673529
673529
673529
COMPUTER-AIDED
EXERGY AND ENERGY ANALYSIS OF
THE VACUUM DISTILLATION UNIT
M.Sc. THESIS
Sena KURBAN
JUNE 2021
ISTANBUL TECHNICAL UNIVERSITY GRADUATE SCHOOL
COMPUTER-AIDED
EXERGY AND ENERGY ANALYSIS OF
THE VACUUM DISTILLATION UNIT
M.Sc. THESIS
Sena KURBAN
(506181025)
JUNE 2021
ISTANBUL TEKNİK ÜNİVERSİTESİ LİSANSÜSTÜ EĞİTİM ENSTİTÜSÜ
Sena KURBAN
(506181025)
HAZİRAN 2021
Sena KURBAN, a M.Sc. student of ITU Graduate School student ID 506181025,
successfully defended the thesis entitled “COMPUTER-AIDED EXERGY AND
ENERGY ANALYSIS OF THE VACUUM DISTILLATION UNIT”, which she
prepared after fulfilling the requirements specified in the associated legislations,
before the jury whose signatures are below.
v
vi
To my dear and pretty family,
vii
viii
FOREWORD
ix
x
TABLE OF CONTENTS
Page
FOREWORD ............................................................................................................. ix
TABLE OF CONTENTS .......................................................................................... xi
ABBREVIATIONS ................................................................................................. xiii
SYMBOLS ................................................................................................................ xv
LIST OF TABLES .................................................................................................. xix
LIST OF FIGURES ................................................................................................ xxi
SUMMARY ........................................................................................................... xxiii
ÖZET..................................................................................................................... xxvii
1. INTRODUCTION .................................................................................................. 1
2. DESCRIPTION AND SIMULATION OF PROCESS ....................................... 7
2.1 Description of Process ........................................................................................ 7
2.1.1 Feed preheat section with heat exchanger network ..................................... 7
2.1.2 Vacuum distillation unit furnace ................................................................. 8
2.1.3 Vacuum distillation unit .............................................................................. 8
2.1.4 Amine units ................................................................................................. 9
2.2 Process Simulation ............................................................................................. 9
2.2.1 Defining the components and the property models .................................. 11
2.2.2 Defining the fluid package ........................................................................ 12
2.2.3 Assay characterization of atmospheric residue ......................................... 14
2.2.4 Building and specifying process flowsheet ............................................... 15
2.2.4.1 Material and energy streams .............................................................. 16
2.2.4.2 Unit operation blocks ......................................................................... 16
3. STUDY OF THE ENERGY AND EXERGY ANALYSIS ............................... 23
3.1 The Exergy Concept ......................................................................................... 24
3.2 The Exergy of Closed Systems......................................................................... 25
3.3 Exergy of Flows ............................................................................................... 27
3.3.1 Exergy of material flow............................................................................. 27
3.3.2 Heat exergy ............................................................................................... 28
3.3.3 Work exergy .............................................................................................. 28
3.4 Energy and Exergy Efficiency.......................................................................... 29
3.4.1 Energy efficiency ...................................................................................... 29
3.4.2 Exergy efficiency ...................................................................................... 29
3.4.2.1 Inlet-outlet exergy efficiency ............................................................. 30
3.4.2.2 Consumed-produced exergy efficiency.............................................. 30
3.5 Exergy Losses ................................................................................................... 31
4. MACHINE LEARNING MODELS ................................................................... 35
4.1 Literature Review of Machine Learning about Exergy Analysis ..................... 38
4.2 Modeling of the Vacuum Distillation Column Using ML Methods ................ 40
4.2.1 Bootstrap aggregated neural network ....................................................... 41
4.2.2 Random forest ........................................................................................... 43
5. RESULTS AND DISCUSSIONS ........................................................................ 45
xi
5.1 Comparison of Simulation Results with Real Values ...................................... 45
5.2 Energy and Exergy Analysis of All Unit Operations ....................................... 47
5.2.1 Energy and exergy calculation of a heat exchanger .................................. 48
5.2.2 Energy and exergy calculation of a furnace or heater ............................... 50
5.2.3 Energy and exergy calculation of a pump ................................................. 50
5.2.4 Energy and exergy calculation of a vacuum distillation column .............. 51
5.3 Exergy Study in Vacuum Distillation Column with Using Machine Learning 57
5.4 Exergy Study in Vacuum Distillation Column with Using Machine Learning 58
6. CONCLUSION ..................................................................................................... 65
REFERENCES ......................................................................................................... 67
CURRICULUM VITAE .......................................................................................... 75
xii
ABBREVIATIONS
xiii
SLT : Second Law of Thermodynamics
SSL : Semi Supervised Learning
TBP : True Boiling Point
UL : Unsupervised Learning
VDU : Vacuum Distillation Unit
VGO : Vacuum Gas Oil
VR : Vacuum Residue
xiv
SYMBOLS
C : Carbon
C : Column
℃ : Degree Celsius
cv : Control Volume
E : Heat Transfer Equipment
𝐄̇𝐞 : Exit Energy Rate
𝐄̇𝐢 : Inlet Energy Rate
Ex : Exergy Rate
𝐞𝐱 : Specific Exergy
𝐄𝐱 𝐜𝐨𝐥𝐝 : Cold Stream Exergy Rate
𝐄𝐱̇ 𝐞 , 𝐄̇𝐱 𝐨 : Exit or Outlet Exergy Rate
𝐄𝐱̇ 𝐢 : Inlet Exergy Rate
𝐄𝐱 𝐟𝐥𝐨𝐰,𝐩𝐡 : Physical Flow Exergy Rate
𝐄𝐱 𝐟𝐥𝐨𝐰 : Flow Exergy Rate
𝐄𝐱 𝐡𝐨𝐭 : Hot Stream Exergy Rate
𝐞𝐱 𝐢𝐜𝐡 : Specific Chemical Exergy
𝐄𝐱 𝐢𝐧 : Inlet Exergy Rate
𝐄𝐱 𝐤𝐢𝐧 : Kinetic Exergy Rate
𝐜𝐡
𝐞𝐱 𝐦𝐢𝐱 : Mixture Specific Chemical Exergy
𝐄𝐱 𝐧𝐨𝐧𝐟𝐥𝐨𝐰 : Nonflow Exergy Rate
𝐄𝐱 𝐨 : Chemical Exergy Rate
𝐄𝐱 𝐨𝐮𝐭 : Outlet Exergy Rate
𝐄𝐱 𝐩𝐡 : Physical Exergy Rate
𝐄𝐱 𝐩𝐨𝐭 : Potential Exergy Rate
𝐄𝐱 𝐐 : Heat Exergy Rate
F : Furnace
f(X) : Aggregated Neural Network Forecaster
G : Pump
g : Gravity Acceleration
xv
𝐇 : Enthalpy
h : Hour
𝐡 : Specific Enthalpy
𝐇𝐢𝐧 : Inlet Enthalpy
𝐇𝟎 : Reference Enthalpy
𝐡𝟎 : Reference Specific Enthalpy
𝐇𝐨𝐮𝐭 : Outlet Enthalpy
H2 : Hydrogen
H2 O : Water
𝐈 : Irreversibility
i, f : Initial and Final Status
i, j : Species
𝐈̇𝐜𝐯 : Irreversible Exergy Loss of a Control Volume
in : Inlet or Input
K : Kelvin
kin : Kinetic
kj : Kilojoule
kPa : Kilopascal
𝐋𝐇𝐕𝐢 : Lower Heating Value
𝐦̇𝐞 : Exit Mass Flow Rate
𝐦̇𝐢 : Inlet Mass Flow Rate
mix : Mixture
n : Neural Networks Number
𝐍𝐢 : Moles Number
N2 : Nitrogen
out : Outlet or Output
O2 : Oxygen
P : Pressure
P0 : Reference Pressure
ph : Physical
pot : Potential
𝐐 : Heat Transfer Interaction
𝐐̇ : Heat Transfer Rate
𝐐̇𝐜𝐯 : Heat Rate into Control Volume
𝐑 : Molar Gas Constant
xvi
R2 : Coefficient of Determination
rev : Reversible
𝐒 : Entropy
s : Second
𝐬 : Specific Entropy
S : Sulfur
SG : Specific Gravity
𝐒𝐠𝐞𝐧 : Entropy Generation
𝐒𝟎 : Reference Entropy
𝐬𝟎 : Reference Specific Entropy
T : Temperature
t : Ton
T0 : Reference Temperature
𝐔 : Internal Energy
𝐔𝟎 : Reference Internal Energy
𝐕 : Volume
𝐯 : Velocity
𝐕𝟎 : Initial Volume
𝐯𝟎 : Initial Velocity
𝐖̇𝐜𝐯 : Control Volume Work
𝐖𝐚𝐢𝐫𝐜𝐨𝐨𝐥𝐞𝐫 : Air Cooler Work
𝐖̇𝐜𝐯
𝐫𝐞𝐯
: Reversible Control Volume Work
𝐰𝐢 : Aggregating Weight of ith Neural Network Combination
𝐖𝐍𝐄𝐓 : Net Work
𝐖𝐩𝐮𝐦𝐩 : Pump Work
𝐗, 𝐱 : Neural Network Inputs Vector
𝐱𝐢 : Mole Fraction
𝐳 : Elevation
𝐳𝟎 : Reference Elevation
𝛃𝐢 : Correction Factor
𝛄𝐢 : Component Activity Coefficient
𝛍𝐢𝟎 : Chemical Potential
𝛍𝐢𝟎𝟎 : Equilibrium State of Chemical Potential
𝛈 : Energy Efficiency
xvii
Σ : Sum
𝛙 : Exergy Efficiency
0 : Reference Point
1,2,3 : States
xviii
LIST OF TABLES
Page
xix
xx
LIST OF FIGURES
Page
xxi
xxii
COMPUTER-AIDED EXERGY AND ENERGY ANALYSIS OF THE
VACUUM DISTILLATION UNIT
SUMMARY
Recently, studies on energy efficiency have become very important due to greenhouse
gas emissions, raw material costs and resource constraints. Energy efficiency
encourages researchers to work on these issues in terms of both reducing costs and
protecting the environment. In order to carry out energy efficiency studies, firstly, the
sources that cause energy consumption should be analyzed, so that the necessary
actions can be taken as a result of these analyzes. Among these actions, the options to
be evaluated include studies such as recovery of process waste heat, prevention of
unwanted leaks, evaluation of new technological possibilities, regular maintenance of
equipment, as well as the implementation of designs and operations that will increase
energy efficiency.
Refineries are complex energy-intensive industry that are both energy producers and
consumers. For this reason, a wide-ranging energy management is required in
refineries. Refineries consist of distillation columns that process crude oil to obtain
different petroleum products. The refinery consists of units such as crude oil
distillation, vacuum distillation, hydrocracker, catalytic cracking, hydrodesulfurizer,
and isomerization. The unit where crude oil is first processed is the crude oil
distillation unit. In the crude oil distillation unit, gas, LPG, light and heavy naphtha,
kerosene, light and heavy diesel and finally atmospheric bottom products are obtained,
respectively. These products can be sold directly or sent to other units to create new
products in the refinery.
Atmospheric bottom products with a very high boiling point coming from the crude
oil distillation unit are sent to the vacuum distillation unit for processing. It is ensured
that the desired products are obtained without thermal decomposition by utilizing the
boiling points drop of the products under vacuum pressure. In the vacuum distillation
unit, atmospheric bottom products first pass through the preheat train and furnace to
reach the desired charging temperature, and then are fed to the vacuum distillation
column. From light to heavy, waste gas, light vacuum gas oil, heavy vacuum gas oil
vacuum bottom product are obtained, respectively. The waste gas is cleaned in amine
units and this cleaned gas used as fuel in the furnace. The other products obtained are
intermediate products and sent to units such as catalytic hydrocrackers or
hydrocrackers for processing.
Energy analysis is a traditional method for evaluating the performance of systems and
is connected to the first law of thermodynamics known as conservation of energy. In
addition, energy efficiency calculations and energy balances are used in the energy
analysis method. However, energy balance cannot give any information about the
degradation of energy resources. Therefore, energy analysis cannot measure energy
losses and energy quality. In addition, energy efficiency calculations used to analyze
the performance of the system can be misleading or confusing in some cases.
xxiii
At this point, the exergy method copes with the limits of the first law of
thermodynamics. Exergy is related to both the first law and the second law of
thermodynamics. The second law of thermodynamics presents the concept of entropy,
which expresses the disorder in the system. Accordingly, the total entropy of a system
either increases or remains constant in any random process, but never decreases.
Consequently, the second law of thermodynamics is concerned with the quality of the
system corresponding to the conversion of energy into work. Exergy is defined as the
maximum theoretical useful work that can be achieved as the system becomes
completely thermodynamic equilibrium with its environment.
Exergy analysis is an important tool in determining the energy losses and measuring
the quality of the process. Because there is always a degradation in energy quality for
real processes. Therefore, exergy analysis results give meaningful results. At this
point, exergy analysis studies provide information to designers, engineers and
researchers about the performance of the system. At the same time, there are many
useful studies about the exergy analysis in the literature to evaluate the performance
of systems and to improve the performance of the system. Exergy has both system and
reference environmental properties. If a system is in equilibrium with its environment,
that system has no exergy and this is called a dead state. If the exergy of the system
increases, exergy deviates from the environment. The total system exergy consists of
a combination of potential, kinetic, physical and chemical exergy. Exergy can be
transferred as mass, heat and work.
Exergy efficiency, one of the system performance indicators, measures the approach
to ideal conditions. The higher the system has exergy efficiency, the higher there is
energy quality of system. There are many equations related to exergy efficiency in the
literature. It is usually expressed as the ratio of the sum of the output exergy to the sum
of the input exergy. Exergy efficiency is always between zero and one. Another
performance indicator is the expression of irreversibility or exergy waste, which
indicates the deviation of the system from ideal conditions. This irreversibility
expression can be calculated by the exergy balance or by using the relevant equations
in the literature.
Some procedures should be followed in order to apply exergy analysis. At the
beginning, the system to be examined is decided and if necessary, system elements can
be evaluated separately. Required operational values such as temperature, pressure and
flow are reached for exergy and energy calculations. Energy calculations are done
using common mass and energy balances. For exergy analysis, a reference
environment is selected and exergy calculations are done according to this reference.
In addition, exergy balance is used to calculate the exergy consumption. Appropriate
equations are also used for the exergy efficiency of the system. Finally, these
calculated values are interpreted. Inefficiencies in the system are determined and
actions are taken to improve system performance depending on the need.
There are many equipment in refineries that can be thermodynamically examined.
These equipment can be heat exchangers, chillers, air coolers, furnaces, distillation
columns, pumps and compressors. Simulation is an important tool for modeling
refineries, and various software products were developed for this. One of them is
ASPEN HYSYS, which is widely used. Thanks to the interactive platform provided
by HYSYS, users set up and run their models effectively and quickly. In addition,
ASPEN HYSYS provides comprehensive thermodynamic infrastructure to accurately
calculate phase behavior, physical and transport properties for refining, oil or gas
xxiv
industries. HYSYS offers users many basic unit operations such as separators, reactors,
rotary equipment, heat transfer and logical operations. It also provides a good
environment for performing exergy and energy analysis calculations and performing
various case studies in different operational conditions.
In this study, energy and exergy analysis of the vacuum distillation unit in TUPRAS
Refinery was performed. The main purpose of the study is to discover the
inefficiencies of all unit operations in the vacuum distillation unit with exergy analysis
studies. In addition, these objectives include interpreting energy losses in operations
with energy inefficiencies and identifying fields that may have potential for
improvement. Another aim of the study is to see the effect of operational parameters
on column exergy efficiency depending on the process control limits. Finally, the
exergy efficiency prediction of vacuum distillation column was studied using
machine-learning models. For this, parameters affecting the column were used as
model inputs. The purpose of developing prediction models is to see and evaluate the
results on the online platform without the need for the continuous ASPEN HYSYS
working environment and intensive exergy calculations in the background. At the same
time, it can be determined whether this operation is efficient or not for any operational
condition by using the predictive models on the column.
In beginning, the study starts with a detailed literature search to understand the exergy
concept and to reach the equations to be used in exergy calculations. Then, a vacuum
distillation unit was simulated in ASPEN HYSYS in order to carry out energy and
exergy studies. For the simulation, a base date which has all data was selected. Actual
plant data were used for simulation. These data are laboratory values and operational
values such as temperature, pressure and flow. In addition, the equipment technical
reports were used in the simulation. Overall unit operations such as heat exchanger,
heater, cooler, air cooler, pump, furnace and column was simulated in HYSYS. The
comparison of the actual values with the HYSYS model results was made for each
operational value. The model was updated continuously until actual values and model
results converge. After the last accurate model simulation was obtained, exergy and
energy analysis calculations were started. In exergy calculations, exergy transfer
equations were taken into consideration. For mass or flow exergy calculations,
calculations were done using the enthalpy, entropy, exergy values of the simulation
streams and the equations in the literature. Also, the exergy efficiency calculations in
the literature were done for each equipment. Equipment with low exergy efficiency
was identified. Evaluations about where the performance degradation of these
equipment may come from were done. In addition, the exergy and energy efficiency
values of the equipment were compared. Moreover, the effect of the parameters on the
column exergy efficiency was evaluated using the case study function of ASPEN
HYSYS according to the process control limits. The most important parameters
affecting the column exergy efficiency were determined.
Synthetic data was produced in ASPEN HYSYS by using the parameters that affect
the column most. Prediction models were developed using this data set with machine
learning methods. For this, BANN, which is one of the artificial neural network
models, and Random Forest models, one of the ensemble learning methods, were
developed in the R program. The "Ipred" and "randomForest" packages of the R
program were used for BANN and the Random Forest model, respectively. The data
for both models are divided into training, testing, and invisible validation data sets.
The accuracy and success of the model was interpreted by looking at the fit between
the last real values and the model predicted values.
xxv
xxvi
VAKUM DİSTİLASYON ÜNİTESİNİN BİLGİSAYAR DESTEKLİ EKSERJİ
VE ENERJİ ANALİZİ
ÖZET
Son zamanlarda sera gazı emisyonları, ham madde maliyetleri, kaynak sıkıntısı
nedeniyle enerji verimliliği konulu çalışmalar oldukça önem arz etmektedir. Enerji
verimliliği, hem maliyetleri azaltması hem de çevreyi koruması açısından
araştırmacıları bu konularda çalışmalara teşvik etmektedir. Enerji verimliliği
çalışmalarının gerçekleştirilmesi için öncelikle enerji tüketimine sebebiyet veren
kaynakların analiz edilmesi gerekmektedir böylece bu analizler sonucunda gerekli
aksiyonlar alınabilir. Bu aksiyonlar arasında değerlendirilmesi gereken seçenekler
proses atık ısılarının geri kazanılması, istenmeyen kaçakların önlenmesi, yeni
teknolojik olanakların değerlendirilmesi, ekipmanlarda düzenli bakımların yer alması
aynı zamanda enerji verimliliğini yükseltecek tasarımların ve operasyonların
uygulanması gibi çalışmalar yer almaktadır.
Petrol rafinerileri enerji yoğun karmaşık yapılar olup hem enerji üreticisi hem de
tüketicisidir. Bu nedenle petrol rafinerilerinde geniş çaplı bir enerji yönetimi
gereklidir. Rafineriler ham petrolü işleyerek farklı petrol ürünlerini elde eden
distilasyon kolonlarından oluşmaktadır. Rafineri, ham petrol distilasyon, vakum
distilasyon, hidrokraker, katalitik kraking, hidrodesülfürizer, izomerizasyon gibi
ünitelerden oluşmaktadır. Ham petrolün ilk işlendiği ünite ham petrol distilasyon
ünitesidir. Ham petrol distilasyon ünitesinde hafiften ağıra sırasıyla gaz, likit petrol
gazı (LPG), hafif ve ağır nafta, kerosen, hafif ve ağır dizel ve son olarak atmosferik
dip ürünleri elde edilir. Bu ürünler doğrudan satışı yapılmakla birlikte rafineride yeni
ürünler oluşturmak için başka ünitelere gönderilebilir.
Ham petrol işleme ünitesinden elde edilen kaynama noktası çok yüksek olan
atmosferik dip ürünler işlenmek üzere vakum distilasyon ünitesine gönderilir. Vakum
basıncında ürünlerin kaynama noktalarının düşmesinden yararlanarak ısıl bozunmaya
uğramadan istenilen ürünlerin elde edilmesi sağlanır. Atmosferik dip ürünler vakum
ünitesinde öncelikle ön ısıtıcı ağlarından ve fırından geçerek istenilen şarj sıcaklığına
getirilir ve vakum distilasyon kolonuna beslenir. Hafiften ağıra sırasıyla atık gaz, hafif
vakum gaz yağı, ağır vakum gaz yağı vakum dip ürünü elde edilir. Atık gaz amin
ünitelerinde temizlenerek tekrardan fırında yakıt olarak kullanılır. Elde edilen diğer
ürünler ara ürün olup işlenmek üzere katalitik hidrokraker veya hidrokraker gibi
ünitelere gönderilir.
Enerji analizi sistemlerin performansını değerlendirmede geleneksel bir yöntem olup
enerjinin korunumu olarak bilinen termodinamiğin birinci kanunuyla ilişkilidir. Enerji
analizi yönteminde enerji verimliliği hesaplamaları ve enerji denkliği işlemleri
kullanılır. Bununla birlikte, kullanılan enerji dengesi, enerji kaynaklarının bozulması
hakkında hiç bir bilgi veremez. Böylece enerji analizi, enerji kayıplarını ve enerji
kalitesini ölçememektedir. Ayrıca sistemin performansını incelemek için kullanılan
enerji verimliliği hesaplamaları bazı noktalarda yanlış yönlendirebilmekte veya kafa
karışıklığı yaratabilmektedir.
xxvii
Ekserji metodu bu noktada termodinamiğin birinci kanunu limitlerini aşmaktadır.
Ekserji konsepti hem termodinamiğin birinci kanunu hem de ikinci kanunuyla
ilişkilidir. Termodinamiğin ikinci kanunu sistemdeki düzensizliği ifade eden entropi
kavramını sunar. Buna göre bir sistemin toplam entropisi herhangi bir rastgele süreçte
ya artar ya da sabit kalır, asla azalmaz. Sonuç olarak termodinamiğin ikinci yasası
enerjinin işe dönüşebildiği yani kalitesini ele almaktadır. Ekserji, sistemin çevresiyle
tümüyle termodinamik denge haline gelirken elde edebilecek maksimum teorik yararlı
iş olarak tanımlanır.
Ekserji analizi prosesteki enerji kayıplarını belirlemede ve prosesin kalitesini ölçmede
önemli bir araçtır. Çünkü gerçek proseslerde her zaman enerji kalitesinde bir bozulma
vardır. Bu nedenle ekserji analizi sonuçları anlamlı çıktılar vermektedir. Bu noktada
ekserji analizi çalışmaları, tasarımcılara, mühendislere ve araştırmacılara sistemin
performansı hakkında bilgi vermektedir. Aynı zamanda literatürde ekserji analizi
kullanılarak sistemlerin performansını değerlendirmek ve sistemin performansını
geliştirmek için birçok faydalı çalışma da bulunmaktadır.
Ekserji sadece termodinamik kavramın ötesinde hem sistem hem de referans çevre
özelliklerine sahiptir. Bir sistem çevresiyle dengedeyse o sistemin ekserjisi yoktur ve
bu durum ölü durum olarak adlandırılır. Eğer sistemin ekserjisi artarsa ekserji çevreden
sapmaktadır. Toplam sistem ekserjisi potansiyel, kinetik, fiziksel ve kimyasal
ekserjilerin birleşiminden oluşur. Ekserji kütlesel, ısı ve iş olarak aktarılabilir.
Sistem performans indikatörlerinden biri olan ekserji verimliliği ideal koşullara
yaklaşımın ölçüsüdür. Ekserji verimliliği enerji verimliliğe göre sistemin
performansını anlamada önemli bir araçtır. Sistem ne kadar yüksek ekserji verimine
sahipse o kadar yüksek enerji kalitesine sahiptir. Literatürde ekserji verimliliği ile ilgili
eşitlikler oldukça fazladır. Genelde, çıkış ekserjilerinin toplamının giriş ekserjilerinin
toplamına oranı olarak ifade edilir. Ekserji verimliliği her zaman sıfır ile bir
arasındadır. Diğer bir performans indikatörü ise sistemin ideal koşullardan sapmasını
gösteren tersinmezlik ya da ekserji atığı ifadesidir. Bu tersinmezlik ifadesi ise ekserji
dengesinden hesaplanarak ya da literatürde ilgili eşitlikler kullanılarak hesaplanabilir.
Enerji analizi çalışmalarında geleneksel olarak kullanılan enerji kavramı ile ekserji
kavramı arasında farklılıklar vardır. Örneğin; enerji her zaman korunurken, ekserjide
tersinmez veya gerçek prosesler için her zaman bir yıkım söz konusudur. Enerji sadece
miktar ölçüsünü dikkate alırken, ekserji hem miktarı hem de kaliteyi ölçer. Enerji
çevreden bağımsız olup ekserji çevre ile bağımlıdır.
Ekserji analizini uygulayabilmek için bazı prosedürler takip edilmelidir. Buna göre
başlangıçta incelenmek istenen sisteme karar verilir, gerekirse sistem elemanları ayrı
ayrı değerlendirilebilir. Ekserji ve enerji hesaplamaları için gerekli olan sıcaklık,
basınç ve akış gibi operasyonel değerler elde edilir. Bilinen kütle ve enerji denklikleri
kullanılarak enerji hesaplamaları yapılır. Ekserji analizi için ise bir referans ortamı
seçilir ve bu referansa göre ekserji hesaplamaları yapılır. Ekserji tüketimlerini
görebilmek için ekserji denkliği kullanılır. Sistemin ekserji verimliliği için de uygun
eşitlikler kullanılır. En son hesaplanan bu değerler yorumlanır. Sistemdeki
verimsizlikler belirlenir ve ihtiyaca bağlı olarak performansı iyileştirebilecek
aksiyonlar alınır.
Petrol rafinerilerinde termodinamik olarak incelenebilecek birçok ekipman vardır. Bu
ekipmanlar ısı değiştiriciler, soğutucular, havalı soğutucular, fırınlar, distilasyon
kolonları, pompalar ve kompresörler olabilir. Petrol rafinerilerini modellemek için
simülasyon önemli bir araçtır ve bunun için çeşitli yazılım ürünleri geliştirilmiştir.
xxviii
Bunlardan biri yaygın olarak kullanılan ASPEN HYSYS'dir. Kullanıcılar, HYSYS'nin
sağladığı interaktif platform sayesinde modellerini etkili ve hızlı bir şekilde kurar ve
çalıştırır. Ek olarak, ASPEN HYSYS, rafinaj, petrol veya gaz endüstrileri için faz
davranışını, fiziksel ve taşıma özelliklerini doğru bir şekilde hesaplamak için kapsamlı
termodinamik altyapı sağlar. HYSYS kullanıcılara, ayırıcılar, reaktörler, döner
ekipmanlar, ısı transferi ve mantıksal işlemler gibi birçok temel işlem seçenekleri
sunmaktadır. Aynı zamanda ekserji ve enerji analizi hesaplamalarını yapmak ve farklı
operasyonel koşullarda çeşitli durum çalışmaları gerçekleştirmek için de iyi bir ortam
sunar.
Bu çalışmada TÜPRAŞ Rafinerisi’nde bulunan vakumlu damıtma ünitesinin enerji ve
ekserji analizi yapılmıştır. Çalışmanın esas amacı enerji analizinin yanı sıra farklı bir
kavram olan ekserji analizi çalışmaları ile vakum distilasyon ünitesinde yer alan tüm
birimlerin verimsizliklerini keşfetmektir. Ayrıca, enerji verimsizlikleri görülen
operasyonlarda enerji kayıplarının neden olabileceğini yorumlamak ve iyileştirme
potansiyeli olabilecek alanları belirlemektir. Çalışmanın diğer amacı ise, vakum
distilasyon kolonunun verimini arttırabilecek şekilde proses kontrol limitlerine bağlı
olarak operasyonel parametrelerin ekserji verimliliği üzerinde etkisini görmektir. Son
olarak, makine öğrenmesi modelleri kullanılarak vakum distilasyon kolonunun ekserji
verimliliği tahmini üzerinde çalışılmıştır. Bunun için kolonu etkileyen parametreler
model girdisi olarak kullanılmıştır. Tahmin modelleri geliştirmenin amacı ise, arka
planda sürekli ASPEN HYSYS çalışma ortamına ve yoğun ekserji hesaplamalarına
gerek kalmadan canlı platformda sonuçları görebilmek ve değerlendirmektir. Aynı
zamanda kolon üzerinde geliştirilen tahmin modelleri ile herhangi bir operasyonel
koşul için bu operasyonun verimli olup olmadığını belirlenebilir.
Çalışma ilk olarak ekserji konseptini anlamak adına ve ekserji hesaplamalarında
kullanılacak eşitliklere ulaşmak için detaylı bir literatür araştırmasıyla başlamaktadır.
Daha sonra enerji ve ekserji çalışmalarını gerçekleştirmek için ASPEN HYSYS’te
vakum distilasyon ünitesi simülasyonu yapılmıştır. Simülasyon için tüm verilerin
ulaşılabildiği bir baz tarihi alınmıştır. Simülasyon için gerçek ünite verileri
kullanılmıştır. Bu veriler laboratuvar değerleri ve sıcaklık, basınç ve akış gibi
operasyonel değerlerdir. Ayrıca ekipmanlara ait ekipman teknik raporunda yazan
bilgiler de ünite simülasyonunda kullanılmıştır. Ünitede yer alan ısı değiştirici, ısıtıcı,
soğutucu, havalı soğutucu, pompa, fırın ve kolon gibi tüm birim operasyonların
simülasyonu HYSYS’te yapılmıştır. Gerçek değerler ile HYSYS model sonuçlarının
karşılaştırılması sıcaklık, basınç ve akış gibi her operasyon değer için yapılmıştır.
Gerçek değerler ile model sonuçları yakınsayana kadar model sürekli güncellenmiştir.
En son doğru model simülasyonu elde edildikten sonra ekserji ve enerji analizi
hesaplamalarına başlanmıştır. Ekserji hesaplamalarında ekserji aktarımı eşitliklerine
dikkat edilmiştir. Kütlesel ya da akış ekserji hesaplamaları için simülasyon akımlarına
ait entalpi, entropi, ekserji değerleri ve literatürde yer alan eşitlikler kullanılarak
hesaplamalar yapılmıştır. Aynı zamanda her ekipman için literatürde bulunan ekserji
verimliliği hesabı yapılmıştır. Ekserji verimliliği düşük olan ekipmanlar belirlenmiştir.
Bu ekipmanların performans düşüşlerinin nereden kaynaklanabileceğine ait
değerlendirmeler yapılmıştır. Bununla birlikte ekipmanlara ait ekserji ve enerji
verimlilik değerlerinin karşılaştırılması yapılmıştır. Ayrıca, proses kontrol limitlerine
göre kolonu kontrol eden parametrelerin ekserji verimliliği üzerine etkisi ASPEN
HYSYS’in durum çalışması fonksiyonu kullanılarak değerlendirilmiştir. Kolonun
ekserji verimini etkileyen en önemli parametreler belirlenmiştir.
xxix
Ekserji verimliliğini etkileyen parametreler değerlendirildikten sonra bu parametreler
kullanılarak ASPEN HYSYS’te sentetik veri üretilmiştir. Makine öğrenmesi
yöntemleri ile bu veri seti kullanılarak tahmin modelleri geliştirilmiştir. Bunun için
yapay sinir ağı modellerinden biri olan Torbalama ve topluluk öğrenmesi
metotlarından biri olan Rassal Orman modelleri R programında geliştirilmiştir. R
programının "Ipred" ve "randomForest" paketleri, sırasıyla Torbalama ve Rassal
Orman modeli için kullanılmıştır. Her iki modelin verileri eğitim, test ve görünmeyen
doğrulama veri setlerine ayrılmıştır. Bu bölünme eğitim seti için %50, test seti için
%30 ve görünmeyen doğrulama veri seti için %20 olacak şekildedir. En son gerçek
değerler ile model tahmini değerler arasındaki uyuma bakılarak modelin doğruluğu ve
başarısı yorumlanmıştır.
xxx
1. INTRODUCTION
The industries cause 36% global energy consumption and 35% global greenhouse gas
emission because they are a big energy consumer. The oil and gas industry is both
major manufacturer and consumer of energy. According to IEA report, oil refineries
are responsible for 9% of energy consumption in Figure 1.1. The refinery operations
that are resource extraction, processing, transformation and transportation extending
to delivery to end users require a large energy consumption [2, 3]. Therefore,
evaluating petroleum industry in energy analysis studies is critical.
1
When energy is used as measuring tool to comprehend and develop energy efficiency,
misdirection and confusion might happen in some cases. The quantity exergy provides
more understandable and useful information in terms of evaluating the waste in energy
systems compared to the information provided by energy [4].
Exergy term is related with both the First Law of Thermodynamics (FLT) and the
Second Law of Thermodynamics (SLT). The FLT is known as the energy conservation
law that means even though energy can change state; it might be neither created nor
exterminated. Exergy analysis technique, which depends on the SLT, cope with the
FLT limitations. The Second Law of Thermodynamics identifies the principal quantity
entropy, which represents the unavailability of randomized energy form to convert
work directly. Also, it states that all physical and chemical spontaneous processes
increase the entropy [5].
The exergy analysis can be used in different fields that can be shown in Figure 1.2.
In addition to this, there are many kinds of exergy process applications on energy
systems like industrial, utility, transportation as well as residential-commercial. The
usage of exergy analysis can mainly be applied to industrial energy processes. Due to
the plenty of different energy usage, determination of overall efficiency is complicated
for the industrial sector that can be grouped as petrochemical, chemical process
(chemical production, water desalinization, distillation, etc.), metallurgical process
like lead smelting and the systems of heating and cooling. Since exergy method is a
powerful tool in terms of analyzing, evaluating, designing, developing and optimizing
process and system, it is used in various industries. Process steam, mechanical driving
2
force, direct and space heating are some of the energy consumptions in the industrial
sector. In the utility sector, not only conventional energy resources including fossil and
nuclear, but also renewable resources including geothermal, biomass, wind, solar and
water are used for processes of electricity production. In addition, residential, public,
commercial and agribusiness activities are involved the sector of residential-
commercial that uses different energy carriers [7, 8].
There are many different field of applications about usage of exergy methods in the
literature. For instance, Dong et al. presented that the exergy analysis method provides
theoretical information about improving engine performance, reducing fuel usage and
optimizing engine combustion for aircraft systems, turbojet and scramjet engines [9].
Fellaou et al. implemented the exergy analysis of desalination facility. They found
largest degradation in the module of membrane, the high-pressure pump and delivery
pump that were 64.28%, 40.84% as well as 38.48% respectively. [10]. In another
research, energy and exergy calculations were conducted to the process of biomass
gasification of pinewood, horse manure and sawdust for hydrogen production using
the gasifying agents. In comparison with results, higher hydrogen exergy efficiency is
obtained with steam gasification for pinewood. When the exergy efficiency of product
gas reached the top value (88.26%), the exergy efficiency of hydrogen was obtained
as 44% [11]. Bai et al. compared the energy and exergy analysis between the cycle
systems of developed supercritical CO2 and existed supercritical CO2 for coal power
station. Whole exergy efficience of the developed supercritical CO2 cycle was found
2.32% more than the existed supercritical CO2 cycle [12]. In other research, an exergy
analysis of marine gas turbine system was conducted for per system element and entire
system. It was concluded that among the all heat exchangers, while combustion gases
heat exchangers had the maximum exergy efficiencies, cooler had the minimum
exergy efficiency. It was also discovered that the turbines have higher exergy
efficiency than compressors. The exergy efficiency of the entire process was 64.12%
[13]. Bühler et al. carried out energy and exergy methods to investigate a production
facility of milk powder. The spray dryer and the gas burner had the maximum exergy
destruction and low exergy efficiencies were appeared in the heaters [14].
Considering the petroleum sector that are responsible for high energy consumption,
the crude oil distillation units have many constituents which are the furnace, the
distillation towers and heat exchanger networks with respect to applying energy and
3
exergy methods thermodynamically. Therefore, many researchers focused on different
components for the exergy analysis studies in the refinery.
One of these studies is the exergy analysis in a crude oil fractionating column. Al-
Muslim et al. carried out exergy analysis of petroleum refining units with one and two
stage, which comprised of furnace and fractionating column for identifying exergy
degradations and efficiencies. The effect of reference condition on exergy
performances of petroleum refining units with one and two-stage was examined in
another study. This study pointed out that the increment of exergy losses occurred
while the reference temperature increased and thus the exergy efficiency decreased
[15, 16]. Tarighaleslami et al. examined the atmospheric refining column in Tabriz
Refinery for a case study to see the feasibility of different retrofit choices in the way
of thermodynamic exergy loss studies. The best retrofit option was found reduction in
exergy loss of 17.16%, resulting in 3.6% reduction in fuel demand [17]. Gu et al.
employed exergy analysis on three crude oil vacuum distillation columns to assist
designers in process choice. This study provided an output to see appropriate furnace
exit temperature and process depending on the expected purposes of the unit [18]. In
another research, multi-stage impact on crude distillation with using energy and exergy
thermodynamically was investigated. It resulted with diminution in the exergy and
energy of heat demand of crude distillation unit occurred when the number of stage
increased [19]. Dincer and Rosen researched operation parameters effects including
temperature and pressure of the crude oil refining column on system efficiencies using
exergy method. As a result, the changes in temperature and pressure profile did not
give rise to important change on the efficiencies of energy and exergy especially for
vacuum distillation unit [20].
Pre-distillation unit containing various heat exchanger networks and furnaces are also
used in exergy analysis studies. For instance, Vilarinho et al. executed both energy and
exergy studies of the pre-distillation unit in Portuguese Refinery. Overall efficiency
results were obtained as 13.4% energy efficiency and 2.3% exergy efficiency in pre-
distillation unit. Also, both energy and exergy losses of furnace were found 15.6% and
56.3% in the unit [21]. Izyan and Shuhaimi applied exergy analysis for performance
assessment of the crude preheating network and furnace inside the petroleum
distillation unit. The presented options for decreasing exergy loss in this study were
cleaning schedule of preheat network and heat loss reduction from furnace [22].
4
Computer aided studies such as process simulation, machine learning models are
important methods for both energy and exergy analysis in terms of evaluating process
analysis, design and optimization. Osuolale and Jie Zhang studied with
thermodynamic tools using ASPEN simulation program and artificial neural network
model in order to help both engineers and designers in terms of designing and
operating energy efficient crude oil distillation units [23]. Osuolale and Jie Zhang also
applied exergy analysis on distillation column for binary systems by means of using
artificial neural network in another research. 32.4% increment of exergy efficiency
was occurred with changing the operating conditions without causing additional
investment costs [24].
In this research, exergy analysis of vacuum distillation unit (VDU) that composed of
heat exchanger network, furnace and distillation column in TUPRAS Refinery was
investigated. Computer aided simulation of VDU was done in order to examine exergy
analysis in ASPEN HYSYS v10 with using equipment, laboratory and operational
data. Both energy and exergy calculations were accomplished for all unit operations
such as pumps, furnace, heat exchangers, air coolers, and vacuum distillation column.
The exergy efficiency in the column has been studied specially. The impact of various
operation parameters on column exergy efficiency analyzed by using “Case Study” of
ASPEN HYSYS. In addition, machine learning (ML) models for column exergy
efficiency were improved using process operation data from simulation. Due to
intensive exergy calculation, the success of ML model predictions were examined by
using two models to satisfy online working environment. In addition to presentation
of exergy analysis studies in the vacuum distillation column, examination the
operational variables affecting the column exergy efficiency is among the aims of this
study. Furthermore, exergy analysis is supported neural network models to enhance
model accuracy and reliability.
5
6
2. DESCRIPTION AND SIMULATION OF PROCESS
The first step of refinery is crude oil distillation that separates crude oil into various
petroleum products known as fractions according to the boiling temperature
difference. Distillation plant comprises of two petroleum distillation towers, which are
the atmospheric and vacuum tower. While the light hydrocarbons are distinguished in
atmospheric tower, the heavy hydrocarbons are distinguished in vacuum tower.
Although, vacuum tower has the same operating basis with atmospheric tower, it has
fewer stages and side cuts [20].
In this study, vacuum distillation unit in TUPRAS Refinery was evaluated as a case
study. The VDU consists of four mainly unit parts such as feed preheat section,
vacuum distillation unit furnace, vacuum distillation unit column with pump-around
(PA) circuits and amine units. The basic representation of the VDU is in Figure 2.1.
The aim of the VDU process is to separate the atmospheric residue by three main
streams consisting of light vacuum gasoil (LVGO), mixture of medium and heavy
vacuum gas oil (MHVGO) as well as vacuum residue (VR). Sour water, waste gas,
slops oil and Heavy Slop Oil (HSO) are also obtained in this process.
The heat exchanger network (HEN) performs two targets that are chilling of product
streams and preheating of the atmospheric crude oil. Shell and tube type heat
exchangers take part in the HEN.
The HEN comprises five unit heat exchangers. Initially, atmospheric residue, which is
cold feed from tank, is sent to VDU for preheating against product stream in the first
heat exchanger. Also, hot atmospheric residue from the available crude distillation
units is sent to stream of first heat exchanger exit. These two atmospheric residues mix
before feeding into the second heat exchanger. Then, this mixed atmospheric residue
is preheated against PA stream in the second heat exchanger. Exit stream of second
7
heat exchanger splits two branches. A branch is routed to third heat exchanger to
preheat against PA stream and other branch is routed to fourth heat exchanger to
preheat against return stream. Two branches that are exit of third and fourth heat
exchangers are mixed with each other. Finally, mixed stream is routed to fifth
exchanger to preheat against VR.
Atmospheric crude oil residues from the exit of fifth heat exchanger is heated in the
VDU furnace using fuel gas. The atmospheric crude oil residues start vaporizing with
combustion. Then, the partially vaporized atmospheric residue mixture is transferred
into the vacuum distillation column.
The vacuum distillation unit is operated with vacuum pressure to provide separating
heavy hydrocarbons and enable to less temperature for distillation of crude oil. Series
of ejectors is placed to create vacuum pressure. The scheme representing the vacuum
distillation column is shown in the Figure 2.1.
The PA circuits remove the heat from the VDU. Liquid is drawn from a specific tray,
is cooled in the heat exchangers and is returned at a lower temperature to the tray above
the tray, which is drawn liquid, by the PA circuits. They are described with the drawn
and the return stage, the return temperature as well as the mass rate of flow.
8
2.1.4 Amine units
The VDU overhead product that is called off gas stream can contain high concentration
of H2S. It is treated in amine units before sent to the VDU furnace for burning. The
amine units have three sections such as amine absorber, amine regenerator and amine
filtering. Amine absorber removes the H2S of vacuum off gas. Amine regenerator
regenerate the amine by removing the H2S. Amine filters remove corrosive products
and other contaminants from the amine.
Various software products that simulate and optimize the refining process were
developed for the petroleum industries. One of them is ASPEN HYSYS, which is
commonly used. Users set up and work their models effectually and quickly by
provided interactive platform of HYSYS. In addition, ASPEN HYSYS enables
extensive thermodynamic substructure to calculate phase behavior, physical and
transport properties for the refining, oil and gas industries accurately. There are many
9
unit operations such as separators, reactors, rotating equipment, heat transfer and
logical operations in HYSYS. Not only the steady state modeling but also dynamic
modeling are done in this environment [25].
a. Unit operations are specified with the input and output material streams, energy
streams as well as required operational and design conditions.
b. The feed flowrates are specified with flowrates, compositions and operating
conditions such as temperature, pressure and vapor fraction.
5. Running the simulation of VDU to produce data for exergy analysis occurs in the
simulation environment.
10
Figure 2.2: ASPEN HYSYS workflow diagram for petroleum distillation process.
When ASPEN HYSYS V10 opens, new button is chosen. The initial step of the
simulation is the creating of component list. Figure 2.3 shows the component list tabs.
If the add button is clicked in the component list tabs, the database list which has pure
and hypothetical components will appear in Figure 2.4.
11
Figure 2.3: ASPEN HYSYS component list tabs.
Figure 2.4: Source databank in the component list tabs of ASPEN HYSYS.
A large amount of unknown hydrocarbons chemicals and compounds form the content
of crude oil. Because of the inadequate information about the crude oil content, it is
frequently qualified in relation to the boiling point of hypothetic constituents [27].
After describing the component list, fluid packages are created with clicking add
button from fluid packages tabs. Appropriate thermodynamic model is selected
between many options and it is ensured that thermodynamic model is related with
Component List - 1. The “decision or selection tree” shown in Figure 2.5 can be used
12
for the choice of the most appropriate thermodynamic package. Some physical
properties of compounds included in the process to decide the thermodynamic model.
These properties are polarity, electrolyte, pressure and real or pseudo constituents. The
Peng Robinson (PR), which is one of the equation of state (EOS), is mostly preferred
among the property packages for petroleum, oil and gas process because of containing
nonpolar or little polar and real components. In addition, the PR Fluid Package is
improved in terms of comprising wide range of pressure and temperature, great
database of binary interaction as well as being particular behaviour for key ingredients
[28].
Figure 2.5: Selection of proper thermodynamic model considering decision tree [28].
13
Also, appropriate fluid package selection can be made with the help of methods
assistant in the property environment of ASPEN HYSYS shown in Figure 2.6.
According to component or process type in the method assistant, this selection is
decided.
Peng-Robinson, Antoine and ASME Steam model were used as thermodynamic fluid
packages. Figure 2.7 shows Peng-Robinson model selection from fluid packages tabs.
Peng-Robinson model is appropriate for oil and gas processing, Antoine model is
recommended for petroleum assay at vacuum conditions and ASME steam model is
used for defining the H2O component in ASPEN HYSYS.
Petroleum has all type of hydrocarbons mixture such as aliphatic, aromatic etc. with
salts, water, sulfur and nitrogen including compositions and metal complexes. Hence,
alternative methods are available for characterizing petroleum like density and
distillation curves instead of using composition directly. Crude oil is characterized by
utilizing American Society for Testing and Materials (ASTM) and True Boiling Point
(TBP) methods. TBP curves estimate about the properties and composition of the
14
crude oil and is plotted between the boiling point of all component of the mix and
cumulative volume percent of distilled. Due to its features of time consuming and
costly, another quicker method was proposed such as ASTM. One of ASTM methods
is ASTM D86 that is used for nearly entire petroleum fractions except for heavy
distillations [29].
In this study, a summary day that holds exact laboratory data was selected to simulate
the unit. Distillation values for this summary day were entered into the simulation.
Atmospheric residue petroleum assay was created by back blending of the VDU
products which consist of LVGO, MVGO, HVGO and VR (Figure 2.8).
Average tagged unit operation values of selected summary day such as flow,
temperature and pressure values were used in order to create process flow diagram of
VDU. Moreover, equipment design values of each equipment belonging to the unit
were entered into the simulation with using equipment data sheet. After the
improvement of PFD, the simulation was run. The comparison was made between the
result of simulation and operation conditions for summary day to check the accuracy
of the simulation. Finally, the simulation was used for exergy analysis.
15
process simulation consists of heat exchanger network for preheat section, VDU
furnace and vacuum distillation column.
Material streams travelling input and output of simulation boundaries and proceeding
among unit operations can be simulated in ASPEN HYSYS. The properties and
composition of material streams must be specified in order to be solved by HYSYS.
While describing material streams, degrees of freedom is used to operate flash
calculation by HYSYS. Therefore, composition and two process specifications
required for flash calculation. These two specifications can be temperature, pressure,
vapor fraction, enthalpy, and entropy, but between two process specifications, one of
them must be either temperature or pressure. Also, one of the flow options is entered
to define a material stream completely. In addition to this, energy streams travelling
input and output of simulation boundaries and proceeding among unit operations can
be simulated in ASPEN HYSYS. Figure 2.9 displays both the view of material and
energy stream property [30].
(a) (b)
Figure 2.9: A screen of (a) Energy (b) Material stream in ASPEN HYSYS.
There are many unit operations used to build flowsheets. These unit operations split
up categories that contain many individual operations. For example, unit operation
blocks can be classified as heat transfer equipment, piping equipment, rotating
equipment, separators, reactors, logicals and user-defined blocks. While modeling
operations, a degrees of freedom (DOF) approach is used by HYSYS [30].
16
Heat transfer operations
ASPEN HYSYS has many unit operations that are indicated in Table 2.1 for heat
transfer modelling. Heater and Cooler are one side of an exchanger that can be used
for heating or cooling, condensing or evaporating and changing pressure. Duty and
pressure drop specifications are necessary in order to solve fully for the Heater and
Cooler in ASPEN HYSYS. Heat transfers between two flows are modelled in the heat
exchangers calculating two-sided material and energy balance. Heat exchanger model
can be selected depending on user’s analysis in Figure 2.10. Two specifications are
essential to be solved by heat exchanger but detailed heat exchangers such as rating
and rigorous models also need exchanger geometry information besides two
specifications. One of these specifications is shell and tube side pressure drops. Other
specification can be one of these specifications: product stream temperature, minimum
temperature approach, overall UA, LMTD or temperature change. Air Cooler uses air
for heating or cooling a process flow by some specified conditions that are pressure
drop and overall UA. Also, air or process exit stream can be specified instead of overall
UA [30].
17
Figure 2.10: Heat exchanger model selection.
Piping operations
18
Table 2.2: Piping operations.
Name Description
Blends multi inlet flows to generate one outlet
Mixer
flow
Split one inlet feed flow into multi outlet flows
Tee
for the same conditions and compositions
19
Figure 2.12: Flow ratios specification of the Tee.
Rotating operations
Different type of rotating operations are defined in Table 2.3. The Pump and
Centrifugal Compressor calculate either an unknown temperature, pressure or
efficiency based on the information specified by user. If the inlet stream is fully
described, just two of the variables should be specified for both the Pump and
Centrifugal Compressor: outlet pressure or pressure drop, energy and efficiency.
Unlike the Centrifugal Compressor, the Expander creates outlet stream with high
velocity and low pressure. The gas internal energy is converted into kinetic energy and
eventually converted into shaft work by an expansion process. To conclude, while
compression process needs energy, expansion process releases energy [30].
20
Separation and column operations
ASPEN HYSYS offers many types of separation and column operation in Table 2.4.
The views of property for the Separator, 3 Phase Separator, and Tank are similar. The
essential differences in the three separator operations are the stream connections that
are described in Table 2.4. The first step in setting up the Column is to decide which
type wanted depending on the equipment of Column like reboiler or condenser.
Column build environment of HYSYS is specified with column configurations
including side strippers and pump arounds sections [30].
21
Logical operations
22
3. STUDY OF THE ENERGY AND EXERGY ANALYSIS
The analysis of energy is a thermodynamic method based on FLT that is linked to the
conservation of energy principle. However, the method of exergy is connected with
the both FLT and SLT. While conservation of energy rule known as the FLT,
nonconservation of entropy rule known as the SLT. The method of exergy can be used
to specify and quantify exergy destruction of process owing to the exergy losses and
irreversibility. Therefore, it is alternating and informative tool in terms of evaluating
and comparing systems and processes. In comparison to energy analysis, exergy
analysis presents more meaningful efficiencies and identifies the thermodynamic
losses more clearly. The detailed comparison for energy and exergy is given in Table
3.1. Also, the same general equations can be written for both of them in any control
volume [1, 4, 31].
23
3.1 The Exergy Concept
Process state: The beginning state of the system under a work is indicated by the
process state (T, P).
Environmental state: The balance of mechanics and heat between the surroundings and
the system is defined as the state of restricted equilibrium. System and ambient
conditions such as pressures and temperatures are equal in the restricted equilibrium.
Environmental state is described as a condition that satisfies the restricted equilibrium
with the surroundings (T0, P0).
Dead State: Chemical potentials of the matters are required within the unrestricted
equilibrium, including temperature and pressure of the system and surroundings to
equal for providing thermodynamic balance totally between the surroundings and the
system. Under these circumstances, the system exergy amount is equal to zero since
the system cannot be exposed to any state changes during interaction with
surroundings. This system state names as the dead state.
24
While analyzing the thermodynamic of the system, three valid equations are used
generally. These equations are the conservation of both energy and mass as well as the
nonconservation of entropy equation. Some assumptions for deriving these equations
are made as follows [15, 16, 20, 34]:
These simplifications are applied for three equations. They are the mass, energy, as
well as exergy balance equations shown in equations 3.1, 3.2 and 3.3 respectively:
𝑇0 (3.3)
∑ 𝐸𝑥̇𝑖 + ∑(1 − )𝑄̇ = ∑ 𝐸𝑥̇𝑒 + 𝑊̇𝑐𝑣 + 𝐼𝑐𝑣
̇
𝑇𝑗 𝑐𝑣
𝑖 𝑗 𝑒
For the streams in equations 3.1, 3.2 and 3.3, ṁ indicates the rate of mass flow, Ė
indicates the energy rate and Ėx indicates the exergy rate. Q̇ cv is the rate of heat into a
control volume, Ẇcv is the work done on the part of a control volume, İcv is irreversible
exergy loss of a control volume as well as inlet and exit of the stream are stated by the
subscripts i and e in equations 3.1, 3.2 and 3.3. Also, the exergy losses from the control
volume can be expressed as equation 3.4:
̇ = 𝑊̇𝑐𝑣𝑟𝑒𝑣 − 𝑊̇𝑐𝑣
𝐼𝑐𝑣 (3.4)
A closed system of mass exergy is stated as nonflow exergy. Equation 3.5 shows the
nonflow exergy that is consisted of physical, chemical, kinetic, and potential exergy
configurations [35].
25
Where
1
𝐸𝑥𝑘𝑖𝑛 = (𝑣 2 − 𝑣02 ) (3.7)
2
𝐸𝑥𝑛𝑜𝑛𝑓𝑙𝑜𝑤,𝑝ℎ = (𝑈 − 𝑈0 ) + 𝑃0 (𝑉 − 𝑉0 ) − 𝑇0 (𝑆 − 𝑆0 ) (3.9)
The maximum acquired work from the system is known as physical nonflow exergy
while it comes to the ambient state. For instance, it is a condition of thermal and
mechanical balance with the surroundings [35].
A maximum acquired work from the system is known as chemical nonflow exergy
while it comes from an ambient condition to a dead condition. To illustrate, it is a
condition of overall balance with the environment [35]. Equation 3.10 calculates a
mixture chemical exergy [36-38].
𝑒𝑥𝑐ℎ 𝑐ℎ
𝑚𝑖𝑥 = ∑ 𝑥𝑖 𝑒𝑥𝑖 + 𝑅𝑇0 ∑ 𝑥𝑖 𝑙𝑛(𝛾𝑖 𝑥𝑖 ) (3.10)
Where xi , exich , γi and R indicate the mole fraction, chemical exergy, the ith component
activity coefficient as well as molar gas constant, respectively.
This term can not be used directly for the mixture of petroleum components. Standard
chemical exergy for the light petroleum components in reference temperature and
pressure can take part in literature. Therefore, they can be substitute for chemical
exergy exich . Since the mixture of heavy petroleum fraction can not be known exactly,
26
it is separated into pseudo-components that need to be calculated depending on the
lower heating value with equation 3.11 [38-40].
βi is described as the correction factor of the i-th component chemical exergy as well
as equation 3.12 shows the β that is related to mass fractions of C, H2, O2, S, and N2
[36, 37, 39, 40].
𝑧𝐻2 𝑧𝑂 𝑧𝑆 𝑧𝐻 𝑧𝑁
𝛽 = 1.0401 + 0.1728 + 0.0432 2 + 0.2169 (1 − 2.0628 2 ) + 0.0428 2 (3.12)
𝑧𝐶 𝑧𝐶 𝑧𝐶 𝑧𝐶 𝑧𝐶
Where zC , zH2 , zO2 , zS , and zN2 correspond to mass fractions of C, H2, O2, S, and N2
respectively. LHV is also calculated by equation 3.13 [41].
𝑀𝐽
𝐿𝐻𝑉 [ ] = 55.5 − 14.4. 𝑆𝐺 (3.13)
𝑘𝑔
Exergy might be transferred to system through mass, heat and work, as well as exergy
might be transmitted from a system.
Summation of a non-flow exergy and a stream work exergy is equal to the exergy of a
material flow as following equation 3.14 [35]:
If equation 3.5 is placed in equation 3.14 without chemical, kinetic, and potential
exergy components, equation 3.15 is obtained. Also, if equation 3.15 is rearranged,
equation 3.16 is reached. Equation 3.16 corresponds to the equation 3.17 that is known
as the physical exergy flow of stream. It is stated as the enthalpy and entropy
differences from the conditions of stream to reference.
𝐸𝑥𝑓𝑙𝑜𝑤 = (𝑈 − 𝑈0 ) + 𝑃0 (𝑉 − 𝑉0 ) − 𝑇0 (𝑆 − 𝑆0 ) + (𝑃 − 𝑃0 )𝑉 (3.15)
27
𝐸𝑥𝑓𝑙𝑜𝑤,𝑝ℎ = (𝑈 + 𝑃𝑉) − (𝑈0 + 𝑃0 𝑉0 ) − 𝑇0 (𝑆 − 𝑆0 ) (3.16)
𝐸𝑥𝑓𝑙𝑜𝑤,𝑝ℎ = (𝐻 − 𝐻0 ) − 𝑇0 (𝑆 − 𝑆0 ) (3.17)
Where H and H0 represent the enthalpy for stream and reference conditions, as well as
S and S0 represent the entropy for stream and reference conditions, respectively in
equations 3.15, 3.16 and 3.17.
Exergy is transferred by heat when a control volume, which is initially dead state, is
heated or cooled in interactions with another system. If the exergy flow correlates with
heat transfer Q, ExQ in equation 3.18 expresses this exergy flow [35].
𝑓
𝑇0
𝐸𝑥𝑄 = ∫ (1 − ) 𝜕𝑄 (3.18)
𝑖 𝑇
The interval of integral of equation 3.18 shows the situation from initial (i) to the final
situation (f), as well as ∂Q and T0 are equal to incremental heat transfer and
environmental temperature, respectively. When temperature is constant, then equation
3.18 becomes as equation 3.19.
𝑇0
𝐸𝑥𝑄 = (1 − )𝑄 (3.19)
𝑇
Exergy is transferred by work when a system does the work because of the volume
change in time between t1 and t2. This is expressed by (WNET)1,2 as equation 3.20 [35].
Where W1,2 states the system work because of volume changing (V2 – V1) and
translocation work is stated by P0(V2 – V1). Work exergy stream equates to work
related to specified energy form. Therefore, shaft work consisting of both mechanical
and electrical work is equal to exergy.
28
3.4 Energy and Exergy Efficiency
Engineers measure the performances of tools and processes with using efficiencies.
These numerous expressions are depended on energy that is related to FLT. Moreover,
SLT based efficiencies are useful in terms of considering limitations that is stemming
from performance measures. When energy input is equal to energy output, there is a
maximum efficiency in the process. Nevertheless, energy based efficiency can
misdirect because of its inability to criterion of approach to ideality. The SLT takes
into account the criterion of approach to ideality. Maximum efficiency is stated as a
reversible process that imply achieving an ideal for this law. Furthermore, considering
the energy efficiencies, they can be bigger than a hundred percent like coefficient of
performance. However, exergy efficiency is always between zero and a hundred
percent [5, 42].
The expression of energy output divided by energy input indicates overall energy
efficiency that can also be expressed in relation to energy loss in equation 3.21.
(1) The result obtained with dividing the total output flow exergy by the input flow
exergy is called as the total, overall, universal or input-output exergy efficiency.
(2) The output obtained with dividing exergy statements coming from the generated
products inside the system to exergy statements coming from expended resources is
29
called as the task, rational, utilitarian functional or consumed-produced exergy
efficiency [44, 45].
The inlet-outlet efficiency exergy shown with equation 3.22 is described as dividing
the total exergy output by the sum of exergy input or in another saying; it equals the
ratio of net amount of generated exergy to net amount of provided exergy. Moreover,
the exergy efficiency might be determined regarding the exergy losses [18, 45, 46].
There are five different exergy efficiency definition in consumed-produced class [47].
However, many exergy efficiency balances as well as exergy-based indicators can be
presented at literature [43, 45, 48].
Grassmann [49] suggested a common expression for exergy yield that is the ratio of
desirable exergy production to reducing of utilized exergy as equation 3.23.
Kotas [50] used equation 3.24, which is called rational efficiency to specify the exergy
performance, which equals the ratio of desired outputs to required inputs.
𝐷𝑒𝑠𝑖𝑟𝑒𝑑 𝑂𝑢𝑡𝑝𝑢𝑡
𝜓2 = (3.24)
𝑁𝑒𝑐𝑒𝑠𝑠𝑎𝑟𝑦 𝐼𝑛𝑝𝑢𝑡
Szargut et al. [51] expressed that exergy performance equaled to the ratio of useful
products exergy to feeding exergy in equation 3.25.
30
The exergy part supplied to the system as well as passing through the system, which
does not undergo any physical and chemical change, is defined as transit exergy.
Tsatsaronis et al. [53] introduced the exergetic efficiency that is the ratio of the desired
exergetic outlet to driving exergetic inlet. In other words, it is the ratio of products
exergy to fuel exergy as indicated equation 3.27.
𝐸𝑥𝑒𝑟𝑔𝑦 𝑜𝑓 𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑠
𝜓5 = (3.27)
𝐸𝑥𝑒𝑟𝑔𝑦 𝑜𝑓 𝐹𝑢𝑒𝑙
𝐼 = 𝑇0 𝑆𝑔𝑒𝑛 ≥ 0 (3.30)
Where I and Sgen indicates the exergy consumption and the entropy generation in a
closed system respectively.
31
Figure 3.1: Balance of energy and exergy [1].
Exergy losses evaluate the deviations from ideality quantitatively. The location, sort,
and reason of a waste or inefficiency are also determined by exergy losses. This
information is important in terms of providing the way to increase exergy efficiency.
Exergy losses connects two types of situation such as external and internal losses.
These are related with residuals of exergy effluents and inner irreversibilities arising
from exergy consumptions in a process or system [42]. These external losses can be
contain as heat transfer with the environment (coolers utilizing air, water), unused
energy component of material flows (exhaust and purge gases) and diffusion of heat
losses from pipelines and operation equipment. However, spontaneous processes like
combustion, limited driving and dissipative forces give rise to irreversibilities known
as internal losses [47].
32
Table 3.2: The equations of exergy lost and exergy efficiency representing some unit
operations [20, 32, 54, 55].
Unit Operation Exergy Lost Exergy Efficiency
Pump 𝐸𝑥𝑖𝑛 − 𝐸𝑥𝑜𝑢𝑡 + 𝑊𝑝𝑢𝑚𝑝 𝐸𝑥𝑖𝑛 − 𝐸𝑥𝑜𝑢𝑡
𝑊𝑝𝑢𝑚𝑝
Two-streams (𝐸𝑥ℎ𝑜𝑡,𝑖𝑛 − 𝐸𝑥ℎ𝑜𝑡,𝑜𝑢𝑡 ) − (𝐸𝑥𝑐𝑜𝑙𝑑,𝑜𝑢𝑡 − 𝐸𝑥𝑐𝑜𝑙𝑑,𝑖𝑛 ) 𝐸𝑥𝑐𝑜𝑙𝑑,𝑜𝑢𝑡 − 𝐸𝑥𝑐𝑜𝑙𝑑,𝑖𝑛
heat exchanger 𝐸𝑥ℎ𝑜𝑡,𝑖𝑛 − 𝐸𝑥ℎ𝑜𝑡,𝑜𝑢𝑡
Heater 𝐸𝑥𝑖𝑛 − 𝐸𝑥𝑜𝑢𝑡 + 𝑄 𝐸𝑥𝑜𝑢𝑡 − 𝐸𝑥𝑖𝑛
𝑄
Air cooler 𝑊𝑎𝑖𝑟𝑐𝑜𝑜𝑙𝑒𝑟
∑ 𝐸𝑥𝑖𝑛 − ∑ 𝐸𝑥𝑜𝑢𝑡 + 𝑊𝑎𝑖𝑟𝑐𝑜𝑜𝑙𝑒𝑟
∑ 𝐸𝑥𝑖𝑛 − 𝑊𝑎𝑖𝑟𝑐𝑜𝑜𝑙𝑒𝑟
Column 𝑄
∑ 𝐸𝑥𝑖𝑛 − ∑ 𝐸𝑥𝑜𝑢𝑡 + 𝐸𝑥𝑢𝑡𝑖𝑙𝑖𝑡𝑦
𝑄 ∑ 𝐸𝑥𝑖𝑛 + 𝐸𝑥𝑢𝑡𝑖𝑙𝑖𝑡𝑦 − 𝐼𝑐𝑜𝑙𝑢𝑚𝑛
∑ 𝐸𝑥𝑖𝑛
33
34
4. MACHINE LEARNING MODELS
Machine learning (ML) permits computer programs to carry out complicated works
such as planning, prediction, recognition and diagnosis with learned from historic data.
ML model performance depends on data and algorithms. The size of large and high
quality data can generally increase ML model accuracy. Applying appropriate
algorithms is also important for solving various problems consisting of different types
of datasets. The types of ML are divided into four categories such as supervised
learning (SL), unsupervised learning (UL), semi supervised learning (SSL) and
reinforcement learning (RL) as shown in Figure 4.1. The task of ML can be regression,
classification, data reduction, clustering and anomaly detection as given in Figure 4.2
[56].
35
For SL, the computer program trains the algorithm between input and output using a
fully set of labeled data. Some of the SL algorithms are Regression, Decision Trees,
Logistic Regression, Random Forest, K Nearest Neighbors (KNN) etc. In contrast to
SL, UL does not use labeled data to train algorithm. Clustering algorithms such as K-
mean and Hierarchical are the examples of UL. SSL that is the mixture of SL and UL
develops a model using both data of labeled and unlabeled. RL enables agents to
examine the environment using feedback mechanism from its self-actions. Online
games such as backgammon and Atari are the examples of RL. Figure 4.3 shows the
detailed classification of ML algorithms [56, 57].
In this study, two different machine-learning algorithms were used. One of them is
artifical neural network (ANN) model and the other one is ensemble-learning method.
The basic concept of ANN comes from the way of biological neurons work. The ANN
are composed of three layer classes like input layer, hidden layer and output layer as
pictured in Figure 4.4. Input signals from an inlet layer are delivered to an outlet layer
throughout hidden layers with aid of activation functions including the vectors of
weight and bias. ANN links data of input with data of output via a series of nodes
36
known as the interconnection of neurons. These connections are consist of some
complex functions such as nonlinear functions. The weight factor of each input
determines the interconnection strength. There are two ANN structure types according
to the connections between their nodes such as the feedforward neural network
(FFNN), and the feedback neural network (FBNN) or recurrent neural network (RNN).
FFNN has one direction connection without backwards loop. Their signals proceed in
the forward direction. In contrast to FFNN, FBNN or RNN has backward loop
connection. Their signals proceed both forward and backward direction. The process
of ANN modeling includes three steps: (1) data collecting and processing, (2) decision
of model configuration, and (3) model training and confirmation [58-60].
Ensemble learning method trains several base learners as members of ensemble and
aggregate their predictions in order to produce a single output. The structure of an
ensemble learning is shown in Figure 4.5. The main target of ensemble learning is to
improve accuracy of classification by combining estimations of multiple classifiers.
Average (in the event of regression) and voting (in the event of classification) based
predictions of each classifier are performed by ensemble learning. The final output
being ensemble classifier has higher performance compared to any single classifier
[62, 63].
37
Figure 4.5: The structure of an ensemble learning [64].
Several studies indicated that ANN models are useful tools for exergy analysis.
Azadbakht et al. applied the ANN model to estimate both energy and exergy variables
of fluid bed dryer. While model inputs became temperature, speed and the beds depth,
the model outputs became loss of exergy, energy usage and the ratio of energy usage.
According to this study, a good overlap was observed between the estimated results
and the measured results. The results of statistical analysis demonstrated that ANN
could be applied for drying process having a large portion of energy utilization. Also,
it was concluded that the prediction of a trained ANN is faster compared to
mathematical models [65]. Since mechanistic models, which have complex processes
consisting of the SLT efficiency are hard and time consuming, Osuolale and Zhang
developed an ANN to model both exergy efficiency and compositions of product for
systems of binary and multiple component in distillation columns. In addition, the
ANN was used for optimization of exergy efficiency depending on quality constraint
of product. According to this study, ANN models, which could be utilized to decide
exergy efficiencies for various operational circumstances of distillation processes,
provided excellent prediction result without complex calculation of the stream
enthalpies and entropies. Also, proposed methods can enhance the distillation column
exergy efficiency. As a result of the optimization, the methanol-water exergy
efficiency increased by 11.2% and for benzene-toluene by 1.8%. In addition,
modelling and optimization related with the models of ANN and BANN can help
deciding of energy effective operations and distillation columns control [24].
38
In another research, the ANN model was developed for crude distillation column by
Osuolale and Zhang. It was found that bootstrap aggregated neural network (BANN)
models predict optimum working conditions of atmospheric distillation unit (ADU).
Also, the result of BANN model gave accurate prediction. Considering this study, the
neural network model can be created by using the historical data of plant. When a
system is formed according to the actual operating conditions of a facility, the system
can be continuously made better. The reason why using proposed system is because
the recent operating conditions of plant can be utilized to train and set the ANN model.
In addition, operators and engineers will have the opportunity to operate the facility in
a very effective status thanks to the ANN model [66]. Osuolale and Anozie improved
the optimization of ADU using ANN model based on exergy. Nine operation variables
were used and optimization of ADU increased from 33% to 53% for exergy efficiency.
The profiles of vapor and fluid exergy ratio in the fractionating column were used as
a retrofit means to reveal inefficiency points inside the column and to propose possible
column change options for energy efficient processes. It was observed that the exergy
ratio profiles in the column intersected with each other. ADU’s optimization enhanced
the exergy efficiency from 53% to 60% provided that the intersection profiles of the
exergy ratio of the column were eliminated.
The consequence of this study showed that the ANN is strong and suitable
optimization model for solution of limited optimization problems like ADU with
restricted operation variables [67]. Keçebaş et al. used the ANN modeling based on
the algorithm of backpropagation learning to estimate the exergy performance of
Afyonkarahisar geothermic district heating system (AGDHS) at a wide operational
state ranges. The mean daily real thermic data obtained from AGDHS during the
between 2009 and 2010 heating season was gathered as well as it is used to analyze
exergy. The ANN prediction performance was evaluated using the coefficient of
correlation; for instance, it can be mean relative and root mean square error. The ANN
model presented good statistic behavior in general by a coefficient of determination
known as R2 in 0.9924-0.9942 interval, RMSs in the interval of 0.0571-0.0810, and
0.0020 for MARE. When the effects of input variables on some parameters such as
flow velocity of heat exchanger output, environmental, well heat and outlet
temperature were analyzed, the network gave reasonable results. According to results,
39
the successful ANN model might be implemented to predict AGDHS’s exergy
efficiency with a good dependability and trueness [68].
The ANN model was built by Ghritlahre to estimate an exergy and energy efficiency
of rough solar air heater with using the values of computed thermic and exergy
efficiency and empirical thermic and exergy efficiency. The ANN model was trained
using two learning algorithm. For the ANN model, the experiments time, flow rate of
mass, intensity of solar radioactivity, environmental temperature, average air
temperature and absorber layer temperature were considered as inputs; energy and
exergy efficiency were evaluated as outputs. The effectiveness of ANN model was
measured with using the analysis of statistical error. The determination coefficient
values for exergy and energy efficiency are 0.95737 and 0.99921 respectively, which
almost equal to one and hence give the correct results of the predictions. Considering
the great accuracy achievement for the energy and exergy yieldance predictions of the
solar air heater, multilayer perceptron ANN model can be preferred to examine energy
and exergy yieldance of any solar power systems [69]. In a different study, Mohanraj
et al estimated both exergy efficiency and destruction of direct extension solar-aided
heat pump with building the ANN model relating to algorithm of backpropagation
learning at various ambient circumstances. The consequences demonstrated that the
network gave a maximal coefficient of correlation along with minimal variance
coefficient and values of root mean square. In addition, the application of ANN model
is quite feasible with reference to results [70]. Sadrameli and Alizadeh applied the
modeling of ANN and generic algorithm (GA) combination to predict the total exergy
performance of the olefin cracking furnaces located in Iran petrochemical plant. While
steam ratio, residence time and coil outlet temperature were taken as input values,
exergy efficiency was taken as output value in the combined ANN-GA model. The
results indicated that the combination of ANN-GA model is very effectual method for
optimizing the model performance and predicting the total exergy efficiency because
correlation coefficient was obtained as 0.928 [71].
40
operational circumstances on the exergy performance of process can be evaluated on
determining by the engineers and operators via this prediction tool. In this study, the
predictive ability of ML models has been examined for different operational conditions
on exergy efficiency of vacuum distillation column. Simulated process operational
data of ASPEN HYSYS was used to both model input and output. Initially, the
parameters that influence exergy efficiency were decided by way of “Case Study” of
ASPEN HYSYS that generated the synthetic data. Considering the ML models,
bootstrap aggregated neural network (BANN) and random forest models were used for
prediction of exergy efficiency in R-Studio for this study. The data of both models are
separated to training, testing as well as unseen validation data. This division is such
that they are 50%, 30% and 20%, respectively. This neural network model for
forecasting the exergy efficiency is given in equation 4.1:
𝜑 = 𝑓(𝑥1 , 𝑥2 , 𝑥3 ) (4.1)
Where φ is exergy efficiency, x1, x2, and x3 are model input variables.
The neural network models could approximate any continuous nonlinear models but
common neural networks may lack the ability of generalization when it is used to
unseen data because of noise overfitting in the data [23]. Various techniques were
proposed to develop the generalization ability of neural network such as regularization,
Bayesian learning, early stopping, multiple network combination, training with static
and dynamic process data. Among these techniques, the multiple network combination
has an encouraging approach in respect to developing model estimation of unseen data.
In addition, the accuracy of estimation about overall input space might be increased
with using multi-neural network combination [72]. Breiman suggested a technique
called as bootstrap aggregating, or bagging in 1996 [73]. Bootstrap aggregated neural
network (BANN) is a resampling method that can be used to decrease the variance
based on prediction with several classification and regression methods so develops the
prediction of nonlinear models iteratively by using multiple models aggregation. It has
a simple concept that shown in Figure 4.6: many samples of bootstrap are pulled from
the present data, some prediction procedure is performed for each sample of bootstrap,
and the outcomes are combined to get the total prediction. A chart of BANN is
demonstrated in Figure 4.7.
41
In substitution for choosing a best accepted singular neural network model, they are
put together to build a robust and accurate model. Total output of the BANN is
obtained by weighted join of the singular neural network outputs. The BANN could
be described with equation 4.2 [72, 73].
The aggregated neural network forecaster, the aggregating weight of ith neural network
combination, ith neural network, neural networks number and the neural network inputs
vector are indicated as f(X), wi, fi(X), n and X, respectively in equation 4.2.
42
Figure 4.7: Bootstrap aggregated neural network [24, 72].
Leo Breiman and Adele Cutler combined the methods of “bootstrap aggregating” and
“random subspace” to develop random forest (RF) creating a group of decision trees
and classifying them. The RF, which is a robust ML algorithm, is an ensemble learning
method for performing tasks such as regression and classifications. The RF benefits
from bagging approach. The RF incorporates the classifiers of multiple decision tree
and the result categories are decided according to the status of decision tree
classification outputs like Figure 4.8. RF algorithms have three major parameters that
have to be adjusted before training. These involve the size of node, the trees number,
and the number of sampled features. For building one decision tree, the RF algorithm
uses two processes of random selection. Firstly, random training samples are selected.
The sample of arbitrary dataset is trained with several times to get a prediction model.
The RF works through improving a large number of decision trees at time of training.
Secondly, the random characteristics qualities of the sample are selected and the RF
gives the mode output (classification) of classes or the average estimation (regression)
of individual trees. After whole decision trees are created, the ending classification
result is decided by the voting method of equal weight. The broad datasets are handled
without loss of dimensionality by the RF. Getting the sum of multiple decision trees
will provide better performance, even if it is sensitive to overfitting. Also, the
importance measurement of the predictor variables is helpful in terms of interpreting
the RF model and the selecting of variable [57, 59, 75-77].
43
The RF model is preferred from a computer viewpoint in terms of some features that
are listed below [77]:
The RF model is preferred statistically in terms of extra features that are listed below
[77]:
44
5. RESULTS AND DISCUSSIONS
Comparison of the simulation results with the actual values is important before exergy
and energy studies in terms of accuracy of calculation results. For this reason, all real
operation values such as temperature, pressure, flow were compared with simulation
results. Exergy and energy calculations were made after the simulation results
converged to real values as much as possible. Table 5.1 indicates the comparison
between simulation and actual values of column tray temperatures in reference to
absolute difference and relative error.
Table 5.1: Comparison between simulation and actual values of column tray
temperatures.
1 0.000 0.000
2
3 0.205 0.174
4
5
6
7
8
9 2.256 0.924
10
11 2.543 0.820
12
13 2.008 0.692
14
15
Also, comparison based on the absolute difference and relative error of column return
temperatures with real and simulation values is given in Table 5.2.
45
Table 5.2: Comparison of column return temperatures with real and simulation
values.
Return Temperature (℃)
PA Type Absolute Difference Relative Error %
PA1 0.0543 0.1713
PA2 0.0184 0.0319
PA3 0.1004 0.0509
PA4 0.3809 0.1423
PA5 0.0037 0.0012
In addition, comparisons between simulation and actual values were made for some
heat exchangers that based on the absolute difference and relative error in Table 5.3.
Table 5.3: Comparisons between simulation and actual values were made for some
heat exchangers.
Stream No. Absolute Difference Relative Error (%) Symbolic HE
1 0.000 0.000
2 0.203 0.086
E1
3 2.513 1.209
4 1.514 1.654
5 0.084 0.047
6 0.055 0.022
E2
7 3.954 1.956
8 1.628 0.898
9 0.008 0.004
10 0.297 0.096
E3
11 1.028 0.387
12 0.077 0.031
13 0.032 0.016
14 0.006 0.002
E4
15 2.657 1.042
16 0.236 0.110
17 0.007 0.003
18 0.041 0.012
E5
19 1.703 0.566
20 2.512 0.842
46
Consequently, very close results were achieved by comparison with simulation and
actual values as sighted in Table 5.1, Table 5.2 and Table 5.3. At the same time, all
these comparisons were made in all unit operations of the plant for all operation values.
The flowchart shown in Figure 5.1 was used to carry out calculations of exergy and
energy analysis for every single unit operations in ASPEN HYSYS. The necessary
data such as equipment design, laboratory and operation values was obtained for
simulating the VDU. Then, simulation results were compared with the actual values.
When the simulation results are appropriate, energy and exergy analysis was
performed for every single unit operations by using the required equations in section
three. Necessary simulation outputs of each stream are used in order to use these
equations calculating energy and exergy.
Figure 5.1: The flowchart of energy and exergy calculation for all unit operations in
ASPEN HYSYS.
47
Selection of exergy transfer method is important for exergy calculation. Equations
3.17, 3.19 and 3.20 are used for each transfer method. After input and output exergy
calculation is done, exergy efficiency calculation is applied by using equation 3.22. In
addition, energy efficiency is calculated equation 3.21. The loss of exergy that is
named irreversibility is done by using equation 3.29. Also, Table 3.2 is used to
calculate exergy efficiency and loss for each unit operation. Both energy and exergy
calculations of each unit operations are done by creating spreadsheets in ASPEN
HYSYS.
Table 5.4 shows the calculation results of exergy and energy for E1 heat exchanger of
VDU. Irreversibility calculation of E1-1 is given by equation 5.1. Also, equation 5.1
can be expressed as equation 5.2 that is the difference of the summation of exergy
input rate and the summation of exergy output rate.
̇
𝐼𝐸1−1 = 𝑚̇1 (𝑒𝑥1 − 𝑒𝑥3 ) + 𝑚̇2 (𝑒𝑥2 − 𝑒𝑥4 ) (5.1)
̇
𝐼𝐸1−1 = (𝑚̇1 𝑒𝑥1 + 𝑚̇2 𝑒𝑥2 ) − (𝑚̇1 𝑒𝑥3 + 𝑚̇2 𝑒𝑥4 ) (5.2)
Where the sum of exergy input and output rate for the E1-1 is given by equation 5.3
and 5.4 respectively.
𝐸̇ 𝑥𝑐𝑜𝑙𝑑,𝑜 − 𝐸̇ 𝑥𝑐𝑜𝑙𝑑,𝑖
𝜓𝐸1−1 = ⁄̇ (5.5)
𝐸 𝑥ℎ𝑜𝑡,𝑖 − 𝐸̇ 𝑥ℎ𝑜𝑡,𝑜
48
Table 5.4: Exergy and energy calculations for E1 heat exchanger.
E1-1 Hin (kj/s) Hout (kj/s) Q (kj/s) Exergy Rate (kj/s) I (kj/s) Energy Efficiency Exergy Efficiency
In (1) -62384.1994 586.8611
In (2) -57368.5349 1673.5318
Out (3) -60672.4068 1146.6473
Out (4) -59080.3275 933.9857
-119752.7343 -119752.7343 0.0000 179.7598 1.0000 0.7569
E1-2 Hin (kj/s) Hout (kj/s) Q (kj/s) Exergy Rate (kj/s) I (kj/s) Energy Eff Exergy Eff
In (5) -60995.1932 1025.6604
In (6) -55152.3595 2887.4090
Out (7) -58779.0177 2000.7221
Out (8) -57368.5349 1673.5318
-116147.5526 -116147.5526 0.0000 238.8155 1.0000 0.8033
E1-3 Hin (kj/s) Hout (kj/s) Q (kj/s) Exergy Rate (kj/s) I (kj/s) Energy Eff Exergy Eff
In (9) -58676.5153 2052.7199
In (10) -49264.5649 7168.4324
Out (11) -52788.7208 5881.9468
Out (12) -55152.3595 2887.4090
-107941.0802 -107941.0802 0.0000 451.7966 1.0000 0.8945
49
Exergy calculations can be done by using standard enthalpy and entropy values instead
of exergy directly given by ASPEN HYSYS. Between equation 5.7 and 5.10 gives
exergy rate of each stream for E1-1. Exergy efficiency calculations can be done with
using these exergy rate equations.
For this study, the furnace of VDU was simulated as a heater in ASPEN HYSYS.
Equation 5.11 gives the exergy efficiency of furnace. Detailed energy and exergy
calculations of furnace are stated in Table 5.5.
Exergy efficiency for G-2 pump is calculated by equation 5.12 or 5.13. In addition,
both energy and exergy calculations are given Table 5.6.
50
(𝑇0 𝑆𝑔𝑒𝑛,𝐺−2 )
𝜓𝐺−2 = 1 − ⁄(𝐻 (5.13)
16 − 𝐻15 )
Where the generation of entropy, Sgen,G-2, for the G-2 pump is given by equation 5.14.
Energy and exergy calculations for the vacuum distillation column are shown in Table
5.7. Exergy efficiency calculation can be done using either equation 5.16 or equation
5.17.
∑𝑜 𝐸̇ 𝑥𝑜
𝜓𝑉𝐷𝑈 = ⁄ ̇ (5.16)
∑𝑖 𝐸 𝑥𝑖
[∑𝑖 𝐸̇ 𝑥𝑖 + 𝐸̇ 𝑥𝑄 − 𝐼𝑉𝐷𝑈
̇ ]
𝜓𝑉𝐷𝑈 = ⁄ ̇ (5.17)
∑𝑖 𝐸 𝑥𝑖
The necessary expressions such as the rate of exergy input and output, exergy
destruction, as well as heat exergy for equations 5.16 and 5.17 are calculated from the
following equations.
51
∑ 𝐸̇ 𝑥𝑜 = 𝑚̇20 𝑒𝑥20 + 𝑚̇21 𝑒𝑥21 + 𝑚̇22 𝑒𝑥22 + 𝑚̇23 𝑒𝑥23 + 𝑚̇24 𝑒𝑥24 + 𝑚̇25 𝑒𝑥25 (5.19)
𝑜
𝑇
𝐸̇ 𝑥𝑄 = (1 − 0⁄𝑇)𝑄̇𝑉𝐷𝑈 (5.20)
̇
𝐼𝑉𝐷𝑈 = ∑ 𝐸̇ 𝑥𝑖 − ∑ 𝐸̇ 𝑥𝑜 + 𝐸̇ 𝑥𝑄 (5.21)
𝑖 𝑜
Where the expressions of equation 5.18 to 5.21 are exergy inlet rate, exergy exit rate,
heat exergy and exergy destruction, respectively.
After doing the exergy and energy calculations for each unit operations of the plant,
Figure 5.2 showing the both exergy and energy efficiency of unit operations is
obtained. It is seen that almost all of the energy efficiency results are equal to one,
except for the column and the furnace. At the same time, the energy efficiency of the
column was greater than one. This proves that the energy efficiency calculation can be
insufficient and incorrect in determining the losses.
According to Figure 5.3, the E-12 heat exchanger using cooling water has the lowest
exergy efficiency due to heat released through hot water into the environment and
bypass flow on the cooling waterside. At the same time, the performance of the E-12
heat exchanger may decreased due to the contamination on the waterside. Another low
exergy efficiency equipment among all unit operations is the furnace with 0.518
exergy efficiency. Moreover, Izyan and Shuhaimi conducted the exergy analysis on
crude preheat train and furnace taking part a crude distillation unit. They found that
the furnace has the highest exergy destruction with 86% of overall exergy destruction
in the system [22]. Waheed et al. performed thermodynamic analysis of a petroleum
distillation unit to specify inefficiencies and they found that the most inefficient
equipment were the furnaces responsible for 51.6% of all exergy destructions [55].
There is a decrease in the efficiency of the furnace due to unwanted heat losses. They
can be caused by loss of heat to the environment from the exterior surface of the walls
via radiation and convection, loss of heat via exhaust gases, loss of heat via furnace
walls as well as loss of heat via gases leaking from clefts, openings [79]. Therefore,
low exergy efficiency due to unwanted losses in the furnace, which is a high energy
intensive process, is an expected result.
52
The E-13 is an air cooler that also has the low exergy efficiency with 0.62. Exergy
efficiency of an air cooler is expected to be low due to the high temperature drop and
the heat released through hot air into the environment so these situations cause exergy
loss. Vilarinho et al. carried out energy and exergy study in the pre-distillation unit. It
was concluded that air coolers are one of the lowest efficient equipment in the unit,
with an energy loss of 59.9% and an exergy loss of 11.2%. They also found that the
greatest exergy loss was in the furnace with 56.3% [21].
Vacuum distillation column with an exergy efficiency of 0.657 follows the E-13 air
cooler in Figure 5.3. Exergy loss of column occurs with entropy generation arising
from irreversibility. Exergy loss can be measured by the irreversibility occurring in the
column because of the loss of momentum, thermal as well as chemical potential that
can be also called as driving force of pressure, temperature and mass, respectively.
Column entropy increases with the mass and heat transfer occurring between flows. In
addition, condenser entropy increases with the just heat transfer. Also, mixing during
separation process causes the irreversibility. Moreover, the total exergy efficiency of
the column consists of the exterior and interior exergy efficiencies. Thermal
integration between units, overhead stream coproduction and recompression for the
reboiler are related to exterior exergy efficiency. However, column interior design,
column stage number, composition and position of feed as well as utility demand are
related to interior exergy efficiencies. Many distillation columns have low exergy
efficiency in the range of 20-25%, whereas specified modifications can enhance the
exergy efficiency by up to 60% [80, 81]. Odejobi applied an exergy analysis in a
petroleum distillation unit. It was concluded that among the other equipment, both
maximum irreversibility and minimum exergy efficiency of 52.1% occurred in the
column [82]. Accordingly, when compared to the general average exergy efficiency of
the columns, it is understood that the vacuum distillation column has a very high
exergy efficiency of 0.657.
Generally, it is seen that the exergy efficiencies of the heat exchangers between E-1
and E-11 are high with the range of 0.792-0.937 in Figure 5.2. Among the preheat
train, the performance of E-1 is lower since it depends on the bypass flow condition.
The exergy efficiency of the pumps also varies between 0.705 and 0.877 in Figure 5.2.
Since G-2 is a vacuum bottom pump, it is normal for its performance to be lower than
other pumps.
53
Table 5.7: Exergy and energy calculations a vacuum distillation column.
VDU Inlet Streams (17-19) Outlet Streams (20-25)
(17) (18) (19) (20) (21) (22) (23) (24) (25)
Exergy
37993.3390 1538.2114 0.0000 -10.2675 246.6981 4332.2077 4674.0268 4073.8481 12649.7015
(kj/s)
ΣExergy 39531.5504 25966.2147
I
8562.3746
(kj/s)
H
-126128.1394 -43739.8034 0.0000 -3860.0002 -18490.9885 -56233.9684 -34344.3975 -21516.2704 -60162.0334
(kj/s)
ΣH -169867.9427 -194607.6585
Q
24739.7157
(kj/s)
Energy Eff. 1.1456
Exergy Eff. 0.6568
54
Equipment vs Energy, Exergy Efficiency
Energy Efficiency Exergy Efficiency
1.20
0.818 0.821
0.792
0.80
0.729 0.712
0.705
0.657
0.620
0.60
0.518
0.379
0.40
0.20
0.00
E-1 E-2 E-3 E-4 E-5 E-7 E-8 E-9 E-10 E-11 E-12 E-13 F1 C1 G-2 G-3 G-4 G-5 G-6
Equipment
Figure 5.2: The exergy and energy efficiencies of all unit operations in the plant.
55
Exergy Efficiency of All Unit Operations
E-4
E-5
E-2
E-3
G-3
G-4
E-10
E-9
E-11
E-7
E-1
E-8
G-5
G-6
G-2
C1
E-13
F1
E-12
0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 0.900 1.000
Figure 5.3: Listing the magnitude of exergy efficiencies for all unit operations.
56
5.3 Exergy Study in Vacuum Distillation Column with Using Machine Learning
There are many process control limits in the vacuum distillation column. Using these
limits, a case study was conducted in ASPEN HYSYS. This case study investigates
the change in exergy and energy efficiency for lower and upper limits range of the
process control variables in the vacuum distillation column. Not only the process
control limits but also the effect of the furnace exit temperature on the column exergy
and energy efficiency was examined. These consequences are given in Table 5.8.
Table 5.8: The effect of operating parameters on exergy efficiencies of the column.
Exergy % Absolute Energy % Absolute
Operating Parameters Efficiency Exergy Efficiency Energy
Change Change Change Change
PA1 flow (t/h) 0.6499-0.6576 1.185 1.138-1.131 0.615
PA2 flow (t/h) 0.6552-0.6567 0.229 1.133-1.132 0.088
PA2 temperature (℃) 0.6570-0.6560 0.152 1.132-1.133 0.088
PA3 flow (t/h) 0.6565-0.6536 0.442 1.133-1.134 0.088
PA3 temperature (℃) 0.6431-0.6560 2.006 1.138-1.133 0.439
PA4 flow (t/h) 0.6555-0.6599 0.671 1.133-1.131 0.177
PA4 temperature (℃) 0.6571-0.6442 1.963 1.132-1.138 0.53
PA5 flow (t/h) 0.6560-0.6556 0.061 1.133-1.133 0
PA5 temperature (℃) 0.6529-0.6560 0.475 1.134-1.133 0.088
PA6 temperature (℃) 0.6560-0.6560 0 1.133-1.133 0
Furnace exit temperature (℃) 0.6655-0.6290 5.485 1.124-1.157 2.936
57
5.4 Exergy Study in Vacuum Distillation Column with Using Machine Learning
In this study, the BANN and the RF models were developed for exergy predictions
according to the most affecting parameters on vacuum distillation column such as
furnace exit temperature as well as PA3 and PA4 temperature, which are the model
inputs. Figure 5.4 and Figure 5.6 shows the scatter plots of the BANN and RF model
predicted versus actual column exergy efficiency on training, testing and validation
datasets, respectively. When the plots are examined, it is seen that the actual outcomes
are close to the predicted values. The accuracy of the model increases if all points are
closer to the diagonal line. In addition, it is realized that the datasets of BANN model
graph in Figure 5.4 is more scattered than the RF model in Figure 5.6. Therefore, this
result gives that the RF model can predict to column exergy efficiency more
accurately. Figure 5.5 and Figure 5.7 demonstrate the BANN and RF model
performances in prediction of column exergy efficiency for training, testing and
validation datasets, respectively. Red dashed lines represent the actual values while
blue lines represent predicted values in Figure 5.5 and Figure 5.7. It can be seen that
the BANN and RF model prediction results in Figure 5.5 and Figure 5.7 are fitted with
the actual values. However, when Figure 5.5 and Figure 5.7 are compared, it is seen
that the predicted values of the RF model overlap the actual values better, so it gives
more accurate prediction results. The decreasing trend in Figure 5.5 and Figure 5.7 is
due to the furnace exit temperature, which is one of the model inputs. As the furnace
exit temperature increases, the column input flow exergy increases and consequently
the exergy efficiency of the column decreases. In the RF model, Figure 5.8 showing
the order of importance variables for training, testing and validation datasets can be
obtained. The RF model uses two calculation methods for this, namely Mean Decrease
Accuracy (%IncMSE) and Mean Decrease Gini (IncNodePurity). According to Figure
5.8, the most important factor affecting column exergy efficiency is the furnace exit
temperature. Also, there are various statistical methods for evaluating the model
performances. The mean squared error (MSE) was used for this study. In Table 5.9, it
is seen that the mean squared error of the RF model is lower than the BANN model,
so it was proved that the RF model predicts exergy efficiency better by the MSE
method. All in all, it was concluded that exergy efficiency can be predicted with ML
models without performing complex exergy calculations.
58
Figure 5.4: Scatter plot of the BANN model predicted versus actual exergy
efficiency of the column for training, testing and validating data.
59
Figure 5.5: The BANN model predicted and actual exergy efficiency of the column
for training, testing and validating data.
60
Figure 5.6: Scatter plot of the RF model predicted versus actual exergy efficiency of
the column for training, testing and validating data.
61
Figure 5.7: The RF model predicted and actual exergy efficiency of the column for
training, testing and validating data.
62
Figure 5.8: Variable importance for (a) training (b) testing (c) validating data.
63
64
6. CONCLUSION
In this study, the exergy and energy investigation of the vacuum distillation unit in
TUPRAS Refinery was performed. The formulations required to perform exergy and
energy calculations were found through literature research. After collecting all the data
required for simulation, the unit was simulated with ASPEN HYSYS. Each equipment
in the VDU was evaluated in both energy and exergy studies. An additional study was
carried out for the exergy efficiency of the column. By using the limit values of the
process control parameters affecting the column, the effect of operational parameters
on the column exergy efficiency was examined with the case study feature of ASPEN
HYSYS. After deciding on the parameters affecting the column, the exergy efficiency
prediction in the column was studied with machine learning models.
65
general column efficiencies in the literature. The reason for the low
performance of G-2 among pumps is that it is a vacuum bottom pump.
4. When all operational parameters such as PA temperature and flow affecting
the column were evaluated, it was understood that these parameters did not
significantly affect the column exergy efficiency. Except for the furnace exit
temperature, only 2% exergy efficiency change was observed. Therefore, it
was understood that there was very little operability part in the column in order
to increase energy efficiency. The main parameters affecting the column were
the furnace exit temperature, PA3 and PA4 temperatures, respectively.
5. The BANN and RF ML models were used to estimate the column exergy
efficiency. It has been found that they are successful in predicting exergy
efficiency. Good overlap was observed between the actual values and the
model predicted values on training, testing and validation data. These results
were also supported by MSE. It was observed that the RF model has a higher
prediction performance compared to the BANN model.
6. The predictability of exergy efficiency in the column was proven without the
need for ASPEN HYSYS simulation study and complex exergy calculations.
At the same time, using these prediction models, it can be decided for which
operational conditions the column is efficient or not.
66
REFERENCES
[1] Taheri, K., Gadow, R., & Killinger, A. (2014). Exergy analysis as a developed
concept of energy efficiency optimized processes: The case of thermal
spray processes. Procedia CIRP, 17(Complete), 511-516.
doi:10.1016/j.procir.2014.01.060.
[2] International Petroleum Industry Environmental Conservation Association
(IPIECA). (2013). Saving energy in the oil and gas industry (pp. 1-13).
[3] Crijns-Graus, W., Yue, H., Zhang, S., Kermeli, K., & Worrell, E. (2020).
Energy efficiency improvement opportunities in the global industrial
sector. Encyclopedia of Renewable and Sustainable Materials, 5, 377-
388. doi:10.1016/B978-0-12-803581-8.10906-3.
[4] Rosen, M. A., & Bulucea, C. A. (2009). Using exergy to understand and improve
the efficiency of electrical power technologies. Entropy, 11(4), 820-
835. doi:10.3390/e11040820.
[5] Dincer, I., & Rosen, M. A. (2013). Chapter 1 - Thermodynamic
fundamentals. Exergy: Energy, Environment and Sustainable
Development (pp. 1-20). Oxford, USA: Elsevier. doi:10.1016/B978-0-
08-097089-9.00001-2.
[6] Gilbert, A., Mesmer, B., & Watson, M. D. (2016). Uses of exergy in systems
engineering. In Proceedings of the 2016 Conference on Systems
Engineering Research, Mar (pp. 22-24).
[7] Terzi, R. (2018). Application of exergy analysis to energy systems. Application of
Exergy (pp. 109-123). doi:10.5772/intechopen.74433.
[8] Rosen, M. A. (2007, October). Exergy concept and its application. In 2007 IEEE
Canada Electrical Power Conference (pp. 473-478). IEEE. doi:
10.1109/EPC.2007.4520378.
[9] Dong, Z., Li, D., Wang, Z., & Sun, M. (2018). A review on exergy analysis of
aerospace power systems. Acta Astronautica, 152, 486-495.
doi:10.1016/j.actaastro.2018.09.003.
[10] Fellaou, S., Ruiz-Garcia, A., & Gourich, B. (2021). Enhanced exergy analysis
of a full-scale brackish water reverse osmosis desalination
plant. Desalination, 506, 114999. doi:10.1016/j.desal.2021.114999.
[11] Samimi, F., Marzoughi, T., & Rahimpour, M. R. (2020). Energy and exergy
analysis and optimization of biomass gasification process for hydrogen
production (based on air, steam and air/steam gasifying
agents). International Journal of Hydrogen Energy, 45(58), 33185-
33197. doi:10.1016/j.ijhydene.2020.09.131.
67
[12] Bai, W., Li, H., Zhang, L., Zhang, Y., Yang, Y., Zhang, C., & Yao, M. (2021).
Energy and exergy analyses of an improved recompression supercritical
CO2 cycle for coal-fired power plant. Energy, 222, 119976.
doi:10.1016/j.energy.2021.119976.
[13] Mrzljak, V., Poljak, I., Prpić-Oršić, J., & Jelić, M. (2020). Exergy analysis of
marine waste heat recovery CO2 closed-cycle gas turbine
system. Pomorstvo, 34(2), 309-322. doi:10.31217/p.34.2.12.
[14] Bühler, F., Nguyen, T. V., Jensen, J. K., Holm, F. M., & Elmegaard, B. (2018).
Energy, exergy and advanced exergy analysis of a milk processing
factory. Energy, 162, 576-592. doi:10.1016/j.energy.2018.08.029.
[15] Al-Muslim, H., Dincer, I., & Zubair, S. M. (2003). Exergy analysis of single-
and two-stage crude oil distillation units. J. Energy Resour.
Technol., 125(3), 199-207. doi: 10.1115/1.1595670.
[16] Al-Muslim, H., Dincer, I., & Zubair, S. M. (2005). Effect of reference state on
exergy efficiencies of one-and two-stage crude oil distillation
plants. International Journal of Thermal Sciences, 44(1), 65-73.
doi:10.1016/j.ijthermalsci.2004.04.015.
[17] Tarighaleslami, A. H., Omidkhah, M. R., Ghannadzadeh, A., & Hesas, R. H.
(2012). Thermodynamic evaluation of distillation columns using
exergy loss profiles: a case study on the crude oil atmospheric
distillation column. Clean Technologies and Environmental
Policy, 14(3), 381-387. doi:10.1007/s10098-012-0465-6.
[18] Gu, W., Huang, Y., Wang, K., Zhang, B., Chen, Q., & Hui, C. W. (2014).
Comparative analysis and evaluation of three crude oil vacuum
distillation processes for process selection. Energy, 76, 559-571.
doi:10.1016/j.energy.2014.08.053.
[19] Li, X., Lin, C., Wang, L., & Li, H. (2013). Exergy analysis of multi-stage crude
distillation units. Frontiers of Chemical Science and Engineering, 7(4),
437-446. doi:10.1007/s11705-013-1349-y.
[20] Dincer, I., & Rosen, M. A. (2007). Exergy analysis of crude oil distillation
systems. Exergy: Energy, Environment and Sustainable Development
(pp. 290-302). Oxford, USA: Elsevier
[21] Vilarinho, A. N., Campos, J. B. L. M., & Pinho, C. (2017). Energy and exergy
analysis of a pre-distillation unit. A case study. International Journal
of Thermodynamics, 20(2), 91-100. doi:10.5541/ijot.5000209345.
[22] Izyan, Z. N., & Shuhaimi, M. (2014). Exergy analysis for fuel reduction
strategies in crude distillation unit. Energy, 66, 891-897.
doi:10.1016/j.energy.2014.01.026.
[23] Osuolale, F. N., & Zhang, J. (2018). Exergetic optimisation of atmospheric and
vacuum distillation system based on bootstrap aggregated neural
network models. In Exergy for A Better Environment and Improved
Sustainability 1 (pp. 1033-1046). Springer, Cham.
[24] Osuolale, F. N., & Zhang, J. (2016). Energy efficiency optimisation for
distillation column using artificial neural network models. Energy, 106,
562-578. doi:10.1016/j.energy.2016.03.051.
68
[25] Azad, A. K., Rasul, M. G., Khan, M. M. K., Mondal, S. K., & Islam, R. (2016).
Modeling and simulation of heat and mass flow by Aspen Hysys for
petroleum refining process in field application. In Thermofluid
Modeling for Energy Efficiency Applications (pp. 227-257). Academic
Press. doi:10.1016/B978-0-12-802397-6.00010-5.
[26] Ajikutira, D. (2016). Jump Start: Aspen Hysys® V8. 0. A Guide for Getting
Started in Aspen HYSYS.
[27] Khalaf, A. N. (2018). Steady state simulation of Basrah Crude Oil Refinery
distillation unit using Aspen Hysys. University of Thi-Qar Journal for
Engineering Sciences, 9(2), 29-39.
[28] Rahmanian, N., Aqar, D. Y., Bin Dainure, M. F., & Mujtaba, I. M. (2018).
Process simulation and assessment of crude oil stabilization unit. Asia‐
Pacific Journal of Chemical Engineering, 13(4), e2219.
doi:10.1002/apj.2219.
[29] Overview of Crude Units. (n.d.). Retrieved March 22, 2021, from
https://www.ou.edu/class/che-design/che5480-
07/Petroleum%20Fractionation-Overview.pdf
[30] Tech, A. (2005). Hysys 2004.2 operations guide. Cambridge: Aspen Technology
Inc.
[31] Poljak, I., Orović, J., & Mrzljak, V. (2018). Energy and exergy analysis of the
condensate pump during internal leakage from the marine steam
propulsion system. Pomorstvo, 32(2), 268-280.
doi:10.31217/p.31.2.12.
[32] Ghannadzadeh, A., Thery-Hetreux, R., Baudouin, O., Baudet, P., Floquet,
P., & Joulia, X. (2012). General methodology for exergy balance in
ProSimPlus® process simulator. Energy, 44(1), 38-59.
doi:10.1016/j.energy.2012.02.017.
[33] Araújo, A. B., Brito, R. P., & Vasconcelos, L. S. (2007). Exergetic analysis of
distillation processes-a case study. Energy, 32(7), 1185-1193.
doi:10.1016/j.energy.2006.07.003.
[34] Waheed, M. A., & Oni, A. O. (2015). Performance improvement of a crude oil
distillation unit. Applied Thermal Engineering, 75, 315-324.
doi:10.1016/j.applthermaleng.2014.10.078.
[35] Dincer, I., & Rosen, M. A. (2013). Chapter 2 - Exergy and energy analyses.
Exergy: Energy, Environment and Sustainable Development (pp. 21-
30). Oxford, USA: Elsevier. doi:10.1016/B978-0-08-097089-9.00002-
4.
[36] Ghasemi, E., Yazdanei, S., Rahimi, A., & Afshar, M. (2017). Thermodynamic
evaluation of a kerosene pre-fraction unit using energy and exergy
analysis. Case Studies in Thermal Engineering, 10, 413-422.
doi:10.1016/j.csite.2017.09.004.
[37] Rivero, R., Rendón, C., & Gallegos, S. (2004). Exergy and exergoeconomic
analysis of a crude oil combined distillation unit. Energy, 29(12-15),
1909-1927. doi:10.1016/j.energy.2004.03.094.
69
[38] Sánchez, Y. A. C., & de Oliveira Jr, S. (2015). Exergy analysis of offshore
primary petroleum processing plant with CO2 capture. Energy, 88, 46-
56. doi:10.1016/j.energy.2015.05.130.
[39] Silva, J. A. M., & Oliveira Jr, S. (2014). An exergy-based approach to determine
production cost and CO2 allocation in refineries. Energy, 67, 607-616.
doi:10.1016/j.energy.2014.01.036.
[40] de Oliveira, S. (2013). Exergy evaluation of petroleum production and refining
processes. In Exergy. Green Energy and Technology (pp. 111-159).
Springer, London. doi:10.1007/978-1-4471-4165-5_4.
[41] Farajzadeh, R., Zaal, C., van Den Hoek, P., & Bruining, J. (2019). Life-cycle
assessment of water injection into hydrocarbon reservoirs using exergy
concept. Journal of Cleaner Production, 235, 812-821.
doi:10.1016/j.jclepro.2019.07.034.
[42] Dincer, I., & Rosen, M. A. (2013). Chapter 5 - Applications of exergy in industry.
Exergy: Energy, Environment and Sustainable Development (pp. 75-
82). Oxford, USA: Elsevier. doi:10.1016/B978-0-08-097089-9.00005-
X.
[43] Marmolejo-Correa, D., & Gundersen, T. (2015). A new efficiency parameter
for exergy analysis in low temperature processes. International Journal
of Exergy, 17(2), 135-170. doi:10.1504/IJEX.2015.069988.
[44] Lior, N., & Zhang, N. (2007). Energy, exergy, and second law performance
criteria. Energy, 32(4), 281-296. doi:10.1016/j.energy.2006.01.019.
[45] Nguyen, T. V., Voldsund, M., Elmegaard, B., Ertesvåg, I. S., & Kjelstrup, S.
(2014). On the definition of exergy efficiencies for petroleum systems:
Application to offshore oil and gas processing. Energy, 73, 264-281.
doi:10.1016/j.energy.2014.06.020.
[46] Rivero, R., Garcia, M., & Urquiza, J. (2004). Simulation, exergy analysis and
application of diabatic distillation to a tertiary amyl methyl ether
production unit of a crude oil refinery. Energy, 29(3), 467-489.
doi:10.1016/j.energy.2003.10.007.
[47] Marmolejo-Correa, D., & Gundersen, T. (2012). A comparison of exergy
efficiency definitions with focus on low temperature processes. Energy,
44(1), 477-489. doi:10.1016/j.energy.2012.06.001.
[48] Magnanelli, E., Berglihn, O. T., & Kjelstrup, S. (2018). Exergy‐based
performance indicators for industrial practice. International Journal of
Energy Research, 42(13), 3989-4007. doi:10.1002/er.4123.
[49] Grassmann, P. (1950). Towards a general definition of efficiencies. Chemie
Ingenieur Technik, 22(4), 77-80.
[50] Kotas, T. J. (1995). The exergy method of thermal plant analysis. 2nd ed.
Malabar, Florida: Krieger Publishing.
[51] Szargut, J., Morris, D. R., & Steward, F. R. (1987). Exergy analysis of thermal,
chemical, and metallurgical processes.
70
[52] Le Goff, P. (1994). The Efficiency of Industrial Processes: Exergy Analysis and
Optimization (Vol. 9). V. M. Brodyansky, & M. V. Sorin (Eds.).
Amsterdam: Elsevier.
[53] Tsatsaronis, G. (1993). Thermoeconomic analysis and optimization of energy
systems. Progress in Energy and Combustion Science, 19(3), 227-257.
[54] Bahadori, F., & Oshnuie, M. N. (2019). Exergy analysis of indirect dimethyl
ether production process. Sustainable Energy Technologies and
Assessments, 31, 142-145. doi:10.1016/j.seta.2018.12.025.
[55] Waheed, M. A., Oni, A. O., Adejuyigbe, S. B., & Adewumi, B. A. (2014).
Thermoeconomic and environmental assessment of a crude oil
distillation unit of a Nigerian refinery. Applied Thermal Engineering,
66(1-2), 191-205. doi:10.1016/j.applthermaleng.2014.02.007.
[56] Kang, Z., Catal, C., & Tekinerdogan, B. (2020). Machine learning applications
in production lines: A systematic literature review. Computers &
Industrial Engineering, 149, 106773. doi:10.1016/j.cie.2020.106773.
[57] Balaji, T. K., Annavarapu, C. S. R., & Bablani, A. (2021). Machine learning
algorithms for social media analysis: A survey. Computer Science
Review, 40, 100395. doi:10.1016/j.cosrev.2021.100395.
[58] Wang, J., He, Y. L., & Zhu, Q. X. (2020). Energy and production efficiency
optimization of an ethylene plant considering process operation and
structure. Industrial & Engineering Chemistry Research, 59(3), 1202-
1217. doi:10.1021/acs.iecr.9b05315.
[59] Paturi, U. M. R., & Cheruku, S. (2021). Application and performance of
machine learning techniques in manufacturing sector from the past two
decades: A review. Materials Today: Proceedings, 38, 2392–2401.
doi:10.1016/j.matpr.2020.07.209.
[60] Xing, J., Wang, H., Luo, K., Wang, S., Bai, Y., & Fan, J. (2019). Predictive
single-step kinetic model of biomass devolatilization for CFD
applications: A comparison study of empirical correlations (EC),
artificial neural networks (ANN) and random forest (RF). Renewable
Energy, 136, 104-114. doi:10.1016/j.renene.2018.12.088.
[61] Simske, S. (2019). Chapter 1 - Introduction, overview, and applications. Meta-
Analytics: Consensus Approaches and System Patterns for Data
Analysis. Morgan Kaufmann, 1-98.
[62] Talia, D., Trunfio, P., & Marozzo, F. (2016). Chapter 1 - Introduction to data
mining. Data Analysis in the Cloud: Models, Techniques and
Applications. Elsevier, 1-25.
[63] Yang, Y. (2017). Chapter 4 - Ensemble learning. Temporal Data Mining via
Unsupervised Ensemble Learning. Elsevier, 35-56.
[64] Guo, Y., Wang, X., Xiao, P., & Xu, X. (2020). An ensemble learning framework
for convolutional neural network based on multiple classifiers. Soft
Computing, 24(5), 3727-3735. doi:10.1007/s00500-019-04141-w.
71
[65] Azadbakht, M., Aghili, H., Ziaratban, A., & Torshizi, M. V. (2017).
Application of artificial neural network method to exergy and energy
analyses of fluidized bed dryer for potato cubes. Energy, 120, 947-958.
doi:10.1016/j.energy.2016.12.006.
[66] Osuolale, F. N., & Zhang, J. (2017). Thermodynamic optimization of
atmospheric distillation unit. Computers & Chemical Engineering, 103,
201-209. doi:10.1016/j.compchemeng.2017.03.024.
[67] Osuolale, F. N., & Anozie, A. N. (2017). Optimization of atmospheric distillation
unit of Warri Refinery using artificial neural network and exergy rate
profiles. Covenant Journal of Engineering Technology, 1(2), 1-15.
[68] Keçebaş, A., Yabanova, İ., & Yumurtacı, M. (2012). Artificial neural network
modeling of geothermal district heating system thought exergy
analysis. Energy Conversion and Management, 64, 206-212.
doi:10.1016/j.enconman.2012.06.002.
[69] Ghritlahre, H. K. (2018). Development of feed-forward back-propagation neural
model to predict the energy and exergy analysis of solar air heater.
Trends in Renewable Energy, 4(2), 213-235. doi:
10.17737/tre.2018.4.2.0078.
[70] Mohanraj, M., Jayaraj, S., & Muraleedharan, C. (2009). Exergy analysis of
direct expansion solar‐assisted heat pumps using artificial neural
networks. International Journal of Energy Research, 33(11), 1005-
1020. doi:10.1002/er.1534.
[71] Alizadeh, M., & Sadrameli, S. M. (2016). Modeling of thermal cracking
furnaces via exergy analysis using hybrid artificial neural network–
genetic algorithm. Journal of Heat Transfer, 138(4).
doi:10.1115/1.4032171.
[72] Zhang, J. (2004). A reliable neural network model based optimal control strategy
for a batch polymerization reactor. Industrial & Engineering Chemistry
Research, 43(4), 1030-1038. doi:10.1021/ie034136s.
[73] Sutton, C. D. (2005). Classification and regression trees, bagging, and boosting.
Handbook of Statistics, 24, 303-329. doi:10.1016/S0169-
7161(04)24011-1.
[74] Tahir, M. F., & Saqib, M. A. (2016). Optimal scheduling of electrical power in
energy-deficient scenarios using artificial neural network and bootstrap
aggregating. International Journal of Electrical Power & Energy
Systems, 83, 49-57. doi:10.1016/j.ijepes.2016.03.046.
[75] Ren, Q., Cheng, H., & Han, H. (2017, March). Research on machine learning
framework based on random forest algorithm. In AIP Conference
Proceedings (Vol. 1820, No. 1, p. 080020). AIP Publishing LLC.
doi:10.1063/1.4977376.
[76] IBM Cloud Education. (2020). Random forest. Retrieved April 10, 2021, from
https://www.ibm.com/cloud/learn/random-forest
[77] Cutler, A., Cutler, D. R., & Stevens, J. R. (2012). Random forests. In Ensemble
Machine Learning (pp. 157-175). Springer, Boston, MA.
doi:10.1007/978-1-4419-9326-7_5.
72
[78] Lets Open the Black Box of Random Forests. (2020). Retrieved April 10, 2021,
from https://www.analyticsvidhya.com/blog/2020/12/lets-open-the-
black-box-of-random-forests/
[79] Bureau of Energy Efficiency. (n.d.). Energy performance assessment of furnaces
(pp. 31-44). Retrieved May 9, 2021, from
https://www.beeindia.gov.in/sites/default/files/4Ch2.pdf
[80] Demirel, Y. (2004). Thermodynamic analysis of separation systems. Separation
Science and Technology, 39(16), 3897-3942. doi:10.1081/SS-
200041152.
[81] Tijani, A. S., Ramzan, N., & Witt, W. (2007). Process and plant improvement
using extended exergy analysis, a case study. In Rev. Chim (Bucureºti).
(pp. 392-396). Retrieved May 9, 2021, from
https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.1086.6721
&rep=rep1&type=pdf
[82] Odejobi, O. J. (2015). Exergy and economic analyses of crude oil distillation
unit. African Journal of Engineering Research, 3(2), 44-55.
73
74
CURRICULUM VITAE
E-Mail : [email protected]
EDUCATION :
PROFESSIONAL EXPERIENCE :
75