Applied Energy 154 (2015) 491499

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Energy conservation improvement and ONOFF switch times reduction

for an existing VFD-fan-based cooling tower
Chun-Cheng Chang a, Shyan-Shu Shieh b,, Shi-Shang Jang a,, Chan-Wei Wu c, Ying Tsou c
a
Department of Chemical Engineering, National Tsing Hua University, Hsinchu 30013, Taiwan
b
Department of Occupational Safety and Health, Chang Jung Christian University, Tainan 71101, Taiwan
c
China Steel Corporation, Kaohsiung 81233, Taiwan

h i g h l i g h t s

 Achieve energy saving and reduce the ON/OFF switching frequency at the same time.
 Propose a PI feedback controller with a temperature zone setting.
 Implement hybrid operations of rule-based and equation-based feedback control.
 Introduce larger approach for the setting of the outlet cooling water temperature.

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

Article history: The increasing economic advantage of replacing traditional two-speed fans with variable-frequency drive
Received 19 December 2014 (VFD) fans has been gaining popularity in the industry. However, concerns regarding frequent ON/OFF
Received in revised form 3 May 2015 switching and the lack of a well-devised controller have discouraged widespread adoption. In this study,
Accepted 9 May 2015
a temperature zone method is proposed to replace the set-point method of fan control. Additionally, the
Available online 29 May 2015
highest output water temperature allowed in the process is set as the upper limit of a zone in order to
further conserve energy. Both strategies are comprehensively analyzed for a virtual cooling tower that
Keywords:
uses operational data from an existing VFD-fan-based cooling tower system in Taiwan. The results show
Cooling tower
Energy conservation
energy savings of 38% for a 0.75 C zone without increasing the ON/OFF switching frequency. The pro-
Variable-frequency drive posed strategies were further veried via an on-line eld experiment. The proposed methods can be uni-
Zone control versally and easily applied to any existing cooling tower, and have signicant implications for energy
Approach conservation if adopted globally.
2015 Elsevier Ltd. All rights reserved.

1. Introduction cooling tower operations would provide signicant opportunity

Cooling towers are a common feature of industrial plants, espe-
cially in energy-intensive sectors such as the iron, steel, or petro-
chemical industries. A cooling tower dissipates waste heat to the
atmosphere through a combination of heat- and mass-transfer pro-
cesses. When considering an individual plant, the potential energy
savings for a cooling tower may not be as great as those of powered
devices, such as compressors, or energy-intensive components
such as boilers or distillation columns. In addition to industrial
applications, cooling towers are also widely used in heating, venti-
lating and air-conditioning systems. Overall, any improvement in
Corresponding authors.
E-mail addresses: sss@mail.cjcu.edu.tw (S.-S. Shieh), ssjang@mx.nthu.edu.tw
(S.-S. Jang).
for energy conservation.
Variable-frequency drive (VFD) devices have been available for
more than four decades, but were not applied to cooling tower fans
until their prices fell sharply over the past decade. Recently,
two-speed fans have been gradually replaced with VFD-fans.
Practical concerns remain, of how to avoid frequent START/STOP
fan operation, which can cause sudden increases in stress due to
the large inertia moment. This drawback has limited cooling sys-
tems based on VFD-fans from fully exploiting their energy-saving
potential, which is more compelling and signicant in
large-scale, multi-cell systems used in energy-intensive industries.
This study proposes two operational strategies to take advantage of
energy-saving potential represented by VFD-fan-based cooling
tower systems to the greatest possibility.
Many studies on cooling towers have focused on design [16]
and performance parameters [79]. Comparatively few studies

http://dx.doi.org/10.1016/j.apenergy.2015.05.025
0306-2619/ 2015 Elsevier Ltd. All rights reserved.
492 C.-C. Chang et al. / Applied Energy 154 (2015) 491499
Nomenclature
CV controlled variable
Dt sampling period (min)
ek error term at the kth run
f fan operating frequency (Hz)
H relative humidity (%)
I total number of fans
KC controller gain (kW/C)
MV manufactured variable
pk power requirement of fan set (kW)
pmin power requirement of fan set at the minimum operating
frequency (kW)
pcutoff the cut-off value of power requirement (kW)
have addressed the optimal operation of cooling towers. Castro
et al. [10] simulated the operating cost of a water-cooling system
considering water make-up, energy consumption, and climatic
effects. Cortinovis et al. [11] minimized operating costs by optimiz-
ing the fan speed, rate of water removal, and the positions of valves
at heat exchanger branches. They also suggested that economical
operation would require the water outlet temperature to remain
as high as possible. The ndings of these studies were based on
single-cell tower operations utilizing two-speed fan mode. Wang
et al. [12] used statistical modeling to predict outlet water temper-
ature and developed a discrete model-based strategy to determine
optimal fan-operating mode in a multi-cell cooling tower with
two-speed fans. In that work, the authors simulated a virtual plant
to verify their proposed method.
Other studies addressed the economic gains of applying VFD to
cooling tower fans. Cohen [13] indicated that variable-speed fans
can signicantly reduce the energy requirements of cooling tower
systems by precisely assigning air ow rates to the required heat
dissipation. Also in his theoretical viewpoint, in a multi-fan VFD
system, i.e., multi-cell, the fan speed must be synchronized,
because a marginal increase in fan speed increases the power con-
sumption, whereas synchronized fan speed put more cells into ser-
vice and maximize the total heat transfer surface while minimizing
the power consumption. Muntean et al. [14,15] implemented tem-
perature control for VFD-fan operation in a cooling tower and
found that energy savings and control of out-ow water tempera-
ture are the two major advantages. They also claimed that using
VFD technology to track wet-bulb temperature throughout the
year resulted in annual energy savings of 83% compared to running
the fans continuously at full speed. These studies addressed the
advantages of VFD installations and the congurations of the sim-
ulation examples. However, as pointing out in this work, synchro-
nizing fan speeds across the cells during low-speed stages is not
practical because it may cause frequent ONOFF switching among
fans. It is the very reason that we propose temperature zone con-
trol, and hybrid operation of rule-based and equation-based feed-
back control in this study.
Yu and Chan [16] used a load-based method to control the
speed of cooling tower fans and condenser water pumps in order
to improve the energy efciency of a cooling tower system, and
highlighted the need to widen the use of such control systems.
Aha and Mitchell [17] proposed optimal supervisory control strate-
gies for the set points (exact values) of controlled variables such as
supply air, chilled water, and condenser water temperatures. The
fan speed was controlled via a proportional-integral-derivative
(PID) feedback controller such that the water temperature reached
the set point designated by the optimal supervisory control strat-
egy. Those studies considered the integration of cooling towers
and chiller systems. However, previous studies have not addressed
the frequency of ON/OFF switching in VFD fans.
sI integral time (min/C)
nfan,k number of fans operated at kth run
Tsp set point temperature (C)
Tair air temperature (C)
Tw,out,k outlet water temperature at kth run (C)
Tw,in inlet water temperature (C)
TUL upper limit temperature in the temperature zone con-
trol (C)
TLL lower limit temperature in the temperature zone con-
trol (C)
yk measured value of the CV at the kth run
Ysp set point of the CV
In traditional cooling towers, two-speed fans, i.e., OFF, low, or
high speed, are the most common type. The limited choices of
fan speed make feedback controllers difcult to control outlet
water temperature at the desired target. As a result, feedback con-
trol is rarely used in the traditional cooling tower. With improving
affordability, VFD fans have gained popularity and gradually
replaced two-speed fans in cooling towers. This provides a great
opportunity to fully exploit the energy-saving potential of VFD
technology by applying feedback control to ensure outlet water
temperatures are maintained at set points. Nevertheless, frequent
ON/OFF switching of fans and the lack of a well-devised controller
prevented industry from widely adopting or wholly utilizing this
technology. As an alternative, rule-coded programmable logic con-
trollers (PLC) were used instead of the feedback controller. The
control of the outlet water temperature at the highest allowable
temperature is then compromised, as is the energy-saving poten-
tial associated with the introduction of VFD-fans. The shortfall is
more apparent in a large-scale cooling tower system comprising
multiple cells. Therefore, this study proposes two operational
strategies to fully exploit the energy-saving potential of
VFD-fan-based multi-cell cooling towers.
2. Operational strategies
Section 2.1 describes the effects of a larger approach value on
energy savings during fan operation. Section 2.2 details the imple-
mentation and theoretical development of zone-setting
proportional-integral (PI) controllers. The rule based control strat-
egy to avoid frequent ONOFF switch is described in Section 2.3.
2.1. Larger approach strategy
In a cooling tower system, heat is removed from the water by
sensible heat, via temperature differences; and by latent heat, via
the evaporation of small amounts of water. The evaporation pro-
cess accounts for 80% of total heat removal. The rate of water evap-
oration in a cooling tower is determined by relative humidity,
ambient air temperature and airow rate. The thermal perfor-
mance of the cooling tower depends principally on the wet-bulb
temperature of inowing air. The lowest outlet water temperature
is limited by ambient air wet-bulb temperatures.
Two kinds of temperature differences are indexed for the design
and operation of tower systems, namely, range and approach. The
temperature difference between water entering and leaving the
cooling tower is called range, (or cooling tower range, or cooling
range). Range is determined using a heat load and water ow rate,
rather than by the thermal capability of the cooling tower. The dif-
ference between the outlet water temperature and the wet-bulb
temperature of inowing air is termed approach. At the design

stage, the approach value is vital for determining the physical size
of the cooling tower.
Water temperature decreases along its path through the tower,
due to evaporative cooling. The specic enthalpy of the saturated
air lm and its variation with water temperature is given by the
saturation curve on the psychometric chart. The difference
between the specic enthalpies of saturated and bulk air is the
enthalpy driving force responsible for evaporative cooling. The
driving force can be visualized in the enthalpytemperature plot
as shown in Fig. 1. The distance between the process line and the
saturation curve is the enthalpy driving force for evaporative cool-
ing at certain points, e.g., line AA0 , line BB0 , line CC0 , line DD0 . [18,19]
It is apparent that the enthalpy driving force varies along the path
from cooling water entering to its leaving. The average enthalpy
driving force can be dened as the area above the process line
and below the saturation curve, as area ABB0 A0 , and CDD0 C0 shown
in Fig. 1, divided by cooling tower range.
At steady state, the heat load and water ow rate are constant,
and thus, the range is also constant. Subsequently, the enthalpy
driving force increases as the outlet water temperature increases.
Taking the selected cooling tower as an example, the operational
conditions are assumed as follows: input water temperature of
35.0 C, air temperature of 25.4 C, relative humidity of 75%, and
ratio of liquid to air ow rate of 2.02. If the output water temper-
ature is 31.5 C, then Range is 3.5 C, the operation line is AB and
the average enthalpy driving force is area ABB0 A0 divided by 3.5,
as shown in Fig. 1. If the output water temperature is maintained
at 33.0 C, then the operation line shifts to the line CD; the average
enthalpy driving force becomes the area CDD0 C0 divided by 3.5. It is
obvious that area CDD0 C0 is larger than ABB0 A0 , so is the average
driving force of the shifting operation line in the latter case of high
output water temperature. Operations with large driving force
consume less energy.
For the operation of an existing cooling tower, a large driving
force means a large approach and, as indicated in Fig. 1, a high out-
let water temperature. Equipped with feedback control, a cooling
tower with variable-speed fans can be automatically operated at
the highest allowable outlet temperature.
2.2. Temperature zone control strategy
For a feedback control system, two process variables must be
identied: the controlled (CVs) and manipulated (MVs) variables.
In a VFD-controlled cooling tower, operators adjust the fan operat-
ing frequency (Hz), which in turn determines the fan speed and
Saturation Curve
140
Operating Line
120 C'
Enthalpy (kJ/kg)
A'
100
80
A
Twout'
60 C where Tw,out,k is the temperature for the kth run; Tsp is the set point
Twout
40
28 30 32
Temperature of water (OC)
Fig. 1. Inuence of range shifting on enthalpy driving force.
C.-C. Chang et al. / Applied Energy 154 (2015) 491499 493
controls the outlet water temperature. Given the relationship
between fan operating frequency (Hz) and power requirement,
the power necessary to achieve the desired outlet water tempera-
ture can be calculated. Hence, outlet water temperature and fan
power are considered as CV and MV respectively. In the following
feedback controller, the objective is to reduce the error to zero:
ek ysp;k  yk 1
where k is the sequence index, ek is the error term on the kth run, yk
is the measured value of the CV at the kth run, and ysp is the set
point of the CV.
The discrete PI control algorithm is considered most appropri-
ate because of its exibility, computational simplicity, cost effec-
tiveness, and transparency [20]. The goal of feedback control is to
maintain the measured value of the CV at its set point.
In this study, the measured value of the CV can be lower than
the set point during periods of cold weather. When the sustained
error occurs, the integral term keeps accumulating and continues
to build up to saturate the controller output element (e.g., fully
open or close of valves) in the traditional PI controller. Further
buildup of the integral term while the controller is saturated is
referred to as reset windup, and usually results in overshooting.
The velocity form of PI control is chosen in order to avoid the
above-mentioned reset windup problem for the proposed temper-
ature control. This approach inherently resolves anti-reset windup,
because the summation of errors is not explicitly calculated. The
velocity form of the formula for discrete PI control is described in
Eq. (2).
 
Dt
Pk1 pk K c ek  ek1 ek 2
s1
Here, p is the value of the MV, namely the power requirement of fan
set; Kc is the controller gain; s1 is an adjustable parameter and is
referred to as integral time; Dt is the sampling period; and K c and
sI are determined by the process reaction curve method. In this
study, these values are also further optimized by using
data-driven estimation and a Matlab function, fmincon.
Zone control strategies have been proposed to stabilize some
feedback control systems [21,22]. Theoretical developments of this
method have been widely discussed in previous works. In order to
reduce the number of ON/OFF events, the temperature zone con-
trol strategy is adopted in this work. The temperature is controlled
within a pre-set zone rather than by a precise target value. When
the measured outlet temperature is outside the upper or lower
limits of the temperature zone, the target is set as the zone
mid-value; the target is set to the present outlet temperature when
its value is within the zone, i.e.,
8
D' < ysp;k T sp ; T w;out;k > T UL
>
B' ysp;k T sp ; T w;out;k < T LL 3
>
:
Enthalpy driving force ysp;k T w;out;k ; T UL > T w;out;k > T LL
combining (1) and (3) gives,
8
< ek ysp  yk ; yk > T UL
>
ek ysp  yk ; yk < T LL 4
Twin
>
:
B ek 0; T UL > yk > T LL
Twin'
D
temperature; and TUL and TLL are the upper and lower limit values,
respectively. The three values have the following relationship:
34 36 38 40 TLL < Tsp < TUL. The power requirement can be calculated using Eqs.
(2) and (4). If the outlet temperature remains within the target zone
no adjustments are necessary, so the power requirement remains
constant.
494 C.-C. Chang et al. / Applied Energy 154 (2015) 491499

2.3. Hybrid operation
For a multi-fan cooling system, there are two types of fan oper-
ations, namely, whole set of fans and partial set of fans in opera-
tion. In the cold weather, only partial set of fans are needed for
service. A slight change of outlet cooling water temperature might
cause frequent ONOFF switch of fans, especially when fans are
operated at the lowest speed. With the additional consideration
of energy saving, we would devise looking-up tables to determine
the number and the frequency of fans needed for operation. This is
the stage of rule-based operation when power requirement are low
and partial set of fans are needed in service.
When in the hot weather, all fans are needed to put in service,
the frequency of fans are determined by computing the power
requirement. This is the stage of equation-based operation. The
cut-off value of power requirement, pcutoff is needed to decide in
which stage the operation is. To avoid frequent ONOFF switching
of fans, we suggest that the cut-off value is computed by the sum-
mation of power of one fan operating at the lowest speed and pow-
ers of the rest fans operated at the next-to-lowest speed.
In Summary, when the required power is larger than pcutoff , all
fans are operated simultaneously at the same frequency. If the
required power is smaller than pcutoff , the number of fans required
and their frequency are based on a rule-based table. There is no
general rule-based table. A rule-based table is decided by number
of fans and their operating frequency. The basic principle of a
rule-based table is to minimize ON/OFF switch possibility. The
rules of turning one more fan ON (or OFF) need to be carefully
devised by the consideration of not turning the very one OFF (or
ON) again if the consequent temperature slightly drops (or rises).
Therefore, the orders of switching fans ON when cooling water
temperature increases and that when temperature decreases are
not the same. The rule-based tables for looking-up in these two
cases are slightly different.
3. Plant description
In this study, the cooling tower of a steel plant was investigated.
This cooling tower is one of relatively few retrotted with
VFD-fans. It provides cold water for a tandem cold mill and contin-
uous annealing line. The designed water circulation rate is
11,000 m3/h. The operating water ow rate is around 4000 m3/h.
The tower has six cells, each of which originally had a two-speed
fan, of which four have been replaced with VFD-fans and the
remaining two are usually operated at low speed throughout the
year. The size of each cooling tower is 11  15  8.6 m. The basin
beneath the cooling tower is 10.8  15  5.9 m. The real system
residence time is about 15 min. The fan diameter is 7.31 m and
maximal fan speed is 150 RPM. The operating frequency of each
VFD-fan is from 30 and 60 Hz. The corresponding energy consump-
tion is from 10 to 55 kW h. The energy consumption has a cubic
relationship with the fan operating frequency. Marginal increase
in fan operating frequencies therefore requires a large increase in
energy consumption. It requires less power to run two fans at
30 Hz, i.e., 10.51 kW  2, than to run one fan at 60 Hz, i.e.,
54.59 kW. Furthermore, two cells, i.e., two fans operating at
30 Hz, provide a larger heat transfer surface than one cell with a
fan operating at 60 Hz. For multiple-fan applications, it is recom-
mended that all fans are operated simultaneously at the same fre-
quency [13] without considering the frequent ONOFF switching of
fans in the situation of low cooling water temperature.
Prior to the improvement strategies, the cooling tower opera-
tion followed a set of heuristic rules based on the experience of
the operators. Table 1 shows the main principles of the rules,
which are written in a PLC. There is no denite target for control-
ling the outlet water temperature. The rules initialize the rst fan
at 30 Hz; when the outlet water temperature exceeds 31.0 C, an
additional fan is operated for every 0.5 C increment until the out-
let water temperature reaches 32.5 C. Subsequently, the outlet
water temperature is checked every 20 min. At each check, the fre-
quency of all fans is simultaneously increased by 5 Hz if the tem-
perature is found to be increasing. The rules for switching the
fans off are similar, but are applied in the reverse order; the only
difference is the target temperature: the last fan is turned off when
the outlet temperature drops below 30.0 C.
Except during hot seasons, the ON/OFF frequency is high when
the original plant operation mode is applied. For example, operat-
ing one fan at 30 Hz is insufcient to cool water, so a second 30-Hz
fan is required. Subsequently, the outlet water may become cooler
than the target temperature, and so one 30 Hz fan is turned off to
conserve energy. As a result, the water temperature may increase
again, and thus the ON/OFF cycle is frequently repeated.
4. Virtual plant simulation
This study uses real operational data from an eight-day period
of March 21 to 29, 2014 to model the energy-saving effect of the
proposed strategies. During that period, average daily energy con-
sumption was 1334.6 kW h, and statistics for cooling water and air
properties are shown in Table 2. The air temperature and humidity
data reect the typical spring season in Taiwan.
The model estimates outlet water temperature for the selected
cooling tower, which is a function of inlet water temperature, air
temperature, relative humidity, water ow rate, and air ow rate.
In this system, water ow rate is usually constant and therefore
does not form an input variable to the model. There are three water
streams returning to the cooling tower, each with its own temper-
ature measurement. Moreover, air ow rate is not measured, but
the total power for fan operation is logged. The air ow rate is
determined by fan speed, which is proportional to fan frequency
and is a cubic root function of fan power. In summary, the outlet
water temperature is modeled using three inlet water tempera-
tures, air temperature, relative humidity, and total fan power.
Table 2
Statistics of ambient air and cooling-water data used to model a virtual plant.
Tw,in,1 Tw,in,2 Tw,in,3 Tair H Tw,out
(C) (C) (C) (C) (%) (C)
Max. 38.1 35.1 36.6 31.9 94.6 32.9
Min. 35.5 32 29 18.8 52.3 30.2
Average 36.8 33.5 33.3 25.4 74.6 31.6
Standard deviation 0.5 0.6 1.8 3.2 7.8 0.5

Table 1
ON/OFF rules of fan operations in the studied plant.

Rules for starting fan

Outlet water temperature T > 31 T > 31.5 T > 32 T > 32.5 T > 33.5
Operation 1st at 30 Hz 2nd at 30 Hz 3rd at 30 Hz 4th at 30 Hz Increase frequency (5 Hz)
Rules for stopping fan
Outlet water temperature T < 30 T < 30.5 T < 31 T < 31.5 T < 33
Operation 1st 2nd 3rd 4th Decrease frequency (5 Hz)
 C.-C. Chang et al. / Applied Energy 154 (2015) 491499
The following equation (6) is obtained from linear regression of the
actual data set, and functions as the virtual cooling tower system:
T w;out 0:539  T w;in;1 0:221  T w;in;2 6:32  102
T w;in;3 3:61  102  T air
3 3
6:49  10  H  2:16  10  P 0:814
where Tw,in,1, Tw,in,2, and Tw,in,3 are the temperatures of each inlet
water stream; Tair is air temperature; H is relative humidity; and
P is total power consumption of the cooling tower fans.
Fig. 2 shows the modeling and prediction results. The data from
the rst seven days are implemented as the modeling set, and the
data of the last day are used as the prediction set. The model gives
an R2 value of 0.85 and the mean square error (MSE) of the predic-
tion set is 0.05. The results indicate that the model can accurately
represent a cooling tower system to examine the implementation
of the proposed operational strategies. In this study, a linear model
is favorable because of its explicit form. And it is accurate for the
simulation of an eight-day operation. To build a general purpose
of a cooling tower system for a whole year operation is not within
the scope of this research. The authors [7,12,23] proposed
multiple-linear-models dealing with the complex nonlinearity of
a cooling tower system.
To validate the proposed strategies, the virtual plant must be
equipped with a PI feedback controller, which replaces the
rule-based PLC used in the existing system. The construction of
the proposed control loop is described in Section 2. To reduce
ON/OFF switching of the fans, this PI feedback controller must
adopt hybrid operations that combine rule- and equation-based
operations.
When ambient air temperature is low, partial startup of the
four-fan set can sufciently cool the water; the rules shown in
Table 3 are adopted in this controller. Table 3 shows that shifting
from three-fan operation to four-fan operation is from the case of
one 40 Hz fan and two 35 Hz fans to the case of one 30 Hz fan
and three 35 Hz fans. This arrangement provides less opportunity
to immediately turn OFF one of the fans in the system when cool-
ing water temperature subsequently decreases. Say, right after
shifting to turn all fans ON, (i.e., one 30 Hz fan and three 35-Hz fans
ON), if the temperature drops slightly, then we may turn one of the
35 Hz fans to 30 Hz, then another one of 35 Hz fans to 30 Hz, until
Modelled Tout Modeling set Prediction set
2 MSE : 0.05
R : 0.85
Predicted Tout
Measured Tout
TCW,out
32
30
03/21 03/23 03/25 03/27
Fig. 2. Modeled results of the virtual plant.
Table 3
Operation rules for power requirement less than 55.36 kW.
Operation mode All close 30 Hz 1
Power requirement (kW) 0 10.51
Operation mode 35 Hz 1, 40 Hz1 40 Hz 2
Power requirement (kW) 35.45 41
495
all four fans are on 30 Hz. Therefore, the cut-off value is based on
the case of one 30 Hz fan and three 35-Hz fans, i.e., 55.36 kW.
When power requirement exceeds the cut-off value, all four
fans are required, and Eq. (6) is executed:
5 f 17:093 1:86  106  p3  1:23  103  p2 0:376  p
6
where f is the operated frequency, and p is the power requirement.
In Eq. (2), the MV (i.e., power requirement) is calculated in the
controller. The number of fans required is determined by compar-
ing the power requirement to pcutoff . If number of operating fans is
less than 4, then one of the rules in Table 3 matching with the
power requirement is executed, otherwise Eq. (6) is computed to
determine appropriate fan frequency.
The proportional and integral parameters, Kc and s1, of the con-
troller in the virtual system are calculated by the process reaction
curve method and further optimized by using data-driven estima-
tion and a Matlab function, fmincon. Kc and sI are obtained as 42.62
and 0.52 respectively. The heat load of the process is assumed to be
constant during the studied period.
Fig. 3 shows the ow chart of detailed operation procedures of
the virtual plant simulation.
5. Results and discussion
This section at rst illustrates how much increasing enthalpy
driving force inuence the reduction of fan energy requirements
by taking the operation of the virtual plant as the case study. The
strategies of large allowable approach and temperature zone con-
trol are employed to operate the virtual cooling tower system,
whose construction is based on the data acquired from the existing
03/29
Fig. 3. Flowchart of operation procedures of the virtual plant simulation.
35 Hz 1 40 Hz 1 30 Hz 1, 35 Hz 1 35 Hz 2
14.95 20.5 25.46 29.9
35 Hz 3 40 Hz 1, 35 Hz 2 30 Hz 1, 35 Hz 3
44.85 50.4 55.36
496 C.-C. Chang et al. / Applied Energy 154 (2015) 491499

system. The data of energy consumption and ON/OFF frequency of 5.2. Implementation of the larger approach strategy
the fans from the existing system are served as the benchmarking
values. The comparison of the results from the virtual plant against The simulation presented in this study demonstrates how a
the benchmarking data validates the effects of the proposed strate- zone-control PI controller maintains the largest allowable
gies in this section. approach in the process by minimizing ON/OFF switching. The sim-
ulation experiments were conducted on the same period of eight
5.1. Relationship between driving force and energy consumption days. The simulated results were compared with those of the exist-
ing system.
Fig. 4 shows how the target outlet water temperature affects For this cooling tower, adjustments to the fan operation fol-
the average driving force and daily energy consumption in the vir- lowed the rules coded in the PLC, as shown in Table 1. The opera-
tual system in the different seasons in Taiwan. The original data of tional histories of the outlet water temperature and power
eight-day period from the existing system represent the typical cli- consumption of the four-fan set are documented in Fig. 5.
mate conditions in spring season. The average temperature and Throughout the entire period, the fans were almost always auto-
humidity of air in this data set are Tair = 25.4 C and H = 74.6% matically controlled by the PLC, except between 17:00 on
respectively. We designate average air properties of Tair = 20.4 C, 03/23/2014 and 11:00 on 03/24/2014 when, for unknown reasons,
H = 69.6% as the typical climate conditions in cold and dry winter they were switched to manual control and set at 50 Hz. For the
and Tair = 30.4 C, H = 79.6% as in hot and humid summer. Then, majority of that time, the outlet water temperature differed from
the difference value, i.e., 5 C and 5% between summer (or winter) the acceptable highest temperature (32.5 C) and at least one of
and spring is added (or subtracted) to (or from) each datum in the fans was operational because the temperature never dropped
original data set to mimic the climate conditions in the summer below 30.0 C, which is the lower limit at which the coded PLC
and winter in Taiwan. Then the datasets are obtained for the case rules would turn OFF all fans. During that period, ON/OFF fre-
study in summer and winter respectively. quency was 60 and daily energy consumption was 1334.6 kW h.
With an outlet cooling water temperature and the average air The green lines in Fig. 5 indicate the results of operating the PI
properties, we can calculate the average enthalpy driving force controller in the virtual system. The estimated outlet temperature
for each dot in Fig. 4. Taking the constant range of 3.5 C and is similar to or less than the target temperature of 32.5 C. For
liquid-to-air ow ratio of 2.02, average driving forces are calcu- those periods when the temperature is less than 32.5 C, water
lated by following the procedures described in Fig. 1 and can be cooled using the two xed-speed fans, and it is not neces-
Section 2.1. Also with the dataset of eight-day in each season, we sary to turn on any of the VFD-fans.
calculate the energy consumption for the whole period and take In the simulation, estimated daily energy consumption was
the daily unit with the rst piece of datum as the initial value in 625.9 kW h, which is only 47% of the real energy consumption, rep-
operating the virtual plant. resenting a signicant saving for the VFD-equipped cooling tower.
Fig. 4 demonstrates that the cooling tower has the largest driv- The results reveal that the approach must be adjusted in order to
ing force and the lowest energy consumption during winter. The conserve energy. With a PI controller, the cooling tower can be
target outlet water temperature strongly affects energy consump- automatically operated at the highest allowable outlet tempera-
tion by the fans. Increasing the target outlet water temperature ture. However, the side effect of attempting to achieve a precise
from 30.0 to 32.5 C reduces daily energy consumption of the fan target temperature, as shown in Fig. 5, is that the simulated
from 2417 to 9 kW h in winter, and from 4851 to 1234 kW h in ON/OFF frequency was 199, which is three times greater than that
summer. From this denition of approach, it is clear that higher of the real operation. This outcome is considered unacceptable;
outlet water temperature is associated with larger approach value. therefore, temperature zone control is proposed, combined with
For an existing cooling tower, whether or not the heat load is con- a PI controller.
stant, it is always practical and cost-effective to set the outlet
water to the highest temperature allowable by the process, in
order to take advantage of the large driving force, i.e., a large
approach, and thereby achieve the required energy savings. 34
estimated Tout
Original Tout
33
T target
TCW,out (O C)

32
100 5000
Driving force - Winter
31
Energy consumption per day, (kWh/day)

Driving force - Spring

Average driving force, (kJ/kg dry air)

Driving force - Summer

Energy consumption - Winter 30
80 Energy consumption - Spring 4000
Energy consumption - Summer
29
03/21 03/23 03/25 03/27 03/29
60 3000
fan power consumption (kW)

300
estimated fan power consumption
40 2000 250 real fan energy consumption

200

20 1000 150

100

50
0 0
29.5 30 30.5 31 31.5 32 32.5 33
0
target of outlet water temperature, (oC) 03/21 03/23 03/25 03/27 03/29
Date
Fig. 4. Relationship between driving force, energy consumption, and outlet water
temperature. Fig. 5. Estimated results of precise target control operation.
 C.-C. Chang et al. / Applied Energy 154 (2015) 491499 497

5.3. Implementation of temperature zone control 34

estimated Tout of 0.5C zone control
Original Tout
33 T target
A temperature zone control strategy is proposed to reduce

TCW,out (O C)
ON/OFF switching. The set point temperature in a PI controller is 32
an interval rather than a specic value. When outlet water temper-
31
ature is outside the zone limits, the target temperature is set to the
zone mid-point. While the outlet water temperature remains 30
within the zone, the set point is the present outlet water temper-
29
ature. When the controller is congured according to zone setting, 03/21 03/23 03/25 03/27 03/29
fan operations remain unchanged unless the CV, i.e., outlet temper-

fan power consumption (kW)

ature, is outside the zone boundaries. The detailed conguration of 300
estimated fan power consumption
a PI zone controller is described in Section 2. 250 real fan energy consumption
A total of six simulation experiments were conducted by replac-
200
ing the set value of 32.5 C with a temperature zone. The temper-
150
ature interval was initially set to 0.25 C, and this interval width
was increased by 0.25 C in each subsequent simulation experi- 100
ment (widest 1.5 C zone). All simulations used 32.5 C as the 50
upper limit of the zone. The results for ON/OFF frequency during
0
the 8-day period and the daily energy consumption for each of 03/21 03/23 03/25 03/27 03/29
the six zone-controlled experiments are listed in Table 4, and com- Date
pared with those of the original operation and traditional (precise
Fig. 6. Estimated results using 0.5 C zone control.
target) PI control. It is apparent that energy consumption increases
and ON/OFF frequency decreases as the zone width increases. For
the widest zone (1.5 C), daily energy consumption is
1118.6 kW h, which is 84% of the original energy consumption; 35
estimated Tout of 0.5C zone control
the ON/OFF frequency is 34, which is almost half the original value 34
estimated Tout of precise control
original Tout
TCW,out ( O C)

of 60. For the 0.75 C zone, daily energy consumption is T target

33
826.2 kW h, which is 62% of the original energy consumption,
and ON/OFF frequency is 63. The simulation results also show that 32
the average, maximum, and minimum outlet water temperatures
31
are similar for the different zones. In other words, by employing
a PI controller operating a temperature zone strategy, a VFD fan 30
03/28 03:36 03/28 09:36 03/28 15:36 03/28 21:36 03/29 03:36
can achieve 38% energy saving in a cooling tower without increas-
fan power consumption (kW)

ing the frequency ON/OFF switching. 150

fan power consumption of zone control
From the results of the six simulation experiments in Table 4, fan power consumption of precise control
the 0.5 C zone was selected as the representative case and com- original fan energy consumption
100
pared against the original operational prole, as shown in Fig. 6.
During the simulation, the zone-based PI controller limits outlet
water temperature within the 0.5 C zone, between 32.0 and 50

32.5 C, except in cold weather, when the outlet water temperature

is less than 32.0 C without the use of the VFD-fans. 0
In Fig. 7, the typical 24-h operational proles show how the 03/28 03:36 03/28 09:36 03/28 15:36 03/28 21:36 03/29 03:36

temperature zone PI controller reduces energy consumption and Date

the ON/OFF frequency of the fans. Fig. 7 reports three cases: origi-
nal operation, precise target, and zone control of 0.5 C width. The
daily energy consumptions, from low to high, are 855, 1041,
1553 kW h, respectively. In the original strategy, as mentioned in
Section 5.2, the rule-coded PLC does not take advantage of setting
the largest approach to conserve energy. The precise strategy sets
an outlet water temperature similar to the target of 32.5 C, and
the lowest energy consumption is at the expense of frequent
Fig. 7. Typical daily outlet water temperature and fan power consumption for three
operational scenarios.
ON/OFF switching (in this case, 34 times within a day). For the
0.5 C zone control, the outlet water temperature varies between
32.0 and 32.5 C with 16 ON/OFF switches. It is worth noting that
there are some instances when the outlet water temperature

Table 4
Number of ON/OFF switching operations and energy consumption for precise and zone-control strategies.

Number of ON/OFF Average daily power Tave Tmax Tmin

operations consumption (kW h/day)
Original operation 60 1334.6 31.5 32.7 30.2
Precise target strategy (32.5 C) 199 625.9 32.1 32.8 31.0
Zone control strategy Zone width (C) Tup (C) Tlow (C)
1.5 32.5 31 34 1118.6 31.8 32.6 30.8
1.25 32.5 31.25 42 1001.7 31.9 32.6 30.9
1 32.5 31.5 43 891.2 31.9 32.6 31.0
0.75 32.5 31.75 63 826.2 32.0 32.6 31.0
0.5 32.5 32 92 735.7 32.0 32.6 31.0
0.25 32.5 32.25 159 668.9 32.1 32.7 31.0
498 C.-C. Chang et al. / Applied Energy 154 (2015) 491499

increases sharply in the precise target scenario in Fig. 7. This is 33

known as an overshoot, and is characteristic of traditional feedback
control. The observations in Table 4 and Fig. 7 suggest the follow-

TCW,out ( O C)
32
ing: implementation of a PI feedback controller and setting the lar-
gest allowable approach as the target temperature achieves
Zone control
maximum energy efciency of an existing VFD-fan system, but at 31
PLC mode
the expense of frequent ON/OFF switching of the fans; the addition Target
of zone control to the PI feedback control signicantly reduces
30
ON/OFF frequency and further conserves energy. 10:04 10:33 11:02 11:31 12:00 12:28 12:57
time

fan power consumption (kW)

5.4. Field experiments
Two short eld experiments were conducted to demonstrate
the validity of the proposed strategies. Since all the facilities has
existed, only one notebook computer with the implementation of
Matlab and Excel software is used to substitute the existing pro-
grammable logic controller. The cooling tower fans were operated
by the control panel in the control room. The eld experiment pro-
cedure was similar to the owchart of virtual plant simulation.
According to cutoff value and power requirement, the operating
frequency was calculated by Table 3 or Eq. (6) The rst experiment
adopted the traditional (precise target) feedback control with a set
point of 32 C, and was run from 10:35 to 15:50 on a day in late
spring. During the experimental period, average air temperature
was 27.2 C, relative humidity was 57.9%, and wet-bulb tempera-
ture was 21.1 C.
Fig. 8 clearly shows that the traditional feedback control keeps
adjusting the fans power, and after one or two cycles, the outlet
temperatures gradually converge to the set point. Fig. 8 shows cor-
responding temperature and power change proles for the simu-
lated rule-coded PLC operation that was run at the same time.
The same virtual plant simulation was repeated for the next case
(shown in Fig. 8). At 12:35, the outlet temperature dropped below
31.5 C, causing one of the fans to be turned OFF. The average
power level was 47 kW, while that for the PLC operation would
be 55 kW. However, the traditional feedback control involves 11
ON/OFF operation, compared with 1 for the PLC operation.
The second eld experiment was run from 9:45 to 13:15 on the
next day, using PI feedback control with a 0.5 C zone as described
in Section 2. During the experimental period, average air tempera-
ture was 31.0 C, relative humidity was 59.3%, and the wet-bulb
33
TCW,out (O C)
100
zone control
80 PLC mode
60
40
20
0
10:04 10:33 11:02 11:31 12:00 12:28 12:57
time
Fig. 9. Experimental zone temperature control strategy.
temperature was 24.5 C. The target temperature zone was set
between 32.0 and 32.5 C.
Fig. 9 shows that the temperature dropped below the lower
bound of the zone interval; at 10:30, the fan frequencies decreased
from 40 Hz to 30 and 35 Hz respectively, with resulting energy
consumption of 55.3 kW. The temperature remained below the
lower zone boundary; at 10:34, one fan turned OFF, and the fre-
quency of the other increased from 35 Hz to 40 Hz, with resulting
energy consumption of 50.4 kW. At 11:00, the temperature
remained below the lower bound of the zone interval; the fre-
quency of the one operational fan decreased from 40 Hz to 35 Hz,
with resulting energy consumption of 44.8 kW. At 11:30, the tem-
perature exceeded the lower bound and the same fan increased
frequency from 35 Hz to 40 Hz. In summary, the fan operation
was only adjusted when the temperature dropped below the lower
limit; otherwise, it remained unchanged. The experimental period
involved 14 fan adjustments, with only 1 ON/OFF operation. The
outlet water temperatures were controlled around the target
region.
Under the same ambient air conditions as the eld experiment,
the operations following the rule-coded PLC are documented in
Fig. 8. At 10:00, outlet water temperature was 32.5 C; according
to the PLC rules, all fans turned ON, with corresponding energy

32
consumption of 63 kW. Subsequently, the temperature uctuated
Precise control slightly but never dropped below 31.5 C, and therefore all the fans
31
PLC mode remained ON. Average energy consumption throughout this period
Target was 56.3 kW compared to 63 kW in the case of PLC operation. The
30 zone control case required 1 ON/OFF operation, whereas the PLC
10:48 12:00 13:12 14:24 15:36
scenario required none.
time Both experiments were conducted at noon with high ambient
air temperature. The resulting energy savings are less than those
fan power consumption (kW)

100
precise control shown in the virtual simulation cases. Nevertheless, both experi-
80 PLC mode ments verify that the proposed strategies, when applied in the
eld, produce similar results to those of the simulation.
60

40
6. Conclusion
20

0 The technical and economic advantages of replacing traditional

10:48 12:00 13:12 14:24 15:36
two-speed fans with VFD-fans in a cooling tower have been gaining
time
popularity in the industry. Concerns over frequent ON/OFF switch-
Fig. 8. Field experiment using precise feedback control with set point of 32 C. ing and the lack of a well-devised controller lead to conservative
 C.-C. Chang et al. / Applied Energy 154 (2015) 491499
operations, which allow for further improvements in energy ef-
ciency. In this work, a PI feedback controller with a temperature
zone setting was proposed for an existing VFD-controlled cooling
tower system to manage the outlet water temperature while
reducing fan energy consumption and solving the problem of fre-
quent ON/OFF switching.
Simulation results showed that the original energy consump-
tion can be reduced by 38% with a 0.75 C temperature zone with-
out increasing ON/OFF frequency. Additionally, 16% of the original
energy consumption can be saved with a 1.5 C temperature zone
and a 50% reduction in ON/OFF switching. An on-line eld experi-
ment was conducted to test the proposed strategies. The results
veried the energy savings and the reduction in ON/OFF switching.
The implementation of the proposed PI feedback controller with
temperature zone settings has been underway at the selected
plant. It is worth noting that the proposed strategies can be imple-
mented without further investment or any additional hardware if
the existing system is operated with a distributed control system
(DCS). Without a DCS, the cost of purchasing the additional hard-
ware needed is still much lower than maintaining the original
operating strategy, given the long-term economic benets of
energy saving, not to mention the reduction in CO2 emissions. It
was concluded that the proposed energy-saving strategies are very
promising for further improvements in any VFD-fan-based cooling
tower system.
Acknowledgement
The authors are grateful for nancial support from the Ministry
of Science and Technology, Taiwan under Grant MOST
103-2622-E007-025.
References
[1] Milosavljevic N, Heikkil P. A comprehensive approach to cooling tower
design. Appl Therm Eng 2001;21:899915.
[2] Kim J-K, Smith R. Cooling water system design. Chem Eng Sci
2001;56:364158.
499
[3] Panjeshahi MH, Ataei A, Gharaie M, Parand R. Optimum design of cooling
water systems for energy and water conservation. Chem Eng Res Des
2009;87:2009.
[4] Zou Z, Guan Z, Gurgenci H. Optimization design of solar enhanced natural draft
dry cooling tower. Energy Convers Manage 2013;76:94555.
[5] Stabat P, Marchio D. Simplied model for indirect-contact evaporative cooling-
tower behaviour. Appl Energy 2004;78:43351.
[6] Yu FW, Chan KT. Improved condenser design and condenser-fan operation for
air-cooled chillers. Appl Energy 2006;83:62848.
[7] Pan T-H, Shieh S-S, Jang S-S, Tseng W-H, Wu C-W, Ou J-J. Statistical multi-
model approach for performance assessment of cooling tower. Energy Convers
Manage 2011;52:137785.
[8] CTI 105-ATC (00) CTI Code Tower Standard Specications for Acceptance Test
Code for Water Cooling Towers: Cooling Technology Institute, 2000.
[9] Sylemez MS. On the optimum performance of forced draft counter ow
cooling towers. Energy Convers Manage 2004;45:233541.
[10] Castro MM, Song TW, Pinto JM. Minimization of operational costs in cooling
water systems. Chem Eng Res Des 2000;78:192201.
[11] Cortinovis GF, Paiva JL, Song TW, Pinto JM. A systemic approach for optimal
cooling tower operation. Energy Convers Manage 2009;50:22009.
[12] Wang J-G, Shieh S-S, Jang S-S, Wu C-W. Discrete model-based operation of
cooling tower based on statistical analysis. Energy Convers Manage
2013;73:22633.
[13] Cohen B. Variable frequency drives: operation and application with
evaporative cooling equipment. CTI J 2007;28:2831.
[14] Muntean N, Volosencu C, Hedes A. A temperature control system for water
cooling towers based on variable speed drives. In: Proceedings of the 9th
WSEAS International Conference on International Conference on Automation
and Information, Bucharest, Romania, 2008. p. 18489.
[15] Muntean N, Hedes A, Scridon S. Practical aspects regarding implementation of
variable speed drives in cooling tower fans. In: Proceedings of the 6-th WSEAS/
IASME Int. Conf. POWER06, Tenerife, Canary Islands, Spain, 2006. p. 705.
[16] Yu FW, Chan KT. Optimization of water-cooled chiller system with load-based
speed control. Appl Energy 2008;85:93150.
[17] Ahn BC, Mitchell JW. Optimal control development for chilled water plants
using a quadratic representation. Energy Build 2001;33:3718.
[18] Cheremisinoff N, Cheremisinoff P. Cooling towers selection, design and
practice. Ann Arbor Science Publishers; 1981.
[19] Picardo JR, Variyar JE. The Merkel equation revisited: a novel method to
compute the packed height of a cooling tower. Energy Convers Manage
2012;57:16772.
[20] Seborg D, Edgar T, Mellichamp D, Doyle F. Process Dynamics and Control:
Wiley, 2010.
[21] Pan T-H, Sheng B-Q, Wong DS-H, Jang S-S. Consumption optimization by Run
to Run control with zone control. J. Taiwan Instit. Chem. Eng. 2010;41:5005.
[22] Gonzlez AH, Odloak D. A stable MPC with zone control. J Process Control
2009;19:11022.
[23] Pan T-H, Shieh S-S, Jang S-S, Wu C-W, Ou J-J. Electricity gain via integrated
operation of turbine generator and cooling tower using local model network.
Energy Conversion, IEEE Trans 2011;26:24555.