Piped Water Cooling of Concrete Dams: J. Charpin, T. Myers, A.D. Fitt, N. Fowkes, Y. Ballim and A.P. Patini
Piped Water Cooling of Concrete Dams: J. Charpin, T. Myers, A.D. Fitt, N. Fowkes, Y. Ballim and A.P. Patini
DAMS
J. Charpin, T. Myers, A.D. Fitt , N. Fowkes, Y. Ballim
and A.P. Patini
Abstract
Piped water is used to remove hydration heat from concrete dams
during construction. By examining simple models we obtain an estimate for the temperature rise along the pipe network and within
the concrete. To leading order, for practically useful networks, the
temperature distribution is quasi-steady, so that exact analytic solutions are obtained. The temperature in the water increases linearly
with distance along the pipe and varies logarithmically with radial distance from the pipe in the concrete. Using these results we obtained
estimates for the optimal spacing of pipes and pipe length. Some preliminary work on optimal network design has been done. This is work
in progress.
Introduction
69
70
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
Obviously this can lead to problems due to thermal stress, such as cracking
and resultant structural weakening.
To alleviate the problem pipe networks are included in the block and
chilled water is pumped through the pipes. Later the pipes are filled with
concrete. The aim of this study was to estimate the temperature profile in
the concrete and the effect of pumping water, with the ultimate goal being
to provide a strategy for improved heat removal.
(2.1)
A crude estimate for the concrete temperature (and water exit temperature)
is thus given by
Tout = T0 +
qV
Qw cw
(2.2)
71
The above discussion assumes the hydration heat release remains fixed
and steady state conditions are realised and appropriate. In fact, the hydration heat release varies with time, so that the aim of the network may
be to simply extract sufficient heat during the significant hydration period
to keep temperature levels within acceptable limits. We now move on to a
more complex, but hopefully more accurate, model.
Tc
r
z=0
=0
Water flux Q
Figure 1: Cylindrical model: water flux Q at temperature T0 through a pipe
cools an insulated concrete cylinder.
We consider a pipe radius a(m) embedded in a long insulated cylindrical
slab of concrete of radius R(m), length L, see Figure 1. The concrete produces hydration heat at a prescribed rate q per unit volume. Cool water, at
temperature T0 , enters the pipe at z = 0 and removes hydration heat from
the slab. The volume flux of water flowing through the pipe is Q. The above
configuration models a section of pipe within a concrete sleeve within the
c
slab. An insulated boundary condition, T
r = 0, is chosen at r = R to take
into account the local periodicity of the pipe network; the situation thus
approximately models a periodic array of pipes with spacing 2R. We will
modify the solutions obtained here to treat the periodic pipe array problem
72
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
later. Information from this model, and its extensions, will enable us to
estimate the optimal spacing R between elements of the pipe network and
the flux levels Q required to maintain concrete temperature levels within
prescribed bounds. Also we will use the model to estimate the appropriate
length of piping.
The heat equations in the concrete and water are, respectively,
c cc
w cw
Tc
= c 2 Tc + q ,
t
Tw
+ u Tw
t
= w 2 Tw ,
(3.1)
(3.2)
where Tc , Tw are the temperature of the concrete and the water in the pipe, q
is the rate of heat production per unit volume in the concrete, and (c , cc , c ),
(w , cw , w ) are the density, specific heat and conductivity of concrete and
water respectively. Due to the low kinematic viscosity of water ( 106 ),
the Reynolds number is high even for low volume fluxes and so the flow in the
pipe will be turbulent (providing the fluid velocity is greater than 1cm/s).
Under such circumstances the average radial velocity is zero and the mean
flow is in the z-direction, u = (0, w). Since the fluid is incompressible we
can state w = Q/a2 is constant, where Q is the water volume flux in the
pipe. We may simplify the water heat equation further by considering the
average temperature, since the flow is axisymmetric we may write
Tw =
Ra
Tw rdr
.
a2
Further, since the flow is turbulent and the fluid well-mixed we expect the r
variation to be small, except perhaps for in a boundary layer near the pipe
wall. Hence, integrating the whole equation, gives
Z a
Tw
1
Tw
2 Tw
Q Tw
2w cw
rdr = 2w
r
+
rdr,
+ 2
t
a z
r r
r
z 2
0
0
Tw
a2 Tw
a2 2 Tw
Q Tw
2w cw
= 2w a
+
+ 2
.
(3.3)
2
t
a z
r r=a
2 z 2
Z
a
Tw
= H(Tc Tw ) .
r
(3.4)
73
Note, the heat transfer coefficient can include the effect of the pipe wall
through the relation
H = 2p
s
+ H0 ,
a
(3.5)
where p , s and H 0 represent the conductivity of the pipe, the pipe thickness
and the heat transfer coefficient of water on the pipe, see Carslaw and Jaeger
(1959). Substituting the boundary condition into equation (3.3) gives the
final form for the dimensional governing equation for heat flow in the water
Q Tw
2 Tw
2 Tw
+ 2
. (3.6)
w cw a
= 2aH Tc |r=a Tw + w a2
t
a z
z 2
Note, since the temperature is constant for a given z, i.e. Tw Tw is a
function of z alone we have replaced the value of Tw at r = a with the
average value across the flow Tw (a, z, t) Tw (z, t).
3.1
Dimensional analysis
Tw
Tc
= 2a
= 2aH(Tc Tw ) .
z
r
z = z0 z 0 ,
t = t0 ,
Tc = T0 + T T 0 ,
Tw = T0 + Tc Tw0 ,
t
R2 r r
r
z02 z 2
74
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
Anticipating the fact that radial diffusion is the dominant method for heat
transferral in the concrete we rearrange this to
c cc R2 Tc
Tc
R2 2 Tc
1
qR2
r
+ 2
=
+
.
(3.8)
c t
r r
r
c T
z0 z 2
In the water we expect energy to be carried along with the fluid and so
rearrange the water heat equation accordingly to give
Tw
2aHz0
a2 w 2 Tw
a2 z0 Tw
+
=
( Tc |r= Tw ) +
.
Q t
z
w cw Q
w cw z0 Q z 2
(3.9)
Here = a/R. The final term on the right hand side of (3.9) is O(109 ),
indicating diffusion does not play a significant role in heat transfer in the
water. This term will be neglected from now on.
There are three unknown scales in equations (3.8, 3.9), the length-scale
z0 , the time-scale and the temperature scale T . Clearly the temperature
rise is driven by heat production in the concrete, so we choose
T = qR2 /c .
(3.10)
In the water the temperature rise is due to forced convection at the boundary,
so we choose
z0 = w cw Q/2aH.
(3.11)
The time derivatives indicate two distinct time scales. In the concrete
= c = c cc R2 /c ,
(3.12)
= w = a2 z0 /Q .
(3.13)
in the water
75
concrete to adjust to the presence of the cooling water. Over this period of
time the heat flux into the water increases, leading eventually to a steady
state situation in which the hydration heat supply rate is balanced by the
heat removal rate by the water. The two time-scales allow us to observe the
cooling process from two different viewpoints. Over the water flow timescale, w , the concrete temperature does not vary, over the much longer
time-scale, c , the concrete temperature will change. We will now briefly
examine the thermal problem over the water time-scale and then move on
to the problem over the longer time-scale, c , which is our main concern.
3.1.1
(3.14)
(3.15)
76
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
c
w
c
q
Q
w
2350
1000
1.37
300
2 104
0.59
kg/m3
kg/m3
W/m C
W/m3
m3 /s
kg/m3
cc
cw
R
H
a
880
4200
0.5
1000
0.025
J/kg C
J/kg C
m
W/m2 C
m
more slowly, i.e. the water is less efficient at removing heat from the concrete
as it becomes hotter itself. The factors affecting the water temperature can
be seen from equation (3.15). The water temperature increases more slowly
with a decrease in heat transfer coefficient or pipe radius or an increase in
the flux. Obviously colder water at the inlet results in colder water at the
outlet.
In general our interest lies with the concrete temperature and this is best
examined by working on the time-scale c .
3.1.2
Note that when we work with this time scale, the time derivative term in
(3.9) is O(104 ) and may be neglected throughout the calculation.
We assume (sensibly) that the pipe-spacing is considerably smaller than
the pipe-length R L and so denote = R/L 1. The leading order heat
equations in the concrete and water may now be written
Tc
t
Tw
z
Tc
r
r
1
r r
Tc |r= Tw .
+ 1,
(3.16)
(3.17)
The neglect of the term involving Tc,zz indicates that diffusion in the zdirection is small. However, this is the largest term so far neglected, so a
more accurate solution could be determined by including this term in an
asymptotic expansion in powers of 2 (in fact the solution obtained below is
accurate to order 2 ).
The problem has now reduced to solving equations (3.16, 3.17). The
water enters the pipe at a dimensional temperature T0 which means (3.17)
must be solved subject to Tw = 0 at z = 0. At the pipe boundary, r = =
77
(3.18)
= HR/c .
(3.19)
where
Note the heat transfer H enters the equations through the dimensionless parameter . Halfway between the current pipe and the next, r = 1, symmetry
requires that the temperature gradient must be zero:
Tc
=0.
r
(3.20)
To simplify the calculations we will now consider the steady-state solution. Since changes in the concrete temperature occur over a time-scale of
c = 4 days which is much less than the time span (weeks) of interest this
makes sense (in practice the hydration heat input varies with time; the solutions obtained below really represent a quasi-steady state approximation).
Under steady-state conditions equation (3.16) may be integrated to give
Tc =
r2
+ A log r + B ,
4
(3.21)
2 2
4
2
that is B depends on the water temperature. Substituting for Tc in equation
(3.17) leads to
Tw
1
1
=
+
.
(3.23)
z
2 2
The temperature in the water and concrete is therefore
1
1
Tw =
+
z,
2 2
r2 1
2 1
1
1
Tc = + log r +
+
(z + 1) +
log ,
4
2
2 2
4
2
2
1
r
1
1
2
r
+
+
+
.
= Tw (z) + + log
4
2
2 2
4
(3.24)
(3.25)
(3.26)
78
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
1 L
1 L
1 1
( ), Tcmax =
( + 1) log ,
2 z0
2 z0
4 2
(3.27)
Vq
R2 Lq
T0 +
,
Qw cw
Qw cw
(3.28)
(3.29)
50
T
80
25
35
z=0
T
50
30
60
25
25
20
40
z=0
15
20
10
0.1
0.2
0.3
0.4
0.5
0.025
0.03
0.035
0.04
0.045
0.05
79
3.1.3
Observations
1
(1 2 ) ,
2
across the pipe skin which simply adds onto the water temperature Tw (z).
Note that the size of the jump depends on the transfer coefficient as reflected
in and the surface area of the pipe as reflected in . We can think of this
term simply as the pipe resistance to heat flow.
3.1.4
Design implications
Recall again that the primary issue for the engineer is to reduce the maximum temperature in the concrete to an acceptable level using water cooling,
80
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
and using a minimal (least costly) network1 . To a very limited extent one
may improve the efficiency of the cooling system by reducing the pipe resistance to heat flow as defined above, for example by increasing (changing
the conductivity of the pipe, or pipe diameter) or (increasing the pipe
diameter). As we have seen above the effect of such a change will be to
1
change the temperature throughout the slab by 2
(1 2 )T in dimensional terms. The effect of reducing the input water temperature T0 will
also be to uniformly reduce the temperature of the concrete slab. However,
it is in the design of the network as a whole (that is the choice R, L and
Q for our simple model) that most gains can be made, and the scales obtained above provide the qualitative answers. When supplemented with the
solutions above quantitative answers result.
Again it is absolutely essential to recognise that we are designing a heat
absorption system, not a water transport system. If the aim was to transport
water, then large diameter (high volume flux) pipes are the answer, whereas
to absorb heat it makes sense to use many small pipes; because the surface
area (for absorption) of such small pipes is much larger than that of the larger
pipe for the same total volume flux. Indeed from a purely heat absorption
point of view very fine pipes are the answer, but the resulting system would
be expensive. This is, however, simplistic; a good design is one in which
the total temperature variation along the pipe is about the same size as
the variation between adjacent pipes (i.e. at r = R), and of course the
maximum concrete temperature must be acceptable; this determines the
spacing R. Our present concern is with the very simple cylindrical network
of Figure 1 with a single pipe running through it, and our aim in this context
is to determine an appropriate flux and pipe radius (Q, a) using all other
parameters as in Table 1.
Thus we have a 250m length of pipe surrounded by a cylindrical sleeve
of concrete radius 0.5m.2 We will parallel observations with calculations to
display the connections. Note:
Scaling arguments indicated a time scale of order c = c cc R2 /c for
significant temperature changes to be brought about by the circulating
water. We found that, at least for the parameters used earlier, this
gave 4.3 days, which suggests that tuning this time scale is not likely
to be important in practice.
1
More correctly the engineer wants to minimise the thermal stress, but one would
expect the maximum temperature to be a good indicator
2
This would correspond to a 4m by 10m by 1m slab of concrete with the above pipe
winding back and forward through its centre.
81
The above cylindrical model is inadequate in two ways, both of which are
easily overcome. Firstly the geometry in practice is not cylindrical; a periodic array of pipes is more realistic. The solution obtained for the cylindrical
model, see (3.25, 3.24) can also be thought of as the first term of the steady
state solution corresponding to a uniform hydration heat input together
with a periodic array (size 2) of matching sinks, see Figure 3. The complete
solution is obtained by adding image sinks in each of the neighbouring cells:
X
1
1
ln rij ) + TR (z),
Tc (x, y, z) = (1 r 2 ) + (ln r +
4
2
i,j
82
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
z
y
x
where
rij =
(x i)2 + (y j)2 , i, j 6= 0 .
X
1
1
ln rij ,
Tc (s, r) = Tw (s) + (1 r 2 ) + ln r +
4
2
i,j
s
where Tw (s) = T0 + 2
s is the temperature of the pipe water at a distance
s measured along the pipe from the inlet.
The second deficiency of the cylindrical model is that in practice the
pipes wind through the slab, and the flow in adjacent pipes will normally
be in opposite directions. The temperature continually increases as one
moves along a continuous pipe; this will tend to make the temperature
distribution in the concrete more uniform. To model this it is useful to notice
that our scaling results strongly suggest that to a high degree of accuracy
the temperature field will be determined by the requirement that all the
83
4.1
Tc (x, z)
O
Figure 4: A Water Network: The steady state concrete temperature increases linearly in x with superimposed periodic variations due to the piped
water.
In order to clarify the above network issue we will examine the simple 2D
network example illustrated in Figure 4. We have a network of 1D pipes
embedded in an infinite concrete slab of width L and height h. The water
flux through the network is Q0 per unit depth (into the page). After a
time scale of order t0 = h2 /c a quasi-steady equilibrium will be reached, as
described above, with the hydration heat being completely absorbed by the
network pipes, resulting in a linear increase in temperature along the pipe
from the entry point; thus the temperature along the pipe is determined to
84
J.P.D. Charpin, T.G. Myers,A.D. Fitt, N.D. Fowkes, Y. Ballim and A.P. Patini
be:
Tw (1, z) = T0 + z ,
Tw (2, z) = (T0 + h) + (h z) ,
Tw (3, z) = (T0 + 2h) + z ,
=
Tw (n, z) = (T0 + (n 1)h) + (h z) , (n even)
Tw (n + 1, z) = (T0 + nh) + z ,
where 1 < n < L/h refers to the nth column, see Figure 3. The quasi-steady
state (cell) equation is
Tc,xx = 1 + (x),
with quadratic solutions in the n cells given by
Tc (x, z) = Tw (n, z) + (x n)
1
(x n)2
2
for
n1< x < n+1 .
Figure 4 indicates the form of the temperature profile. The base state is a
linear increase across the block, superimposed on this is quadratic solution
which exhibits peaks in the temperature in the region furthest from the
pipes. Generally speaking this solution will not be in global equilibrium.
Certainly the temperature levels near x = 0 will be less than at x = L.
There will thus normally be a redistribution of heat driven by surface driving
conditions. The time scale for global equilibrium to be reached will be of
order L2 /c t0 . If the slab is insulated the above solution is compatible
with this requirement, so the solution is correct. If the surface temperature
around the block is for example required to be constant then the adjustments
will occur to accommodate this requirement. A perturbation procedure can
be used to determine the transient.
Conclusions
The basic aim of water cooling is to decrease the maximum temperature
reached in the concrete to an acceptable level. For the simple cylindrical
model we have obtained explicit expressions for the maximum concrete temperature in the concrete as a function of the driving parameters, and have
85
determined expressions for the pipe length and separation distance required
to limit the temperature rise in the concrete to a prescribed level. Our
calculations give a pipe length of 60m for a typical pipe size, flux level and
slab. The general results obtained for this model will carry through for more
realistic models. Preliminary investigations on more realistic networks have
been carried out and surprisingly analytic results can be obtained. When
combined with a financial model, optimum design parameters can be determined. This is ongoing work appropriate for a postgraduate student.
Acknowledgement
T.G. Myers acknowledges support of this work under the National Research
Foundation of South Africa grant number 2053289. J.P.F. Charpin acknowledges the support of the Claude Leon Harris Foundation.
References