Applied Thermal Engineering 105 (2016) 519–530

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Research Paper

A feedback model to predict parameters for controlling the performance

of a mechanical draft cooling tower q
Kuljeet Singh, Ranjan Das ⇑
Department of Mechanical Engineering, Indian Institute of Technology Ropar, Punjab 140001, India

h i g h l i g h t s

A feedback and control model for the forced draft cooling tower performance.

Model simulation is presented for three different cases under varying heat load.

Useful for solar cogeneration, ORC, diesel engine and HVAC applications.

GA optimized results with feedback control save pump and blower power.

Performance parameters also improve with feedback.

a r t i c l e i n f o a b s t r a c t

Article history: In this work, a feedback model has been proposed in order to control the performance of a mechanical
Received 24 December 2015 draft cooling tower under varying heat load conditions suiting diverse applications such as solar power
Accepted 3 March 2016 generation, HVAC and diesel engine. The model uses specific objective functions to meet the operating
Available online 21 March 2016
conditions related with the process requirement under varying heat load. These objective functions have
been formulated using empirical correlations corresponding to different performance parameters devel-
Keywords: oped from experimental data. Thereafter, an inverse method-based feedback model is made to control
Forced draft
the cooling tower’s performance for different outputs. The simulated model considers different cases clo-
L/G ratio
Inverse optimization
sely relating to practical use of the cooling tower. Further, by monitoring the inlet water temperature
Feedback with time, the controlling variable governed by water to air ratio has been adjusted to optimize the tower
Tower control operation. It is concluded that for different situations, the proposed inverse feedback model is an opti-
mized tool to control the tower performance.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction Similar literatures covering the optimization part of cooling towers

Cooling towers are used in several industrial processes involv-
ing power generation, petrochemical, refrigeration and air-
conditioning and many more [1]. Therefore, the performance of
the cooling tower needs to be optimally in line with the process
to which it is coupled. Considerable amount of studies addressed
the optimization issue of cooling towers. A work estimating the
optimum value of heat and mass transfer area has been presented
by Söylemez [2] to minimize the cost. Further, the thermo-
hydraulic performance has been optimized for counter flow tower
[3]. Cortinovis [4] minimized the single objective of operation cost.
q
The ISIRD grant received from Indian Institute of Technology Ropar, Punjab for
this work is duly acknowledged.
⇑ Corresponding author. Tel.: +91 1881 242245; fax: +91 1881 223395.
E-mail address: ranjandas81@gmail.com (R. Das).
for various performance aspects are available [5–7].
The above discussion suggests that the performance control of a
cooling tower is an important objective. Although, several studies
optimizing the tower performance have been reported, but its con-
trol for desired performance in line with connected system also
needs to be addressed. For example, if a cooling tower operates
under certain ambient conditions, then it is possible that the heat
load on the cooling tower imposed by a certain process may be
continuously varying. Such systems include low-grade power gen-
eration cycles such as solar power generation, Organic Rankine
Cycle (ORC) and other low temperature operating cycles in which
the heat input may vary depending upon the supply of waste heat
[8–11]. Therefore, to control the performance of cooling tower, a
feedback model is proposed based on the inverse method that
monitors the varying heat load and adjusts the water to air ratio
for satisfying the desired output. Firstly, the performance of the

http://dx.doi.org/10.1016/j.applthermaleng.2016.03.030
1359-4311/Ó 2016 Elsevier Ltd. All rights reserved.
520 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530

Nomenclature

A approach ð CÞ   u uncertainty  
a interfacial area m2 =m3 V volume of exchange core m3
cp specific heat at constant pressure ½kJ=ðkg  CÞ v independent variable in uncertainty analysis
c coefficient
e effectiveness (%) Greek symbols
F objective function x specific humidity ðkg=kg of dry airÞ
G air flow rate ðkg=sÞ rT standard deviation
H head ðmÞ
h specific enthalpy ðkJ=kgÞ Subscripts
0
h saturated air enthalpy at bulk water temperature a air
ðkJ=kgÞ    abs absolute
K mass transfer coefficient kg= m2 s cr. current
L water flow rate ðkg=sÞ ex. exact/required
Me tower characteristic ratio fg fluid–gas mixture
meV water evaporation rate ðkg=sÞ
g guessed value
N number of samples i inlet
n number of independent variables k index for iteration
P power ðkW  Þ  o outlet
p pressure N=m2 wb wet bulb

R range ð CÞ v vapour
T temperature ð CÞ

cooling tower has been estimated at different levels of heat loads ferent points are recorded using DAQ system. Performance param-
(inlet water temperature) and water to air ratios. Then, empirical eters are explained next.
correlations for performance parameters have been developed
from experiments. These correlations are then used in relevant 3. Performance characteristics and empirical correlation
objective functions, which have been formulated to get a desired
output from the tower under varying heat load conditions. In the In this work, the objective is to establish a feedback system
next section, the experiment setup and procedure have been based on inverse method to achieve a desired performance from
discussed. a forced-draft cooling tower. So, first the important performance
parameters at different levels of T L;i and L=G ratio are estimated
2. Experimental setup and procedure as discussed below.
½1 Range : R ¼ T L;i  T L;o ð1Þ
A counter flow forced-draft cooling tower with wire mesh pack-
ing (Fig. 1) is studied having dimensions 0:30  0:30  1:35 m with Depending upon the process, a desired T L;o is needed for differ-
1:2 m fill height. Warm water from the tank is supplied at top of ent chemical processes [12] For example, in ORC, the condenser
the tower by a centrifugal pump through a gate valve and rotame- should be operated at a particular temperature difference between
ter having range 0—11 LPM: Water is distributed over the packing working fluid and cooling water to ensure optimum performance
and droplets of water come into contact with air supplied from the [13]. Moreover, low water temperature is advisable in applications
bottom by means of a centrifugal blower. Air flow rate is measured such as HVAC, but in engines, relatively higher temperatures are
using orifice plate alongwith U-tube manometer containing water required. It is found that low cold water temperature also increases
as working fluid. The cold water is again collected in tank, where the total cost of the tower [14].
water is heated and re-circulated. To maintain a desired tempera- ½2 Approach : A ¼ T L;o  T wb;i ð2Þ
ture in tank, a heater of 2 kW is installed in the tank that is pow-
ered through a digital temperature controller. After calibration, The operating cost of the cooling tower is inversely proportional
six K-type thermocouples are used to measure the temperature to approach [14]. So, to minimize the cost of the entire system, a
at different locations i.e. water inlet, water outlet, DBT and WBT balance between range and approach needs to be maintained.
at air inlet and outlet. The data acquisition system (DAQ) of Range T L;i  T L;o
National Instruments is used to record the temperature data from ½3 Effectiveness : e ¼ ¼ ð3Þ
Range þ Approach T L;i  T wb;i
thermocouples. Experimental procedure is described below.
L=G ratio and T L;i are two major factors which strongly govern During heat transfer process within tower, water droplet is sur-
the cooling tower performance. In this study, five levels of rounded by film of air. The enthalpy difference between film and
T L;i ð36:4—45:3  CÞ have been maintained, which are in line with air provides driving force for cooling. The tower characteristic ratio
different applications such as HVAC, petrochemical industry and is expressed as [7,15],
other plants [12]. At each T L;i , nine levels of L=G ratio ð1—5Þ are Z T L;i
KaV cp;L dT
used, which are widely used in different processes. First, the water ½4 Tower characteristic ratio : Me ¼ ¼ 0 ð4Þ
L T L;o h  hG
tank heater is lighted through a digital temperature controller hav-
ing desired setting. After achieving the desired water temperature, Eq. (4) is solved with Chebyshev four point method [16],
pump and air blower are powered. The warm water supplied at top Z
is cooled by air entering from the bottom through sensible and KaV T L;i
cp;L dT T L;i  T L;o X4
1
¼ 0 ffi cp;L ð5aÞ
evaporative cooling, which is collected in tank, where it is again L T L;o h  hG 4 j¼1
D hj
heated and re-circulated. Under steady state, temperatures at dif-
 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530 521

Fig. 1. Schematic diagram of cooling tower setup.

where where pv ;wb ¼ 10 z N=m2 ð8bÞ

 
0
Dh j ¼ hj  hGj ð5bÞ 
T wb
  z ¼ 10:79586 
0 T wb þ 273:15
h1 calculated at T L;o þ 0:1 T L;i  T L;o and hG1 
  273:15
¼ hG;i þ 0:1 hG;o  hG;i ð5cÞ þ 5:02808  log 10
T wb þ 273:15
  
0   4
T wb
h2 calculated at T L;o þ 0:4 T L;i  T L;o and þ 1:50474  10 1  10 8:29692273:15
 
hG2 ¼ hG;i þ 0:4 hG;o  hG;i ð5dÞ  
T wb
þ 4:2873  104 4:76955  1
0   T wb þ 273:15
h3 calculated at T L;o þ 0:6 T L;i  T L;o and
  þ 2:786118312 ð8cÞ
hG3 ¼ hG;i þ 0:6 hG;o  hG;i ð5eÞ
  In Eq. (6), hfg can be evaluated using,
0
h4 calculated at T L;o þ 0:9 T L;i  T L;o and
 
hG4 ¼ hG;i þ 0:9 hG;o  hG;i ð5fÞ hfg ¼ 3:4831814  106  5:8627703  103 ð273:15Þ

The enthalpy of the air can be estimated from below [7,17], þ 12:139568  273:152  1:40290431
   
hG ¼ cp;G ðT G Þ þ x hfg þ cp;v ðT G Þ ð6Þ  102 273:153 ð9Þ

where cp;G ðairÞ and cp;v ðwater vapourÞ are calculated at  

T ¼ ½ðT G þ 273:15Þ þ 273:15=2 as mentioned below, ½5 Water evaporation rate : meV ¼ G xG;o  xG;i ð10Þ
3 1
cp;G ¼ 1:045356  10  3:161783  10 ðTÞ þ 7:083814 After estimating different performance characteristics at each
4 2 7 3 levels of T L;i and L=G ratio; the correlation for each parameter has
 10 ðT Þ  2:705209  10 ðT Þ ð7aÞ
been proposed. The 3rd order polynomial is used to obtain the fit
for all data points, with T L;i and L=G ratio being considered as
cp;v ¼ 1:3605  103 þ 2:31334  ðTÞ  2:46784  1010 ðT 5 Þ
variables,
þ 5:91332  1013 ðT 6 Þ ð7bÞ    
f L=G; T L;i ¼ c0 þ c1 ðL=GÞ þ c2 T L;i þ c3 ðL=GÞ2 þ c4 ðL=GÞT L;i
and x can be calculated using,
   ! þ c5 T 2L;i þ c6 ðL=GÞ3 þ c7 ðL=GÞ2 T L;i þ c8 ðL=GÞT 2L;i
2501:6  2:3263  ðT wb Þ 0:62509  P v ;wb  
x¼  
2501:6 þ 1:8577  ðT db Þ  4:184  ðT wb Þ Pabs  1:005  P v ;wb þ c9 T 3L;i ð11Þ

1:00416  ðT db  T wb Þ
 In this way each performance parameter can be easily related to
2501:6 þ 1:8577  ðT db Þ  4:184  ðT wb Þ
T L;i and L=G ratio. The inverse feedback model has been explained
ð8aÞ
below.
522 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530

4. Inverse feedback model
In actual situation, the heat load imposed on cooling tower by
an industrial system may vary continuously with time. This vary-
ing load demands a feedback system that controls the tower for
a desired output. L=G ratio is a parameter that can be easily con-
trolled to get desired performance. In this study, three different
cases are considered to address the cooling tower operation under
varying heat load. The feedback model is based upon inverse opti-
mization that uses certain objective functions (as per require-
ment), which are formulated using above empirical correlations.
4.1. Objective functions
Case-1 and 2: First case has been taken for the situation where a
constant T L;o is required under continuously varying
36:4  C 6 T L;i 6 46:3  C. The below objective function is formulated
to estimate required L=G ratio to get a constant T L;o ,
h i2 h n  
F 1 ¼ ðT L;o Þex:  ðT L;o Þg ¼ ðT L;o Þex:  T L;i  R T L;i ; L=G
Constraints : 1 6 L=G 6 5
 
where R is function of T L;i ¼ T L;i cr: and L=G ratio; whereas, ðT L;o Þex:
is the required constant outlet water temperature. The constraints
are driven by lower and upper limits of T L;i and L=G ratio for this
setup. T L;i may continuously vary with time (as heat load varies with
process), hence it needs to be updated regularly in the above objec-
tive function. Then objective function is minimized using an opti-
mization algorithm to obtain required L=G ratio for desired
ðT L;o Þex: . In Case-1 the objective is to obtain constant ðT L;o Þex: over a
period of cooling tower operation with varying heat load for solar
and ORC cogeneration systems [9,13]. Case-2 considers a process
where heating load is continuously increasing with time and outlet
water temperature is required to be gradually decreased with it.
Case-2 can be better understood keeping I.C. engine application in
mind, where at lower engine temperature, the warmer water is
needed to be supplied, whereas, with increasing engine tempera-
ture, the water temperature needs to be lowered [18]. Therefore,
in both cases, the desired outlet water temperature is required with
respect to time and the same has been obtained by continuously
adjusted L=G ratio through inverse feedback system (Eq. (12)). Fur-
thermore, it is possible that different combinations of L and G sat-
isfy a particular L=G ratio: So, depending upon the requirement,
this model also enables the option to select an optimum combina-
tion of L and G for a desired L=G ratio that results in minimum
power consumption. To get multiple combinations of L=G ratio:
the following objective function is formulated,
"  #2
L L
F2 ¼ 
G g G ex:
ð13Þ
 oi2 Constraints : 0:2183 kg=s 6 L 6 0:109 kg=s
g 0:0173 kg=s 6 G 6 0:217 kg=s
To minimize F 2 , the Genetic Algorithm (GA) is used because it
ð12Þ intelligently searches the variables satisfying a desired objective
function without violating the above constraints (lower and upper
limits of water and air flow rates). After each run of GA, different
combinations of variables ðL and GÞ are obtained by minimizing
the objective function ðF 2 Þ. Finally, the combination corresponding
to minimum power consumption can be selected. Power consump-
tion expression is represented below,
Lðor GÞgH
P¼ kW ð14Þ
1000

Enable optimum
User input
combination option with case 1 or 2

Case 1 Case 2 Case 3

constant TL, o ex. decreasing TL, o ex.with optimum operation
increasing, TL, i

Select objective function 2 Select objective function

F1 TL, o TL, i R TL, i , L F3 R A
2
ex. g G g

Update current TL , i and required TL, o ex.

Update current TL, i
Get multiple
Generate Initial Population combinations using
2
L L
F2
Evaluate Fitness G G ex.
TL , i TL , i Minimize
cr.
objective
Termination Yes Best Select optimum L and
function
Condition? Individual G using power
using GA
No consumption
Select Parents Adjust
Crossover,
Mutation L and G
Generate New Offspring

Yes
Adjust Cooling tower
Required L/G ratio
No L/G ratio
abs. TL, i TL, i 0? TL , i
cr. cr.

Fig. 2. Flow chart of inverse feedback model.

 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530 523

For the water-pump, lift height of 1:80 m is considered as the of cooling tower or by decreasing the approach. But, a decreased
required head, whereas for blower, head is taken as the tower approach enhances the energy consumption due to increased fan
height ð1:35 mÞ. power [14]. Therefore, a balance needs to be maintained between
Case-3: As explained earlier, the cooling tower is always used in A and R which is possible at A ¼ R. Consequently, for a given
conjunction with other systems like HVAC, chillers and many 36:4  C 6 T L;i 6 46:3  C appropriate determination of L=G ratio is
more. The power consumption of the system to which cooling governed by,
tower feeds water strongly depends upon the cold water tempera-
F 3 ¼ ½R  A2
ture. In applications like condensers of HVAC, low cold water tem- ð15Þ
perature is desirable to save the energy consumption [19]. Low Constraint : 1 6 L=G 6 5
water temperature can be obtained by increasing either the range

Table 1
Uncertainties involved in different measurements and performance parameters.

Measurements
Absolute uG ¼ 1  104 kg=s uL ¼ 0:002 kg=s uT ¼ 0:007  C
uG uL uT
G ¼ 0:01 L ¼ 0:09 ¼ 0:0002
Relative T

Performance parameters
Absolute uR ¼ 0:01  C uA ¼ 0:01  C ue ¼ 0:07% uMe ¼ 0:003 umeV ¼ 4  106 kg=s
uR uA ue uMe umeV
Relative R ¼ 0:003 A ¼ 0:002 e ¼ 0:003 Me ¼ 0:003 ¼ 0:006
meV

TL,i = 36.4°C TL,i = 38.7°C TL,i = 40.3°C TL,i = 43.5°C TL,i = 45.3°C

1. 0 1.5 2. 0 2.5 3. 0 3.5 4. 0 4.5 5. 0

14
12
Range, R ( C)

10
8
6
4
2
16
Approach, A ( C)

14
12
10
8
6
Effectiveness, e ( )

60

50

40

30
Tower characteristic

1.36
ratio, Me

1.02

0.68

0.34
-4
7.70x10
Water evaporation
rate, meV (kg/s )

-4
6.60x10

-4
5.50x10

-4
4.40x10

1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

L/G ratio

Fig. 3. Variation of performance parameters with L=G at different T L;i .

524 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530

Fig. 4. Proposed correlation surface fitting for different parameters.

 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530 525

where R and A are functions of T L;i and L=G ratio; At a given T L;i ,rel-

5:977  107
evant L=G ratio at the intersection point serves as the optimal

9:85  103
5:51  104
4:09  103

1:169  102
because at that point neither the range nor the approach is compro-
mised, which in turn results in overall energy saving. In this case
too, the inverse feedback system can be implemented to operate

c9
the cooling tower at the optimum L=G ratio under varying heat load
 
or T L;i . Fig. 2 depicts actual implementation of the inverse based

5:324  103

1:316  103

2:286  107
1:46  103

1:894  103
feedback model under continuously varying T L;i . The user can select
the appropriate case depending upon the requirement and conse-
quently the corresponding objective function is chosen. Then, the

c8
values of T L;i and required outlet water temperature ðT L;o Þex: are
updated in the desired objective function, which is further mini-

1:895  103
2:281  102
1:034  103

2:08  106
1:111  102
mized using GA to get the required value of L=G ratio; so that the
cooling tower operation can be controlled by adjusting L=G. More-
 
over, current inlet water temperature T L;i cr: with respect to time

c7
is continuously monitored and absolute difference

 

e ¼
T L;i cr:  T L;i
is calculated regularly. If e ¼ 0, same L=G ratio is

8:925  102
4:137  105

4:669  106
maintained, whereas, if e > 0, then the feedback loop recalculates

4:552  102
the required L=G ratio to meet the desired objective. Furthermore,

0.3936
if the option of selecting optimum L=G ratio is selected
ðinvolving F 2 Þ, then the desired loop would be followed as shown

c6
in Fig. 2. Next, details of GA have been explained.

7:626  105
7:4  102
4.2. Genetic algorithm (GA)

1.465
0.5299
1.226
c5
GA falls under the category of evolutionary methods that is
based upon biological evolution, which repeatedly modifies the

7:74  102
3:74  105
population using biological rules [20]. It initially creates a random
population within lower and upper limits and satisfying searching

0.2644
0.3077
0.124
constraints that corresponds to number of possible solutions in

c4
binary strings of 10 bits (either 0 or 1). Then binary strings are con-
verted into real number using below relation,

1:146  104
ðUpper Limit  Lower LimitÞ
0.6184

1.259
L=G ¼ Lower Limit þ  

0.1195
210  1 0.9901
c3

 ðdecimal equivalent of binary coded numberÞ ð16Þ

The set of such solution (individual) is referred as population 3:233  103

21.83

3.322
50.58

(50 in this work) and fitness function is evaluated for this popula-
61.33

tion. In case of unsatisfied termination condition, two GA operators

c2

known as crossover and mutation are used generate the new pop-
Coefficients used to develop the correlation for different performance parameters.

ulation. The termination condition is assumed when the average

1:147  103

change in the objective function, F of 50 consecutive iterations

becomes either less than or equal to the order, O (106). For cross-
9.279
3.322
8.172
3.906

over, a random number is generated for each string and compared

c1

with crossover probability (0.8 in present work), if it is less than

crossover probability, the bits at random location are exchanged
4:59  102

as represented,
784.5
299.3
696.6
51.44

Parents : 110 0111000 10 Children ðafter crossoverÞ 110 1101010 10

c0

010 1101010 00 010 0111000 00

ð17Þ
R value

0:9936
0:9829
0:9914
0:9862
0:9940

The selected portions (within the quote) of parent string pair

2

are exchanged after crossover to generate children. After crossover,

a random number is again generated, if it is less than mutation
Tower characteristic ratio, Me

Evaporation rate, meV ðkg=sÞ

probability (0.03 in this problem), then mutation of string will be

Performance parameter

carried out flipping a single bit,

Effectiveness, eð%Þ

Before mutation 11110100 10 10

Approach, Að CÞÞ

ð18Þ
After mutation 11110100 00 10
Range, Rð CÞ
Coefficient

In this way, over the number of generations, the population

Table 2

evolves toward an optimal solution till the termination condition

remains unsatisfied. Each run of GA may result in multiple
526 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530

Table 3
Sensitivity of different correlations to uncertainties.

Performance parameter R A Me e meV

True value 6.991 8.537 0.768 45.614 4:318  104
Considering uT 6.987 8.533 0.768 45.611 4:316  104
Error 4:067  103 4:425  103 8:836  105 2:274  103 2:336  107
Considering uL 7.179 8.346 0.794 46.793 4:265  104
Error 0.188 0.191 0.026 1.179 5:31  106
Considering uG 6.967 8.561 0.765 45.466 4:325  104
Error 2:34  102 2:386  102 3:166  103 0.147 6:62  107
Considering all uncertainties 7.152 8.365 0.791 46.647 4:269  104
Error 0.161 0.172 0.023 1.034 4:89  106

with inverse feedback without inverse feedback

2 4 6 8 10 12 14 16 18 20
46
Temperature ( C)

44
TL,i
42

34

33

2.4
L/G ratio

2.2

2.0
54
Effectiveness, e ( )

52

50

48

46
Tower characteristic

0.874
ratio, Me

0.836

0.798

0.760
-4
Water evaporation

5.95x10
rate, meV (kg/s )

-4
5.60x10

-4
5.25x10

-4
4.90x10
2 4 6 8 10 12 14 16 18 20
Time (min)

Fig. 5. Variation of different parameters with time for the cooling tower with and without inverse feedback (Case-1) (dashed and solid lines represent outlet water
temperature, T L;o in the first sub-figure).

combinations of variables optimizing particular objective function, 5. Experimental uncertainties

because the algorithm initiates with a random population and
keeps on updating it to get an optimal solution. Next, the uncer- Experimental results always involve some uncertainty due to
tainty analysis is presented. different instruments, therefore, an uncertainty analysis is
 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530 527

required [21]. Let n is the final estimated result based upon n mea- temperature uncertainty (0.007 °C) is in line with available
sured variables, literature, which is 0.01 °C [23]. Next section covers the results
and discussion part.
n ¼ f ðv 1 ; v 2 ; v 3 ; . . . ; v n Þ ð19Þ
6. Results and discussion
Further, for each of the measured variable v, if the correspond-
ing uncertainty is u, then the contribution of each uncertainty in The experimental results for different performance parameters
final result is calculated as follows, are presented in Fig. 3. The range, R decreases with increase in
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi L=G ratio for each T L;i , because, increase in L=G ratio infers that L
 2  2  2  2
un 1 @n 1 @n 1 @n 1 @n increases for a given G. Therefore, the amount of air available for
¼ u1 þ u2 þ u3 þ  þ un
n n @v 1 n @v 2 n @v 3 n @v n water decreases, which directly decreases heat transfer between
two fluids [24]. Moreover, with the increase in T L;i ; R increases,
ð20Þ
because, as T L;i increases, the heat transfer potential between the
The uncertainties involved in each measurement and different warm water and cold air also increases. Further, if approach, A is
estimated performance parameters have been presented in Table 1. considered, for different values of T L;i , at a low L=G ratio, more air
pffiffiffiffi
The standard deviation of mean rT = N has been used to calculate is available for given amount of water that enables the tower to
 
the temperature uncertainty [22]. The temperature is recorded supply water at lower outlet temperature T wb;i . Therefore, A
after each 0.1 s, therefore high sampling rate results in low is less with lower L=G, which increases with L=Gratio. Moreover,
relative uncertainty value, whereas, the present value of absolute A increases with T L;i , because if T L;i increases at any value of

with inverse feedback without inverse feedback

2 4 6 8 10 12 14 16 18 20
44
42
Temperature ( C)

40
38 TL,i

34
32

4
L/G ratio

60
Effectiveness, e ( )

50

40

30
Tower characteristic

1.08
ratio, Me

0.81

0.54

-4
5.1x10
Water evaporation

-4
rate, meV (kg/s )

4.8x10
-4
4.5x10
-4
4.2x10
-4
3.9x10

2 4 6 8 10 12 14 16 18 20
Time (min)

Fig. 6. Variation of different parameters with time for the cooling tower with and without inverse feedback (Case-2) (dashed and solid lines represent outlet water
temperature, T L;o in the first sub-figure).
528 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530

L=G ratio; T L;o also increases. Further, as effectiveness, e is the ratio
of R to the difference of T L;i and T wb;i , it also decreases as L=G ratio
increases. Trends of Me reveals that it decreases with increase in
L=G ratio, because less air promotes the heat transfer resistance
and lowers Me [7]. Moreover, Me decreases as T L;i increases, which
0 model as discussed earlier.
is due to increased air saturation enthalpy, h corresponding to bulk
water temperature (because of increased inlet water temperature),
0
which ultimately rises h  hG . Evaporation rate, meV first increases
with increasing L=G ratio and then decreases, because, as L=G ratio
increases, the water temperature over the tower height increases,
resulting in increased meV . But, at increased L=G, the amount of
air available for a given amount of water becomes comparatively
less, so, the interfacial area between the water droplet and air
decreases, thereby lowering meV . Using 3rd order polynomial, the
fitted surfaces for different parameters have been presented in
Fig. 4 with relevant coefficients being mentioned in Table 2. The
goodness of each correlation is also presented in Table 2, which
indicate that the R2 value is greater than 0.98 for each correlation.
The sensitivity of each correlation to uncertainties in various mea-
surements has been revealed in Table 3. Each correlation is more
sensitive to the uncertainty involved in water flow rate ðuL Þ as
compared to other parameters, because the relative uncertainty
in L is more as compared to other measured quantities. Next, these
correlations have been used in the inverse method-based feedback
For demonstrating the inverse feedback model, results are pre-
sented for three different cases as discussed in Section 4. Fig. 5 cov-
ers the detailed analysis for Case-1, where a constant, T L;o is
required against continuously varying T L;i ðor heat loadÞ. In this
example, the tower is required to supply water at a constant tem-
perature ðT L;o Þex: ¼ 33:7  C; where T L;i continuously varies ran-
domly with time. If tower operates without any feedback system
ði:e:; at fixed L=G ¼ 2:25Þ, it fails to deliver the water at desired
temperature ðT L;o Þex: . The cooling tower equipped with feedback
system continuously senses T L;i and adjusts L=G ratio by minimiz-
ing the function, F 1 . The update of L=G ratio against time is also
shown, that enables the tower to supply the water at desired tem-
perature, i:e:; T L;o ¼ ðT L;o Þex: . It is also found that the cooling tower

5 10 15 20 25 30 35 40 45 50 55 60
2.47
Case 1
2.46

2.45 TL,i = 44.5 C; TL,o = 33.7 C

L/G ratio

2.12 TL,i = 42 C; TL,o = 33.7 C

2.10
2.08
2.06
2.04
10 -1 Case 1
Objective function, F1

10 -3
TL,i = 44.5 C; TL,o = 33.7 C
-5
10 TL,i = 42 C; TL,o = 33.7 C

10 -7

10 -9

5.0 Case 2

4.9
TL,i = 38 C; TL,o = 34.6 C
L/G ratio

4.8 TL,i = 40 C; TL,o = 32.6 C

2.286

2.283

2.280

2.277
10 -3 Case 2
Objective function, F1

10 -5
TL,i = 38 C; TL,o = 34.6 C
10 -7
TL,i = 40 C; TL,o = 32.6 C
10 -9

10 -11

10 -13

5 10 15 20 25 30 35 40 45 50 55 60
Generation

Fig. 7. Variation of estimated parameter ðL=G ratioÞ and objective functions with generations of GA for Case-1 and Case-2.
 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530 529

Mass flow rate, (kg/h) 2 4 6 8 10 12 14 (a) 18

(a) Range TL,i = 36.5 C
160 16
Approach TL,i = 44.5 C
140
14

Temperature ( C)
L
120 G
12
80
70 10
Power consumption, P (kW)

60 8 Optimum L/G

1.0x10 -3 6
8.0x10 -4 4
6.0x10 -4 Pump
Optimum 2
Blower 1 2 3 4 5
4.0x10 -4 combination
Total L/G ratio
2.0x10 -4
2 4 6 8 10 12 14 (b) 5 10 15 20 25 30 35 40 45 50 55 60
2.21
Combination
2.20
2.19

L/G ratio
5 10 15 20 25 30 35 40 45 50 55 60
(b)
Mass flow rate, (kg/h)

144 2.18 TL,i = 36.5 C

L 1.908 TL,i = 44.5 C
136 G
1.905
128 1.902
66
Objective function, F3 10 -3
63 TL,i = 36.5 C

-3
10 -5 TL,i = 44.5 C
10
Objective function, F2

10 -7
10 -5
10 -9
10 -7
-11
10
10-9 5 10 15 20 25 30 35 40 45 50 55 60
Generation
-11
10
5 10 15 20 25 30 35 40 45 50 55 60
(c) 2.25
Generation
2.20
Fig. 8. Different combinations of (a) mass flow rate of water and air ðL and GÞ for a
Optimum L/G ratio

constant L=G ratioð2:0846Þ and corresponding power consumption, (b) variation of 2.15
L, G and objective function F 2 for optimum combination during GA optimization.
2.10

having inverse feedback system has better e for most of the time 2.05
period as compared to tower without feedback. A similar trend is
2.00
also observed for Me. Moreover, meV also reduces by implementing
the inverse feedback system in comparison with the tower without 1.95
any feedback, which saves the cost due to less requirement of
make-up water. Therefore, results explained in context of Case-1 1.90
37 38 39 40 41 42 43 44 45
justify the use of inverse based feedback model to get a constant Water inlet temperature, TL,i ( C)
ðT L;o Þex: .
 
Fig. 6 reveals results for Case-2, where the heat load or T L;i Fig. 9. Optimization study (a) approach and range based optimum L=G ratio (b)
continuously increases with time and the water temperature, T L;o variation of L=G ratio and objective function during GA optimization (c) optimum
L=G ratio with the variation in T L;i (Case 3).
needs to be decreased with this [18]. The cooling tower having
feedback is able to supply the water at the desired temperature,
T L;o ¼ ðT L;o Þex: , whereas, in the absence of feedback, the tower fails, for minimizing F 1 . Further, as the number of generations increases,
as it operates at a fixed L=G ratio: The performance comparison for the objective function corresponding to each case decreases. Next,
both arrangements in terms of e and Me has been also presented. It it is obvious that a given value of L=G ratio can be satisfied by dif-
can be observed that as process (say diesel engine) picks up, the ferent combinations of L and G. So, the present inverse feedback
performance of the tower gets improved by the inverse feedback model also provides the option to select different combinations
arrangement. Further, meV is considerably less as the process picks. of L and G to meet certain performance requirement. For constant
The feedback model continuously updates T L;i and ðT L;o Þex: in the L=G ¼ 2:0846, covering this concept, results (involving 15 different
objective function F 1 and use GA to get required L=G ratio. After combinations) obtaining multiple combinations of L and G have
calculating desired L=G, required adjustments can be made in cool- been presented in Fig. 8a. Moreover, out of these combinations
ing tower by varying either L or G. At two random time levels, the one optimum combination can be selected that results in least
variation of L=G ratio and corresponding objective function, F 1 power consumption. For optimum combination, the iterative vari-
with GA generations for Case-1 and 2 have been presented in ation of two variables ðL and GÞ and the objective function, F 2 are
Fig. 7. GA takes 53–63 generations to determine L=G ratio ratio presented in Fig. 8b.
530 K. Singh, R. Das / Applied Thermal Engineering 105 (2016) 519–530
Fig. 9 presents the case where the cooling tower is required to
operate at an optimum L=G ratio: As explained previously, the
performance of certain processes (like HVAC) can be improved
by lowering the cooling tower’s outlet water temperature. But,
at the same time high range results in decreased approach, which
further increases the operating cost. So, the optimum values of
L=G ratio have been proposed for the entire range of cooling
tower inlet water temperature, T L;i ð36:4—45:3  CÞ. The intersec-
tion point of range and approach is considered to be the optimum
point (Fig. 9a), because if the cooling tower is operated at this
value of L=G ratio: neither the performance of tower nor the
industrial process will be compromised. In Fig. 9a sample results
are shown for two different values of T L;i ¼ 36:5 and 44:5  C. Sim-
ilarly, this optimum L=G ratio is calculated under different heat
loads and the optimum output from whole setup can be obtained
by minimizing the objective function, F 3 . Variations of L=G ratio
along with objective function, F 3 during optimization process
are presented in Fig. 9b. Furthermore, the inverse feedback model
can be also implemented to operate the cooling tower at opti-
mum point under continuously varying heat load, where it has
to be operated along the curve given in Fig. 9c. In this curve
the optimum value of L=G ratio has been shown for different
T L;i , along which cooling tower needs to be operate for optimum
performance.
7. Conclusion
In this work, a forced draft cooling tower has been studied
under varying inlet water temperature and water to air ratio
ðL=G ratioÞ. New empirical correlations are proposed to develop
certain objective functions to meet the desired performance
depending upon diverse operating conditions. The present inverse
feedback model involving the Genetic Algorithm (GA) predicts the
appropriate values of L=G ratio with time under varying heat load
conditions. The suggested feedback model can be useful for differ-
ent cases such as solar cogeneration, HVAC and diesel engine appli-
cations. Further, it can be also implemented to operate the cooling
tower at optimum performance reducing the input energy. It can
be concluded that inverse feedback model in conjunction with
GA is an optimized, simple and effective tool for adjusting water
to air ratio in order to control the performance of the cooling
tower.
References
[1] N. Milosavljevic, P. Heikkilä, A comprehensive approach to cooling tower
design, Appl. Therm. Eng. 21 (2001) 899–915.
[2] M.S. Söylemez, On the optimum sizing of cooling towers, Energy Convers.
Manage. 42 (2001) 783–789.
[3] M.S. Söylemez, On the optimum performance of forced draft counter flow
cooling towers, Energy Convers. Manage. 45 (2004) 2335–2341.
[4] G.F. Cortinovis, J.L. Paiva, T.W. Song, J.M. Pinto, A systemic approach for
optimal cooling tower operation, Energy Convers. Manage. 50 (2009) 2200–
2209.
[5] E. Rubio-Castro, M. Serna-González, J.M. Ponce-Ortega, Optimal design of
effluent-cooling systems using a mathematical programming model, Appl.
Therm. Eng. 30 (2010) 2116–2126.
[6] E. Rubio-Castro, M. Serna-González, J.M. Ponce-Ortega, M.A. Morales-Cabrera,
Optimization of mechanical draft counter flow wet-cooling towers using a
rigorous model, Appl. Therm. Eng. 31 (2011) 3615–3628.
[7] R.K. Singla, K. Singh, R. Das, Tower characteristics correlation and parameter
retrieval in wet-cooling tower with expanded wire mesh packing, Appl. Therm.
Eng. 96 (2016) 240–249.
[8] D. Manolakos, G. Papadakis, S. Kyritsis, K. Bouzianas, Experimental evaluation
of an autonomous low-temperature solar Rankine cycle system for reverse
osmosis desalination, Desalination 203 (2007) 366–374.
[9] B. Twomey, P.A. Jacobs, H. Gurgenci, Dynamic performance estimation of
small-scale solar cogeneration with an organic Rankine cycle using a scroll
expander, Appl. Therm. Eng. 51 (2013) 1307–1316.
[10] S. Abdelhady, D. Borello, E. Tortora, Design of a small scale stand-alone solar
thermal co-generation plant for an isolated region in Egypt, Energy Convers.
Manage. 88 (2014) 872–882.
[11] F. Calise, M.D. D’Accadia, M. Vicidomini, M. Scarpellino, Design and simulation
of a prototype of a small-scale solar CHP system based on evacuated flat-plate
solar collectors and Organic Rankine Cycle, Energy Convers. Manage. 90 (2015)
347–363.
[12] J.J. McKetta Jr., Encyclopedia of Chemical Processing and Design, Abrasives to
Acrylonitrile, vol. 1, CRC Press, 1976.
[13] J. Freeman, K. Hellgardt, C.N. Markides, An assessment of solar-powered
organic Rankine cycle systems for combined heating and power in UK
domestic applications, Appl. Energy 138 (2015) 605–620.
[14] B.G. Lipták, Process Control: Instrument Engineers’ Handbook, Butterworth-
Heinemann, 2013.
[15] F. Merkel, Verdunstungskühlung, VDI-Verlag, 1925.
[16] A.K.M. Mohiuddin, K. Kant, Knowledge base for the systematic design of wet
cooling towers. Part I: selection and tower characteristics, Int. J. Refrig. 19
(1996) 43–51.
[17] J.C. Kloppers, A Critical Evaluation and Refinement of the Performance
Prediction of Wet-cooling Towers, University of Stellenbosch, 2003.
[18] M. Abdulqader, Diesel Generator Auxiliary Systems and Instruments, 2006.
<www.Lulu.com>.
[19] J.F. Kreider, Handbook of Heating, Ventilation, and Air Conditioning, CRC Press,
2000.
[20] K. Deb, Optimization for Engineering Design: Algorithms and Examples, PHI
Learning, 2012.
[21] R.J. Moffat, Contributions to the theory of single-sample uncertainty analysis, J.
Fluids Eng. 104 (1982) 250.
[22] J.R. Taylor, An Introduction to Error Analysis: The Study of Uncertainties in
Physical Measurements, second ed., University Science Books, Sausalito,
California, 1983.
[23] M. Gao, F.Z. Sun, K. Wang, Y.T. Shi, Y. Bin Zhao, Experimental research of heat
transfer performance on natural draft counter flow wet cooling tower under
cross-wind conditions, Int. J. Therm. Sci. 47 (2008) 935–941.
[24] K. Singh, R. Das, An experimental and multi-objective optimization study of a
forced draft cooling tower with different fills, Energy Convers. Manage. 111
(2016) 417–430.