a) shows the
damage on the topmost 45◦ ply. The fiber kinking perpendicular to the 45◦ direction and matrix splitting along the 45◦
direction are observed. The fiber kinking damage is very clear in micro-CT scannings since the damage is geometrically
larger than that of matrix splitting, and hence more exposed to the dye penetrant, which make the color brighter in
micro-CT results. From figure 13(b), for the slice 0.2mm below the topmost slice, enriched damage modes are observed.
As mentioned before, the slice is planar and the top few plies are deformed significantly. Therefore, in the slice, damage
from multiple plies can be seen. The red solid lines represent fiber kinking and matrix splitting from the topmost ply.
The green solid lines represent the fiber kinking and matrix splitting of the second 0◦ ply. The blue solid line represents
the fiber kinking from the third -45◦ ply. Enclosed by these solid lines, delaminations are seen. The fiber kinking
and matrix splitting both serve as initiation for delamination. In 13(C), for the slice 0.26mm below the topmost slice,
fiber kinking from the first, second and third plies are observed, and matrix splitting from the third -45◦ ply is seen.
Delamination formation enclosed by fiber kinking and matrix splitting is shown in the sub-figure. A sketch is drawn to
illustrate the enriched damage modes in the top part of the specimen in figure 14.
Figure 15 shows the typical damage modes in the middle part of the specimen. From figure 15(a) and (c), matrix
splitting in the ply -45◦ and 90◦ are observed and highlighted by the red solid line. In figure 15(b), a fan-shaped
delamination is seen and found to be enclosed by the upper and lower ply matrix splitting. Since in the middle plies, no
fiber kinking is present, and matrix splitting is not extensive, the formation of the delamination is simpler and hence the
delamination is better shaped than that in the top and bottom plies. Again, a sketch is drawn to illustrate the damage
modes in the middle part of the specimen in figure 16. It should be noted that only a quarter of the specimen is drawn to
reveal the internal damage.
Figure 17 shows the damage modes in the bottom part of the specimen. Figure 17(a) shows the matrix splitting in
the bottom-most 45◦ ply. Extensive matrix splitting is observed and a clear crack spacing pattern is found. Figure 17(b)
shows the slice 0.22mm above the bottom slice. Delamination between 0◦ ply and 45◦ ply is seen, as well as the matrix
splitting in both plies. Again, the matrix cracks widely spread out with uniform crack spacings. Figure 17(c) shows
the delamination between -45◦ ply and 0◦ ply, and the matrix splitting from both plies. Compared with figure 15, the
delamination in the bottom slices are not as well-shaped as in the middle slices. One reason is that, in the middle part,

10
Interface [45/0 ] Interface [0/-45 ] Interface [-45/90 ] Interface [90/45 ] Interface [45/0 ] Interface [0/-45 ]

1 2 3 4 5 6
3
Interface [-45/90 ] Interface [90/45 ] Interface [45/0 ] Interface [0/-45 ] Interface [-45/90 ] Interface [90/90 ]

7 8 9 10 11 12

Interface [90/-45 ] Interface [-45/0 ] Interface [0/45 ] Interface [45/90 ] Interface [90/-45 ] Interface [-45/0 ]
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

13 14 15 16 17 18

Interface [0/45 ] Interface [45/90 ] Interface [90/-45 ] Interface [-45/0 ] Interface [0/45 ] 40 mm

40 mm

19 20 21 22 23

Fig. 11 Delamination slices of the micro-CT scannings

each ply will only have two matrix splitting on both sides of the impact circle. The delamination can be neatly formed
following the mechanism shown in figure 16. For the bottom part of the specimen, the matrix splitting is extensive,
and multiple matrix splitting can serve as initiation and boundaries for the delamination. Therefore, the outline of the
delamination is not as clear as that in the middle part. An alternative explanation is from a perspective of energy. In the
bottom plies, since more matrix splitting exist, more energy is released by matrix cracking and hence the extent of
delamination is reduced. Since the interaction between damage modes is not clean, illustration is not drawn here.
Sectional view of internal damage is provided in figure 18. The section is 20mm from the edge of the observed area
and crossing the center of the specimen. At several locations, corresponding top-to-bottom views are attached. From
figure 18, a pine tree shaped damage pattern is observed. In the top part, relatively large shiny dots denote fiber kinking
and induced large-size delaminations, as seen in subfigure (d). In the middle part, staircase-like pattern is observed,
which correlate to subfigure (b) and (d). The staircase-like pattern is often referred to as delamination migration. In the
bottom part, extensive matrix splitting is observed, corresponding to subfigure (c).

C. A Summary of Experimental Findings


Three drop tower tests of 25J impact energy and 7.5Kg impactor mass for the stacking sequence [45/0/-45/90]3s
have been performed. The load-time and load-displacement curves are highly repeatable. 3D DIC to measure the
in-situ displacement field of the bottom surface of the specimens are carried out and the displacement-time responses at
the center of the bottom surface and upper surface are compared. The deformation of the specimen is found to be a
composition of global and local flexural deformations. From the C-scanning results, the impact damage footprint is
close to circular, since the layup is quasi-isotropic. From the B-scanning, only top surface permanent indentation is
found, and the indentation depth is identical to the in-situ value found in the load-displacement curve, meaning there is
minimal relaxation of the top-surface indentation. Micro-CT results revealed abundant damage modes and mechanisms.
The mechanisms vary among three parts of the specimen. In the top part, both fiber kinking and matrix splitting serve as
initiation for the delamination. In the middle part, delamination is well bounded by shear matrix cracks. In the bottom
part, extensive matrix splitting is present and hence the shape of the delamination is not as neat as that in the middle. A
rotating-fan pattern of delamination is observed, due to the design of the stacking sequence, to embody the interaction

11
40 mm

Ply [45] Ply [0 ] Ply [-45] Ply [90] Ply [45 ] Ply [0]

1 2 3 4 5 6

Ply [-45] Ply [90] Ply [45] Ply [0] Ply [-45] Ply [90]
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

7 8 9 10 11 12

Ply [90] Ply [-45] Ply [0] Ply [45] Ply [90] Ply [-45]

13 14 15 16 17 18

Ply [0] Ply [45] Ply [90] Ply [-45] Ply [0] Ply [45]

19 20 21 22 23 24

Fig. 12 Matrix splitting slices of the micro-CT scannings

Fig. 13 Slices in the top part of the specimen: (a) The top most slice, (b) the slice 0.2mm below the top slice,
and (c) the slice 0.26mm below the top slice

12
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 14 Damage mechanisms in the top part of the specimen

Fig. 15 Slices in the middle part of the specimen: (a) slice for the ply -45◦ , (b) slice for the interface [-45/90],
and (c) slice for the ply 90◦

Fig. 16 Damage mechanisms in the top part of the specimen

13
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 17

Fig. 18

14
Sectional view of internal damage
Slices in the bottom part of the specimen
between matrix splitting and delamination. From the sectional view, typical features including pine tree shaped damage
distribution and staircase like pattern are observed.

IV. Methodology of the Numerical Model


Numerical modeling of the LVI damage is established based on FEM. For intra-ply damage prediction, damage
models are based on the crack band (CB) method [24]. CB lies within the category of the weak discontinuity method.
Progressive failure analyses for laminated composite can also be done using the strong discontinuity method, including
variational multiscale continuum method (VMCM) [25], and a category bridging weak and strong discontinuity analyses,
such as the continuum decohesive finite element (CDFE) method, [26–28]. The pre-peak nonlinearity due to matrix
micro cracks and post-peak damage are captured with the EST method. For inter-ply delamination prediction, cohesive
contact behavior in-built in Abaqus is used. The following part of this section will provide details for the intra-ply
damage model and inter-ply damage model.
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

A. Intra-ply Damage Model: Enhanced Schapery Theory


EST [29] is a a thermodynamically based work potential model originated from the Schapey Theory (ST) [17].
EST captures pre-peak nonlinearity due to matrix micro-cracking and post-peak degradation due to macro-cracks in
composite materials. The theoretical background of EST is given below.
The total work potential, WT , is composed of an elastic potential WE , and the dissipated potential WS . The elastic
potential is contributed by the elastic deformation of the material while the dissipated part is both caused by matrix
micro-cracks and macro-cracks.

1. Pre-peak Damage
The term ’pre-peak’ represents the part on the stress-strain curve before the tangent stiffness decreases to zero, as
shown in figure 19. Pre-peak nonlinearity usually happens in matrix shear and tensile cracking. In the pre-peak part,
WS is solely composed of energy dissipation S of micro cracking, namely WS = S. And, the work-potential balance
becomes:

WT = WE + S (1)
where, S and WE are illustrated in figure 19. According to [17], the total work potential WT is stationary with regard to
S, which is:
dWT
=0 (2)
dS
Combining equation 1 and 2:
dWE
= −1 (3)
dS
Apply law of thermodynamics, it follows that:

SÜ ≥ 0 (4)
In equation 1 and 3, the elastic potential can be expanded as follows according to classical lamination theory (CLT)
in a 2D plane stress state.
1
WE = (E11 11 2 + E22 22 2 + G12 γ12 2 ) + Q12 11 22 (5)
2
In equation 5, elastic modulus in the fiber direction E11 should not be affected by the matrix micro-damage and
hence remain constant in the pre-peak region. However, the transverse and shear moduli E22 and G12 are determined
with regard to the micro-damage state of the matrix following equations

E22 = E220 es (S) (6)

15
G12 = G120 gs (S) (7)
where, E220 and G120 are pristine material parameters, es (S) and and gs (s) are transverse and shear Schapery
micro-damage functions, which are determined from tests [29].
By assuming Q12 = ν12 E22 is independent of and therefore remains constant, the micro damage evolution equation
is (derived from equation 3,5,6 and 7),

1 des (S) 2 dgs (S) 2


(E220 22 + G120 γ12 ) = −1 (8)
2 dS dS
According to [17, 29], S varies linearly with  3 where  is applied strain in the experimental coordinate frame. A
reduced damage variable, Sr is therefore introduced:
√3
Sr = S (9)
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Substituting the reduced damage variable to equation 8, the evolution function is transformed into:

1 des (Sr ) 2 dgs (Sr ) 2


(E220 22 + G120 γ12 ) = −3Sr 2 (10)
2 dSr dSr
Equation 10 is an ordinary differential equation and can be solved with regard to Sr . Finally, the pre-peak constitutive
relation considering micro-damage of matrix in the transverse and shear direction can be written as:
σ11   E110 ν12 E220 es (Sr ) 0   11 
    
σ22  = ν12 E220 es (Sr ) E220 es (Sr ) 0   22  (11)
    
 τ12   γ
    
   0 0 G g (S )
120 s r   12 
  

The Schapery micro-damage functions are acquired through experiments. +/-45 angle-ply laminate is recommended
for the shear micro-damage function gs determination. In this paper, the matrix transverse micro-damage function
es is assumed identical to gs for the easiness of implementation. The +/-45 angle-ply laminate test results of the
T800S/3900-2B material have been reported in [30, 31].

2. Damage Initiation
Damage initiation in the intra-ply damage model is following the Hashin criteria. From the initiation time point, the
stress-strain response would enter the post-peak region, meaning the tangent stiffness starts to become negative. In the
current model, the initiation of fiber direction in tensile and compressive modes are:
11 2
( ) =1 (11 ≥ 0) (12)
XT
11 2
( ) =1 (11 < 0) (13)
XC
where XT and XC are the fiber direction tensile and compressive critical strains. While for matrix cracking, mode mixity
is introduced as:
22 2 γ12 2
( ) +( ) =1 (22 ≥ 0) (14)
YT Z
22 2 γ12 2
( ) +( ) =1 (22 < 0) (15)
YC Z
where YT and YC are the matrix transverse tensile and compressive critical strains. Z is the longitudinal shear critical
strain of the lamina. It should be specifically pointed out that, as the fiber failure initiation criteria get satisfied, the
matrix mode would automatically start degrading, whereas the matrix initiation does not start fiber degradation.

16
3. Post-peak Damage
In the post-peak model, CB method [24] is implemented. The work-potential relationship originally in equation 1 is
now becoming:
f
WT = WE + S + SI + SIm + SImI (16)
f
where, SI , SIm andSImI
are energy dissipated due to mode I fiber damage, mode I and mode II matrix damage. The current
work-potential relationship is illustrated in figure 20.
The degraded moduli of the material after entering the post-peak region are:
−1
1 11 − 11
C
E11 = [ − f
] (17)
E110 f t I C lel , θ +π/2
tIC (1 + f (11 − 11
C
))
2G I C
−1
22 − 22 C
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

1
E22 =[ ∗ − m ] (18)
E22 t m (1 + tI C lel ,θ
2G m (22 − 22 ))
C
IC IC
−1
1 γ12 − γ12 C
E22 = [ − m ] (19)
G∗12 2t m (1 + tI I C lmel , θ (γ − γC ))
I IC 4G 12 12
IC

where, E22 ∗ is the value of matrix transverse modulus at the time of initiation and G ∗ is the shear matrix modulus
12
f m and t m are fiber mode I and II strengths. Correspondingly, G f , G m and G m are fracture
at initiation. tIC , tIC I IC IC IC I IC
toughnesses.
The damage factors which represent the degree of damage are defined as:

f E11
DI = 1 − (20)
E110
E22
DIm = 1 − ∗ (21)
E22
G12
DImI = 1 − (22)
G∗12
The matrix relating stress and strain is defined as:
σ11   E110 (1 − D f ) ν12 E22
∗ (1 − D m ) 0   11 
   I I   
σ22  = ν12 E22
∗ (1 − D m ) E ∗ (1 − D m ) 0  22  (23)
     
I 22 I 
 τ12   γ12 
   
∗ (1 − D m )
 
   0 0 G 12 II   
The failure criterion for matrix is also mixed-mode:
GmI Gm II
+ =1 (24)
Gm
IC G m
I IC

B. Inter-ply Damage Model: Cohesive Contact Behavior


The inter-ply delamination modeling is based upon the cohesive contact behavior in-built in the commercial software
Abaqus. According to the model, as two adjacent plies contact with each other, the cohesive contact is automatically
turned into a general contact definition while when the two adjacent plies become separated, the cohesive contact
enters the mode dictated by the traction-separation law. The traction-separation law is based upon a mixed-mode law
introduced in [32]. In this mixed-mode law, an effective separation is defined as a combination of separations of all
modes. An effective traction-separation law is also defined to dictate the effective traction from the effective separation.
The last step is to project the effective traction back to physical tractions of all fracture modes. This model has been
widely applied to many problems. Nguyen and Waas [33] developed a novel mixed-mode cohesive formulation, in
which the definition of the effective separation has more practical physical meanings and the tractions of all modes are
always guaranteed to vanish even with the mixed-mode ratio keeps changing.

17
Pre-peak
𝜏12

S
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

WE

𝛾12

Fig. 19 Pre-peak shear stress-strain relationship

𝜎 Traction-separation Stress-strain
𝜎

S S

Gc
Gc/lel

𝛿 𝜖
𝜖𝐶

Fig. 20 Pre-peak and post-peak stress-strain relationship

18
Impactor
Roller support

Area II with
micro damage
Area I with
micro and
macro damage

Area II
mesh, each
layer
Area I mesh,
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Tie attached
each layer
constraint with tie
attached with
connecting constraint
cohesive
contact area I and
II

Fig. 21 FEM model for LVI

C. Modeling Strategy
The current FE model is illustrated in figure 21. As shown, the intra-ply elements are continuum shell elements
SC8R, while the inter-ply model is based on cohesive contact definition. A global-local meshing strategy is utilized to
reduce the computing time. Area I is under the impactor which can undergo both micro and macro damage. Area II is
far from the damage zone, and hence only micro damage is allowed to happen in area II. Between each layer of area I,
the cohesive contact behavior is defined for delamination prediction. Between each layer of area II, tie constraint of all
degrees of freedom (DOFs) is defined, meaning that no delamination can happen between plies. Area I and area II are
connected by tie constraint of all DOFs. The impactor and roller support are modeled by rigid surface elements R3D3
and R3D4. Initial velocity is given to the impactor according to the impact energy and mass property is assigned to the
impactor. All DOFs other than that along the z direction is constrained for the impactor and all DOFs are fixed for the
roller supports. In the current analyses, mesh size for Area I and 2 are 0.6 × 0.6mm2 and 1.0 × 1.0mm2 . It should be
noted that, unlike Abaqus in-built cohesive element and user element based upon the discrete cohesive zone model
(DCZM) [5], the cohesive contact definition does not require the nodes to match between contacting surfaces, FEM
model with fiber-aligned mesh can be easily implemented. This is one of the future directions for the current work.

V. Numerical Results
Numerical prediction is performed on the LVI of the [45/0/-45/90]3s specimen. The modeling strategy is explained
in section IV. The initial velocity of the impactor is 2.58m/s. The material parameters used are listed in table 3. The
material parameters come from in-house tests [30, 31] and [5, 7]. Tension tests on [±45◦ ] specimens were done to back
out the fitting parameters for the Schapery function gS . For the easiness of implementation, the transverse micro damage
function eS is set equal to gS . A fifth order polynomial is used to fit the Shapery function and details of the fitting
procedure can be found in [3]. The Schapery function describing the degradation of the stiffness due to micro damage is
shown in figure 22. Except for the material parameters, the computational details of the model is provided in table 4.

A. Load and Deformation Predictions


The predicted load-time and load-displacement responses compared with experimental results are provided in figure
23 and 24. As seen in figure 3 and 4, the test results are almost identical, therefore only the results of specimen 1 is
taken for comparison here. From figure 23, the predicted load-time response is almost on top of the test curve. Two
differences are noticed. The first difference is that on the predicted curve, after 1.15ms, the modeling result starts to get
more oscillatory than the test result. The reason for this is believed to be caused by damage initiation. With a explicit
numerical method, the oscillation is magnified. Another difference is after the peak of the curve, between 2.85ms and

19
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 22 The Schapery function of T800S/3900-2B for matrix shear micro damage

Table 3 Material properties for T800S/3900-2B used in numerical predictions

Fiber mode I toughness


E1 1-direction modulus 167542 MPa GTf I 40 Nmm
(tension)
Fiber mode I toughness
E2 2-direction modulus 8480 MPa GCfI 10 Nmm
(compression)
G12 Shear modulus 4481 MPa GI M Matrix mode I toughness 0,54 Nmm
ν12 Poisson’s ratio 0.3 GI I M Matrix mode II toughness 1.64 Nmm
1-direction initiation strain
X1T 0.0176 σn Interface normal fracture strength 66.9 MPa
(tension)
1-direction initiation strain
X1C 0.0106 σt Interface shear fracture strength 100 MPa
(compression)
2-direction initiation strain
Y2T 0.007 GI Interface mode I toughness 0,54 Nmm
(tension)
2-direction initiation strain
Y2C 0.0251 GI I Interface mode II toughness 1.64 Nmm
(compression)
Z12 Shear initiation strain 0.0323

Table 4 Numerical details of the model

Number of elements 589,643


Mass scaling No
Penalty stiffness of cohesive contact 2e5 N/mm
Frictional coefficient of general contact 0.3
Total step time 7.0 ms
Stable time increment 4.34e-8 s
Number of cpus 64
Wall-clock time for running simulation 36 h

20
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 23 Predicted load-time response compared with test result

4.5ms, the load is slightly higher than that of the test result. The reason has been explained in figure 5. The residual
strain effect that is not captured in the numerical model is preventing the tested specimen from bouncing back as fast
as that predicted. However, since the LVI is within the BVID limit, such effect is insignificant. It is expected that,
with higher impact energy, this difference will be more significant between test and prediction results. From figure 24,
before the damage-induced oscillation strikes in, the predicted curve is very close to the test curve, because the current
model is capable of predicting the elastic response of the specimen very well. After the peak value, the residual strain
effect starts to play a role. In the end, the predicted curve returns to the origin, but the test curve still has a residual
displacement of 1.24mm. As shown, the predicted load-displacement curve is almost enclosed by the test curve, which
means the absorbed impact energy is lower than the test result. The impact energy comparison is shown in figure 25.
From the plot, the absorbed energy first increases and as the impactor bounces back, the residual absorbed energy is
mostly released by impact damage. The predicted absorbed energy is lower than the test result. The comparison of the
critical parameters are compared in table 5. Excellent agreement is found in terms of maximum load, impact duration
and maximum displacement. However, since the residual strain effect is not captured, the variance of the absorbed
impact energy is great.
The bottom surface displacement field acquired by 3D DIC is also used for validating the predicted results. The
out-of-plane displacement fields are compared in figure 26. Both the displacement fields are taken from time point
2.8ms, corresponding to the peak displacement. The colormaps adopt the same maximum and minimum values. The
displacement contours are very similar. The gradient of the numerical displacement field is found to be higher, which is
due to the resolution of the DIC result being not as high as the numerical result. More quantified comparison can be
drawn in terms of displacement history of area A and displacement distribution on section B, as denoted in figure 6.
The comparison of the center displacement history for both top and bottom surface are shown in figure 27. As seen, the
displacement histories of both surfaces between numerical and test results are almost identical, especially for the bottom
surface. For the top surface, the out-of-plane displacement is slightly over-predicted, and the bouncing back of the
displacement is faster than the test results, contributed by the residual strain effect. However, since most of the damage
that can cause residual strain happens near the impacted area on the top surface, the bottom surface displacement
is predicted very accurately. For the time point 1.4ms, 2.8ms, 4.2ms and 5.6ms, the displacement distributions on
section B are plotted. The curves indicate the deformed shapes of the bottom surface. Time points 1.4ms and 2.8ms are
plotted in figure 28 as the deformation gets larger. Time points 2.8ms, 4.2ms and 5.6ms are plotted in figure 29 as the
deformation bounces back. As seen in figure 28, the deformed shapes in test and predictions are right on top of each
other. From time point 2.8ms, local deformation with higher curvature is seen. As in figure 29, at time point 4.2ms and
5.6ms, the predicted deformations are slightly higher than the test curves. Such over-prediction can also be found in
figure 27, caused by the residual strain effect.

21
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 24 Predicted load-displacement response compared with test result

Fig. 25 Predicted energy-time response compared with test result

Table 5 Critical parameter comparison between predicted and test results

Test result Numerical result Error


Max impact load (N) 10571.7 10857.7 2.6%
Time at the max load (ms) 2.57 2.64 2.6%
Impact duration (ms) 5.51 5.64 2.8%
Max impactor displacement (mm) 4.46 4.58 2.6%
Absorbed impact energy (J) 11.9 8.2 31.1%

22
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

(a) (b)

Fig. 26 Bottom surface out-of-plane displacement field: (a) test result, (b) numerical result

Fig. 27 Tested and predicted center displacement history

23
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 28 Tested and predicted section displacement when deforming

Fig. 29 Tested and predicted section displacement when bouncing back

24
B. Damage Predictions
With the damage models illustrated in section IV, the damage modes including fiber kinking, matrix splitting and
delamination are captured. Internal damage inspected by NDIs including C-scanning and micro-CT are compared with
the numerical predictions.
The comparison of the C-scanning and predicted delamination footprint is provided in figure 30. The predicted
delamination footprint is plotted with the color map "white to black" in Abaqus, to achieve the similar color pattern of
the micro-CT. The overall damage variable "CSDMG" is plotted to visualize the delamination. The damage variable
represents the overall damage extent at the contact point in the cohesive contact model. It initially has the value of zero,
meaning perfect bonding. After delamination initiation, it would monotonically evolve from 0 to 1. When CSDMG is
equal to 1, there is no traction transferred between adjacent plies. In figure 30, the profile of the C-scanning damage
footprint is overlapped onto the predicted delamination footprint. An excellent agreement between predicted and
C-scanning footprints is seen. The global shape of the footprint is circular, as the stacking sequence is quasi-isotropic.
Several parts of the delamination near the top of the specimen stick out the circular shape. Such local morphology is also
observed in C-scanning and believed to be caused by fiber kinking. The maximum width of the predicted delamination
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

footprint in 1 direction is 32.95mm, as a comparison to 30.2mm in C-scanning and that in 2 direction is 23.26mm. as a
comparison to 23.1mm in C-scanning. The variance are correspondingly 8.34% and 0.68%.
A section view of predicted delamination damage compared with micro-CT scanning is provided in figure 31. The
damage variable CSDMG is used to indicate the delamination. In the top part of the specimen, delaminated plies are
clearly seen. These delaminaions are induced by fiber kinking. As in the micro-CT result in figure 18, large shinny
dots are indicating the fiber kinking and induced delaminations. In the middle part of the specimen, a cone without
damage is seen. This is one of the signatures of the delamination damage. Numerical models capable of predicting such
damage-free cone has hardly been reported. In the bottom plies, no delamination is predicted and in the micro-CT
result, only matrix crackings are observed. The overall size and shape of the delamination "pine tree" is well predicted.
However, permanent indentation is not captured.
Apart from the load curves and overall damage, interface-by-interface delamination details should also be compared.
The stacking sequence is specially designed to have the rotating-fan pattern with regard to delamination. Whether
this pattern can be captured is one of the most important criterion for the accuracy of the numerical model. As a
comparison to figure 11, predicted interface-by-interface delamination is shown in figure 32. Same as before, the
predicted delamination is indicated by the damage variable CSDMG of the cohesive contact model. In the figure, from
interface 1 to 5, some parts of the delamination stick sharply and far out of the impact circle, these parts are induced by
fiber kinking in the top part of the specimen. From interface 6 to 23, the delaminations are fan-shaped. After careful
checking with the upper and lower adjacent ply angles, the "fans" are correctly bounded from interface 8 to 11, and from
interface 13 to 22. The interface 6 and 7 have vague bounds that is dictated by the adjacent ply angles. However, the
delaminations in these two interfaces are not well propagated enough to reach the boundaries. The interface 12 is the
symmetric plane of the specimen. The adjacent ply angles are [90◦ /90◦ ]. Theoretically, no delamination should form in
this interface. However, the delamination is predicted here, mostly believed to be caused by numerical error. Very clear
fan shapes are seen in interface 18, 19 and 20. The major reason is that they are close to the bottom of the specimen,
and hence have higher bending tensile stress during impact deformation. In this way, the adjacent plies would have
longer matrix splitting and higher tendency to form larger and more well-shaped delaminations. The capturing of the
rotating-fan pattern in figure 32 is astonishing and has not been reported in relative works according to the author’s
knowledge. However, several issues still need to be pointed out. From figure 11, all of the fan-shaped delaminations
are centro-symmetric with respect to the center of the specimen. In the predicted results, such centro-symmetry is
not captured for a few interfaces, and believed to be caused by numerical error. In the micro-CT results, the neat fan
shapes start to form from interface 5, but in the numerical prediction, they start to form from interface 8. This is mostly
because the current model is not capable of capturing transverse matrix shear splitting. In the top most plies, due to
local flexural deformation, transverse shear splitting of matrix is dominating. In the lower plies, when in-plane tensile
matrix splitting starts to take charge, the fan shapes are becoming salient. In the micro-CT results, most of the sizes of
the delaminations remain relatively constant compared to numerical results. For example, in figure 32, the delamination
in interface 17 is apparently smaller than interface 16 and 18. This again is believed to be due to numerical error. Due
to the overshadowing effect, although in some interfaces, the delamination sizes are over-predicted or less-predicted, the
overall footprint is very close to the C-scanning result, as shown in figure 30.
Fiber kinking predicted are shown in figure 33. Only topmost 5 plies have fiber kinking damage. Damage variable
ranging from 0 to 2 represents the damage extent. 0 means pristine state and 2 means total failure. As seen in the figure,
in ply 1, 2, 4 and 5, fiber kinking is concentrated to bands spanning over only the width of one element. In addition, the

25
32.95 mm

23.26 mm

2
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

10 mm
1

Fig. 30 Comparison of the C-scanning footprint and predicted foot print

fiber kinking is perpendicular to the fiber directions. However for ply 3, the predicted direction is not correct. This
might be caused by matrix splitting and fiber kinking in adjacent plies. In figure 12, fiber kinking is seen in the first 4
plies. Although it is not possible to separate fiber kinking out of the other damage modes in micro-CT results, the fiber
kinking predictions except for ply 3 are found to be satisfying in terms of size and orientation.
Matrix splitting predicted is shown in figure 34. It should be noted that, in the current EST model, if fiber kinking
happens in one element, the matrix splitting of this element will automatically happen. This is implemented in the
EST user subroutine. From ply 1 to 6, the matrix splitting is coupled with fiber kinking. From ply 7 to 21, matrix
splitting is well aligned with corresponding fiber direction of each ply. Damage free zone under the impactor is also
observed. One difference from the micro-CT results should be noted is the predicted width of matrix splitting is not
as "thin" as that in micro-CT. One of the reason is due to the fact that the nature of crack band is to smear the crack
within a finite-width band. The lack of sharpness of the crack representation might lead to erroneous prediction of the
delamination since the damage modes are interacted with each other. Since it is due to the nature of the damage model,
simply refining the mesh does not make the matrix splitting damage sharper. However, with correct splitting angle
prediction, the delamination has already been modeled rather successfully. From ply 22 to 24, large matrix damage
zones are seen. It might be to the observer’s surprise that matrix splitting seems to be spaced. However, no matrix crack
spacing mechanism is implemented in the EST subroutine, unlike what has been done in the Intra-Inter Crack Band
Model (I2CBM) [34–36]. The phenomenon is believed to be caused by numerical error.
As pointed out in figure 14, delamination is highly relative to fiber kinking. The interaction is also found in numerical
predictions. Interface 4 in figure 32 and ply 4 and 5 in figure 33 are grouped and shown in figure 35. The delamination
shape is clearly influenced by fiber kinking in the upper and lower plies. Due to the fiber kinking, the delamination sizes
in the topmost interfaces are larger than that in the lower interfaces.
Interaction between matrix splitting and delamination is also captured by the model. The interaction is salient in the
plies close to the bottom of the specimen. Interface 18 together with ply 18 and 19 are grouped in figure 36. Clear
fan-shaped delamination is bounded by matrix splitting from adjacent layers. The mechanism has been shown in figure
16. However, it should be noted that not every interface show such a clear interaction and delamination shape, Many
factors contribute to the unsatisfactory part of the predictions, including smeared matrix splitting, erroneous damage
direction prediction and numerical errors. Also, these factors are coupled with each other. Lastly, it should also be
pointed out that, to the knowledge of the authors, such damage modes interactions have hardly been reported from the
existing literature.

26
40 mm

40 mm
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

Fig. 31 Comparison of the section view of predicted and scanned delaminations

Interface [45/0 ] Interface [0/-45 ] Interface [-45/90 ] Interface [90/45 ] Interface [45/0 ] Interface [0/-45 ]

1 2 3 4 5 6

Interface [-45/90 ] Interface [90/45 ] Interface [45/0 ] Interface [0/-45 ] Interface [-45/90 ] Interface [90/90 ]

7 8 9 10 11 12

Interface [90/-45 ] Interface [-45/0 ] Interface [0/45 ] Interface [45/90 ] Interface [90/-45 ] Interface [-45/0 ]

13 14 15 16 17 18

Interface [0/45 ] Interface [45/90 ] Interface [90/-45 ] Interface [-45/0 ] Interface [0/45 ] 40 mm

40
mm
19 20 21 22 23

Fig. 32 Interface by interface delamination prediction

27
[45] [0] [-45]

1 2 3

[90] [45] 40 mm
4
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

2 0
m
m
4 5
1
Fig. 33 Fiber kinking prediction

40 mm

4 [45] [0] [-45] [90] [45] [0]


0
m
m
[4
1 2 3 4 5 6

5]
[-45] [90] [45] [0] [-45] [90]

1
7 8 9 10 11 12

[90] [-45] [0] [45] [90] [-45]

13 14 15 16 17 18

[0] [45] [90] [-45] [0] [45]

19 20 21 22 23 24

Fig. 34 Matrix splitting prediction

28
[90]

Fiber kinking s of ply 4


Fiber kinking s of ply 5
Ply 4
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

[45]

[90/45]

Ply 5

Fig. 35 Predicted interaction between fiber kinking and delamination

[-45]

Fiber kinking s of ply 18


Fiber kinking s of ply 19

Ply 18

[0]

[-45/0]

Ply 19

Fig. 36 Predicted interaction between matrix splitting and delamination

29
C. A summary of Numerical Results
EST combined with the cohesive contact behavior was used to predict the LVI event with 25J impact energy and
7.5Kg impactor mass. The predicted results are compared with test results in terms of load-time, load-displacement,
absorbed energy-time responses, bottom surface displacement field, and internal damage. Excellent agreement between
numerical predictions and test results are found between load-time, load-displacement curves, with errors of critical
parameters less than 3%. The absorbed energy is under-predicted, due to the residual strain effect being not captured
in the current model. From bottom surface comparisons, both the center displacement history and deformed shapes
are well predicted according to 3D DIC results. Before the impactor reaches its lowest point, the curves are nearly
right on top of each other. After the peak, slight differences are seen due to the residual strain effect not captured.
Delamination footprint predicted is compared with the C-scanning damage footprint. Circular footprint is predicted,
similar to the C-scanning result. The maximum error in sizes is 8.34%. Sectional view of predicted delaminations and
that in micro-CT are compared. One significant phenomenon well captured is the damage-free cone in the center of the
specimen. The overall sizes and shapes from a the sectional view are close. The rotating-fan pattern of delamination is
accurately predicted by checking interface-by-interface delamination prediction. Ply-by-ply fiber kinking are predicted
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

well, although the kinking direction in one ply erroneously predicted. Matrix splitting predictions are aligned correctly
with ply angle of each plies. By checking delaminations together with fiber kinking and matrix splitting of it’s adjacent
plies, clear interaction between these damage modes are found.

VI. Conclusions
A shear dominated stacking sequence [45/0/-45/90]3s is tested with 25J impact energy and 7.5Kg impactor mass.
3D DIC is applied to measure in-situ bottom surface deformation. Post-impact NDIs include ultrasound C-scanning
and X-ray micro-CT. Excellent consistency among test results is found. By checking the 3D DIC results, the top
surface deformation becomes greater than bottom surface deformation after damage sets in. After damage initiation, the
deformation of the specimen is regarded as a composition of global flexrual deformation and local flexrual deformation.
Same values of permanent indentation is found from both load-displacement curves and B-scanning, meaning there
is no relaxation happening for the indentation after LVI. From C-scanning, the impact damage footprint is almost
circular, since the stacking sequence is quasi-isotropic. Micro-CT results provide abundant details about internal damage
including fiber kinking, matrix splitting and delamination. A rotating-fan pattern of delamination is well observed. The
interactions between damage modes are seen. Damage mechanisms in top, middle and bottom part of the specimen are
individually explained.
The numerical model is established based on EST and cohesive contact definition in-built in Abaqus. Intra-ply
damage is modeled with EST, which is a combination of the Schapery theory to capture matrix pre-peak nonlinearity
caused by micro-damage, and the Crack Band model to capture post-peak material degradation caused by macro-cracks.
Inter-ply damage is modeled with cohesive contact behavior. Excellent agreement is found in terms load-time,
load-displacement responses, with the maximum error of critical parameters less than 3%. Bottom ply deformation
predicted is almost on top of the 3D DIC results. Delamination footprint predicted is also circular and with the maximum
error in sizes being 8.34% when compared with C-scanning result. The highlight of this work is detailed predictions on
ply-by-ply and interface-by-interface damage modes. Rotating-fan pattern of the delamination is accurately predicted
and the damage-free cone is seen from a sectional view. Fiber kinking in the top plies is predicted, although with one
ply having incorrect kinking orientation. Matrix splitting predicted is well aligned with ply angles. Interaction between
fiber kinking, matrix splitting and delamination is correctly modeled, and is the key for the correct prediction of the
rotating-fan pattern.
Some improvements to the current model are still needed. One is the incapability of capturing residual strain. This is
the major reason for the absorbed energy-time response to vary from experimental results and slight inconsistency seen
in deformation responses. Another is that the matrix splitting predicted is not sharp. This comes from the methodology
of the Crack Band method which smears the effect of a sharp crack. Also, crack spacings in bottom plies are not
predicted. Non-local algorithms implemented in the I2CBM [34–36] to capture crack spacing can resolve this aspect.
Compared with load and displacement responses, the ply-by-ply and interface-by-interface damage is more
challenging to capture with numerical models. The current study may be improved by using fiber-aligned mesh and user
defined material model with 3D stress state. In addition, user defined cohesive contact behavior with novel mixed-mode
cohesive formulation [33] might also help. Despite these drawbacks, the predictions from an efficiency (computational
time) view point are quite striking.
One of the major purpose of this work is to provide a benchmark case to challenge proposed numerical models. An

30
accurate high fidelity model should be able to capture load and displacement responses very well. In addition, features
such as damage-free cone, fan-shaped delamination and rotating-fan pattern of delamination should be reproduced in
numerical predictions.

Acknowledgments
The authors gratefully acknowledge financial support from Mitsubishi Heavy Industries (MHI). We are grateful for
the interest and support of Toshio Abe of MHI. We acknowledge with thanks the material and panels provided by Toray
CMA. The study reported here was partially completed at the University of Washington, Seattle, when both authors
were affiliated with the William E Boeing Department of Aeronautics and Astronautics.

References
[1] D7136/D7136M-12, A., “Standard test method for measuring the damage resistance of a fiber-reinforced polymer matrix
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

composite to a drop-weight impact event,” 2015.

[2] Waas, A. M., Thorsson, S. I., and Rassaian, M., “Prediction of Low-Velocity Face-on Impact Response and Compression after
Impact (CAI) of Composite Laminates using EST and Cohesive Modeling (DCZM),” 2018, p. 1704.

[3] Thorsson, S. I., “Experimental and numerical investigation of fiber reinforced laminated composites subject to low-velocity
impact,” 2017.

[4] Thorsson, S. I., Waas, A. M., and Rassaian, M., “Numerical investigation of composite laminates subject to low-velocity
edge-on impact and compression after impact,” Composite Structures, Vol. 203, 2018, pp. 648–658.

[5] Thorsson, S. I., Waas, A. M., and Rassaian, M., “Low-velocity impact predictions of composite laminates using a continuum
shell based modeling approach part A: Impact study,” International Journal of Solids and Structures, 2018.

[6] Tan, W., Falzon, B. G., Chiu, L. N., and Price, M., “Predicting low velocity impact damage and Compression-After-Impact
(CAI) behaviour of composite laminates,” Composites Part A: Applied Science and Manufacturing, Vol. 71, 2015, pp. 212–226.

[7] Ebina, M., Yoshimura, A., Sakaue, K., and Waas, A. M., “High fidelity simulation of low velocity impact behavior of CFRP
laminate,” Composites Part A: Applied Science and Manufacturing, Vol. 113, 2018, pp. 166–179.

[8] McElroy, M., Jackson, W., Olsson, R., Hellström, P., Tsampas, S., and Pankow, M., “Interaction of delaminations and matrix
cracks in a CFRP plate, Part I: A test method for model validation,” Composites Part A: Applied Science and Manufacturing,
Vol. 103, 2017, pp. 314–326.

[9] Hongkarnjanakul, N., Bouvet, C., and Rivallant, S., “Validation of low velocity impact modelling on different stacking sequences
of CFRP laminates and influence of fibre failure,” Composite Structures, Vol. 106, 2013, pp. 549–559.

[10] González, E., Maimí, P., Camanho, P., Turon, A., and Mayugo, J., “Simulation of drop-weight impact and compression after
impact tests on composite laminates,” Composite Structures, Vol. 94, No. 11, 2012, pp. 3364–3378.

[11] Donadon, M., Iannucci, L., Falzon, B. G., Hodgkinson, J., and de Almeida, S. F., “A progressive failure model for composite
laminates subjected to low velocity impact damage,” Computers & Structures, Vol. 86, No. 11-12, 2008, pp. 1232–1252.

[12] Bouvet, C., Rivallant, S., and Barrau, J.-J., “Low velocity impact modeling in composite laminates capturing permanent
indentation,” Composites Science and Technology, Vol. 72, No. 16, 2012, pp. 1977–1988.

[13] Shi, Y., Swait, T., and Soutis, C., “Modelling damage evolution in composite laminates subjected to low velocity impact,”
Composite Structures, Vol. 94, No. 9, 2012, pp. 2902–2913.

[14] Abrate, S., Impact on composite structures, Cambridge university press, 2005.

[15] Shi, Y., Pinna, C., and Soutis, C., “Modelling impact damage in composite laminates: a simulation of intra-and inter-laminar
cracking,” Composite Structures, Vol. 114, 2014, pp. 10–19.

[16] Thorsson, S. I., Sringeri, S. P., Waas, A. M., Justusson, B. P., and Rassaian, M., “Experimental investigation of composite
laminates subject to low-velocity edge-on impact and compression after impact,” Composite Structures, Vol. 186, 2018, pp.
335–346.

31
[17] Schapery, R., “A theory of mechanical behavior of elastic media with growing damage and other changes in structure,” Journal
of the Mechanics and Physics of Solids, Vol. 38, No. 2, 1990, pp. 215–253.

[18] Xie, D., and Waas, A. M., “Discrete cohesive zone model for mixed-mode fracture using finite element analysis,” Engineering
fracture mechanics, Vol. 73, No. 13, 2006, pp. 1783–1796.

[19] Xie, D., Salvi, A. G., Sun, C., Waas, A. M., and Caliskan, A., “Discrete cohesive zone model to simulate static fracture in 2D
triaxially braided carbon fiber composites,” Journal of Composite Materials, Vol. 40, No. 22, 2006, pp. 2025–2046.

[20] Petit, S., Bouvet, C., Bergerot, A., and Barrau, J.-J., “Impact and compression after impact experimental study of a composite
laminate with a cork thermal shield,” Composites Science and Technology, Vol. 67, No. 15-16, 2007, pp. 3286–3299.

[21] Chang, F.-K., and Chang, K.-Y., “A progressive damage model for laminated composites containing stress concentrations,”
Journal of composite materials, Vol. 21, No. 9, 1987, pp. 834–855.

[22] Yu, B., Bradley, R., Soutis, C., Hogg, P., and Withers, P., “2D and 3D imaging of fatigue failure mechanisms of 3D woven
Downloaded by IOWA STATE UNIVERSITY on January 9, 2019 | http://arc.aiaa.org | DOI: 10.2514/6.2019-1269

composites,” Composites Part A: Applied Science and Manufacturing, Vol. 77, 2015, pp. 37–49.

[23] Hull, D., and Shi, Y. B., “Damage mechanism characterization in composite damage tolerance investigations,” Composite
Structures, Vol. 23, No. 2, 1993, pp. 99–120.

[24] Bažant, Z. P., and Oh, B. H., “Crack band theory for fracture of concrete,” Matériaux et construction, Vol. 16, No. 3, 1983, pp.
155–177.

[25] Rudraraju, S., Salvi, A., Garikipati, K., and Waas, A. M., “Predictions of crack propagation using a variational multiscale
approach and its application to fracture in laminated fiber reinforced composites,” Composite Structures, Vol. 94, No. 11, 2012,
pp. 3336–3346.

[26] Prabhakar, P., and Waas, A. M., “A novel continuum-decohesive finite element for modeling in-plane fracture in fiber reinforced
composites,” Composites Science and Technology, Vol. 83, 2013, pp. 1–10.

[27] Nguyen, N., and Waas, A. M., “Continuum Decohesive Finite Element Modeling of Fiber-Reinforced Polymer Composites:
Mesh-Objectivity and Sensitivity Studies,” 58th AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials
Conference, 2017, p. 0431.

[28] Lin, S., and Waas, A. M., “Using the Continuum Decohesive Finite Element for Crack Growth Analysis in Fiber Reinforced
Composites,” 2018 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2018, p. 0731.

[29] Pineda, E. J., and Waas, A. M., “Numerical implementation of a multiple-ISV thermodynamically-based work potential theory
for modeling progressive damage and failure in fiber-reinforced laminates,” International journal of fracture, Vol. 182, No. 1,
2013, pp. 93–122.

[30] Nguyen, M., Vijayachandran, A., Davidson, P., and Waas, A. M., “Effect of Automated Fiber Placement (AFP) Manufacturing
Signature on Composite Performance (In press),” 2019 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials
Conference, 2019, p. 0200.

[31] Nguyen, M., Vijayachandran, A., Davidson, P., and Waas, A. M., “Interlaminar Toughened Interfaces Enhance Mechanical
Response of Composites (In press),” 2019 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference,
2019, p. 0200.

[32] Camanho, P. P., Davila, C. G., and De Moura, M., “Numerical simulation of mixed-mode progressive delamination in composite
materials,” Journal of composite materials, Vol. 37, No. 16, 2003, pp. 1415–1438.

[33] Nguyen, N., and Waas, A. M., “A novel mixed-mode cohesive formulation for crack growth analysis,” Composite structures,
Vol. 156, 2016, pp. 253–262.

[34] Joseph, A. P., Davidson, P., and Waas, A. M., “Open hole and filled hole progressive damage and failure analysis of composite
laminates with a countersunk hole,” Composite Structures, Vol. 203, 2018, pp. 523–538.

[35] Joseph, A. P., Davidson, P., and Waas, A. M., “Progressive damage and failure analysis of single lap shear and double lap shear
bolted joints,” Composites Part A: Applied Science and Manufacturing, Vol. 113, 2018, pp. 264–274.

[36] Joseph, A. P., Davidson, P., and Waas, A. M., “Intra-inter Crack Band Model (I2CBM) for Progressive Damage and Failure
Analysis of Joints,” 58th AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2017, p. 0200.

32

You might also like