Purdue University

Purdue e-Pubs
International Refrigeration and Air Conditioning
Conference

School of Mechanical Engineering

2002

A Universal Engineering Model For Cooling

Towers
L. Lu
Nanyang Technological University

W. Cai
Nanyang Technological University

Follow this and additional works at: http://docs.lib.purdue.edu/iracc

Lu, L. and Cai, W., "A Universal Engineering Model For Cooling Towers" (2002). International Refrigeration and Air Conditioning
Conference. Paper 625.
http://docs.lib.purdue.edu/iracc/625

This document has been made available through Purdue e-Pubs, a service of the Purdue University Libraries. Please contact [email protected] for
additional information.
Complete proceedings may be acquired in print and on CD-ROM directly from the Ray W. Herrick Laboratories at https://engineering.purdue.edu/
Herrick/Events/orderlit.html

R20-2
A UNIVERSAL ENGINEERING MODEL FOR COOLING TOWERS
*Lu Lu, Research Associate, Nanyang Technological University, Nanyang Avenue,
Singapore 639798; Tel.: (0065) 6790-5472; Fax: (0065) 6790-5471
E-Mail: [email protected]
*Author for Correspondence
Wenjian Cai, PhD, Associate Professor, Nanyang Technological University, Nanyang Avenue,
Singapore 639798; Tel.: (0065) 6790-6862; Fax: (0065) 6790-5471
E-Mail: [email protected]

ABSTRACT
This paper presents a universal engineering model, which can be used to formulate both counterflow and
crossflow cooling towers. By using fundamental laws of mass and energy balance, the effectiveness of heat
exchange is approximated by a second order polynomial equation. Gauss-Newton and Levenberg-Marquardt
methods are then used to determine the coefficients from manufactures data. Compared with the existing models,
the new model has two main advantages: (1) As the engineering model is derived from engineering perspective, it
involves fewer input variables and has better description of the cooling tower operation; (2) There is no iterative
computation required, this feature is very important for online optimization of cooling tower performance.
Although the model is simple, the results are very accurate. Application examples are given to compare the
proposed model with commonly used models.

NOMENCLATURE
Cpw: specific heat of water under constant pressure;
Cs: derivative of saturated ha with respect to Tw;
c0-c5: curve-fitting constants;
dk: search direction of optimization method;
Fdata: computation value of a;
ha: enthalpy of air;
hs,w: saturation air enthalpy at Tw;
hs,wb: saturation air enthalpy at Twb;
J(uk): Jacobian matrix of uk;
ma: mass flow rate of air;
me: mass flow rate of evaporation;
ma: mass flow rate of air;
mm: mass flow rate of makeup water;
mw: mass flow rate of water;
NTU: number of transfer units;
m*: ratio of air to water effective capacitance rate;
Qe: heat evaporation rate of loss water;
Qrej: heat rejection rate of cooling tower;
Tdb: dry-bulb temperature of air;
Tm: temperature of makeup water;
Tw: temperature of water;
Twb: wet-bulb temperature of air;
uk: the value of c0-c5 of the kth iteration;
a: heat transfer effectiveness of Brauns model;
h: enthalpy difference with respect to T;
T: approach of cooling tower;
k: control coefficients;
Subscript
i: inlet;
o: outlet;

INTRODUCTION
Cooling towers are commonly used to dissipate heat from heat sources to heat sink (ambient environment).
Their applications are typically in Heating Ventilation and Air Conditioning (HVAC) systems and power
generators, etc. Heat rejection of cooling towers is accomplished by heat and mass transfer between hot water
droplets and ambient air. Although cooling towers are relatively inexpensive and normally consume around ten
percent of the whole system energy, their operation has significant effect to the energy consumption of other
related subsystems (RMIRA 1995; Michel 1995). Therefore, optimizing cooling tower performance will not only

increase the tower efficiencies but also has direct effect to other subsystems. As such, there has been some
research interest in this area. Austin (1997) recommended regression methods to create the models of each
component in air conditioning systems for predicting and optimizing the system performance. Flake (1997)
utilized a different regression technique to determine parameters of the cooling tower model developed by Braun
(1989) and to build a predictive model for optimal supervisory control strategies. However, due to the lacking of
an effective and precise model for cooling towers, which is essential to estimate and verify the energy savings by
different optimization strategies, the research on optimization of cooling tower performance is still in the primary
stage.
Attempts to develop the cooling tower models have a relative long history, the first such work may trace
back to 1925, when Merkel developed a practical model for cooling tower operation, which has been the basis for
most modern cooling tower analyses. In his model, the water loss of evaporation is neglected and the Lewis
number is assumed to be one in order to simplify the analysis. However, as evaporate water cannot be neglected in
cooling tower operation, Merkels model is not accurate enough and not suitable for real applications. A more
rigorous analysis of a cooling tower model that relaxed Merkels restriction was given by Sutherland (1983). In
1989, Braun developed effectiveness models for cooling towers, which utilized the assumption of a linearized
air saturation enthalpy and the modified definition of number of transfer units. The models were useful for both
design and system simulation and has been adopted by the simulation software TRNSYS (SEL 2000). However,
Brauns model needs iterative computation to obtain the output results and is not suitable for online optimization.
Bernier (1994) reviewed the heat and mass transfer process in cooling towers at water droplet level and analyzed
an idealized spray-type tower in one-dimension, which is useful for cooling tower designers, but no much
information is provided to plant operators. Soylemez (1999) presented a quick method for estimating the size and
performance of forced draft countercurrent cooling towers and experimental results were used to validate the
prediction formulation. Unfortunately, this model also need iterative computation and not suitable for online
optimization.
In this paper, a universal engineering model, which can be used to formulate both counterflow and crossflow
cooling towers, is proposed. Extending the methods provided by Merkel and Braun and using fundamental laws of
mass and energy balance, the effectiveness of heat exchange is approximated by a second order polynomial
equation. Gauss-Newton and Levenberg-Marquardt methods are then used to determine the coefficients from
manufactures data. Compared with the existing models, the new model has two main advantages: (1) As the
engineering model is derived from engineering perspective, it involves fewer input variables and has better
description of the cooling tower operation; (2) There is no iterative computation required, this feature is very
important for online optimization of cooling tower performance. Although the model is simple, the results are
very accurate. Application examples are given for both counterflow and crossflow to compare the proposed model
with commonly used models.

COOLING TOWER MODEL ANALYSIS

The mechanism of heat and mass transfer between ambient air and condenser water inside a cooling tower is
illustrated in Figure 1.

h a , o ma ,o

T w ,i mw,i

evaporative water

Tdb,o m e
AIR

W ATER

Q& rej

T m mm
m akeup water

h a , i ma , i

Tw ,o mw,o

Figure 1. Schematic representation of heat and mass transfer in the cooling tower

Four governing equations can be used to express the mass and energy balance in the system:
(1) Mass conservation of air:
m a ,i + me = m a , o

(1)

(2) Heat conservation of air:

m a ,i ha ,i + Qrej Qe = m a ,o ha ,o

(2)

(3) Mass conservation of condenser water:

m w ,i me + m m = m w , o

(3)

(4) Heat conservation of condenser water:

m w,i Tw,i C pw Qrej + m m Tm C pw = m w,o Tw,o C pw

(4)

In the governing equations, there are nine known parameters including: six input variables, ha ,i , ma ,i , mw,i , mm ,
Tm , Tw,i ; a constant C pw ; and two measurable output variables mw,o , Tw,o , and five unknowns: three output
variables: ha ,o , ma ,o , and Qrej ; and two unmeasurable variables me and Qe . As the unknown variables are more
than the number governing equations, it is insufficient to determine all outlet conditions by the four governing
equations alone, additional equations that could depict the characteristics of the cooling tower should be added. In
Brauns model with effectiveness coefficient (1989), the derivative of saturation air enthalpy with respect to
temperature, Cs, is introduced and used to formulate the cooling load model. a is also added as a ratio of the
actual heat transfer amount to the theoretical maximum amount.
Qrej = a m a (hs , w,i ha ,i )
(5)
Analogous to a dry counterflow heat exchanger, the effectiveness, a, is evaluated by

a =
with NTU, m*, and Cs calculated respectively by

1 e

NTU 1 m*

1 m*e

NTU 1 m*

m
NTU = c a
mw
m C
m* = a s
m w C pw

(6)

(1+ n )

(7)
(8)

h hs ,w,o
dh
Cs = s
s , w ,i
(9)
Tw,i Tw,o
dT T = Tw
where, c and n are empirical constants specific to a particular tower design derived from the manufacturer. These
two parameters are correlated as a straight line on a log-log plot of NTU vs. the flow rate ratio. Since Cs depends
on outlet conditions of cooling tower Tw,o and hs,w,o , it cannot be computed directly. Consequently, the outlet
conditions of cooling tower need to be guessed initially at the reasonable values, and iterative computation is
engaged for Equation (1)-(9) to calculate the ultimate results.
Although Brauns model is more accurate than Merkels one, it also has several problems.
The computations are very complicated, it needs iterative computation to get the final results, and the
estimated outlet water temperature is needed before calculation;
It is hard to find the function derivatives, which are useful in real-time optimization analysis;
The model was derived based on mechanical principles, it only suitable for the counterflow cooling
towers. For the crossflow cooling towers, a different model is needed.

ENGINEERING MODEL DEVELOPMENT

Since the main difficulties in real-time application of Brauns model are the initial estimation of Cs and
highly nonlinearities of a, which resulted in a complicated and time consuming computation. To develop an
effective engineering model, lets analysis both Cs and a from fundamental laws of mass and energy balance.

Analysis Cs :

Water Oultlet
Ambient Air

(3)

Water Inlet
(1)

Air Saturation Enthalpy

In Brauns model (1989), a straight line between water inlet temperature and water outlet temperature on the
air saturation enthalpy with respect to temperature is used to approximate the curve between water inlet point and
water outlet point (Figure 2), where Cs is the ratio of length of line (1) hs ,w,i hs ,w,o to line (2) Tw,i Tw,o .

(2)
(4)

Saturation Temperature

Figure 2. Saturation air enthalpy vs. temperature

For control and optimization purpose, however, Tw,o and hs , w,o are output variables, which need to be
controlled, therefore, these two variables should not be used as input variables to calculate the heat rejection
ration. Instead, we may express Equation (9) with measurable variables as.
(h hs ,wb ) + f1 (h)
dh
= s , w ,i
Cs = s
(10)
(Tw,i Twb ) + f 2 (T )
dT T = Tw
where, T is the approach of the cooling tower and represents the difference between line (4) and line (2) in
Figure 2; h is the saturated air enthalpy difference with respect to T. By energy and mass conservation laws, the
m
approach, T, is a function of a and (Tw,i Twb ) , as the approach is affected by the mass flow rate of both air
mw
and water and the temperature difference between inlet water and ambient air. h can be considered as a function
m
of T, also the function of a and (Tw,i Twb ) . Therefore, Cs can be described as:
mw
m

C s = f 3 a , (Tw,i Twb )
m

(11)

Analysis a:
From the Equation (6), (7), and (8), it clear shows that the heat transfer effectiveness, a, is the function of
m
m
NTU and m* , where NTU is the function of a , and m* is the function of a and Cs. By Equation (11), Cs
mw
mw
m
is the function of a
mw

and (Tw,i Twb ) . Then, we can obtain a general expression for a as:

a = function a , (Tw,i Twb ) = function(x, y )

m

(12)

m
where x = a and y = (Tw,i Twb ) . The heat transfer effectiveness is the function of two variables, which are
mw
the inlet conditions of the cooling tower. As finding the exact function for Equation (12) is neither practical nor
necessary for real-time application, the following engineering solution is proposed.

Engineering model:
In order to solve the problem above, Taylors series expansion is used as an approximation of the unknown
function in Equation (12). It is clear that a is a continuous variable under normal operating conditions, its
derivative and high-order derivatives exist. Thus, we can apply Taylors series expansion for two variables into a
function. Because the characteristics of cooling towers are highly nonlinear, second-order Taylors series
expansion is used to better reflect the nonlinearity.
f (x0 , y0 )
(x x0 ) + f (x0 , y0 ) (y y0 )
f (x, y ) = f (x0 , y0 ) +
y
x

(13)
2
2
2

f (x0 , y0 )
f (x0 , y0 )
1 f (x0 , y0 )
2
2
(y y0 ) + 2
(x x0 )( y y0 )
(x x0 ) +
+
x 2
y 2
2!

xy

Where, (x0, y0) is any reasonable operating point of cooling tower near (x, y). Once the point (x0, y0) is determined,
f (x0 , y0 ) f (x0 , y0 ) 2 f (x0 , y0 ) 2 f (x0 , y0 )
2 f (x0 , y0 )
,
,
, and
can be treat as the constants.
x0, y0, f (x0 , y0 ) ,
,
2
2
x
x
y
y
xy
To express the equation in neat way, Equation (13) is rearranged and written as a function of two variables
form.
2

m
m
m
2
a = c 0 + c1 a + c 2 (Tw,i Twb ) + c3 a + c 4 (Tw,i Twb ) + c5 a (Tw,i Twb )
(14)
mw
mw
mw
Where, the coefficients, c0 - c5, are constants, and determined only by the cooling tower characteristics, which
depend on the towers structure and design.

ALGORITHMS FOR DETERMINING ENGINEERING MODEL

The real performance data of the cooling tower provided by manufacturers are used in our method. The
objective function is given as:
N 1
2
min ( function(c0 , c1 , c2 , c3 , c4 , c5 ) Fdatai )
(15)
i =1 2
where the function(.) is the right hand side of Equation (14) and the real performance data of cooling tower are
represented by Fdatai. N is the number of the sampling points. Fdatai can be derived from manufacturers data by
lookup-table or interpolation. In order to obtain accurate results, the number of sampling points must more than
that of coefficients, i.e. N > 5. Furthermore, the sampling points should be distributed evenly among the whole
range of operation.
Nonlinear least square method for curve fitting is used to solve Equation (15), both Gauss-Newton and
Levenberg-Marquardt methods are implemented in the optimization algorithms (Coleman et al. 1999). In GaussNewton method, a search direction dk is obtained at each major iteration step. The search direction is expressed as:
T
(16)
J (u k ) J (u k )d k = J (u k )F (u k )
where,

u = [c0 , c1 , c2 , c3 , c4 , c5 ] ;
uk is the u value of the kth iteration;
T

1
( function(u ) Fdatai )2 ;
2
T
F (u k ) = [F1 (u k ), F2 (u k )L, FN (u k )] ;
Fi (u k ) =

F1
c0 c0

c F2
J (u k ) = J 1 = c0

M M
c5 F
k N
c0

F1
c1
F2
c1
M
FN
c1

F1
c5
F2
L
c5
O
M
FN
L
c5
L

J(uk) is the Jacobian matrix with respect to uk . In the case of H(uk) (Hessian Matrix of Fi(uk)) is significant,
Levenberg-Marquardt method is adopted. It uses a search direction between the Gauss-Newton direction and the
steepest descent. This makes it less effective but more robust than the Gauss-Newton method. The LevenbergMarquardt method is given by
T
(17)
J (u k ) J (u k ) + k I d k = J (u k )F (u k )
In this equation, k controls both magnitude and direction of dk . When k is zero, the direction dk is identical to
that of the Gauss-Newton method. As k tends to infinity, dk tends towards a vector of zeros and a steepest descent
direction.

Remarks
1. In this method, the coefficients c0 - c5 are determined offline by curve fitting in the whole operating
range. Therefore, the real-time output calculation is straightforward once the input variables are
measured.
2. For more accurate results, it is possible to construct a look up table for coefficients c0 - c5 by dividing the
whole operating range into sub-regions. One set of coefficients is selected at one time according to the
cooling tower operation conditions.
3. In Bruans model, both NTU and a are exponential functions, which requires substantial computing
effort. Whereas, in the new model a is in a polynomial form which is much easier to calculate and
suitable for on-line optimization.
4. For crossflow cooling towers, the analysis is almost same except Equation (6), which takes the following
form according to the heat exchange principle.
m* 1 e NTU
1

(18)
a = * 1 e

However, this change will not affect the model structure. The differences of the different cooling tower
models are determined by coefficients of Equation (14). Therefore, both counterflow and crossflow
cooling towers can be represented by the same model.
5. In practice, it is very hard to measure the inlet and outlet airflow rate ( ma ,i and ma ,o ) accurately. This
problem could be solved as follows:
Using energy conservation principle, we can replace ma (hs , w,i ha ,i ) by m w C pw (Tw,i Twb ) ;

Writing

a =

m w,i C pwTw,i + m m C pwTm m w,o C pwTw,o

m w,o C pw (Tw,i Twb )

(19)

in Equation (19), according to the known variables: a , Tw,i , Twb , and m w,o , Equation (14) is again used
inversely to find the value of the mass airflow rate, m& a . The value will then be employed to determine
the overall heat rejection rate at the next sample time.

MODEL VALIDATION
To validate the proposed model, the outputs of new model are compared with model provided by Braun
(1989). Both counterflow and crossflow design cooling towers are used to illustrate its universeness. The
parameters of cooling tower are given in following:
Air flow rate:
10.7-32.7 kg/s (1.41-4.32*105 gpm);
Water flow rate:
21.7 kg/s (344 gpm);
Inlet water temperature:
38C (100.4F);
Ambient dry-bulb temperature:
35C (95F);
Ambient wet-bulb temperature:
21-31C (69.8-87.8F);
c in Equation (7):
2.3 (dimensionless);
n in Equation (7):
-0.72(dimensionless).

KHDWWUDQVIHUHIIHFWLYHQHVVGLPHQVLRQOHVV

For the counterflow cooling tower, the heat transfer effectiveness, a ,varied with mass flow ratio of air to
water and the ambient wet-bulb temperature shown in figure 3.

(D
&XUYH
(D
&XUYH












PDVVIORZUDWHUDWLRRIDLUWRZDWHUGLPHQVLRQOHVV

Figure 3. Comparison of heat transfer effectiveness of two models








+HDWUHMHFWLRQRIQHZPRGHO

+HDWUHMHFWLRQRIQHZPRGHONZ

where Ea1 and Ea2 are heat transfer effectiveness of Brauns model in ambient wet-bulb temperature 26C and
30C respectively. Curve1 is the heat transfer effectiveness given by the new model under 26C wet-bulb
temperature, and Curve2 under 30C wet-bulb temperature. According to the figure, the results of two models are
almost same.






















+HDWUHMHFWLRQRI%UDXQ
VPRGHONZ



Figure 4. Counterflow cooling tower models







+HDWUHMHFWLRQRI%UDXQ
VPRGHO



Figure 5. Crossflow cooling tower models

The results of heat rejection for the counterflow and crossflow cooling towers are shown in figure 4 and 5
respectively. There are totally 121 points on each figure. From the figures, it is clearly that the new model can
predict the performance of cooling tower very well.

CONCLUSIONS AND FUTURE WORK

The new engineering model for cooling towers, which can be used to formulate both counterflow and
crossflow cooling towers, has been presented in this paper. The methods of Merkel and Braun and fundamental
laws of mass and energy balance are used to develop the effectiveness of heat exchange with polynomial form.
Nonlinear least square curve-fitting methods are used to determine the coefficients of the model. Some
engineering considerations are also discussed. The comparison study of existing and the new model is given to
show that the new model can predict the performance of both counterflow and crossflow cooling tower accurately
with less computation. As the manufacture data are used to determine the coefficients for the model, it is predicted
that it should have better performance compared with the existing ones.
In practice, many unpredictable factors affect the performance of the cooling towers, such as outdoor airflow
rate, interior problems of cooling tower, and measurement errors, etc. Therefore, the coefficients of cooling tower
model may not be constant during the operational life span. Fault detection or adaptive scheme should be added to
accommodate these changes; these aspects are also subject to future study. Utilizing the model for on-line
optimization of both cooling towers and chillers as well as condenser water loop for HVAC systems is currently
under study; the research results will be published soon.

REFERENCES
Ahmed, O., DDC applications in variable-water-volume systems, ASHRAE Transactions, Vol. 97(1), pp. 751758, 1991.
ASHRAE, ASHRAE HandbookHVAC Systems and Equipment, American Society of Heating, Refrigerating, and
Air-Conditioning Engineers Inc., 1996.
Austin, S.B., Regression analysis for savings verification, ASHRAE Journal, Vol. 39, No. 10, pp. 57-63, 1997.
Bernier, M.A., Cooling tower performance: theory and experiments, ASHRAE Transactions, Vol. 100(2), pp.
114-121, 1994.
Braun, J.E., Klein, S.A. and Mitchell, J.W., Effectiveness models for cooling towers and cooling coils, ASHRAE
Transactions, Vol. 95(2), pp. 164-174, 1989.
Burger, R., Cooling Tower TechnologyMaintenance, Updating and Rebuilding, Fairmont Press Inc., 1995.
Coleman, T., Branch, M.N. and Grace, A., Optimization Toolbox for Use with MATLAB, MathWorks Inc., 1999.
Flake, B.A., Parameter estimation for multiresponse nonlinear chilled-water plant models, ASHRAE Transactions,
Vol. 97(1), pp. 470-485, 1997.
Merkel, F., Verdunstungskuhlung, VDI Forschungsarbeiten, No. 275, Berlin, 1925.
Michel, A.B., Thermal performance of cooling towers, ASHRAE Journal, Vol. 37, No. 4, pp. 56-61, 1995.
Mick, S., Take it to the limit or just halfway? ASHRAE Journal, Vol. 40, No. 7, pp. 32-39, 1998.
Rocky Mountain Institute Research Associates, Commercial Space Cooling and Air Handling, Technology Atlas, E
Source Inc., 1995.
Solar Energy Laboratory, Reference Manual of TRNSYS, University of Wisconsin-Madison, 2000.
Soylemez, M.S., Theoretical and experimental analyses of cooling towers, ASHRAE Transactions, Vol. 105(1),
pp. 330-337, 1999.
Sutherland, J.W., Analysis of mechanical draught counterflow air/water cooling towers, Journal of Heat Transfer,
Vol. 105, pp. 576-583, 1983.