rspa.royalsocietypublishing.

org

Research
Cite this article: Chakraborty S, Shaw A,
Banerjee B. 2015 An axisymmetric model for
Taylor impact test and estimation of metal
plasticity. Proc. R. Soc. A 471: 20140556.
http://dx.doi.org/10.1098/rspa.2014.0556
Received: 21 July 2014
Accepted: 17 November 2014

Subject Areas:
Structural engineering, mechanical
engineering
Keywords:
Taylor impact test, radial bulging, strain
hardening, estimation of metal plasticity
Author for correspondence:
Amit Shaw
e-mail: abshaw@civil.iitkgp.ernet.in

An axisymmetric model for

Taylor impact test and
estimation of metal plasticity
Sukanta Chakraborty, Amit Shaw and
Biswanath Banerjee
Department of Civil Engineering, Indian Institute of Technology,
Kharagpur, West Bengal, India
The impact of a at-ended cylindrical rod onto a
rigid stationary anvil, often known as the Taylor
impact test, is studied. An axisymmetric model is
developed to capture the deformation behaviour of
the rod after impact. The most distinctive feature of
the proposed model is that it takes into account the
spatial and temporal variation of both longitudinal
and radial deformation and consequently the strains
and strain rates. The nal deformed shapes and time
histories of different eld variables, as obtained from
the model, are found to be in good agreement with
corresponding experimental and numerical results
reported in the literature. The proposed model is
then used to formulate an inverse framework to
estimate the JohnsonCook constitutive parameters.
In the inverse formulation, the objective function
is constructed using the nal deformed length and
diameter at the impact end of the retrieved rod.
Finally, the potential of the proposed model in
estimating material parameters is illustrated through
some examples.

1. Introduction
The impact of a cylindrical deformable rod on a rigid
stationary surface is a classical problem in impact
mechanics and has been the subject of interest for many
years to both experimentalists and numerical analysts.
This problem is frequently referred to as the Taylor
impact problem in view of the pioneering work by Taylor
[1]. His ndings with those of Whiffen [2], Carrington &
Gayler [3], which curved an elegant approach to
estimate the plastic ow stress, still bears signicance
as a base line for understanding physical principles
behind ductile metal behaviour under high strain
rate. Their analysis uses data from recovered samples
2014 The Author(s) Published by the Royal Society. All rights reserved.

2
...................................................

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

after several experiments on a at-ended cylindrical rod impacting on a rigid massive anvil. The
simplistic mathematical foundation made the test a reference point thereafter to estimate dynamic
yield strength, as a function of density, impact velocity, original and nal length, and length of the
un-deformed region of the retrieved specimen and also as a validation test case for constitutive
model development [4,5].
Since then, several attempts have been made to develop correlation between rate-dependent
material constants at a high strain rate and the deformation characteristic of the Taylor specimen.
Material constants were determined by feeding the experimentally measured deformation
parameters (nal deformed length, length of plastic zone, etc.) into the developed relation.
Hawkyard et al. [6] equated kinetic energy in the impact event with the plastic work and obtained
the mean dynamic yield strength of copper and mild steel. Therein a direction to determine
static yield stress was also proposed. In his further study [7], Hawkyard simplied the idea
of Taylors theory by replacing the momentum equilibrium by an energy equilibrium equation
across the plastic front. The strain hardening effect was also considered. Therein it was shown
that the derived model exhibits a closer proximity to experimental observations when compared
with Taylors theory. Jones et al. [8] modied Taylors equation of motion for the un-deformed
portion of the rod by introducing a term relating to mass loss from the un-deformed part to
the plastic zone. In [9], a method was presented to accommodate a more complex material
description including the strain-hardening effect. It was assumed that the material on the impact
side of the rigid-plastic interface is brought instantaneously to rest for low-velocity impact. After
logical investigation and careful insight into the plastic front propagation, Jones et al. [10] tried to
avert one very uncertain measurement of nal un-deformed section length in the Taylor test by
replacing it with a position measurement corresponding to a particular value of strain. In their
subsequent work [11], an elementary theory of the Taylor impact test was developed wherein
the state of stress in the cylindrical rod at high strain rate was obtained by establishing a linear
relationship between the length of the rigid part and the nal deformed length.
Being one-dimensional in nature, the above-mentioned models are unable to accurately
capture the three-dimensional deformation behaviour of the impacting rod, especially at the
impact end. Moreover, while determining the material constants, these models rely on a nal
un-deformed section length, which is one of the most unreliable measurements in the Taylor
test [11,12]. Though some of the uncertainty associated with the un-deformed section length
was removed in [10], it is desirable to have a correlation between the material constants and
the deformation parameters that can be measured accurately in the Taylor test. Two such
measurements are the nal deformed length of the rod and deformed diameter at the impact
end. This motivated the present study.
In this paper, an axisymmetric model for Taylor impact is derived. The plastic zone at the
impact end is approximated by the frustum of a cone. Longitudinal elastic wave propagation
and the follow-up by a more severe plastic front are considered. The condition of incompressible
plastic deformation is made use of in order to derive expression for longitudinal and transverse
velocity in the plastic zone. The procedure to incorporate strain-hardening and thermal softening
effect is discussed. Strain rates are computed over each time-step to obtain better accuracy, unlike
the commonly taken [1,11] average value over the entire process. The accuracy of the proposed
model in capturing the deformation behaviour of the Taylor rod is demonstrated by comparing
the prediction with some experimental and numerical results reported in the literature.
Next the potential of the proposed model in estimating JohnsonCook (JC) material
parameters [13] is explored. The distinctive effects of strain hardening, strain-rate and thermal
softening in the J-C model were estimated by earlier researchers using a combination of pseudostatic, split-Hopkinson bar and torsion tests at ambient and elevated temperatures. But noting
the unique strain and strain rate variations along the length of the Taylor specimen over time,
Johnson & Holmquist [14] devised an empirical method under bounded strain and strain rate.
Batra & Kim [15] tried to characterize parameters for metals plastic ow by recursive trials
on reconstruction of an experimentally obtained stressstrain curve. In the present study, J-C
parameters are determined through an inverse formulation of the proposed model based on

(a)

(b)

l0

D0
q = tan1

Dx
D D0
= tan1 b
2s
2

Figure 1. Assumed deformed shape for the Taylor impact test. (a) Experimentally obtained deformed configuration [17] and
(b) axisymmetric profile.

box-constrained nonlinear optimization. The nal deformed length and diameter at the impact
end of the rod are used to construct the objective function. This is unlike the work by Ozel
& Karpat [16], wherein the objective function is constructed through a stress function which is
experimentally expensive to obtain. Estimated J-C parameters for 4340 steel via the proposed
model are found to be in good accordance with their reference values.
The paper is organized as follows. The model is derived and its salient features are discussed
in 2. In 3, the developed model is validated through some experimental and numerical test
cases. The inverse analysis based on the proposed model is shown in 4. Finally, conclusions are
drawn in 5.

2. Theory
When a rod-shaped specimen impacts a rigid stationary anvil an elastic compressive wave
generates at the impact interface and travels through the rod in longitudinal direction with a
speed equal to the sound speed in the medium. When this compressive wave reaches the free
trailing end, it is reected as a tensile wave causing deceleration of the rod. For a sufciently
severe shock, i.e. once the magnitude of the compressive wave reaches the ow stress, the impact
end undergoes plastic deformation. The plastic front with maximum stress magnitude equal to
the ow stress starts propagating from the impact interface at a much lower speed. The travelling
plastic front interacts with the reected precursor elastic wave at some intermediate point along
the length of the rod. Elastic wave then get reected at the elasticplastic interface and travels
towards the rare end of the rod. This back and forth movement of the elastic wave causes
deceleration of the rod.
A typical deformed shape of the Taylor specimen obtained through experiments [17] is shown
in gure 1a. The entire rod may be divided into two parts separated by an elasticplastic interface:
the mushroom zone near the impact end where plastic deformation takes place and the undeformed zone at the rear end, which is characterized by back and forth movement of the elastic
wave. In the present model, the mushroom zone is approximated by a frustum of a cone which is
attached to a cylindrical part representing the un-deformed zone as shown in gure 1a. Density,
Youngs modulus, Poissons ratio and dynamic yield strength of the rod are denoted as , E,
and Yd , respectively. The initial velocity of the rod is v0 .
Owing to symmetry with respect to the longitudinal axis, the three-dimensional model may
be reduced to an axisymmetric model with l0 and D0 as the initial length and diameter of the
rod, respectively, as shown in gure 1b. Let at any instant of time, h be the axial shortening, l be
the extent of the un-deformed part, s be the extent of the plastic part and D be the diameter at
the impact end of the rod, all being time-dependent entities. So overall length L at any instant of

...................................................

s
h

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

D0

In the subsequent subsections, a mathematical model which relates the material constitutive
(such as, , E, , Yd along with strain hardening and thermal softening parameters) with the
geometrical parameters (l0 , D0 , h, l, s, D), especially the mushrooming at the impact end, is
derived. First, a kinematically admissible velocity eld is assumed and an incompressibility
condition is invoked in order to arrive at the longitudinal and radial velocity distribution in
the mushrooming zone (2a,b). The length of the un-deformed part of the rod is estimated by
considering axial shortening (h) and plastic front movement in 2c. Finally in 2d, an expression
for overall deceleration is derived by computing the change in velocity in the un-deformed part of
the rod after a round trip of stress wave. To incorporate the important notion of strain hardening,
the current state of strain is obtained via the spatial gradients (averaging spatially over plastic
zone but preserving the temporal variation) of the velocity components in 2e. A critical aspect of
implementing the derived model is discussed in 2f.

(a) Velocity profiles

Consider the rod is axisymmetric with the axis in x-direction (gure 1b). Let vx and vr be
the velocity in the longitudinal and radial directions, respectively. The longitudinal velocity is
assumed to remain uniform spatially in the un-deformed part of the rod. It gradually decreases
from the plastic front and becomes zero at the impact interface. This spatial distribution of the
longitudinal velocity has been found in good accordance with the numerical simulations through
ABAQUS [19] and simulations in [18] (gure 2).
Since there is no transverse ow occurring, the radial velocity is zero in the un-deformed part
of the rod. In the plastic region, the radial velocity, which causes the transverse deformation
(bulging), is symmetric with respect to the longitudinal axis (axisymmetric assumption).

...................................................

Deformation due to elastic wave propagation is small when compared with its plastic
counterpart and therefore neglected. Consequently, the trailing end remains un-deformed
and the mushrooming zone deforms plastically.
No unloading takes place in the mushrooming zone and the impact end of the rod always
remain in contact with the anvil.
The plastic deformation in the mushrooming zone remains incompressible (no signicant
change in material density except local oscillations to allow stress wave propagation) over
the entire impact process.
Carrington & Gayler [3] showed through changes in micro-structure in the stressed
zone that stress wave travels with a non-planar front. However, within the present
mathematical framework in order to obtain integrated volume of the temporally varying
shape and spatial extent of the plastic zone, the plastic front is assumed to be moving
in a plane surface to the longitudinal direction. This assumption implicitly implies that
the integrated volume would be equivalent to the volume of the plastic zone with a
non-planar front.
When the reected precursor elastic wave interacts with the plastic front, ideally it gets
reected partially from the elasticplastic interface and transmitted partially to reach the
impacting interface [18]. However, in the present formulation it is assumed that the elastic
wave completely reects back from the plastic front, while the plastic wave continues
propagating towards the rear end.
Impact velocity is less than speed of the plastic front and also in a range such that the rod
does not experience any fracture or material separation during the deformation process
(limiting our focus to capturing of dynamic plastic behaviour only).

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

time is given by L = l + s = l0 h. The initial (t = 0) and nal (t = tf ) conditions of (L, l, h, s, D) are

(l0 , l0 , 0, 0, D0 ) and (Lf , lf , hf , sf , Df ), respectively.
The present development is based on some assumptions, required to preserve the conciseness
of the model, as mentioned below.

0.5

x
x1

plastic
zone

assumed profile
profile from ABAQUS
0

0.25

0.50

0.75

1.00

relative distance from trailing edge x1/lf

Figure 2. Assumed longitudinal velocity distribution. (Exact functional form of vx is derived in equation (2.14) of 2a.)

Therefore, the boundary conditions on velocity proles may be summarized (with known l, s
and velocity of the rod v at a particular time instant) as follows. Longitudinal velocity is v all over
the trailing un-deformed part, i.e.


D0 D0
(2.1)
vx (x, r) = v, x [l, 0], r ,
2 2
and is zero at the impact surface, i.e.


D D
.
at x = s, r ,
2 2


vx (x, r) = 0,

The radial velocity is zero all over the trailing un-deformed part, i.e.


D0 D0
vr (x, r) = 0, x [l, 0], r ,
2 2

(2.2)

(2.3)

and is also zero along the centroidal axis in the plastic zone, i.e.
vr (x, r) = 0,

x [0, s], r = 0.

(2.4)

Let Dx be the diameter of the rod at a distance x from the plastic front (gure 1b) towards the
impact surface. To begin with, for the radial velocity at any point {x, r} {[0, s] [Dx /2, Dx /2]}
a complete quadratic polynomial as given in equation (2.5) is assumed. The dimensionless
entities x/s and r/Dx are chosen as the independent arguments of the polynomial. The unknown
constants, i may be determined by satisfying equations (2.1)(2.4)
vr
x
r
x r
x2
r2
= 0 + 1 + 2
+ 3
+ 4 2 + 5 2 .
v
s
Dx
s Dx
s
Dx

(2.5)

Substituting equations (2.3) and (2.4) into equation (2.5), one may have 0 = 1 = 2 = 4 = 5 = 0.
Therefore, the expression for vr may be reduced to vr = 3 (x/s)(r/Dx )v. As for the assumed prole,
radial velocity vr , which causes the bulging at the impact end varies linearly with x, Dx may be
written as


D D0
D D0
x = D0 + x where, =
.
(2.6)
Dx = D0 +
s
s
Finally,
vr = 3

r
x
v ( Dx = D0 + x).
s D0 + x

(2.7)

As mentioned earlier, the plastic deformation in the mushrooming zone is assumed to

be incompressible. In axisymmetric formulation, the condition for incompressibility (zero

...................................................

vx = f (v, x, s, D)

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

relative velocity ux /u

1.0

divergence of the velocity eld) may be approximated as

and
vx = 2

3 v
s


D0 + x D0

ln(D
+
x)
+ f (r),
0
2
2

(2.8)

(2.9)

(2.10)

where, f (r) is the integration constant independent of x.

Satisfying equation (2.1) (i.e. vx = v at x = 0), f (r) may be obtained as
f (r) = v +

23 v
D0 [1 ln D0 ].
2s

Then, substituting f (r), equation (2.10) may be re-written as




23 v
D0 + x
.
x D0 ln
vx = v 2
D0
s

(2.11)

(2.12)

Next satisfying equation (2.2) (i.e. vx = 0 @ x = s), 3 may be obtained as

3 =

2s
.
2[(D D0 ) D0 ln(D/D0 )]

(2.13)

Finally, substituting 3 from equation (2.13) into equations (2.12) and (2.7), the longitudinal and
radial velocity prole in the plastic zone (with, = D/D0 ) may be, respectively, obtained as



v
D0 + x
x D0 ln
(2.14)
vx = v
D0 ( 1 ln )
D0
and
vr =

2 vxr
.
2D0 (D0 + x)( 1 ln )

(2.15)

Here, a point to note is that the local proles of longitudinal and radial velocity components
(equations (2.14) and (2.15)) in the mushrooming zone are clearly functions of the velocity (v)
of the un-deformed part. Though not explicit in the above equations, the velocity distributions
will also depend on the other obvious inuencing material constitutive parameters, because the
current diameter, D (and, hence, and explicitly appearing in the equations), is a direct outcome
governed by those material properties as will be seen in the subsequent subsections.

(b) Mushrooming at impact end

The radial velocity at (x = s, r = 0.5D) represents the rate of change of the radius at the impact
end. Therefore,
d(radius)
2 vs(0.5D)
1 v( 1)
=
=
,
dt
2D0 (D0 + s)( 1 ln ) 4 ( 1 ln )

(2.16)

where = D/D0 , = (D D0 )/s. Finally, the rate of change of the diameter at the impact end,
currently known as the bulging rate, may be obtained as
d(radius) 1 v( 1)
dD
=2
=
.
dt
dt
2 ( 1 ln )

(2.17)

...................................................

Using equation (2.7) in equation (2.8), one may get



vr
vr
v
x
vx
=
+
= 23
x
r
r
s D0 + x

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

vr
vr
vx
+
+
= 0.
r
r
x

(c) Length of un-deformed part and axial shortening

dh
= vx |x=0 vx |x=s = v.
dt

(2.19)

The length of the plastic zone increases with movement of the plastic front and decreases due to
axial shortening in the plastic zone. Therefore, the rate of change of the length of the plastic zone
may be written as
d
ds
= Cp (h) = Cp v.
(2.20)
dt
dt
To have meaningful positive evolution of plastic zone s, the model is only considered to be valid
when v < Cp .

(d) Deceleration model

Deceleration of the un-deformed part is caused by the repeated back and forth travelling of

the longitudinal elastic wave with front speed Cl = ( E/). From the one-dimensional wave
propagation theory, the relation between the magnitude of stress wave and the longitudinal
velocity at any material point yields v = Cl /E. As the maximum stress magnitude in the plastic
medium can at most reach the ow stress value, is replaced by Yd , safely ignoring the initial
magnitude build-up in a very quick time-span. Then from [18], the change in particle velocity
over one round trip of the shock-wave may be written as
 


Yd
Yd
Yd
Cl
= 2Cl .
(2.21)
v = Cl
E
E
E
Now, let t1 be the time taken by the elastic compressive wave to travel from the plastic front to
the free end. Therefore,
Cl t1 = l.

(2.22)

The movement of the plastic front towards the free end results in reduction of the un-deformed
zone, and therefore when returning the distance travelled by the reected tensile wave is less. If
t2 be the time taken by the reected tensile wave to travel from the free end to the plastic front
one may write
 t1 + t2  
dl
dt.
(2.23)
Cl t2 = l +
dt
0
Combining equations (2.22) and (2.23), one may have

 t   
dl
Cl t = l + l +
dt ,
dt
0

(2.24)

where t = t1 + t2 is the total time for the round trip of the wave. Substituting equations (2.18),
equation (2.24) may be recast as
 t
Cl t = 2l
Cp dt = 2l Cp t.
(2.25)
0

...................................................

Now, as shown in gure 2, the longitudinal velocity remains uniform (= v) in the un-deformed
part and hence no axial shortening takes place there. In the plastic zone, the longitudinal velocity
varies according to equation (2.14). Therefore, the rate of axial shortening which takes place only
in the plastic zone may be denoted by the velocity difference at either end of the plastic zone as

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

The length of the un-deformed part of the rod reduces due to the movement of the plastic front
towards the free end. Therefore, the rate of change of length of the un-deformed part may be
related to the plastic front speed as
dl
(2.18)
= Cp .
dt

Cl (Cl + Cp )Yd .
dv
v
2Cl Yd /E
 lim
= lim
=
dt t0 t t0 2l/(Cl + Cp )
lE

(2.27)

It is shown in appendix A that this nal form conserves the linear momentum.

(e) Thermo-mechanical behaviour

In many practical cases, the material may not behave as perfectly plastic and the ow stress
may signicantly depend on plastic strain and strain rate. Furthermore, in a highly dynamic
event such as this Taylor impact case, the material may even undergo thermal melting.
The need of incorporating thermo-mechanical strain-rate sensitivity and subsequent material
hardening/softening in Taylors impact problem was presented by Woodword et al. [20].
The procedure to incorporate such a constitutive law in the proposed model is discussed in
this section. Towards this, two commonly used material models, viz. J-C model [13] and the
dislocation-based ZerilliArmstrong (Z-A) model for body-centred cubic (BCC) slip [21] as given,
respectively, in equations (2.28) and (2.29), are considered.

  

 

pl
T Tr m
n
1
(2.28)
Yd = (A + B
pl ) 1 + C ln

0
T m Tr
and

Yd = C0 + C1 exp(C3 T + C4 T log
pl ) + C5 (
pl )n ,

(2.29)

where (A B n C m) and (C0 C1 C3 C4 C5 n) are material parameters for the J-C and Z-A
models, respectively. In the J-C model (equation (2.28)),
0 is the reference strain at which A, B,
n are characterized, Tr is the ambient temperature and Tm is the melting temperature. Now the
temporal variation of plastic strain
pl , plastic strain rate
pl and temperature T are computed as
follows.
Knowing the spatial distribution of the longitudinal and the radial velocity (equations (2.14)
and (2.15), respectively) in the plastic zone, components of strain rate may be obtained as

xx =

2 vx
vx
=
,
x
D0 ( 1 ln )(D0 + x)

vr vr
2 vx
=
=
=
r
r
2D0 ( 1 ln )(D0 + x)


vr
2 vr
1 vx

xr =
+
=
.
2 r
x
2( 1 ln )(D0 + x)2

rr =

and

(2.30)
(2.31)
(2.32)

Ignoring the initial rapid elastic build-up, these strain rates actually denote the plastic
components of the strain tensor effective within the plastic zone. Though the components of the
strain rate are spatial function of (x, r), an average strain rate value computed at the centre of the
plastic zone is used for calculating effective plastic strain. Thus substituting (x = 0.5s, r = 0) in
equations (2.30)(2.32), strain rate values at central point of the plastic zone are obtained as

xx =

rr =
=
and

2
dD
v( 1)
=
,
D0 ( + 1)( 1 ln )
D + D0 dt

1
v( 1)
=
xx
2D0 ( + 1)( 1 ln ) 2

xr = 0.

(2.33)
(2.34)
(2.35)

...................................................

Now from equations (2.21) and (2.26), with t 0 in the un-deformed part of the rod

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

Here, Cp is assumed to be invariant within the small time-frame of t. So, t may be nally
expressed as
2l
.
(2.26)
t =
Cl + Cp

For the given spatial distribution of strains (equations (2.30)(2.32)), the expression of effective
strain rate at (x = 0.5s, r = 0) is found to be coincident with the volume averaged (over the whole
mushrooming zone, the size varying with time) effective rate value. Then the effective plastic
strain
pl may be obtained by integrating this above rate over time. Finally, the computed effective
plastic strain and strain rate are used in equations (2.28) and (2.29).
Next the remaining required input variable is current temperature T . Assuming an adiabatic
condition, the temperature evolution from ambient value is computed from the rate of heat
generation as
:
pl
Yd .
pl
dT
=
=
,
(2.37)
dt
Cp
Cp
where, Taylor-Quinney [22] empirical constant denotes the fraction of plastic work converted
into heat (normally taken as 0.9) and , Cp denote current density and specic heat density at
constant pressure, respectively. The numerator indicates the current state of stress bounded by

a closed convex set of admissible stresses E = int(E ) E = {(Yd ) S | f := ( 3J2 Yd ) 0},
where J2 is the second invariant of deviatoric stress components. So at yield surface E , during
the plastic ow, the plastic work is computed as (Yd .
pl ).

(f) Corrections for bulging rate

The changing diameter along with changing plastic front speed (Cp ) as a result of strain hardening
make it essential to integrate the governing ODEs (equation (a)(f) in table 1), numerically with
respect to time. Now the changing diameter inuences [= (D D0 )/s] and [= D/D0 ] directly.
As shown in gure 1b, is actually the measure of the angle created due to bulging from the
original rectangular projected shape of the plastic zone. Therefore, ideally increases from zero
at the time of impact and reaches a nal stable value f [= (Df D0 )/sf ] when impact is ceased.
However, the presence of s in the denominator of may result in instability in computation,
especially during the initial phase of the impact, when s is very small but increasing very rapidly.
In order to avoid such instability, an average value av = 0.5f is reached in the following way.
First equations (2.17)(2.19), (2.27), (2.36) and (2.37) are integrated in time with assumed av =
0.5 [or f = 1] and the nal un-deformed length lf and the nal length of the plastic zone sf are
obtained. Next incompressibility condition is imposed to estimate av equating initial and nal
volume as

2
D l0 = D20 lf + sf (D20 + D2f + D0 Df ).
(2.38)
4 0
4
12
Solving equation (2.38), the value of Df and that of f are obtained as

 

12
D0
Df =
(l0 lf ) 3 1
2
sf
and
1 Df D0 D0
av =
=
2
sf
4sf

 


12
(l0 lf ) 3 3 .
sf

(2.39)

(2.40)

Computed av as per equation (2.40) after the rst run is then used and another set of timeintegrations is performed to obtain the actual response time-history and nal deformed shape.
For the better understanding of the proposed model, the governing equations are summarized
and implementation steps are given in table 1. In all steps, time integration is to be performed
till the velocity of the rear end becomes zero and the entire rod comes to rest. For a given
material, the speed of plastic front Cp may
 be computed from the square root of the density
averaged slope of the stressstrain curve ( ( /
)/). However, as in the present case, the plastic

...................................................

(2.36)

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

Now the effective plastic strain rate may be computed as

2
2 2
dD
2
2 +
2 + 2
2 ) = |

pl =

xr

xx | =

=
(
xx +
rr
.

3
3
D + D0 dt

Table 1. Equation summary and implementation steps.

10

(b)

dh
dt

=v

(c)

dv
dt

C (C +C )Y
= l l lE p d

D0

(d) dD
=
dt

0.5v( 1)
1ln

d
pl
dt

2 dD
D+D0 dt

(e)

l0

d pl
(f) dT
= C
dt
p

s
r

D0
q = tan1

Dx
D D0
= tan1 b
2s
2

[ = DD0 , Cl : elastic wave speed, Cp : plastic front speed and Cp : Sp. heat]
Yd is updated @ each t as
(g) Yd = f (
pl ,
pl , T )

Note: f (
pl ,
pl , T ) may be
different for different

and accordingly Cp is also updated

implementation steps

constitutive model

determination of av
integrate equations (a)(g) until v = 0 with assumed = 0.5




12
(h) av = 4sD0f
(l

l
)

3
f
sf 0
computation of time-histories and final deformed shape
integrate equations (a)(g) with = av from equation (g).

front is compressive in nature and moves as a plane in the longitudinal direction; the simplied

expression Cp = Yd / as given in [23] is taken.

3. Validation of the proposed model

The model developed in the previous section is validated through some experimental and
numerical results reported in the literature. A series of experiments on a cylindrical rod made of
4340 steel impacting a rigid stationary anvil with different impact velocities was conducted in [12].
Later hydro-code (MSPH) simulations of similar impact processes were performed in [24]. The
proposed model is rst validated with these test cases (table 2a). To start with, the J-C constitutive
behaviour (equation (2.28)) is incorporated, as was used in MSPH [24]. Corresponding J-C
parameters are given in table 2b.
Computed nal length of the rod and diameter of the impact face are compared with their
corresponding experimental values [12] in table 3. Numerically computed results through MSPH
[24] and ABAQUS simulations are also shown in the same table. Next the Z-A constitutive
behaviour (equation (2.29)) is also incorporated in the present model and corresponding
deformations are shown in table 3. The Z-A material parameters as taken from [16] are
given in 2c.
It can readily be seen that the predictions via the proposed model are in very good agreement
with their experimental and numerical counterparts. It is observed that the present set of
ODEs yields closer agreement with the experimental data when Z-A constitutive behaviour is
incorporated. The marginal deviations in all the cases are in comparable scales with the small

...................................................

(a) dtdl = Cp

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

equation summary

Table 2. Material parameters and test specimens for 4340 steel.

11

l0 (mm)

10

11

12

11.39

15.19

15.19

15.19

25.29

25.29

37.97

37.97

37.97

37.97

37.97

56.96

285

234

275

302

170

215

181

183

224

234

270

242

..........................................................................................................................................................................................................

v0 (m s )

(b) J-C material properties [24]

(kg m3 )
7830

(GPa)

K (GPa)

A (MPa)

B (MPa)

0 (s1 )

Tr (K)

Tm (K)

82.9

169.1

792

510

0.26

1.0

0.014

1.03

293

1293

(c) Z-A material properties [16]

C0 (MPa)

C1 (MPa)

C3 (K 1 )

C4 (K 1 )

C5 (MPa)

2073.6

0.0015

0.0000485

1029.4

0.531

89.9

..........................................................................................................................................................................................................

Table 3. Deformed geometry compared with experimental [12] and with MSPH [24], ABAQUS simulations results for the Taylor
impact test on 4340 steel (D0 = 7.595 mm).
Lf (mm)

Df (mm)
MSPH

ABAQUS

Sl. No.

test

9.20

9.60

10.97

13.10

13.30

12.4

model

(J-C)

MSPH

ABAQUS

model

model (Z-A)

test

(J-C)

model (Z-A)

9.14

9.50

10.9

10.3

9.24

12.88

11.97

13.08

13.1

13.48

10.2

10.1

10.41

11.32

10.72

12.8

12.47

12.37

12.83

11.2

10.7

11.29

12.55

11.72

12.0

12.5

12.05

11.84

12.35

12.2

11.1

11.90

13.55

12.53

23.4

23.3

23.20

23.36

23.75

9.2

9.5

9.47

9.91

9.54

22.4

22.5

22.21

22.29

22.85

10.3

10.4

10.42

10.86

10.35

34.6

34.6

35.50

34.67

35.34

9.5

10.0

9.76

10.13

9.73

34.7

34.6

34.37

34.60

35.28

9.8

10.0

9.80

10.17

9.77

33.0

33.3

32.95

33.09

33.98

10.5

11.0

10.72

11.09

10.56

10

33.1

32.9

32.58

32.68

33.62

10.6

11.2

10.97

11.35

10.78

11

31.3

32.0

31.19

31.09

32.19

12.1

11.8

11.90

12.44

11.70

12

47.8

48.7

48.30

48.48

49.92

11.6

11.6

11.18

11.60

11.01

..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................

measurement error in experiments and theoretical assumptions made to derive the present
mathematical formulation. This strengthens the claim of resolving the nal deformed shape
subjected to certain material parameters and initial conditions but with more signicant ease than
that of computationally exhaustive simulations.
Now, other than the nal geometry, the actual time-evolution of deformation in the Taylor
specimen with varying strain and strain rates must be validated before venturing condently
into predicting material constitutive behaviour [25]. So, to examine the time-resolved behaviour,
the temporal variation of the desired parameters (diameter, length and velocity) are investigated.
To this end, time histories for the diameter of the impact end, the deformed length and the

...................................................

Sl.No.

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

(a) test cases

39

8.5
7.5

L (mm)

MSPH (J-C)
ABAQUS (J-C)
present model (J-C)
present model (Z-A)

37
35
33

10

15 20
time (ms)

25

30

35

10

15 20
time (ms)

25

30

35

Figure 3. Time-histories of deformations obtained from the simulation and the present models.

200
MSPH (J-C)
ABAQUS (J-C)
present model (J-C)
present model (Z-A)

V (m s1)

150
100
50

10

15
20
time (ms)

25

30

35

Figure 4. Velocity time-history obtained from simulations and present model.

spatially averaged velocity of un-deformed part for 37.97 mm long rod with initial striking
velocity 181 m s1 are computed with the proposed model (with J-C and Z-A material models)
and compared with that of obtained via hydro-code simulation [24] and with that of FEM
simulations through ABAQUS [19], in gures 3 and 4.
It may be seen that the predicted time histories are also in very good agreement with their
numerical counterparts. The small discrepancies in the velocity time history (gure 4) are due
the fact that the MSPH result, as digitized from Batra & Zhang [24], indicates the velocity time
history of a point at the rear end of the rodso it is stepwise, while the velocity computed from
the proposed model is the average value over the un-deformed part of the rod and hence a
continuous smooth deceleration is obtained. Whereas the velocity time history obtained from
ABAQUS follows the MSPH trend, but with oscillations due to several reections of stress-waves
as the velocity prole is extracted on a single node at the trailing edge and on the axis of symmetry.
In the case of the Z-A model, the time-histories depict somewhat hardened behaviour as
compared to simulations with the J-C model, as was also observed by Forde et al. [26]. But,
investigating table 3, the Z-A model is seen to be closer to experimental observations, though with
a slight softening; i.e. the Z-A model-based deformation is between J-C model-based deformation
and experimental test data. In the absence of grain-size denition in the present macro-level
model, the HallPetch (H-P) relation of deformation twinning, as suggested by Armstrong [27] to
nullify this difference, could not be incorporated. However, for the test cases considered in this
study, it is observed that the deviation of plastic deformation with respect to their experimental
counterpart is in a negligible range. Thus, in the present context of a simplied approach with
macro-level plastic behaviour, these micro-mechanics-based ne details yielding nonlinearity in
a bulged prole were ignored and the conoid shape of a plastic zone with straight edges was
assumed, as depicted earlier in gure 1b.

...................................................

9.5

12

MSPH (J-C)
ABAQUS (J-C)
present model (J-C)
present model (Z-A)

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

D(mm)

10.5

12 000

13

(37.97 7.595 mm) @

pl
dxx
/dt

4000

4
3
(s-x) mm

pl

Figure 5. Spatial distribution of

xx .

The nonlinearity in the above time-histories is due to the treatment of variable strain rate
and corresponding strain hardening in place of assuming a constant strain rate as explained in
2e, generally in coherence with that of high-delity MSPH and ABAQUS simulations. These
temporal variation of strains and strain rates are important as they do not remain the same
throughout the impact event. When impact occurs, the impact end of the rod instantaneously
comes to rest and experiences a steep velocity gradient which results in high strain rates at
the impact end. Moreover, during the initial phase, since the plastic front has not yet moved
sufciently towards the rear end of the rod, the thickness of the plastic zone (s) is small, and
therefore the plastic deformation is localized over a smaller region. However, as time progresses,
owing to the increase in thickness of the plastic zone and deceleration of the rear end of the rod,
the velocity gradient over the plastic zone decreases, which results in a decrease in strain rate.
In order to illustrate whether the proposed model is able to capture this phenomenon, spatial
distribution of longitudinal strain along the rod axis at different time instants is shown in gure 5.
It is evident from the gure that the information regarding spatial and temporal variation of
strain ratesunique attributes related to the Taylor testcan be extracted from the present model.
Furthermore, their nonlinear distribution is found to be consistent with the physical phenomena
experimentally monitored using high-speed photography and VISAR [26].
Now for the considered tests case (i.e. 37.97 mm dia : 7.595 mm rod impacting at 181 m s1 )
the sensitivities of the parameters = (A B n C m) on Df and Lf are investigated and compared
with that of an axisymmetric model simulated through ABAQUS [19]. As evident from gure 6,
the sensitivity of different material parameters as obtained from the proposed model is found to
be coherent with that of ABAQUS. In the case of parameter m, the sensitivities on Df and Lf are
found to be much less both in ABAQUS and the present model and are thus not shown here.
For the further validation of the proposed model, next the predictions are compared with some
other test results given in [28]. The rod is composed of aluminium alloy 7A04-T6 (E = 69.35 GPa,
= 0.31 and = 2850 kg m3 ) with D0 = 12.62 mm. The material constitutive law to capture the
plastic behaviour is taken to be a modied J-C constitutive law with Voce empirical formula [29] as

  

 

pl
T Tr m
1
,
(3.1)
Yd = fpl 1 + C ln

0
T m Tr
where fpl is dened with A1 , A2 , t1 , u ,
u and w as the material parameters, given in table 4, as
follows:



A1

pl /t1 ],

A
+
A
[1

exp
if

<

u
1
2
pl

E
(3.2)
fpl =









pl
A1
A1

u w 1 +
pl
u +
+ (1 w)
.
, if
pl
u
E

u
E

...................................................

@ 1 ms
@ 5 ms
@ 10 ms
@ 15 ms
@ 20 ms
@ 25 ms

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

8000

181 m s1

% deviation in A

% deviation in deformation

10 10
50
50

25

50

25

% deviation in C

% deviation in deformation

5
50

25

25

% deviation in B

25

50

25

% deviation in n

25

50

5
50

Figure 6. Sensitivity of material parameters on deformed geometry: (Df , Lf ).

Table 4. Modified J-C parameters for 7A04-T6 Al alloy [28].

A1 (MPa)

u (MPa)

A2 (MPa)

t1

0 (s1 )

Tr (K)

Tm (K)

0.014

1.015

293

878

..........................................................................................................................................................................................................

602.5

681.5

0.0638

119

0.05

0.1

1.1 10

..........................................................................................................................................................................................................

Table 5. Experimental [28] and computed deformed geometry for 7A04-T6 Al alloy.
l0 (mm)

v0 (m s1 )

Lf,test (mm)

Lf,model (mm)

Df,test (mm)

Df,model (mm)

50.82

176.5

48.47

48.41

14.64

15.35

50.50

184.5

48.29

47.89

14.81

15.52

..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................

50.79

199.6

47.76

47.72

15.60

15.86

50.83

201.9

47.62

47.68

15.37

15.92

50.82

204.3

47.69

47.60

15.56

15.98

50.82

211.3

47.42

47.37

15.74

16.15

..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................

Comparisons of the computed nal length of the rod and the diameter of the impact face with
their corresponding experimental values are given in table 5. For the present focus of capturing
the plastic response only in a simplistic form, the velocities considered are such that they will not
initiate material damage or fracture. A very good agreement between the experimental values
and the prediction via the proposed model is again observed.

4. Parameter estimation via an inverse analysis

If Taylors original idea [1] is to be followed, the average dynamic yield limit can be estimated by
feeding the deformed length parameters (lf and Lf ) in the following Taylor formula as
Yd =

v02 l0 lf
1
.
2 l0 Lf ln(l0 /lf )

(4.1)

In addition to the problem in precise determination of lf the major limitation of the mathematical
model that leads to equation (4.1) is that it fails to capture the thermo-mechanical behaviour of

...................................................

25

14

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

50

10

10

Lf: ABAQUS
Lf: model
Df: ABAQUS
Df: model

()

()

where, Df,i and Lf,i are, respectively, the nal diameter and overall length (= l0 hf ) of the rod
for ith test situation. Superscripts m and c denote the measured data from experiments and
computed data from the governing forward equations, respectively (table 1). Each test case here
represents a particular set of initial diameter, length and impact velocity. = (A B n C m)T denotes
the decision variable in the minimization problem. It is to be noted that the objective functional
is constructed using nal total length and diameter at the impact end and thus the uncertainties
associated with measurement of the un-deformed part is avoided.
The inverse problem is cast as a bound constrained optimization problem as follows:
min {()| L U }

(subject to equation (a)(h) in table 1),

(4.3)

where, U and L denotes the upper and lower bounds, respectively, of the decision variable. The
optimization is performed via a standard gradient based quasi-Newton method [30]. The gradient
of the cost functional (equation (4.2)) is computed via rst-order optimality criteria as
 c  c
 c  c

Lf,i (Lf,i Lm
Df,i (Df,i Dm

f,i )
f,i )
=
+
,
(4.4)
g=
2
2

(Dm
(Lm
f,i )
f,i )
i

(Dcf,i /)

(Lcf,i /)

and
represents parameter sensitivities at the nal time for the ith test
where
case. These sensitivities are shown in gure 6 and some patterns for each of the parameters
were obtained. Following Taylors idea [1], these sensitivity patterns may be used to inversely
calculate the dynamic yield limit. But a more rational approach would be to incorporate accurate
mathematical sensitivity under an optimization framework. Thus to arrive at the nal value
of these sensitivities at the end of the time-evolution of D and L(= l0 h), a sensitivity based

...................................................

15

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

the material. Therefore, equation (4.1) is restricted only to material which exhibits rigid-perfect
plasticity. Since most of the materials do not behave as perfectly plastic especially at high strain
rate the Taylor test in its classical form is being used only for validation rather than a material
test. However, the beauty of Taylors work perhaps lies in the fact that it showed how an estimate
of dynamic plasticity can be made through a simple test, and therefore it still bears signicance
in the context of metal plasticity. Keeping this in mind, an improved mathematical model for the
Taylor test was derived in the previous sections. It was shown that the proposed model not only
predicts the nal deformed shape accurately, it also captures their time evolution reasonably well.
Now the main purpose of this section is to further explore the potential of the proposed model in
order to have better understanding of material behaviour especially at high strain rate.
To start with in order to draw the similitude between the proposed model and the Taylor
approach for perfectly plastic material, the differential equations ((a)(h) in table 1) are solved
with constant yield value. By parametric variation of the constant yield limit ((500 600 700 800 900
1000 1100) MPa), the deformed geometries (Lf and lf ) are obtained from the present model and
fed into equation (4.1). The obtained estimate ((485 573 676 780 885 990 1097) MPa) are indeed
very close to the assumed values. This shows that Taylors formula is a special case of the present
model if the yield value is assumed to be constant.
Now, as stated above, real materials do not behave as perfectly plastic and thus it is important
to understand the actual evolution of yield strength as a function of different thermo-mechanical
variables. Owing to nonlinearity in the equation, closed form expressions for different material
parameters are not possible. Therefore in this study, material parameters estimation is performed
through an inverse technique based on the proposed model. For demonstration purposes,
estimations of J-C constitutive parameters (A B n C m) are considered here. However, the strategy
may be extended to other material models such as the Z-A model or the MTS model.
The inverse problem is posed under an optimization framework with a standard least-squares
based objective functional as follows:
 c

m 2
2
(Lcf,i () Lm
1  (Df,i () Df,i )
f,i )
+
,
(4.2)
() =
2
2
2
(Dm
(Lm
f,i )
f,i )

(4.5)

where
y k

Ckp







Ckp Ckp
Ckp Ckp Ckp lk

2
1+

Cl
Cl

lk
lk

v k




 
=

Dk
1 (D0 /Dk ) ln(Dk /D0 )
v k Dk0

+
v D0

k
2
( k )2





k
k

(Dk0 )2
2D0

k ) v
k D

v
+
(D

k k
0

Dk k
(D0 )
D







Ydk
(Dk0 )2

Dk 2D0
v k

k
k
k
k

k k + (D0 )

v
+ Yd k

) k
k
Cp pl

D
(D
D
0
0

g :=

(4.6)


In equation (4.6), the superscript k denotes the kth time-step, Ckp / = (1/2 Ydk )(Ydk /) and

Ydk

k ][1 (T k )m ]
[1 + C ln
pl

k )n [1 + C ln
k ][1 (T k )m ]
(
pl
pl

k )n ln
k [1 + C ln
k ][1 (T k )m ]
B(
pl
pl
pl

k
n
k )m ] ln
k

)
][1

(T
[A
+
B(

pl
pl

[A + B(
k )n ][1 + C ln
k ](T k )m ln(T k )

pl

(4.7)

pl

k = Dk + D0 , k = Dk D0 D0 ln(Dk /D0 ) and T k = (T k Tr )/(Tm Tr ).

with Dk0 = Dk D0 , D
0
Starting with a zero initial value (i.e. y0 / = 0), equation (4.5) is solved along with the
governing equations (a)(g) of table 1 and the nal value of y/ is used to estimate Dcf /
and Lcf /. After obtaining sensitivities, material parameter is updated and the procedure
is repeated till the cost function is sufciently reduced. The iterations are performed via the
function fmincon available in commercial software package Matlab ([31]), wherein the cost
function (equation (4.2)) and gradient values (equation (4.4)) are supplied for obtaining material
parameter update.
Next the potential of the model along with the inverse formulation is illustrated through J-C
parameter estimation of 4340 steel. Density, shear and bulk modulus are assumed to be known
a priori as given in table 2. Length and diameter of the un-deformed projectile are taken from
l0 (20, 40) mm and D0 (10, 15) mm, respectively, with varying L/D ratio between 1.33 and 4.
Two velocity ranges 75115 m s1 and 230270 m s1 are taken for denoting low strain rate and
high strain rate, respectively, generating a total of 180 test cases altogether. Now in the absence of
real-life experimental observations for all these test data, two methods are employed to generate
m
Dm
f s and Lf s.
First, the proposed forward model itself (table 1) is used to generate the experimental data
for a known material, 4340 steel. Now there are three measurements to be made in each of the

...................................................

yk+1 yk
=
+ tg(yk , , tk )

16

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

differential equation is formed and solved recursively. Considering the time discretized form of
k T k }T , the discrete form of the sensitivity
the proposed forward model with yk = {lk v k hk Dk
pl
differential equation is as follows:

Table 6. Estimated material parameters for 4340 steel.

17

output
1184.2

J-C parameters
A (MPa)

reference values
792

without noise
823.5

with 1% noise
828.3

with 5% noise
904.8

B (MPa)

510

493.1

493.3

410.2

0.26

0.014

0.0121

0.0122

0.0102

0.0132

1.03

1.045

1.072

1.062

1.263

..........................................................................................................................................................................................................

596.2

..........................................................................................................................................................................................................

0.289

0.292

0.292

0.272

..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................

test casesnal length, deformed diameter at the impact end and initial impact velocity. Noting
the accuracy feasible in measuring geometric dimensions of the retrieved samples through Laser
transducers [32], the tolerance of velocity measurement does pose scope for pollution in the
experimental data. In order to handle such possibilities, a perturbation (i.e. measurement noise)
is added to the initial velocities for all test cases while simulating the inverse problem. Thus, data
pollution comes in terms of initial velocities which nally affect the computed geometry after
impact. The measurement noise is added as follows:
vnoisy = vexact [1 + (r)],

(4.8)

where is noise intensity and r denotes a random variable uniformly distributed in [1, 1].
Secondly, an axisymmetric nite-element model (FEM) is implemented in the ABAQUS [19]
software package with J-C plasticity for 4340 steel to generate synthetic data. Computed Dm
f and
lm
f through FE analysis are used as measurements in the inverse problem. It may be noted here that
the mathematical model used to generate measurements is essentially a numerical approximation
(via FEM) of the original problem and the proposed model used for material parameter estimation
is entirely different. This is in sharp contrast to the previous experiment, wherein the same model
is used for generating measurements and parameter estimation.
Thus, for the present inverse procedure, the model dependencies in estimating plasticity
parameters are completely avoided. In both cases, bounds on the decision variable are taken
as L = [0.5 109 0.3 109 0.2 0.1 101 0.1 101 ]T and U = [1.0 109 0.7 109 0.3 0.2
101 0.2 101 ]T . The mean of the upper and lower bounds is taken as the initial guess of the
material parameters to start the inverse problem. However, in test cases considered in the present
analysis, no signicant sensitivity of the initial guess to the reconstructed parameters is observed.
Table 6 shows reconstructed parameters for two cases. For the rst set of measurements 0%, 1%
and 5% (Cases I, II and III) noise intensities are considered. It is evident from the reconstruction
result that the inverse procedure can appropriately identify plasticity parameters with different
orders of magnitude.
It is also noted that the choice of proper bounds is very important for identifying parameters
of different magnitude which can be done a priori from a crude knowledge about the material
model to be identied. As expected, with the increase in data pollution the solution also deviates
from the true value. The order of deviation depends on the sensitivity of the decision variables
on the constructed objective function and thus affects the reconstructed solution. In the case
of measurement obtained from ABAQUS, the reconstruction results deviate from the reference
value especially for parameter A. As shown previously in gure 3, the nal length and diameter
obtained through ABAQUS do deviate a bit from the proposed model. This results deviation
(though not out of range) of the estimated values from the reference values when experimental
values are simulated through ABAQUS. However, as noted from table 6, inverse formulation with
the proposed axisymmetric model can reasonably identify the J-C parameters of 4340 steel. The
parameter estimation may further be improved with an efcient inverse algorithm [33,34].

...................................................

measured data (ABAQUS)

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

measured data (axisymmetric model)

(a) 1.2

(b)

18

1.5

0.9
0.8

0.05

0.10

0.15
0.20
pleff

0.25

0.30

Yd (GPa)

Yd (GPa)

known f
f w/o noise
f with 1% noise
f with 5% noise

1.0
known f
f from ABAQUS

0.5

0.05

0.10

0.15

0.20
pleff

0.25

0.30

Figure 7. Yd as obtained from inverse analysis. (a) Measurement generated from present model and (b) measurement
generated from ABAQUS simulations.

Finally, the estimated dynamic yield strengths of 4340 steel (rod of l0 = 37.97 mm,
D0 = 7.595 mm impacting with a velocity of 181 m s1 ) for different measurement data are
compared with their reference behaviour in gure 7. Obtained stressstrain behaviour as apparent
from gure 7 is adequately indicative of the robustness of the proposed algorithm in estimating
J-C plasticity parameters. As expected, measurements generated from ABAQUS dynamic yield
strengths do deviate from reference value. However, the characteristics remain the same with the
increase in effective plastic strain.

5. Closure
An axisymmetric model for the impact of a at-ended cylindrical rod on a rigid stationary surface
is developed in this study. Plastic deformation causing bulging at the impact end, propagation of
plastic front along the length of the rod, spatial and temporal variation of plastic strains and strain
rates constitute the basis of the model. The proposed model gives an implicit relationship between
material yield strength and deformation of the impacting rod in the form of a set of ODEs in time.
The derived model is based on some assumptions which are directly or indirectly related to the
impact velocity. Higher impact velocity may cause hydrodynamic effects (if, v > Cp ) or shock

propagation (if, v > K/, K being bulk modulus) or even fracture, which violate the premise
of the present model. On the other hand, if the impact velocity is very small such that elastic
deformation and unloading play a dominant role, the prediction may be affected since these
effects are neglected in the present analysis. It is important to note that the efcacy of the present
model has to be assessed in the context of Taylors impact test, where impact velocities are always
taken within a range which is different for different materials. In that range the present model
gives a reasonably good prediction. This is demonstrated by comparing the obtained estimate
with some experimental and numerical results reported in the literature. It is shown that the
proposed model is efcient in not only predicting the nal deformed shape of the rod but also in
capturing the response development over time.
Next the proposed model is taken to identify J-C parameters via a constrained nonlinear
optimization scheme. The relationship between material strength and bulged diameter at the
impact end is explored in the inverse formulation. This is in contrast with the one-dimensional
models which rely on the length of an un-deformed section whose experimental measurement
in the Taylor test is uncertain. Moreover, the proposed model takes into account the temporal
variation of strain, strain rates and temperature effect unlike the existing model [1,11], wherein a
time-average strain rate value is used. The major advantage of such temporal variation is that, in
place of only mean dynamic yield strength, the actual evolution of dynamic yield strength as a
function of strain and strain rate may be obtained. The accuracy of the model with the inverse
framework in estimating J-C parameters to capture that evolution is demonstrated. It is also
shown that the model is capable of estimating true material constants with reasonable accuracy
when a different forward model (here FEM-based ABAQUS simulation) is used to generate the

...................................................

1.0

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

1.1

Taking the trailing rigid part (length, l) of the rod, with cross-sectional area (Ar = (/4)D20 ) under
consideration, the mass M(= Ar l) moves at v and subjected to yielding pressure [ (= Cl v) = Yd ]
at the elasticplastic interface. So the deceleration equation (2.27) is rearranged as


Yd Cp
Yd
E
dv
+
=
Cl =
(A 1)
dt
Cl l
l

Substituting, Yd = Cl v in LHS

Ar l
and

Yd
dv
vCp =
dt

dl
Yd
dv
+v =
dt
dt

(A 2)


dl
= Cp
dt

d
dv
+ v (Ar l) = Yd Ar
dt
dt
d
(Mv) = Yd Ar,

dt


(A 3)
(A 4)
(A 5)

which, invariably, is the statement of linear momentum conservation.

References
1. Taylor G. 1948 The use of at-ended projectiles for determining dynamic yield stress.
I. Theoretical considerations. Proc. R. Soc. Lond. A 194, 289299. (doi:10.1098/rspa.1948.0081)
2. Whiffen AC. 1948 The use of at-ended projectiles for determining dynamic yield stress
II. Tests on various metallic materials. Proc. R. Soc. Lond. A 194, 300322. (doi:10.1098/
rspa.1948.0082)
3. Carrington W, Gayler ML. 1948 The use of at-ended projectiles for determining dynamic
yield stress. III. Changes in microstructure caused by deformation under impact at highstriking velocities. Proc. R. Soc. Lond. A 194, 323331. (doi:10.1098/rspa.1948.0083)
4. Chapman DJ, Radford DD, Reynolds M, Church PD. 2005 Shock induced void nucleation
during Taylor impact Int. J. Fract. 134, 4157. (doi:10.1007/s10704-005-7151-1)
5. Chapman DJ, Radford DD, Walley SM. 2005 A history of the Taylor test and its present use
in the study of lightweight materials. In Design and use of light-weight materials (eds F TeixeiraDias, B Dodd, E Lach, P Schultz), pp. 1224. Aveiro, Portugal: University of Aveiro.
6. Hawkyard JB, Eaton D, Johnson W. 1968 The mean dynamic yield strength of copper and low
carbon steel at elevated temperatures from measurements of the mushrooming of at-ended
projectiles. Int. J. Mech. Sci. 10, 929948. (doi:10.1016/0020-7403(68)90048-9)
7. Hawkyard JB. 1969 A theory for the mushrooming of at-ended projectiles impinging on
a at rigid anvil using energy considerations. Int. J. Mech. Sci. 11, 313333. (doi:10.1016/
0020-7403(69)90049-6)
8. Jones SE, Gillis PP, Foster JC. 1987 On the equation of motion of the undeformed section of a
Taylor impact specimen. J. Appl. Phys. 61, 499502. (doi:10.1063/1.338249)
9. Gillis PP, Jones SE, Dehn MH. 1987 Some further results on the Taylor impact test. Mech. Mater.
6, 293301. (doi:10.1016/0167-6636(87)90028-7)

...................................................

Appendix A. Momentum conservation

19

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

measurement data. While the present inverse formulation effectively illustrates the potential
of the proposed model, a more detailed exploration with different material models (including
micro-mechanical effects) is yet to be addressed. The present inverse formulation relies on the
deformation parameters obtained from the nal recovered rod alone. As the proposed model can
also predict deformation time history with reasonable accuracy, it is anticipated that an inverse
formulation enriched with intermediate proles obtained from high-speed photography [26]
would provide more information about the material behaviour. This may constitute the content
of a future work.

20
...................................................

rspa.royalsocietypublishing.org Proc. R. Soc. A 471: 20140556

10. Jones SE, Maudlin PJ, Foster JC. 1997 An engineering analysis of plastic wave propagation in
the Taylor test. Int. J. Impact Eng. 19, 95106. (doi:10.1016/S0734-743X(96)00020-6)
11. Jones SE, Drinkard AJ, Rule WK, Wilson LL. 1998 An elementary theory for the Taylor impact
test. Int. J. Impact Eng. 21, 113. (doi:10.1016/S0734-743X(97)00036-5)
12. House JW, Lewis JC, Gillis PP, Wilson LL. 1995 Estimation of ow stress under high rate plastic
deformation. Int. J. Impact Eng. 16, 189200. (doi:10.1016/0734-743X(94)00042-U)
13. Johnson GR, Cook WH. 1983 A constitutive model and data for metals subjected to large
strains, high strain rates and high temperatures. In Seventh Int. Symp. on Ballistics, The Hague,
The Netherlands 1921 April, pp. 541547. International Ballistics Society.
14. Johnson G, Holmquist T. 1988 Evaluation of cylinderimpact test data for constitutive model
constants. J. Appl. Phys. 64, 39013910. (doi:10.1063/1.341344)
15. Batra RC, Kim CH. 1990 Effect of viscoplastic ow rules on the initiation and growth of shear
bands at high strain rates J. Mech. Phys. Solids 38, 859874. (doi:10.1016/0022-5096(90)90043-4)
16. zel T, Karpat Y. 2007 Identication of constitutive material model parameters for high-strain
rate metal cutting conditions using evolutionary computational algorithms. Mater. Manuf.
Process. 22, 659667.
17. Woodward RL, ODonnell RG, Flockhart CJ. 1992 Failure mechanisms in impacting
penetrators. J. Mater. Sci. 27, 64116416. (doi:10.1007/BF00576292)
18. Walker JD, Anderson CE. 1995 A time-dependent model for long-rod penetration. Int. J. Impact
Eng. 16, 1948. (doi:10.1016/0734-743X(94)00032-R)
19. ABAQUS -Explicit 2010 Abaqus, users manual version 6.10. USA: Hibbit, Karlsson and
Sorensen Inc.
20. Woodword RL, Burman NM, Baxter BJ. 1994 An experimental and analytical study of the
Taylor impact test. Int. J. Impact Eng. 15, 407416. (doi:10.1016/0734-743X(94)80025-5)
21. Zerilli FJ, Armstrong RW. 1987 Dislocation mechanics based constitutive relations for material
dynamics calculations. J. Appl. Phys. 61, 18161825. (doi:10.1063/1.338024)
22. Taylor GI, Quinney H. 1934 The latent energy remaining in a metal after cold working. Proc.
R. Soc. Lond. A 143, 307326. (doi:10.1098/rspa.1934.0004)
23. Vaughn DG, Canning JM, Hutchinson JW. 2005 Couple plastic wave propagation and column
buckling. J. Appl. Mech.-T. ASME 72, 139146. (doi:10.1115/1.1825437)
24. Batra RC, Zhang GM. 2008 Modied smoothed particle hydrodynamics (MSPH) basis
functions for meshless methods, and their application to axisymmetric Taylor impact test.
J. Comput. Phys. 227, 19621981. (doi:10.1016/j.jcp.2007.10.001)
25. Forde LC, Walley SM, Peyton-Jones M, Proud WG, Cullis IG, Church PD. 2009 The use of
symmetric Taylor impact to validate constitutive models for an FCC metal (copper) and a BCC
alloy (RHA steel). In Proc. 9th Int. Conf. on the Mechanical and Physical Behaviour of Materials
under Dynamic Loading (DYMAT 2009), Brussels, Belgium, 711 September, pp. 12451250. Les
Ulis, France: EDP Sciences.
26. Forde L, Proud W, Walley S. 2009 Symmetrical Taylor impact studies of copper. Proc. R. Soc. A
465, 769790. (doi:10.1098/rspa.2008.0205)
27. Armstrong RW. 2014 Bertram Hopkinsons pioneering work and the dislocation mechanics
of high rate deformations and mechanically induced detonations. Phil. Trans. R. Soc. A 372,
20130181. (doi:10.1098/rsta.2013.0181)
28. Xiao X, Zhang W, Wei G, Mu Z, Guo Z. 2011 Experimental and numerical investigation
on the deformation and failure behavior in the Taylor test. Mater. Des. 32, 26632674.
(doi:10.1016/j.matdes.2011.01.016)
29. Voce E. 1948 The relationship between stress and strain for homogeneous deformation. J. Inst.
Metals 74, 537562.
30. Nocedal J, Wright SJ. 1999 Numerical optimization, vol. 2. New York, NY: Springer.
31. MATLAB 2009 version 7.8.0 (r2009a). Natick, MA: The MathWorks Inc.
32. Jing L, Wang Z, Zhao L. 2014 Response of metallic cylindrical sandwich shells
subjected to projectile impactexperimental investigations. Compos. Struct. 107, 3647.
(doi:10.1016/j.compstruct.2013.07.011)
33. Banerjee B, Roy D, Vasu R. 2009 A pseudo-dynamical systems approach to a class of inverse
problems in engineering. Proc. R. Soc. A 465, 15611579. (doi:10.1098/rspa.2008.0501)
34. Corigliano A, Mariani S. 2004 Parameter identication in explicit structural dynamics:
performance of the extended Kalman lter. Comput. Methods Appl. Mech. Engg. 193, 38073835.
(doi:10.1016/j.cma.2004.02.003)