IOP Conference Series: Earth and Environmental Science

PAPER • OPEN ACCESS

Prediction of Thermal Performance of Cooling Tower of a Chiller Plant

Using Machine Learning
To cite this article: K Karunamurthy et al 2020 IOP Conf. Ser.: Earth Environ. Sci. 573 012029

View the article online for updates and enhancements.

This content was downloaded from IP address 111.88.231.140 on 28/03/2021 at 14:41

SESBT 2020 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 573 (2020) 012029 doi:10.1088/1755-1315/573/1/012029

Prediction of Thermal Performance of Cooling Tower of a

Chiller Plant Using Machine Learning
1
K Karunamurthy*, 2Debdutta Chatterjee, 2Shivam Pattanashetti, 2Aditya Chhetri,
2
Aswin Sevugan, , 1G Suganya
1
Associate Professor, Vellore Institute of Technology Chennai
2
B.Tech Student, Vellore Institute of Technology Chennai

Abstract. Chiller plants (CP) are accountable for regulating the comfort levels of most indoor environments.
The CP uses water as the working medium and acts as a centralized cooling system for the controlled cooling of
products, in the different production environments, machine tool industries, 3D printing, packaging, heat
exchanging systems and to preserve agricultural produce, dairy products, and other edible items. The CP under
study is used for providing cooling to sixteen storeyed hostel building at VIT Chennai. The refrigerant used in
this system is Tetra fluoro ethane (R134a). The chilled water (ChW) from the evaporator is circulated to the
rooms in the hostel through the secondary circuit and the air handling unit exchanges chillness from the chilled
water and finally supplies to the hostel rooms. The chilled water from the hostel returns to the evaporator in a
closed loop. The cooling water (CW) from the condenser of the refrigeration system rejects heat to the cooling
tower (CT). Thus the performance of the CT is directly linked to the performance of the cooling provided to the
hostel rooms. The objective of this research is to predict the temperature at the outlet of the CT integrated with
the CP using machine learning algorithms. The predicted values are compared with the measured values and
with the values calculated theoretically. The results are analyzed using the standard metrics and are observed to
be appreciable.

Keywords: Machine Learning, Linear Regression, Cooling Tower, Thermal Performance Prediction,
Chiller Plant.

1. Introduction
Temperature regulation is an important requirement for almost every industrial and agricultural
process. The purpose of a CP is to transfer heat from one system (control volume) to another (or
ambient) using water as working medium. Their applications range from cooling of products in cold
storage, comfort cooling under indoor conditions and also other industrial processes & environments.
A CP‟s highest cooling capacity varies according to its operating environments [1]. The CP can use
air, water or brine solution as working medium. Chiller Plants using water has advantage over air-
cooled as they are not vulnerable to outdoor environmental conditions such as rain, sunshine; dusty
winds etc. and hence have a longer lifespan. They are also quiet while operating since flow of water
through ducts and vents are not noisy. To meet the energy efficiency goals of the plant, it is required
to conduct frequent checks for leakages and perform scheduled preventive maintenance [2]. In the
centrally air-conditioned high rise buildings greater than 50% of the overall energy consumption is
consumed by the chiller plant. It is very important to optimize the operating parameters of the chiller
plant to augment its operating efficiency, reduce the energy source depletion and stimulate the exergy
preservation of chiller plants and its associated systems [3]. The properties of the refrigerant are listed
in Table 1 [4]. The primary focus of this article is on operation of the CT which is an integral part of
CP, the CT under study is of induced draft type. The operation of cooling tower is dependent upon the
principle thermal energy exchange from hot water in direct exposure to comparatively cool and arid
air [5]. The performance characteristics of a cooling tower can also be influenced by on the size of the
dewdrops in the spray area along with other parameters such as temperature, humidity and wind
velocity [6]. The effectiveness of a cooling tower can be improved by atomisation of water droplets
into fine size [7]. In another study, it has been observed that for 5-10°C rise in the dry bulb

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
SESBT 2020 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 573 (2020) 012029 doi:10.1088/1755-1315/573/1/012029

temperature, the evaporation losses are increases by 20% [8]. It is quite challenging to model the
performance of a cooling tower quite precisely and accurately using classical modelling techniques
that cope with the solution of complex differential equations [9]. Henceforth Supervised Machine
Learning algorithm has been used to train the dataset and the outlet water temperature of the CT was
predicted. The properties of the refrigerant R134a are listed in Table 1 and the specification of the
chiller is mentioned in Table 2.

Table 1. Properties of R134a Table 2. Chiller Plant Specifications

Properties Values Model 30XW1712P Refrigerant R-134a
Boiling Point -26.1C Voltage 415 V Max Power 446 kW
Auto Ignition Temperature 770C Input
Ozone Depletion Level 0 Frequency 50 Hz Max Current 674 A
Solubility in water at 25C 0.11% wt Phase 3 Weight 9368 kg
Critical Temperature 122C Pressure 0-15 bar Pressure 14.5 bar
Global Warming Potential 1200 (Min/Max) Setting

2. The Chiller Plant

The CP taken for study is a system which exchanges heat from atmospheric air with ChW liquid
through a vapour compression refrigeration (VCR) cycle/system. The CP operates with three sections
namely air conditioning section, chiller section, and cooling tower section which have been
schematically illustrated in the following Figure 1.

Figure 1. Schematic diagram of Chiller Plant

Air conditioning section is located in the hostel which contains pipelines and it receives the ChW
from evaporator section of CP. After the thermal exchange between the air and ChW, the warm ChW
returns back to the chiller (evaporator of VCR system). VCR system consists of four major
components viz., compressor unit, condenser unit, throttling valve, and evaporator coil. The VCR
system uses the refrigerant R134a as working medium in the primary circuit. The warm ChW leaving
from the air conditioning unit transfers heat to the refrigerant of the VCR system at its evaporator
section. The heat exchange medium in the cooling tower is CW from the condenser of CP and

2
SESBT 2020 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 573 (2020) 012029 doi:10.1088/1755-1315/573/1/012029

atmospheric air. This CW dissipates thermal energy to the atmospheric air after receiving the heat
from the refrigerant at the condenser section of the CP. The chiller is made by Carrier Corporation
and its specifications provided in Table 2. . In order to be efficient the chiller section should operate
with reduced/increased refrigeration capacity with the decrease/increase in cooling load [10].

3. Data Reduction

The various performance parameters of the cooling tower and its mathematical relations are
provided in this section. The performance parameters depend on the CW outlet temperature.

(i) Theoretical value of CW outlet temperature( )

The theoretical outlet temperature of cooling water from CT was calculated using the energy
balance,

Heat lost by CW = Heat gained by air + Heat lost due to evaporation

(1)
Where
m1 , m2 - Mass flow rate of CW, air (kg/s)
Cp1 , Cp2 - Specific heat of water, air (kJ/kgK),
, T2 - Inlet and outlet temperature of CW (°C)
T3, T4 - Inlet and outlet temperature of air (°C)
m3 - Mass of water evaporated (kg/s)
Hv - Latent Heat of evaporation (kJ/kg)
(ii) Range, R= T1-T2act (°C) (2)
(iii) Approach = T2act-Twb (°C) (3)
Where Twb – Wet bulb temperature (°C)
(iv) Effectiveness = (4)
(v) Cooling Capacity = (kW) (5)
(vi) Blowdown loss, B = –
(m3/s) (6)
E - Evaporative flow loss = 0.00085 1.8 R Q1 (m3/s) (Evaporative Loss is
calculated for a Range of 2.3 °C and kept constant throughout.)
CoC - Cycle of Concentration = 3
(vii) Drift Loss, (m3/s) (7)
Q1 - Volume flow rate of water ( /s)
(viii) Make-up Water Requirement, M = E + B + D (m3/s) (8)

4. Prediction Using Machine Learning Algorithm

a. Linear Regression:

Let the input vector x ∊ {x1, x2, . . . , xm} represent the input features (otherwise called as
independent variables or predictor variables) and the output or response or dependent variable, y

3
SESBT 2020 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 573 (2020) 012029 doi:10.1088/1755-1315/573/1/012029

represents the required output. Linear regression works by creating a regression line equation
involving all training data using least square method. The impact of every input feature towards
the predicted output is observed and calculated using least square method. These weights that are
computed for all input features are defined to be the coefficients, β1, β2, …, βm. The common
bias of the data is observed and stored as β0 (called as intercept) of the system. The representation
can be understood by the following equation.

The parameters and intercept „c‟ is calculated using least square method to ensure
minimal difference between predicted and actual values.

b. Dataset creation:
The following are observed to be the input parameters viz., CW inlet water temperature, Dry bulb
& Wet bulb temperature of air entering, exit air temperature, mass flow rate of air & water and the
output variable will be the CW outlet temperature of CT. The control unit of chiller plant is
observed over a period of 46 days continuously at regular intervals of 1 hour to capture the
defined inputs making a total of 414 experiences. The dataset is collected from a live environment
using the electronic gadgets and hence prone to have some errors. Hence, pre-processing becomes
mandatory before using the dataset for training the model.

Feature wise, the data is verified to have any missing values. If some values are missing, they are
filled with mean of the values of corresponding feature. The dataset is then normalized. This is
done since the measuring units are different for various features. Hence, feature values are
standardized using min-max normalization to fall in the standard range from 0 to 1.

Figure 2: Correlation between features

Once completed, the dataset is checked for linearity to justify the usage of linear regression. The
linearity check is performed using correlation analysis. A positive (+0.8 to +1) or negative (-0.8 to
-1) correlation is witnessed to suit well with regression application whereas the correlation value
near to zero indicates absence of relation between independent and dependent feature. The
correlation values are presented in Fig.2.

4
SESBT 2020 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 573 (2020) 012029 doi:10.1088/1755-1315/573/1/012029

c. Creation of Machine Learning model

The pre-processed dataset is splitted into training and testing data in the ratio of 80-20. The model
is implemented in Python using Jupyter notebook as the platform.

Input Data
set

Missing values are filled using mean of the corresponding feature

Normalization of i

Splitting into training and testing dataset (80% - 20%)

Modelling using Multiple Linear Regression

Change hyper-
parameters

Testing of Model using testing dataset

False Accuracy
>= threshold

True

Stop

Figure 3. Process flow diagram of Predictive Model

A predictive model for determining the CW outlet temperature is illustrated in Figure 3 in the form of
a flow chart. 80% data chosen at random is used for training the ML model and is resulted with the
following regression line.

Where, {x1, x2, . . . , xm} are the values residing in training dataset. From the equation it is evident
that the feature titled x6 doesn‟t have any correlation with output and the impact or coefficient
becomes zero. The computed coefficients and intercept are tested using the testing dataset. The
input data is passed to equation and output is calculated for the separated 20% data. The difference

5
SESBT 2020 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 573 (2020) 012029 doi:10.1088/1755-1315/573/1/012029

between actual value and the predicted value is known as the error. To remove sign imbalance in the
result, calculated differences are squared and taken average and is called as mean square error.

A sample dataset with measured and predicted values is shown in Table 3.

Table 3. Sample Dataset (Temperature in C and flow rate in kg/s)

CW Outlet
CW Inlet Wet Bulb Dry Bulb Outlet Air Flow Rate Flow Rate Temperature
S.No.
Temperature Temperature Temperature Temperature of Air of Water
Measured Predicted

0 30.1 24 30 40.2 33.067623 55.111944 28.5 28.554

1 30.2 24 30 40.2 32.710661 55.111944 29 28.64
2 30.5 23.5 29.7 40.2 31.802032 55.111944 28.5 28.85
3 30.9 22.5 30 40.2 32.029189 55.111944 29.5 29.056
4 32 24 30 40.2 33.132525 55.111944 30 29.664

5. Results and Summary

The accuracy of the trained model is tested using standard metrics, Root Mean Square Error (RMSE),
and R2 value. The model results in RMSE value of 0.96 and R2 score of 0.71 indicating a minimal
error rate. Figure 4a represents the relation between actual and predicted values of outlet temperature
against the inlet temperature. Figure 4b illustrates the 3d view of outlet water temperature (Measured
and Predicted) distribution against inlet water temperature and wet bulb temperature of incoming air.
From these figures 4a and 4b, it can be inferred visually that, the deviation of predicted outlet
temperature is very small.

Figure 4a. Actual V/s Predicted values Figure 4b. Actual vs Predicted

6. CONCLUSION
The results of the predictive model using ML algorithm are compared with measured values and it is
convincing. In any heat exchange system the energy loss is inevitable and its performance decreases
as the system‟s life time increases. Once the decrease in performance is alarming then the cause of the
problem is identified and suitable solution is provided. However, continuous performance monitoring
of the CP is possible by comparing the predicted CW outlet temperature and the measured CW outlet

6
SESBT 2020 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 573 (2020) 012029 doi:10.1088/1755-1315/573/1/012029

temperature. If the deviation is too much then the cause can be identified and suitable control
measures or preventive measures for energy loss can be carried out instantly. Hence, the performance
of the CP can be at its best, throughout the period of operation in its life span. This allows optimum
circulation of cooling water to achieve optimum heat transfer, thereby minimizing the pumping losses
& thermal energy losses and ultimately sustainability of the environment is achieved.

References

[1] Yongjun Sun, Shengwei Wang , Fu Xiao “In situ performance comparison and evaluation of three chiller
sequencing control strategies in a super high-rise building”, Energy and Buildings, Vol 61, (2013), pp. 333-343.
https://www.sciencedirect.com/science/article/abs/pii/S0378778813001357?via%3Dihub

[2] Sandy Younger, “Advantages and Disadvantages of Water-Cooled Chillers”, (2016).

https://www.jcyounger.com/advantages-disadvantages-watercooled-chillers/

[3] Xuan Zhou, Bingwen Wang, Liequan Liang, Junwei Yan and Dongmei Pan, “ Optimization Method for the
Chiller plant of Central Air-conditioning System Parameters on Association Rules Analysis for Energy
Conservation”, International High Performance Building Conference, Purdue University, (2018).
https://docs.lib.purdue.edu/cgi/viewcontent.cgi?article=1292&context=ihpbc

[4] Air conditioning systems https://www.airconditioning-systems.com/R134a.html

[5] Nenad Milosavljevic , Pertti Heikkila, “ A comprehensive approach to cooling tower design”, Applied
Thermal Engineering vol.21 (2001) pp.899-915
https://www.sciencedirect.com/science/article/abs/pii/S1359431100000788

[6] S.P. Fisenko, A.I. Petruchik, A.D. Solodukhin, “Evaporative cooling of water in a natural draft cooling
tower“, International Journal of Heat and Mass Transfer, Vol. 45(23), (2002), pp. 4683-4694.
https://doi.org/10.1016/S0017-9310(02)00158-8

[7] Xuehong Chen, Fengzhong Sun, Dongqiang Lyu, “Field test study on water droplet diameter distribution in
the rain zone of a natural draft wet cooling tower”, Applied Thermal Engineering, Vol. 162(5), (2019).
https://doi.org/10.1016/j.applthermaleng.2019.114252 .

[8] Bilal Ahmed Qureshi, Syed M. Zubair , “Prediction of Evaporation Losses in Wet Cooling Towers” Journal
Heat Transfer Engineering, Vol 27(9), (2006), pp. 86-92.
https://doi.org/10.1080/01457630600846372

[9] M. Hosoz, H.M. Ertun , H. Bulgurcu, “Performance prediction of a cooling tower using artificial neural
network, Energy Conversion and Management, Vol.48, (2007), pp. 1349–1359.
https://doi.org/10.1016/j.enconman.2006.06.024 .

[10] Demis Pandelidis, Marlena Drag˛, Paweł Drag, William Worek, Sabri Cetin, “Comparative analysis
between traditional and M-Cycle based cooling tower”, International Journal of Heat and Mass Transfer,
Vol.159, (2020), pp. 1-13.
https://doi.org/10.1016/j.ijheatmasstransfer.2020.120124