Inverse estimation of heat ux and temperature in multi-layer gun barrel

Tsung-Chien Chen
a,
*
, Chiun-Chien Liu
a
, Horng-Yuan Jang
b
, Pan-Chio Tuan
b
a
Department of System Engineering, Chung Cheng Institute of Technology, National Defense University, Tao Yuan 335, Taiwan, ROC
b
Department of Computer Science & Information Engineering, Nan Kai Institute of Technology, Taiwan, ROC
Received 19 February 2006; received in revised form 11 November 2006
Available online 16 January 2007
Abstract
In this paper, we present an inverse method, an input estimation method, to recursively estimate both the time varied heat ux and the
inner wall temperature in the chamber. The algorithm includes the use of the Kalman lter to derive a regression model between
the biased residual innovation and the heat ux through a given heat conduction state space model. Based on this regression model,
the Recursive Least Squares Estimator (RLSE) is proposed to extract the time-varying heat ux on-line as the input. Computational
results show that the proposed method exhibits a good estimation performance and highly facilitates practical implementation.
2006 Elsevier Ltd. All rights reserved.
Keywords: Input estimation method; Heat ux
1. Introduction
Gun barrel heating from multiple rings continues to be
a subject of concern to ordnance engineers. Continuous
gun ring rises the barrel temperature, producing several
adverse eects on system performance. In rapid ring situ-
ation, combustion gases and the projectiles sliding friction
will produce chemical, mechanical and temperature aspect
variation. The propellant gas temperature is the major fac-
tor to make gun material melting, cracked, erosion, and
wear, etc. [14]. An excellent review of gun barrel erosion
was given by Ahmad [5]. Ebihara and Rorabaugh [6]
describe various erosion mechanisms and Bracuti [7] dis-
cusses the role of wear reducing additives. After repeated
rings the inner wall temperature increases quickly and
reaches the chrome layer melting point. Coating embrittle-
ment by propellant gas product absorption can adversely
aect barrel wall resistance to thermal shock and sliding
wear [8]. High temperatures will reduce the normal barrel
fatigue life.
A number of investigators have modeled barrel heating
[915]. Some resent publications include: Gerber and
Bundy (1991,1992,1993,1995) and Conroy (1991). In these
studies, the nite-dierence scheme used to solve the prob-
lem numerically and described in detail. The reliable gun
tube design strongly depends on theoretical analysis and
experimental experience. The chamber temperature is
related to the barrel material and re frequency. Thermal
eects resulting from high temperature propellant gas and
friction action between the bullet band and gun bore result
in barrel melting, erosion and wear, thermal stress and
deformation, propellant self-ignition or combustion in the
cartridge case. These failures lead to reduction in muzzle
velocity, range and accuracy, limiting gun performance of
guns, and become restrictions to improvements of gun
power.
Unknown heat source or heat ux estimation utilizing a
measured temperature inside a heat-conducting solid is
called the inverse heat conduction problem (IHCP). Inverse
heat conduction problems have been of interest to many
researchers in recent years. It is sometimes necessary to cal-
culate the transient surface heat ux and the surface tem-
perature from a temperature measured at some location
inside a body. The primary heat transfer mechanism to
0017-9310/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2006.11.022
*
Corresponding author. Tel.: +886 3 3809257; fax: +886 3 3906385.
E-mail address: chojan@ccit.edu.tw (T.-C. Chen).
www.elsevier.com/locate/ijhmt
International Journal of Heat and Mass Transfer 50 (2007) 20602068
the gun barrel from ring is thermal convection from the
propellant gas behind the projectile to the bore surface.
The bullet res within a short time, the propellant gas pro-
duces heat ux convection to the wall. In high rate machine
gun continuous ring the time interval is very short, allow-
ing the heat ux to accumulate so fast that the inner wall
temperature increases sharply. This situation produces heat
input from projectile passage mechanisms that can melt the
bore surface material.
The time varied heat ux acting on the inner wall is dif-
cult to measure from the inner wall temperature. There-
fore, how to estimate the inner wall temperature is very
important. If we know the inner wall temperature, better
gun material and strength can be designed. This paper con-
cerns the boundary heat ux estimation in a one-dimen-
sional heat conduction domain with heat ux in the
chamber wall and convection in the outer wall. The
cross-sectional average temperature of the ow in the bore
T
g
and the coecient of heat transfer between the gas-par-
ticle mixture in the bore and the inner wall of the barrel h
g
were obtained from experiments. The T
g
and h
g
values were
dicult to obtain. In this paper, we present an inverse
method in which the input estimation method, including
a nite-element scheme, is used to solve the estimating
time-varying unknown surface heat ux inverse problem.
The nite-element discretization concept has been applied
to the inverse heat conduction problem [16]. The input esti-
mation method uses the Kalman lter to generate the resid-
ual innovation sequence. Based on the regression model,
the Recursive Least Squares Estimator (RLSE) is proposed
to on-line extract the time-varying heat ux named as the
input. The simulation results demonstrate good perfor-
mance and accuracy in tracking the unknown boundary
heat ux of a thermal system. Estimating the unknown
heat ux presents a very important reference for under-
standing the inner wall melting and material phase change
necessary to design propellant grain types and barrel
materials.
2. The weighting input estimation algorithm
The recursive input estimation algorithm consists of two
parts. The rst part is a Kalman lter. The second part is a
real-time least squares algorithm. The input parameter is
the unknown time-varying heat ux. The Kalman lter
requires an exact knowledge of the process noise variance
Q and the measurement noise variance R, R depends on
the sensor measurements. The Kalman lter is used to gen-
erate the residual innovation sequence. This recursive real
time least-squares algorithm is derived by residual sequence
Nomenclature
B sensitivity matrix
[B] gradient matrix
[C] capacitance matrix
C
p
specic heat
C
c
chrome specic heat
C
s
steel specic heat
[D] matrix of conductivity values
{F} thermal load vector
{} coecient vector
H measurement matrix
I identity matrix
J functional
J
e
element functional
k time (discretized)
k
c
chrome thermal conductivity
k
s
steel thermal conductivity
K Kalman gain
K
b
steady-state correction gain
l element length
M sensitivity matrix
[M] global conductance matrix
N total number of nodes
[N] shape function matrix
P lters error covariance matrix
P
b
error covariance matrix
Q process noise covariance
q(t) heat ux
R measurement noise covariance
R
o
radius of outer wall
R
i
radius of inner wall
s innovation covariance
t time
T temperature
T
0
initial temperature
v measurement noise vector
r radial coordinate
Z(k) observation vector
c forgetting factor
a thermal diusivity
a
c
thermal diusivity of the chrome
a
s
thermal diusivity of the steel
b impulse duration time
C input matrix
d Dirac delta function
q density
q
c
density of the chrome
q
s
density of the steel
U state transition matrix
W coecient matrix
X coecient matrix
Dt sampling time interval
x process noise vector
r standard deviation
T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068 2061
to compute the value of the input heat ux. The actual val-
ues derived in the paper [17].
The Kalman lter equations are given by
X(k=k 1) = UX(k 1=k 1) (1)
P(k=k 1) = UP(k 1=k 1)U
T
CQC
T
(2)
s(k) = HP(k=k 1)H
T
R (3)
K(k) = P(k=k 1)H
T
s
1
(k) (4)
P(k=k) = [I K(k)H[P(k=k 1) (5)
Z(k) = Z(k) HX(k=k 1) (6)
X(k=k) = X(k=k 1) K(k)Z(k) (7)
The equations for a recursive least-squares algorithm are
B(k) = H[UM(k 1) I[C (8)
M(k) = [I K(k)H[[UM(k 1) I[ (9)
K
b
(k) = c
1
P
b
(k 1)B
T
(k)[B(k)c
1
P
b
(k 1)B
T
(k) s(k)[
1
(10)
P
b
(k) = [I K
b
(k)B(k)[c
1
P
b
(k 1) (11)
^q(k) = ^q(k 1) K
b
(k)[Z(k) B(k)^q(k 1)[ (12)
where ^q(k) is the estimated input vector, P
b
(k) is the error
covariance of the estimated input vector, B(k) and M(k) are
the sensitivity matrices, and K
b
is the Kalman gain. Z(k) is
the bias innovation caused by measurement noise and
input disturbance. s(k) is the covariance of the residual.
c is a forgetting factor.
In this paper, we choice c = 0.925 [18] to compromise
between fast adaptive capability and the estimate accuracy
loss.
3. Mathematical model
Assume that there is a gun barrel is a hollow cylinder.
The radii of the inner and outer walls are R
i
and R
o
, respec-
tively. The radius of the interface between the two metal
layers is R
L
: and d
c
(= R
L
R
i
) is the thickness of the
chrome layer. The measured temperature z(t) is from the
thermocouple at x = R
o
. Fig. 1 shows the geometry and
coordinates. The following restrictions apply here [12]:
(1) The temperature gradients in the axial direction are
neglected in comparison to those in the radial
direction.
(2) The temperature is axi-symmetrical in the plane nor-
mal to the bore axis. This implies axi-symmetrical
heat input, as well neglecting gravity in the barrel wall
and chrome thickness variation along with other
eects that would cause azimuthal dependence.
(3) The barrel feedback heating to the gas ow in the gun
bore is neglected.
(4) Barrel thermal expansion is not considered to have an
eect on the heat transfer process.
(5) The densities, specic heats, and thermal conductivi-
ties of the barrel steel and chrome layer are all
constants.
convection
boundary q (t)
r=Ri r=R0
Node1 Node2 Last Node
i j
Steel
Chrome
Sensor
convection
boundary
2
6 /
25
h W m C
T C

=
=
l
R0
Steel
X
Chrome
R
L
R
i
O
q
Sensor
Fig. 1. Geometry and coordinates.
2062 T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068
(6) Thermally perfect contact is assumed at the chrome-
steel interface (i.e., temperature and radial conductive
heat ux are continuous).
(7) The latent heat eects due to phase change in the steel
from martensite to austenite are neglected.
The mathematical formulation of the one-dimensional,
transient, heat conduction problem can be generalized as:
o
2
T(r; t)
or
2

1
r
oT(r; t)
or
=
1
a
c
_ _
oT(r; t)
ot
in R
i
6 r 6 R
L
;
(13)
and
o
2
T(r; t)
or
2

1
r
oT(r; t)
or
=
1
a
s
_ _
oT(r; t)
ot
in R
L
6 r 6 R
o
; (14)
T(r; 0) = T
0
for t = 0; in R
i
6 r 6 R
o
(15)
k
c
oT
or
= q(t) in r = R
i
; (16)
k
s
oT
or
= h(T T

) in r = R
o
; (17)
k
c
oT
or
_ _
c
= k
s
oT
or
_ _
s
in r = R
L
; (18)
Z(r; t) = T(r; t) v(t) measured temperature (19)
where T is the temperature distribution as is function of r
and t, t = time measured from firing initiation. The con-
stants a
c
= k
c
=(q
c
c
p
c
) and a
s
= k
s
=(q
s
c
p
s
) are the ther-
mal diusivities of the chrome and steel, respectively.
Here k
c
,q
c
and c
p
c
are the thermal conductivity, density,
and specic heat, respectively of the chrome; k
s
, q
s
and
c
p
s
are the corresponding properties of the steel.
Where T
0
is the uniform initial temperature, q(t) is
the unknown heat ux input to be estimated, and Z(t) is
the noise-corrupted measurement, v(t) is the measurement
noise assumed zero mean and white Gaussian, h is the coef-
cient of convective heat transfer between the barrel wall
and the surrounding atmosphere, T

is outer wall
temperature.
In chrome layer R
i
6 r 6 R
L
, The calculus of variations
provides an alternative method for formulating the govern-
ing Eq. (13) and boundary conditions (15), (16), and (18).
Variational calculus states that the minimization of the
functional J
c
[19]:
J
c
=
1
2
_ _ _
V
rk
c

oT
or
_ _
2
2rq
c
C
c

oT
ot
_ _
T
_ _
dV

_
S
qT dS (20)
The element equation for the temperature is T
e
=
N
e
i
T
i
N
e
j
T
j
= [N
e
[Twhere T
i
and T
j
are the nodal tem-
peratures to be determined.
[N
e
[ = [N
e
i
N
e
j
[ =
R
j
r
R
j
R
i
r R
i
R
j
R
i
_ _
the temperature gradient matrix {g
e
}, is given by
g
e
=
oT
e
or
=
oN
e
i
or
oN
e
j
or
_ _
T =
1
R
j
R
i
T
i

1
R
j
R
i
T
j
=
1
R
j
R
i
1
R
j
R
i
_ _
T
i
T
j
_ _
= [B
e
[T
where [N] is the shape function matrix, [B] is obtained by
dierentiating [N] with respect to r.
Eq. (20) must be minimized with respect to the set of
nodal values {T}
oJ
c
oT
=
o
oT

E
e=1
J
e
c
=

E
e=1
oJ
e
c
oT
= 0
The minimization process produces the following system of
equations.
[C[
c
oT
ot
[M[
c
T F
c
= 0
The [C]
c
matrix is the global capacitance matrix, [M]
c
is
the global conductance matrix, {F}
c
is the thermal load
vector. The element contributions to [C]
c
, [M]
c
, {F}
c
are
summed in the usual manner.
[D
e
[
c
= [rk
e
c
[ (21)
[C
e
[
c
=
_
V
rq
c
C
c
[N
e
[T[N
e
[
oT
ot
dV =
2pq
c
C
c
60l
2
(2R
5
j
20R
2
j
R
3
i
30R
j
R
4
i
12R
5
i
) (3R
5
j
5R
4
j
R
i
5R
j
R
4
i
3R
5
i
)
(3R
5
j
5R
4
j
R
i
5R
j
R
4
i
3R
5
i
) (12R
5
j
30R
4
j
R
i
20R
3
j
R
2
i
2R
5
i
)
_
_
_
_
(22)
[M
e
[
c
=
_
V
[B
e
[
T
[D
e
[
c
[B
e
[dV
=
2pk
c
l
2
_
Rj
R
i
1
1
_ _
[1 1[ r
2
dr =
2pk
c
(R
3
j
R
3
i
)
3(R
j
R
i
)
2
1 1
1 1
_ _
(23)
f
e

c
=
_
S1
q[N
e
[
T
dS = ff q = 2pR
i
q
1
0
_ _
(24)
where R
i
, R
j
are the single element left node and right node.
l = R
j
R
i
is the length of single element. [D] is matrix con-
sists of the conductivity values. [] is the coecient matrix.
In the steel layer, The functional formulation that is
equivalent to (14) and its boundary conditions (15), (17)
and (18) is J
s
:
J
s
=
1
2
_ _ _
V
rk
s

oT
or
_ _
2
2rq
s
C
s

oT
ot
_ _
T
_ _
dV

_
S
1
1
2
h(T T

)
2
dS (25)
J =

E
e=1
_
V
e
1
2
g
e

T
[D
e
[g
e
dV
_
V
r(q
s
C
s
)
e
oT
e
ot
_ _
T
e
dV

_
S
e
1
1
2
h(T
e
T
e

)
2
dS (26)
T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068 2063
We take a single element
J
e
s
=
_
V
e
1
2
T
T
B
e

T
[D
e
[B
e
TdV

_
V
e
rq
s
C
s
[N
e
[T[N
e
[
oT
ot
dV

_
S
e
1
h
2
T
T
N
e

T
N
e
TdS

_
S
e
1
hT

N
e
TdS
_
S
e
1
h
2
hT
2

dS (27)
In the steel layer, Eq. (27) must be minimized with
respect to the set of nodal values {T}:
oJ
s
oT
=
o
oT

E
e=1
J
e
s
=

E
e=1
oJ
e
s
oT
= 0
We can get steel layer single element matrix [C
e
[
s
, [M
e
[
s
,
f
e

s
, respective:
[D
e
[
s
= [rk
e
s
[ (28)
[C
e
[
s
=
_
V
rq
s
C
s
[N
e
[T[N
e
[
oT
ot
dV
=
2pq
s
C
s
60l
2
(2R
5
j
20R
2
j
R
3
i
30R
j
R
4
i
12R
5
i
) (3R
5
j
5R
4
j
R
i
5R
j
R
4
i
3R
5
i
)
(3R
5
j
5R
4
j
R
i
5R
j
R
4
i
3R
5
i
) (12R
5
j
30R
4
j
R
i
20R
3
j
R
2
i
2R
5
i
)
_ _
(29)
[M
e
[
1
s
=
_
V
[B
e
[
T
[D
e
[
s
[B
e
[dV
=
2pk
s
l
2
_
Rj
Ri
1
1
_ _
1 1 [ [r
2
dr =
2pk
s
(R
3
j
R
3
i
)
3(R
j
R
i
)
2
1 1
1 1
_ _
(30)
In the right node (N
i
= 0, N
j
= 1, r = R
j
)
[M
e
[
2
s
=
_
S
e
2
hN
e

T
N
e
dS =
_
S
e
2
h
N
i
N
i
N
j
N
i
N
i
N
j
N
j
N
j
_ _
dS
= 2pR
o
h
0 0
0 1
_ _
Merging [M
e
[
1
s
ms for element 1; 2; . . . ; n and [M
e
[
2
s
.
[M
e
[
s
= [M
e
[
1
s
[M
e
[
2
s
=
m
1
m
1
0 0 0 0
m
1
m
1
m
2
m
2
0 0 0
0 m
2
m
2
m
3
m
3
0 0
0 0 m
3
m
3
m
4


.
.
.
m
n
0 0 0 0 m
n
m
n
2pR
o
h
_

_
_

_
(31)
f
e

c
=
_
S
e
1
hT

N
e

T
dS =
_
S
e
1
hT

0
1
_ _
dS = 2pR
o
hT

0
1
_ _
(32)
Assemble the chrome layer and steel layer element, then
minimized with respect to the set of nodal values {T}:
oJ
oT
=
o
oT

E
e=1
J
e
=

E
e=1
oJ
e
oT
= 0 (33)
where Eq. (33) is a system of rst-order linear dierential
equations.
[C[
oT
ot
[M[T F = 0 (34)
The [C] matrix is the capacitance matrix, [M] is the con-
ductance matrix, {F} is the thermal load vector. From Eq.
(34) and to account for process noise inputs [20], the con-
tinuous time state equation can be written as:
_
T(t) = WT(t) X[q(t) w(t)[ (35)
W = (1)[C[
1
[M[; X = (1)[C[
1
ff
where the state vector T(t) is N 1, N is the total nodes, W
and X are the coecient matrix. W is N N, X is N 1, q
is the unknown input heat ux of the boundary, x(t) is the
process noise. This noise term represents the modeling
error. We replace the time derivative with a forward
dierence, discrete over time intervals of length Dt is
X(k) = UX(k 1) C[q(k 1) x(k 1)[ (36)
where
X(k 1) = [ T
1
T
2
T
3
. . . T
N1
T
N
[
T
;
U = e
WDt
C =
_
(k1)Dt
kDt
expW[(k 1)Dt s[Xds
Here X represent the state vector, U is the state transi-
tion matrix, C is the input matrix, q is the deterministic
input sequence and x is the process noise vector, assumed
to be zero mean and white noise with variance
Ex(k)x
T
(j) = Qd
kj
where d
kj
is a Dirac delta function. In order to compare the
results for situations involving measurement errors, using
Eq. (19), the discrete measure equation becomes:
Z(k) = HX(k) m(k) (37)
where Z is the observation vector at time kDs, H is the
measurement matrix, v is the measurement noise vector,
assumed to be zero mean and white. The variance of v(k)
is given by Em(k)m
T
(j) = Rd
kj
.
The obtained state equation is combined with the Kal-
man lter and recursive least squares method to progress
the inverse estimation.
4. Results and discussion
Using simulation, a 5.56 mm T65K2 rie three-shot and
5.56 mm machine gun under continuous re situations
were modeled. The thermal and mechanical properties of
the gun steel (AISI 4340) and its associated chrome layer
were taken as [12]:
Specic heat, C
s
= 469:05 J=(kg

C), C
c
= 505:03
J=(kg

C)
Density, q
s
= 7827 kg=m
3
, q
c
= 7191kg=m
3
Thermal conductivity, k
s
= 38:07 J=(m s

C), k
c
= 83:75
J=(m s

C)
2064 T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068
Thermal diusivity, a
s
= 1:037 10
5
m
2
=s, a
c
=
2:305 10
5
m
2
=s
Melting temperature (T
m
)
s
= 1504

C, (T
m
)
c
= 1857

C
Here we consider two gun barrels, a T65K2 rie barrel,
and a 5.56 mm machine gun barrel
T65K2 rie, (R
i
= 0:00278 m and R
o
= 0:00720 m)
5.56 mm machine gun (R
i
= 0:00278 m and R
o
=
0:00828 m).
To illustrate the accuracy of the proposed approach in
predicting input heat ux ^q(k), we used a representative
one dimension unknown triangle wave heat ux to check
the feasibility of the weighting input estimation method
combined nite element method. A thermocouple was
placed at the surface x = R
o
, the coecient of convective
heat transfer between the barrel wall and the surrounding
atmosphere h = 6 W=m
2
C [12], T

= 25

C is outer wall
temperature. Since P(1/1) and P
b
(1) are normally
not known, the estimator was initialized with P(1/1)
and P
b
(1) as very large numbers, such as 10
8
, respectively.
This had the eect of treating the initial errors as very large,
so the estimator ignored the rst few initial estimates. The
initial input estimator CONDITIONS were given
by X(1= 1) = [ 0 0 . . . 0 [
T
and P(1= 1) =
diag[10
8
[ for the Kalman lter. The real-time least-squares
algorithm initial conditions were given by ^q(1) =
0; P
b
(1) = 10
8
and M(1) was set using a zero matrix.
The Kalman lter for the recursive input estimation algo-
rithm requires exact knowledge of the process noise covari-
ance matrix Q and the measurement noise covariance
matrix R. R depends on the sensor measurements. Both
the lter Q value and sequential least-square c value interac-
tively aect the fast adaptive capability for tracking the
time-varying parameter. The test input heat ux is given
by
Example 1. Triangle waveform in q(t) (W=m
2
)
q(t) =
0 0 6 t < 4; 10 < t 6 t
f
10
6
(0:25t 1) 4 6 t 6 8
10
6
(0:5t 5) 8 6 t 6 10
_

_
(38)
The sampling interval Dt = 0:01 s, the estimation input
heat ux q(t) on the boundary x = R
i
, at x = R
o
is convec-
tion situation h = 6 W=m
2
C, T

= 25

C, the sensor loca-
tion is at x = R
o
, the element number E = 550, the initial
temperature T
0
= 0

C, process noise covariance Q = 10,
and measurement noise covariance r = 10
4
, r = 10
3
.
The estimates of q(t) are plotted in Fig. 2 for r = 10
4
and r = 10
3
, respectively. This indicates that a large mea-
surement error can cause estimate accuracy degradation. In
this case, although the measurement error inuences the
estimate resolution, the results are still good.
Example 2. Square waveform in q(t) (W=m
2
)
q(t) =
0 0 6 t < 2; 27 < t 6 t
f
7 10
5
2 6 t 6 27
_
(39)
We used the following test condition and parameters:
5.56 mm machine gun, continuous re at 25 s, if we assume
the heat ux become reduce the ring time can be prolong,
the round-for-round time was very short and neglected, the
sampling interval Dt = 0:01 s, elements number E = 550,
process noise covariance Q = 10, and measurement noise
covariance r = 10
4
, r = 10
3
. Figs. 3 and 4 demonstrate
the temperature gure. The estimates of q(t) are plotted
in Fig. 5 for r = 10
4
and r = 10
3
, respectively. From
Fig. 3a we can nd the chrome layer melting point at
1857 C. Under continuous re the melting point will be
reached in about 22 s. From Fig. 5 it is evident that the
recovery of q(t) using the weighting input estimation
method combined with the nite element method is good.
The estimation results show excellent agreement with the
exact value. Using this method, we can use the measured
temperature to recursively estimate both the time varied
heat ux in the chamber and the inner wall temperature.
Fig. 2. (a) Estimated triangle wave as input heat ux with r = 10
4
,
Q = 10. (b) Estimated triangle wave as input heat ux with r = 10
3
,
Q = 10.
T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068 2065
The proposed scheme is helpful in developing future
nondestructive experiments. The temperature inuence on
a shooter, gun material and propellant testing are fast
and economical. These applications can be useful in mak-
ing quick and ecient identication of unknown heat ux
in the chamber surface.
Example 3. Three-shot square waveform in q(t) (W=m
2
)
q(t) =
7 10
5
1 6 t 6 1:3; 1:4 6 t 6 1:7;
1:8 6 t 6 2:1; 2:2 6 t 6 2:5
2:6 6 t 6 2:9; 3:0 6 t 6 3:3
0 others
_

_
(40)
The following test condition and parameters: 5.56 mm
TK65K2 rie gun, three-shot six times, round-for-round
time was very short and neglected, sampling interval
Dt = 0:001 s, elements number E = 450, process noise
covariance Q = 10, and measurement noise covariance
r = 10
4
. The estimates of q(t) are plotted in Fig. 6a. It
was found that Fig. 6a shows larger uctuations in the esti-
Fig. 3. (a) Continuous re inner wall temperature. (b) Continuous re
outer wall temperature.
Fig. 4. Continuous re 3D temperature.
Fig. 5. (a) Estimated continuous re with r = 10
4
, Q = 10. (b) Estimated
continuous re with r = 10
3
, Q = 10.
2066 T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068
mated value than Fig. 5. This indicates that fast pulse heat
ux can cause estimate accuracy degradation. In Fig. 6b,
Q = 100 the estimates of q(t) were better than those in
Fig. 6a. In this case, although the fast pulse heat ux
inuences the estimate resolution, the results are still
acceptable.
In order to analyze the inuences in estimation results
produced by using dierent impulse duration and sampling
time, the results of tests are shown in Fig. 7. The three sets
of sampling time, Dt = 0.01, 0.001, and 0.0005 s, and the
three sets of impulse duration time, b = 0.1, 0.2, and
0.3 s. According to the gure, the estimation is precise for
dierent impulse duration when the sampling time (Dt) is
0.001 s. The estimation has slightly better result when
Dt = 0:0005 s. However, the improvement is not signi-
cant. As a result, when Dt is shorter than 0.0005 s, the com-
puting time in the simulation will be longer. The results of
estimation may not even converge due to the relatively
greater inuence of noise than the change of temperature
in a very short time interval.
The above simulation results demonstrates that the pro-
posed method has good performance in tracking unknown
heat ux imposed on a gun barrel wall in a one-dimen-
sional body.
Fig. 6. (a) Estimated three-shot re with r = 10
4
, Q = 10. (b) Estimated
three-shot re with r = 10
4
, Q = 100.
Fig. 7. (a) Comparison between the estimated b = 0:1 s with Dt = 0.01,
0.001, 0.0005 s. (b) Comparison between the estimated b = 0:2 s with
Dt = 0.01, 0.001, 0.0005 s. (c) Comparison between the estimated b = 0:3 s
with Dt = 0.01, 0.001, 0.0005 s.
T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068 2067
5. Conclusions
This research used simulations of the measured temper-
ature on a gun barrel outer wall to estimate the heat ux in
the chamber on-line with accuracy. The measurement data
can be used to understand all temperature elds. We found
that the heat ux input from projectile passage mechanisms
can melt/erode the bore surface material. If we know the
temperature variability, we can then avoid reaching the
gun material melting point. The proposed scheme is helpful
in developing future nondestructive experiments and design
propellant grain types and barrel materials. The time var-
ied heat ux of a gun barrel was estimated precisely on-line
in this study. The total temperature eld inside the gun bar-
rel was estimated to determine the gun material melting
point using the temperature variation. This kind of nonde-
structive experiment can provide fast economical gun bar-
rel and propellant research. The heat pulse ux produced
by the bullet occurs simultaneously after shooting from
the barrel in continuous shooting cases. Therefore, the pro-
posed method is sensitive to the temperature delay eect
and measurement accounting. We must choose a large Q
and to acquire better estimated results.
Further, systematic study can focus on how to avert the
temperature eect on the gun barrel inner wall, determining
methods to control heat, such as tantalum or improved
chromium coatings to prolong barrel life and reduce main-
tenance costs. The temperature inuence on a shooter, gun
material and propellant testing can be determined econom-
ically and quickly. These applications can be useful in mak-
ing quick and ecient identication of unknown heat ux
on the chamber surface. The conclusions obtained are valu-
able in improving the service life of gun barrels and oer a
scientic basis for the rational and eective choice of
materials.
Acknowledgement
This work was supported by the National Science Coun-
cil of the Republic of China under Grant NSC93-2212-E-
014-009.
References
[1] J. Richard Ward, Timothy L. Brosseau, Bertram B. Grollman. Heat
Transfer in Guns-Determination of Friction Factor From Heat Input
Measurements (ADA105430), U.S. Army Armament Research and
Development Command, U.S. Army Ballistic Research Laboratory,
Aberdeen Proving Ground, MD, September, 1981.
[2] R.M. Fisher, A. Szirmae, W.T. Huang, Paul J. Conroy. Metallo-
graphic Studies of Erosion and Thermo-Chemical Cracking of
Cannon Tubes (ADA135816), U.S. Army Armament Research &
Development Command Benet Weapons Laboratory, DRDAR-
LCB-TL Watervliet, NY 12189, May, 1983.
[3] James N. Blecker. Small Arms Gun Barrel Thermal Experimental
Correlation Studies (AD786509), Research Directorate, SARRI-LR
GEN Thomas J. Rodman Laboratory Rock Island Arsenal, Rock
Island, IL 61201, June, 1974.
[4] Leonard H. Caveny. Steel Erosion Produced by Double Base, Triple
Base, and RDX Composite Propellants of Various Flame Temper-
ature, (ADA092344), U.S. Army Armament Research and Develop-
ment Command, Large Caliber Weapon Systems Laboratory, Dover,
New Jersey, October, 1980.
[5] I. Ahmad, in: L. Steifel (Ed.), Gun Propulsion Technology, Progress
in Astronautics and Aeronautics, vol. 109, AIAA, Washington, 1988,
pp. 311356.
[6] W.T. Ebihara, D.T. Rorabaugh, in: L. Steifel (Ed.), Gun Propulsion
Technology, Progress in Astronautics and Aeronautics, vol. 109,
AIAA, Washington, 1988, pp. 357376.
[7] A.J. Bracuti, in: L. Steifel (Ed.), Gun Propulsion Technology,
Progress in Astronautics and Aeronautics, vol. 109, AIAA, Wash-
ington, 1988, pp. 377412.
[8] P.J. Cote, Gas-metal reaction products in the erosion of chromium-
plated gun bores, Wear (241) (2000) 1725.
[9] N. Gerber, M. Bundy, Heating of a Tank Gun Barrel: Numerical
Study. BRL-TR-3932 (ADA241136), U.S. Army Ballistic Research
Laboratory, Aberdeen Proving Ground, MD, August, 1991.
[10] N. Gerber, M. Bundy. Eect of Variable Thermal Properties on Gun
Tube Heating. BRL-TR-3984 (ADA253066), U.S. Army Ballistic
Research Laboratory, Aberdeen Proving Ground, MD, July, 1992.
[11] N. Gerber, M. Bundy, Cross-Barrel Temperature Dierence Due to
Wall Thickness Variation, ARL-TR-100 (ADA262509), U.S. Army
Ballistic Research Laboratory, Aberdeen Proving Ground, MD,
March, 1993.
[12] N. Gerber, M.L. Bundy, Gun Barrel Heating With Heat Input
During Projectile Passage, Journal of Ballistics 12 (4) (1995) 267
281.
[13] Paul J. Conroy, Gun Tube Heating. BRL-TR-3300 (ADA243265),
U.S. Army Ballistic Research Laboratory, Aberdeen Proving
Ground, MD, December, 1991.
[14] Tony W.H. Sheu, S.H. Lee, Numerical Study of Two-Dimensional
SolidGas Combustion Through Granulated Propellants, Numerical
Heat Transfer, Part A 27 (1995) 395415.
[15] J.K. Clutter, W. Shyy, Computation of High-Speed Reacting Flow
for Gun Propulsion Applications, Numerical Heat Transfer, Part A
31 (1997) 355374.
[16] T.C. Chen, H.Y. Jang, P.C. Tuan, Input Estimation Method
Combining the Finite-Element Scheme for Inverse Heat Flux
Estimation in a Barrel Inner Wall, Journal of Explosives and
Propellants (2004).
[17] P.C. Tuan, L.W. Fong, W.T. Huang, Analysis of On-Line Inverse
Heat Conduction Problems, Journal of Chung Chen Institute of
Technology 25 (1) (1996) 5973.
[18] P.C. Tuan, L.W. Fong, W.T. Huang, Application of Kalman
Filtering with input estimation technique to on-line cylindrical inverse
heat conduction problems, JSME International Journal, Series B 40
(1) (1997) 126133.
[19] J.N. Reddy, Applied Functional Analysis and Variational in Engi-
neering, Malabar, Florida, 1991.
[20] A.H. Jazwinski, Stochastic Processes and Filtering Theory, Academic
Press, New York, 1970.
2068 T.-C. Chen et al. / International Journal of Heat and Mass Transfer 50 (2007) 20602068