EPJ Web of Conferences 26, 04011 (2012)

DOI: 10.1051/epjconf/20122604011


c Owned by the authors, published by EDP Sciences, 2012

Determination of Johnson-holmquist constitutive model parameters

for fused silica
A. Ruggiero1 , G. Iannitti1 , N. Bonora1,a , and M. Ferraro2
1
University of Cassino, DiMSAT, Cassino, Italy
2
MBDA Italia Spa, Bacoli, Italy

Abstract. Johnson-Holmquist constitutive model parameters for Fused Silica under dynamic conditions were determined by
means of an inverse calibration technique. The identification of the parameters was performed with an multi-object optimizer
using experimental data generated by two different validation tests: Taylor Cylinder Impact tests and Drop Weight tests, with
impact velocity and strain rate ranging from 1 to 100 m/s and from 102 to 104 s−1 , respectively. The validity of the parameters set
determined in this way was verified comparing numerical predictions and experimental results for an independent designed test,
given by a fused silica tile impacted at prescribed velocity by a steel sphere.

1 Introduction the most widely used constitutive models for simulating

the post-yield response of ceramic materials is the JH-2
Fused silica is a high purity synthetic amorphous silicon ceramic model. This constitutive equation, developed by
dioxide characterized by low thermal expansion coeffi- Johnson and Holmquist, incorporates the effect of damage
cient, excellent optical qualities and exceptional transmit- on residual material strength and bulking during failure
tance over a wide spectral range. Because of its wide in compression. The relevant equations describing the
use in the military industry as window material, it may response of the material are summarized in the following.
be subjected to high-energy ballistic impacts. Under such The strength of the material is described by a smoothly
dynamic conditions, post-yield response of the ceramic as varying function of the intact strength, fractured strength,
well as the strain rate related effects become significant and damage:
and should be accounted for in the constitutive modeling.

Today, one of the most widely used constitutive model for σ∗ = σ∗i − D σ∗i − σ∗f (1)
simulating dynamic behavior of ceramic materials is the
later version of the Johnson-Holmquist model [1], usually where the stresses are normalized respect to σHEL that is
indicated as JH-2 ceramic model. In this study, for the defined as follows,
identification of the model parameters, a procedure based
on the inverse calibration technique of experimental vali- σHEL = 3
(HEL − PHEL ) (2)
2
dation tests was developed. To this purpose, Taylor impact
tests and drop weight tests were designed and performed in which, HEL is the Hugoniot elastic limit and PHEL is the
at different impact velocities ranging from 1 to 100 m/s pressure at the HEL. The normalized intact and fractured
and a strain rate range ranging from 102 up to 104 s−1 . strength are given, respectively, by:
Numerical simulation of both tests was performed with
explicit finite element code. Identification of the model σ∗i = A(P∗ + T ∗ )N (1 + C ln ε̇∗ ) (3)
parameters was performed by means of structural opti-
mization using as objective functions multiple validation
metrics. The validity of the parameters set determined with σ∗f = B(P∗ ) M (1 + C ln ε̇∗ ) (4)
the proposed procedure was verified comparing numerical
predictions and experimental results for an independent in which the actual pressure, P, and the maximum tensile
designed test consisting in a fused silica tile impacted at hydrostatic pressure that the material can withstand, T ,
prescribed velocity by a steel sphere. are normalized respect to PHEL . The dimensionless strain
rate is ε̇∗ = ε̇/ε̇0 , where ε̇ is the actual strain rate and
ε̇0 = 1.0 s−1 is the reference strain rate.
2 JH-2 Model parameters The damage variable accumulates with plastic strain
according to:
 
Ceramic materials are commonly used in protective armor   ∆ε p 
D= 
 f 
 (5)
applications and may be subject to high-energy ballistic
impacts in these situations. Under simple loading condi- ∆ε p
tions, ceramics may be regarded as elastic-brittle materials.
However, when considering ballistic impacts, the post- where ∆ε p is the plastic strain increment and ∆ε pf is the
yield response of the ceramic becomes significant. One of plastic strain at fracture function of the actual pressure,

a e-mail: [email protected] ε pf = D1 (P∗ + T ∗ )D2 (6)

This is an Open Access article distributed under the terms of the Creative Commons Attribution License 2.0, which permits unrestricted use, distribution,
and reproduction in any medium, provided the original work is properly cited.

Article available at http://www.epj-conferences.org or http://dx.doi.org/10.1051/epjconf/20122604011

 EPJ Web of Conferences

Moreover, the hydrostatic behavior is modeled by

means of the following equation,

P = K1 µ + K2 µ2 + K3 µ3 + ∆P (7)

where, µ = ρ/ρ0 −1 is the hydrostatic compression and ∆P

is the bulking pressure of the material that is determined
by the amount of the accumulated damage. For the range
of pressures investigated in this work, K2 and K3 can be
Fig. 1. Fragmentation in Taylor cylinder impacted at high veloc-
neglected, while K1 is the bulk modulus, K, that can be
ity (v = 83.3 m/s).
derived by the values of the elastic constants of the material
E and ν.
As a result of the linear EOS, PHEL can be expressed
as a function of HEL and two elastic moduli, K and G, and
then eliminated from the set of unknown parameters, using
the following expression:


4G
PHEL = HEL 1 − (8)
3K + 4G

Then, in order to completely describe the mechanical

behavior of a brittle ceramic with the JH-2 model the
following nine parameters have to be determined: A, B, C,
M, N, T , HEL, D1 , D2 .

3 Experimental tests
In order to generate data for the calibration of the JH-2
model parameters, two types of tests were designed and
performed:
a) Taylor Cylinder impact test; Fig. 2. Spall fracture in Taylor cylinder impacted at low velocity
b) Drop Weight test. (v = 4.2 m/s).
In the Taylor test, a cylinder, made of the material of
interest, is launched, at a given velocity, along its axial axis
against a rigid anvil. Input parameters for the test are the and 2 the high and low impact velocity failure modes are
impact velocity and the initial geometry of the cylinder. shown.
Traditionally, the output of this test is the deformed shape In the Drop Weight test, a ceramic tile, supported by
of the impacted cylinder. Deformed profile, diameter of a rigid plate, is hit by an impactor with hemispherical
the impact surface and final length are obtained post head of 16.0 mm diameter. Ceramic tiles were available
mortem. In case of breakup, test outputs are the number with different geometries: a rectangular tile with nominal
and the dimensions of fragments. When dealing with dimensions L = 97.0 mm, W = 28.0 mm, T = 4.0 mm,
brittle materials, such as ceramics, fragments may also be and a square tile with L = 50.0 mm and T = 8.0 mm. A
generated from secondary impacts. A clearer insight of the total of 19 tests were performed with 5 different configu-
dynamics of fragments formation, as a result of the stress rations: rectangular tile at two different velocities, square
waves travel and interaction, can be better obtained by tile at two different velocities, and another configuration
means of high speed video-recording. To this purpose, it in which two rectangular tiles are stacked. For all tests the
was decided to document each test with the support of high Load vs. Displacement curve at the tup of the impactor
speed camera. This technique allowed the determination were recorded providing quantitative data for the JH-2
of the exact time and location of fragments formation. model parameters calibration. In Figure 3 the response
Taylor tests were performed at the design velocity impact obtained in one test configuration is given showing the
of ∼100 m/s. Since complete rupture of the material occurs two behaviors (high and low peak load) that the material
at that velocity, slower impact tests were performed in exhibited in all tests.
order to determine the threshold impact velocity for which
no failure occurs. The nominal diameter of the cylinders
was 19 mm with a length ranging between 90 and 167ṁm. 3.1 JH-2 model parameters optimization
A total of 21 tests were performed with impact velocity
ranging between 83.0 and 3.0 m/s using two different All experimental configurations were simulated by means
facilities: a gas gun for the higher velocities and a drop of FEM. Initially, inverse calibration technique was used
weight facility, consisting in a vertical launch pipe with a for the drop weight test configurations. Since for all these
diameter of 22 mm, for the lower velocities. In Figures 1 type of tests the strain rate is similar, the strain rate

04011-p.2
 DYMAT 2012

7000 5000
Experiment 1
6000
Experiment 2 4000
5000 Experiment 3
Load (N)

3000

Load (N)
4000

3000 2000
2000
1000 Experiment 1
1000 Experiment 2
Numeric
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.2 0.4 0.6 0.8
Displacement (mm) Displacement (mm)
Fig. 3. Load vs. Displacement curve for Drop weight test on Fig. 5. Comparison between numerical and experimental results
rectangular tile at 1.5 m/s. for drop weight test on rectangular stacked tiles at 1.1 m/s.

7000
Experiment 1
6000 Experiment 2
Experiment 3
5000 Numeric
Load (N)

4000

3000

2000

1000

0
0 0.1 0.2 0.3 0.4 0.5 0.6
Displacement (mm)
Fig. 4. Comparison between numerical and experimental results
for drop weight test on rectangular tile at 1.5 m/s.

parameter C cannot be determined at this time. All other

Fig. 6. Qualitative comparison of damage pattern and distribution
parameters were optimized minimizing the following ob-
in Taylor impact at 83.3 m/s for two time frames. In the second
jective functions: frame the destroyed material near to impact surface is graphically
obj1 the square root of the summation of the differences removed by the erosion algorithm.
between the calculated and the experimental Load
vs. Displacement curve (indicated as ErrMin);
obj2 the difference between the numerical and the exper- from each other. However, this was a necessary step in
imental values of force peak (Fdelta); order to reduce the range and the step of the variables for
obj3 the difference between the times at which the force the second refinement phase of the task.
peak is reached in the computational and experi- Successively a new optimization was performed ac-
mental curves (T delta); counting for all configurations at the same time. The
obj4 the difference between the integrals of the com- analysis (with 20 objective functions) resulted in 1275
putational and experimental load vs. displacement finite element simulations, with a total of 1120h CPU time,
curve, i.e. the deformation work (IntMin). that led to the optimized material parameters set whose
The optimization task was carried out in two phases. First, high-quality performances were demonstrated by the com-
the objectives were minimized separately for each of the parison between experimental and numerical results shown
5 configurations using large ranges and large steps for the in Figures 4 and 5.
varying variables; the mean value for each parameter was Finally, the Taylor impact test configuration was simu-
chosen based on data available for similar material, [2]. A lated using the set of parameters determined for calibrating
DoE using Sobol algorithm with 38 designs was generated, the strain rate sensitivity parameter C. In this phase the
[3]. The multiobjective genetic algorithm NSGA-II, [4], time resolved fragmented length (extracted from the video-
was used for the optimization with 19 generations for a recording and shown for comparison in Figure 6) as a
total of 722 designs for each of the 5 configurations. The 5 function of the impact velocity was used as target function
data sets found in this way resulted to slightly differ one for the determination of C.

04011-p.3
 EPJ Web of Conferences

30.0 µs after impact. Fracture occurs as a result of two main

cracks running, orthogonally one to each other, from side
to side of the tile, Figure 7. The comparison of numerical
and experimental results confirmed the accuracy of the
calibrated JH-2 model parameters in predicting failure
occurrence under dynamic conditions different than those
used in the structural optimization process.

4 Conclusion
In this work the determination of the JH-2 mode parame-
Fig. 7. Qualitative comparison of damage pattern and distribution ters for fused silica was performed using both quantitative
in the sphere impact test at 30 m/s. and qualitative data obtained on two different dynamic
test configurations. Inverse calibration technique, based on
extensive numerical simulation of both type of tests, was
3.2 JH-2 model parameters validation implemented. In order to determine the optimum set for
the material model parameters multi-objective optimiza-
The determined JH-2 parameters set was verified planning tion was used. Finally, an independent validation test was
an independent test with a strain rate of ∼103 s−1 , which planned and performed. Results seem to indicate that the
is in between that of Drop weight and Taylor test char- determined material parameters for the JH-2 model can
acteristic strain rate values. This test consists in a steel be confidently used for strain rate ranging from 102 to
sphere 6.3 mm in diameter launched against a fused silica 104 s−1 .
tile supported by a rigid plate. Scope of the test is to predict
the critical velocity for which failure of the tile occurs and
the resulting failure mode. References
Running a parametric FEM analysis, it was found that
brittle fracture occurs at 30.0 m/s, while below 20.0 m/s 1. Johnson G. R. and Holmquist T. J., High-Pressure
no visible damage is expected to occurs. Based on this, Science and Technology (American Institute of Physics,
two impact tests, respectively at 20.0 and 30.0 m/s, were 1994)
planned and performed using a 6.5 mm gas-gun. Fused 2. Cronin D. S., Bui K., Kaufmann C., McIntosh
silica tile behavior was again monitored using high speed G. and Berstad T., Proceedings of the 4th
camera. As expected, failure in brittle silica is found to European LS-DYNA Users Conference, (2003)
be controlled by the superposition of tensile stress waves, pp.47-60
generated by reflection of the compression waves at free 3. Bratley P. and Fox B. L., ACM Transactions on Mathe-
surfaces. matical Software, 14-1, (1988) 88-100
As predicted by FEM, high speed video recording 4. Deb K., Pratap A., Agarwal S. and Meyarivan T.,
confirmed that brittle failure starts to occur approximately KanGAL Report Number 2000001, (2000)

04011-p.4