pp 5 - 28: Dominika Bandoła

Transcription

pp 5 - 28: Dominika Bandoła
Archiwum Instytutu Techniki Cieplnej
Vol. 1 Nr 1(2016)
Praca zbiorowa pod redakcja:
˛
Ziemowita Ostrowskiego,
Marka Rojczyka,
Arkadiusza Ryfy,
Wojciecha Stanka
www.itc.polsl.pl
Gliwice, 2016
Archiwum Instytutu Techniki Cieplnej Vol. 1 Nr 1(2016)
Praca zbiorowa pod redakcja:
˛ Z. Ostrowskiego, M. Rojczyka, A. Ryfy, W. Stanka
Recenzenci:
prof. dr hab. inż. Andrzej J. Nowak, prof. dr hab. inż. Ireneusz Szczygieł, dr hab. inż. Jacek Smołka,
dr hab. inż. Wojciech Stanek, dr inż. Lucyna Czarnowska, dr inż. Ziemowit Ostrowski,
dr inż. Marek Rojczyk, dr inż. Arkadiusz Ryfa
Archiwum Instytutu Techniki Cieplnej
www.itc.polsl.pl/ArchiwumITC
[email protected]
Kolegium redakcyjne:
Ziemowit Ostrowski, Andrzej Szl˛ek, Andrzej J. Nowak, Zbigniew Buliński
Wydawca:
Instytut Techniki Cieplnej
Wydział Inżynierii Środowiska i Energetyki, Politechnika Ślaska
˛
ul. Konarskiego 22, 44-100 Gliwice tel. +48 32 237 1661, fax +48 32 237 2872
www.itc.polsl.pl
W Archiwum ITC publikowane sa˛ wybrane prace powstałe podczas realizacji projektów magisterskich oraz inżynierskich realizowanych pod opieka˛ pracowników naukowych Instytutu Techniki Cieplnej Politechniki Ślaskiej.
˛
Wszystkie prace naukowe sa˛ recenzowane. Przyjmowane sa˛ prace w j˛ezyku polskim lub angielskim. Wybór j˛ezyka
publikacji pozostawiono autorom. Manuskrypty należy przesłać na powyżej podany adres poczty elektronicznej
redakcji w postaci sformatowanej – wzór formatu zamieszczony jest na stronie internetowej. Zgłoszenie powinno
zawierać plik tekstu w formacie *.docx (*.doc) oraz *.pdf ze wstawionymi rysunkami. Redakcja nie przepisuje
tekstów i nie wykonuje rysunków. Monografia publikowana jest w trybie Open Access. Nadesłanie manuskryptu
jednoznaczne jest z wyrażeniem zgodny na publikacj˛e pracy na warunkach poniższej licencji.
Opublikowano na warunkach licencji/Published under licence
by Instytut Techniki Cieplnej, 2016.
Zawartość tej publikacji może byc wykorzystana na warunkach licencji Uznanie Autorstwa 3.0 Polska. Licencja
pozwala na kopiowanie, zmienianie, rozprowadzanie, przedstawianie i wykonywanie utworu jedynie pod warunkiem oznaczenia autorstwa: autora(-ów), tytułu rozdziału, nazwy serii, tomu, strony. (Content from this work may
be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work
must maintain attribution to the author(s), chapter title, series title, volume, pages.)
Projekt okładki: M. Sokalska, K. Kalina & Z. Ostrowski
Skład: Z. Ostrowski. Wydrukowano z plików elektronicznych w postaci przekazanej przez autorów.
ISBN 978-83-61506-33-1
Druk: Centrum Poligrafii Politechniki Ślaskiej,
˛
ul. Kujawska 1D, 44-100 Gliwice
PRINTED IN POLAND
Spis treści
Identification and modeling the pulsatile blood flow in the cardiovascular system using a zero-dimensional model in an electrical analogy
(Identyfikacja i modelowanie pulsacyjnego przepływu krwi w systemie krwionośnym przy użyciu zero–wymiarowego modelu w analogii elektrycznej)
Dominika Bandoła . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
Multi-fluid Euler-Euler model of the blood flow within the blood vessel with
rigid walls
(Wielofazowy model przepływu krwi w naczyniu krwionośnym ze sztywnymi
ścianami w podejściu Euler-Euler)
Maria Gracka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
Identification and modelling of blood flow processes in section of large blood
vessel using hybrid Euler-Lagrange multiphase approach
(Identyfikacja oraz modelowanie procesu przepływu krwi w cz˛eści dużego naczynia krwionośnego z wykorzystaniem wielofazowego modelu hybrydowego
Euler-Lagrange)
Artur Knopek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
Analiza i modelowanie wymiany ciepła w obiektach przemysłowych ogrzewanych promiennikami podczerwieni
(Analysis and identification of radiation heat transfer in the industrial hall)
Adrian Orłowski . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
Wykorzystanie egzergii kriogenicznej skroplonego gazu ziemnego do produkcji energii elektrycznej
(Exploiting the cryogenic exergy of liquefied natural gas in production of electricity)
Tomasz Simla . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Numerical analysis of the orthotropic sample for conductivity tests
(Analiza numeryczna pola temperatury w ortotropowej próbce)
Mieszko Tokarski . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Indeks autorów . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Indeks słów kluczowych . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3
c
Instytut
Techniki Cieplnej, Politechnika Ślaska
˛
Archiwum Instytutu Techniki Cieplnej Vol. 1 nr 1(2016)
www.itc.polsl.pl/ArchiwumITC
Identification and modeling the pulsatile blood flow in the cardiovascular system using a zero-dimensional model in an electrical
analogy
Dominika Bandoła*†
e-mail: [email protected]
Key words: Windkessel Model, boundary conditions, arterial pressure, blood flow, aorta
Abstract
Advanced computer technology tools, like computational fluid dynamics (CFD) and knowledge
about the functioning of the human blood circulatory system, structure of blood, behavior
of vessels – allow improving understanding the process of blood distribution in human body.
Complex simulation has to assume multiphase approach, walls elasticity and pulsating blood
flow conditions, resulting from work of the heart [1]. In the presented work the blood flow and
pressure were simulated.
The characteristic impedance, peripheral resistance, capacitance in the systemic peripheral vascular beds had been taken into the consideration in electrical analogues as a lumped parameter
model (LPM) of the circulatory system which will be implemented as a boundary conditions in
the complete CFD model on the inlet of ascending aorta and on the outlets
of descending aorta, innominate artery, left common artery and left subclavian artery [2].
The resistance in a vessel was modeled by electronic component – resistor.
The blood flow is not stationary, it is stored (e.g. in the vessels and kidneys) just like energy in
the capacitors. Coil is an analog to the inert tendency of blood, which mass resists to move due
to the pressure difference. Furthermore it can be assumed that flow has only one direction – so
vessels act like a diodes in electronic circles. Additionally the conservation of the mass principle
has been applied converted into the Kirchhoff’s law [3].
*
This chapter was prepared during master thesis project at the Institute of Thermal Technology, Department
of Energy and Environmental Engineering of the Silesian University of Technology, under the supervision of
Dr Marek Rojczyk.
†
This research is supported by National Science Centre, Poland, within the project No 2014/13/B/ST8/04225. This
help is gratefully acknowledged herewith.
Zawartość tej publikacji może byc wykorzystana na warunkach licencji Uznanie Autorstwa 3.0 Polska. Licencja pozwala na
kopiowanie, zmienianie, rozprowadzanie, przedstawianie i wykonywanie utworu jedynie pod warunkiem oznaczenia autorstwa: autora,
tytułu rozdziału, nazwy serii, tomu, strony. Content from this work may be used under the terms of the Creative Commons Attribution
3.0 licence. Any further distribution of this work must maintain attribution to the author, chapter title, series title, volume, pages.
5
The realistic, time-course, lumped parameter (0D) model represents the cardiovascular system
and is written in Matlab code, however it can be implemented in the external numerical modeling application (CFD) in the future.
1
Introduction
The design and development of the medical equipment requires the precise and accurate numerical model reproducing the physical phenomena inside the vessel. Moreover computer analysis are less harmful then in-vivo methods and overall less expensive. Information about vulnerable zones can enhance diagnose. But the first step to generate specific models for an individual is to prepare model as a generalization of the average person.
To prepare any blood flow model a vast database is necessary; it has been understood as
a knowledge about properties of blood components, a description of interactions between them
and interactions between blood and vessels. Moreover blood flow model requires also transient
boundary conditions which are capable to describe the pulsatile blood flow.
1.1
The scope of the work
The work presents process of creating the lumped parameter model of the blood flow in the
cardiovascular system based on its electrical analog. LPM, including main characteristics of
blood and vascular tree is able to mimic response of the whole system on the analyzing section.
The heart rate, duration of the systolic and diastolic phases can be adopted to patient data.
The resistances and compliances of vessels must be assumed so the validation of the model is
dependent on further data. Model was prepared in the Matlab application.
1.2
Cardiovascular system
The heart, deformable vessels and blood are essential components of the cardiovascular system.
The greatest function of the system is the transport of oxygen, nutrients, metabolites and hormones and removal of the carbon dioxide, waste products and toxins. It also regulates the temperature and protects body from pathogens.
It is divided for pulmonary and systemic circuits. The function of the pulmonary part is to deliver blood to the lungs for exchange of carbon dioxide and oxygen. It is then returned into the
left part of the heart. The left chamber pumps the blood through the systemic circuit for circulation around the body so that oxygen can be used and harmful substances can be taken away
and then returns it to the right heart’s atrium. Blood is carried again to the lungs [4].
The heart is a two synchronized muscular pumps with four chambers. Atria collect the blood
from the veins and ventricles pump the blood to the arteries. The heart is equipped with four
valves – one per each cavity – which prevent blood from flowing backwards. When blood is
being pushed by contracting left ventricular the mitral valve is closed preventing blood
back – flow to atrium and the aortic valve is opened allowing blood to get to aorta. Valves in
the right part of the heart behave similarly. The ventricles are separated by a septum so oxygenated and de-oxygenated blood do not mix [4].
6
Each of the circuits begin with relatively high pressure with the pulsatile flow from the ventricles through a tree-like network of arteries. The size of cross-section of the vessels decrease
gradually from the aorta to the arteries then arterioles and finally they turn into the capillaries
which task is to exchange substances between blood and tissues. The mass transfer in this microcirculation includes, inter alia, transport though the intercellular gaps, active transport across
the membranes or via vesicles. Then blood with modified composition is collected and returned
to the heart at low pressure by veins. The pressure in vena cava equals 0.5 kPa so it is much
lower than pressure of blood in aorta, which is 12.5 kPa [5]. Thus the structure of vessels is
different, those vessels have to resist specific hydrostatic pressure, tension and shear stress.
Arteries carrying blood away from the heart have thick walls, thin veins carry blood to it and
unlike the arteries have valves. Moreover vessels respond to changes in pressure and flow rate
changing temporarily the diameter by smooth muscle cells. It can be described as a pulse wave
propagation and characterized by specific value of Reynolds number:
𝑅𝑒 =
𝐷𝑤
𝜗
(1.1)
where D is a diameter of vessel, m; w mean velocity of the blood, m∙s–1 and 𝜗 is its kinematic
viscosity, m2∙s–1.
Considering the blood viscosity Fahraeus – Lindquist effect should be mentioned. It turns out
that in the capillary the plasma create the layer covering the vessel wall and red cells stay in the
central part which helps erythrocytes to move and decrease the apparent viscosity. So blood
shows viscoelastic behavior which is a result of cells deformation and shear rate. In other words
it has non-Newtonian model, where the viscosity is not constant and depends on the strain rate.
In the large and medium sized vessels the Navier – Stokes equation can be applied.
Blood is suspension of particles in the matrix. Plasma which is the fluid component of the blood
represents 55% of its volume. The other parts are: red and white blood cells, platelets and inter
alia, electrolytes, small sugars, carbohydrates, lipids, proteins, hormone and others. All of them
interact with each other and with vessel’s wall. Blood transports most of the substances in the
body, removes heat and is responsible for acidic balance. Blood pressure depends on the oxygen
need. During exercises or stress heart pump it faster.
One normal cardiac cycle lasts about 0.8s and is divided for two stages: systolic and diastolic.
During systole the blood is ejected from the heart, during diastole the heart muscle is relaxing.
Both of this phases can be recorded as an electrical activity of the heart on the commonly performed cardiology tests which is electrocardiography. The results shows specific points:
P, R, Q, S, T per each rhythm [4].
7
Figure 1: Systolic and diastolic phases presented against ECG [4].
Stroke volume of the ejected blood is the difference between the end diastolic volume (which
means the volume remaining in the ventricle after diastole) and end systolic volume. The difference between pressure during systole and diastole, so called: pulse pressure, is usually given
in millimeters of mercury unit (mmHg).
During systole the pressure generated by heart creates the pressure wave that distend the arterial
walls due to maintenance a steady mean arterial pressure. Because of their elasticity, vessels
buffer the blood flow and hence the wave can be observed as a pulse. For children over 10 years
and adults normal pulse rate is 60–100 beats per min, for younger children: 70–130 [6].
1.3
Hydraulic analogy
The real cardiac cycle can be described as the hydraulic Windkessel Model developed by Otto
Frank in 1899 [8]. It consists a water pump and an air–filled chamber, which affect the velocity
and pressure of fluid in pipes. The arterial compliance is represented by compressible air in
the pocket of air-chamber (in German language – Windkessel).
The water is pumped to the chamber, compresses the air which pushes water out from the chamber, working against the resistance water encounters in the canals – just like arterial compliance.
This resistance refers to the peripheral resistance.
8
Figure 2: Hydraulic analogy of cardiovascular system. Figure made on the basis of [7].
The Windkessel model takes into account an unsteady flow, conservation of mass and assumes
that the water (as a blood) is incompressible.
Considering the section marked on the Figure 2. between point ‘inlet’ and ‘outlet’ during systole
(0 ≤ 𝑡 ≤ 𝑡𝑠 ),
where:
t is the current time, s; and ts is the systole duration, s; the relation between the water flow and
volume of the water that is stored in the chamber (windkessel) can be described as the principle
of the mass conservation:
𝑄(𝑡)𝑖𝑛 = 𝑄(𝑡)𝑜𝑢𝑡 + 𝑄(𝑡)𝑤 = 𝑄(𝑡)𝑜𝑢𝑡 +
𝑑𝑉
𝑑𝑡
(1.2)
where:
Q(t)in is a water volumetric flow as a function of time at the point ‘inlet’, m3∙s-1;
Q(t)out is a water volumetric flow at the point ‘outlet’, m3∙s-1; Q(t)w is a water volume inside the
𝑑𝑉
chamber, m3∙s-1; and 𝑑𝑡 is the time derivative of the volume, m3∙s-1. Moreover the Hagen –
Poiseuille law says that:
𝑄(𝑡) =
∆𝑃𝜋𝑟 4
8𝑙𝜂
(1.3)
And the resistance, Pa∙s∙m–3 can be calculated:
𝑅=
8𝑙𝜂
𝜋𝑟 4
(1.4)
9
where: ∆P is a pressure loss, Pa; r is the geometric factor depending on cross-section of the
canal, m; l is the length of the canal, m; 𝜂 is dynamic viscosity, Pa·s.
Therefore the flow in general equals:
𝑄(𝑡) =
𝑃(𝑡) 𝑑𝑉(𝑡) 𝑑𝑃(𝑡) 𝑃(𝑡)
𝑑𝑃(𝑡)
+
=
+𝐶
𝑅
𝑑𝑃(𝑡) 𝑑𝑡
𝑅
𝑑𝑡
(1.5)
where:
𝑑𝑉
𝐶 = 𝑑𝑃 is a constant ratio,
𝑚4 ∙𝑠
𝑔
. In case of cardiovascular system it is an arterial compliance
connected to the vasodilation which is the widening of elastic vessel [8].
1.4
Electrical analogy
In the project the relationship between pressure and volume in certain points is considering
therefore the model can be simplified into a lumped parameter model to get reliable results and
to decrease computational cost at the same time.
In zero-dimensional models the space is discretized and solution is based on differential equations.
In this approach the blood flow rate is presented by a current, and the pressure difference analogy by a voltage. The elements corresponding to cardiovascular parameters are presented in the
Table 1 where relations are presented to prove the validity of using the analogy.
They are simplified because of assumption of the linearity of the equations and of the laminar
blood flow [3].
Comparison of both analogies can be made with the same assumptions:
Figure 3: Hydraulic analog of the circulatory
system.
Figure 4: Electrical analog of the circulatory
system. Figure made on the basis of [5].
For the flow in the section presented above the governing equation of the system are:
𝐶𝑐
𝐿𝑐
𝑑𝑝̂
𝑑𝑡
𝑑𝑄̂
𝑑𝑡

𝐶𝑒
+ 𝑈2 − 𝑈1 = 0
(1.6)
+ 𝑅𝑄̂ + 𝑃2 − 𝑃1 = 0

𝐿𝑒 𝑑𝑡 + 𝑅𝐼̂ + 𝑈2 − 𝑈1 = 0
(1.7)
with the initial values:
Qi and Pi with i as 1 and 2 [5].
10
̂
𝑑𝑢
+ 𝑄2 − 𝑄1 = 0
𝑑𝑡
𝑑𝐼̂
Table 1: The summary of the corresponding equations in cardiovascular system and its electrical analog [3].
Parameters
of the real
model
Electrical
analog
The flow
resistance
Resistor
Vessel
compliance
Capacitor
The symbol
of the analog
Characteristic of the
cardiovascular system
Description of the
analog
∆𝑃
𝑅𝑐 =
𝑄
𝑈
𝐼
(the Ohm’s law)
𝑄 = 𝐶𝑐
𝑅𝑒 =
𝑑𝑃
𝑑𝑡
𝐼 = 𝐶𝑒
𝑑𝑄
𝑑𝑡
Relationship referring to
the Newton’s second
law
𝑑𝑈
𝑑𝑡
∆𝑃 = 𝐿𝑐
Blood
inertia
Coil
𝑄 = 0 𝑖𝑓 ∆𝑃 < ∆𝑃𝑐𝑟
The valve
position
∆𝑃
𝑄=
𝑖𝑓 ∆𝑃 ≥ ∆𝑃𝑐𝑟
𝑅𝑐
Diode
The valve position depends on the pressure
difference. If it is lower
than critical value the
valve is closed.
𝑈 = 𝐿𝑒
𝑑𝐼
𝑑𝑡
𝐼 = 0 𝑖𝑓 𝑈 < 𝑈𝑐𝑟
𝐼=
𝑈
𝑖𝑓 𝑈 ≥ 𝑈𝑐𝑟
𝑅𝑒
The diode behaves
similarly as it blocks
the current flow in
the wrong direction
due to the potential
difference.
𝑃 = 𝑃𝑚𝑖𝑛
Reference
pressure
Ground
Minimal pressure occurring in the cardiovascular system
𝑈=0
where: 𝑡 is the time, s; 𝑅 is the resistance, Ω; 𝑃 − pressure, 𝑄 − flow rate, 𝑈 – potential difference, V; 𝐼 − current flow, A; 𝐶𝑐 − compliance coefficient, 𝐶𝑒 − capacitance, F;
𝐿𝑐 − inertia coefficient, 𝐿𝑒 – inductance,Ω; c’ subscript refers to cardiovascular system,
‘e’ subscript refers to electrical analogy. The units are written only for the electrical analogue,
because hydraulic analogy was already mentioned.
11
The most often used electrical models are the variants of Windkessel Models [8]:
a) 2–element Windkessel Model
𝐼(𝑡) =
𝑃(𝑡)
𝑑𝑃(𝑡)
+𝐶
𝑅
𝑑𝑡
(1.8)
when 𝐼(𝑡) = 0 (during diastole):
−(𝑡)
𝑅𝐶
𝑃(𝑡) = 𝑃(𝑡𝑑 ) ∙ 𝑒
(1.9)
Resistor mimes the peripheral resistance of the systemic circuit.
Capacitor refers to the vessel compliance.
Figure 5: Two-element Windkessel Model. Figure made
on the basis of [8].
b) 3–element Windkessel Model
(1 +
𝑅1
𝑑𝐼(𝑡) 𝑃(𝑡)
𝑑𝑃(𝑡)
) 𝐼(𝑡) + 𝐶𝑅1
=
+𝐶
𝑅2
𝑑𝑡
𝑅2
𝑑𝑡
(1.10)
when 𝐼(𝑡) = 0 (during diastole):
𝑃(𝑡) = 𝑃(𝑡𝑑 ) ∙
−(𝑡)
𝑒 𝑅2 𝐶
(1.11)
Figure 6: Three element
Windkessel Model. Figure
made on the basis of [8].
Additional resistance is an analogue to the resistance occurring
due to the blood flow through the valve.
a) 4–element Windkessel Model
(1 +
𝑅1
𝐿 𝑑𝐼(𝑡)
𝑑2 𝐼(𝑡)
) 𝐼(𝑡) + (𝑅1 𝐶 + )
+ 𝐿𝐶
𝑅2
𝑅2 𝑑𝑡
𝑑𝑡 2
𝑃(𝑡𝑃
𝑑𝑃(𝑡)
=
+𝐶
𝑅2
𝑑𝑡
(1.12)
Figure 7: Four element
Windkessel Model. Figure
made on the basis of [8].
when 𝐼(𝑡) = 0 (during diastole):
−(𝑡)
𝑃(𝑡) = 𝑃(𝑡𝑑 ) ∙ 𝑒 𝑅2 𝐶
Inductor mimes the blood inertia.
12
(1.13)
2
Comparison between two- and three-element Windkessel Models and
analysis of the analytical and numerical approaches.
In the present work the comparison between two- and three-element Windkessel Models has
been done including the analytical and numerical approaches in case of two–element Windkessel Model.
The calculations using the electrical analogy base on the relations between the pressure and
flow. The outflows are read from the CFD model (what will be discussed later) however on the
inlets the function describing the ejection of the blood from the heart must be assumed.
2.1
Analytical solution of the two-element Windkessel Model [9]
Before calculating the value of the blood flow rate I(t), ml∙s–1, is important to define in which
phase the heart is at the current time t, s. According to the literature systole takes about
two–fifths of the cardiac cycle. Thus if the remainder of division t by cardiac period – Tc, s; is
2
smaller than 𝑇𝑠 = 5 𝑇𝑐 , 𝑠 it means that systole is in progress [7] and I(t) can be modeled as
an sinusoidal flow:
𝑡
𝐼(𝑡) = 𝐼0 𝑠𝑖𝑛 (𝜋 ∙ )
(2.1)
𝑇𝑠
Where I0 [ml] is a half of the amplitude of the sinusoidal function within the limits: 0 – Tc.
Blood flow during diastole had been assumed to be zero. Integral of I(t) represents cardiac output (CO).
𝑇𝑠
𝑡
𝐶𝑂 = ∫ 𝐼0 𝑠𝑖𝑛 (𝜋 ∙ ) 𝑑𝑡
(2.2)
𝑇𝑠
0
𝑇𝑠
𝐶𝑂 = 𝐼0 ∫ 𝑠𝑖𝑛 (𝜋 ∙
0
𝑡
) 𝑑𝑡
𝑇𝑠
(2.3)
If the heart rate and normal cardiac output are given and with assumptions that the cycle period
60
2
is 𝑇𝑐 = ℎ𝑒𝑎𝑟𝑡 𝑟𝑎𝑡𝑒 s and systole lasts 𝑇𝑠 = 5 ∙ 𝑇𝑐 𝑠, maximum value of sinusoidal function can
be counted:
𝐶𝑂
𝐼0 = 𝑇𝑠
𝑡
(2.4)
∫0 sin (𝜋 ∙ 𝑇 ) 𝑑𝑡
𝑠
𝐼0 =
𝐶𝑂
𝑇
𝑇
− 𝜋𝑠 𝑐𝑜𝑠(𝜋) + 𝜋𝑠 𝑐𝑜𝑠(0)
(2.5)
13
Windkessel first order ordinary differential equation:
𝑃(𝑡)
𝑑𝑃(𝑡)
+𝐶
= 𝐼(𝑡)
𝑅
𝑑𝑡
(2.6)
To find function of pressure Integrating Factor Method was used. The method was described
below (2.2.7 –2.2.13). W and Y are functions of the same variable
𝑎
𝑌
+𝑏∙𝑌 =𝑊
𝑑𝑡
(2.7)
With both sides divided by a:
𝑌 𝑏
𝑊
+ ∙𝑌 =
𝑑𝑡 𝑎
𝑎
(2.8)
Method involves defining the integrating factor:
𝑏
𝐼𝐹 = 𝑒 ∫𝑎𝑑𝑡
(2.9)
and multiplying every element of the equation by integrating factor:
𝐼𝐹 ∙
𝑊
𝑦𝑌
𝑏
= 𝐼𝐹
+ 𝐼𝐹 ∙ ∙ 𝑌
𝑎
𝑑𝑡
𝑎
(2.10)
Then equation should be integrated:
∫ 𝐼𝐹
𝑊
𝑑𝑌
𝑏
𝑑𝑡 = ∫ (𝐼𝐹 ∙
+ 𝐼𝐹 ∙ 𝑌) 𝑑𝑡
𝑎
𝑑𝑡
𝑎
(2.11)
taking into account the product rule :
𝑑
𝑌
𝐼𝐹
(𝐼𝐹 ∙ 𝑌) = 𝐼𝐹 + 𝑌
𝑑𝑡
𝑑𝑡
𝑑𝑡
𝑑
𝑑𝑡
(𝐼𝐹 ∙ 𝑌) = 𝐼𝐹
𝑑𝑌
𝑑𝑡
′
+ 𝐼𝐹 ∙ 𝑌 since: 𝐼𝐹 = (𝑒
𝑏
∫𝑎𝑑𝑥
)′ = 𝑒
𝑏
(2.12)
∫𝑎𝑑𝑥
General solution can be calculated by integrating:
𝑦 = 𝐼𝐹 −1 ∫
𝑊
𝐼𝐹𝑑𝑡
𝑎
(2.13)
Rearranged Windkessel equation analogically to (2.2.8) is:
𝑑𝑃(𝑡)
1
𝐼(𝑡)
+
𝑃(𝑡) =
𝑑𝑡
𝐶𝑅
𝐶
14
(2.14)
For sinusoidal function of the blood flow:
𝑑𝑃(𝑡) 𝑃(𝑡) 𝐼0
𝑡
+
= ∙ 𝑠𝑖𝑛 (𝜋 ∙ )
𝑑𝑡
𝐶𝑅
𝐶
𝑇𝑠
(2.15)
Analogically parameters a, b and integrating factor are:
𝑎=𝐶,
𝑏=
1
𝑅
1
𝑡
, 𝐼𝐹 = 𝑒 ∫𝑅𝐶𝑑𝑡 = 𝑒 𝑅𝐶
(2.16)
General solution for two-element Windkessel Model:
𝑡
𝑃(𝑡) = 𝑒 −𝑅𝐶 ∫
𝑡
𝐼0
𝜋∙𝑡
𝑠𝑖𝑛 (
) 𝑒 𝑅𝐶 𝑑𝑡
𝐶
𝑇𝑠
(2.17)
The integral was solved by parts and the solution can be presented as follows:
a) for systole:
𝑡
𝜋𝑡
𝜋𝑡
) − 𝑇𝑠 𝑠𝑖𝑛 ( ))
−𝑡
𝑇𝑠
𝑇𝑠
𝑅𝐶
+
𝑐
𝑒
1
2
𝜋 2 𝐶 2 𝑅 2 + 𝑇𝑠
𝐼0 𝑅𝑇𝑠 𝑒 𝑅𝐶 (𝜋𝐶𝑅 𝑐𝑜𝑠 (
𝑃(𝑡) = −
(2.18)
Integration constant was calculated for pressure at the beginning of the systole which equals to
the pressure at the end of the diastole from previous cycle. This value is assumed for initialization of calculations.
𝑐1 =
𝑡
𝑃𝑠𝑠 𝑒 𝑅𝐶
+
𝑡
𝜋𝑡
𝜋𝑡
𝑒 𝑅𝐶 𝐼0 𝑅𝑇𝑠 (𝜋𝐶𝑅 𝑐𝑜𝑠 ( 𝑇 ) − 𝑇𝑠 𝑠𝑖𝑛 ( 𝑇 ))
𝑠
𝑠
𝜋 2 𝐶 2 𝑅 2 + 𝑇𝑠 2
(2.19)
where Pss is pressure at the start of systole and tss is a time when systole starts so it equals 0 for
the first cycle. It gives us:
𝑐1 = 𝑃𝑠𝑠 +
𝐼0 𝑅𝑇𝑠 𝐶𝜋𝑅
𝜋 2 𝐶 2 𝑅 2 + 𝑇𝑠 2
(2.20)
b) for diastole:
𝑡
𝑃(𝑡) = 𝑒 −𝑅𝐶 ∫ 𝐼(𝑡)𝑑𝑡
(2.21)
Since the blood flow for this part of the cycle is assumed to be zero:
−𝑡
𝑃(𝑡) = 𝑐2 𝑒 𝑅𝐶
(2.22)
15
Pressure at the start of diastole equals the pressure at the end of systole so the constants c1 and
c2 are different.
The calculations for the three-element Windkessel model can be done using Integrating Factor:
𝑡
(2.23)
𝐼𝐹 = 𝑒 𝑅2 𝐶
where the peripheral resistance is described with the R2 symbol.
2.2
Numerical solution of the two-element Windkessel Model [10]
Numerical model of the inlet pressure was performed using the Runge –Kutta method for solving differential equations. This method involves the initial value and approximates solution for
each step using FSAL strategy (First Same as Last) and has a wide range of versions.
ODE23 is a version of Runge –Kutta method implemented in Matlab software. In the ODE23
algorithm, pressure is calculated iteratively, each step gives pressure for the time enlarged
by the time step (h). Step contains three stages.
Initial condition is the blood pressure at the start of systolic phase. Blood flow is calculated
using equation (2.2.1).
Derivative of pressure is counted:
𝑠1 = −
𝑃(𝑡) 𝐼(𝑡)
+
𝑅𝐶
𝐶
(2.24)
For the next stage with the time enlarged by half of the time step (h) pressure equals:
ℎ
ℎ
𝑃 (𝑡 + ) = 𝑃(𝑡) + 𝑠1
2
2
blood flow for the same time:
ℎ
𝐼 (𝑡 + 2) = 𝐼0 𝑠𝑖𝑛
(2.25)
ℎ
2
𝜋(𝑡+ )
𝑇𝑠
(2.26)
And it is used to count next derivative:
ℎ
ℎ
𝑃 (𝑡 + ) 𝐼 (𝑡 + )
2
2
𝑠2 = −
+
𝑅𝐶
𝐶
16
(2.27)
Next stage is calculated for the time enlarged by three fourth of the time step:
3
ℎ
3
𝑃 (𝑡 + ℎ) = 𝑃 (𝑡 + ) + ℎ ∙ 𝑠2
4
2
4
(2.28)
3
𝜋 (𝑡 + ℎ)
3
2
𝐼 (𝑡 + ℎ) = 𝐼0 𝑠𝑖𝑛
2
𝑇𝑠
(2.29)
3
3
𝑃 (𝑡 + ℎ) 𝐼 (𝑡 + ℎ)
4
2
𝑠3 = −
+
𝑅𝐶
𝐶
(2.30)
The pressure for next step equals:
ℎ
𝑃(𝑡 + ℎ) = 𝑃(𝑡) + (2𝑠1 + 3𝑠2 + 4𝑠3 )
9
Next derivative must be calculated:
𝑠4 = −
𝑃(𝑡+ℎ)
𝑅𝐶
+
(2.31)
𝐼(𝑡+ℎ)
𝐶
(2.32)
it is used in next step as s1 due to FSAL strategy [10].
The estimation error equals:
𝑒𝑛+1 =
ℎ
(−5𝑠1 + 6𝑠2 + 8𝑠3 − 9𝑠4 )
72
(2.33)
where n is the number of the step. The numerical calculations for three-element Windkessel
Model can be solved using the same method.
17
2.3
Model of the healthy heart
The model was prepared for the data used in the [9] and presented in the Table 2.
Table 2: Data used for model validation.
heart rate
72 beats per minute
cardiac output
90
duration of cardiac cycle
duration of systolic phase
𝑚𝑙
𝑐𝑦𝑐𝑙𝑒
60
5
of cardiac cycle s
𝑔
resistance
0.95
compliance
1.0666
𝑐𝑚4 ∙𝑠
2WM
3WM
𝑠
ℎ𝑒𝑎𝑟𝑡 𝑟𝑎𝑡𝑒
2
𝑐𝑚4 ∙𝑠
𝑔
𝑔
resistance 1
0.05
resistance 2
0.9
compliance
1.0666
𝑐𝑚4 ∙𝑠
𝑔
𝑐𝑚4 ∙𝑠
𝑐𝑚4 ∙𝑠
𝑔
where: 2WM and 3WM are respectively: two- and three-element Windkessel Model. With step
size: 0.01.
Pressure (mmHg)
Results for every used method discussed in the present chapter are close to those obtained by
the authors of the article [9] and are presented on the graph (Figure 8.). Figure presents one of
calculated cycle.
Numerical results for
2WM
Numerical results for
3WM
Analytical results for
2WM
Blood Pressure
140
130
120
110
100
90
80
70
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Time (s)
Figure 8: Results for analytical and numerical approaches to two- and three-element Windkessel
Models of the healthy cardiovascular system during first calculated cycle.
18
Percenage difference (%)
Comparison of the models
Percentage difference
between numerical
results for 2WM and
3WM
Percentage difference
between the analytical
and numerical results
for 2WM
8
6
4
2
0
0
-2
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Time (s)
Figure 9: Comparison of the results for analytical and numerical approaches to two- and threeelement Windkessel Models.
As it is presented on the Figure 9. the difference between numerical solutions for 2–element
and 3-element Windkessel Models are negligible. The absolute value of maximum percentage
difference is 1.51% at the time 0.83s (which is the end of the cycle). At the time 0.28s the
difference is 1.27%. On the beginning of diastole (0.34s of the cycle) the difference is only
0.16%.
Results for the diastolic phase (0.33s – 0.83s) calculated numerically and analytically for the 2–
elements Windkessel Model are close to each other. The absolute values of percentage differences are within the limits: 0.49% – 0.73%. For the systole those limits are higher:
0.89% – 8.38%. This difference reaches its peaks at 0.15s, which the half of the systole.
19
3
Case study
In the project simulations of the blood flow for eight years old girl is being made.
Girls weight was assumed to be 25 kg thus cardiac output (CO) can be calculated
according to [11]:
𝑚𝑙
𝑚𝑖𝑛
(3.1)
𝑚𝑙
𝑐𝑦𝑐𝑙𝑒
(3.2)
𝑐𝑜 = 933 ∙ 𝑤𝑒𝑖𝑔ℎ𝑡 0,38 ,
𝐶𝑂 = 𝑐𝑜 ∙
𝑇𝑐
= 36.9,
60
The heart rate has a value measured on the patient and equals 86 beats per minute [12].
With the assumptions [12] that:
𝑇𝑐 =
𝑇𝑠 =
60
= 0.7,
𝐻𝑅
𝑠
2
∙ 𝑇 = 0.28,
5 𝑐
(3.3)
𝑠
(3.4)
the maximum blood flow according to the equation (2.2.5) equals:
𝐶𝑂
𝐼0 =
−
𝑇𝑠
𝑇
𝑐𝑜𝑠(𝜋) + 𝑠 𝑐𝑜𝑠(0)
𝜋
𝜋
= 207.49,
𝑚𝑙
𝑠
(3.5)
Blood pressure at rest in the upper-body had been tested via a sphygmometer and for systole
was equal to 115 mmHg, for diastole to 65 mmHg [12].
High blood pressure is one of the consequences of the pathology. Using this specific data the
resistance and compliance of the circulatory system were chosen in order to obtain values of
pressure mentioned above.
3.1
Estimation of resistance and compliance coefficients
The choice of the R and C parameters was made using the Matlab script prepared for this purpose. Assumptions were done:
-the minimum values of R and C: 0.1,
-the maximum values of R and C: 1.9,
-the maximum tolerance: 5,
-the expected results for diastolic and systolic pressure: 65±5 mmHg and 115±5 mmHg.
20
Program chose the set of random values of R and C from the range: 0.1 to 1.9, and run the
algorithm based on Windkessel Models to compute the pressures. If the values of pressure
matched the chosen ranges, results were written into the table. 100000 iterations were calculated. Through this time-consuming process the set of R and C were obtained. For the calculations the following values were chosen:
𝑔
-
resistance = 1.812621 𝑐𝑚4 ∙𝑠,
-
compliance = 0.550099
𝑐𝑚4 ∙𝑠
𝑔
,
which provide the result for diastolic and systolic pressure respectively: 62.12 and 114.60
mmHg.
3.2
Results for the case study
Summarized data and assumptions that were used in computing the model of blood flow out of
the heart are presented in the table 3.
Table 3: Summarized blood flow and pressure data.
systolic pressure
114.60 mmHg
2WM
3WM
diastolic pressure
62.12 mmHg
heart rate
86 beats per minute
weight
25 kg
cardiac output
36.9 𝑐𝑦𝑐𝑙𝑒
duration of cardiac cycle
0.7 s
duration of systolic phase
0.28 s
maximum blood flow
207.49
resistance
1.812621
compliance
0.550099
𝑚𝑙
𝑚𝑙
𝑠
𝑔
𝑐𝑚4 ∙𝑠
𝑐𝑚4 ∙𝑠
𝑔
𝑔
resistance 1
0.05 𝑐𝑚4 ∙𝑠
resistance 2
1.812621 𝑐𝑚4 ∙𝑠
compliance
0.550099
𝑔
𝑐𝑚4 ∙𝑠
𝑔
where: 2WM and 3WM are respectively: two- and three-element Windkessel Model. With the
step size: 0.01.
21
Figure 10: Blood flow for the parameters presented in Table 1. Chart includes four cycles.
Figure 11: Results of the analizys of the case study.
Multivariate analyze included observation of pressure changes due to changes in heart rate.
According to [11] among the girls under the age of seventeen heart rate that occurs is between
52 and 136 beats per minute. In Table 4. the results for this range have been summarized.
22
Table 4: Maximum and minimum values of cardiac pressure in respect of changing pulse.
heart
bpm
rate, maximum value of minimum value of
pressure, mmHg
pressure, mmHg
50
131.34
58,77
60
122.55
60.8
70
115.77
61.95
80
110.42
62,71
90
106.05
63.16
100
102.47
63.55
110
99.45
63.75
120
96.87
63.93
130
94.62
64.1
140
92.68
64.26
systolic pressure, mmHg
The systolic pressure decreases exponentially with increase of heart rate. The opposite behavior
is observed in case of diastolic pressure. The changes are illustrated on the Figure 12 and
Figure 13.
Systolic pressure with respect to the heart rate
140
130
120
110
100
90
80
40
60
80
100
120
140
heart rate bpm
Figure 12: Changes of systolic pressure with heart rate increase.
23
diastolic pressure, mmHg
Diastolic pressure with respect to the heart rate
65
64
63
62
61
60
59
58
40
60
80
100
120
140
heart rate bpm
Figure 13: Changes of diastolic pressure with heart rate increase.
Coupling outlet boundary conditions for the model which will be discussed later is carried out
iteratively. The blood flow in domain is computed based on predetermined velocity inlet profile.
On the respective outflows the volumetric flows are read every time step. Using information
about the outflows and coupled Lumped Parameters Model the pressure is periodically calculated. When the results are reliable time step is finished and pressure becomes new boundary
condition.
Two – or three – element Wondkessel Model will be used. Parameters of the flow resistance
due to the blood flow through the valve (R1), the peripheral resistance (R2) and compliance of
the patient vessels are presented into the Table 5.
Parameters of the Windkessel Model on the outlets differ from those on the inlets, because they
represent distribution of the blood to the certain branches of the vascular system.
Table 5: Parameters of the Windkessel Models coupled on the outlets of the model [12].
Artery name
R1, 𝑘𝑔 ∙ 𝑚−4 ∙ 𝑠 −1 C, 𝑚4 ∙ 𝑠 2 ∙ 𝑘𝑔−1 R2, 𝑘𝑔 ∙ 𝑚−4 ∙ 𝑠 −1
Right Subclavian Artery
0.885 ∙ 108
7.13 ∙ 10−10
1.3858 ∙ 109
Right Common Carotid Artery
1.122 ∙ 108
5.62 ∙ 10−10
1.7571 ∙ 109
8
−10
Left Common Carotid Artery
1.124 ∙ 10
5.62 ∙ 10
1.7611 ∙ 109
Left Subclavian Artery
2.895 ∙ 108
2.12 ∙ 10−10
4.5361 ∙ 109
Descending Thoracic Aorta
0.271 ∙ 108
2.922 ∙ 10−9
0.3122 ∙ 109
24
Using LPM based on those data and blood flow read from the CFD model the boundary conditions are described.
4
Conclusions
The time – depended heart work was identified and modeled with the Lumped Parameter Model
which will serve as a boundary conditions of the complete CFD model in the future work. Obtained results are:
– the time dependent blood flow which can be used to prepare the velocity profile as
an inlet boundary condition,
– the functions describing the pressure of the blood ejected from the heart,
with the assumptions of heart rate and cardiac output. Moreover the model can be used with
another assumptions as it was presented
The analytical and numerical approaches were taken into the consideration.
According to the literature the blood pressure of a healthy adult should not exceed 135/85
mmHg (respectively systolic and diastolic pressure). For the numerical three – element Windkessel Model this rate is equal to: 132/78 mmHg. Hence the claim that the results are reliable.
Moreover three methods used to calculate the pressure give the similar results.
The case of eight years old girl with coarctation of the descending aorta was analyzed. The
assumed blood pressure was: 115/65 mmHg, what is a consequence of the pathology. Using
estimation of the model parameters close results were reached: 114.6/62.1 mmHg.
The work shows that Lumped Parameter Model based on the Windkessel analogue is sufficient
tool, give the reliable results and make investigations of the cardiovascular system more precise.
Future work
Figure below presents the geometry of three-dimensional model of aorta with indicated places
of the computational domain where boundary conditions have been attached. The data had been
taken from the Gadolinium-enhanced MR angiography. Model includes: ascending aorta which
starts with ‘inlet’ (without coronary artery), aorta arch attached with upper branches:
– brachiocephalic artery – 1th outlet (it is a common branch for right subclavian artery
and right common carotid artery),
– left common carotid artery - 2nd outlet,
– left subclavian artery – 3th outlet,
and part of descending thoracic aorta (forth outlet) where narrowing called ‘coarctation’ (approximately 65% aortic area reduction). The geometry is presented on the Figure 14.
25
Figure 14: The geometry of the 3D model with indicated boundary conditions as a:
a) three-element Windkessel Model, b) two-element Windkessel model. Figure made on the
basis of [13].
Acknowledgments
This research is supported by National Science Centre (Poland) within project
No 2014/13/B/ST8/04225. This help is gratefully acknowledged herewith.
I would like to express sincere gratitude to Dr Marek Rojczyk for the invaluable contribution
to this work and tremendous support.
26
References
[1] Eric J. Topol, Robert M. Califf, Textbook of Cardiovascular Medicine, Lippincott Williams
& Wilkins, 2007
[2] Marwan A. Simaan, George Faragallah, Yu Wang and Eduardo Divo. Left Ventricular Assist
Devices: Engineering Design Considerations, New Aspects of Ventricular Assist Devices,
Dr. Guillermo Reyes (Ed.), 2011
[3] Hassanain Ali Lefta Mossa, Engineering Modeling of Human Cardiovascular System,
The 1st Regional Conference of Eng. Sci. NUCEJ Spatial ISSUE vol.11, No.2, pp 307-314,
2008
[4]Red. Jaroszyk F, Biofizyka. Podręcznik dla studentów, Wydawnictwo Lekarskie PZWL,
ISBN:9788320036763, 2013
[5] Formaggia L., Quarteroni A., Veneziani A., Cardiovascular Mathematics. Modeling and
simulation of the circulatory system, Springer-Verlag Italia, Milano, 2009
[6] Wikipedia The Free Encyclopedia, Pulse [online] [access 2.02.2016] https://en.wikipedia.org/wiki/Pulse
[7] Tracey Morland , MBW: Womersley Arterial Flow, [online] [access 2.02.2016]
https://mathbio.colorado.edu/index.php/MBW:Womersley_Arterial_Flow#cite_notewindkessel-7
[8] Kerner D., Solving Windkessel Models with MLAB, [online] [access: 2.02.2016]
http://www.civilized.com/mlabexamples/windkesmodel.htmld/
http://www.springer.com/us/book/9788847011519
[9] Catanho M., Sinha M. Vijayan V., Model of Aortic Blood Flow Using the Windkessel Effect,
2012
[online]
[access
2.02.2016]
http://isn.ucsd.edu/classes/beng221/problems/2012/BENG221_Project%20-%20Catanho%20Sinha%20Vijayan.pdf
[10] Cleve Corner, Ordinary Differential Equation Solvers ODE23 and ODE45, [online]
http://blogs.mathworks.com/cleve/2014/05/26/ordinary-differential-equation-solvers-ode23and-ode45/
[11] Giovanni de Simone, MD; Richard B. Devereux, MD; Stephen R. Daniels, MD, PhD;
GianFrancesco Mureddu, MD; Mary J. Roman, MD; Thomas R. Kimball, MD; Rosanna Greco,
MD; Sandra Witt, RDCS; Franco Contaldo, MD, Stroke Volume and Cardiac Output in Normotensive Children and Adults, Assessment of Relations With Body Size and Impact of Overweight, 1996
[12] Figueroa A., Sharma P., Mansi T., Wilson N., CFD Challenge: Simulation of Hemodynamics
in
a
Patient-Specific
Aortic
Coarctation
Model,
2012 [online]
[access: 14.02.2016 ] http://www.vascularmodel.org/miccai2012/miccai-2012-cfd-challengedescription.pdf
[13] Rojczyk M., Ostrowski Z., Adamczyk W., Melka Bartłomiej, Bandoła D., Gracka M.,
Knopek A., Nowak A. J.,Gołda A., CFD analysis of blood flow within aorta of patient with
coarctation of aorta, 2016
27
Identyfikacja i modelowanie pulsacyjnego przepływu krwi w
systemie krwionośnym przy użyciu zero–wymiarowego modelu w
analogii elektrycznej.
Dominika Bandoła
Słowa kluczowe: Model Windkessel, warunki brzegowe, ciśnienie tętnicze, przepływ krwi,
aorta
Streszczenie
Zaawansowane komputerowe narzędzia wspomagania inżynierskiego, jak numeryczna mechanika płynów (CFD) oraz wiedza na temat funkcjonowania układu krążenia człowieka, struktury
krwi, zachowania naczyń krwionośnych pozwalają na lepsze zrozumienie procesu dystrybucji
krwi po organizmie. Kompleksowe symulacje muszą zakładać wielofazowość przepływu krwi,
elastyczne ściany naczyń krwionośnych oraz pulsacyjny przepływ wynikający z charakterystyki pracy serca [1].
W prezentowanej pracy zamodelowano przepływ oraz ciśnienie krwi.
Charakterystyczna impedancja, rezystancja oraz pojemność zostały wzięte pod uwagę
w elektrycznej analogii jako model skupiony dużego układu krążenia, który zostanie zaimplementowany jako warunek brzegowy w kompletnym modelu CFD na wlocie do aorty wstępującej oraz na wylotach z pnia ramienno-głowowego, tętnicy szyjnej wspólnej lewej
i tętnicy podobojczykowej lewej.
Opór przepływu krwi w naczyniach jest analogiczny do oporu elektrycznego rezystora. Niestacjonarny przepływ regulowany przez elastyczne naczynia krwionośne można zasymulować
używając kondensatora. Za pomocą cewki można opisać inercję krwi.
Co więcej, zakładając, ze krew płynie tylko w jednym kierunku mimo zmiennego ciśnienia
dzięki działaniu zastawek, w modelu można użyć w ich miejsce diod, które podobnie działają
na przepływ prądu. Dodatkowo analogia elektryczna umożliwia zastosowanie zasady zachowania masy dzięki prawu Kirchhoffa [3].
Powstały model skupiony opisujący układ krążenia został napisany w programie Matlab, jednakże może być przepisany na język programowania C i zaimplementowany w zewnętrznym
oprogramowaniu CFD.
28
c
Instytut
Techniki Cieplnej, Politechnika Ślaska
˛
Archiwum Instytutu Techniki Cieplnej Vol. 1 nr 1(2016)
www.itc.polsl.pl/ArchiwumITC
Multi-fluid Euler-Euler model of the blood flow within the blood
vessel with rigid walls
Maria Gracka*†
Key words: CFD, Euler-Euler model, multi-fluid, blood flow, aorta
Abstract
The cardiovascular diseases and disorders such as atherosclerosis, strokes and heart attacks
are the leading causes of death in the world mainly in the developed and industrialized
societies. Understanding of basic mechanisms and phenomena occurring in the cardiovascular
system could be useful in early detection of the development of lesions in blood vessels. In
the presented work numerical analysis of blood flow within aorta has been made. Analysis
included two models: single- and multi-phase approaches. In first blood was treated as a
homogenous, non-Newtonian fluid with averaged rheological properties of viscosity and
density. Second model uses an Eulerian multiphase approach in model of blood flow which
assumes blood as a mixture of three phases (plasma, erythrocytes and leukocytes). To develop
numerical model of blood flow within the human blood vessel the commercial software
ANSYS Fluent (ANSYS Inc., USA) has been used. In the project the geometry of 8-year old
patient with moderate thoracic aortic coarctation (approximately 65% area reduction) has
been used. The geometry was created from data generated during Gadolinium-enhanced MR
angiography (MRA). Model of the geometry includes ascending aorta, arch, descending aorta
und upper branches such as innominate artery, left common artery, left subclavian artery. To
reproduce periodic cardiac cycle as an inlet boundary condition velocity profile was used.
Velocity profile was calculated from the conversion of the volumetric flow that was measured
by a phase-contrast (PC) MRI sequence with through-plane velocity encoding. As the outlet
boundary conditions for all branches and descending aorta the outflow condition was used.
Volumetric share of blood flow through various outflows of the aortic model was measured
via PC-MRI. Results of numerical simulation are presented for two characteristic points
(during systole and diastole) of the cardiac cycle.
*
This chapter was prepared during master thesis project at the Institute of Thermal Technology, Department
of Energy and Environmental Engineering of the Silesian University of Technology, under the supervision of
Dr Ziemowit Ostrowski.
†
This research is supported by National Science Centre, Poland, within project No 2014/13/B/ST8/04225. This
help is gratefully acknowledged herewith.
Zawartość tej publikacji może byc wykorzystana na warunkach licencji Uznanie Autorstwa 3.0 Polska. Licencja pozwala na
kopiowanie, zmienianie, rozprowadzanie, przedstawianie i wykonywanie utworu jedynie pod warunkiem oznaczenia autorstwa: autora,
tytułu rozdziału, nazwy serii, tomu, strony. Content from this work may be used under the terms of the Creative Commons Attribution
3.0 licence. Any further distribution of this work must maintain attribution to the author, chapter title, series title, volume, pages.
29
1. Introduction
The cardiovascular diseases and disorders such as atherosclerosis, strokes and heart
attacks are the leading causes of death in the world mainly in the developed and industrialized
societies. Understanding of basic mechanisms and phenomena occurring in the cardiovascular
system could be useful in early detection of the development of lesions in blood vessels,
where pharmacologic treatment can be used instead of expensive and complicated surgical
procedures [12]. Computer models offer many advantages over experimental work therefore,
in recent years Computational Fluid Dynamics (CFD) methods [14], widely used in various
engineering applications, have become a tool and proved its applicability in the biomedical
flows modeling. Computer models can be used to perform virtual experiments that can give
answers related to flow conditions in human body. Increasingly numerical modeling is used in
the design of medical devices such as stents, artificial hearts and heart valves. The accurate
velocity and pressure fields obtained in simulations could help in early medical diagnosis for
instance to indicate on zones vulnerable to cardiac diseases.
1.1 The scope of the work
The scope of the present work shows an Eulerian multiphase approach in model of
blood flow in human blood vessel (artery) section. Proposed numerical model assumes blood
flow in rigid vessels and blood properties as a non-homogeneous fluid which consists of
blood plasma, red blood cells (erythrocytes) and white blood cells (leukocytes). Multi-fluid
Euler-Euler technique [8] allows to model the flow behavior for each of blood components
and treat each phase as a separate interpenetrating continua with defined density, volume
fraction and other physical properties [3].
1.2 ANSYS Fluent
To develop numerical model of blood flow within the human blood vessel the ANSYS
Fluent (ANSYS Inc., USA), commercial CFD software has been used. The software allows
modeling of all phenomena associated with the flows (combustion, turbulence, multiphase
flows, chemical reactions, heat transfer, radiation and as well as flow condition in human
body) and preparing geometry and mesh corresponding to the simulated phenomenon.
The numerical model of human artery and multi-fluid blood flow was prepared using
ANSYS Fluent v.17, while flow equations of pulsatory boundary conditions were
implemented by using User Defined Function (UDF). This approach covers the influence of
the human arterial system and allows to reproduce periodical cardiac cycles that determinate
the pulsating (transient) blood flow conditions.
2. Geometrical model
2.1 Geometry
In presented work the geometry of 8-year old patient with moderate thoracic aortic
coarctation (approximately 65% area reduction) has been used. The geometry was created
from data generated during Gadolinium-enhanced MR angiography (MRA) using 1.5 GE
30
Sigma scanner [17]. Figure 1 shows model of the aorta that includes ascending aorta (AAO),
arch, descending aorta (DAO) und upper branches such as innominate artery (IA), left
common artery (LCCA), left subclavian artery(LSA).
Figure 1: The geometry of the aorta with selected inlet and outlets used in CFD model
2.2 Mesh
In the modeling of phenomena characterized by a high irregularity significant impact
on the accuracy, computation time and convergence of the solution has a number of numerical
mesh divisions of the considered geometry. Mesh generated in the project is shown in Fig. 2
and Fig. 3 consists about 750 000 of hybrids elements. In regions such as inlet and outlets the
structural mesh has been created with hexahedral elements to achieve exact inlet velocity
profile and stabilized flow. In part of aorta arch unstructured mesh with tetrahedral element
has been used.
31
Figure 2: Generated hybrid mesh
Figure 3: Generated mesh in the inlet of the
domain
Before the start of the numerical simulation of the blood flow the quality of mesh has
been verified. Elements of mesh with poor quality has been improved. Figure 4 shows the
percentage of cells with a specific orthogonal quality.
Figure 4: Histogram of the percentage of cells with a specific orthogonal quality
32
3. Numerical model
3.1 Model description
In presented work a comparison of numerical models of blood flow in human artery
has been made. The blood was treated as a multiphase fluid that consists plasma, RBC (red
blood cells) and WBC (white blood cells) and also, as alternative, the blood was treated as a
single phase non-Newtonian fluid.
Multiphase model
For multiphase model the principle of mass conservation equation for fluid phase
(plasma) is presented by Eq. (1) while for each solid phase (RBC and WBC) by Eq. (2). Both
mass conservation equations are equal zero because mechanism of mass exchange between
phases has not been used in current model, as well as no mass source are present. In the
formulas presented below index f describes plasma as a fluid phase and s describes
erythrocytes and leukocytes as solid phases [16]:
𝜕(𝜌𝑓 𝜀𝑓 )
+ ∇ ∙ (𝜌𝑓 𝜀𝑓 ⃗⃗⃗⃗
𝑣𝑓 ) = 0
𝜕𝑡
𝜕(𝜌𝑠 𝜀𝑠 )
+ ∇ ∙ (𝜌𝑠 𝜀𝑠 ⃗⃗⃗
𝑣𝑠 ) = 0
𝜕𝑡
(1)
(2)
where:
ρ - density, ε - volume fraction of i-phase, 𝑣- interphase velocity.
The sum of volume fraction for each phase is presented by Eq. (3):
𝜀𝑓 + 𝜀𝑊𝐵𝐶 + 𝜀𝑅𝐵𝐶 = 1
(3)
The equations for conservation of momentum is respectively written for fluid phase
(4) and applied twice separately for each solid phases RBC (5) and WBC (6):
𝑁𝑠
𝜕(𝜌𝑓 𝜀𝑓 ⃗⃗⃗⃗
𝑣𝑓 )
+ ∇ ∙ (𝜌𝑓 𝜀𝑓 ⃗⃗⃗⃗
𝑣𝑓 ⃗⃗⃗⃗
𝑣𝑓 ) = 𝜌𝑓 𝜀𝑓 𝑔 − 𝜀𝑓 ∇𝑃 + ∇ ∙ ⃗⃗⃗
𝜏𝑓 + ∑(𝐾𝑠𝑓 (𝑣
⃗⃗⃗
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 ))
𝜕𝑡
𝑠=1
𝜕(𝜌𝑅𝐵𝐶 𝜀𝑅𝐵𝐶 𝑣𝑅𝐵𝐶 )
+ ∇ ∙ (𝜌𝑅𝐵𝐶 𝜀𝑅𝐵𝐶 𝑣𝑅𝐵𝐶 𝑣𝑅𝐵𝐶 )
𝜕𝑡
= 𝜌𝑅𝐵𝐶 𝜀𝑅𝐵𝐶 𝑔 − 𝜀𝑅𝐵𝐶 ∇𝑃 − ∇𝑃𝑅𝐵𝐶 + ∇ ∙ 𝜏𝑅𝐵𝐶 + 𝐾𝑓,𝑅𝐵𝐶 (𝑣
⃗⃗⃗⃗𝑓 − 𝑣𝑅𝐵𝐶 )
+ (𝐾𝑊𝐵𝐶,𝑅𝐵𝐶 (𝑣𝑊𝐵𝐶 − 𝑣𝑅𝐵𝐶 ))
(4)
(5)
𝜕(𝜌𝑊𝐵𝐶 𝜀𝑊𝐵𝐶 𝑣𝑊𝐵𝐶 )
+ ∇ ∙ (𝜌𝑊𝐵𝐶 𝜀𝑊𝐵𝐶 𝑣𝑊𝐵𝐶 𝑣𝑊𝐵𝐶 )
𝜕𝑡
= 𝜌𝑊𝐵𝐶 𝜀𝑊𝐵𝐶 𝑔 − 𝜀𝑊𝐵𝐶 ∇𝑃 − ∇𝑃𝑊𝐵𝐶 + ∇ ∙ 𝜏𝑊𝐵𝐶 + 𝐾𝑓,𝑊𝐵𝐶 (𝑣
⃗⃗⃗⃗𝑓 − 𝑣𝑊𝐵𝐶 )
+ (𝐾𝑅𝐵𝐶,𝑊𝐵𝐶 (𝑣𝑅𝐵𝐶 − 𝑣𝑊𝐵𝐶 ))
(6)
where:
index q describes RBC or WBC and depends on configuration (while q≠s), Ns - number of
solid phase (Ns=2), P- pressure of fluid, Ps - granular pressure, 𝑔 - gravitational acceleration,
𝜏 - stress tensor, K - interphase exchange momentum coefficient
33
The momentum exchange coefficient between plasma and RBC was calculated using
the Gidaspow model [4] which is a combination of two equations and is recommended for
dense flows. The Gidaspow model uses Wen and Yu correlation (6) for low solid volume
fraction (εs<0,2) and switches to Ergun's law (8) for flow in a porous medium for larger solid
volume fraction [15,2].
When εf>0,8 the fluid-solid momentum exchange coefficient K is of the following form:
𝐾=
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 | −2,65
3 𝜀𝑠 𝜀𝑓 𝜌𝑓 |𝑣
𝐶𝐷
𝜀𝑓
4
𝑑𝑠
(7)
24
[1 + 0,15(𝜀𝑓 𝑅𝑒𝑠 )0,687 ]
𝜀𝑓 𝑅𝑒𝑠
(8)
where:
𝐶𝐷 =
when εf≤0,8:
𝐾 = 150
𝜀𝑠 (1 − 𝜀𝑓 )𝜇𝑓
𝜌𝑓 𝜀𝑓 |𝑣
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 |
+ 1,75
2
𝜀𝑓 𝑑𝑠
𝑑𝑠
(9)
The momentum exchange coefficient between plasma and WBC was calculated using
Wen and Yu model (6) which is applicable for dilute flows where the total secondary phase
volume fraction is significantly lower than that of the primary phase. To describe solid-solid
interactions the symmetric model has been used. For this model viscosity and density are
calculated from volume averaged properties [13]:
𝜌𝑝𝑞 = 𝜀𝑝 𝜌𝑝 + 𝜀𝑞 𝜌𝑞
(10)
𝜇𝑝𝑞 = 𝜀𝑝 𝜇𝑝 + 𝜀𝑞 𝜇𝑞
(11)
and the diameter is defined as:
1
𝑑𝑝𝑞 = (𝑑𝑝 + 𝑑𝑞 )
2
Then the exchange coefficient is calculated from the Eq. (12):
𝐾=
𝜌𝑝𝑞 𝑓
𝑑 𝐴
6𝜏𝑝𝑞 𝑝 𝑖
(12)
(13)
where:
𝜏𝑝𝑞
𝜌𝑝𝑞 (𝑑𝑝𝑞 )
=
18𝜇𝑝𝑞
2
(14)
and
𝐶𝐷 𝑅𝑒
18
In turn Re is defined by Eq. (15) and CD by Eq. (16):
𝑓=
34
(15)
𝜌𝑝𝑞 |𝑣
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 |𝑑𝑝𝑞
𝜇𝑝𝑞
0,687
)/𝑅𝑒
24(1 + 0,15𝑅𝑒
𝑅𝑒 ≤ 1000
𝐶𝐷 = {
𝑅𝑒 =
0,44
𝑅𝑒 > 1000
(16)
(17)
Model of turbulence used during simulation was standard k-ε and has been calculated
per phase [1]. Kinetic energy and its rate of dissipation are obtained from the following
transport equations [7]:
𝜕
𝜕
𝜕
𝜇𝑡 𝜕𝑘
(𝜌𝑘) +
(𝜌𝑘𝑢𝑖 ) =
[(𝜇 + )
] + 𝐺𝑘 + 𝐺𝑏 − 𝜌𝜀 − 𝑌𝑀 + 𝑆𝑘
𝜕𝑡
𝜕𝑥𝑖
𝜕𝑥𝑗
𝜎𝑘 𝜕𝑥𝑗
(18)
and
𝜕
𝜕
𝜕
𝜇𝑡 𝜕𝜀
𝜀
𝜀2
(𝜌𝜀) +
(𝜌𝜀𝑢𝑖 ) =
(19)
[(𝜇 + )
] + 𝐶1𝜀 (𝐺𝑘 + 𝐶3𝜀 𝐺𝑏 ) − 𝐶2𝜀 𝜌 + 𝑆𝜀
𝜕𝑡
𝜕𝑥𝑖
𝜕𝑥𝑗
𝜎𝜀 𝜕𝑥𝑗
𝑘
𝑘
where:
Gk represents the generation of turbulence kinetic energy due to the mean velocity gradients,
Gb is the generation of turbulence kinetic energy due to buoyancy, YM represents the
contribution of the fluctuating dilatation in compressible turbulence to the overall dissipation
rate. 𝐶1𝜀 , 𝐶2𝜀 and 𝐶3𝜀 are constants.
For red blood cells Partial Differential Equation as a granular temperature model was
used [16]. Granular temperature for the sthsolid phase is proportional to the kinetic energy of
the particles random motion and is calculated from the Eq. (19):
1
𝜃𝑠 = 𝑢𝑠,𝑖 𝑢𝑠,𝑖
3
(20)
where:
us,i represents the ith component of the fluctuating solids velocity in the Cartesian coordinate
system.
While the transport equation that are derived from kinetic theory takes the form as Eq. (20):
3 𝜕
[ (𝜌 𝛼 𝜃 ) + ∇ ∙ (𝜌𝑠 𝛼𝑠 ⃗⃗⃗
𝑣𝑠 𝜃𝑠 )] = (−𝑝𝑠 𝐼 ̿ + 𝜏̿𝑠 ): ∇𝑣
⃗⃗⃗𝑠 + ∇ ∙ (𝑘𝜃𝑠 ∇𝜃𝑠 ) − 𝛾𝜃𝑠 + 𝜑𝑙𝑠
2 𝜕𝑡 𝑠 𝑠 𝑠
where:
(−𝑝𝑠 𝐼 ̿ + 𝜏̿𝑠 ): ∇𝑣
⃗⃗⃗𝑠 = generation of energy by the solid stress tensor,
𝑘𝜃𝑠 ∇𝜃𝑠 = the diffusion of energy (𝑘𝜃𝑠 is the diffusion coefficient)
𝛾𝜃𝑠 = the collisional dissipation of energy [10]
𝜑𝑙𝑠 = the energy exchange between the lth fluid or solid phase and the sth solid phase
(21)
Single phase model
In one phase model the principle of mass conservation as well as momentum
conservation have been calculated as in equations above Eq. (1) and Eq. (4) in multiphase
35
model for fluid phase and also model standard k-ε for turbulence model has been chosen.
Blood was treated as a non-Newtonian fluid. Density of blood was calculated as a weighted
average of density values used for each phase of multi-fluid model.
𝜌𝑏 = 𝜀𝑝 𝜌𝑝 + 𝜀𝑅𝐵𝐶 𝜌𝑅𝐵𝐶 + 𝜀𝑊𝐵𝐶 𝜌𝑊𝐵𝐶
(22)
3.2 Blood as a non-Newtonian fluid
Considering blood as a non-Newtonian fluid assumes that viscosity is not constant
under isobaric conditions but its value changes over time. To calculate changing value of
blood viscosity the Share-Rate-Dependant Method has been used. Non-Newtonian fluid flow
is modeled according to the following power law for the non-Newtonian viscosity [16]:
𝜂 = 𝐾𝛾̇ 𝑛−1 𝐻(𝑇)
(23)
where:
K is the consistency factor, λ is the natural time or the reciprocal of a reference shear rate, n is
the power-law index, H(T) is the temperature dependence, known as the Arrhenius law and is
calculated from Eq. (23):
1
1
𝐻(𝑇) = 𝑒𝑥𝑝 [𝛼 (
−
)]
𝑇 − 𝑇0 𝑇𝛼 − 𝑇0
(24)
where:
α is the ratio of the activation energy to the thermodynamic constant, Tα is a reference
temperature for which H(T)=1, T0 which is the temperature shift and is set to 0 (corresponds
to the lowest temperature that id thermodynamically acceptable).
3.3 Boundary conditions
Particularly significant in such a detailed numerical simulation of the blood flow is to
implement appropriate boundary conditions at the inlet of the computational domain that will
be able to reproduce periodical cardiac cycle. At the inlet of the aorta the velocity inlet was
used. Velocity profile was calculated from the conversion of the volumetric blood flow which
was measured by a phase-contrast (PC) MRI sequence with through-plane velocity encoding
[17]. During the procedure the measurement of the volumetric blood flow has been made 20
times. Twenty measuring points were plotted on the graph and the linear interpolation
between them has been added. Figure 5 shows an ascending aortic volumetric flow profiles
for all blood phases. Two black vertical lines on the Fig. 5 are representations of
characteristic points (during systole and diastole) of the cardiac cycle.
36
Figure 5: Volumetric blood flow at the inlet of the aorta with share of the specific phases
Pulsating and repeated boundary condition at the inlet of the computational domain
was implemented by using User Defined Functions (UDFs). Profile of blood velocity is
shown in Fig. 6.
1
0.9
0.8
inlet velocity, m/s
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
time, s
Figure 6: Velocity profile at the inlet of aorta
37
As the outlet boundary conditions for all branches and descending aorta the outflow
condition was used. Volumetric share of blood flow through various outflows of the aortic
model was measured via PC-MRI. Measurements based on [17] are presented in Table 1.
Table 1: The volumetric share of blood flow on the outlets
Location
IA
LCCA
LSA
DAO
Ascending aortic
flow, %
25,6
11,3
4,3
58,8
Other important element of the blood flow simulation is to defined appropriate
material properties of specific phases. The characteristic parameters (such as density [6],
volume fraction and viscosity [11]) used in multiphase numerical model are shown in Table 2.
It contains also coefficients describing wall behavior such as restitution coefficient that
defines the amount if momentum that is retained by the particle after the collision with the
boundary and specularity coefficient that is a measure of the fraction of collisions which
transfer momentum to the wall [16].
Table 2: The characteristic parameters of plasma, RBC and WBC [5]
Plasma density, kg/m3
Plasma viscosity, kg/(m∙s)
RBC density, kg/m3
WBC density, kg/m3
Inlet RBC volumetric fraction, %
Inlet WBC volumetric fraction, %
1020
0,0035
1090
1050
44
1
Restitution coefficient
Specularity coefficient
Time step size, s
Operating pressure, Pa
0,99999
0,6
0,01
101325
Table 3 presents characteristic parameters used in single phase numerical model where
blood was treated as a non-Newtonian fluid [9]. Flow rates through the upper branch vessels
as well as the descending aorta are shown in Table 1.
Table 3: Characteristic parameters of blood in single phase model [9]
Blood density, kg/m3
Consistency index, kg∙sn-2/m
Power-Law index, n
38
1051
16
0,63
4. Results
Received results present relative values of static pressure that are referred to the
pressure at the inlet of the ascending aorta where the operating point was located. Possible
obtained negative pressure values are relative to the adopted operating pressure outside the
domain. Obtained results for single-phase and three-phase model are shown for two
characteristic points of the cardiac cycle during systole (0,14s) and diastole (0,62s) as shown
in Fig. 5.
Single-phase model
The pressure fields on the wall of the domain for both characteristic points for singlephase model are shown in Fig. 7. During systole the highest relative pressure occurred in the
area of the inlet of the domain and it reached about 840Pa while the lowest static pressure can
be noticed in the area of the coarctation of the aorta and reached about -12.4kPa. During
diastole pressure at the inlet reached 4Pa and -900Pa in the stenosis of the aorta. Difference
between pressure at inlet of the ascending aorta and outlet of the descending aorta was about
6kPa during systole and 80Pa during diastole.
39
Figure 7: Comparison of pressure fields on walls during characteristic points of systole (left)
and diastole (right) for single-phase model
40
Figure 8 shows pressure field of aortic arch with upper branches where appeared the
greatest pressure changes. During systole the highest pressure values are obtained at the inlet
of the domain and also at the inlet to the branches. During the diastole pressure changes are
negligible.
Figure 8: Comparison of pressure fields on arch of the aorta during characteristic points of
systole (left) and diastole (right) for single-phase model
Wall shear stress caused by the blood flow on the rigid walls of human vessel reached
the highest value of about 390Pa during systole at the bottom part of the aortic arch. High
value of wall shear stress are directly related to the highest values of the velocity which are
located in the same area, see Fig. 10. During diastole wall shear stress caused by blood flow
was close to 2Pa on the entire surface of blood vessel. Obtained results for single-phase model
are shown in Fig. 9.
41
Figure 9: Comparison of wall shear stress caused by blood flow during systole (left) and
diastole (right) for single-phase model
Comparison of the velocity vectors in blood vessel during systole and diastole are
shown in Fig. 10. The highest velocity values appeared in the area of the aortic coarctation
and on the opposite wall to the subclavian artery bifurcation on the arch of the aorta. The
maximum velocity values reached almost 5m/s. During systole blood flow was mostly
laminar the only eddies were located at the end of the aortic arch and near the pathological
narrowing. The carried out analysis of the velocity field showed that blood flowed to the
specific outflows according to the implemented boundary conditions.
42
During diastole velocity of blood and its volumetric flow were close to zero
throughout the vessel. Blood flow was turbulent and highly irregular it is caused by the low
pressure gradient between inlet and outlets of the domain, a lot of eddies are clearly visible.
The reversed flow appeared in every outflow. The highest velocity values during diastole
reached 0.13m/s in the region before coarctation.
Figure 10: Comparison of velocity vectors in characteristic points during systole (left) and
diastole (right) for single-phase model
Multi-phase model
Analysis of the multi-phase model shows that comparing to the single-phase model
difference between pressure at the inlet to the ascending aorta and outlet of the descending
aorta for mixture is lower and reaches about 4.9kPa (systole) and 30Pa during diastole. The
highest static pressure of about 260Pa appeared at the inlet of the aorta and the -13.4kPa in the
region of the aortic coarctation. According to the pressure profile for the characteristic point
43
of diastole the maximum pressure occurred at the outlet of the geometry and reached 26Pa
while pressure at the inlet was about -4Pa. Pressure fields during systole and diastole are
presented in Fig. 11.
Figure 11: Comparison of pressure fields on walls during characteristic points of systole (left)
and diastole (right) for multi-phase model
In multi-phase model likewise in single-phase model the largest pressure changes
occur at the arch of the aorta and at the inlets to branches during the systole. While during the
diastole values of the pressure changed by only few kilopascals. Obtained results are shown in
Fig. 12.
44
Figure 12: Comparison of pressure fields on arch of the aorta during characteristic points of
systole (left) and diastole (right) for single-phase model
Results of the wall shear stress caused by the plasma phase flow on rigid walls of the
artery show that as well as for single-phase analysis the highest value of about 95Pa appeared
underneath the aortic arch on the opposite wall to the left subclivian artery. Obtained results
showed in Fig. 13 are significantly lower than achieved in one-phase model. During diastole
wall shear stress on entire surface of the rigid vessel was close to 2kPa. As in single-phase
model areas where appeared the highest values of wall shear stress are related with locations
where are obtained the highest values of velocity.
45
Figure 13: Comparison of wall shear stress caused by blood flow during systole (left) and
diastole (right) for multi-phase model
Wall shear stress caused by erythrocytes during systole is shown in Fig. 14. Calculated
values are almost three times bigger than for plasma flow and the highest value reaches about
230Pa in the region of arch of the aorta. Wall shear stress caused by leukocytes during systole
is negligible and the highest value does not exceed 1Pa, results are shown in Fig. 15.
46
Figure 14: Wall shear stress during systole for
RBC phase
Figure 15: Wall shear stress during systole for
WBC phase
The velocity vectors obtained at characteristic points of systole and diastole has been
compared in Fig. 16. The simulated flow and calculated velocity profiles for plasma flow in
three-phase model looks similar to blood flow in one-phase model. The highest velocity
values are present at the area of coarctation and aortic arch and during systole reached over
4m/s. The lowest values appears to be located close to the wall surface. According to the
velocity field analysis, during systole blood flow was stable and mostly laminar, while during
diastole flow was highly irregular and turbulent, also in every outlet occurred the reversed
flow.
47
Figure 16: Comparison of velocity vectors in characteristic points during systole (left) and
diastole (right) for multi-phase model
5. Conclusions
The main scope of the present work was to show an Euler-Euler multiphase approach
in model of blood flow in human blood vessel (artery) section. In paper blood flow within real
geometry of a 8-year old female patient diagnosed with moderate thoracic aortic coarctation
has been simulated. Numerical analysis consists of multi-phase model where blood was a
non-homogenous mixture of three phases. As primary phase plasma as a Newtonian fluid was
used and as a secondary phases two granular phases RBC and WBC which represents
morphotic elements of blood. In multi-fluid model each phase was treated as a separate
continuum and modeled using Eulerian approach. Second numerical analysis consists single-
48
phase model where blood was treated as a homogenous, non-Newtonian mixture with average
properties. To reproduce periodical cardiac cycle the velocity profile as boundary condition
on the inlet of the domain has been implemented using User Defined Functions. Boundary
conditions applied in project are simplification of the influence that circulatory system exerts
on modeling section of the aorta. To fully compensate the influence of the human circulatory
system the numerical model should be coupled with an electrical analog (lumped model).
Presented model assumed rigid walls of the blood vessel and thus that the wall motion affects
the flow insignificantly. This simplification could have impact on the obtained pressure and
velocity profiles. Presented results show pressure field, wall shear stress and velocity vectors
for two characteristic points of cardiac cycle (during systole and diastole) for both singlephase and multi-phase models.
Acknowledgments
This research is supported by National Science Centre (Poland) within project
No 2014/13/B/ST8/04225. This help is gratefully acknowledged herewith.
I would like to express my sincere gratitude to Dr Wojciech Adamczyk and Bartłomiej Melka
for the invaluable support and help throughout this project.
Refrences
[1] Anderson J., Computational fluid dynamics. The basics with applications McGrawHill, New York (1995)
[2] Ergun S. Fluid Flow through Packed Columns, Chemical Engineering Progress
(1952) 48(2):89–94
[3] Gidaspow. D., Flow and Fluidization, Academic Press, Boston, MA., Reaction
Engineering International (1994)
[4] Gidaspow D., Bezburuah R. and Ding J., Hydrodynamics of Circulating Fluidized
Beds, Kinetic Theory Approach, In: Fluidization VII, Proceedings of the 7th
Engineering Foundation Conference on Fluidization, 75–82 (1992).
[5] Huang J., Lyczkowski R. W., Gidaspow D., Pulsatile flow in a coronary artery using
multiphase kinetic theory, Journal of Biomechanics (2009) 42: 743-754
[6] Kenner T., The measurement of blood density and its meaning, Basic Research in
Cardiology (1989) 84:111-124
[7] Launder B. E. and Spalding D. B., Lectures in Mathematical Models of Turbulence.
Academic Press, London, England (1972)
[8] Laurent F., Massot M., Villedieu P., Eulerian multi-fluid modeling for the numerical
simulation of coalescence in polydisperse dense liquid sprays, Journal of
Computational Physics, (2004) 197:505-543.
49
[9] Liu B., Tang D., Influence of non-Newtonian Properties of Blood on the Wall Shear
Stress in Human Atherosclerotic Right Coronary Arteries, Mol Cell Biomech (2011)
8(1):73-90, doi: 10.3970/mcb, 2011
[10] Lun C. K. K., Savage S. B., Jeffrey D. J., Chepurniy N., Kinetic Theories for Granular
Flow: In elastic Particles in Couette Flow and Slightly Inelastic Particles in a General
Flow Field, J. Fluid Mech.(1984) 140:223–256
[11] Siebert M. W., Fodor P. S., Newtonian and Non-Newtonian Blood Flow over a
Backward-Facing Step - A Case Study, COMSOL Conference, Boston (2009)
[12] Stergiopulos N., Computer simulation of arterial blood flow. Retrospective Theses and
Dissertations. Paper 9896, 1990
[13] Syamlal M The Particle-Particle Drag Term in a Multiparticle Model of Fluidization
National Technical Information Service, Springfield (1987)
[14] Tu J., Yeoh G.H., Liu Ch., Computational Fluid Dynamics, A Practical Approach,
Elsevier (2008)
[15] Wen C.-Y. and Yu Y. H., Mechanics of Fluidization, Chemical Engineering Progress
Symposium Series 62:100-111(1966)
[16] ANSYS Academic Research, Release 17.0, Help System, UDF Manual, ANSYS, Inc.
[17] CFD Challenge problem: http://www.vascularmodel.org/miccai2012 (accessed May
16th 2016)
Wielofazowy model przepływu krwi w naczyniu krwionośnym ze
sztywnymi ścianami w podejściu Euler-Euler
Maria Gracka
Słowa kluczowe: CFD, model Euler-Euler, przepływ wielofazowy, aorta, przepływ krwi
Streszczenie
Choroby oraz zaburzenia układu krążenia, takie jak miażdżyca tętnic, udary i zawały serca są
główną przyczyną zgonów na świecie, w szczególności w rozwiniętych i uprzemysłowionych
społeczeństwach. Zrozumienie podstawowych mechanizmów i zjawisk występujących w
układzie sercowo-naczyniowym może być przydatne do wczesnego wykrycia i diagnozy
rozwijających się zmian w naczyniach krwionośnych. W prezentowanej pracy
przeprowadzono analizę numeryczną przepływu krwi w aorcie. Wykonano dwa modele
numeryczne pierwszy model traktujący krew jako jednorodną ciecz nienewtonowską z
reologicznymi właściwościami średnimi dla lepkości i gęstości składników. Drugi model
obejmuje analizę przepływu krwi będącej mieszaniną trzech faz (osocza, czerwonych krwinek
oraz białych krwinek). Symulacje przeprowadzono przy użyciu komercyjnego
oprogramowania ANSYS Fluent (ANSYS Inc., USA). W projekcie wykorzystano geometrię
aorty 8-letniej pacjentki z koarktacją za łukiem aorty (zwężenie ok. 65%) wygenerowaną ze
skanu wykonanego podczas wzmocnionej angiografii (MRA). Model obejmuje aortę
wstępującą, łuk aorty, aortę zstępującą oraz górne odgałęzienia naczyń (pień ramiennogłowowy, tętnicę szyjną wspólną oraz podobojczykową lewą). W celu odwzorowania cyklu
50
pracy serca na wlocie do aorty jako warunek brzegowy przyjęto profil prędkości przeliczony z
przepływu objętościowego krwi, który zmierzono w trakcie badania kontrastem fazowym.
Warunek zaimplementowano wykorzystując tzw. procedury własne (UDF - User Defined
Function). Jako warunek brzegowy na wylotach przyjęto wypływy przez odgałęzienia oraz
aortę zstępującą, których wartość wyrażona jest udziałem procentowym ze strumienia krwi na
wlocie do aorty wstępującej. Wyniki symulacji numerycznej przedstawiono dla dwóch
punktów charakterystycznych podczas skurczu i rozkurczu serca.
51
c
Instytut
Techniki Cieplnej, Politechnika Ślaska
˛
Archiwum Instytutu Techniki Cieplnej Vol. 1 nr 1(2016)
www.itc.polsl.pl/ArchiwumITC
Identification and modelling of blood flow processes
in section of large blood vessel
using hybrid Euler-Lagrange multiphase approach1
Artur Knopek2
Key words: aorta, CFD, blood flow, ANSYS Fluent, multiphase, DPM
Abstract
Computational fluid dynamics (CFD) in past known only in highly specialized technical engineering branch is nowadays one of main engineering tool in solving numerous complex problems in order to get crucial information and extend general knowledge in many fields. CFD
allows to create new, more advanced systems and also optimize already created to enhance
efficiency and/or reduce costs of production and operating. Actual situation demands from engineers to face difficult competition - fighting for minor fractions of efficiency due to construction and materials limitations. That operations do not concentrate only on that obvious disciplines like heat transfer, fluid dynamics or power-generation, but also new uncharted areas like
automotive, chemical, aerospace, environmental engineering etc.
One of that innovative field of CFD application is bio-engineering. In medicine, computer simulations can provide necessary, life-saving information with no interfere in patient body (in
vivo), that allows to avoid later complications, application collisions and dangerous unpredictable after-effects. What more in several cases, in vitro analyses cannot be used through to life
threats of treatment.
The main objective of current project is to develop and test novel approach of accurate modelling of human blood flow in arteries.
Currently available research reports do not cover the spatial interaction of individual blood
phases and walls of blood vessels. Such approach could significantly reduce accuracy of such
models. Proper simulations enriches general knowledge with specific details which could be
1
Research carried out under the project No 2014/13/B/ST8/04225 funded by National Science
Center.
This section was prepared as a Master Thesis project by Artur Knopek at Institute of Thermal
Technology at Energy and Environmental Engineering Faculty of Silesian University of Technology, under supervision of Dr Ziemowit Ostrowski
2
Zawartość tej publikacji może byc wykorzystana na warunkach licencji Uznanie Autorstwa 3.0 Polska. Licencja pozwala na
kopiowanie, zmienianie, rozprowadzanie, przedstawianie i wykonywanie utworu jedynie pod warunkiem oznaczenia autorstwa: autora,
tytułu rozdziału, nazwy serii, tomu, strony. Content from this work may be used under the terms of the Creative Commons Attribution
3.0 licence. Any further distribution of this work must maintain attribution to the author, chapter title, series title, volume, pages.
53
crucial in early diagnosis of potential cardiac problems showing vulnerable zones (e.g. narrowed blood vessels). Such precise information are extremely difficult to obtain experimentally.
Apart from multiphase concept of the project (that is considering every component of blood as
separate phase assigning exceptional properties to each of them and determines relations between them) special attention was paid to the realism of geometry - considering the real system
of the aortic segment (part of ascending aorta, aortic arch and part of thoracic aorta) including
bifurcations. In addition a pulsating blood flow is being considered and implemented using built
in UDF (User Defined Function) functionality of CFD code.
1. Introduction
1.1
Main goal of project
The main objective of current project is to develop and test novel approach of accurate modelling of human blood flow in arteries.
Currently available research reports usually do not cover the spatial interaction of individual
blood phases and walls of blood vessels. Such approach could significantly reduce model accuracy. Proper simulations enriches general knowledge with specific details which could be
crucial in early diagnosis of potential cardiac problems showing vulnerable zones (e.g. narrowed blood vessels). Such precise information are extremely difficult to obtain experimentally.
1.2
Computational Fluid Dynamics and ANSYS CFD
The permanent development of the technique enabled applying numerical methods to wide
range of applications, thanks to enhancing the possibility of modern computers. The appropriate
equipment and software are capable to deal with complicated problems and make advanced
simulations which could not be solved with analytical methods. However, numerical methods
through the process of the discretization does not provide accurate calculated answer, but its
approximation with accuracy balanced according to needs. Nowadays thanks to high computing
power one can explore even so sophisticated field of life as bioengineering [8]. Usage of Computational Fluid Dynamics (CFD) solver [A] allows to assemble all essential information. A hybrid multifluid Euler-Lagrange approach is proposed in current study, being a part of wider
research project targeted at modelling of blood flow in human artery section. Calculations was
based on bundled software (ANYS Inc., USA) – highly developed engineering tool allowing to
investigate problems with many approaches to get real behavior of blood flow (velocity, pressure, particles deposition etc.)
1.3
Concept of blood as a multiphase fluid
In proposed model, the multifluid [9] approach is used to model blood consisted of plasma and
RBCs (Red Blood Cells) as interpenetrating phases. In addition, the leukocytes (which volume
fraction is about 1% of the overall blood volume) are represented as dispersed phase. In proposed methodology, where collisions of WBCs (White Blood Cells) can be modeled, the advantages of hybrid Euler Lagrange [15] model are used.
The interaction between phases and within phases are calculated in Eulerian grid, where the
calculated interaction stress tensor, is used by dispersed phase to take into account collision
effect. As a result, the 3D structure of mutually interacting particles will be modelled.
54
1.4
Blood and geometry properties
Apart from multiphase concept of the project, that is considering every component of blood as
separate liquid assigning unique properties to each of them and determines relations between
them. Special attention was paid to the realism of geometry - considering the real system of the
aortic segment (part of ascending aorta, aortic arch and part of thoracic aorta) including bifurcations. In addition, a pulsating blood flow was implemented with using a built in UDF (User
Defined Function) module which will employ lumped model of human circulatory system.
2. Biomedical specifications
2.1
Basic definitions
Heart – muscular organ build of two chambers and two atria (right and left). Atria are separated
from each other by interatrial septum and chambers by interventricular septum. Atria are separated from chambers by atrioventricular septum with two holes in it (left and right) with 2 cm
diameter each. Chambers and atria works as a 4 series pumps (right chamber and atrium; left
chamber and atrium) which supplies all human cardiovascular system. There is one direction
of blood flow – from veins reservoir to aortic reservoir thanks to build of heart and existence
of valves which prevents backflow of blood [17].
Cardiac cycle – consist of two phases: systole - period of cycle when chambers contracts and
ejects part of contained blood to large aortic reservoirs and diastole - period of cycle when heart
chambers muscles are under relaxation, pressure and blood fulfillment in them is lowering.
Phases are repeated cyclically with frequency of 1,2Hz (72 times per minute). One cycle last
about 800 ms (systole – 270ms; diastole – 530ms) [7].
Stroke Volume (SV) is the amount of blood pumped by heart chambers to specific aortic reservoirs. Every of each chamber pumps about 75 mL of blood in one cardiac cycle (adult male
with total mass of 70 kg, in quiescence and lying position). At the end of cycle in heart remains
about 50 mL of residual blood volume, which conditions end-systolic ventricular volume [17].
Heart work – in physic work (W) is the force (F) working on the certain distance (L)
𝑊 =𝐹 ∙𝐿
(1)
Essential feature of heart is volume of ejected blood and pressure in aorta, which heart must
overcome to pump mentioned amount and accelerate it to required speed. There is an external
and internal work of heart. Internal responds for contraction tension generation, while external
responds for pumping blood. Total heart energy Wt consist of energy necessary to sustain pressure in blood vessels – potential energy (Ep), energy necessary to accelerate blood flow after its
been pump from heart to vessels – kinetic energy (Ek), energy necessary to overcome gravitation force (Eg).
𝑊𝑡 = 𝐸𝑝 + 𝐸𝑘 + 𝐸𝑔
(2)
55
In normal conditions Ek is equal to 5% of total energy and is feasible to calculate it by equation:
𝐸𝑘 =
𝑚 ∙ 𝑣2
2
(3)
where: 𝑚 – mass, kg; 𝑣 – velocity, m/s;
Potential energy is equal to 95% of total energy and is feasible to calculate it by eq. (4):
𝐸𝑝 = 𝑃𝐴 ∙ 𝑉𝑠
(4)
where: 𝑃𝐴 - pressure in arteries, Pa; 𝑉𝑠 - stroke volume, m3;
2.2
Large Blood Vessel – Aorta
Vascular system is a systemic arterial reservoir which includes: aorta, arteries, arteriole, capillaries, wenule and veins. Main function of the aorta, arteries and arteriole (fig.2.1) is blood
transport under high pressure to tissue and changing of pulsatile flow to continuous flow. Aorta
leaves from the left chamber of a heart and subdivide to: ascending aorta, arch of aorta with
bifurcations, descending aorta which divides to the thoracic and the abdominal aorta [10].
Fig.2.1. Aorta placement in human body (left3 and right4)
3
Creative Commons Attribution-Share Alike 3.0 Unported. After author:
AoDissekt_scheme_StanfordB.png
by
JHeuser;
media.org/wiki/File:AoDissekt_scheme_StanfordB_en.png#filelinks
4
Fvasconcellos.
Image:
https://commons.wiki-
Creative Commons Attribution 3.0 Unported. After author: OpenStax College; https://commons.wikimedia.org/wiki/File:2121_Aorta.jpg
56
Arterial flow characterized by: pressure, velocity of blood flow, capacity, arterial pulse. Arterial
reservoir contains about 550 mL of blood, what is about 11% of all blood flowing in human
organism. Cardiac output is in quiescence about 0,9 l/min. The same amount of blood must
flow in aortic, arteries, arterioles etc. cross sections, causing gradual enlarging of summary
cross section which must be supplied with blood. Due to that velocity of blood is lowering,
referring to the equation [10, 17]:
𝑣 = 𝑄/ ∑ 𝐴
(5)
where: v – velocity, m/s; Q – blood flow, kg/s; A – summary area of cross sections, m2;
The biggest summary cross section area is in capillaries where velocity in quiescence is the
lowest – almost 0. In common adult aorta, which cross section is about 4,5 cm2 and total maximum flow about 0,2 l/s, velocity is about:
0,2 𝑑𝑚3 /𝑠
𝑚
𝑘𝑚
𝑣𝐴 =
= 4,44 ≈ 16
2
0,045 𝑑𝑚
𝑠
ℎ
Arterial blood pressure depends on inflow and outflow from arterial reservoir. When inflow
and outflow is stabilized and there is no change in walls tension of aorta, average pressure in
that reservoir do not change and depend on cardiac cycle. For maximum ejection of left chamber pressure is the highest (systolic arterial pressure) and its equal to 16 kPa (120 mm Hg),
when the diastolic arterial pressure is the lowest and equal to 9,3 kPa (70 mm Hg) [17].
2.3
Blood components
Blood is a liquid tissue, filling bloodstream, separated from other organism tissues with a layer
of endothelium with total area of about 100m2. Blood is in constant flow and comprise 6-7% of
total human mass. Blood consist of blood cells which is about 45% of blood volume (leukocytes, erythrocytes, thrombocytes) and plasma [10].
Plasma is partially transparent, pale straw (yellow) colored liquid, iridescent fluid, which
mostly consist of water (90-92%) and contains dissolved organic and nonorganic proteins [7].
Density of plasma is equal to 1022 – 1026 kg/m3.
Erythrocytes (Red Blood Cells) are produced by bone marrow. Their main purpose is transporting oxygen along the organism. Theirs unique shape connected with flexibility is highly important in that proses. Average diameter of one RBC is about 8 μm and its round, biconcave
disc (Fig.2.1), its area equals to 120μm. The amount of RBCs in 1μl of blood is about 5mln in
men and 4,5mln in women, density is equal to 1095 – 1101 kg/m3 [7].
57
Fig.2.1. RBCs structure (left5 and right6)
Hematocrit (Hct) is crucial parameter (6) describing volume of RBCs (VRBCs) ratio to total
blood volume (VB). It has significant influence on numerous blood parameters. Usually in percentage 45% for men and 40% for women.
𝐻𝑐𝑡 =
2.4
𝑉𝑅𝐵𝐶𝑠
∙ 100%
𝑉𝐵
(6)
Leukocytes
Mainly function of WBCs is to defend the organism from pathogenic germs. The amount of
WBCs in 17,5 μml of blood is about 4000 – 11000 what gives amount less than 1% in total
blood volume. Divided on many different kinds (tbl.2.2), it is possible to specify them by structure on two main groups granulocytes (presence of granules in its cytoplasm) or agranulocytes
(absence of granules in its cytoplasm) [7].
Tbl.2.2. Chosen characteristic feature of WBCs
Amount in 1l of blood ∙ 109 Approx. % Diameter μm
Type
Neutrophil 4,4
62
10-12
Granulocytes Eosinophil 0,2
2,3
10-12
0,04
0,4
10-15
Lymphocyte
2,5
30
8-9
Monocyte
0,3
5,3
15-30
Basophil
5
Creative Commons Attribution-Share Alike 4.0 International. After author: Cancer Research UK uploader;
https://commons.wikimedia.org/wiki/File:Diagram_of_a_red_blood_cell_CRUK_467.svg
6
Creative Commons Attribution-ShareAlike License. After author: NHLBI; https://commons.wikimedia.org/wiki/File:Modified_sickle_cell_01.jpg
58
3. Numerical model
3.1
General information about CFD
Computational fluid dynamics (CFD), [8] in past known only in highly specialized technical
engineering branch, is nowadays one of main engineering tool in solving several complex problems in order to get crucial information and extend general knowledge in many fields. CFD
allows to create new, more advanced systems and also optimize already created to enhance
efficiency and/or reduce costs of production and operating. Actual situation demands from engineers to content with difficult competition - fighting for percentage of efficiency due to construction and materials limitations. Operations not concentrates only on that obvious disciplines
like heat transfer, fluid dynamics or power-generation, but also new uncharted areas like automotive, chemical, aerospace, environmental engineering etc.
One of that innovative field of CFD application is bio-engineering [16]. In medicine, computer
simulations can provide necessary, life-saving information with no interfere in patient body (in
vitro), allowing to avoid later complications, application collisions and dangerous unpredictable
after-effects. What more in several cases in vitro analyses can be impossible to use through to
life threats of treatment.
3.2
Finite volume method
The FVM is main discretization model in ANSYS Fluent. Domain is subdivided on finite control volumes called cells. Mesh (set of cells) allows program to obtain continuous calculation
of momentum exchange law through domain. Program uses the cell-centered (CC) method for
calculating conservation equations – whole cell value is equal to its center (fig.3.1). First, average values of chosen parameters are stored in the centroids and then, discrete equations of conservations laws are calculated in order to obtain whole value of all volume (cell). Methods used
to calculate average values of the centroids are described in section 3.8. [13]
Fig.3.1. Presentation of Cell Centered Finite Volume Method
59
3.3
Euler – Euler Approach
In Eulerian approach, the base characteristic of the solid phase is calculated as a continuous
medium with main properties assumed as dispersed phase of fluid - and is treated as a continuum. Important feature of mixture consisting of two fluids is individual fraction volume concentration. To obtain phases transport equations terms of fluid continuity and momentum conservation must be provided. It is impossible to consider every point of time and space of the
two continua flow. Therefore conservation laws are solved separately and then combined in
discretized geometry, taking in account volume concentration of each phase [3].
3.3.1 Specific project usage
Eulerian approach was used in calculations of interactions with diffused phases and within
phases of plasma and erythrocytes. Assumption of granular second phase in multiphase model
[A] requires to define flow as a solid-fluid mixture. for the case where assumption of inelastic
granular phase is used, it is necessary to found the combination of the random molecules motion
among particles collisions and thermal motion of particles in a fluid when calculating solid
phase stresses. Pressure, stresses and viscosity of the solid phase are determined by intensity of
molecules velocity fluctuations. Therefore it is necessary to implicate particle velocity fluctuations (which is a result of particle kinetic energy) with granular temperature (equal to random
particle motions mean square).
The general momentum conservation law includes interactions between more than two phases.
In this case main phase law (fluid) is shown in (7), when the second phase in (8) including
exchange of moment between main and secondary phase.
𝜕(𝜌𝑓 𝜀𝑓 ⃗⃗⃗⃗
𝑣𝑓 )
⃗⃗⃗
(7)
+ ∇ ∙ (𝜌𝑓 𝜀𝑓 ⃗⃗⃗⃗
𝑣𝑓 ⃗⃗⃗⃗
𝑣𝑓 ) = 𝜌𝑓 𝜀𝑓 𝑔 − 𝜀𝑓 ∇𝑃𝑓 + ∇ ∙ ⃗⃗⃗
𝜏𝑓 + 𝐾𝑠𝑓 (𝑣
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 )
𝜕𝑡
𝜕(𝜌𝑠 𝜀𝑠 ⃗⃗⃗
𝑣𝑠 )
(8)
+ ∇ ∙ (𝜌𝑠 𝜀𝑠 ⃗⃗⃗
𝑣𝑠 ⃗⃗⃗
𝑣𝑠 ) = 𝜌𝑠 𝜀𝑠 𝑔 − 𝜀𝑠 ∇𝑃𝑓 − ∇𝑃𝑠 + ∇ ∙ ⃗⃗⃗
𝜏𝑠 + 𝐾𝑓𝑠 (𝑣
⃗⃗⃗
⃗⃗⃗⃗𝑓 − ⃗⃗⃗
𝑣𝑠 )
𝜕𝑡
where: f – fluid phase; s – solid phase; 𝑔 – gravitational acceleration, m/s2; ⃗τ⃗ – stress tensor,
N/m2; 𝑃𝑠 - granular pressure, Pa; 𝑃𝑓 – fluid pressure, Pa; 𝜌𝑠 – solid density, kg/m3;
𝑣𝑠/𝑓 – solid/fluid velocity, m/s; 𝑡 – time, s; 𝐾 – interphase exchange momentum coefficient
⃗⃗⃗⃗⃗⃗⃗
3.3.2 Interphase exchange momentum coefficient (General Model)
One of the crucial implementation in conservation law [A] is interphase exchange momentum
coefficient which determines behavior of phases in domain. Interphase exchange momentum
coefficient general model equation is presented below.
𝐾𝑓𝑠 =
𝜀𝑠 𝜌𝑠 𝑓
𝜏𝑟𝑒𝑙
(9)
where: 𝑓 – drag function, 𝜏𝑟𝑒𝑙 – particulate relaxation time, s;
𝜏𝑟𝑒𝑙
𝜌𝑠 𝑑𝑠2
=
18𝜇𝑙
where: 𝑑𝑠 –diameter of solid particle, m; 𝜇𝑙 – viscosity of liquid phase, Pa∙s;
60
(10)
3.3.3 Drag model - Gidaspow
Different models formulates f function in other way usually using the drag coefficient CD (based
on the Reynolds number). To calculate momentum exchange within RBC’s and Plasma
Gidaspow [5] model approach was used. General assumption of models are:
Drag function must be multiplied by main phase volume fraction to obtain no momentum exchange if there is no main phase, there is no momentum exchange and the only determinant is
fractions volume.
Gidaspow model approach highly depends on volume fraction by using two different equations:
Wen and Yu (11) [18] or Ergun equation (12) [4]. Momentum exchange coefficient 𝐾𝑓𝑠 equates:
when 𝜀𝑓 > 0,8 (Wen and Yu):
𝐾𝑓𝑠 =
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 | −2,65
3 𝜀𝑠 𝜀𝑓 𝜌𝑓 |𝑣
𝐶𝐷
𝜀𝑓
4
𝑑𝑠
(11)
when 𝜀𝑓 ≤ 0,8 (Ergun):
𝐾𝑓𝑠 = 150
𝜀𝑠 (1 − 𝜀𝑓 )𝜇𝑓
𝜌𝑓 𝜀𝑓 |𝑣
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 |
+ 1,75
2
𝜀𝑓 𝑑𝑠
𝑑𝑠
(12)
where Cd is calculated from (13):
𝐶𝐷 =
24
[1 + 0,15(𝜀𝑓 𝑅𝑒𝑠 )0,687 ]
𝜀𝑓 𝑅𝑒𝑠
Where:
(13)
𝑅𝑒 - relative Reynolds number, presented in equation below
𝑅𝑒𝑠 =
𝜌𝑓 𝑑𝑠 |𝑣
⃗⃗⃗𝑠 − ⃗⃗⃗⃗
𝑣𝑓 |
𝜇𝑓
(14)
Project assumes constant RBC’s fraction volume 𝜀𝑠 for about 44% of total flow volume what
had an effect in usage mostly Erguns equation.
3.4
Standard 𝒌 − 𝜺 model
Assumption of turbulent standard 𝑘 - 𝜀 model was made in which transport equations are calculated separately: 𝑘 – turbulent kinetic energy and 𝜀 – rate of dissipation. By coupling exact
equations of (k) and partially empirical equations of (𝜀) model received acceptable accuracy in
large range of turbulent flows which occurred in studied case. Equations calculated per phase
are presented below in (15) and 16) [11]:
61
𝜕
𝜕
𝜕
𝜇𝑡 𝜕𝑘
(𝜌𝑘) +
(𝜌𝑘𝑢𝑖 ) =
[(𝜇 + )
] + 𝑆𝑘 + 𝑆𝑏 − 𝜌𝜀 − 𝐺𝑀
𝜕𝑡
𝜕𝑥𝑖
𝜕𝑥𝑗
𝜎1 𝜕𝑥𝑗
(15)
𝜕
𝜕
𝜕
𝜇𝑡 𝜕𝜀
𝜀
𝜀2
(𝜌𝜀) +
(𝜌𝜀𝑢𝑖 ) =
[(𝜇 + )
] + 𝐺1 (𝑆𝑘 + 𝐺3 𝑆𝑏 ) − 𝐺2 𝜌
𝜕𝑡
𝜕𝑥𝑖
𝜕𝑥𝑗
𝜎2 𝜕𝑥𝑗
𝑘
𝑘
(16)
and
where:
𝑆𝑘 - turbulence kinetic energy generation due to the mean velocity gradients, J/kg; 𝑆𝑏 - turbulence kinetic energy generation due to buoyancy, J/kg; 𝐺𝑀 - contribution of the fluctuating
dilatation in compressible turbulence to the overall dissipation rate, J/kg.
Eddy viscosity 𝜇𝑡 is calculated from (17).
𝜇𝑡 =
𝑘2
𝜌𝐺𝜇
𝜀
(17)
Constants 𝐺1 , 𝐺2 , 𝐺3 , 𝐺𝜇 , 𝜎1 , 𝜎2 used in above equations are empirical values obtained from
experiments and observations founded to work properly in practical engineering.
3.5
Partial Differential Equation - Granular Temperature Model
The granular temperature 𝜃𝑡 of the solid phase in the Euler approach is proportional to the
kinetic energy of the random motion of the particles and is calculated from (18):
𝜃𝑡 =
1
𝑢 𝑢
3 𝑆,𝑗 𝑆,𝑗
(18)
where:
𝑢𝑆,𝑗 - Cartesian coordinates (𝑗) of the fluctuating solid (𝑆) velocity
While the transport equation [2] as (19):
3 𝜕
[ (𝜌 𝜀 𝜃 ) + ∇ ∙ (𝜌𝑠 𝜀𝑠 ⃗⃗⃗
𝑣𝑠 𝜃𝑡 )] = (−𝑝𝑠 𝐼 ̿ + 𝜏̿𝑠 ): ∇𝑣
⃗⃗⃗𝑠 + ∇ ∙ (𝑘𝜃𝑡 ∇𝜃𝑡 ) − 𝛾𝜃𝑡
2 𝜕𝑡 𝑠 𝑠 𝑡
(19)
where:
(−𝑝𝑠 𝐼 ̿ + 𝜏̿𝑠 ): ∇𝑣
⃗⃗⃗𝑠 - generation of energy by the solid stress tensor, J/kg;
𝛾𝜃𝑡 - the collisional dissipation of energy, J/kg; 𝑘𝜃𝑡 ∇𝜃𝑡 - the diffusion of energy (𝑘𝜃𝑡 is the
diffusion coefficient), J/kg.
For obtaining 𝑘𝜃𝑡 Gidaspow et al. model was assumed [6] and calculated (20):
𝑘𝜃𝑡 =
150 𝜌𝑠 𝑑𝑠 √𝜃π
6
𝜃𝑡
[1 + 𝜀𝑠 𝑔0,𝑠𝑠 (1 + 𝑒𝑠 )]2 + 2𝜌𝑠 𝜀𝑠 2 𝑑𝑠 (1 + 𝑒𝑠𝑠 )𝑔0,𝑠𝑠 √
348(1 + 𝑒𝑠𝑠 )𝑔0,𝑠𝑠
5
𝜋
(20)
where:
𝑔0,𝑠𝑠 - radial distribution function; 𝑒𝑠𝑠 – solid-solid restitution coefficient; 𝑒𝑠 - solid restitution
coefficient,; 𝛾𝜃𝑡 - collisional dissipation of energy, J/kg.
62
Energy dissipation within the solids phases as a result of interparticle collisions and is equal to
(Lun et al.) [12]:
𝛾𝜃𝑡 =
3.6
12(1 + 𝑒𝑠𝑠 2 )
𝑑𝑠 √𝜋
𝜌𝑠 𝜀𝑠 2 𝜃𝑡 1,5
(21)
Euler – Lagrange Approach
Lagrange approach describes particle trajectory which has momentum, mass and energy. Particle is tracked in moving coordinate frame following the particle in time (22) [3]:
V = V(x,y,z,t)
(22)
Fig.3.2. Particle streamline in domain
ANSYS Fluent [A] combines both of approaches (Eulerian, Lagrangian) by calculating Eulerian phases as a continuum, using Navier-Stokes equations. Dispersed phase as a dependent
numerous particles flowing though calculated flow field (Eulerian grid) and exchanging momentum, mass, and energy with the main phase. Euler – Lagrange approach in hybrid model
allows to obtain all crucial information about blood behavior (e.g. velocity vectors, pressure
distribution, particles positions in blood vessel)
Change of dispersed phase momentum is calculated from (23):
𝐹 = ∑(
18𝜇𝐶𝐷 𝑅𝑒
(𝑢𝑝 − 𝑢) + 𝐹𝑜 )𝑚̇𝑝 ∆𝑡
𝜌𝑝 𝑑𝑝2 24
(23)
where:
𝜇 – fluid viscosity, Pa∙s; 𝜌𝑝 – particle density, kg/m3; 𝑑𝑝 – particle diameter, m; Re – Reynolds
number; 𝑢𝑝 – particle velocity, m/s; 𝑢 – fluid velocity, m/s; 𝐶𝐷 – drag coefficient, 𝑚̇𝑝 – mas
flow rate of particles, kg/s; ∆𝑡 – time step, s; 𝐹𝑜 – other forces
63
Drag coefficient 𝐶𝐷 , for smooth particles equals to
𝛼2
𝛼3
(24)
+ 2
𝑅𝑒 𝑅𝑒
Where 𝛼1,2,3 are constants related to Re numbers Given by Morsi and Alexander [14].
𝐶𝐷 = 𝛼1 +
3.7
Particle injection
The Particle Type was chosen as inert. Moreover, the surface injection type was determined as
particle stream releasing particles from each cells face on chosen boundary condition. The particle injection visualization is presented in (Fig.3.3).
Fig.3.3. Injecting particles parcels from faces
In chosen approach, momentum equation is not calculated for an individual particles [1]. Program mechanism for tracking is based on parcels – groups of particles. Particles in each parcel
have equal value of main properties (position, mass, velocity). Number of particles in parcel
depends on individual particle mass and can be obtained from (25):
64
𝑛𝑝 =
𝑚̇𝑝
∆𝑡
𝑚(𝑑, 𝜌)
(25)
where: Δt - time step, s; 𝑚̇𝑝 - single parcel source mass flow rate, kg/s; 𝑚 - mass of an individual particle, kg; 𝑑 - particle diameter of an individual particle, d; 𝜌 – density of an individual
particle, kg/m3.
3.8
Rosin-Rammler Diameter Distribution Method (RRDDM)
To better represent the variation of leukocytes diameters, RRDDM was used [A]. Full range of
available diameters (continuous function) is divided on narrow intervals. Representation of
each interval is its mean diameter injected as a separated stream. If particle diameter is > d,
mass fraction is calculated from (26):
𝑌𝑑 = 𝑒
𝑑
−( ̅ )𝑛
𝑑
(26)
𝑌𝑑 – mass fraction, kg; 𝑑 – droplet diameter, m; 𝑑̅ – size constant – mean diameter, m; 𝑛 - size
distribution parameter – spread parameter.
According to paragraph 2.6, the largest part of WBSs are neutrophils what determinates choosing its diameter as a general mean diameter of a model (Table 3.1).
Specific values of diameter distribution parameters used in RRDDM for obtain acceptable diversity of WBCs are shown in tbl.3.1:
Table.3.1. RRDDM characteristic parameters
Parameter
Value
Max diameter
15 μm
Mean diameter
11 μm
Min diameter
7 μm
Spread
8
Number of Diameters
6
65
3.9
Green-Gauss Node-Based Gradient Evaluation (G-GN-B)
To obtain scalar value on cells’ faces, secondary diffusion terms and velocity derivatives are
necessary to calculate gradients [A]. Gradients are important part of conservation equations (in
discretization of diffusion and convection condition):
𝑁𝑓𝑐
𝜑̅𝑓𝑐
1
=
∑ 𝜑̅𝑛
𝑁𝑓𝑐
(27)
𝑛
where:
𝜑𝑓𝑐 – value of 𝜑 at the cell face centroid, 𝜑̅𝑓𝑐 – arithmetic average of the nodal values on the
cells faces, 𝜑̅𝑛 – nodal values constructed from the weighted average of the cell values surrounding the nodes, 𝑁𝑓𝑐 – number of nodes on face.
Node-based gradient scheme is more accurate than cell-based on skewed and distorted meshes
but may be more computational expensive
4. Model properties
4.1
Discrete Phase Model (DPM)
Discrete phase model is a practical answer of ANSYS Fluent for Lagrangian reference frame.
Model ignores inner-phase, particle-particle impact with assumption that discrete phase has a
low volume fraction according to main phase. However, high mass loading in model is possible.
Trajectories of dispersed phase particles are calculated separately in chosen steps. Model calculations were synchronized with general frame time step. Computations of DPM were made
at the beginning of iteration proses based on previous step calculations. No particles at the beginning of simulation was assumed. Injection begins in first time step. The amount of WBCs is
a percentage mass fraction of the continuous phases (RBCs + Plasma) and it is equal to 1%.
WBCs are threated similarly to RBCs – like a granular suspension in plasma. The difference is
(despite of total volume WBCs fraction which is negligible) that Leukocytes do not exchange
momentum with main phase but are carried without any influence on main phase [A].
4.2
Geometry
Geometry used in the model was obtained from CFD Challenge [C] and presents real geometry
adopted from 8-year old female patient MRA imagining (Gadolinium-enhanced Magnetic Resonance Angiography), performed using 1.5-T GE Signa scanner. The model geometry is presented in (fig.4.1). Patient had congenital defect of cardiovascular system – Coarctation of the
aorta (CoA) which was about 65% of the blood vessel cross-section area reduction. Examined
regions were: ascending aorta (AAO), aortic arch with bifurcation - innominate artery (IA), left
common artery (LCCA), left subclavian artery (LSA) and descending aorta (DAO).
66
Fig.4.1. Geometry of the model used in the numerical simulations.
4.3
Boundary Conditions
The importance and difficulty of proper setting the BCs (boundary conditions) is one of the key
feature in CFD modeling process. Used geometry demands well adaptation of 5 BCs - one
inflow and four outflow. Ansys Fluent [A] determinates restrictions and limitations regarding
to boundary conditions due to them not all of inlet-outlet relations are compatible. When considered domain has a multiple outlets, the recommended outlet boundary condition is OUTFLOW what reduce possible inlet conditions to velocity inlet condition only.
4.3.1 Inlet BC
At inlet to domain in Fig.4.1 Inlet from heart modeled BC was velocity inlet. To obtain more
realistic behavior of human cardiovascular model pulsatile flow was implicated. Data was obtained from CFD Challenge in which blood flow was measured using phase-contrast (PC) cine
sequence with through-plane velocity encoding [C]. The measured (shown in Table 4.1) blood
flow presented is in time dependent points of cardiac cycle which was assumed as 0,7s. In effect
volumetric flow was obtained. Due to chosen inlet BC volumetric blood flow was recalculated
basing on stream continuity equation (4.1). To get constant blood flow in cardiac cycle data
was interpolated within chosen periods of time to derive interpolation functions.
𝑉(𝑡) = 𝐴𝑣
(4.1)
67
Tbl.4.1. Time dependent mass flow measurement and velocity
T
0
0,035
0,07
0,105
0,14
0,175
0,21
0,245
0,28
0,315
0,35
0,385
0,42
0,455
0,49
0,525
0,56
0,595
0,63
0,665
0,7
mm3/s
3214,5
12736,1
70423
171359,4
201975,2
176127,7
141663,4
100536,8
73060,3
36444,2
10113
-257,1
6611,8
14545,6
21754,2
18761,3
13006,3
3472,9
2223,4
4190,7
3214,5
m/s
0,0157
0,0623
0,3445
0,8382
0,9880
0,8616
0,6930
0,4918
0,3574
0,1783
0,0495
-0,0013
0,0323
0,0712
0,1064
0,0918
0,0636
0,0170
0,0109
0,0205
0,0157
Highlighted (bold) characteristic measure point shows common occurring in heart process of
blood backflow, which was zeroed (essential negative influence on model stability with outflow
BC).
Effect of undertaken operations is shown in Figure 4.2, where two characteristic periods are
specified: systole (when the left heart ventricle contracts, and blood flows to the aorta with
maximum speed) and diastole (when the semilunar valve is closed, the atrioventricular valves
are open, and the whole heart is relaxed).
68
Fig.4.2. Inlet velocity obtained in model
Implication of the inlet pulsatile flow was possible with usage of UDF (User Define Function)
– special feature in ANSYS products allowing to adjust program standard commercial code
features and options in order to get more advantageous results. Periodic flow is obtained
through time dependence of auto-implicated flow velocity in every repeatable cycle. UDF uses
particular internal functions based on C# programming language.
Fraction percentage share on inlet assumed in model is 44% of RBC and its complement equal
to 56% is Plasma.
4.4
Outlet BC
As said in previous paragraph BC appropriate in chosen system is Outflow, which is used to
calculate model flow exits if pressure and velocity magnitude is unknown. Outflow BC was
chosen in all outlets (BCA, LCCA, LSA, DA – cf. Fig. 1). Amount of fluid volume leaving
domain depends percentage on inlet flow volume and its shown in Table 4.2.
Tbl.4.2. Outlet boundary conditions fractions distribution
Outlet Surface
Fraction, %
Brachiocephalic artery (BCT)
25,6
Left Common Carotid Artery (LCCA)
11,3
Left Subclavian Artery (LSA)
4,3
Descending aorta (DA)
58,8
69
4.5
Blood properties used in model
Taking in account blood phases it is necessary to show implemented values in model (tbl.4.3)
Tbl.4.3. Modeled blood parameters
4.6
Plasma density, kg/m3
1026
Plasma viscosity, Pa∙s
0,0035
RBC density, kg/m3
1101
RBC viscosity, Pa∙s
0,00417
WBC density, kg/m3
1125
Mesh
One of inherent element of CFD modeling is the discretization of the domain. Not only mesh
element quality is crucial but also mesh element size must be adjusted to selected timestep.
Optimum mesh element size is obtained when particles flow thru subsequent cells without passing any of them (causing information loos and inaccuracy). Second limitation is not to calculate
the same particle multiple times in one control volume in order to overburden the computing
power.
In the thesis, hybrid mesh was created with approximately 720 000 elements (hexagonal, tetrahedral) (fig.4.4). Hexagonal part situated on inlet and descending aorta was created in order to
better track discrete phase elements. In parts when intensive fluctuation occurs mesh was desilicated to avoid numerical errors (aortic arch). More precise view on inlet and outlet is presented in Fig. 4.3, showing inflation of mesh near boundaries.
Fig. 4.3. Precise insight on inlet and outlet mesh
70
Fig.4.4. Hybrid mesh
5. Results
Calculation was made on 3D model with rigid walls presented in section 4. Transient simulation
was performed with timestep Δt = 0,01. Operating fluid was incompressible. Point of reference
is localized in descending aorta.
5.1
Pressure
Obtained relative static pressure on the walls of domain is presented on contours, in two of
characteristic points of cardiac cycle: systole (see Fig. 5.1) and diastole (see Fig 5.2). Highest
pressure for about 6,8 kPa is localized on inlet of domain in systole (0.14s of calculation). In
the same period of time the lowest pressure equal to -7.7 kPa is localized on CoA. Systolic
pressure difference is equal to 14.5kPa. Diastolic maximum pressure was about 600 Pa on descending aorta outlet. The lowest diastolic pressure was noticed before CoA inlet, equals to - 2
kPa.
71
Fig.5.1. Contours of systolic static pressure Fig.5.2. Contours of diastolic static pressure
5.2
Wall stress tensor
Plasma wall stress tensor of two characteristic points was presented inFig.5.3 and Fig. 5.4. The
magnitude of about 260 Pa was obtained on aortic arch, on maximum angulation, before CoA.
Noticeable magnification was located in CoA equal to 90Pa. Diastolic primary phase maximum
wall stress tensor is located in the same part of aortic arch (Fig. 5.4) like in systolic cycle period,
43 Pa.
72
Fig.5.3. Primary phase diastolic wall shear stress
Fig.5.3. Primary phase systolic wall shear stress
73
Secondary phase (RBCs) distribution of wall stress tensors have comparable space distribution
but vary with value: for systole (Fig. 5.5) it is 62 Pa and for diastole (Fig. 5.5) is 6 Pa.
Fig.5.5. Secondary phase systolic wall shear stress
Fig.5.6. Secondary phase diastolic wall shear stress
74
Velocity
The highest velocity magnitude was obtained during systole on outflow from CoA (4,15 m/s),
shown on velocity vectors (Fig. 5.7). Volume surrounding maximum velocity area shows also
significantly increased velocity in narrowing cross section at aortic arch, before CoA. At diastole (Fig. 5.7), highest velocity magnitude occurs at aortic arch, equal to 1,44 m/s.
Fig.5.7. Vectors of systole with CoA (zoomed)
Fig.5.8. Vectors of diastole with CoA (zoomed)
75
5.3
Particles residence time
According to positions of tracked in Lagrangian reference frame partitions shown in (fig.5.9)
is visible that particles eddy in domain due to turbulence caused by CoA. Part of (fig.5.9) where
particles residence time of total flow time is 0.9 sec (beginning of second calculated cycle),
shows that particles after first cardiac cycle still remains in domain reaching residence time
near 0.9 sec.
Fig.5.9. Particle residence time in chosen times of cardiac cycle
76
5.4
Diameter distribution
Particles diameter distribution in 0,7s of simulation (Fig.5.10) presents variety of diameter generated by EEDDM. Predominance of diameters near to 11 μm is noticeable due to assumed
mean diameter on that level.
Fig.5.11. Particle Residence time in 0,7s of simulation
6. Conclusions
In presented work practical applications of CFD modeling was shown. Blood flow in 8 years
old human female section of aorta and its main bifurcations was computed. Real geometry contains a congenital condition – Coarctation of the aorta (CoA), with about 65% of vasoconstriction. Two main calculation models was considered. Eulerian grid was used to calculate two
main fractions of morphological elements in blood: Plasma and RBCs, when Lagrangian grid
77
was used to track particles of WBCs. Work shows blood velocity distribution, static pressure,
wall share stress in two main cardiac cycle characteristic points (diastole and systole). The particle residence time of WBCs in domain was presented. Significant extension of time flow in
domain was revealed due to eddy caused by CoA.
An innovative way of modeling blood flow includes in pioneering discipline which is bio-technology. Great labor input is undertaking to constant improving interdisciplinary approach, nevertheless there is still need of engage more different fields specialists to get a wider picture.
Acknowledgements
I would like to acknowledge all project group members: Bartlomiej Melka, Wojciech
Adamczyk, Marek Rojczyk, for great time spent together during work on case, especially my
thesis supervisor Dr Ziemowit Ostrowski. I would like also to thank Adam Golda and Andrzej
J. Nowak for help and cooperation.
Literature:
[A] ANSYS, Inc. ANSYS Fluent. 2016. htpp://www.ansys.com
[C] CFD Challenge problem: http://www.vascularmodel.org/miccai2012 (accessed May 16th
2016).
[1] Wojciech P. Adamczyk, Adam Klimanek, Ryszard A. Białecki, Gabriel Węcel,Paweł
Kozołub, Tomasz Czakiert, Comparison of the standard Euler–Euler and hybrid Euler–Lagrange approaches for modeling particle transport in a pilot-scale circulating fluidized bed,
Particuology (2016).Elsevier B.V.,
[2] J. Ding and D. Gidaspow. A Bubbling Fluidization Model Using Kinetic Theory of Granular Flow AIChE J. (1990) 36(4):523–538
[3] F. Durst, D. Milojevic, B. Schönung, Eulerian and Lagrangian predictions of particulate
two phase flows: a numerical study, Appl. Math. Modelling (1984)8,
[4] S. Ergun. Fluid Flow through Packed Columns. Chem. Eng. Prog.(1952) 48(2):89–94
[5] D. Gidaspow, R. Bezburuah, and J. Ding. Hydrodynamics of Circulating Fluidized Beds,
Kinetic Theory Approach. In Fluidization VII, Proceedings of the 7th Engineering Foundation
Conference on Fluidization. 75–82 (1992)
[6] D. Gidaspow, V. Chandra, Unequal granular temperature model for motion of platelets to
the wall and red blood cells to the center, Chemical Engineering Science (2014) 117: 107-113
[7] J. Górski, Fizjologia człowieka, Wydawnictwo lekarskie PZWL (2010)
[8] J. Tu, G. Heng Yeoh, C. Liu, Computational Fluid Dynamics: A Practical Approach, Butterworth-Heinemann, (2007)
[9] Jung J and Hassanien A., Three-phase CFD analytical modelling of blood flow, Medical
Engineering & Physics 20:91-103, 2008.
[10] S.J. Konturek, Fizjologia czowieka, ELSEVIER Urban & Partner, 2007
[11] B. E. Launder and D. B. Spalding. Lectures in Mathematical Models of Turbulence. Academic Press, London, England (1972)
78
[12] C. K. K. Lun, S. B. Savage, D. J. Jeffrey, and N. Chepurniy. Kinetic Theories for Granular
Flow: Inelastic Particles in Couette Flow and Slightly Inelastic Particles in a General Flow
Field. J. Fluid Mech.(1984) 140:223–256
[13] F.D. Molina-Aiz, H. Fatnassi, T. Boulard, J.C. Roy, D.L. Valera, Comparison of finite
element and finite volume methods for simulation of natural ventilation in greenhouses, Computers and Electronics in Agriculture (2010) 72(2):69-86
[14] S. A. Morsi and A. J. Alexander. An Investigation of Particle Trajectories in Two-Phase
Flow Systems. J. Fluid Mech. 55(2). 193–208. September 26 1972.
[15] Patankar N.A and Joseph D.D. Lagrangian numerical simulation of particulate flows. International Journal of Multiphase Flow (2001) 27.
[16] A. Polanczyk, A. Piechota, CFD simulations of blood flow through abdominal part of
aorta, Challenges of modern technology (2010) 1
[17] W. Z. Traczyk, Fizjologia człowieka w zarysie, Wydawnictwo lekarskie PZWL, 2002.
[18] C.-Y. Wen and Y. H. Yu. Mechanics of Fluidization. Chem. Eng. Prog. Symp. Series. 62.
100–111. 1966.
Identyfikacja oraz modelowanie procesu przepływu krwi
w części dużego naczynia krwionośnego
z wykorzystaniem wielofazowego modelu hybrydowego Euler-Lagrange
Słowa kluczowe: aorta, CFD, przepływ krwi, ANSYS Fluent, Euler-Lagrange, Euler-Euler,
wielofazowy, DPM
Abstrakt
Obliczeniowa mechanika płynów (ang. CFD – Computational Fluid Dynamics) znana niegdyś
tylko w wysoce wyspecjalizowanej technicznie branży jest jednym z podstawowych narzędzi
inżynieryjnych w rozwiązywaniu wielu złożonych problemów, celem zdobycia kluczowych informacji i poszerzenia wiedzy ogólnej w wielu dziedzinach. CFD pozwala na tworzenie nowych, bardziej zaawansowanych systemów oraz na udoskonalanie już istniejących – poprawiając ich wydajność i/lub obniżając koszty produkcji oraz eksploatacji. Aktualna sytuacja wymaga od inżynierów zmierzenia się w trudnej dyscyplinie – walce o ułamki wydajności z powodu ograniczeń materiałowych i konstrukcyjnych. Przedsięwzięcia te nie koncentrują się jedynie na oczywistych dyscyplinach, takich jak przepływ ciepła, mechanika płynów czy wytwarzanie energii, ale także na nowych, niezbadanych sferach jak inżynieria motoryzacyjna, chemiczna, kosmiczna czy środowiska itd.
Jednym z innowacyjnych zastosowań CFD jest bio-inżynieria. W medycynie, symulacje komputerowe są w stanie dostarczyć niezbędnych, nierzadko ratujących życie informacji, bez ingerencji w ciało pacjenta (in vitro), co pozwala uniknąć późniejszych komplikacji, zagrożeń występujących w trakcie wprowadzania przyrządów w ciało pacjenta czy niebezpiecznych nieprzewidywalnych powikłań. Ponad to w wielu przypadkach metody in vivo są niemożliwe do
zastosowania ze względu na zagrożenie życia pacjenta.
79
Głównym celem powyższego projektu było stworzenie i testy innowacyjnego, dokładnego modelu przepływu krwi w ludzkiej aorcie.
Aktualnie dostępne badania nie uwzględniają przestrzennych interakcji pomiędzy poszczególnymi fazami krwi i ścianami naczyń krwionośnych. Takie podejście zdecydowanie
zmniejsza dokładność tego typu modeli. Odpowiednie badania wzbogacają wiedzę ogólną o
dokładne informacje , które mogą okazać się kluczowe we wczesnym diagnozowaniu problemów układu sercowo-naczyniowego, wskazując na potencjalnie podatne obszary (np. kurczące
się naczynia krwionośne). Tak dokładne informacje są trudno dostępne do uzyskania na drodze
badań.
Poza wielofazowa koncepcją projektu, który rozpatruje każdy komponent krwi jako oddzielną
fazę, przyporządkowując poszczególne właściwości do każdej z nich i uwzględniając ich wzajemne relacje, szczególną uwagę zwrócono na realistykę geometrii – zakładając rzeczywisty
układ aortalny (część aorty wstępującej, łuk aortalny i część aorty zstępującej) uwzględniający
bifurkację. Ponadto wprowadzono do modelu przepływ pulsacyjny za pomocą wbudowanej
wewnętrznej funkcji programu. (ang. UDF – User Defined Function).
80
c
Instytut
Techniki Cieplnej, Politechnika Ślaska
˛
Archiwum Instytutu Techniki Cieplnej Vol. 1 nr 1(2016)
www.itc.polsl.pl/ArchiwumITC
Analiza i modelowanie wymiany ciepła w obiektach
przemysłowych ogrzewanych promiennikami podczerwieni
Adrian Orłowski*
Słowa kluczowe: promienniki podczerwieni, promieniowanie, hala przemysłowa, ogrzewanie
Streszczenie
Przedmiotem pracy jest analiza i identyfikacja wymiany ciepła przez promieniowanie
w hali przemysłowej, opracowanie programu obliczeniowego oraz implementacja
opracowanego kodu w języku VisualBasic (VBA/MS Excel). Praca opisuje dwuwymiarowy
model hali, uwzględniający strefy zacienienia. Daje możliwość ustawienia w hali
prostokątnego przedmiotu, a więc najczęściej jest to używana w takich miejscach paleta Euro,
na której znajdują się przechowywane w hali przedmioty. Taki układ może być
zamodelowany z jednym lub dwoma promiennikami. Poddano analizie położenie palety
względem promiennika w celu jak najlepszego ogrzania hali przemysłowej oraz
wykorzystania dostępnej mocy. Omówiono także wyniki obliczeń zamodelowanego układu
w stanie ustalonym z różnymi wariantami, tzn. układ z zadaną temperaturą wewnętrzną na
stałym poziomie, układ z wymianą ciepła w hali wyłącznie na drodze promieniowania czy też
z temperaturą wewnętrzną w hali wyznaczaną na podstawie konwekcji z elementów budynku.
Uzyskane wyniki omówiono pod kątem wykorzystania algorytmu do wstepnego oszacowania
opromieniowania i odczuwalnej temperatury w hali przemysłowej.
Rozdział przygotowano podczas pracy nad projektem dyplomowym magisterskim wykonywanym przez autora
w Instytucie Techniki Cieplnej na Wydziale Inżynierii Środowiska i Energetyki Politechniki Śląskiej, pod opieką
Dra inż. Ziemowita Ostrowskiego.
*
Zawartość tej publikacji może byc wykorzystana na warunkach licencji Uznanie Autorstwa 3.0 Polska. Licencja pozwala na
kopiowanie, zmienianie, rozprowadzanie, przedstawianie i wykonywanie utworu jedynie pod warunkiem oznaczenia autorstwa: autora,
tytułu rozdziału, nazwy serii, tomu, strony. Content from this work may be used under the terms of the Creative Commons Attribution
3.0 licence. Any further distribution of this work must maintain attribution to the author, chapter title, series title, volume, pages.
81
Spis oznaczeń
a – absorpcyjność, A – powierzchnia, m2


B – opromieniowanie ( B i – opromieniowanie powierzchni i), W

E i – emisja własna powierzchni i, W
Fi  j – stosunek konfiguracji pomiędzy powierzchniami i-j, 
H – jasność promieniowania (całej powierzchni), W
L1, 2,3, 4 – długość odcinka 1,2,3,4, m

q – gęstość strumienia ciepła, W/m2
r – refleksyjność, T – temperatura, K
t – transmisyjność, W j – szerokość j-tej płytki, m
x, y - współrzędne geometryczne miejsca, m
 – współczynnik wnikania ciepła (przejmowania ciepła), W/(m2 .K)
 – grubość, m
 – emisyjność,  – współczynnik przewodzenia ciepła, W/(m·K)
 – stała Stefana-Boltzmana,   5,67  10 8 , W/(m2 ·K4)
Indeksy
g – gruntu (warstwy ziemi)
i – odnoszący się do i-tej powierzchni
j – odnoszący się do j-tej powierzchni
odcz  i – odczuwalna i-tej powierzchni
Ri – promieniowania i-tej powierzchni
śr – średnioroczna powietrza zewnętrznego
w – wewnętrzny
z – zewnętrzny
82
1
1.1
Wprowadzenie
Problematyka projektowania instalacji promiennikowych
Ogrzewanie obiektów budowlanych o dużych wymiarach, takich jak hale przemysłowe,
wymaga zwrócenia szczególnej uwagi na koszty energii zużytej do tego celu oraz zapewnieniu komfortu i bezpieczeństwa w miejscu przebywania osób. Istnieje wiele systemów rozwiązań ogrzewania obiektów przemysłowych, jednak z powodu cech promieniowania, takich jak
przenikanie przez powietrze prawie bez strat, możliwość ukierunkowania energii na wybraną
powierzchnię czy krótki czas nagrzewania, zużycie energii przy ogrzewaniu przez promieniowanie jest niższe niż w konwencjonalnych systemach grzewczych [4].
Przy projektowaniu instalacji ogrzewania promiennikami trzeba wziąć pod uwagę wiele
czynników. Należy określić warunki brzegowe, takie jak wymiary budynku, właściwości materiałów czy parametry otoczenia odpowiednie dla lokalizacji obiektu [3]. Ważnym elementem jest ustalenie charakterystyki użytkowania danego budynku, ponieważ to ten element
wpływa na temperaturę pomieszczenia jaka ma być osiągnięta. W zależności od przeznaczenia może to być tylko 17  C [4].
W wysokich, słabo izolowanych lub użytkowanych okresowo obiektach, takich jak hale
przemysłowe należy uwzględnić wyraźne różnice pomiędzy temperaturą powietrza i temperaturą promieniowania [4]. Są to podstawowe parametry projektowe komfortu cieplnego [3].
Komfort cieplny jest określany za pomocą średniej arytmetycznej temperatury powietrza oraz
temperatury promieniowania. Średnia ta nazywaną temperaturą odczuwalną została przedstawiona we wzorze (17). Ze wzoru tego wynika, że zwiększenie temperatury odczuwalnej może wynikać ze zwiększenia temperatury promieniowania o wartość, o którą została zmniejszona temperatura powietrza. Zależnie od aktywności fizycznej niższa temperatura powietrza
jest postrzegana jako bardziej komfortowa [4]. Do pewnego momentu odbywa się to bez
wpływu na odczuwalny komfort cieplny i prowadzi do oszczędności energii [9][3]. Temperatura odczuwalna na odpowiednim poziomie jest podstawowym i koniecznym warunkiem dla
uzyskania komfortu cieplnego. Należy jednak zwrócić uwagę na asymetrię promieniowania
i dopuszczalne natężenie promieniowania. Asymetria, która jest określana jako różnica temperatury efektywnej promieniowania, np. lewej i prawej półprzestrzeni, wpływa na wskaźnik
PPD (Predicted Percentage Dissatisfied) mówiący o przewidywanym procencie niezadowolonych z warunków cieplnych [5]. Przy projektowaniu ogrzewania promiennikami należy także
unikać nadmiernego natężenia promieniowania [4].
1.2
Modelowy obiekt
Niniejsza praca uwzględnia model dwuwymiarowy widoczny na Rysunku 1, dlatego też
najlepsze efekty daje rozpatrywanie w niej promienników rurowych, których to długości
sięgają do 20 metrów [9]. Dzięki tak dużym wymiarom w porównaniu z promiennikami
jasnymi zmniejsza się wpływ błędu jaki mógłby wynikać z braku uwzględnienia odbić
z każdego kierunku. Wzięte pod uwagę strefy zacienienia rzucane przez paletę oraz straty
ciepła przez przegrody dają możliwość na odpowiednie określenie temperatury,
opromieniowania oraz temperatury promieniowania każdego elementu.
83
Rysunek 1 przedstawia model hali przemysłowej z promiennikiem ustawionym nad
paletą. Opisany został każdy element wzięty pod uwagę przy analizowaniu i modelowaniu
przepływu ciepła, a także została zaznaczona przykładowa droga promieni, które pozwalają
pokazać powstałe strefy zacienienia. Nazwy te będą używane w dalszej części tekstu.
Rysunek 1: Dwuwymiarowy model hali przemysłowej.
1.3
Mechanizm wymiany ciepła w hali przemysłowej
W rozpatrywanym przypadku jedynym źródłem ciepła jest promiennik lub dwa promienniki podczerwieni umieszczone na suficie. W obliczeniach przyjęto, że promiennik jest
w tej samej płaszczyźnie co sufit. Z trzech stron, tj. ścian i sufitu następuje przenikanie ciepła
do otoczenia, którego temperaturę domyślnie przyjęto dla III strefy klimatycznej [7]. Natomiast wyznaczanie strat ciepła do gruntu, które opisuje norma [6] sprowadzono do podziału
na jednowymiarowe elementy przenikania ciepła, czyli podłogę i warstwę ziemi. Więcej
szczegółów o sposobie przenikania ciepła do gruntu opisano w rozdziale 1.5, w którym zostały opisane regulacje prawne dotyczące wymiany ciepła w budownictwie.
Określenie współczynnika przewodzenia ciepła materiału na palecie mogłoby być kłopotliwe z uwagi na to, że jest to element bez bezpośredniego kontaktu z powietrzem zewnętrznym. Jego wpływ na ogólną temperaturę w obiekcie determinowany jest głównie przez
jego kształt i rozmiar, które to wpływają na powstawanie stref zacienienia. Dlatego też przyjęto na stałym poziomie dwa różne współczynniki przewodzenia ciepła palety w zależności od
kierunku przepływu ciepła. Współczynnik przewodzenia ciepła dobrze przewodzących materiałów  =100 W/(m·K) dla kierunku pionowego, aby z jednej strony nie izolować całkowicie
podłogi w miejscu postawienia palety, a z drugiej strony uwzględnić straty ciepła przez paletę
do gruntu w tym miejscu. W kierunku poziomym przyjęto współczynnik przewodzenia ciepła
dla materiałów stanowiących pewną izolację równy  =0,1 W/(m·K), aby wyższe temperatury boku palety osiągnięte w przypadku większego padającego na nie opromieniowania nie
84
wpływały znacząco na temperaturę na drugim brzegu (przegrody) palety. Co w stanie ustalonym powodowałoby wyrównywanie temperatury boków palety i utrudniałoby odpowiednią
interpretację wpływu promieniowania.
W celu wygody opisu geometrii i bardziej obrazowego przedstawieni układu w dalszej
części tekstu podstawowa część podziału hali będzie nazywana na zmianę „elementem podziału” lub „płytką” zaznaczona na Rysunku 2 jako „i”.
Aby przedstawić równania użyte w algorytmie posłużono się pomniejszoną i uproszczoną o brak palety geometrią widoczną na Rysunku 2. Prezentuje on pomieszczenie z centralnie umieszczonym na suficie promiennikiem. Po lewej stronie hali na ścianie zaznaczono
jeden z elementów, na który podczas liczenia zostaje podzielona hala. Wszystkie elementy
podziału mają zawsze identyczne wymiary.
Rysunek 2: Hala przemysłowa z zaznaczonymi elementem podziału.
Pierwszym krokiem jest wyznaczenie stosunków konfiguracji pomiędzy elementami
podziału. Korzystając z [8] w równaniach (1) oraz (2) uwzględniono występowanie dwóch
możliwości ustawienia elementów hali, to znaczy przedstawione na Rysunku 3 w przypadku
a) dwie nieskończenie długie płytki które są do siebie równoległe oraz przedstawione na Rysunku 3 w przypadku b) dwie nieskończenie długie płytki, które są do siebie prostopadłe.
Możliwe ustawienia elementów przedstawiono dla elementu i z Rysunku 2. Dla przejrzystości
proporcje z Rysunku 2 nie zostały zachowane.
85
Rysunek 3: Ustawienia płytek wraz z oznaczeniami wykorzystanymi w równaniach (1) i (2).
Dwie płytki równoległe przypadek a) oraz dwie płytki prostopadłe przypadek b).
W przypadku a) z Rysunku 3, kiedy płytki są do siebie równoległe stosunek konfiguracji przyjmuje postać:
F j i 
L1  L2  L3  L4
2 W j
(1)
W j – szerokość elementu „j”, m
natomiast w momencie, kiedy płytki są do siebie prostopadłe to znaczy kąt   90  ,
jak zostało przedstawione na Rysunku 3 w przypadku b), to stosunek konfiguracji określamy
jako:
1
F j i
1
1
1
( x 2  y 22 ) 2  ( x 22  y12 ) 2  ( x22  y 22 ) 2  ( x12  y 22 ) 2
 1
2  ( x2  x1 )
(2)
W algorytmie w celu skrócenia obliczeń wykorzystana została zasada wzajemności
[2], którą zgodnie z wcześniejszym założeniem, że wszystkie płytki mają takie same wymiary,
a więc Ai  A j opisuje równanie:
Fi  j  F j i
(3)
Fi  j - część jasności powierzchni „i” trafiającej na powierzchnię „j” (stosunek konfiguracji), F j i - część jasności powierzchni „j” trafiającej na powierzchnię „i” (stosunek konfiguracji), -
86
Kolejny etap to rozpisanie równań bilansu jasności [2], jest on układany dla wszystkich płytek poza płytkami promiennika, którego jasność promieniowania jest znana, a opromieniowanie pomijalnie małe, ponieważ to promiennik jest jedynym źródłem ciepła, a jego
powierzchnia jest niewielka w stosunku do powierzchni całego układu:



H i  E i  ri  B i
(4)

B i – opromieniowanie powierzchni i, W

E i – emisja własna powierzchni i, W

H i – jasność promieniowania powierzchni i, W
ri – refleksyjność powierzchni i, -
gdzie emisja własna jest równa [2]:

E i   i    Ti  Ai
4
(5)
 – emisyjność,  – stała Stefana-Boltzmana,   5,67  10 8 , W/(m2 ·K4)
Ti – temperatura powierzchni i, K
Ai – powierzchnia i-tej płytki, m2
w obliczeniach pomijane jest opromieniowanie powierzchni promiennika. Dla pozostałych
elementów jest liczone, jako [2]:


B i   H j  F j i
(6)
j
Przyjęto, że transmisyjność jest równa zero t  0 oraz emisyjność jest równa absorpcyjności
  a [2], dlatego refleksyjność określa zależność:
r  1 
(7)
Sposób przenikania ciepła przez przegrody został przedstawiony na dwóch przykładach dla sufitu i ścian oraz osobno dla podłogi. Na Rysunku 4 zostały przedstawione kierunki
strumieni ciepła oraz gęstości strumieni ciepła w bilansie energii powierzchni ściany lub sufitu. Przykład dla i-tej płytki z Rysunku 2.
87
Rysunek 4: Bilans energii dla ściany lub sufitu. Przykład dla i-tej płytki.

q i – gęstość strumienia ciepła wymienianego pomiędzy elementem i a powietrzem wewnątrz
hali na drodze konwekcji, W/m2

q zi – gęstość strumienia ciepła wymienianego pomiędzy elementem i a otoczeniem, W/m2
 i – współczynnik wnikania ciepła pomiędzy elementem i a powietrzem w hali, W/(m2 .K)
 z – współczynnik wnikania ciepła między elementem hali a powietrzem otoczenia,
W/(m2 .K)
 i – grubość elementu i, m
i – współczynnik przewodzenia ciepła elementu i, W/(m.K)
Gęstości strumienia ciepła na Rysunku 4 wynoszą odpowiednio:

qi 
Ti  Tw
1
(8)
i

q zi 
Ti  Tz
i 1

i
88
z
(9)
dlatego bilans energii wykorzystując równania (7) i (8) jest równy:
Ai 


Ti  Tw
T  Tz
 Ai  i
 ( H i  B i )
1
i 1

i
i
(10)
z
Na Rysunku 5 podobnie jak na Rysunku 4 zostały opisane oznaczenia i zaznaczone
kierunki przepływu ciepła w bilansie energii dla podłogi. Przykład dla i-tego elementu położonego na podłodze.
Rysunek 5: Bilans energii dla podłogi. Element i znajduje się na podłodze.
 g – grubość warstwy ziemi, m
 g – współczynnik przewodzenia ciepła gruntu, W/(m.K)
Tśr – temperatura średnioroczna powietrza zewnętrznego jako temperatura gruntu [1], K
Gęstości strumienia ciepła z Rysunku 5 wynoszą odpowiednio:

qi 
Ti  Tw
1
(11)
i
89

q zi 
Ti  Tśr
(12)
i  g

i  g
Zapisując bilans energii dla podłogi z wykorzystaniem równań (11) i (12) otrzymujemy:
Ai 


Ti  Tw
T  Tśr
 Ai  i
 ( H i  B i )
1
i  g

i
i
(13)
g
Równanie wiążące temperaturę wewnętrzną w hali z temperaturami elementów budynku oraz jasnością promieniowania promiennika:



H promiennika    ( H i  B i )   Ai 
i
i
Ti  Tw
1
(14)
i
Wyznaczona została także temperatura promieniowania TR , która jest zastępczą temperaturą wszystkich powierzchni wewnętrznych hali (widocznych z danego miejsca). Temperatura ta odpowiada zastępczej średniej temperaturze dużego otoczenia danej powierzchni hali
odpowiadającej za wymianę ciepła na drodze promieniowania pomiędzy wybranym elementem a tym otoczeniem. Temperatura promieniowania została wyznaczona, z bilansu energii:

  
Ai  ei  b i   Ai      Ti 4  TRi4



(15)
po przekształceniu równania (15), TRi jest równe:

TRi  4
Bi
 i    Ai
(16)
temperaturę promieniowania wykorzystuje się do obliczenia temperatury odczuwalnej, która
określona jest wzorem:
Todczi 
1.4
Tw  TRi 
2
(17)
Model EES
Model przedstawiony na Rysunku 6 prezentuje pomieszczenie o wymiarach 2x3 m
z centralnie umieszczonym na suficie promiennikiem reprezentowanym przez płytkę numer 2.
90
został wykonany w programie Engineering Equation Solver, wykorzystując równania omówione w punkcie 1.2. Obliczony przykład zostanie porównany przy wykorzystaniu identycznych danych z wynikami uzyskanymi w zaimplementowanym algorytmie w języku VBA
w programie MS Excel. Porównanie ma na celu określenie dokładności i poprawności stworzonego algorytmu.
Rysunek 6: Schemat pomieszczenia użyty do obliczeń sprawdzających dokładność algorytmu.
Przyjęte dane:
Tz  253 K – temperatura zewnętrzna otoczenia dla III strefy klimatycznej,
Tśr  280,6 K – temperatura średnioroczna powietrza zewnętrznego,
 z  25 W/(m2 .K) – współczynnik wnikania ciepła do powietrza zewnętrznego,
 g  7 m – grubość warstwy ziemi,
  0,95 – emisyjność dla każdej płytki,
 g  2 W/(m.K) – współczynnik przewodzenia ciepła dla gruntu,
Dla płytek sufitowych, tj. i  1, 2, 3 :
 i  10 W/(m2 .K) – współczynnik wnikania ciepła pomiędzy powietrzem w hali a sufitem,
 i  0,1 m – grubość sufitu,
91
i  0,03 W/(m.K) – współczynnik przewodzenia ciepła dla sufitu,
Dla płytek na bokach hali, tj. i  4, 5, 6, 7 :
 i  7,69 W/(m2 .K) – współczynnik wnikania ciepła pomiędzy powietrzem w hali a ścianą,
 i  0,1 m – grubość ścian,
i  0,03 W/(m.K) – współczynnik przewodzenia ciepła dla ścian,
Dla płytek na podłodze, tj. i  8, 9, 10 :
 i  10 W/(m2 .K) – współczynnik wnikania ciepła pomiędzy powietrzem w hali a podłogą,
 i  0,3 m – grubość podłogi,
i  1,7 W/(m.K) – współczynnik przewodzenia ciepła dla podłogi.
Współczynniki przewodzenia ciepła i grubości ścian i sufitu przyjęto według danych
dla płyt warstwowych znajdujących się w kartach katalogowych producentów hal przemysłowych [10][11]. Płyty te mają grubości od 0,04 m do ponad 0,1 m i składają się z dwóch okładzin z blachy stalowej oraz z rdzenia konstrukcyjno-izolacyjnego [11]. Zastępcze współczynniki przewodzenia ciepła dla płyt mieszczą się w granicach   0,04 W/(m.K) dla płyt z rdzeniem styropianowym, do około   0,022 W/(m.K) dla płyt ze sztywna pianką poliuretanową.
Do najważniejszych elementów konstrukcji podłogi należą płyta betonowa oraz podbudowa. Grubość płyty betonowej wynosi od 0,12 do 0,3 m, natomiast grubość podbudowy
0,15-0,4 m [12]. Podbudowa może się składać z odpowiednio przygotowanej mieszanki
piaszczysto-żwirowej, dlatego w przyjętych danych została zaliczona do warstwy ziemi
o współczynniku przewodzenia ciepła  g  2 W/(m.K). Współczynnik przewodzenia ciepła
dla podłogi został więc określony dla betonu zwykłego z kruszywa kamiennego według normy [6].
Na Rysunku 7 zgodnie z numeracją z Rysunku 6 umieszczono wyniki uzyskane
w programie EES, tj. temperaturę powierzchni płytek oznaczoną jako Ti , temperaturę radia
cyjną TRi wyznaczoną według wzoru (16) oraz opromieniowanie oznaczone jako B i . Linią
przerywaną zaznaczono promiennik (płytka numer 2), nad którym wpisano założoną gęstość

strumienia ciepła oznaczoną jako H 2 .
Nie jest to geometria rzeczywista, a jedynie służąca do określenia poprawności obliczeń, dlatego uzyskane wyniki nie będą omówione pod kątem wymiany ciepła, a jedynie pod
kątem zgodności wyników.
92
Rysunek 7: Wyniki uzyskane w programie EES dla przykładowej geometrii.
Dla zwiększenia przejrzystości, jednostki dla temperatury powierzchni płytek, temperatury promieniowania i opromieniowania opisano osobno na środku Rysunku 7.
Wyniki uzyskane w programie MS Excel przedstawia Rysunek 8 z numerowaniem
płytek analogicznym do Rysunku 6. Promiennik został oznaczony kolorem czarnym, wewnątrz pól opisano do czego odnoszą się wyniki i jakie mają jednostki.
93
Rysunek 8: Wyniki dla przykładowej geometrii uzyskane w programie Excel.
Porównując Rysunki 7 i 8 można zauważyć, że wyniki są do siebie bardzo zbliżone.
Temperatury nie różnią się bardziej niż o 1 K a opromieniowanie o 2 W/m2. Różnice wynikają z dokładności wprowadzonej w narzędziu programu MS Excel - Solver, który jest wykorzystywany do rozwiązywania układu równań oraz w szczególności z wprowadzonego punktu
startowego. Trzeba jednak zauważyć, że jest to nieduży symetryczny układ, gdzie różnice
wartości nie są duże. W układach ze strefami zacienienia, ze względu na sposób działania
Solvera, może miejscami dochodzić do zwiększenia błędu, np. na granicy płytek zasłoniętej
i opromieniowanej.
Dodatkowy błąd w układzie ze strefami zacienionymi może także wynikać z niespełnienia we wszystkich płytkach zasady zamkniętości. W programie obliczeniowym przyjęto
zasadę, że jeśli płytki nie są dla siebie widoczne w całości to stosunek konfiguracji pomiędzy
nimi wynosi zero. Im większa ilość płytek, na którą podzielona jest hala, tym poszczególne
stosunki konfiguracji między płytkami są mniejsze, a to z kolei wpływa na mniejszy błąd całego układu. Suma wszystkich stosunków konfiguracji według zasady zamkniętości powinna
być równa liczbie płytek. W rzeczywistości jednak z powodu przyjętego sposobu wyznaczania stref zacienienia różnica pomiędzy teoretyczną sumą stosunków konfiguracji a rzeczywistą jest równa około 0,7-1,1. W porównaniu do całego układu, który ma np. 200 elementów
(płytek) to niewiele. Pamiętać jednak trzeba, że błąd ten nie jest rozdzielony równo na każdą
płytkę a największa niedokładność występuje na suficie nad paletą, ponieważ bywa, że
z dwóch stron palety, uznane za niewidzące się płytki, pada stosunkowo duże potencjalne
promieniowanie.
94
1.5
Regulacje prawne w budownictwie dotyczące wymiany ciepła
Producenci podając parametry cieplne ścian i sufitów hal wykorzystują współczynnik
przenikania ciepła U W/(m2 .K) w literaturze technicznej (z dziedziny wymiany ciepła) oznaczany zwykle jako k . Aby odpowiednio zinterpretować podane współczynniki i uzyskać dodatkowe informacje na temat użytych do budowy hal materiałów należy zapoznać się
z normami obowiązującymi w Polsce. Oprócz uwzględnienia przenikania przez ściany należy
także uwzględnić przenikanie ciepła do gruntu.
W tabeli 1 zamieszczono określone normą [6] opory przejmowania ciepła na powierzchni, które to zostały następnie przeliczone na współczynniki wnikania ciepła. Dla
omawianych przypadków geometrii oraz temperatur przyjęto, że kierunek strumienia ciepła
dla sufitu i podłogi będzie zawsze pionowy o zwrocie w górę. Założenie to wynika ze stałego
usytuowania promiennika na suficie, dlatego podłoga na drodze radiacji nagrzewa się i oddaje
ciepło na drodze konwekcji swobodnej do powietrza w hali. Natomiast na sufit trafia promieniowanie wyłącznie odbite z innych elementów hali, a jego temperatura rośnie także przez
unoszące się w górę podgrzane powietrze. Prowadzi to do sytuacji, gdzie pionowy kierunek
strumienia ciepła będzie miał zawsze zwrot w górę. Dla ścian natomiast przyjęto poziomy
kierunek strumienia ciepła.
Tabela 1: Opory przejmowania ciepła na powierzchni [6].
Kierunek strumienia ciepła
Opór przejmowania ciepła, (m2 .K)/W
W górę
Poziomo
W dół
Wewnątrz
Na zewnątrz
1

1

 Rsi
0,1
0,13
0,17
 Rse
0,04
0,04
0,04
W przypadku strat ciepła do gruntu nie uwzględniano całego algorytmu przedstawionego
w normie [7] z uwagi na rozpatrywanie układu dwuwymiarowego hali, ale dokonano porównania przykładowego obiektu, tj. hali o wymiarach podłogi 30x80 m i wyznaczono grubość
warstwy ziemi na poziomie około 7,5 metra. Nie odbiega to znacznie od funkcjonującego
w polskiej literaturze z zakresu budownictwa i inżynierii środowiska przybliżenia, które mówi
o stałej temperaturze gruntu na głębokości poniżej 8 metrów równej średniorocznej temperaturze powietrza [1].
2
2.1
Prezentacja programu obliczeniowego
Opis programu
Przed pierwszym włączeniem koniecznie należy sprawdzić czy Excel posiada zainstalowany dodatek Solver, czy w oknie „Project – VBAProject” edytora VisualBasic istnieje
opcja „Reference to solver.xla” oraz czy są włączone odwołania cykliczne (iteracje).
95
Po otworzeniu pliku pojawi się ekran informujący o tytule pracy, imię i nazwisko autora
oraz opiekuna pracy magisterskiej. Należy kliknąć przycisk „Przejdź do obliczeń”. Nastąpi
wtedy otworzenie się okna głównego zatytułowanego „Wymiary hali”, zawierającego dwie
zakładki: „Dane” przedstawione na Rysunku 9 oraz „Dodatkowe” przedstawione na Rysunku
10.
W pierwszej zakładce, którą przedstawia Rysunek 9, istnieje możliwość doboru najważniejszych parametrów, takich jak: ilość palet i promienników, rozmiarów hali, rozmiarów
promienników i gęstości strumieni ciepła. Po prawej stronie pokazuje się odpowiedni do zaznaczonej ilości palet i promienników rysunek z oznaczeniami. Po wpisaniu właściwych danych, aby uruchomić obliczenia należy kliknąć przycisk „LICZ”.
Rysunek 9: Okno główne zawierające najważniejsze dane odnośnie modelowanego obiektu.
96
Rysunek 10: Zakładka „Dodatkowe” zawierająca szereg parametrów określających układ oraz
wpływających na działanie programu.
Zakładka „Dodatkowe”, przedstawiona na Rysunku 10, zawiera 6 pól umożliwiających zmianę sposobu liczenia, parametrów otoczenia i materiałów, z których zbudowany jest
obiekt. W każdym polu znajduje się pomoc opisująca znaczenie poszczególnej opcji.
W polu numer 1 można określić dokładność działania programu poprzez określenie liczby
płytek, na które ma być podzielona hala. Maksymalna liczba płytek może być równa 200. Jest
to ograniczenie wbudowanego w program MS Excel Solvera. Trzeba pamiętać jednak, że
zwiększając liczbę elementów, na które zostanie podzielona hala zwiększa się także czas obliczeń.
Znajduje się tam również opcja „Szybkie liczenie”, która pozwala na zmniejszenie
czasu obliczeń o blisko połowę kosztem dokładności wyników. W przypadku temperatur jest
to najczęściej 3-5 K. Jest to przydatna opcja, jeśli użytkownik chce sprawdzić czy podane
założenia dają rzeczywiste wyniki. Polega ona na zmianie ustawień narzędzia Solver na mniej
dokładne. Mogą wystąpić skrajne sytuacje zależne od ustawień parametrów hali, kiedy wyniki
uzyskane przy włączonej opcji „Szybkie liczenie” będą równie dokładne jak przy wyłączonej,
a także, gdy będą posiadały większy błąd niż 3-5 K dla pola temperatur. Dlatego nie zaleca się
używania tej opcji przy ostatecznych obliczeniach.
Opcja „wykresy czarno białe” pozwala od razu przedstawić wyniki w odcieniach szarości. Istnieje możliwość późniejszej zmiany przedstawienia wyników bez konieczności przeliczania układu.
W obszarze oznaczonym jako numer drugi można zmienić temperaturę zewnętrzną
powietrza, średnioroczną powietrza zewnętrznego oraz wewnętrzną hali. Dwie pierwsze moż-
97
liwości pozwalają na umiejscowienie hali w dowolnej strefie klimatycznej. Użytkownik może
wybrać także trzy sposoby liczenia temperatury wewnętrznej. To znaczy, że może ją określić
na podstawie konwekcji z nagrzanych przez promieniowanie elementów hali, określić stałą
wartość temperatury, np. kiedy hala jest często wietrzona lub całkowicie pominąć wymianę
ciepła z poszczególnych elementów do wnętrza hali.
W polach 3, 4, 5, 6 określane są poszczególne parametry elementów obiektu, otoczenia i gruntu pod budynkiem. Domyślne wartości odpowiadają obowiązującym normom oraz
parametrom najczęściej występującym na rynku halom przemysłowym.
2.2
Wyniki
Wyniki są prezentowane w czterech Arkuszach nazwanych „Temperatura”, „Opromieniowanie”, „Temperatura promieniowania” oraz „Wykres”. Za pomocą przycisków, znajdujących się w trzech pierwszych arkuszach, można wybrać sposób kolorowania wyników między
barwami od niebieskiego do czerwonego lub w odcieniach szarości. W arkuszu „Wykres”
przedstawione zostało porównanie rozkładów temperatury, opromieniowania i temperatury
promieniowania na podłodze.
Po najechaniu kursorem na każdą z płytek otwiera się komentarz z szeregiem informacji
o parametrach zamodelowanej hali oraz o parametrach konkretnego elementu. Opcja ta umożliwia szybki podgląd wyników bez konieczności przełączania arkuszy za każdym razem.
W każdym rysunku zawierającym wyniki dodano zakresy wyników nad, obok i pod
płytkami ze zwiększoną czcionką w celu lepszej czytelności. W programie Excel nie jest to
konieczne ze względu na większe pole, na którym jest wyświetlany schemat oraz możliwość
przybliżenia dowolnej części wyników.
Przy przedstawianiu wyników w odcieniach szarości promiennik zawsze oznaczany jest
kolorem białym, jako puste miejsce na suficie.
98
Pierwszą zamodelowanym obiektem będzie hala bez palety o parametrach:
Wysokość: X  10 m
Szerokość: Y  20 m
Odległość lewej ściany do krawędzi promiennika: L  9,5 m
Szerokość promiennika: E  0,5 m. Rozmiary promiennika z [9].

Gęstość strumienia ciepła promiennika: q  1000 W/m2. Gęstość strumienia ciepła promiennika została oszacowana na podstawie bilansu energii w hali dla stanu ustalonego.
Oznaczenia zgodne z opisem z Rysunku 9. Wartości innych współczynników przyjęto
jak na Rysunku 10, zmieniono jedynie ilość płytek na 120 z powodu wystarczającej dokładności dla hali bez palety. Obiekt został podzielony na 120 płytek o wymiarze 0,5 m każda.
Może wystąpić sytuacja, kiedy ilość płytek, na które zostanie podzielony układ będzie mniejsza niż założona przez użytkownika. Jest to spowodowane większą dbałością o proporcjonalność układu niż zmniejszania poszczególnych elementów.
Temperatura powietrza w hali wyniosła 299,4 K.
Rozkład temperatury płytek pokazano na Rysunku 11:
Rysunek 11: Rozkład temperatury płytek w hali. Wartości w K.
Na Rysunku 11, zawierającym rozkład temperatury płytek, można zauważyć, że największe wartości mają płytki na podłodze bezpośrednio pod promiennikiem. Temperatury te
maleją wraz z oddalaniem się od środka hali. Najniższe temperatury są natomiast na górnej
części lewej i prawej ściany a nie na suficie. Dzieje się tak, ponieważ wewnętrzny współczynnik wnikania ciepła z powietrza do ścian jest mniejszy niż pionowy wewnętrzny współczynnik wnikania ciepła dla sufitu. Dodatkowo na sufit trafia odbite wprost promieniowanie
z podłogi, która ma wyższą temperaturę niż ściany.
Rysunek 12 przedstawia opromieniowanie w hali:
99
Rysunek 12: Rozkład opromieniowania w hali. Wartości podane w W/m2.
Podobnie jak w rozkładzie temperatury największe opromieniowanie jest bezpośrednio pod promiennikiem. W narożnikach u dołu hali większa wartość opromieniowania występuje na ścianach, a nie jak to było w przypadku temperatury na podłodze. Ma to związek
z brakiem wpływu przenikania ciepła. Podobną sytuację można zauważyć w górnych narożnikach, gdzie wpływ konwekcji przesądza o wyższej temperaturze płytek sufitu niż sąsiadującej z nim płytki ściany a opromieniowanie jest tam najmniejsze.
Rysunek 13 przedstawia temperaturę promieniowania w hali:
Rysunek 13: Rozkład temperatury promieniowania w hali. Wartości w K.
Rozkład temperatury promieniowania pokrywa się z rozkładem opromieniowania. Jak wynika
z wzoru (16) duże znaczenie w tym przypadku ma emisyjność każdego elementu, która została przyjęta na poziomie ciemnej matowej farby, tj. 0,95. Wartości temperatury promieniowania są większe od wartości temperatury płytek z Rysunku 11.
100
Na Rysunku 14 przedstawiono wykres porównujący rozkład uzyskanych wyników na
podłodze hali. Temperaturę zaznaczono linią z kołami, temperaturę promieniowania jako linię
z trójkątami, a punkty na linii opromieniowania zaznaczono kwadratami.
Rysunek 14: Wykres rozkładu temperatury płytek, temperatury promieniowania
i opromieniowania na podłodze.
Po lewej stronie znajduje się skala w K dla temperatur natomiast po prawej stronie została
dodana skala w W/m2 dla opromieniowania. Na osi poziomej umieszczono odległości od lewej strony hali. Na wykresie można zauważyć, że temperatura opromieniowania jest wyższa
i bardziej zmienna na szerokości całej podłogi niż temperatura poszczególnych płytek. Natomiast opromieniowanie, co było też widoczne na Rysunku 12, osiąga najwyższą wartość równą 466 W/m2 na środku hali pod promiennikiem.
Drugim zamodelowanym obiektem będzie hala z paletą o parametrach:
Wysokość hali: X  10 m
Szerokość hali: Y  20 m
Wysokość palety: I  4,2 m
Szerokość palety: J  2,8 m
Odległość lewej ściany hali do krawędzi palety: D  14 m
Odległość lewej ściany hali do krawędzi promiennika: L  9,5 m
Szerokość promiennika: E  0,5 m. Rozmiary promiennika z [9].

Gęstość strumienia ciepła promiennika: q  1000 W/m2.
Oznaczenia zgodne z opisem z Rysunku 9. Wartości innych współczynników przyjęto jak na
Rysunku 10. Ilość płytek wyznaczono na 180, co spowodowało podzielenie obiektu równo na
180 elementów o wymiarze 0,38 m każdy. Temperatura powietrza w hali wyniosła 299,5 K.
Rozkład temperatury płytek w hali z paletą pokazuje rysunek 15:
101
Rysunek 15: Rozkład temperatury płytek w hali przemysłowej z paletą. Wartości w K.
Na rysunku widać wyraźnie wyższą temperaturę na podłodze pod promiennikiem oraz
na lewej ścianie palety. Temperatura podłogi wynika głównie z największego w tym miejscu
opromieniowania, co widać na Rysunku 16. Natomiast wysoka temperatura po lewej stronie
palety jest spowodowana tym, że straty ciepła na tych płytkach są małe, ponieważ są skutkiem
wnikania ciepła do powietrza, którego temperatura jest równa 299,5 K oraz przewodzenia
przez paletę, której temperatura na drugim końcu jest podobnie wysoka i wynosi około 299,3
K. Najniższe temperatury płytek z powodu braku bezpośredniego padania promieni znajdują
się w prawym dolnym rogu hali i wynoszą około 298 K.
Rysunek 16 przedstawia opromieniowanie w hali z paletą:
Rysunek 16: Rozkład opromieniowania w hali przemysłowej z paletą. Wartości w W/m2.
Na rozkładzie opromieniowania z Rysunku 16 widać oznaczone jaśniejszym kolorem
miejsca, na które rzucany jest cień przez paletę. Widać też mniejsze opromieniowanie na pra-
102
wej ścianie hali oraz na suficie. Oprócz stref zacienienia jest to spowodowane pozycją promiennika, który nie jest ustawiony na samym środku hali, lecz przesunięty jest o półtora metra
w lewą stronę.
Rysunek 17 przedstawia temperaturę promieniowania w hali:
Rysunek 17: Rozkład temperatury promieniowania w hali z paletą. Wartości w K.
Temperatury promieniowania mieszczą się w granicach od 302 K w dolnym prawym
rogu hali, gdzie nie dociera promieniowanie do 305,8 K w miejscu na podłodze, gdzie opromieniowanie jest największe. Są też mniej wyrównane w porównaniu do temperatur powierzchni płytek z Rysunku 15, co najlepiej widać u dołu lewej ściany hali. Wpływ na to ma
brak wpływu przenikania ciepła do otoczenia jak w przypadku temperatury powierzchni elementów oraz większe znaczenie stosunków konfiguracji, które w związku ze sposobem ich
wyznaczania zależnie od miejsca płytki na ścianie posiadają na przemian większy lub mniejszy błąd. Błąd ten nie jest znaczący jednak w związku z automatycznym dobieraniem kolorów
widoczny w postaci raz za razem zmieniających się odcieni.
Rysunek 18 przedstawia wykres temperatur i opromieniowania na podłodze.
103
Rysunek 18: Wykres rozkładu temperatury płytek, temperatury promieniowania
i opromieniowania na podłodze.
Puste miejsce na wykresie z Rysunku 18, przedstawiającym rozkład temperatur i opromieniowania na podłodze hali, oznacza miejsce, w którym jest ustawiona paleta. Można zauważyć mniejszą zmienność temperatury płytek w porównaniu do temperatury promieniowania.
Widać również podwyższone temperatury i opromieniowanie na podłodze blisko palety. Może to być spowodowane w niewielkim stopniu, jak wspomniano w odniesieniu do Rysunku
17, błędami przy wyznaczaniu stref zacienienia, a także sytuacją gdzie płytka na podłodze jest
w szczególności dobrze opromieniowana ze ścian palety.
Trzecim zamodelowanym obiektem będzie hala z paletą o parametrach:
Wysokość hali: X  10 m
Szerokość hali: Y  20 m
Wysokość palety: I  5,2 m
Szerokość palety: J  2,8 m
Odległość lewej ściany hali do krawędzi palety: D  5 m
Odległość lewej ściany hali do krawędzi promiennika: L  9,5 m
Szerokość promiennika: E  0,5 m. Rozmiary promiennika z [9].

Gęstość strumienia ciepła promiennika: q  5000 W/m2. Gęstość strumienia promiennika
została wyznaczona na podstawie danych ciemnych gazowych promienników rurowych z [9].
Temperatura powietrza w hali została określona na stałym poziomie 283 K.
Oznaczenia zgodne z opisem z Rysunku 9. Wartości innych współczynników przyjęto jak
na Rysunku 10. Układ został podzielony na 200 płytek o wymiarze 0,352 m każda.
Rysunek 19 przedstawia temperaturę powierzchni elementów w budynku. Widoczny jest
znaczny wpływ określonej na stałym poziomie temperatury w hali. Tylko na najmniej opromieniowanych elementach w górnym i dolnym rogu lewej strony hali temperatura płytek jest
niższa niż 283 K.
104
Rysunek 19: Rozkład temperatury z ustaloną temperaturą powietrza wewnątrz hali na poziomie 283 K. Wartości w K.
Na rysunku 20 przedstawiono rozkład opromieniowania. Jak we wcześniejszych przykładach
widoczne jest największe opromieniowanie na podłodze pod promiennikiem, a także na prawym górnym rogu palety. Jeśli celem dla montującego promiennik było zapewnienie komfortu cieplnego na podłodze, to wysokie opromieniowanie znajdującego się na wysokości ponad
5 metrów prawego narożnika palety świadczy o złym ułożeniu palety lub promiennika.
Rysunek 20: Rozkład opromieniowania z ustaloną temperaturą powietrza wewnątrz hali na
poziomie 283 K. Wartości w W/m2.
Temperatura promieniowania została przedstawiona na Rysunku 21. Można zauważyć, że
paleta o wysokości 5,2 m całkowicie zasłania całą podłogę po jej lewej stronie. A największe
wartości temperatury są osiągane są na prawym górnym narożniku palety.
105
Rysunek 21: Rozkład temperatury opromieniowania z ustaloną temperaturą powietrza na poziomie 283 K. Wartości w K.
Na Rysunku 22 przedstawiony został wykres temperatury płytek, opromieniowania oraz temperatury promieniowania na podłodze. Widać dużo większy przyrost temperatury promieniowania niż temperatury powierzchni płytek porównując podłogę po lewej i prawej stronie palety.
Rysunek 22: Wykres rozkładu temperatury płytek, temperatury promieniowania
i opromieniowania na podłodze w hali ze stałą temperaturą wewnętrzną równą 283 K.
Czwartym zamodelowanym obiektem będzie hala z paletą o parametrach:
Wysokość hali: X  10 m
Szerokość hali: Y  20 m
Wysokość palety: I  5,2 m
Szerokość palety: J  2,8 m
106
Odległość lewej ściany hali do krawędzi palety: D  5 m
Odległość lewej ściany hali do krawędzi promiennika: L  9,5 m
Szerokość promiennika: E  0,5 m. Rozmiary promiennika z [9].

Gęstość strumienia ciepła promiennika: q  1000 W/m2. Gęstość strumienia promiennika
została wyznaczona na podstawie bilansu energii dla stanu ustalonego.
Wnikanie ciepła między elementami hali a powietrzem wewnętrznym zostało pominięte,
dlatego temperatura w hali nie została określona a wymiana ciepła w hali odbywa się tylko na
drodze promieniowania.
Oznaczenia zgodne z opisem z Rysunku 9. Wartości innych współczynników przyjęto jak
na Rysunku 10. Układ został podzielony na 200 płytek o wymiarze 0,352 m każda.
Na Rysunku 23 przedstawiono rozkład temperatury płytek w hali:
Rysunek 23: Rozkład temperatury w hali bez uwzględnienia konwekcji do powietrza wewnętrznego. Wartości w K.
Porównując Rysunek 23 z Rysunkiem 15, który pomimo innej pozycji palety ma podobne
parametry, takie jak wymiary hali oraz gęstość strumienia ciepła promiennika, można zauważyć, że temperatura płytek bezpośrednio pod promiennikiem jest niższa o około 6 K a w miejscach nieopromieniowanych różnica ta jest znacznie większa i wynosi 24 K. Świadczy to
o dużym wpływie wnikania ciepła pomiędzy elementami obiektu a powietrzem wewnątrz
budynku oraz temperatury powietrza w hali na ogólny rozkład temperatur.
Rozkład promieniowania został opisany na Rysunku 24. Widać na nim jak miejsce
ustawienia palety wyznacza wyraźną granicę, od której na lewo opromieniowanie hali jest
mniejsze. Można zauważyć, że nawet odbite promieniowanie w dużo mniejszym stopniu trafia do lewej części hali porównując z nieopromieniowanym bezpośrednio przez promiennik
sufitem nad prawą częścią podłogi.
107
Rysunek 24: Rozkład opromieniowania w hali bez uwzględnienia konwekcji do powietrza
wewnętrznego. Wartości w W/m2.
Na Rysunku 25 widać rozkład temperatury promieniowania w hali. Największa temperatura
jest osiągana na podłodze pod promiennikiem oraz na prawym narożniku palety. Można zauważyć, że cała hala po prawej stronie palety ma wyższe temperatury promieniowania.
Rysunek 25: Rozkład temperatury promieniowania w hali bez uwzględnienia konwekcji do
powietrza wewnętrznego. Wartości w K.
Na rysunku 26 widoczny jest wykres temperatury płytek, opromieniowania i temperatury
promieniowania na podłodze w hali. Widać dużo większy przyrost temperatury płytek w porównaniu do przykładów hali z uwzględnieniem wnikania ciepła do powietrza wewnętrznego.
108
Rysunek 26: Wykres rozkładu temperatury płytek, temperatury promieniowania
i opromieniowania na podłodze bez konwekcji do wewnątrz hali.
3
Podsumowanie i wnioski
Wyniki uzyskane w niniejszej pracy magisterskiej potwierdzają konieczność odpowiedniej analizy promieniowania w celu najbardziej korzystnego wykorzystania energii, co widać
na przykładzie promiennika ustawionego za blisko palety, który ogrzewa niepotrzebnie jej
nieużywaną część. Pamiętać jednak trzeba, że algorytm bierze pod uwagę stan ustalony
w hali, który w rzeczywistych warunkach nie jest osiągalny. Stan ustalony nie jest osiągalny
nie tylko z powodu wpływu takich czynników jak wietrzenie hali, ale też ze sposobu wykorzystania samych promienników. W rzeczywistych warunkach wykorzystywane są zalety
promienników, takie jak krótki czas reakcji, natychmiastowy miejscowy efekt grzania, co
skutkuje brakiem ciągłości pracy promienników. Najważniejsza jednak dla osób w hali jest
odczuwalna temperatura w miejscu pracy, na którą składa się temperatura otoczenia oraz temperatura promieniowania. Dlatego też istnieje możliwość symulacji różnych warunków.
W przykładzie z ustaloną temperaturą wewnątrz hali na poziomie 283 K i gęstości strumienia
ciepła odpowiadającej rzeczywistym urządzeniom, temperatura odczuwalna, jako średnia
temperatury powierzchni i promieniowania wynosi około 299 K, co pozwala na komfort
cieplny nawet w pracy niewymagającej ruchu.
109
Literatura
[1] B. Biernacka, Miesięcznik CIEPŁOWNICTWO, OGRZEWNICTWO, WENTYLACJA
9/2012, Badania rozkładów temperatury w gruncie na różnych głębokościach.
[2] E. Kostowski, Przepływ ciepła, Wydawnictwo Politechniki Śląskiej, Gliwice (2006).
[3] H. Koczyk, Ogrzewnictwo Praktyczne. Projektowanie Montaż Eksploatacja, Wydawnictwo Systherm Serwis, Poznań 2005
[4] H. Recknagel, E.Sprenger, E.R. Schramek, Kompendium wiedzy. Ogrzewnictwo,
klimatyzacja,ciepła woda,chłodnictwo, wydanie trzecie polskie, tłumaczone w oparciu o
73. wydanie niemieckie, Omni Scala, (2008).
[5] P. O. Fanger, Komfort cieplny, Wydawnictwo Arkady, Warszawa (1974)
[6] Polska norma PN-EN ISO 6946.
[7] Polska norma PN-EN 12831.
[8] Algebra
stosunków
konfiguracji,
strona
http://www.thermalradiation.net/sectionc/C-2a.htm
http://www.thermalradiation.net/sectionc/C-5a.html, dostęp 25 maja 2016.
internetowa
oraz
[9] Dane techniczne promienników i materiały informacyjne, Strona internetowa firmy
Schwank,
http://www.schwank.pl/pl/produkty/gama-produktow/ciemne
promienniki/calorschwank/dane-techniczne.html
oraz
http://www.schwank.pl/fileadmin/00_customer/pl/pdf/Podstawy_promienniki_Schwank.pdf
, dostęp 25 maja 2016r.
[10] Dane techniczne płyt warstwowych, Strona internetowa firmy Balexmetal,
http://www.balex.eu/architekt/pliki/katalogi/katalog_produktow.pdf, dostęp 03.06.2016
[11] Dane techniczne płyt warstwowych, Strona internetowa
http://www.nowhal.pl/hale-z-plyty-warstwowej, dostęp 03.06.2016
firmy
Nowhal,
[12] Konstrukcja warstw podłogi przemysłowej, Strona internetowa Inzynier budownictwa,
http://www.inzynierbudownictwa.pl/technika,materialy_i_technologie,artykul,konstrukcja
_warstw_podlogi_przemyslowej,7309
110
Analysis and identification of radiation heat transfer
in the industrial hall
Adrian Orłowski
Key words: radiation, radiant heater, industrial hall, heating
Abstract
The aim of this paper is the analysis and identification of radiation heat transfer in the
industrial hall, creating algorithm and implementation this code in the MS VBA programming
language. The paper describes two dimensional model of the industrial hall with shading zones. It allows to place a rectangular object in the industrial hall, in such places in most cases it
is a standard europalette with different kinds of material. This configuration can be heated by
one or two infrared heater. Because of the consideration of the two dimensional configuration
this calculation program referred to gas fired radiant tube heater. This paper show analysis of
various calculation. Settings allow to change many calculation options and set technical specifications of modeling building for example thermal conductivity coefficient, surface film conductance, emissivity and more for each parts of hall. In this paper there are also included currently valid building standards concerning thermal insulation in Poland.
111
c
Instytut
Techniki Cieplnej, Politechnika Ślaska
˛
Archiwum Instytutu Techniki Cieplnej Vol. 1 nr 1(2016)
www.itc.polsl.pl/ArchiwumITC
Wykorzystanie egzergii kriogenicznej skroplonego gazu ziemnego
do produkcji energii elektrycznej
Tomasz Simla1
e-mail: [email protected]
Słowa kluczowe: skroplony gaz ziemny, regazyfikacja, egzergia kriogeniczna, odzysk
energii, terminal LNG w Świnoujściu
Streszczenie
Gaz ziemny jest paliwem kopalnym o największej dynamice wzrostu udziału w światowym
miksie energetycznym. Transport gazu w postaci skroplonej (LNG, ang. liquefied natural gas)
stanowi alternatywę dla tradycyjnego transportu rurociągowego. Polska dołącza do światowego rynku LNG dzięki wybudowanemu w Świnoujściu terminalowi regazyfikacyjnemu. Proces
skraplania gazu jest bardzo energochłonny. Część energii wykorzystanej w tym procesie zostaje zmagazynowana w LNG jako egzergia kriogeniczna. W konwencjonalnym procesie regazyfikacji egzergia ta jest tracona poprzez uwalnianie do wody morskiej lub innego czynnika
służącego jako zewnętrzne źródło ciepła. Istnieje wiele koncepcji wykorzystania egzergii
kriogenicznej LNG. Wśród możliwych zastosowań jest wykorzystanie LNG do produkcji
energii elektrycznej poprzez użycie go jako dolnego źródła ciepła w obiegach termodynamicznych lub bezpośrednio jako czynnika obiegowego. W ramach niniejszej pracy zamodelowano cztery układy technologiczne regazyfikacji LNG: dwa układy bez odzysku „zimnej”
egzergii oraz dwa układy z odzyskiem, produkujące energię elektryczną. Podstawowe dane
wejściowe do modelu (strumień masowy, ciśnienie gazu) odpowiadają rzeczywistym parametrom pracy terminalu w Świnoujściu. Wykonano symulację działania wszystkich układów dla
zmiennej w skali roku temperatury otoczenia. Obliczono szereg wskaźników służących do
porównania między sobą poszczególnych układów, takich jak średnioroczne zużycie paliwa,
sprawność egzergetyczna i wskaźnik skumulowanego zużycia energii.
Rozdział przygotowano podczas pracy nad projektem dyplomowym magisterskim wykonywanym przez autora
w Instytucie Techniki Cieplnej na Wydziale Inżynierii Środowiska i Energetyki Politechniki Śląskiej, pod opieką
dra hab. inż. Wojciecha Stanka, profesora nadzwyczajnego na Politechnice Śląskiej.
1
Zawartość tej publikacji może byc wykorzystana na warunkach licencji Uznanie Autorstwa 3.0 Polska. Licencja pozwala na
kopiowanie, zmienianie, rozprowadzanie, przedstawianie i wykonywanie utworu jedynie pod warunkiem oznaczenia autorstwa: autora,
tytułu rozdziału, nazwy serii, tomu, strony. Content from this work may be used under the terms of the Creative Commons Attribution
3.0 licence. Any further distribution of this work must maintain attribution to the author, chapter title, series title, volume, pages.
113
1
Wprowadzenie
Gaz ziemny (ang. natural gas, NG) jest paliwem kopalnym pochodzenia organicznego. Jego
skład jest zróżnicowany (zależny od złoża), jednak głównym składnikiem gazu ziemnego jest
metan. Udział molowy metanu wynosi najczęściej pomiędzy 70 a 98%. Obok metanu w składzie gazu ziemnego mogą występować wyższe węglowodory (etan, propan, butan) w ilościach nie przekraczających 20% oraz gazy inertne i zanieczyszczenia: dwutlenek węgla, azot,
tlen i siarkowodór. W procesie oczyszczania gazu usuwane są z niego składniki niepożądane
(przede wszystkim siarkowodór) oraz takie, które mogą stanowić osobny produkt handlowy
(propan i butan) [1]. Oczyszczony gaz ziemny składa się niemal wyłącznie z metanu, dlatego
jego spalanie, w przeciwieństwie do spalania węgla i w mniejszym stopniu produktów przeróbki ropy naftowej, nie wiąże się z emisją żadnych szkodliwych substancji pochodzenia paliwowego. Ponadto uzyskanie jednostki energii ze spalenia gazu ziemnego obarczone jest
emisją dwutlenku węgla o około połowę mniejszą niż przy uzyskaniu tej samej ilości energii
ze spalenia węgla. Co więcej, elektrownie gazowo-parowe opalane tym paliwem z łatwością
uzyskują sprawności konwersji energii przewyższające sprawności bloków kondensacyjnych
opalanych węglem. Są to tylko niektóre z powodów, dla których gaz ziemny staje się coraz
bardziej popularnym nośnikiem energii, co odzwierciedla się w zwiększającym się udziale
tego paliwa w światowym miksie energetycznym.
1.1
Gaz ziemny w kontekście światowej gospodarki energetycznej
Wydobycie gazu ziemnego na przestrzeni ostatnich 40 lat wykazuje stale rosnący trend. Na
rysunku 1.1 przedstawiono światowe zużycie energii pierwotnej (wyrażone w milionach ton
oleju ekwiwalentnego) w latach 1971÷2013 w rozbiciu na poszczególne źródła.
Rysunek 1.1: Zużycie energii pierwotnej na świecie [2]
114
W tabeli 1.1 porównano wielkości światowego wydobycia węgla, ropy naftowej i gazu ziemnego w latach 1973 i 2013. Wydobycie gazu ziemnego w analizowanym okresie wzrosło niemal trzykrotnie. W roku 1973 gaz ziemny zaspokajał 16% światowego zapotrzebowania na
energię pierwotną, w roku 2013 było to już 21,4% [2].
Tabela 1.1: Statystyki wydobycia paliw kopalnych w latach 1973 i 2013 (opracowanie własne
na podstawie [2]).
Jednostka
Wydobycie w 1973 r.
Udział w miksie energetycznym w 1973 r.
Wydobycie w 2013 r.
Udział w miksie energetycznym w 2013 r.
Przyrost wydobycia na przestrzeni 40 lat
Mtoe
%
Mtoe
%
%
Paliwo kopalne
Ropa
Gaz
Węgiel
naftowa ziemny
1474,35 2938,38
990,98
24,5
46,2
16
3958,1 4215,64 2908,64
28,9
31,1
21,4
168
43
194
Światowe rezerwy gazu ziemnego na koniec roku 2014 szacowane były na 187 bilionów metrów sześciennych. Globalny wskaźnik R/P (rezerwy do produkcji) przyjmuje wartość około
54 lat, jednak utrzymuje się na stałym poziomie od ponad 30 lat [3]. Zmiany wskaźnika R/P
dla gazu ziemnego na przestrzeni lat 2004÷2014 zobrazowano na rysunku 1.2.
Rysunek 1.2: Wartości wskaźnika R/P dla gazu ziemnego w poszczególnych regionach świata
(opracowanie własne na podstawie [3])
1.2
LNG jako metoda transportu gazu ziemnego
Tradycyjną, stosowaną od ponad stulecia metodą transportu gazu ziemnego jest transport rurociągowy. Gazociągi stanowią stabilne i bezpieczne źródło dostaw gazu ziemnego z dużych,
dogodnie zlokalizowanych złóż tego surowca. W miarę wyczerpywania się starych złóż i od-
115
krywania nowych, oddalonych znacznie od odbiorców i położonych niekorzystnie z punktu
widzenia możliwości budowy gazociągu, konieczne stało się opracowanie alternatywnych
metod transportu gazu ziemnego.
W warunkach normalnych 1 m3 metanu ma masę 0,66 kg i wartość opałową 32,8 MJ. Dla
porównania, 1 m3 ropy naftowej ma masę około 835 kg i wartość opałową około 41 GJ, czyli
ponad tysiąc razy większą niż gazowy metan. Ze względu na niską gęstość energii nieracjonalne byłoby transportowanie metanu w postaci gazowej w tankowcach lub cysternach. Jedyną alternatywną wobec rurociągów technologią transportu gazu ziemnego, która rozwinęła się
na komercyjną skalę jest skraplanie gazu ziemnego (LNG, ang. liquefied natural gas) [4].
Metan o temperaturze rzędu −162°C pod ciśnieniem normalnym jest bezbarwną, bezwonną,
nietoksyczną i niepowodującą korozji cieczą o gęstości około 650 razy większej niż w warunkach normalnych. Skroplenie metanu pozwala zatem na zwiększenie gęstości energii do około
21 GJ na metr sześcienny. Po dostarczeniu surowca do miejsca przeznaczenia jest on na powrót doprowadzany do postaci gazowej.
Zaletą transportu gazu ziemnego w formie LNG jest jego elastyczność. Dostawca gazu ziemnego nie jest sztywno związany z odbiorcą poprzez rurociąg. Skroplony gaz można drogą
morską dostarczyć do dowolnego terminalu regazyfikacyjnego na świecie. W ten sposób LNG
przyczynia się do zwiększenia konkurencyjności na rynku gazu ziemnego. Odbiorca gazu
dysponujący terminalem regazyfikacyjnym może dowolnie wybrać, od kogo zakupi to paliwo.
Na rysunku 1.3 przedstawiono porównanie jednostkowych kosztów transportu gazu ziemnego
rurociągami i w postaci skroplonej. Jak wynika z wykresu, transport w postaci LNG charakteryzuje się dużymi początkowymi nakładami inwestycyjnymi, lecz dystans między fabryką
LNG a odbiorcą nie wpływa znacząco na koszt transportu. Jednostkowy koszt transportu gazu
drogą morską w postaci LNG jest niższy niż dla transportu rurociągowego na dystansie powyżej 1000 km dla rurociągów podwodnych i powyżej 3000 km dla rurociągów lądowych.
Rysunek 1.3: Porównanie kosztów transportu gazu ziemnego [5]
W styczniu 2016 r. łączna wydajność istniejących na świecie terminali skraplających wynosiła 301 milionów ton rocznie. Łączna wydajność terminali regazyfikacyjnych była dwa i pół
razy większa i wynosiła 757 milionów ton rocznie. W roku 2015 przetransportowano w formie LNG 245 milionów ton gazu ziemnego, co stanowiło około 10% światowego zużycia
116
tego surowca [6]. Na rysunku 1.4 przedstawiono wielkości eksportu LNG przez wszystkie 17
państw, które w roku 2015 uczestniczyły w światowym rynku LNG po stronie dostawców.
Największym eksporterem LNG jest od kilku lat Katar, który zaspokaja niemal jedną trzecią
światowego zapotrzebowania na ten surowiec.
Rysunek 1.4: Eksporterzy LNG w roku 2015 (opracowanie własne na podstawie [6])
W roku 2015 33 państwa brały udział w światowym rynku LNG po stronie odbiorców. Na
rysunku 1.5 przedstawiono wielkości importu LNG przez poszczególne państwa. Największym importerem jest Japonia, która zapewnia aż jedną trzecią światowego popytu na ten surowiec. W Europie największymi importerami LNG są Wielka Brytania i Hiszpania [6].
Rysunek 1.5: Importerzy LNG w roku 2015 (opracowanie własne na podstawie [6])
117
Rynek LNG będzie nadal się rozwijał wraz z budową nowych terminali skraplających i regazyfikacyjnych. Do roku 2021 łączna zdolność wysyłkowa fabryk LNG ma wzrosnąć do 400
milionów ton rocznie [6].
Gaz ziemny i LNG w kontekście polskiej gospodarki energetycznej
1.3
Podobnie jak w skali świata, również w przypadku Polski prognozowany jest wzrost udziału
gazu ziemnego w miksie energetycznym. Będzie to powiązane ze stopniowym zmniejszaniem
się roli węgla w krajowej gospodarce energetycznej. W roku 2013 w Polsce zużyto
14,08 Mtoe energii pierwotnej gazu ziemnego, co stanowiło 12% krajowego zapotrzebowania
na energię pierwotną wynoszącego 116,86 Mtoe [7]. Projekt polityki energetycznej Polski do
2050 roku zakłada, że zapotrzebowanie na energię pierwotną gazu ziemnego wyniesie
15,3 Mtoe w roku 2030 i 15,5 Mtoe w roku 2050 [8]. Wartości te będą musiały być wyższe
w wypadku opóźnień w realizacji polskiego programu atomowego.
Polska dysponuje własnymi złożami gazu ziemnego, są one jednak niewystarczające do pokrycia krajowego zapotrzebowania na to paliwo. Głównymi regionami występowania konwencjonalnych złóż gazu ziemnego w Polsce są Niż Polski (69% wydobywalnych zasobów),
przedgórze Karpat (26%) i strefa ekonomiczna na Morzu Bałtyckim (4%). Złoża gazu w tych
trzech regionach kraju stanowią aż 99% wszystkich wydobywalnych zasobów. Na Niżu Polskim dominują złoża gazu zaazotowanego (zawierającego od 30 do ponad 80% metanu),
z kolei na przedgórzu Karpat występuje głównie gaz ziemny wysokometanowy. W 2014 roku
stan wydobywalnych konwencjonalnych zasobów gazu ziemnego wynosił 129,75 mld m3
(wartość dotycząca gazu zwykłego, nieprzeliczona na gaz wysokometanowy). Na terytorium
Polski występują również niekonwencjonalne złoża gazu – gaz łupkowy. Łączne zasoby wydobywalne z tego źródła szacowane są maksymalnie na 1920 mld m3, przy czym z największym prawdopodobieństwem zasoby te mieszczą się w przedziale 346÷768 mld m3 [9]. Na
chwilę obecną eksploatacja tych zasobów jest jednak uznawana za nieopłacalną.
Rysunek 1.6: Struktura pochodzenia gazu ziemnego zużywanego w Polsce w roku 2014
(opracowanie własne na podstawie [7, 10])
118
Wielkość krajowego wydobycia gazu ziemnego w roku 2014 wyniosła 3,17 Mtoe co stanowiło blisko 28% zużycia tego surowca [7]. Pozostałe 72% gazu pochodziło z importu. Dotychczas całość importu gazu ziemnego do Polski odbywała się za pośrednictwem gazociągów
międzysystemowych. Strukturę pochodzenia gazu ziemnego w Polsce przedstawiono na rysunku 1.6. Większość importowanego gazu pochodzi z Rosji i jest kupowana w ramach obowiązującego do 2022 roku kontraktu długoterminowego z firmą Gazprom [10].
Dzięki wybudowanemu w Świnoujściu terminalowi regazyfikacyjnemu (zwanemu też gazoportem) Polska dołącza do światowego rynku LNG. Decyzja o budowie terminalu zapadła
w roku 2006, a w lipcu 2010 roku podpisano umowę na jego budowę. Pierwsza, testowa dostawa skroplonego gazu ziemnego miała miejsce 11 grudnia 2015 roku, natomiast 27 kwietnia
2016 roku wydano pozwolenie na użytkowanie obiektu [11]. Pierwsza komercyjna dostawa
LNG miała miejsce 17 czerwca 2016 roku [12]. Znaczna część dostaw realizowana będzie na
podstawie długoterminowego kontraktu między PGNiG a katarską spółką Qatargas. Umowa
dotyczy dostawy 1 mln ton LNG rocznie przez okres 20 lat. Za eksploatację terminalu odpowiedzialna jest należąca do skarbu państwa spółka Polskie LNG. W związku z powyższym,
w bieżącym roku dojdzie do dywersyfikacji źródeł dostaw gazu ziemnego do Polski. Na rysunku 1.7 przedstawiono prognozę struktury pochodzenia zużywanego w Polsce gazu ziemnego w roku 2020.
Rysunek 1.7: Prognozowana struktura pochodzenia gazu ziemnego zużywanego w Polsce
w roku 2020 (opracowanie własne na podstawie [13])
Świnoujski terminal posiada zdolność regazyfikacyjną 5 mld mn3 gazu ziemnego rocznie.
Największym elementem terminalu są dwa zbiorniki na LNG o pojemności 160 000 m3 każdy. Na terenie zakładu jest pozostawiona rezerwa na trzeci zbiornik, którego budowa pozwoli
na zwiększenie przepustowości do 7,5 mld mn3 gazu ziemnego rocznie [14]. Gazoport ma
możliwość przyjmowania statków (metanowców) o pojemności od 120 000 do 216 000 m³.
LNG będzie odparowywany przy użyciu metody Submerged Combustion Vaporizer. Technologia Open Rack Vaporizer była rozważana jako metoda uzupełniająca, jednak zrezygnowano
z niej z uwagi na możliwość wykorzystania jej tylko przez kilka miesięcy w roku i koniecz-
119
ność uzyskania dodatkowego pozwolenia środowiskowego [15]. Opis poszczególnych metod
regazyfikacji zawarto w podrozdziale 2.3.
W ramach projektu, oprócz samego terminalu regazyfikacyjnego, wybudowano także: nowy
falochron zewnętrzny przy ujściu rzeki Świny, infrastrukturę cumowania i nabrzeża dla metanowców wewnątrz nowego basenu portowego oraz wysokociśnieniowy (8,4 MPa) gazociąg
przyłączeniowy o długości 6 km i gazociąg przesyłowy o długości 74 km łączący terminal
z krajową siecią przesyłową gazu [14].
2
Łańcuch produkcji LNG, metody regazyfikacji
Łańcuch produkcji i dostaw LNG przedstawiono schematycznie na rysunku 2.1. Składa się on
z czterech głównych etapów: wydobycia gazu ziemnego, oczyszczania i skraplania, transportu
morskiego oraz regazyfikacji.
Rysunek 2.1: Schemat łańcucha produkcji i dostaw LNG [16]
Poszczególne etapy opisano w kolejnych podrozdziałach.
2.1
Przygotowanie i skraplanie gazu
Przed przystąpieniem do skraplania gazu ziemnego musi on zostać oczyszczony ze składników niepożądanych. Surowiec dostarczany z pola wydobywczego jest najpierw rozdzielany
na frakcję gazową i ciekłą. Następnie gaz przechodzi kilka etapów oczyszczania. Pierwszym
etapem jest usuwanie dwutlenku węgla i związków siarki metodami aminowymi. Zawartość
CO2 jest zmniejszana do 50 ppm, zawartość H2S do 4 ppm. W kolejnym etapie sita molekularne absorbują wilgoć oraz pozostałe związki siarki. Zawartość wilgoci w osuszonym gazie
jest niższa od 0,1 ppm, a całkowita zawartość siarki (siarkowodoru, merkaptanów i tlenosiarczku węgla) nie powinna przekraczać 30 ppm. Jeśli gaz ziemny zawiera rtęć, musi ona być
również usunięta. Stężenie rtęci powinno być niższe od 10 ng/m3. Wymagania dotyczące
maksymalnych stężeń wymienionych powyżej substancji wynikają zarówno ze specyfikacji
handlowych gazu ziemnego jak i ze szkodliwego wpływu tych substancji na elementy konstrukcyjne rurociągów i zbiorników na gaz ziemny. Ostatnim etapem przed skraplaniem gazu
jest separacja węglowodorów cięższych od metanu. Związki te powodowałyby problemy eksploatacyjne w procesie skraplania. Etan może, lecz nie musi być separowany. Ponadto propan,
butan i wyższe węglowodory mogą zostać sprzedane osobno z większym zyskiem, niż gdyby
pozostały w składzie LNG. W trakcie skraplania gazu ziemnego jest z niego dodatkowo usuwany azot [4].
120
Najbardziej energochłonnym procesem w łańcuchu produkcji LNG jest skraplanie gazu.
Technologia skraplania bazuje na zasadzie działania cykli chłodniczych. Gaz ziemny musi
zostać ochłodzony do temperatury nasycenia, która przy ciśnieniu normalnym wynosi około
−160°C. Istniejące technologie skraplania opierają się na trzech podstawowych metodach [4, 17]:



Klasyczny cykl kaskadowy. Gaz ziemny jest chłodzony w kilku kolejnych cyklach
chłodniczych, w których czynnikami chłodniczymi są propan, etan i metan. Parametry
cykli chłodniczych są dobrane tak, aby uzyskać jak najlepsze zbliżenie profili temperaturowych skraplanego gazu i czynnika chłodniczego. Do zalet tej metody należą relatywna energooszczędność i prostota technologiczna. Wadami są duża liczba maszyn
i urządzeń skutkująca wysokim nakładem inwestycyjnym oraz zapotrzebowanie na
czysty etan i propan.
Cykl z mieszanym czynnikiem chłodniczym. Czynnikiem chłodniczym jest mieszanina węglowodorów dobrana tak, aby jej profil temperaturowy był jak najbardziej zbliżony do profilu temperaturowego skraplanego gazu ziemnego. Metoda ta jest nieco
bardziej energochłonna niż klasyczny cykl kaskadowy, ale jej zaletą jest mniejsza
liczba sprężarek i wymienników ciepła. Charakteryzuje się dłuższym czasem rozruchu
niż metoda kaskadowa.
Cykl rozprężenia z zastosowaniem turboekspandera. Wykorzystuje on efekt Joule'aThomsona. W procesie tym część gazu ulega rozprężeniu, co skutkuje ochłodzeniem
do bardzo niskiej temperatury. Schłodzony gaz wykorzystuje się następnie do skroplenia kolejnej porcji gazu przepływającej przez instalację. Metoda ta jest stosunkowo
prosta i nie wymaga dużych nakładów inwestycyjnych, lecz jest bardziej energochłonna niż metody kaskadowe i z mieszanym czynnikiem chłodniczym. Z tego powodu
jest stosowana w małych instalacjach skraplających i do pokrywania szczytowych zapotrzebowań na przepustowość instalacji skraplającej.
Najpopularniejszą metodą skraplania gazu jest proces AP-C3MR (Air Products propane precooled mixed refrigerant), bazujący na metodzie z mieszanym czynnikiem chłodniczym. Gaz
ziemny jest w tej metodzie wstępnie chłodzony propanem. W roku 2015 proces ten i jego modyfikacje odpowiadały za 79% światowej zdolności produkcyjnej LNG. 14% pokrywała metoda ConocoPhillips Optimised Cascade, czyli proces z kaskadowymi cyklami chłodniczymi [6].
Skroplenie 1 kg gazu ziemnego wymaga zużycia średnio 1,2 MJ energii. To zapotrzebowanie
na energię pokrywane jest poprzez spalanie około 8% surowca dostarczanego do fabryki
LNG [4]. Część energii wykorzystanej w procesie skraplania gazu ziemnego zostaje zmagazynowana w LNG jako egzergia kriogeniczna. Egzergia ta może zostać częściowo odzyskana
w procesie regazyfikacji. Definicję egzergii zawarto w podrozdziale 3.5.
2.2
Transport
Małe dostawy LNG do lokalnych odbiorców mogą być realizowane przy użyciu cystern samochodowych, jednak główną metodą transportu LNG jest transport morski w specjalnie do
tego zaprojektowanych statkach zwanych metanowcami lub gazowcami. LNG o temperaturze
około −169°C jest przechowywany w zaizolowanych zbiornikach przy nadciśnieniu około
121
0,3 bar. Ponieważ nie jest możliwe idealne zaizolowanie zbiorników, LNG pobiera ciepło
z otoczenia i odparowuje w ilości około 0,1÷0,15% dziennie. Odparowany gaz (ang. boil-off
gas, BOG), w zależności od rozwiązań technicznych zaimplementowanych na statku, może
zostać ponownie skroplony lub wykorzystany jako paliwo do napędu statku (w silnikach
dwupaliwowych lub kotłach parowych) [4].
Pojemność statków mieści się w zakresie 30÷265 tysięcy m3, przy czym większość statków
ma pojemność w przedziale 125÷140 tysięcy m3. Statki nowszego typu Q-Flex (takie, jakie
będą dostarczały katarski gaz do Polski) mają pojemność 216 tysięcy m3. Największe metanowce (265 tysięcy m3) noszą oznaczenie Q-Max. W styczniu 2016 roku światowa flota metanowców liczyła 410 statków o łącznej pojemności 60 milionów m3 [4, 6].
2.3
Regazyfikacja
Zadaniem terminalu regazyfikacyjnego jest doprowadzenie LNG do stanu gazowego i wysłanie go pod wymaganym ciśnieniem do sieci gazociągów odbiorczych. LNG wyładowany ze
statków jest najpierw przesyłany do zaizolowanych zbiorników, gdzie jest przechowywany
pod niewielkim nadciśnieniem w temperaturze o kilka stopni niższej od temperatury parowania. Standardową pojemnością zbiorników jest 160 tysięcy m3. LNG odparowuje w nich
w ilości około 0,05% dziennie. Z tego powodu terminal regazyfikacyjny powinien być wyposażony w instalację do skraplania BOG. LNG pobierany ze zbiornika jest najpierw sprężany
do ciśnienia wymaganego w gazociągu odbiorczym. Jest to podyktowane faktem, że praca
sprężania cieczy jest o wiele niższa niż praca sprężania gazu przy takim samym stosunku ciśnień. Sprężony LNG trafia do wymiennika ciepła, w którym jest odparowywany i ogrzewany
do temperatury zbliżonej do temperatury otoczenia. Ze względu na źródło ciepła pobieranego
przez LNG metody regazyfikacji można podzielić na następujące grupy [4]:

122
ORV (Open Rack Vaporizers), STV (Shell and Tube Vaporizers) – ciepło potrzebne
do odparowania LNG pobierane jest z wody morskiej. Metody ORV i STV różnią się
miedzy sobą rozwiązaniem konstrukcyjnym wymiennika ciepła. Są to najpopularniejsze metody regazyfikacji, stosowane w ponad 70% terminali regazyfikacyjnych. Ponieważ terminale są najczęściej usytuowane na wybrzeżach, woda morska jest tanim
źródłem ciepła dostępnym w dużych ilościach. Układy regazyfikacji tego typu są proste konstrukcyjnie, bezpieczne, łatwe w konserwacji i charakteryzują się niskimi kosztami operacyjnymi. Wymagają jednak częstej konserwacji ze względu na zanieczyszczanie powierzchni wymiany ciepła przez osady obecne w wodzie morskiej. Możliwość zastosowania tej metody regazyfikacji jest zdeterminowana warunkami klimatycznymi i środowiskowymi w otoczeniu terminalu. Należy zbadać skład wody morskiej. Zbyt duża zawartość metali ciężkich wpływa na przyspieszoną korozję wymienników ciepła, natomiast zbyt duża zawartość osadów stałych pociąga za sobą konieczność dokładnego filtrowania wody. Jeżeli temperatura wody morskiej jest zbyt niska,
nie może ona być wykorzystana jako źródło ciepła. Przyjmuje się, że temperatura wody nie powinna być niższa niż 5°C, a spadek temperatury zużywanej wody nie powinien przekraczać 5 K. W chłodnej strefie klimatycznej stosowanie metody ORV może
nie być możliwe przez cały rok i wówczas konieczna jest obecność zapasowego systemu regazyfikacji, mogącego działać niezależnie od temperatury otoczenia. Ponadto
mogą wystąpić przeciwwskazania środowiskowe związane z wrażliwością morskiej
flory i fauny na obniżoną temperaturę wody.
2.4

AAV (Ambient Air Vaporizers) – ciepło potrzebne do odparowania LNG pobierane
jest z powietrza atmosferycznego. Metoda AAV jest bardziej przyjazna środowisku
niż metoda ORV i jej stosowanie nie jest ograniczone przez temperaturę otoczenia.
Jednak ze względu na małą w porównaniu do metody ORV wartość współczynnika
przenikania ciepła wymaga o wiele większej powierzchni wymienników, co skutkuje
dużą zajętością terenu. Z tego powodu metoda AAV nie jest optymalna dla terminali
o dużej projektowanej zdolności regazyfikacyjnej. Głównym problemem eksploatacyjnym tej metody jest obladzanie powierzchni wymienników, w związku z czym co
4 do 8 godzin należy usuwać nagromadzony lód. Dodatkowym problemem, zwłaszcza
w wilgotnym klimacie, może być powstawanie mgły ograniczającej widoczność w sąsiedztwie terminalu.

SCV (Submerged Combustion Vaporizers) – ciepło potrzebne do odparowania LNG
pochodzi ze spalania części odparowanego gazu. Jest to druga pod względem popularności metoda regazyfikacji, stosowana w około 20% terminali. Ze względu na wysokie koszty operacyjne (zużycie paliwa wynosi około 1,5% regazyfikowanego LNG),
proces ten jest używany tylko tam, gdzie nie są dostępne inne, darmowe źródła ciepła.
Wymiennik regazyfikacyjny jest umieszczony w zbiorniku z wodą, na dnie którego
znajdują się palniki. Spaliny w formie bąbelków wędrują w górę, omywając powierzchnię wymiany ciepła i zapewniając dobre warunki przekazywania ciepła. Spaliny są ochładzane do niskich temperatur i zawarta w nich wilgoć ulega wykropleniu,
zwiększając efektywność energetyczną procesu, która sięga powyżej 98%. Dzięki dużej pojemności cieplnej wody system ten dobrze radzi sobie ze zmianami strumienia
LNG i jest w stanie dostarczać ciepło przez pewien czas od momentu wyłączenia palników. Problemem eksploatacyjnym jest kwasowy odczyn kąpieli wodnej powodowany rozpuszczanymi w niej składnikami spalin. Nadmiar kwasów należy neutralizować,
aby zapobiec korozji. Innym problemem może być powstawanie tlenków azotu podczas spalania gazu ziemnego.

IFV (Intermediate Fluid Vaporizers) – czynnik pośredniczący w wymianie ciepła krąży w obiegu zamkniętym, a ciepło potrzebne do regazyfikacji może pochodzić
z otoczenia lub z innego procesu przemysłowego (np. ciepło odpadowe z elektrowni
cieplnej). Najczęściej stosowanymi czynnikami obiegowymi są wodne roztwory glikolu i ciekłe węglowodory. Czynnik pośredniczący może dodatkowo pracować w obiegu
prawobieżnym (np. obiegu Rankine’a) i napędzać turbinę, produkując w ten sposób
energię elektryczną.
Możliwości odzysku egzergii
Aby oszacować potencjał energetyczny LNG dostarczanego do terminalu w Świnoujściu
przeprowadzono następujące obliczenia. Przy temperaturze otoczenia równej 𝑡0 = 9°C (średnioroczna temperatura w Świnoujściu) egzergia fizyczna LNG przechowywanego w zbiornikach (𝑡𝐿𝑁𝐺 = −162°C, 𝑝𝐿𝑁𝐺 = 0,15 MPa) wynosi 978 kJ/kg. Egzergia ciśnieniowa gazu wysyłanego do sieci (𝑝𝑁𝐺 = 8,4 MPa) wynosi 621 kJ/kg (definicje egzergii i sposób ich obliczania
123
opisano w podrozdziale 3.5). Różnica tych egzergii przemnożona przez zdolność wysyłkową
terminalu (𝑚̇𝐿𝑁𝐺 = 113,5 kg/s) daje wynik około 40 MW. W konwencjonalnych procesach
regazyfikacji cała egzergia kriogeniczna zmagazynowana w LNG jest uwalniana do wody
morskiej lub innego czynnika służącego jako zewnętrzne źródło ciepła i tracona. Może ona
jednak zostać częściowo odzyskana. Opłacalność zastosowania odzysku „zimnej” egzergii
musi być w każdym przypadku przedmiotem analiz termodynamicznych i ekonomicznych, ale
zastosowanie tych technologii zawsze prowadzi do obniżenia energochłonności skumulowanej importu LNG i jednostkowej emisji CO2 obciążającej importowany gaz.
Istnieje wiele koncepcji wykorzystania egzergii kriogenicznej LNG. Można je podzielić na
dwie grupy. Metody pierwszej grupy opierają się na bezpośrednim wykorzystaniu niskiej
temperatury LNG. Wśród możliwych zastosowań są: skraplanie i separacja powietrza, przemysł spożywczy (chłodzenie żywności lub jej zamrażanie), klimatyzacja, kriogeniczne odsalanie wody morskiej lub wykorzystanie w różnorodnych procesach przemysłowych, np.
w petrochemii [18]. Niska temperatura LNG może też służyć do chłodzenia skraplaczy w siłowniach opartych na obiegu Rankine’a lub do chłodzenia powietrza sprężanego w obiegach
turbin gazowych. Druga grupa metod polega na wykorzystaniu egzergii kriogenicznej LNG
do produkcji energii elektrycznej. Skroplony gaz ziemny może posłużyć jako dolne źródło
ciepła w obiegach termodynamicznych lub bezpośrednio jako czynnik obiegowy. Najczęściej
rozpatrywane są metody oparte o [19]:






Bezpośrednie rozprężanie (ang. direct expansion cycle, DEC),
Obiegi Rankine’a (RC),
Obiegi Braytona (BC),
Obiegi Kaliny,
Silniki Stirlinga,
Układy skojarzone (połączenie powyższych rozwiązań).
W dalszej części niniejszej pracy skupiono się na trzech pierwszych spośród powyższych
rozwiązań.
Najprostszą metodą wytwarzania energii elektrycznej poprzez odzysk egzergii kriogenicznej
LNG jest bezpośrednie rozprężanie, czyli otwarty obieg Rankine’a. Polega ona na podniesieniu ciśnienia skroplonego gazu do poziomu wyższego niż wymagany w sieci przesyłowej
gazu, odparowaniu gazu i rozprężeniu go w turbinie. Ideę tego rozwiązania przedstawiono na
rysunku 2.2. Układy tego typu są proste konstrukcyjnie, ale cechują się niską sprawnością
wykorzystania egzergii kriogenicznej LNG. Jest to spowodowane faktem, że niemal cała egzergia temperaturowa LNG jest tracona, a turbina rozprężna wykorzystuje tylko mechaniczną
część egzergii. Metoda ta jest najczęściej stosowana w połączeniu z układami opartymi na
obiegach Rankine’a i Braytona.
124
Rysunek 2.2: Schemat układu odzysku egzergii bazującego na bezpośrednim rozprężaniu
W układach z zamkniętymi obiegami Rankine’a czynnikami obiegowymi mogą być te same
płyny, które są stosowane w chłodnictwie (czyli np. węglowodory). Schemat najprostszego
układu tego typu przedstawiono na rysunku 2.3. LNG jest dolnym źródłem ciepła (chłodzi
skraplacz), a górnym źródłem ciepła jest najczęściej otoczenie.
Rysunek 2.3: Schemat układu odzysku egzergii bazującego na obiegu Rankine’a
Układy te charakteryzują się wyższą sprawnością niż proste układy DEC, ponieważ wykorzystywana jest w nich temperaturowa część egzergii LNG. Struktura układu może być optymalizowana, przykładowo poprzez wprowadzenie regeneracji ciepła między wylotem z turbiny
a wlotem do górnego wymiennika ciepła. Można też zastosować kilka osobnych obiegów
Rankine’a połączonych kaskadowo, szczególnie wówczas, gdy do dyspozycji jest ciepło na
wysokim poziomie temperaturowym. Możliwość zastosowania układów RC ograniczona jest
przez ciśnienie gazu ziemnego dostarczanego do sieci przesyłowej. Nadkrytyczne ciśnienie
odparowywanego gazu dyskwalifikuje układy oparte na obiegach Rankine’a ze względu na
brak dopasowania profili temperaturowych procesów odparowywania LNG i skraplania czynnika roboczego w obiegu Rankine’a. Ciśnienie krytyczne metanu wynosi około 4,6 MPa [20].
Przy wysokich ciśnieniach wyjściowych gazu ziemnego lepsze dopasowanie profili temperaturowych zapewnia obieg Braytona. Układy z obiegami Braytona mogą występować w róż-
125
nych konfiguracjach. Podstawową konfigurację układu przedstawiono na rysunku 2.4. LNG
służy w nim do chłodzenia gazu obiegowego przed wlotem do sprężarki. Praca sprężania gazu
jest tym mniejsza, im niższa jest jego początkowa temperatura. Układy oparte na obiegach
Braytona nie osiągają wysokiej sprawności, jeśli górnym źródłem ciepła jest otoczenie. Dlatego najczęściej, jeżeli nie jest dostępne wysokotemperaturowe źródło ciepła, wysoką temperaturę uzyskuje się poprzez spalanie części odparowywanego gazu ziemnego. W literaturze proponowane są różne układy, między innymi z otwartym obiegiem turbiny gazowej, w którym
powietrze wlotowe i międzystopniowe jest chłodzone przez LNG oraz połączenia kaskadowe
otwartego obiegu turbiny gazowej z obiegiem zamkniętym [19].
Rysunek 2.4: Schemat układu odzysku egzergii bazującego na obiegu Braytona
Pierwsze instalacje do odzysku egzergii kriogenicznej LNG powstały w latach 70. Pionierem
w tej dziedzinie jest Japonia, tam też znajduje się najwięcej instalacji tego typu. Jest to skutek
bardzo dużego importu LNG do Japonii. Większość tamtejszych układów produkcji energii
elektrycznej zintegrowanych z terminalami regazyfikacyjnymi bazuje na metodach DEC
i RC [19].
3
Opis analizowanych układów i metodyka obliczeń
W ramach niniejszej pracy zamodelowano w programie Engineering Equation Solver (EES)
cztery układy technologiczne regazyfikacji LNG: dwa układy bez odzysku egzergii kriogenicznej oraz dwa układy z odzyskiem, produkujące energię elektryczną. Układy z odzyskiem
bazują na metodach DEC i BC. Wartość ciśnienia roboczego w gazociągu przesyłowym nie
sprzyja stosowaniu układów opartych na metodzie RC. Wspólne założenia dla wszystkich
układów są następujące [16, 21]:
𝑇𝐿𝑁𝐺 = 111,15 K
𝑝𝐿𝑁𝐺 = 151 kPa
𝑚𝐿𝑁𝐺 = 3579,3 Gg
𝑚̇𝐿𝑁𝐺 = 113,5 kg/s
126
-
temperatura LNG w zbiorniku,
ciśnienie LNG w zbiorniku,
roczna ilość LNG dostarczanego do terminalu,
strumień masowy LNG,
𝑝𝑁𝐺 = 8400 kPa
𝑇𝑁𝐺 = 274,15 K
𝜂𝑖 𝑃𝑁𝐺 = 0,665
𝜂𝑒𝑚 𝑃 = 0,9
𝜂𝑒𝑚 𝐶 = 0,9
-
ciśnienie gazu wysyłanego do rurociągu,
temperatura gazu wysyłanego do rurociągu,
sprawność wewnętrzna pomp LNG,
sprawność elektromechaniczna pomp,
sprawność elektromechaniczna wentylatorów powietrza.
Założenie dotyczące strumienia masowego LNG wynika ze zdolności regazyfikacyjnej terminalu w Świnoujściu (5 mld mn3 na rok). Ciśnienie wysyłanego gazu ziemnego jest równe ciśnieniu roboczemu gazociągu Świnoujście – Szczecin.
Poszczególne układy opisano w kolejnych podrozdziałach.
3.1
Układ referencyjny (SCV)
Referencyjny układ bazuje na metodzie SCV zastosowanej w terminalu w Świnoujściu.
Schemat układu zaprezentowano na rysunku 3.1. Skroplony gaz ziemny (1) jest sprężany
w pompie do ciśnienia panującego w rurociągu przesyłowym. LNG o wysokim ciśnieniu (2)
trafia do wymiennika regazyfikacyjnego, gdzie jest odparowywany i ogrzewany. Większość
gazu (3) jest wysyłana do sieci, pozostała część (4) jest zużywana jako paliwo do procesu
regazyfikacji. Powietrze atmosferyczne (6) jest sprężane do ciśnienia niezbędnego w palnikach metody SCV. Sprężone powietrze (7) jest mieszane ze zdławionym gazem ziemnym (5)
i spalane. Spaliny (8) są odprowadzane do otoczenia. Na moc napędową układu (11) składają
się moc pompy LNG (9) oraz moc wentylatora powietrza (10).
Rysunek 3.1 Schemat układu referencyjnego (SCV)
Dodatkowe założenia do modelu układu referencyjnego:
- różnica między temperaturą spalin a temperaturą podgrzanego gazu,
∆𝑇𝐹𝐺 = 15 K
- stosunek nadmiaru powietrza do spalania,
𝜆 = 1,05
∆𝑝𝐴 = 30 kPa - przyrost ciśnienia powietrza w wentylatorze,
- sprawność wewnętrzna wentylatora.
𝜂𝑖 𝐶 = 0,9
127
3.2
Układ hybrydowy (SCV+ORV)
Hybrydowy układ regazyfikacyjny stanowi połączenie metody SCV i metody regazyfikacji
opartej na pobieraniu ciepła z wody morskiej. Schemat układu zaprezentowano na rysunku 3.2. Od układu referencyjnego różni się jedynie obecnością dodatkowej linii dostarczania
i odprowadzania wody morskiej. Woda (9) jest pobierana z morza i tłoczona do wymiennika
ciepła. Ciśnienie za pompą (10) wynika z różnicy poziomów między punktem poboru wody
a wymiennikiem regazyfikacyjnym oraz z oporów przepływu. Woda po opuszczeniu wymiennika (11) odprowadzana jest z powrotem do morza. Na moc napędową układu (15) składają się moc pompy LNG (12), moc wentylatora powietrza (13) oraz moc pompy wody morskiej (14).
Rysunek 3.2 Schemat układu hybrydowego (SCV+ORV)
Dodatkowe założenia do modelu układu hybrydowego są następujące:
- różnica między temperaturą spalin a temperaturą podgrzanego gazu,
∆𝑇𝐹𝐺 = 15 K
- stosunek nadmiaru powietrza do spalania,
𝜆 = 1,05
- przyrost ciśnienia powietrza w wentylatorze,
∆𝑝𝐴 = 30 kPa
- sprawność wewnętrzna wentylatora,
𝜂𝑖𝐶 = 0,9
- przyrost ciśnienia wody morskiej w pompie,
∆𝑝𝑊 = 200 kPa
- maksymalny spadek temperatury wody morskiej,
∆𝑇𝑊 = 5 K
𝑇𝑔𝑟 = 275,15 K
- minimalna temperatura wody morskiej odprowadzanej do otoczenia,
- sprawność wewnętrzna pompy wody.
𝜂𝑖 𝑃𝑊 = 0,8
Zdefiniowano trzy stany pracy układu w zależności od temperatury wody morskiej. Jeśli temperatura wody morskiej jest wyższa od 𝑇𝑔𝑟 + ∆𝑇𝑊 , całe ciepło potrzebne do regazyfikacji
pochodzi z wody morskiej i układ spalania jest wyłączony. Gdy temperatura morza spadnie
poniżej 𝑇𝑔𝑟 + ∆𝑇𝑊 , strumień pobieranej wody morskiej pozostaje bez zmian, natomiast włączają się palniki i spalana jest taka ilość paliwa, aby utrzymać temperaturę wody odprowadzanej do morza na poziomie 𝑇𝑔𝑟 . Gdy temperatura morza spadnie poniżej 𝑇𝑔𝑟 , wyłączane są
pompy tłoczące wodę morską i całe ciepło potrzebne do regazyfikacji pochodzi ze spalania
gazu ziemnego.
128
3.3
Układ z kaskadowym rozprężaniem gazu ziemnego (DEC)
Pierwszy z zamodelowanych układów z odzyskiem egzergii kriogenicznej bazuje na rozwiązaniu zaproponowanym w [20]. Jest to metoda oparta o bezpośrednie, kaskadowe rozprężanie
gazu ziemnego. Schemat układu zaprezentowano na rysunku 3.3.
Rysunek 3.3 Schemat układu z kaskadowym rozprężaniem (DEC)
LNG (1) po pierwszym stopniu sprężania (2) miesza się z parą gazu ziemnego (12) rozprężoną w turbinie niskoprężnej. LNG w stanie nasycenia na wylocie z mieszalnika (3) przechodzi
przez drugi stopień sprężania (4), po czym miesza się z parą gazu ziemnego (9) pochodzącą
z upustu turbiny wysokoprężnej. LNG w stanie nasycenia na wylocie z mieszalnika (5) jest
sprężany do ciśnienia wyższego niż wymagane w rurociągu przesyłowym (6). Następnie
w wymienniku regazyfikacyjnym jest odparowywany i ogrzewany (7). Para gazu ziemnego
jest rozprężana w turbinie do ciśnienia roboczego rurociągu przesyłowego (8), a następnie
ogrzewana ponownie do temperatury wymaganej w rurociągu (13). Cyrkulująca w układzie
część gazu ziemnego jest dalej rozprężana w turbinie. Strumień (10) jest przegrzewany (11)
i rozprężany w turbinie niskoprężnej. Układ dostarczania ciepła na potrzeby regazyfikacji
i przegrzewu gazu ziemnego działa na takich samych zasadach, jak w modelu SCV+ORV
(trzy stany pracy, zużywane ciepło pochodzi od wody morskiej lub ze spalania części gazu
ziemnego). Temperatura w punktach 7 i 11 jest równa 𝑇𝑁𝐺 jeśli źródłem ciepła są spaliny lub
równa temperaturze wody morskiej pomniejszonej o ∆𝑇𝑁𝐺 jeżeli gaz jest ogrzewany wyłącznie wodą morską. Moc turbin (25, 26) napędza generator prądu elektrycznego. Na potrzeby
własne układu składają się moce pomp LNG (22, 23, 24), moc pompy wody morskiej (27)
oraz moc wentylatora powietrza do spalania (28). Reszta wytworzonej mocy elektrycznej (29)
jest odprowadzana do sieci. Założenia do modelu układu DEC są następujące:
- ciśnienie LNG po pierwszym stopniu sprężania,
𝑝2 = 532 kPa
- ciśnienie LNG po drugim stopniu sprężania,
𝑝4 = 2533 kPa
𝑝6 = 11410 kPa - ciśnienie LNG po trzecim stopniu sprężania,
129
∆𝑇𝐹𝐺 = 15 K
𝜆 = 1,05
∆𝑝𝐴 = 30 kPa
𝜂𝑖𝐶 = 0,9
∆𝑝𝑊 = 200 kPa
∆𝑇𝑊 = 5 K
𝑇𝑔𝑟 = 275,15 K
𝜂𝑖 𝑃𝑊 = 0,8
∆𝑇𝑁𝐺 = 6 K
𝜂𝑖 𝑇𝑊 = 0,85
𝜂𝑖 𝑇𝑁 = 0,9
𝜂𝑚 𝑇 = 0,98
𝜂𝐺 = 0,99
-
różnica między temperaturą spalin a temperaturą podgrzanego gazu,
stosunek nadmiaru powietrza do spalania,
przyrost ciśnienia powietrza w wentylatorze,
sprawność wewnętrzna wentylatora,
przyrost ciśnienia wody morskiej w pompie,
maksymalny spadek temperatury wody morskiej,
minimalna temperatura wody morskiej odprowadzanej do otoczenia,
sprawność wewnętrzna pompy wody,
różnica między temperaturą wody a temperaturą przegrzewanego gazu,
sprawność wewnętrzna turbiny wysokoprężnej,
sprawność wewnętrzna turbiny niskoprężnej,
sprawność mechaniczna turbin,
sprawność generatora.
Wartości ciśnień w punktach 2, 4 i 6 są wynikiem przeprowadzonej przy pomocy programu
EES optymalizacji. Za kryterium optymalizacji przyjęto maksymalizację mocy elektrycznej
netto układu.
3.4
Układ z dwoma obiegami Braytona (BC)
Drugi z zamodelowanych układów z odzyskiem egzergii kriogenicznej bazuje na modelu
przedstawionym w [16]. W jego skład wchodzą trzy obiegi. Górny obieg jest zwykłym obiegiem otwartym turbiny gazowej (OBC, open Brayton cycle). Dolny obieg jest otwartym obiegiem bezpośredniego rozprężania LNG. Obieg środkowy jest zamkniętym obiegiem Braytona
(CBC, closed Brayton cycle), w którym czynnikiem obiegowym jest azot. Schemat układu
zaprezentowano na rysunku 3.4.
LNG (1) jest sprężany do ciśnienia wyższego niż wymagane w rurociągu przesyłowym (2), po
czym trafia do wymiennika regazyfikacyjnego, gdzie odparowuje i ogrzewa się pobierając
ciepło od czynnika w zamkniętym obiegu Braytona (CBC). Para gazu ziemnego (3) jest rozprężana w turbinie NG do ciśnienia roboczego rurociągu przesyłowego (4). Przed wysłaniem
do sieci jest dogrzewana do wymaganej temperatury (5) przez spaliny wylotowe. Część LNG
jest zużywana jako paliwo zasilające turbinę gazową. Strumień ten (6) jest sprężany do ciśnienia panującego w komorze spalania (7), a następnie odparowuje, pobierając ciepło od
powietrza międzystopniowego. Odparowane paliwo (8) kierowane jest do komory spalania.
Powietrze atmosferyczne do spalania w turbinie gazowej (9) jest sprężane dwustopniowo. Po
pierwszym stopniu sprężania (10) jest chłodzone w wymienniku powietrze-powietrze (11)
oraz w odparowywaczu paliwa, na wylocie z którego ma swoją początkową temperaturę (12).
Powietrze chłodzące jest pobierane z otoczenia (22) i sprężane (23), aby pokonać opory przepływu. Po odebraniu ciepła jest odprowadzane do otoczenia (24). Powietrze do spalania jest
następnie sprężane w drugim stopniu sprężarki i kierowane do komory spalania (13). Spaliny
(14) rozprężają się w turbinie OBC (15), po czym oddają ciepło czynnikowi w zamkniętym
obiegu Braytona (16) i po ogrzaniu dodatkowo gazu ziemnego odprowadzane są do otoczenia
(17). Po oddaniu ciepła w wymienniku regazyfikacyjnym czynnik w obiegu zamkniętym (18)
jest sprężany w sprężarce CBC. Sprężony czynnik (19) pobiera ciepło od spalin z obiegu
otwartego (20) i rozpręża się w turbinie CBC. Po opuszczeniu turbiny (21) czynnik schładza
się przekazując ciepło do regazyfikowanego LNG. Moc turbin (26, 30, 32) napędza sprężarki
130
(28, 29, 31) oraz generator prądu elektrycznego. Po zaspokojeniu potrzeb własnych – mocy
pomp LNG (25, 27) i mocy wentylatora powietrza chłodzącego (33) – reszta wytworzonej
mocy elektrycznej (34) jest odprowadzana do sieci.
Rysunek 3.4 Schemat układu z obiegami Braytona (BC)
Założenia do modelu układu BC:
- ciśnienie LNG po sprężeniu,
𝑝2 = 27000 kPa
- ciśnienie międzystopniowe w obiegu otwartym,
𝑝10 = 579 kPa
- stosunek sprężania w obiegu otwartym,
𝜉 = 43,5
- stosunek nadmiaru powietrza do spalania,
𝜆 = 2,166
- dolne ciśnienie w obiegu zamkniętym,
𝑝18 = 280 kPa
- górne ciśnienie w obiegu zamkniętym,
𝑝19 = 4270 kPa
∆𝑇1𝑔 = 15 K
- górne spiętrzenie temperatur w wymienniku CBC-NG,
- dolne spiętrzenie temperatur w wymienniku CBC-NG,
∆𝑇1𝑑 = 36 K
∆𝑇2𝑔 = 20 K
- górne spiętrzenie temperatur w wymienniku OBC-CBC,
131
∆𝑇2𝑑 = 20 K
∆𝑇3𝑔 = 20 K
∆𝑇4𝑔 = 10 K
∆𝑝𝐴 = 5 kPa
𝜂𝑖 𝐶𝑂𝐵𝐶 = 0,9
𝜂𝑖 𝐶𝐶𝐵𝐶 = 0,85
𝜂𝑖 𝐶𝐴 = 0,9
𝜂𝑖 𝑇𝑁𝐺 = 0,85
𝜂𝑖 𝑇𝐶𝐵𝐶 = 0,85
𝜂𝑖 𝑇𝑂𝐵𝐶 = 0,94
𝜂𝑚 𝐶 = 0,98
𝜂𝑚 𝑇 = 0,98
𝜂𝐺 = 0,99
3.5
-
dolne spiętrzenie temperatur w wymienniku OBC-CBC,
górne spiętrzenie temperatur w wymienniku powietrze-powietrze,
górne spiętrzenie temperatur w wymienniku powietrze-paliwo,
przyrost ciśnienia powietrza chłodzącego w wentylatorze,
sprawność wewnętrzna sprężarek OBC,
sprawność wewnętrzna sprężarki CBC,
sprawność wewnętrzna wentylatora powietrza chłodzącego,
sprawność wewnętrzna turbiny NG,
sprawność wewnętrzna turbiny CBC,
sprawność wewnętrzna turbiny OBC,
sprawność mechaniczna sprężarek,
sprawność mechaniczna turbin,
sprawność generatora.
Metodyka obliczeń, definicja egzergii
Opracowane modele oparte są o równania bilansu masy i energii, równania przemian adiabatycznych w maszynach przepływowych oraz równania stechiometrii spalania. Założono, że
stałe są sprawności maszyn, ciśnienia w poszczególnych punktach układu oraz wybrane spiętrzenia temperatur w wymiennikach ciepła. W modelach założono, że LNG jest czystym chemicznie metanem (niewielkie udziały innych składników nie wpływają znacząco na wyniki
równań, na których bazuje model obliczeniowy). Pominięto straty ciśnienia oraz straty ciepła
w wymiennikach ciepła i rurociągach. Założono, że spalanie gazu ziemnego jest całkowite
i zupełne. Jako temperaturę odniesienia dla obliczania entalpii oraz egzergii przyjęto średnioroczną temperaturę powietrza w Świnoujściu, czyli 282,18 K. Poziom odniesienia dla obliczania egzergii jest zatem stały, niezależny od zmiennych w skali roku parametrów otoczenia.
Skutkuje to dodatnimi wartościami egzergii powietrza i wody morskiej pobieranej z otoczenia, jeśli temperatura tych czynników różni się od temperatury średniorocznej. Przyjęcie
zmiennego w ciągu roku poziomu odniesienia utrudniałoby interpretację efektywności działania analizowanych układów w funkcji temperatury otoczenia, gdyż zmienna byłaby wartość
egzergii LNG w zbiornikach (występująca w mianowniku wzoru 3.5).
Wykonano symulację działania wszystkich układów dla zmiennej w skali roku temperatury
otoczenia. Dane klimatyczne dla Świnoujścia zaczerpnięto z obowiązującej bazy danych klimatycznych [22]. Za temperaturę wody morskiej w poszczególnych dobach przyjęto średnią
dobową temperatury powietrza. Na rysunku 3.5 zaprezentowano uporządkowany wykres temperatury powietrza dla Świnoujścia. Sposób jego interpretacji (oraz innych wykresów uporządkowanych występujących w tej pracy) wyjaśniono na następującym przykładzie: temperatura powietrza w zakresie 5÷15°C występuje przez około (7000−3000), czyli około 4000 godzin w roku.
132
Rysunek 3.5 Uporządkowany wykres temperatur powietrza dla Świnoujścia
Dla każdego układu zestawiono w tabelach parametry termodynamiczne wszystkich strumieni
dla wybranego punktu pracy. Sporządzono także uporządkowane względem temperatury otoczenia wykresy mocy elektrycznej 𝑁𝑒𝑙 pobieranej lub oddawanej przez układ, strumienia gazu
ziemnego spalanego w instalacji 𝑚̇𝐹 oraz sprawności egzergetycznej 𝜂𝐵 . Obliczono średnio̅𝑒𝑙 𝑖𝑛 lub oddawaną 𝑁
̅𝑒𝑙 𝑜𝑢𝑡 ), średnioroczną sprawność
roczną moc elektryczną (pobieraną 𝑁
egzergetyczną 𝜂̅𝐵 , całkowitą ilość gazu ziemnego spalonego 𝑚𝐹 oraz wysłanego do sieci 𝑚𝑁𝐺
w skali roku, średnioroczny wskaźnik zużycia paliwa 𝜀𝐹 , całkowitą ilość wyprodukowanej lub
zużytej energii elektrycznej 𝐸𝑒𝑙 i skumulowany wskaźnik zużycia energii pierwotnej na jednostkę energii chemicznej gazu dostarczonego do systemu przesyłowego 𝑤𝑁𝐺 . Dla trzech
układów obliczono też roczną oszczędność energii pierwotnej 𝑃𝐸𝑆 oraz roczną oszczędność
gazu ziemnego ∆𝑚𝑁𝐺 w porównaniu do układu referencyjnego.
Sprawność energetyczna nie jest odpowiednim wskaźnikiem do określania efektywności
układów regazyfikacji z kilku powodów. Po pierwsze, celem działania terminalu regazyfikacyjnego jest dostarczenie do sieci przesyłowej gazu ziemnego pod odpowiednim ciśnieniem.
Entalpia gazów idealnych zależy tylko od temperatury, zatem entalpia fizyczna gazu wysyłanego przez terminal jest bliska zeru. Po drugie, entalpia LNG ma wartość ujemną. Po trzecie,
analiza energetyczna nie uwzględnia różnicy jakościowej pomiędzy różnymi postaciami energii (energia elektryczna, energia chemiczna, ciepło). Do określania potencjału termodynamicznego substancji o temperaturach niższych od temperatury otoczenia lub o ciśnieniu różnym od ciśnienia otoczenia służy egzergia. Egzergia jest miarą jakości energii. Jest definiowana jako maksymalna zdolność rozpatrywanej porcji energii do wykonania pracy z wykorzystaniem ciepła pobranego z otoczenia oraz powszechnie występujących i wzajemnie niezależnych składników otoczenia [23]. Można wyodrębnić różne składowe egzergii; na potrzeby
analizy termodynamicznej układów regazyfikacji LNG wystarcza egzergia termiczna, która
dzieli się na część fizyczną i chemiczną.
133
Egzergia fizyczna 𝑏𝑓 wynika z różnicy parametrów termodynamicznych danej substancji
(temperatury 𝑇 i ciśnienia 𝑝) względem otoczenia. Obliczana jest ze wzoru:
𝑏𝑓 (𝑇; 𝑝) = [𝑖(𝑇; 𝑝) − 𝑖(𝑇0 ; 𝑝0 )] − 𝑇0 ∙ [𝑠(𝑇; 𝑝) − 𝑠(𝑇0 ; 𝑝0 )]
(3.1)
gdzie:
- ciśnienie otoczenia,
𝑝0
- temperatura otoczenia,
𝑇0
- entropia właściwa,
𝑠
- entalpia właściwa.
𝑖
W egzergii fizycznej można wyodrębnić część temperaturową 𝑏𝑓 𝑇 i ciśnieniową 𝑏𝑓 𝑝 :
𝑏𝑓 𝑇 (𝑇) = [𝑖(𝑇; 𝑝0 ) − 𝑖(𝑇0 ; 𝑝0 )] − 𝑇0 ∙ [𝑠(𝑇; 𝑝0 ) − 𝑠(𝑇0 ; 𝑝0 )]
𝑏𝑓 𝑝 (𝑝) = [𝑖(𝑇0 ; 𝑝) − 𝑖(𝑇0 ; 𝑝0 )] − 𝑇0 ∙ [𝑠(𝑇0 ; 𝑝) − 𝑠(𝑇0 ; 𝑝0 )]
(3.2)
(3.3)
Egzergia chemiczna 𝑏𝑐ℎ danej substancji wynika z ilości wydzielanej energii w reakcjach
chemicznych prowadzących do przekształcenia tej substancji w substancję odniesienia oraz
z różnicy pomiędzy składem chemicznym danej substancji a stężeniem konkretnych jej składników w otoczeniu. Obliczana jest ze wzoru:
∑(𝑧𝑖 ∙ 𝑀𝑏𝑐ℎ 𝑖 ) + 𝑇0 ∙ (𝑀𝑅) ∙ ∑(𝑧𝑖 ∙ ln 𝑧𝑖 )
(3.4)
𝑏𝑐ℎ =
𝑀
gdzie:
- udział molowy i-tego składnika substancji,
𝑧𝑖
𝑀𝑏𝑐ℎ 𝑖 - normalna egzergia chemiczna i-tego składnika (odczytywana z tablic [23]),
(𝑀𝑅) - uniwersalna stała gazowa,
- zastępcza masa molowa substancji.
𝑀
Chwilową (3.5) i średnią (3.6) sprawność egzergetyczną układów zdefiniowano następująco [16]:
𝑚̇𝑁𝐺 ∙ (𝑏𝑝 𝑜𝑢𝑡 − 𝑏𝑝 𝑖𝑛 ) + 𝑁𝑒𝑙 𝑜𝑢𝑡
𝑚̇𝐿𝑁𝐺 ∙ 𝑏𝑡 𝑖𝑛 + 𝑚̇𝐹 ∙ 𝑏𝑐ℎ 𝐹 + 𝑁𝑒𝑙 𝑖𝑛
𝑚𝑁𝐺 ∙ (𝑏𝑝 𝑜𝑢𝑡 − 𝑏𝑝 𝑖𝑛 ) + 𝐸𝑒𝑙 𝑜𝑢𝑡
𝜂̅𝐵 =
𝑚𝐿𝑁𝐺 ∙ 𝑏𝑡 𝑖𝑛 + 𝑚𝐹 ∙ 𝑏𝑐ℎ 𝐹 + 𝐸𝑒𝑙 𝑖𝑛
𝜂𝐵 =
gdzie:
𝑏𝑝 𝑖𝑛
𝑏𝑝 𝑜𝑢𝑡
𝑏𝑡 𝑖𝑛
𝑏𝑐ℎ 𝐹
-
(3.5)
(3.6)
egzergia ciśnieniowa LNG w zbiorniku,
egzergia ciśnieniowa LNG w rurociągu,
egzergia temperaturowa LNG w zbiorniku,
egzergia chemiczna metanu.
Osłona bilansowa obejmuje część terminalu regazyfikacyjnego pomiędzy zbiornikami LNG
a początkiem gazociągu przesyłowego. Jako efekt użyteczny działania układu uznano przyrost
egzergii ciśnieniowej gazu ziemnego i (tylko dla układów z odzyskiem energii) wytwarzaną
moc elektryczną netto. Za nakład przyjęto egzergię temperaturową LNG, egzergię chemiczną
spalanego gazu i (tylko dla układów bez odzysku energii) pobieraną z sieci moc elektryczną.
We wzorze na sprawność nie jest ujęta egzergia chemiczna całości gazu ziemnego przepływającego przez instalację, a jedynie części spalanej. Ponieważ egzergia fizyczna LNG stanowi
poniżej 1% jego egzergii chemicznej, uwzględnienie egzergii chemicznej przy wszystkich
134
strumieniach gazu ziemnego prowadziłoby do wyników przekraczających 99% i utrudniałoby
porównanie sprawności układów między sobą.
Wskaźnik zużycia gazu w terminalu regazyfikacyjnym 𝜀𝐹 zdefiniowano następująco:
𝑚𝐹
𝜀𝐹 =
(3.7)
𝑚𝐿𝑁𝐺
Wskaźnik skumulowanego zużycia energii pierwotnej na jednostkę energii chemicznej gazu
dostarczonego do systemu przesyłowego 𝑤𝑁𝐺 zdefiniowano następująco:
∗
𝐸𝑖𝑛
(3.8)
𝑤𝑁𝐺 =
𝐸𝑜𝑢𝑡
gdzie:
∗
- skumulowana energia pierwotna dostarczana do terminalu,
𝐸𝑖𝑛
- energia chemiczna gazu wysyłanego do sieci.
𝐸𝑜𝑢𝑡
∗
Wielkości 𝐸𝑖𝑛
i 𝐸𝑜𝑢𝑡 są zdefiniowane poniżej. Jeżeli układ wytwarza energię elektryczną, to
∗
𝐸𝑖𝑛 jest obliczana ze wzoru 3.10, w przeciwnym przypadku ze wzoru 3.9:
∗
𝐸𝑖𝑛
= 𝑤𝐿𝑁𝐺 ∙ 𝑚𝐿𝑁𝐺 ∙ 𝑊𝑑 CH4 + 𝑤𝑒𝑙 ∙ 𝐸𝑒𝑙 𝑖𝑛
(3.9)
∗
𝐸𝑖𝑛 = 𝑤𝐿𝑁𝐺 ∙ 𝑚𝐿𝑁𝐺 ∙ 𝑊𝑑 CH4 − 𝑤𝑒𝑙 ∙ 𝐸𝑒𝑙 𝑜𝑢𝑡
(3.10)
𝐸𝑜𝑢𝑡 = 𝑚𝑁𝐺 ∙ 𝑊𝑑 CH4
(3.11)
gdzie:
𝑊𝑑 CH4 - wartość opałowa gazu ziemnego równa 50 MJ/kg,
- wskaźnik skumulowanego zużycia energii pierwotnej na produkcję jednostki
𝑤𝑒𝑙
energii elektrycznej, w Polsce równy 3,22 [24],
- wskaźnik skumulowanego zużycia energii pierwotnej na dostawę jednostki
𝑤𝐿𝑁𝐺
LNG.
Przyjęto wartość 𝑤𝐿𝑁𝐺 równą 1, ponieważ przeprowadzana analiza ma na celu porównanie
poszczególnych układów regazyfikacji między sobą. Aby porównać energochłonność skumulowaną dostaw LNG z energochłonnością skumulowaną dostaw gazu do Polski z innych źródeł (np. import rosyjskiego gazu poprzez gazociągi) należałoby przyjąć rzeczywistą wartość
𝑤𝐿𝑁𝐺 , większą od 1. Wartość ta powinna uwzględniać zużycie energii pierwotnej obarczające
produkcję i dostawę LNG.
Roczna oszczędność energii pierwotnej 𝑃𝐸𝑆 oraz roczna oszczędność gazu ziemnego ∆𝑚𝑁𝐺
(bezwzględne i względne) są obliczane ze wzorów:
𝑚𝑁𝐺 0
∗
∗
𝑃𝐸𝑆 = 𝐸𝑖𝑛
(3.12)
0 − 𝐸𝑖𝑛 ∙
𝑚𝑁𝐺
𝑃𝐸𝑆
𝑃𝐸𝑆% = ∗
(3.13)
𝐸𝑖𝑛 0
𝑚𝑁𝐺 0
∆𝑚𝑁𝐺 = 𝑚𝐿𝑁𝐺 ∙ (1 −
)
(3.14)
𝑚𝑁𝐺
∆𝑚𝑁𝐺
(3.15)
∆𝑚𝑁𝐺 % =
𝑚𝐿𝑁𝐺
gdzie indeks „0” odnosi się do układu referencyjnego (SCV).
135
4
Wyniki obliczeń
W rozdziale przedstawiono wyniki obliczeń dla poszczególnych układów oraz porównanie ich
między sobą. Dla każdego układu zestawiono w tabelach parametry termodynamiczne
wszystkich strumieni w wybranym punkcie pracy i wszystkie wartości mocy mechanicznych
lub elektrycznych występujących w układzie. Zamieszczono uporządkowane względem temperatury otoczenia wykresy mocy elektrycznej generowanej lub pobieranej przez poszczególne układy, strumienia spalanego paliwa i sprawności egzergetycznej. Pod koniec rozdziału
zamieszczono wykresy zbiorcze dla analizowanych układów oraz wartości wskaźników oceny
ich pracy.
4.1
Układ SCV
Wyniki symulacji działania układu przy średniorocznej temperaturze i wilgotności powietrza
zestawiono w tabeli 4.1. W tabeli 4.2 zestawiono moce poszczególnych maszyn. Na rysunkach 4.1, 4.2 i 4.3 przedstawiono wykresy uporządkowane 𝑁𝑒𝑙 𝑖𝑛 , 𝑚̇𝐹 oraz 𝜂𝐵 . Wartości 0 na
osi odciętych odpowiada najniższa temperatura otoczenia.
Tabela 4.1 Parametry termodynamiczne wszystkich strumieni w układzie SCV
𝑗
1
2
3
4
5
6
7
8
𝑇𝑗 ,
𝑡𝑗 ,
K
°C
111,15 −162,0
116,14 −157,0
274,15
1,0
274,15
1,0
226,53 −46,6
282,18
9,0
306,34
33,2
289,15
16,0
𝑝𝑗 ,
𝑠𝑗 ,
kPa kJ·kg−1·K−1
151
−6,571
8400
−6,484
8400
−2,631
8400
−2,631
131
−0,609
101
0,000
131
0,008
101
0,025
𝑖𝑗 ,
𝑏𝑓 𝑗 ,
1
−
kJ·kg
kJ·kg−1
−876,3 977,8
−847,0 982,7
−120,3 621,9
−120,3 621,9
51,5
−120,3
0,0
0,0
24,5
22,2
7,0
0,1
𝑏𝑐ℎ 𝑗 ,
kJ·kg−1
51800,9
51800,9
51800,9
51800,9
51800,9
0,0
0,0
55,6
𝑚̇𝑗 ,
kg·s−1
113,50
113,50
112,02
1,48
1,48
26,93
26,93
25,18
Tabela 4.2 Zestawienie mocy występujących w układzie SCV
9
𝑗
𝑁𝑗 , kW −3697
10
−733
11
−4430
Pompa LNG pobiera moc około 3,7 MW, co stanowi około 83% zapotrzebowania układu na
energię elektryczną, które wynosi 4,4 MW. Ujemne wartości entalpii i entropii właściwej
LNG są skutkiem przyjęcia jako stanu odniesienia średniorocznej temperatury otoczenia. Nie
ma to jednak znaczenia z punktu widzenia wyników analizy egzergetycznej.
136
Rysunek 4.1 Uporządkowany wykres poboru mocy elektrycznej przez układ SCV
Pobór mocy elektrycznej przez układ jest rosnącą funkcją temperatury otoczenia. Wpływa na
to zmienna moc wentylatora powietrza do spalania. Wynika to ze zmiany gęstości powietrza.
Przy wyższych temperaturach gęstość powietrza jest mniejsza i zwiększa się jego strumień
objętościowy.
Rysunek 4.2 Uporządkowany wykres zużycia paliwa przez układ SCV
137
Zużycie paliwa w układzie referencyjnym jest praktycznie stałe przez cały rok i wynosi około
1,48 kg/s. Przy niskich temperaturach powietrza trzeba spalić nieznacznie więcej paliwa, aby
uzyskać wymaganą temperaturę gazu ziemnego.
Rysunek 4.3 Uporządkowany wykres sprawności egzergetycznej układu referencyjnego
Sprawność egzergetyczna układu jest również niemal niezmienna przez cały rok. Wyższa
sprawność przy wyższych temperaturach otoczenia wynika z mniejszego strumienia spalanego gazu. Rozmyty charakter wykresów wynika z faktu, że temperatura otoczenia nie jest jedyną zmienną niezależną w analizie. Zmienną niezależną jest także wilgotność powietrza.
4.2
Układ SCV + ORV
W tabeli 4.3 zestawiono wyniki symulacji działania układu przy temperaturze otoczenia równej 276 K, czyli w stanie gdy pracują jednocześnie oba układy dostarczania ciepła na potrzeby
regazyfikacji. W tabeli 4.4 zestawiono moce poszczególnych maszyn. Na rysunkach 4.4, 4.5
i 4.6 przedstawiono wykresy uporządkowane 𝑁𝑒𝑙 𝑖𝑛 , 𝑚̇𝐹 oraz 𝜂𝐵 . Wartości 0 na osi odciętych
odpowiada najniższa temperatura wody morskiej.
138
Tabela 4.3 Parametry termodynamiczne wszystkich strumieni w układzie SCV+ORV
𝑗
1
2
3
4
5
6
7
8
9
10
11
𝑇𝑗 ,
𝑡𝑗 ,
K
°C
111,15 −162,0
116,14 −157,0
274,15
1,0
274,15
1,0
226,53 −46,6
276,00
2,9
299,65
26,5
289,15
16,0
276,00
2,9
276,01
2,9
275,15
2,0
𝑝𝑗 ,
𝑠𝑗 ,
kPa kJ·kg−1·K−1
151
−6,571
8400
−6,484
8400
−2,631
8400
−2,631
131
−0,609
101
−0,023
131
−0,015
101
0,025
101
−0,093
301
−0,093
101
−0,106
𝑖𝑗 ,
𝑏𝑓 𝑗 ,
1
−
kJ·kg kJ·kg−1
−876,3 977,8
−847,0 982,7
−120,3 621,9
−120,3 621,9
−120,3
51,5
−6,5
0,1
17,5
21,6
7,0
0,1
−25,9
0,3
−25,7
0,5
−29,5
0,4
𝑏𝑐ℎ 𝑗 ,
𝑚̇𝑗 ,
1
−
kJ·kg
kg·s−1
51800,5 113,50
51800,5 113,50
51800,5 112,28
51800,5
1,22
51800,5
1,22
0,0
22,10
0,0
22,10
55,3
20,72
0,0 3891,36
0,0 3891,36
0,0 3891,36
Tabela 4.4 Zestawienie mocy występujących w układzie SCV+ORV
12
𝑗
𝑁𝑗 , kW −3697
13
−588
14
−1081
15
−5366
Strumień wody morskiej jest około 34 razy większy od strumienia odparowywanego LNG.
Pompa wody morskiej pobiera około 1,1 MW mocy, czyli więcej niż wentylator powietrza do
spalania. Układ działając w trybie ORV zużywa więcej energii elektrycznej, niż gdy pracuje
w trybie SCV.
Rysunek 4.4 Uporządkowany wykres poboru mocy elektrycznej przez układ SCV+ORV
139
Na wykresach można wyraźnie rozróżnić trzy stany pracy układu. W trybie pracy ORV układ
regazyfikacji pobiera około 4,8 MW mocy, a w trybie SCV około 4,4 MW mocy. Szczytowe
zapotrzebowanie na energię elektryczną (w trybie mieszanym) wynosi około 5,5 MW.
Rysunek 4.5 Uporządkowany wykres zużycia paliwa przez układ SCV+ORV
Przez około 5000 godzin w roku układ regazyfikacji pracuje w trybie ORV i wówczas nie
zużywa paliwa. Okres pracy w trybie SCV wynosi około 1500 godzin, a w trybie mieszanym
około 2200 godzin.
Rysunek 4.6 Uporządkowany wykres sprawności egzergetycznej układu SCV+ORV
140
Zwiększanie udziału ciepła pochodzącego od wody morskiej w procesie regazyfikacji prowadzi do wzrostu sprawności egzergetycznej do wartości około 55%, gdy układ pracuje całkowicie w trybie ORV. Wpływ ilości spalanego gazu na sprawność egzergetyczną jest większy
niż wpływ ilości zużywanej energii elektrycznej, dlatego mimo większego poboru mocy
w trybie ORV niż w trybie SCV, sprawność egzergetyczna jest wyższa w trybie ORV.
4.3
Układ DEC
W tabeli 4.5 zestawiono wyniki symulacji działania układu DEC przy temperaturze otoczenia
równej 276 K, czyli w stanie gdy pracują jednocześnie oba układy dostarczania ciepła na potrzeby regazyfikacji. W tabeli 4.6 zestawiono moce poszczególnych maszyn. Na rysunkach
4.7, 4.8 i 4.9 przedstawiono wykresy uporządkowane 𝑁𝑒𝑙 𝑖𝑛 , 𝑚̇𝐹 oraz 𝜂𝐵 . Wartości 0 na osi
odciętych odpowiada najniższa temperatura wody morskiej.
Tabela 4.5 Parametry termodynamiczne wszystkich strumieni w układzie DEC
𝑗
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
𝑇𝑗 ,
K
111,15
111,38
136,48
138,07
172,36
185,46
274,15
254,49
187,33
187,33
274,15
192,03
274,15
274,15
226,53
276,00
299,58
289,15
276,00
276,01
275,15
𝑡𝑗 ,
𝑝𝑗 ,
𝑠𝑗 ,
kPa kJ·kg−1·K−1
°C
−162,0
151
−6,569
−161,8
532
−6,565
−136,7
532
−5,836
−135,1 2533
−5,817
−100,8 2533
−4,881
−87,7 11410
−4,796
1,0 11410
−2,894
−18,7 8400
−2,875
−85,8 2533
−2,795
−85,8 2533
−2,795
1,0 2533
−1,806
−81,1
532
−1,713
1,0 8400
−2,629
1,0 8400
−2,629
−46,6
131
−0,609
2,9
101
−0,023
26,4
131
−0,015
16,0
101
0,025
2,9
101
−0,093
2,9
301
−0,093
2,0
101
−0,106
𝑖𝑗 ,
𝑏𝑓 𝑗 ,
1
−
kJ·kg kJ·kg−1
−876,3 977,3
−874,9 977,5
−784,8 861,9
−776,9 864,4
−631,4 746,0
−587,4 766,0
−158,5 658,0
−185,3 626,0
−269,6 519,0
−269,6 519,0
−46,0 463,5
−202,0 281,5
−120,3 621,5
−120,3 621,5
−120,3
51,4
−6,5
0,1
17,4
21,6
7,0
0,1
−25,9
0,3
−25,7
0,5
−29,5
0,4
𝑏𝑐ℎ 𝑗 ,
𝑚̇𝑗 ,
1
−
kJ·kg
kg·s−1
51800,5 113,50
51800,5 113,50
51800,5 131,06
51800,5 131,06
51800,5 183,77
51800,5 183,77
51800,5 183,77
51800,5 113,50
51800,5
52,70
51800,5
17,56
51800,5
17,56
51800,5
17,56
51800,5 112,16
51800,5
1,34
51800,5
1,34
0,0
24,17
0,0
24,17
55,3
22,65
0,0 4238,90
0,0 4238,90
0,0 4238,90
Tabela 4.6 Zestawienie mocy występujących w układzie DEC
22
𝑗
𝑁𝑗 , kW −171
23
−1144
24
−8982
25
10633
26
2685
27
−1177
28
−641
29
1070
141
W układzie cyrkuluje 52,7 kg/s gazu ziemnego w części wysokoprężnej i 17,56 kg/s gazu
w części niskoprężnej. Strumień wody morskiej jest większy niż w przypadku układu
SCV+ORV. Wynika to ze zwiększonego strumienia LNG, który trzeba odparować oraz dogrzać po rozprężeniu w turbinach. Pompy LNG pobierają łącznie 10,3 MW; turbiny produkują 13,3 MW.
Rysunek 4.7 Uporządkowany wykres produkcji mocy elektrycznej przez układ DEC
W trybie pracy SCV układ DEC produkuje około 2,1 MW mocy elektrycznej. Po włączeniu
pompy wody morskiej moc układu spada do 1 MW, a w momencie wyłączenia wentylatora
powietrza do spalania wynosi 1,7 MW. Wraz ze wzrostem temperatury wody morskiej moc
układu rośnie i osiąga 2,7 MW dla najwyższej występującej w ciągu roku temperatury.
Rysunek 4.8 Uporządkowany wykres zużycia paliwa przez układ DEC
142
Kształt uporządkowanego wykresu zużycia paliwa w funkcji czasu jest jednakowy, jak dla
układu SCV+ORV bez odzysku egzergii. Nieznacznie większe jest tylko zużycie paliwa
w trybie pracy SCV.
Rysunek 4.9 Uporządkowany wykres sprawności egzergetycznej układu DEC
Uporządkowany wykres sprawności ma również podobny przebieg jak w przypadku układu
SCV+ORV bez odzysku energii, ale osiągane sprawności są nieco wyższe. Przy wysokich
temperaturach otoczenia sprawność egzergetyczna układu DEC zbliża się do 60%.
4.4
Układ BC
Wyniki symulacji działania układu przy średniorocznej temperaturze i wilgotności powietrza
zestawiono w tabeli 4.7. W tabeli 4.8 zestawiono moce poszczególnych maszyn. Na rysunkach 4.10, 4.11 i 4.12 przedstawiono wykresy uporządkowane 𝑁𝑒𝑙 𝑖𝑛 , 𝑚̇𝐹 oraz 𝜂𝐵 . Wartości
0 na osi odciętych odpowiada najniższa temperatura otoczenia.
143
Tabela 4.7 Parametry termodynamiczne wszystkich strumieni w układzie BC
𝑗
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
𝑇𝑗 ,
K
111,15
127,49
341,72
263,57
274,15
111,15
113,72
294,41
282,18
481,81
304,41
282,18
524,21
1562,84
693,96
411,95
403,27
163,49
391,95
673,96
356,72
282,18
286,52
461,81
𝑡𝑗 ,
°C
−162,0
−145,7
68,6
−9,6
1,0
−162,0
−159,4
21,3
9,0
208,7
31,3
9,0
251,1
1289,7
420,8
138,8
130,1
−109,7
118,8
400,8
83,6
9,0
13,4
188,7
𝑝𝑗 ,
𝑠𝑗 ,
kPa kJ·kg−1·K−1
151
−6,569
27000
−6,300
27000
−2,842
8400
−2,620
8400
−2,629
151
−6,569
4408
−6,524
4408
−1,972
101
0,000
579
0,043
579
−0,427
579
−0,504
4408
−0,456
4408
0,881
101
0,986
101
0,405
101
0,382
280
−0,876
4270
−0,782
4270
−0,196
280
−0,058
101
0,000
106
0,002
101
0,504
𝑖𝑗 ,
𝑏𝑓 𝑗 ,
1
−
kJ·kg
kJ·kg−1
977,3
−876,3
996,8
−780,9
775,8
−26,1
585,1
−154,2
−120,3
621,5
977,3
−876,3
−861,2
979,7
539,2
−17,3
0,0
0,0
204,6
192,4
22,5
143,2
0,0
142,3
249,0
377,6
1545,0 1296,4
454,1
175,8
138,9
24,7
129,5
21,8
122,3
−125,0
110,1
330,9
415,4
470,7
77,4
93,8
0,0
0,0
4,4
4,0
183,8
41,7
𝑏𝑐ℎ 𝑗 ,
kJ·kg−1
51800,9
51800,9
51800,9
51800,9
51800,9
51800,9
51800,9
51800,9
0,0
0,0
0,0
0,0
0,0
25,1
25,1
25,1
25,1
20,1
20,1
20,1
20,1
0,0
0,0
0,0
𝑚̇𝑗 ,
kg·s−1
103,75
103,75
103,75
103,75
103,75
9,75
9,75
9,75
364,97
364,97
364,97
364,97
364,97
374,72
374,72
374,72
374,72
386,82
386,82
386,82
386,82
370,36
370,36
370,36
Tabela 4.8 Zestawienie mocy występujących w układzie BC
25
𝑗
𝑁𝑗 , kW −10099
30
𝑗
𝑁𝑗 , kW 400634
26
13024
31
−92725
27
−164
32
128120
28
−76209
33
−1811
29
−92736
34
264273
Strumień czynnika w zamkniętym obiegu Braytona wynosi 387 kg/s. Podobnej wielkości są
strumienie powietrza do spalania (365 kg/s), spalin (375 kg/s) i powietrza chłodzącego
(370 kg/s). Stosunkowo wysokie są temperatury spalin wylotowych (130°C) i powietrza chłodzącego (189°C). Strumienie te stanowią źródło możliwego potencjalnie do dalszego wykorzystania ciepła odpadowego. Obieg NG (maszyny 25 i 26) odpowiada za produkcję blisko
3 MW mocy, obieg CBC (maszyny 31 i 32) 36 MW, a obieg OBC (maszyny 27, 28, 29, 30
i 33) 230 MW.
144
Rysunek 4.10 Uporządkowany wykres produkcji mocy elektrycznej przez układ BC
Wartość mocy elektrycznej wytwarzanej przez układ BC zależy od temperatury otoczenia.
Waha się ona od 258 MW w miesiącach letnich do 269 MW w miesiącach zimowych. Przy
wyższych temperaturach otoczenia większa jest praca sprężania powietrza do spalania
i mniejsza jest moc netto wytwarzana przez zespół turbiny gazowej.
Rysunek 4.11 Uporządkowany wykres zużycia paliwa przez układ BC
145
Zużycie paliwa maleje wraz ze wzrostem temperatury otoczenia. Jest to spowodowane faktem, że im wyższa jest temperatura wlotowa powietrza, tym wyższa będzie temperatura powietrza przed komorą spalania i tym mniej paliwa należy spalić, aby zapewnić wymagany
strumień ciepła przekazany przez spaliny do kolejnych obiegów.
Rysunek 4.12 Uporządkowany wykres sprawności egzergetycznej układu BC
Sprawność egzergetyczna układu BC tylko w niewielkim stopniu zależy od temperatury otoczenia. Niższe wartości sprawności dla wyższych temperatur są skutkiem mniejszej mocy
elektrycznej wytwarzanej przez układ.
4.5
Porównanie układów
Na rysunku 4.13 zestawiono wykresy uporządkowane produkcji mocy elektrycznej dla układów SCV, SCV+ORV i DEC (ze względu na czytelność wykresu pominięto układ BC, który
charakteryzuje się znacznie większą wartością generowanej mocy niż pozostałe układy). Na
rysunku 4.14 zestawiono razem wykresy uporządkowane sprawności egzergetycznej dla
wszystkich zamodelowanych układów.
146
Rysunek 4.13 Uporządkowany wykres mocy elektrycznej pobieranej lub oddawanej przez
analizowane układy
Na podstawie porównania wykresów dla układów SCV+ORV bez odzysku energii i układu
DEC można wywnioskować, że układ DEC pozwala odzyskać z egzergii kriogenicznej i przetworzyć na elektryczność około 6,5 MW.
Rysunek 4.14 Uporządkowany wykres sprawności egzergetycznej analizowanych układów
147
Przy wysokich temperaturach otoczenia najlepszą sprawność egzergetyczną osiągają układy
regazyfikacji wykorzystujące metodę ORV. Układ BC mimo dużej mocy elektrycznej ma od
nich mniejszą sprawność egzergetyczną. Spowodowane jest to dużymi stratami egzergii
w procesie spalania paliwa. Jednak pozostałe układy działając w trybie SCV charakteryzują
się o wiele niższą sprawnością niż układ BC.
W tabeli 4.9 porównano wskaźniki oceny pracy badanych układów.
Tabela 4.9 Zestawienie wskaźników oceny wszystkich układów
Parametr
Jednostka
̅𝑒𝑙
𝑁
𝐸𝑒𝑙
𝑚𝐹
𝑚𝑁𝐺
𝜀𝐹
𝜂̅ 𝐵
𝑤𝑁𝐺
𝑃𝐸𝑆
𝑃𝐸𝑆%
∆𝑚𝑁𝐺
∆𝑚𝑁𝐺 %
MW
GWh
Gg
Gg
%
%
MJ*/MJ
PJ
%
Gg
%
Układ
Układ
SCV
SCV+ORV
−4,43
−4,81
−38,806
−42,137
46,81
14,01
3532,53
3565,33
1,31
0,39
32,80
47,96
1,0158
1,0067
1,612
0,90
32,93
0,92
Układ
DEC
1,95
17,041
15,33
3564,01
0,43
50,94
1,0032
2,226
1,24
31,61
0,88
Układ
BC
264,34
2315,626
307,52
3271,81
8,59
52,39
0,9299
15,170
8,46
−285,22
−7,97
Zastosowanie hybrydowego układu regazyfikacji (SCV+ORV) pozwala w warunkach Świnoujścia (przy poczynionych założeniach dotyczących warunków poboru wody morskiej)
zmniejszyć zużycie paliwa ponad trzykrotnie względem układu opartego wyłącznie o metodę
SCV. Roczna oszczędność gazu ziemnego wynosi około 33 tysiące ton, czyli 0,92% zdolności
wysyłkowej terminalu. Zastosowanie zaproponowanego układu odzysku energii bazującego
na metodzie DEC tylko nieznacznie zwiększa zużycie paliwa względem metody SCV+ORV
bez odzysku, za to pozwala z niewielką nadwyżką zaspokoić potrzeby własne terminalu na
moc elektryczną. Układ BC jest w gruncie rzeczy elektrownią gazową sprzęgniętą z terminalem regazyfikacyjnym. Układ ten zużywa prawie siedmiokrotnie więcej gazu niż układ referencyjny, co w skali roku przekłada się na 285 tysięcy ton mniej gazu ziemnego dostarczonego do krajowego systemu gazowego. Rozwiązanie to nie jest zatem optymalne, jeśli za funkcję celu działania terminalu przyjąć dostarczanie jak największej ilości gazu do systemu.
Wszystkie układy charakteryzują się niższym wskaźnikiem skumulowanego zużycia energii
niż układ referencyjny i przyczyniają się do oszczędności energii pierwotnej. Układ z turbiną
gazową (BC) charakteryzuje się wskaźnikiem skumulowanego zużycia energii mniejszym od
jedności. Jest to spowodowane faktem, że sprawność wytwarzania elektryczności w tym układzie jest wyższa od średniej systemowej dla Polski.
148
5
Podsumowanie
W ramach niniejszej pracy omówiono rolę gazu ziemnego w światowej i polskiej gospodarce
energetycznej. Skupiono się na metodzie transportu gazu ziemnego w postaci skroplonej
(LNG). Opisano łańcuch produkcji i dostawy LNG oraz aktualny stan światowego rynku tego
surowca. Dokonano przeglądu metod regazyfikacji LNG i możliwości odzysku zmagazynowanej w nim egzergii kriogenicznej. Głównym elementem pracy było wykonanie w programie Engineering Equation Solver modeli obliczeniowych czterech układów regazyfikacji:
standardowego układu SCV (bez odzysku egzergii), połączenia układów SCV i ORV (bez
odzysku egzergii), układu produkującego energię elektryczną przy użyciu metody bezpośredniego rozprężania gazu ziemnego oraz układu będącego połączeniem terminalu regazyfikacyjnego z elektrownią gazową. Główne dane wejściowe do modeli wynikały z założenia, że
układy te działałyby w terminalu regazyfikacyjnym w Świnoujściu. Wykonano symulację
działania wszystkich układów dla zmiennej w skali roku temperatury otoczenia. Obliczono
szereg wskaźników służących do porównania między sobą analizowanych układów. Wykorzystano w tym celu elementy analizy egzergetycznej.
Wykazano potencjał gazoportu w Świnoujściu do wytwarzania energii elektrycznej przez odzysk egzergii kriogenicznej LNG oraz związany z tym potencjał oszczędności energii pierwotnej. Na otrzymane wartości liczbowe duży wpływ miały przyjęte założenia, zwłaszcza te
dotyczące warunków wykorzystania wody morskiej jako źródła ciepła i sprawności poszczególnych maszyn. Niezależnie jednak od wartości założonych parametrów modelu, wnioski
płynące z porównania analizowanych układów byłyby podobne. Zastosowanie zaproponowanego układu odzysku egzergii bazującego na metodzie bezpośredniego, kaskadowego rozprężania LNG tylko nieznacznie zwiększa zużycie paliwa względem układu regazyfikacji bez
odzysku, za to pozwala zaspokoić potrzeby własne terminalu na moc elektryczną. Ewentualna
nadwyżka mocy byłaby niewielka (rzędu kilku megawatów). Główną wadą tego układu jest
brak wykorzystania temperaturowej części egzergii kriogenicznej, ponieważ jest ona w całości oddawana do otoczenia. W tym zakresie można zatem upatrywać możliwości zwiększenia
efektywności układu – na przykład poprzez sprzęgnięcie z innym obiegiem termodynamicznym wykorzystującym LNG jako dolne źródło ciepła. Drugi z analizowanych układów odzysku egzergii, będący połączeniem terminalu regazyfikacyjnego z elektrownią z turbiną gazową charakteryzuje się wysoką wartością sprawności wytwarzania energii elektrycznej, wyższą
niż osiągalna w elektrowniach węglowych. Moc elektryczna netto tego układu jest rzędu
265 MW. Z tego powodu znacznie bardziej niż pozostałe układy przyczyniałby się do oszczędności energii pierwotnej w skali kraju. Zużywa on jednak kilkukrotnie więcej paliwa niż
pozostałe układy i dlatego nie jest dobrym rozwiązaniem z punktu widzenia maksymalizacji
dostaw gazu ziemnego z gazoportu do krajowego systemu gazowego.
Decyzja o budowie przy terminalu regazyfikacyjnym układu odzysku zimnej energii powinna
być oparta nie tylko o analizę termodynamiczną, ale również analizę ekonomiczną. Zaprezentowane w niniejszej pracy układy są na etapie koncepcji, w związku z czym trudno jest oszacować ich koszty inwestycyjne i eksploatacyjne, a co za tym idzie, opłacalność. W przypadku
wyrażenia przez inwestora chęci budowy w ramach świnoujskiego terminalu jednego z przeanalizowanych w ramach tej pracy układu odzysku egzergii, dalszym etapem prac powinno być
wykonanie bardziej szczegółowego modelu obliczeniowego danego układu, a następnie przeprowadzenie studium wykonalności oraz analizy ekonomicznej.
149
Literatura
[1] Background of natural gas, strona internetowa NaturalGas.org, http://naturalgas.org
/overview/background/, dostęp 23 kwietnia 2016r.
[2] International Energy Agency, Key World Energy Statistics, Paris (2015).
[3] Natural gas reserves, strona internetowa British Petroleum, http://www.bp.com/en/global/
corporate/energy-economics/statistical-review-of-world-energy/natural-gas-review-byenergy-type/natural-gas-reserves.html, dostęp 24 kwietnia 2016r.
[4] Mokhatab S., Mak J., Valappil J., Wood D., Handbook of Liquefied Natural Gas, Gulf
Professional Publishing, Houston (2013).
[5] Foss M. M., Introduction To LNG - An overview on liquefied natural gas (LNG), its properties, organization of the LNG industry and safety considerations, Center for Energy
Economics, Austin (2007).
[6] International Gas Union, 2016 World LNG report, Oslo (2016).
[7] Główny Urząd Statystyczny, Energia 2015, Warszawa (2015).
[8] Ministerstwo Gospodarki, Projekt polityki energetycznej Polski do 2050 roku,
Warszawa (2015).
[9] Czapigo-Czapla M., Bilans zasobów złóż kopalin w Polsce według stanu na dzień 31
grudnia 2014 r., Państwowy Instytut Geologiczny / Państwowy Instytut badawczy, Warszawa (2015).
[10] Import, strona internetowa PGNiG, http://www.pgnig.pl/pgnig/segmenty-dzialalnosci/
obrot-i-magazynowanie/import, dostęp 29 kwietnia 2016r.
[11] Aktualności, strona internetowa Polskiego LNG, http://www.polskielng.pl/nc/biuroprasowe/aktualnosci/, dostęp 17 czerwca 2016r.
[12] Aktualności, strona internetowa PGNiG, http://www.pgnig.pl/aktualnosci/-/newslist/id/w-czerwcu-pierwsze-dostawy-gazu-lng-dla-pgnig/newsGroupId/10184?changeYear
=2016&currentPage=1, dostęp 17 czerwca 2016r.
[13] Kaliski M., Nagy S., Rychlicki S., Siemek J., Szurlej A., Gaz ziemny w Polsce –
wydobycie, zużycie i import do 2030 roku, Górnictwo i Geologia tom 5 zeszyt 3 (2010)
[14] Terminal LNG w Polsce, strona internetowa Polskiego LNG, http://www.polskielng.pl/
lng/terminal-lng-w-polsce/, dostęp 29 kwietnia 2016r.
[15] Polskie LNG, Projekt terminalu LNG Świnoujście – Streszczenie nietechniczne, Szczecin (2010).
[16] Tsatsaronis G., Morosuk T., LNG–Based Cogeneration Systems: Evaluation Using Exergy-Based Analyses, InTech, Rijeka (2012).
[17] Molenda J., Gaz ziemny. Paliwo i surowiec, Wydawnictwa Naukowo-Techniczne, Warszawa (1996).
[18] Sharrat C., LNG terminal cold energy integration opportunities offered by contractors,
LNG Journal (marzec 2012)
[19] Romero Gómez M., Ferreiro Garcia R., Romero Gómez J., Carbia Carril J., Review of
thermal cycles exploiting the exergy of liquefied natural gas in the regasification process,
Renewable and Sustainable Energy Reviews 38 (2014)
150
[20] Franco A., Casarosa C., Thermodynamic and heat transfer analysis of LNG energy recovery for power production, 32nd UIT Heat Transfer Conference, Pisa (2014)
[21] Polskie LNG, Terminal LNG w Świnoujściu - szansa dla regionu, IX konferencja „Energetyka przygraniczna Polski i Niemiec – doświadczenia i perspektywy”, Sulechów (2012).
[22] Obowiązująca baza danych klimatycznych, strona internetowa Certyfikat-energetyczny
http://certyfikat-energetyczny.powiat.pl/CE_P/baza_danych_klimatycznych.html, dostęp
28 lutego 2016r.
[23] Szargut J., Egzergia. Poradnik obliczania i stosowania, Wydawnictwo Politechniki
Śląskiej, Gliwice (2007)
[24] Stanek W., Analiza termoekonomiczna w energetyce, wykład 4, Gliwice (2015)
Exploiting the cryogenic exergy of liquefied natural gas in
production of electricity
Tomasz Simla
Key words: LNG, regasification, cryogenic exergy, energy recovery, Świnoujście LNG
terminal
Abstract
Natural gas is a fossil fuel, the share of which in the global energy mix is growing the fastest.
Transportation of natural gas in liquefied form (LNG) is an alternative to traditional pipeline
transport. Poland joins the global LNG market through the receiving terminal which was built
in Świnoujście. The liquefaction process is very energy-consuming. Some energy utilised in
this process gets stored in LNG as cryogenic exergy. In a conventional regasification process
this exergy is destroyed by releasing to sea water or other fluid serving as an external heat
source. There are numerous ideas to recover the cryogenic exergy of LNG. Among possible
applications, the use of LNG to produce electricity by using it as a lower heat source in thermodynamic cycles or directly as a working fluid can be considered. In the present paper, an
analysis of four regasification systems was carried out: two systems without cold exergy recovery and two systems that produce electricity. Main input data to the analysis (mass flow,
pressure) correspond to real parameters of natural gas in the Świnoujście LNG receiving terminal. A simulation of operation of the systems for the whole year (with varying ambient
temperature) was performed. In order to compare the analysed systems, a number of coefficients, such as average fuel consumption, exergetic efficiency and coefficient of cumulative
energy consumption, was calculated.
151
c
Instytut
Techniki Cieplnej, Politechnika Ślaska
˛
Archiwum Instytutu Techniki Cieplnej Vol. 1 nr 1(2016)
www.itc.polsl.pl/ArchiwumITC
Numerical analysis of the orthotropic sample for conductivity
tests
Mieszko Tokarski1
e – mail: [email protected]
Key words: MATLAB, Ansys, FEM, Neumann's Boundary Condition, Factorization
Abstract
This work describes the development of a program based on the Finite Element Method for
the calculation of a temperature field in orthotropic sample with use of the Neumann's
Boundary Condition. Such a program has been created for the purposes of the project carrying
out in the Intitute of Thermal Technology in Gliwice, Poland. It is an important part of the
fully – automated algorithm for determining the sample's thermal conductivity by fitting
numerically obtained temperature field with its counterpart provided by the measurements.
Because of the specific nature of the measurement process as well as the main algorithm
itself, the developed program is characterized by high efficiency (comparable to Ansys),
sufficient accuracy and preparation for cooperation with the mentioned before fully –
automated algorithm. Most important features of the program are: module for geometry data
import (data is provided by the Ansys), module for the results export, the two control text files
for easy management by external procedures, logging and error reporting module.
1
The author have created the following chapter during the work on the master thesis carrying out in the Institute
of Thermal Technology in the Department of Energy and Environmental Engineering of the Silesian University
of Technology under the supervision of dr ing. Arkadiusz Ryfa. The work has been realised within the project
carrying out by the Intitute of Thermal Technology in Gliwice, Poland.
Zawartość tej publikacji może byc wykorzystana na warunkach licencji Uznanie Autorstwa 3.0 Polska. Licencja pozwala na
kopiowanie, zmienianie, rozprowadzanie, przedstawianie i wykonywanie utworu jedynie pod warunkiem oznaczenia autorstwa: autora,
tytułu rozdziału, nazwy serii, tomu, strony. Content from this work may be used under the terms of the Creative Commons Attribution
3.0 licence. Any further distribution of this work must maintain attribution to the author, chapter title, series title, volume, pages.
153
Nomenclature
Greek symbols
ξ, η, μ
Ρ
Δ
∂
Α
- natural coordinates
- density, kg/m3
- increase
- derivative
- thermal diffusivity, m2/s
Latinsymbols
N
C
k
𝑞
- shape function
- specific heat, J/kg K
- thermal conductivity, W/(m °C)
- heat flux, W/m2
Superscripts
E
N
T
1D, 2D, 3D
S
- element number
- Gaussian Point number
- transposition
- 1, 2 ,3 dimensions
- step number
Subscripts
i, j
- nodal number
Abbreviations
BC(s)
BDM
CDM
CMM
DBC
DLMM
FDM
FEM
GP(s)
NBC
PCG
TC
Notation
vectors and matrices are set in boldface
154
- Boundary Condition(s)
- Backward Difference Method
- Central Difference Method
- Consistent Mass Matrix
- Dirichlet's Boundary Condition
- Diagonally Lumped Mass Matrix
- Forward Difference Method
- Finite Element Method
- Gaussian Point(s)
- Neumann's Boundary Condition
- Preconditioned Conjugate Gradient
- Thermal Conductivity
1 Introduction.
1.1 Background and motivation.
Most of widely known and used contemporary methods of determination of thermal
conductivities (TC) are, in most cases, very inaccurate, time – consuming, complicated or
have destructive character. A precisely determined TC is crucial in many engineering issues,
for instance: assessment of heat losses and gains, definition of allowable thermal working
conditions of a component or the entire machine and evaluation of thermal strains and
stresses. In the case of insulating materials, the TC has decisive impact on the material's
quality. In general, in the literature are described many TC measuring methods with their
numerous variants, but certainly one can distinguish several major ones:





Guarded hot plate – a solid sample is placed between two plates of which one is heated
while the other is cooled (or heated in lesser extent). After reaching steady – state,
necessary measurements and calculations are performed [1].
Hot wire – a heated wire is inserted into the sample and then temperature change is
recorded. Density and heat capacity have to be known. At the end, plot of the wire's
temperature change versus logarithm of the time is used to calculate the TC [1].
Modified hot wire – the wire is supported on backing so it does not require sample's
penetration [1].
Laser flash diffusivity – sample's surface is heated with laser's pulse and infrared
camera records the temperature field [1].
Magnetic resonance imaging techniques – relation between temperature and magnetic
field has been used in order to determine the thermal diffusivity [2].
Guarded hot plate method requires a lot of time until steady – state is reached. Hot wire
method, in turn, because of sample's penetration, can be applied to fluids, foams and melted
plastics [1] but not for solids. While the Laser flash method (see also Parker flash method [3])
is fast, reliable but requires preparation of the samples that are damaged during the
measurements due to high temperature. It is important to notice that most of them allow for
determining TC in isotropic materials only. Hence, need of developing more reliable as well
as more accurate methods arises, particularly in orthotropic or anisotropic materials case. An
example of one of the promising and innovative methods is described in [2]. In brief, it
constitutes a development of the Parker flash method [3] and also consists of heating a sample
with use of a laser flash (Figure 2) and recording (with Infrared Camera) the temperature
distribution on a heated surface. In the next stage numerical analysis is applied comprising
modeling this phenomena with use of dedicated algorithm and comparing temperature fields
(obtained both by experiment and by computations). Figure 1 shows schematic block diagram
of the described procedure. It should be stressed that, in order to properly fit experimental
temperature field with its numerically obtained counterpart, TC must be controlled during the
analysis. Briefly, TC has to be guessed and experimental results are used as a benchmark.
155
Figure 1: Schematic block diagram of the algorithm for determining the thermal conductivity
in orthotropic material. The program's place in the procedure is marked with ellipse. The
image has been taken from [2].
Usage of a commercial software like Ansys Mechanical, Ansys Fluent or MSC.Patran is, at
least, troublesome. Mainly due to the specific requirements provided by the nature of the
analysis. For instance, some difficulties occurred during modeling of heat transfer based on
the Neumann's Boundary Condition (NBC) in Fluent. Most probably the Finite Volume
Method (FVM) Fluent is using, is unsuitable for such phenomenon or there is a necessity for
some specific settings of the analysis. For now, Fluent provides far higher temperatures than
Ansys Transient Thermal (using Finite Element Method FEM). Such a situation confirms the
fact that using the commercial software can be problematic due to its own specific features
and requirements regarding the, for example, mesh or analysis setup. Furthermore,
cooperation of this software with external procedures can provide a lot of trouble as well. It
should be stressed that, for the purposes of the project it is necessary to allow for efficient
cooperation between different programs belonging to the procedure described in the next
paragraph. In this regard, developing of a self – made FEM code may be much better solution
than use of the complicated commercial software. Besides that, the licenses of such software
are expensive so in a situation when one has choose between commercial and free program
with comparable performance and accuracy, the latter is better choice.
First of all, whole numerical procedure has to be fully automated. Experimental results are
loaded into analysis setup file, then main program calls for a subprogram responsible for the
computations which are carried out in order to get numerical temperature field. This
subprogram prepares specific output files formatted in a desired way that next subprogram
can compare specified temperatures and, if it is necessary, change the TCs and repeat whole
procedure until convergence is reached.
156
Figure 2: A sample holder and dot of the laser beam.
The subject of this work is to create a fully automated program using the Finite Element
Method to solving heat transfer problems with use of the NBC. Such program will be used in
a procedure of determination of the TCs within the project described in [2]. Besides that, it
will have several important advantages over the Ansys. At first, in contrast to the complex
Ansys and its highly developed interface, managing this program will be much simpler due to
ordinary text files which are easy to create and edit with use of user defined procedures
written, for instance, in MATLAB or in FORTRAN. Secondly, results provided by the
program can be easily formatted in a required manner so their further processing will not
cause any additional problems – it is easy to achieve with use of user defined procedures
designed directly for this task and for this particular kind of the program. Additionally it is
always better when there is possibility of customization both of cooperating programs than
only one of them like it would have been if the Ansys was used instead of program developed
as the subject of this work.
First task that had to be done, was to choose the programming environment which would
allow relatively quick, easy and efficient developing of desired program. The choice fell on
the MATLAB [4] because of several important factors:






a wide range of efficient and already implemented functions for managing data,
easy developing own procedures on the basis of the MATLAB's functions,
efficient computing of large matrices,
easy and user – friendly debugging module,
'Profiler' – tool facilitating the optimization of the code,
File Exchange – free service for sharing files between MATLAB's users [5].
157
The next stage was proper application of the Finite Element Method (briefly described in
Chapter 2.1). To be sure, the FEM written in MATLAB works properly, several test cases
have been solved. Each of them had its own contribution into final result being finished
program for the purposes of the project. It was a long journey through MATLAB
programming environment, FEM's features and linear algebra needed to find a way of
efficient and accurate solving systems of thousands equations.
As it was mentioned before, the program is for the purposes of the project realized at the
Institute of Thermal Technology in Gliwice, Poland. A core problem is not a development of
such program, but achievement of desired accuracy and efficiency. It is problematic due to the
number of phenomena accompanying the process of laser's radiation. These phenomena are:



radiation of the warmed sample' surface due to the laser beam,
conduction inside the sample's material (TC is unknown),
carbonization of the sample due to high temperatures.
Convection's impact is negligible. Radiation heat flow depends on a temperature to the fourth
power. That temperature rises to about several hundred Celsius degrees in less than 0.2 s.
Thermal conductivities are unknowns whereby they are additional complicating factor.
Moreover, due to high temperatures local carbonization of the sample occurs thus chemical
composition changes so material properties do as well. Besides that, sample's density, TC and
specific heat, all depend on the temperature in real case.
Another factor complicating the task is short time interval. Mentioned above phenomena are
very dynamic. Whole measurement process consists of three main stages:



heating with use of the laser flash – lasts for about 0.2 second,
removing the laser and replacing it with the Infrared (IR) camera in order to keep
precisely the same angle as during previous stage – lasts for about 0.3 s,
temperature field recording – lasts for about two second, time steps depend on
recording times of IR camera (frames per second to be exact) and on the speed of
saving those temperature fields on a hard drive.
At this stage, the program for numerical modeling of the temperature field is under the
validation. In addition to that, importing required data, file for managing whole program,
monitoring of the executed procedures (logs) as well as extracting results in desired form are
already implemented. However, loss of the heat due to radiation and temperature – dependent
material properties still require further investigation on their impact on the solution.
2 Governing equations.
In the FEM transient heat transfer equation [6] is solved numerically. For orthotropic bodies
and constant TC, specific heat and density such equation takes the form
∇𝐤∇T + q v = ρc
158
DT
Dt
(1)
T is the temperature, 𝐤 is thermal conductivity represented by the following vector
𝐤 = kx
ky
kz
(2)
q v is internal heat source (in this work q v = 0), ρ and cstand for the specific heat and density
respectively.
2.1 The Finite Element Method.
Today solving complex mathematical problems of physics and engineering is possible using
numerical methods only. Amongst these methods certainly worthy of distinction is the Finite
Element Method – a powerful and widely used tool in the field of mechanics and heat
transfer. To get more information about FEM, please refer to [7], [8], [9] and[11]. Figure 3
presents main idea of the FEM used in this work. First, main domain – a cylinder as an
example of a solid that can be easily presented and simultaneously has irregular shape – is
divided into sub domains (finite elements – that is the first approximation of the domain).
Then it is assumed that all finite elements are hexahedrons (parent elements – this is second
approximation of the domain). For each parent element Jacobian together with its
determinants is computed as well as stiffness matrix of the element Ke. Simultaneously mass
matrix of the element Me is determined. Next, these matrices are assembled into global K and
M respectively. Right Hand Side (RHS) vector F stores coefficients of the NBC determined
with GPs, nodal coordinates, shape functions for a 2D case and the same mesh. In the next
stage, time step and initial temperature are applied and equation BDM [(Backward Difference
Method) (16)] is solved. Each subsequent iteration has its own initial temperature from the
previous time step.
Figure 3: Schematic block diagram showing main idea of the FEM used in this work.
159
2.2 Shape functions.
Shape functions (SF) are polynomials and are used for interpolating continuous filed quantity
as well as to define shape of the parent element. In this work hexahedral 20-node elements
have been used, hence it follows the order of such functions together with their form. Such
functions belong to, so called, serendipity shape functions family [9]. Shape functions are
more precisely described in [7,9,10]. Figure 4 shows adopted manner of numbering of the
parent element's nodes.
Figure 4: Hexahedral element in global coordinate system with its node numbering manner.
Nodal numbers correspond to the proper shape functions. The choice of 20 – node element
(and those functions) is not accidental because of two important factors. Namely, shape
functions of the second order are sufficient enough to model temperature distribution
precisely, moreover use of serendipity elements allows to remove interior nodes of all
hexahedron's walls as well as one right in the middle. As a result significant reduction of
degrees of freedom occurs what leads to smaller computational effort – instead of 27 nodes
there are only 20 in a single element. In addition to that, such reduction has no serious impact
on accuracy [12].
2.3 Gaussian quadrature – integral approximation.
In numerical methods integrals are approximated instead of directly solved. There are many
methods to do such approximation, some of them are more efficient than others, for instance,
Gaussian quadrature. The idea is to pick such set of points (let say Gaussian Points - GP) that
after the insertion of them into approximating integral, an accurate solution is obtained.
Each GP has its own wage. The wages define shares of approximated domain. In 3
dimensions (3D) it is a volume, in 2 dimensions (2D) it is a share of surface whereas in 1D
case it is a section. In 3D case GP's coordinates has three components (x, y, z or ξ, η, μ),
hence volume around the GP consists of a combination of three wages γx, γy, γz.
Figure 5 shows 1D example of the Gaussian quadratures. Despite the fact that this work
concerns 3D domain, one dimensional case is far more readable. Each GP (dots) approximates
160
one piece of the bar. Each corresponding wage describes length of a section that is
approximated by a given GP.
Figure 5: Gaussian quadrature of the third order in 1D case.
For more specific information about Gaussian quadrature, please, refer to [8,9,13].
2.4 The Jacobian – a link between global and natural coordinate systems.
The Jacobian [J] is a measure of the distortion of the given parent element defined in local
coordinates in comparison to its counterpart in global coordinates [14]. Whereas determinant
of the Jacobian, det[J], is numerically equal to the length of a section of its counterpart in 1D,
surface in 2D case and equal to its volume in 3D case. The Jacobian is necessary to keep a
link between sub - domains and their global counterparts.
Figure 6 shows simple 2D case while the exemplary element (gray) is transformed into
quadrilateral parent element. Using the Jacobian allows to determine surface A of the given
element
𝐴 𝑥, 𝑦 = det 𝑱 𝜕𝜉𝜗𝜂
(3)
Figure 6: Finite element (gray) in global coordinate system and its counterpart in local.
If the element had been, for example, rectangle with the same length in x direction as its
𝜕𝑥 𝜕𝑦
parent element, the two main (diagonal) Jacobian coefficients 𝜕𝜉 , 𝜕𝜂 would be non – zeros.
161
𝜕𝑥
𝜕𝜉
𝜕𝑦
[𝑱] =
𝜕𝜉
𝜕𝑧
𝜕𝜉
𝜕𝑥
𝜕𝜂
𝜕𝑦
𝜕𝜂
𝜕𝑧
𝜕𝜂
𝜕𝑥
𝜕𝜇
𝜕𝑦
𝜕𝜇
𝜕𝑧
𝜕𝜇
(4)
𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑥 𝜕𝑦 𝜕𝑧
where 𝐉 is the Jacobian matrix and
,
,
,
,
,
,
,
,
𝜕𝜉 𝜕𝜉 𝜕𝜉 𝜕𝜂 𝜕𝜂 𝜕𝜂 𝜕𝜇 𝜕𝜇 𝜕𝜇
are the distortions in a
specific directions. Calculating such derivatives carries out in the following manner
𝜕𝑥
=
𝜕𝜉
𝜕𝑦
=
𝜕𝜉
𝜕𝑧
=
𝜕𝜉
where
𝜕𝑁𝑖𝑒
𝜕𝜉
𝑁𝑜𝐸𝑁
𝑖=1
𝑁𝑜𝐸𝑁
𝑖=1
𝑁𝑜𝐸𝑁
𝑖=1
𝜕𝑁𝑖𝑒
𝑥
𝜕𝜉 𝑖
𝜕𝑁𝑖𝑒
𝜕𝜉
𝑦𝑖
𝜕𝑁𝑖𝑒
𝑧
𝜕𝜉 𝑖
𝜕𝑥
=
𝜕𝜂
𝜕𝑦
=
𝜕𝜂
𝜕𝑧
=
𝜕𝜂
𝑁𝑜𝐸𝑁
𝑖=1
𝑁𝑜𝐸𝑁
𝑖=1
𝑁𝑜𝐸𝑁
𝑖=1
𝜕𝑁𝑖𝑒
𝑥
𝜕𝜂 𝑖
𝜕𝑥
=
𝜕𝜇
𝜕𝑁𝑖𝑒
𝜕𝑦
=
𝜕𝜇
𝜕𝜂
𝑦𝑖
𝜕𝑁𝑖𝑒
𝑧
𝜕𝜂 𝑖
𝜕𝑧
=
𝜕𝜇
𝑁𝑜𝐸𝑁
𝑖 =1
𝑁𝑜𝐸𝑁
𝑖=1
𝑁𝑜𝐸𝑁
𝑖 =1
𝜕𝑁𝑖𝑒
𝑥
𝜕𝜇 𝑖
𝜕𝑁𝑖𝑒
𝑦
𝜕𝜇 𝑖
(5)
𝜕𝑁𝑖𝑒
𝑧
𝜕𝜇 𝑖
is an 𝑒 − 𝑡ℎ element's shape function's derivative, 𝑥𝑖 , 𝑦𝑖 , 𝑧𝑖 are proper components
of global coordinate of 𝑖 − 𝑡ℎ nodal point.
In the method adopted in this project Gaussian integration was used to estimate triple integral
limiting parent element's domain. Determination of such matrix is quite simple and proceeds
as follows
𝜕𝑥 𝜕𝑥 𝜕𝑥 𝑛,𝑒
𝜕𝜉 𝜕𝜂 𝜕𝜇
𝑥1𝑒 𝑦1𝑒 𝑧1𝑒
𝜕𝜉
𝜕𝜂
𝜕𝜇
𝜕𝑦 𝜕𝑦 𝜕𝑦
(6)
⋮
⋮ =
× ⋮
⋮
⋮
⋮
𝜕𝜉
𝜕𝜂
𝜕𝜇
𝑛
𝑛
𝑛
𝑒
𝑒
𝑒
𝜕𝑁20
𝜕𝑁20
𝜕𝑁20
𝑥20 𝑦20 𝑧20
𝜕𝑧 𝜕𝑧 𝜕𝑧
𝜕𝜉
𝜕𝜂
𝜕𝜇
𝜕𝜉 𝜕𝜂 𝜕𝜇
where first matrix contains set of all nodes' derivatives calculated with 𝑛 − 𝑡ℎ Gaussian point,
middle one contains global coordinates of 𝑒 − 𝑡ℎ element and last one is the component
Jacobian matrix. In order to determine 𝐉 𝑒 a simple summation of all components[𝑱]𝑛 ,𝑒 must
be carried out hence
𝜕𝑁1𝑛
𝑇
𝜕𝑁1𝑛
𝜕𝑁1𝑛
𝑁𝑜𝐺𝑃
𝑒
[𝑱] =
[𝑱]
𝑛 =1
162
𝑁𝑜𝐺𝑃
𝑛,𝑒
=
𝑛 =1
𝜕𝑥
𝜕𝜉
𝜕𝑦
𝜕𝜉
𝜕𝑧
𝜕𝜉
𝜕𝑥
𝜕𝜂
𝜕𝑦
𝜕𝜂
𝜕𝑧
𝜕𝜂
𝜕𝑥
𝜕𝜇
𝜕𝑦
𝜕𝜇
𝜕𝑧
𝜕𝜇
𝑛,𝑒
(7)
where 𝑁𝑜𝐺𝑃 is number of the Gaussian points (in this work 𝑁𝑜𝐺𝑃 = 27).
Determinant of the Jacobian matrix of the element 𝑒 and 𝑛 − 𝑡ℎ GP is calculated as follows
𝑁𝑜𝐺𝑃
𝑒
det⁡
[𝑱]𝑛,𝑒
det[𝑱] =
(8)
𝑛 =1
where det⁡[𝑱]𝑛,𝑒 is computed with use of the Sarrus' rule [15].
2.5 Element's Stiffness Matrix [K]e.
Element's Stiffness Matrix [K]e (or conductance matrix) contains a set of coefficients
indicating mutual relationship between all nodes in the considering sub-domain. Such a
matrix is diagonal having a size of Number of the Element's Nodes x Number of the Element's
Nodes (NoEN x NoEN). In case of 20 – node hexahedron such a matrix is 20 x 20.
Determination of the matrix [K]e is as follows
𝑁𝑜𝐺𝑃
𝒆
[𝑲] =
det 𝑱
𝑛=1
𝑛,𝑒
+ 𝑘𝑦
𝜕𝑁
𝜕𝑦
𝜕𝑁
𝜕𝑦
𝑛 ,𝑒
(𝑘𝑥
𝜕𝑁
𝜕𝑥
𝑛,𝑒,𝑇
+𝑘𝑧
𝑛,𝑒
𝜕𝑁
𝜕𝑧
𝜕𝑁
𝜕𝑥
𝑛,𝑒
𝑛,𝑒,𝑇
𝜕𝑁
𝜕𝑧
(9)
𝑛,𝑒,𝑇
)
and
𝜕𝑁
𝜕𝑥
where
𝜕𝑁
𝜕𝑥
,
𝜕𝑁
𝜕𝑦
,
𝑛,𝑒
𝜕𝑁
𝜕𝑧
𝜕𝑁1𝑛,𝑒
𝜕𝑥
=
,
⋮
𝑛,𝑒
𝜕𝑁20
𝜕𝑥
𝜕𝑁
𝜕𝑦
𝑛,𝑒
𝜕𝑁1𝑛 ,𝑒
𝜕𝑦
=
,
⋮
𝑛 ,𝑒
𝜕𝑁20
𝜕𝑦
𝜕𝑁
𝜕𝑧
𝑛,𝑒
𝜕𝑁1𝑛,𝑒
𝜕𝑧
=
⋮
𝑛,𝑒
𝜕𝑁20
𝜕𝑧
(10)
are arrays of derivatives transformed to global coordinate system. In
order to gather more specific information about such transformation, please refer to chapter 4
p. 154-157 of [8], chapter 2 p. 42-44 of [7] and chapter 2 p. 58-61 of [11]. Moreover, [K]e is
full.
2.6 Element's Mass Matrix [M]e.
Element's Mass Matrix [M]e (or capacitance matrix) contains approximated masses in nearest
surroundings of all nodes of the element after multiplication by density. Otherwise sum of all
values is equal to det[J]. In other words, each node is approximation of a part of the parent
element it belongs to. In 3D case, such node represents some volume of a solid which has
some density (mass) and specific heat. Such a matrix has the same dimensions as [K]e and its
computation is as follows [11]
163
𝑁𝑜𝐺𝑃
𝒆
[𝑴] =
det 𝑱
𝑛 ,𝑒
𝑛,𝑒
𝑵
𝑵
𝑛,𝑒,𝑇
(11)
𝑛 =1
where
𝑵
𝑛,𝑒
𝑁1𝑒
= ⋮
𝑒
𝑁20
𝑛
(12)
It should be stressed that multiplication by density and specific heat occurs after global mass
matrix assemblage. Moreover, similarly to [K]e, [M]e is full.
2.7 Assembly of a global stiffness [K] and mass [M] matrices.
Assemblage of such matrices consists of a rewriting of all local[K]e and [M]e into one global
matrix [K]e and [M]e respectively. Size of the [K] and [M] depends on a total number of
nodes and is equal to 𝑇𝑁𝑜𝑁 × 𝑇𝑁𝑜𝑁. It would be convenient to discuss assemblage on the
example. Element's stiffness matrix
𝐾𝑒 =
𝐾1,1
⋮
𝐾𝑁𝑜𝐸𝑁,1
⋯
𝐾𝑖,𝑗
⋯
𝐾1,𝑁𝑜𝐸𝑁
⋮
(13)
𝐾𝑁𝑜𝐸𝑁 ,𝑁𝑜𝐸𝑁
Provides us with pattern of local nodal indexing, hence
𝑖𝑖 = 1: 1: 𝑁𝑜𝐸𝑁 1: 1: 𝑁𝑜𝐸𝑁 ⋯ 1: 1: 𝑁𝑜𝐸𝑁
𝑗𝑗 = [ 1,1, … , 𝑁𝑜𝐸𝑁 2,2, … , 𝑁𝑜𝐸𝑁 ⋯ 𝑁𝑜𝐸𝑁, 𝑁𝑜𝐸𝑁, … 𝑁𝑜𝐸𝑁 ]
(14)
where𝑖𝑖 and 𝑗𝑗 are local i – indices and j- indices respectively. For the first element (e = 1)
global indices equals local ones, but for any other element do not. Because of matrix
dimensions 20x20, global indices increase by 400 for each subsequent element.
At this point it suffices to state that, method adopted in this work is fast and effective. It
should be noted that [K] and [M] are sparse, which means a lot of their elements are equal to
zero. MATLAB's function sparse() allows to store only non – zero elements in the matrix.
Such solution saves a lot of memory. Example of stiffness and mass matrices are shown in
Figure 7. To get more specific information about stiffness and mass matrices assembly, please
refer to [7] or [11].
164
Figure 7: Stiffness matrix [K] and Consistent Mass Matrix [M] (left hand side) and
Diagonally Lumped Mass Matrix (right hand side). Black dots denote non – zero values.
Generated with MATLAB's spy() function.
It is important to state that, those matrices (left hand side of the Figure 7) are highly
decomposed. In order to save computational effort one should strive for narrowing all off –
diagonal elements and concentrate them along main diagonal by optimization of nodal
numbering. Mass matrix with identical non – zero elements' distribution as well as the same
dimensions as [K] is known in the literature as Consistent Mass Matrix (CMM). Besides
CMM, in the literature known is also Diagonally Lumped Mass Matrix (DLMM) being
approximated version of CMM. Such a matrix has significantly reduced number of elements
what leads to faster computations but, due to additional approximation, lowers the accuracy.
Generally DLMMs are used in cases of very large geometries with millions of nodes when
computation time is more important than the accuracy [16].
Such approximation consists of a reduction of all off – diagonal elements to one lying in the
main diagonal, row – wisely. However, there are some requirements that have to be met. For
instance, in case of serendipity element, rows corresponding to their corner nodes result in
negative masses after summation of all elements which belong to them. In such a case,
conservation condition is not satisfied so use of different method of lumping may be
necessary. In the literature one can distinguish three main ways of mass matrix lumping:



summation of all off – diagonal elements into one lying on the main diagonal,
scaling diagonal elements by proper factor – in this method det[J] is divided by sum of
all diagonal elements and next is multiplied by each of them separately,
Lobatto's method – is very similar to Gaussian quadrature.
Each of mentioned above methods, has its own pros and cons. In all of them lumping is
applied to local (element's) mass matrices which are assembled later. In order to get more
details, please refer to [16].
165
2.8 Neumann's Boundary Condition.
In the work at hand boundary condition of the second kind is used to model the laser flash. In
case of a 3D geometry, laser flash (Heat Flux) can be prescribed to the surface only (see
Figure 8).
Figure 8: Example of applying HF and new 2D nodal numbering (grey).
As a result appears a necessity of determination of the Jacobian 2D as well as use of proper
shape functions together with Gaussian Points. Determination of the [𝑱]𝟐𝑫 carries out in the
same way as the [𝑱] for 3D case. More information about 2D case can be found in [7], [8], [9]
and [11].
Such surface has 8 nodes and, in case of 3rd order quadrature, 9 GPs. Heat flux coefficients
stored in RHS vector F are determined as follows
9
𝑭2𝐷
𝑖 =
𝑁𝑖 2𝐷 (𝑛) det 𝐽
2D
(𝑛)
(15)
𝑛 =1
where𝑖 = 1: 8 is nodal number, 𝑛 = 1: 9 is GP number and 𝑭𝟐𝑫 denotes temporary array with
the BC coefficients. Next, all of these coefficients are multiplied by desired heat flux.For
single element, there is 8 coefficients determined in this way that are (with use of the
connectivity matrix) inserted into proper rows of the RHS array F (according to their global
order).
2.9 Backward Difference Method.
BDM assumes that, nodal temperature in step s+1 depends on all other nodal temperatures
from step s. In this work BDM has been chosen because of three main reasons:



166
as an implicit scheme, is always convergent,
there is no 'inversion' of [M] ([6], [11]),
is widely used by commercial software.
First point refers to the time step ∆t which has no impact on convergence of the solution [11].
The second one means that, in case of dense mesh, lumping of CMM will not be necessary to
keep reasonable calculation time because there is no mass matrix 'inversion' in Eq.(16).
Another advantage of this method is simplicity, because implementation of the BDM does not
require much effort in comparison to Central Difference Method (CDM) or to Forward
Difference Method (FDM) [11]. System of equations can be written as
𝑻𝑠+1 =
(𝑴𝑻𝑠 + ∆𝑡𝑭)
𝑴 + ∆𝑡𝑲
(16)
where 𝑻𝑠+1 is vector of nodal temperatures in step s+1, 𝑻𝑠 is vector of nodal temperatures in
previous step s.
It is important to notice that in the denominator of Eq.(16) is [M] together with [K] which has
to be 'inversed' with no exception. [M] in the nominator, in this particular scheme, does not
have to be inversed. Eq.(16) is recomputed each time when time step Δ𝑡 changes and can be
rewritten
𝑻𝑠+1 = 𝑴 + ∆𝑡𝑲 \(𝑴𝑻𝒔 + ∆𝑡𝑭)
(17)
where \ is MATLAB's left division. Behind \ is hidden advanced equation solver using a wide
range of numerical methods [17], but has one essential drawback. Namely, solving many
systems of equations like 𝑨𝒙 = 𝒃 and 𝑨𝒚 = 𝒄 requires factorization each time, when \
command is executed. There is no possibility of storing factorized matrix in order to reuse it
in another system of equations. Matrix A inversion could be carried out once and then reused
but such solution should be excluded at the very beginning because of three main reasons:



inversion process is highly inaccurate – a lot of elementary operations are burdened
with numerical errors,
such process requires a lot of time,
in case of sparse matrices (huge systems of equations), so called, fill – in phenomenon
occurs, what leads directly to full matrix instead of sparse and to a lot of memory
requested in order to store such matrix.
There are two ways of solving huge systems of equations of type 𝑨𝒙 = 𝒃 and 𝑨𝒚 = 𝒄:


direct methods - fast and accurate but requiring a lot of memory,
o factorization (Cholesky, LU, LDLT, QR [20], Gaussian Elimination),
iterative methods - not so fast and not so accurate but saving a lot of memory.
o Preconditioned Conjugate Gradient (PCG),
o Jacobi Conjugate Gradient (JCG),
o Cholesky Conjugate Gradient (CCG) [18].
Geometry and mesh, in this work and in the project [2] as well, are rather small and simple so
use of factorization is possible. Because of the decomposition of matrices [K] and [M]
Gaussian Elimination is inefficient, but works goods for diagonal matrices with narrowly
167
distributed off – diagonal entries. Besides that, PCG has been tested also (because of the best
performance described in [18]) and it turned out it is comparable to factorization in MATLAB
2011a in contrast to 2009b.
Through Mathworks' File Exchange [5],Timothy A. Davis of the University of Florida has
released his own factorization algorithm [19] in 2009 that has been permanently implemented
to MATLAB 2014b as toolbox. This algorithm has been used, in this work, to solve Eq.(17).
2.10
Errors.
All errors presented in this work are calculated as follows
2
𝐸𝑟𝑟𝑖 =
2
𝑇𝑖,𝐷𝐿𝑀𝑀
−
2
𝑇𝑖,𝐶𝑀𝑀
∙ 100%
(18)
2
𝑇𝑖,𝐶𝑀𝑀
where 𝑇 is nodal temperature, 𝑖 denotes node number, 𝐷𝐿𝑀𝑀 and 𝐶𝑀𝑀 are Diagonally
Lumped Mass Matrix and Consistent Mass Matrix respectively.
In the case of mixed BC described in Chapter 3.1.2 (steady – state), nodal temperatures
denoted with use of DLMM and CMM subscripts in the Eq.(18) are replaced by nodal
temperatures obtained with 2nd and 3rd order quadratures respectively using Eq.(19).
3 Numerical study – benchmark.
Before the program has been written, two benchmarks had been solved. First one in 2D,
second one in 3D to make sure the FEM works properly. It was very important to implement
new features step by step because of the complexity of the Finite Element Method so finding
potential mistake could be extremely difficult. At first, learn about integration between local
and global systems of coordinates was necessary (Gaussian quadrature, Jacobian). After that
proper assemblage of global stiffness matrix had to be done.
3.1 2D benchmark.
Pattern example has been generated with PATRAN and consisted of two cases of steady –
state heat transfer:


pure Dirichlet's Boundary Condition (DBC),
mix of Dirichlet's and Neumann's BCs.
This benchmark was carried out in order to learn how to apply BCs properly. Applying of
these BCs is discussed in chapters 3.1.1 and 3.1.2 using the specific examples for the
educational purposes. Also implementation of the BCs is described more specifically when
discussing the two following examples, mainly due to greater readability.
The geometry from PATRAN (Figure 9) was a simple plate composed of 27 elements and 106
nodes of length 0.5 m and width 0.2 m. All elements were 8 – nodal hexahedrons and were
equal to each other. Nodal coordinates as well as connectivity matrix were exported from
168
PATRAN to a text file. It was, and it meant to be, a simple test and its purpose was to validate
the developed program.
Figure 9: The geometry generated with PATRAN.
PATRAN's nodal numbering were adopted so all nodes correspond to each other. Another
thing worth of mention is that, to approximate double integral Gaussian quadrature of the
second order was used. As shown later research, third order quadrature met the requirements
concerning the accuracy. To solve steady – state problem, the following equation is used
(19)
𝑻 = 𝑲\𝑭
𝑊
Results in the following subsections have been obtained for thermal conductivity 𝑘 = 25 𝑚 °𝐶 .
3.1.1 Dirichet's Boundary Condition.
Boundary condition of the first kind, also known as Dirichlet's BC, has been applied in the
way described in [8] on pages 103 – 104 with use of, so called, penalty method.
To the left hand side of the elements 1, 10 and 19 (see Figure 9) 50˚C has been prescribed, to
the right hand side of the elements 9, 18 and 27, 150˚C. Figure 10 proves temperature
distribution in both cases is the same and isotherms are arranged into symmetrical, vertical
strips of equal width. Gaussian order of the second order was used.
Figure 10: Temperature distribution from:
MATLAB (left hand side), PATRAN (right hand side).
169
It should be stressed, that presenting the results provided by different programs, in a similar
manner, is problematic. For instance, in MATLAB it is easy to generate such figure (Figure
10) in grayscale in contrast to PATRAN, where it is much more complicated. During the
analysis has been discovered that, in this particular case of pure DBC, results do not depend
on values of coefficients in a stiffness matrix as well as on Gaussian quadrature's order. But
they do depend on sign – negative or positive and to be exact, on distribution of negative and
positive coefficients in [K]. Nevertheless, investigation of the nodal temperatures (chosen
ones has been gathered in Table 1) has confirmed the fact that, the results are correct.
Table 1: Comparison of chosen nodal temperatures (second order quadrature).
Self-made code in MatLAB
Patran
Node:
Nodal temperature, ˚C:
106
149.9999
150.0000
95
88.8889
88.8888
80
72.2222
72.2221
56
127.7777
127.7780
14
122.2221
122.2220
1
50.0000
50.0000
As one can notice, the results are nearly the same with some negligible differences after third
decimal place. Obviously, the code, in this particular case, works properly.
3.1.2 Mixed Dirichlet's and Neumann's Boundary Condition.
Next thing that had to be done, was implementation of the BC of the second kind (imitating
the heat flux), crucial to the next stages of this work. In this case DBC remained unchanged
𝑊
and NBC has been added. Normal heat flux (𝑞𝑠 = 40000 𝑚 2 ) has been prescribed to northern
edge of the 24 – th element (see Figure 9).
From the point of view of the heat flux (normal to the boundary), the edge to which it is
assigned is a single line described by nodes 98, 99 and 100. Single line means one
dimensional case. For that reason an approximation of the edge in local coordinates system
has been done. That requires different shape functions, new Jacobian and new set of the
Gaussian points. Further implementation of the NBC (RHS coefficients) is carrying out
according to Eq.(15).
Temperature fields obtained from MATLAB and PATRAN are presented in Figure 11. On the
left hand side of the plate DBC (50 °C) is prescribed that keeps applied there temperature
(cools the plate), second DBC (on the right side) holds there 150 °C. Applied heat flux to the
northern edge of the 24 – th element is sufficient to locally heat the plate to temperature over
200 °C. The isotherms are unsymmetrical due to two different (but symmetrically oriented)
DBCs.
170
Figure 11: Temperature distribution obtained from: MATLAB (left hand side) and PATRAN
(right hand side). Mixed BC.
Table 2 shows chosen nodal temperatures of the considering case obtained with 2nd and 3rd
Gaussian quadrature's order. Nodes to which the NBC is applied are set in bold.
Table 2. Chosen nodal temperatures obtained with MATLAB and PATRAN.
Self – made code in MATLAB
Quadrature's
order
Node:
106
100
99
98
95
80
56
14
1
2
149.9999
200.8242
212.5521
192.5137
127.5782
92.6374
156.7796
150.0737
50.0000
PATRAN
3
Nodal temperature, ˚C:
150.0000
200.7404
211.8979
192.4287
127.6529
92.6137
156.7883
150.0577
50.0001
3
150.0000
200.7380
211.8959
192.4279
127.6530
92.6132
156.7879
150.0570
50.0000
It is easy to notice that, second order quadrature is not accurate enough in case of mixed BC
or pure NBC, but in case of pure DBC is sufficient. Moreover, the closer to the nodes that
NBC is prescribed to (98, 99, 100), the higher differences between the nodal temperatures are.
All errors were calculated with use of Eq.(18) and take the maximum values directly at NBC
– nodes and are negligible after exceeding a certain distance (from the NBC), see Figure 12.
It is clear that error's distribution concentrates in region of NBC and spreads in all available
directions. Despite the fact that, applied DBCs are at varying distances from the NBC, error
distribution is almost perfectly symmetrical. At the nodes where DBC is prescribed, those
differences are negligible (see Table 2, nodes 1 and 106).
171
Figure 12: Graphical interpretation of the errors' distribution due to applying the NBC with
use of the 2nd order Gaussian quadrature in comparison to the 3rd order quadrature.
Obviously, 2nd order quadrature does not provide satisfying accuracy in contrast to the 3rd
order quadrature that simultaneously involves slightly higher computational effort (instead of
4 GP points there are 9 for 2D case). Because of the accuracy, 3rd order quadrature has been
permanently implemented in the program.
3.2 3D benchmark.
At this phase, extending 2D FEM to 3D was necessary. Main difficulty was finding proper
formulas. In the literature 2D cases are described in detail, but in most cases when it was
coming to 3D, authors just limited themselves to a perfunctory statement, that all formulas are
similar to 2D. As it turned out later, they were right. Nevertheless, stiffness matrix assemblage
provided a lot of difficulties. At this phase, again, development of the code proceeded step by
step in order to evade mistakes that could be hard to find.
Analysis setup and material properties:
𝑊

Thermal conductivity 𝑘 = 25 𝑚 °𝐶 .

Density 𝜌 = 1091

Specific heat 𝑐 = 900 𝑘𝑔 °𝐶 .



𝑘𝑔
𝑚3
.
𝐽
Time step Δ𝑡 = 20 𝑠.
Time of analysis (total time)𝑡𝑡 = 2500 𝑠.
𝑊
Heat flux 𝑞𝑠 = 40000 𝑚 2 .
Applying the BCs has been carried out in the same way as for 2D benchmark in the
MATLAB. The geometry has been created in PATRAN together with the BCs and imported
to the MATLAB. A cuboid, presented in Figure 13, consists of 60 hexahedral elements and
406 nodes.
172
Figure 13: 3D geometry prepared with PATRAN and prescribed all boundary conditions.
Heat flux were applied later, in the next analyzes.
3.2.1 Dirichlet's Boundary Condition, steady – state.
In this case DBC was ascribed at three regions:



lower X – Y plane: 50˚C (first),
upper X – Y plane: 150˚C (second),
one element in X – Z plane: 250˚C (third).
Three different DBCs are prescribed in order to eliminate special case described in chapter
3.1.1concerning pure DBC in 2D, where values of coefficients in stiffness matrix did not have
any impact on the solution. Third DBC prevents such situation. Figure 14 shows temperature
distributions obtained from PATRAN and MATLAB. Unfortunately, rotating a 3D figure in
MATLAB is limited so setting similar view to PATRAN's figure is not possible. Third DBC
(250˚C) makes isotherms irregular and much more concentrated on the wall it is prescribed to,
especially between the first DBC and the third.
173
Figure 14: Temperature distribution obtained from: MATLAB (left hand side), PATRAN
(right hand side). Steady – state, triple DBC.
Table 3 presents chosen nodal results from MATLAB and PATRAN. As one can see, 3 rd
order quadrature applied to MATLAB's code is sufficient and allows for getting accurate
temperatures in comparison to PATRAN.
Table 3: Results from MATLAB (3rd order quadrature) and PATRAN.
Self-made code in MATLAB
PATRAN
Node:
Nodal temperature, ˚C:
1
49.9999
50.0000
50
50.0000
50.0000
100
92.9081
92.9082
150
166.0193
166.0200
200
168.7079
168.7079
250
138.8943
138.8939
264
135.7371
135.7370
300
159.2606
159.2610
350
149.0673
149.0670
400
150.0000
150.0000
406
150.0000
150.0000
This benchmark proved that, the stiffness matrix [K]in 3D case has been assembled correctly.
Otherwise, corresponding nodal temperatures would not be consistent. The differences
between them are negligible.
174
3.2.2 Mixed Boundary Condition, steady - state.
In this case, the correctness of applying mixed BC has been verified. To triple DBC described
in chapter 3.2.1, NBC has been added for the following elements: 34, 35, 46, 47 (presented on
Figure 13). Figure 15 presents graphical interpretation of the obtained temperature fields
provided by MATLAB and PATRAN. Applied heat flux, is sufficient for increasing the
maximum temperature to higher value than third (the highest) DBC.
a)
b)
c)
d)
Figure 15: Temperature distribution with triple DBC and single NBC (mixed BC). a) NBC
side, MATLAB, b) NBC side, PATRAN, c) 3rd DBC side, MATLAB, d) 3rd DBC side,
PATRAN.
175
In Figure 15 c) and d), one can notice rectangular shape slightly brighter than its nearest
surroundings – it is third DBC (250 °C) applied to the four chosen elements' surfaces. To the
lower base (lower X – Y plane) first DBC is prescribed, what is confirmed by dark shade of
gray. Similarly upper base (upper X – Y plane) to which second DBC is prescribed and which
shade corresponds to 150 °C. Big bright stains in Figure 15 a) and b) present applied heat flux
and are located near the same edge. Except for the number of isotherms, figures from the
MATLAB look similarly to those from the PATRAN. Chosen nodal temperatures are
presented in Table 4 and are nearly the same.
Table 4: Results from MATLAB (3rd order quadrature) and PATRAN. Mixed BC.
Self-made code in MATLAB
PATRAN
Node:
Nodal temperature, ˚C:
1
49.9999
50.0000
50
50.0000
50.0000
100
125.7416
125.7409
150
192.7073
192.7070
200
206.0997
206.0989
250
266.3079
266.3060
264
269.3892
269.3880
300
187.2779
187.2780
350
188.7256
188.7250
400
149.9999
150.0000
406
149.9999
150.0000
This benchmark proved that 3rd order quadrature is sufficient to obtain results that are
comparable to PATRAN's. In addition to that, stiffness matrix assembly (3D) has been carried
out properly as well as applying of the NBC (2D – surface).
3.2.3 Mixed Boundary Condition, transient, CMM and DLMM.
The next step was an implementation of the BDM – transient heat transfer equation (Eq.(16))
in place of steady – state equation (Eq.(19)). Second and third DBCs (150˚C and 250˚C) have
been removed. First DBC (50˚C) remained and simultaneously this temperature was an initial
value also. NBC remained unchanged as well as density, TC and all others. The following
figure (Figure 16) is a set of two figures from MATLAB and two from PATRAN. It presents
temperature distribution at 20 and 2500 second of heating a cuboid with heat flux. The
temperature increases gradually around the area to which NBC is applied. Because of the first
DBC that holds 50˚C on the basis of the presented cuboid, upper part of the solid heats up to a
much greater extent. Back side, however, still remains 'cold' but it is only a matter of time
before the temperature become more uniform. It is noteworthy that steady – state has not been
achieved.
176
a)
b)
c)
d)
Figure 16:Temperature distribution from MATLAB (left hand side) and PATRAN (right
hand side): a) and b) after 20 s; c) and d) after 2500 s.
As one can see, PATRAN's shade interpolation looks smoother. Nevertheless, the temperature
distribution in both cases is similar. In
Table 5 are presented results from MATLAB and PATRAN. Setup of the analysis was the
same in both cases. Node to which the NBC is applied is set in bold. Calculations have been
carried out with CMM and with use of factorization algorithm written by Timothy Davis [20].
177
Table 5: Results from MATLAB and PATRAN. Single DBC and NBC. CMM.
MATLAB
Time, s:
Node:
1
50
100
150
200
250
264
300
350
400
406
20
50.0000
50.0000
50.1069
49.9937
49.9918
51.9898
82.2932
50.0164
51.0058
49.3207
48.6141
1500
2500
Temperature, ˚C:
50.0000
50.0000
50.0000
50.0000
61.9589
72.1512
57.41512
71.0562
63.5700
80.9149
133.0161
161.7480
288.2891
319.2235
63.5048
84.1807
103.2670
134.5854
95.0306
128.3418
109.3724
143.0194
PATRAN
20
50.0000
50.0000
50.1068
49.9937
49.9918
51.9896
82.2923
50.0164
51.0057
49.3207
48.6141
1500
2500
Temperature, ˚C:
50.0000
50.0000
50.0000
50.0000
61.9588
72.1509
57.4151
71.0560
63.5698
80.9146
133.0149
161.7469
288.2869
319.2210
63.5046
84.1801
103.2669
134.5850
95.0303
128.3410
109.3720
143.0180
Nodal temperatures in both cases are nearly the same. Because of the negligible impact of the
mass matrix lumping on a calculation time it has not been taken into account at this stage.
However, in this particular case lumping has serious impact on results. Table 6 shows nodal
temperatures after mass matrix lumping involved.
Table 6: Results from MATLAB and PATRAN. Single DBC and NBC. DLMM.
MATLAB
PATRAN
Time, s:
20
1500
2500
20
1500
2500
Node:
Temperature, ˚C:
Temperature, ˚C:
1
50.0000
50.0000
49.9999
50.0000
50.0000
50.0000
50
50.0000
50.0000
50.0000
50.0000
50.0000
50.0000
100
50.0047
61.1171
71.2368
50.0047
61.1170
71.2367
150
50.0000
57.6899
70.8089
50.0000
57.6898
70.8087
200
50.0033
64.0506
81.0321
50.0033
64.0503
81.0317
250
50.0082
130.0064
159.7537
50.0082
130.0050
159.7519
264
51.3226
282.8759
316.1018
51.3230
282.8739
316.0989
300
50.0141
62.9922
82.888695 50.0141
62.9921
82.8883
350
50.2543
102.7736
134.0306
50.2542
102.7730
134.0299
400
49.8913
91.7538
125.2727
49.8913
91.7536
125.2720
406
50.8818
105.5617
139.8922
50.8816
105.5609
139.8910
It should be noticed that lumping introduces some errors. In case of early iterations such error
reaches even 37 % (Table 7) and decreases over time. The temperature at node 264 in the first
178
iteration (20 s) with CMM equals to about 82.3˚C while the temperature at the same node and
iteration, but with DLMM, is equal to about 51.3˚C. This gives the difference of 30˚C (37 %).
Research project (described in chapter 1.1) requires as much accuracy as possible and,
unfortunately, lumping noticeably decreases program's reliability simultaneously without
providing any benefits (for instance, shorter calculation time).
Table 7: Errors generated by mass matrix lumping.
Node:
1
50
100
150
200
250
264
300
350
400
406
0
0
0.20
0.01
0.02
3.81
37.63
0.00
1.47
1.16
4.66
MATLAB
Error, %:
0
3.8E-07
0
0
1.36
1.27
0.48
0.35
0.76
0.14
2.26
1.23
1.88
0.98
0.81
1.53
0.48
0.41
3.45
2.39
3.48
2.19
0
0
0.20
0.01
0.02
3.81
37.63
0.00
1.47
1.16
4.66
PATRAN
Error, %:
0
0
1.36
0.48
0.76
2.26
1.88
0.81
0.48
3.45
3.48
0
0
1.27
0.35
0.14
1.23
0.98
1.53
0.41
2.39
2.19
3.3 Target geometry.
Previously carried out tests have shown that, the program works properly so from this
moment it was treated as a 'black box'. This approach is dictated by large number of nodes
(and data in general) what indicates much more difficult debugging. An additional function
had been implemented for adjusting the diameter of the laser's spot which, during the
measurements, may be changed. Function for proper locating of the heat flux had been added
also. At this phase, real sample's geometry (that will be used in the future in the project) and
real heat flux (heat flux that actually can be applied during the measurements) have been
modeled in Ansys Workbench and were used as final benchmark which includes:




mass matrix lumping impact on the solution,
comparison of the Ansys' and the developed code's calculation times and memory
usage,
comparison of different factorization methods,
comparison of factorizations and PCG,
Additional example including quarter of a cube has been investigated also.
All of the following analyzes were carried out with use of the same hardware and software:
Intel® Core™ i7-3770 CPU @ 3.40GHz 3.4GHz with 8Gb RAM on 64-bit operating system
Windows 7 SP1, Ansys 16.2 (student license), MATLAB 7.9 2009b (academic license).
179
The geometry includes whole sample showed at Figure 17thatcube with dimensions 0.05 x
0.05 x 0.05 m. The mesh is obtained with Sweep method. Number of elements: 7085. Number
of nodes: 31150.
Figure 17: Whole sample. Emission face is marked with the circle.
It should be stressed that, the elements of the mesh presented above has low quality and the
mesh is insufficiently fine so the results are not accurate. However, main purpose of this
analysis was to check if the developed program gives the accurate results in comparison with
Ansys.
Benchmark sample's material is overwritten structural steel in Ansys Workbench with the
following properties:
𝑘𝑔

Density – 1091

Orthotropic TC:
o X: 5.5 𝑊 𝑚 °𝐶 .
o Y: 5.5 𝑊 𝑚 °𝐶 .
o Z: 5.5 𝑊 𝑚 °𝐶.
𝐽
Specific Heat – 900 𝑘𝑔 °𝐶 .

𝑚3 .
All calculations were carried out with the same setup which consists of:




180
Laser's beam radius: 0.0005 m.
Laser's heat flux: 8.5158e+007 𝑊 𝑚 .
Initial temperature: 18˚C.
Laser emission time: 0.19 s.
However, two cases were analyzed with different number of iterations and time steps:


Short – 362 iterations.
Long – 436 iterations.
More information about considered analysis' variants can be found in Table 8. In case of long
analysis, IR camera recording time steps are in random order, so factorization had to be
performed almost every two iterations – this means increased computational effort.
Table 8: Specification of the performed analyzes.
Analysis' variants.
Time, s:
0.19
0.1912
0.49
0.4918
0.4937
0.4957
0.4978
0.5
-
Long:
Steps: Time step, s:
190
0.001
1
0.0009
1
0.0011
125
0.0018
1
0.007
1
0.008
1
0.015
42
0.016
68
0.017
3
0.018
1
0.019
2
0.026
Time, s:
0.19
0.1909
0.192
0.417
0.424
0.432
0.447
1.119
2.275
2.329
2.348
2.4
Laser's emission.
Indirect time step.
Recording the temperature
field (random time steps).
Short:
Steps: Time step, s:
190
0.001
1
0.0012
166
0.0018
1
0.0018
1
0.0019
1
0.002
1
0.0021
1
0.0022
-
3.3.1 Results – full cube.
Figures below (Figure 18, Figure 19and Figure 20) present results of most accurate methods.
For MATLAB it is factorization with CMM and for the Ansys it is direct solver with CMM as
well. Only heated surface is presented. Additionally, Figure 18, Figure 19 and Figure 20 have
different scales. If they had not been different, due to high temperature difference (over
5400°C) Figure 18 and Figure 20 would not have been readable (there would not have been
anything to see).
Figure 18 shows numerically obtained temperature distributions in first time step (after 0.001
s) in MATLAB and in Ansys respectively. This figure provides important information: in
MATLAB, location of the applied heat flux is the same as in Ansys (what means previously
added function works well), diameters of the obtained bright dots are not noticeably different
from each other.
181
Figure 18: Temperature distribution at 0.001 s.
MATLAB (left hand side): Max: 317.878˚C. Min: -34.432˚C.
Ansys (right hand side)Max: 318.64˚C. Min: -36.861˚C.
Negative temperatures occurred due to way of applying NBC. In case of serendipity 8 – node
quadrilateral elements (surface of hexahedron to which NBC is prescribed), coefficients
corresponding to corner nodes are negative. When NBC is applied to several adjacent
elements (so they have common nodes) the negative coefficient in RHS (in case of common
corner node) is sum of negative coefficients from all elements that this particular node
belongs to. In addition to that, those coefficients in RHS depend on heat flux (in this analysis
it is 8.5158e+007 𝑊 𝑚) and on det[J]2D. So huge heat flux in comparison to the surface (it is
66.83 W) and very short time step results in negative temperatures at the corner nodes of
specified surfaces of the boundary elements.
Figure 19 and shows numerically obtained temperature distributions in last iteration when
NBC is applied (0,19 s) – end of laser's emission (maximum temperature during the whole
process).
Figure 19: Temperature distribution at 0.19 s.
MATLAB (left hand side): Max: 5477.259˚C. Min: -8.904˚C.
Ansys (right hand side) Max: 5479.3˚C. Min: -8.825˚C.
182
Considering now again the temperatures, one should notice that the maximum temperatures
are extremely high. Main reason for this situation is that only NBC was applied, with no heat
losses due to radiation and convection. Another thing worth mentioning is the negative
temperatures. Obviously 0.19 s is too short time interval to model heat flux properly
(physically).
Figure 20 shows numerically obtained temperature distributions in last iteration of the short
analysis' variant.
Figure 20: Temperature distribution at 0.5 s.
MATLAB (left hand side): Max: 191.329˚C. Min: 17.999˚C.
Ansys (right hand side): Max: 191.62˚C. Min: 17.999˚C.
In this case all temperatures are positive, moreover minimum temperatures are equal to the
initial temperature. Differences between corresponding to each other nodal temperatures are
negligible also. This leads to the conclusion that, the results become more reliable after
switching off the laser's source (NBC). Obviously use of the results from the final iterations
(some time after the end of the laser's emission) for determining the TC will provide
maximum accuracy. Moreover, spot's diameter noticeably increased, mainly due to time (the
heat has time to spread) and smaller temperature difference (18 – 191˚C) which affects the
color scale (displaying the results).
3.3.2 Solvers' performances – full cube.
Simulations in MATLAB have been carried out with use of two methods:


Direct method with use of factorizations (LU, LDLT, Cholesky, QR).
Iterative method with use of PCG provided by MATLAB [21] with no preconditioner.
Somehow, at least in MATLAB 2009b, preparation of the preconditioner takes much longer
than the analysis with use of PCG without preconditioners, so it has been used without them.
Another disadvantage of MATLAB's PCG is its syntax, namely, while no preconditioner is
used, change of the initial guess becomes impossible due to the lack of input arguments
required by the function to work. When initial guess is not defined, PCG starts each iteration
183
from 0 what leads directly to lose of the efficiency. In other words, to define the initial guess
all other input arguments of the function are necessary.
In further analyzes Ansys' direct method with Consistent Mass Matrix has been assumed as
the one providing an exact solution. All other methods were referenced to it. Moreover, in the
following part several abbreviations appear with common pattern Software_method_mass
matrix:












A_d_cmm (Ansys_direct_Consistent Mass Matrix),
A_d_dlmm (Ansys_direct_Diagonally Lumped Mass Matrix),
A_it_cmm (Ansys_iterative_CMM),
A_it_dlmm (Ansys_iterative_DLMM),
ML_chol_cmm (MATLAB_Cholesky_CMM),
ML_chol_dlmm (MATLAB_Cholesky_DLMM),
ML_ldl_cmm (MATLAB_LdLT_CMM),
ML_ldl_dlmm,
ML_lu_cmm (MATLAB_LU_CMM),
ML_lu_dlmm,
ML_qr_cmm (MATLAB_QR_CMM),
ML_pcg_cmm.
Figure 21 presents accuracy of Ansys' direct and iterative solvers with and without mass
matrix lumping. All errors were calculated with Eq.(18). Differences have been computed for
all nodes corresponding to each other and then mean value was taken into account. As the
A_d_cmm has been chosen as a reference solution, it lies directly on the abscissa.
Figure 21: Average differences in chosen time steps. Only Ansys' solvers included.
As one can notice, all of the Ansys' solvers are very accurate because of insignificant results'
differences. Nevertheless, those differences increase over time until NBC is switched off and
then their stabilizes themselves.
184
Figure 22 presents accuracy of solvers used in MATLAB. Average differences are calculated
in the same manner as in case of Ansys' solvers described above.
Figure 22: Average differences in chosen time steps. MATLAB.
It is important to notice that, all factorizations with use of CMM have the same accuracy.
Similarly factorizations with use of DLMM. Differences between them are much higher than
in case of Ansys' solvers. Moreover, mass matrix lumping has significant impact on the results
contrary to the lumping in Ansys. It also should be noted that in both cases differences grow
with time due to the growth of the nodal temperatures. Obviously, NBC applied in MATLAB
is burdened with inaccuracy.
Table 9 contains the measured analyzes times and required memory of all tested methods of
solving systems of linear equations. In case of ML_qr_cmm and ML_pcg_cmm long analysis
variant has not been carried out due to very long analysis time of the short variant. Required
memory as well as analyzes times for Ansys' solvers have been found in solver output files
(located at the bottom of folder with project's files). In case of MATLAB, required memory
has been computed as a sum of memory used by all variables existing in the workspace right
after factorization or first iteration (in case of PCG).
Table 9: Analysis time and required memory by tested methods.
Method:
A_d_cmm
A_d_dlmm
A_it_cmm
A_it_dlmm
ML_chol_cmm
ML_chol_dlmm
Short:
Analysis time, Requiredmemory,
s:
Mb:
649.0
174.0
628.8
174.0
629.4
174.0
624.4
174.0
537.7
541.2
530.3
515.7
Long:
Analysis time, Requiredmemory,
s:
Mb:
759.3
174.0
746.9
174.0
743.4
174.0
753.5
174.0
882.2
541.2
884.0
515.7
185
ML_ldl_cmm
ML_ldl_dlmm
ML_lu_cmm
ML_lu_dlmm
ML_qr_cmm
ML_pcg_cmm
594.7
592.0
410.6
401.8
2512.8
1246.9
547.6
522.1
992.1
966.7
1325.5
55.0
1000.4
1000.1
983.8
991.3
-
547.6
522.1
992.1
966.7
-
Figure 23 presents analyzes times obtained with all tested methods. QR factorization and
MATLAB's PCG (with no preconditioner) have significantly worse performances so they
have been omitted in the short analysis with DLMM and in the whole long variant. Presented
times contain all computational processes from the very beginning to the end of the analyzes.
In Ansys it is from the start of the solution to its end and in MATLAB from import data from
Ansys' solver output file .dat through preparing all necessary matrices (building system of
equations) to the end of calculations. It is important to keep in mind that, the MATLAB used
to the analyzes has no parallel computing toolbox, so only one core has been used. For this
reason all times concerning analyzes carried out in Ansys are the total CPU time summed for
all threads (it can be also found in solver output file).
Figure 23: Analyzes times of all tested methods – short variant.
It is important to notice that, mass matrix lumping has no significant effect on the analyzes'
times. It confirms the fact known from the literature [16], that in case of implicit scheme (no
mass matrix 'inversion') lumping has no major sense. What is more, the program with use of
LU, LdLT and Cholesky factorizations is slightly faster than Ansys (short variant).
Figure 24 presents times of analyzes used in long variant of analysis. Five methods were
tested in total.
186
Figure 24: Analysis time of all tested methods – long variant.
In contrast to the short variant (Figure 23), Ansys in this case has better performance than the
code written in MATLAB. Furthermore, once again times of the analyzes hardly depend on
the mass matrix lumping which, in turn, seriously affects the results in case of the MATLAB.
Additionally, LU factorization was the fastest method (in MATLAB) in short variant but in
this case the Cholesky turned out to be faster. Explanation to this is as follows – solver written
in MATLAB saves factorized matrices of different time steps to the hard drive as .mat files in
temporary directory. Time consumed by saving as well as loading these matrices highly
depends on their size. LU factorization is almost twice the size of the Cholesky factorization
so saving and loading takes more time. Such an approach has sense in case of small meshes
(like in this analysis) when saving and loading factorized matrices takes less time than
carrying out whole new factorization and when time steps are not very diverse (infrequent
saving, frequent loading of necessary factorized matrices). For instance, Cholesky
factorization takes about 10 s, saving takes about 15 s but loading only about 2 s (these times
are estimated, because each loading or saving takes different time due to temporary CPU
usage by other processes).
Figure 25 presents memory used by variables. Ansys' solvers alone are far more economical
than those written in MATLAB except for PCG but, as it was mentioned before, in MATLAB
2009b it had poor performance in comparison to the factorizations.
187
Figure 25: Memory used by tested methods.
Figure 25 does not show this clearly, but mass matrix lumping allows to save about 25 Mb of
memory in case of considered small mesh (see Table 9). It is negligible advantage. To state if
Ansys actually is more economical than the program written in MATLAB, one should take
into account one additional factor, namely, memory requested by the software itself. In Table
10is summarized minimum amount of memory needed to carry out the necessary calculations.
Table 10: Memory required to run the analysis.
Ansys:
Workbench:
Mechanical:
Solver:
Total:
MATLAB:
Variables:
Total:
Memory, Mb:
280*
240*
2112
2632
500
600**
1100
* estimated value.
** estimated value on the assumption of the Cholesky or LDLT factorization in case of similar
number of elements as in already considered mesh.
To run the thermal analysis with use of the Ansys, one has to open Workbench, Mechanical
and additionally start the solver. Unfortunately, during starting the solver Ansys allocates
2112 Mb of RAM although in cases described earlier, uses only 174 Mb. MATLAB, in turn,
requires more memory during computations but, after all, smaller amount of available
memory will met the requirements. What is more, if MATLAB is started as a command line
alone, required memory reduces from about 500 Mb to a little over 100 Mb. Such approach
allows to perform even complex numerical analyzes with use of old hardware when the
available memory is limited.
188
In order to check impact of saving factorized matrices to the hard drive on the solver's
performance, two most promising methods (LU and Cholseky) have been chosen and tested
once again. In this analysis, solver written in MATLAB does not save any matrices to the hard
drive. In Table 11are gathered analyzes times and memory usage of the considered methods.
Table 11: Analyzes times and memory usage of LU and Cholesky factorizations in case of
modified solver.
Long:
Analysis time, s: Required memory, Mb:
759.3
174.0
A_d_cmm
ML_lu_cmm
1044.1
992.1
ML_chol_cmm
749.4
541.2
Method:
Figure 26 presents graphical comparison of the analyzes times of Ansys and two chosen
factorizations in MATLAB. Long analysis.
Figure 26: Solver performance with saving factorizations to the hard drive and without in
comparison to Ansys' direct solver.
In case of LU factorization, solver which saves factorized matrices to the hard drive is slightly
faster than one without saving. However, in case of Cholesky factorization the situation is
contrary. As it was mentioned before, memory usage of LU is almost twice as the Cholesky,
so saving and loading prepared matrices improves efficiency of the solver (instead of doing
whole new factorization), but it is still much slower than Ansys. On the other hand Cholesky
factorization based solver improved its efficiency after switching off saving and loading
factorized matrices. Furthermore, its efficiency is slightly better than Ansys' direct solver.
Memory usage of these methods remained unchanged. From the viewpoint of the project, the
analysis time and accuracy of the described program are crucial. However, one cannot forget
about memory usage. In case of more complex meshes with increasing number of elements,
memory requirements of the direct solver grow rapidly – double number of nodes results in,
189
approximately, fourfold increase in the size of the [K] and [M]. Excluding all other arrays, it
gives 8 times more memory required to store these matrices (not mentioning the time required
to build them), what affects the final performance (time needed to assembly and later for
solving system of equations). Analysis of the solver always has to be supported by the
investigation of the memory usage.
3.3.3 Analysis setup, geometry and mesh –quarter of a cube.
Carrying out the calculations with use of the full cube is inefficient when geometry together
with boundary conditions are symmetric. In such a case, usage of part of a whole domain is
far more economical because it significantly affects the computational effort simultaneously
providing the same results. To do so, all body parts of the full cube have been suppressed
except for the ones belonging to chosen quarter. Because of Ansys that keeps suppressed
nodes, the selected quarter was re – meshed. As side research showed, the program written in
MATLAB coped with that case (suppressed nodes are present in stiffness and mass matrices,
but their coefficients are empty) and as a result, correct temperature field was obtained.
Whereas for all suppressed nodes it was "NaN" (Not a Number) instead of temperature value.
Analysis time differs from the one obtained after re – meshing and is a little bit higher.
Figure 27: Mesh of a quarter of the cube. 2015 elements and 9782 nodes.
Mesh of a quarter of the cube (Figure 27) consists of: 2015 elements and 9782 nodes and has
been generated in the same way as the mesh of the full cube. Whole analysis' setup remained
unchanged. To check, if the program works properly, short variant (362 iterations) has been
used.
190
3.3.4 Results – quarter of a cube.
In general, results from both analyzes (full and quarter of the cube, see Table 12) are almost
identical which means that the program works properly in case of reduced domain along its
axes of symmetry. Once again, too short time intervals result in negative temperatures due to
negative coefficients in RHS vector. It should be noted that maximum and minimum
temperatures hardly differs from those obtained in case of the full cube. Similar mesh gave
similar results.
Table 12: Minimum and maximum temperatures of the full and quarter of the geometry.
Geometry:
Time, s:
Ansys
MATLAB
0.001
0.19
0.5
0.001
0.19
0.5
Full.
Max.
318.64
5479.30
191.62
317.88
5477.26
191.33
Quarter.
Temperature, ˚C.
Min.
Max.
Min.
-36.86
318.80 -36.79
-8.82
5488.50 -11.09
18.00
191.39
18.00
-34.43
318.06 -34.47
-8.90
5486.38 -12.42
18.00
191.17
18.00
One should keep in mind that, the quarter has been re – meshed so its mesh differs from the
one just after suppressing all other unnecessary bodies (after re – meshing number of elements
increased by about 400) despite the fact that none of the mesh setup parameters has been
changed. The maximum temperatures in case of the quarter are slightly higher than in case of
the full cube. Obviously, in considered case, the results depend on the mesh so in the future
further investigation will be necessary.
Figure 28 shows average percentage differences between nodal temperatures corresponding to
each other. Once again, solution provided by Ansys direct solver without mass matrix
lumping has been assumed as the accurate one.
191
Figure 28: Accuracy comparison of the selected factorizations and Ansys' direct solver.
Again, accuracy decreases in the course of applying NBC and reaches its lowest value in the
last step before turning off the heat flux. Both factorizations have the same accuracy.
Table 13 contains analyzes times and memory usage of the selected factorizations in
comparison to the Ansys' direct solver. MATLAB clearly became more efficient than Ansys.
Table 13: Analyzes times and memory usage of the Ansys' direct solver, LU and Cholesky
factorizations. Quarter of the geometry. Short variant.
Method:
A_d_cmm
ML_chol_cmm
ML_lu_cmm
Analysis time, s:
158.4
51.2
33.3
Memory used, Mb:
55
76.58
125.9
Figure 29 shows obtained times of the performed analyzes of the quarter of the geometry.
These times are calculated in the same manner as times provided by case of whole geometry
discussed before.
192
Figure 29: Analyzes times of the tested methods of solving linear systems of equations.
Reduced mesh (and geometry) results in improved performance of the program. It should be
stressed that, use of the hard drive as a way of storing factorized matrices, in this analyzes, has
been permanently turned off. Figure 30 presents memory usage of the considered methods.
Used memory has been obtained in the same way as in the previous case.
Figure 30: Memory usage of the tested methods of solving linear systems of equations.
As one can see, there is no major difference between Cholesky factorization carried out in
MATLAB and Ansys' direct solver. LU factorization requires memory the most
(approximately two times more than Ansys).
Finally, Cholesky factorization has been chosen as the method of solving Eq.(17) because of
its satisfying performance and memory usage in both tested variants. Mass matrix will not be
lumped due to serious loss of the accuracy and negligible impact on performance. Storing
temporary factorized matrices will not be used because of the Cholesky factorization
performance degradation.
193
4 Conclusions.
Performance of the code is one of the most important factors in the developing program for
determining thermal conductivity in a way described at the beginning of this work. Such a
program have to be efficient, because during the analysis it will be called and executed many
times until convergence of the TC is reached. For example, the analysis takes 10 minutes for
poorly optimized program. To determine the TC the program has to be called 100 times. This
brings a total of 1000 minutes for one full analysis. Another program, much better optimized,
needs 7 minutes only for completing the same one – run analysis. It also has to be called 100
times, but total time required to complete the full analysis amounts to 700 minutes only. This
gives saving of 300 minutes which equals to 5 hours. This is a lot of time and that is why a
good performance is so important to this program.
Mass matrix lumping is one of the available options provided by commercial software so it
was tested in this work. Main cause for which it was not implemented, was lower accuracy of
the code and negligible impact on the performance.
Obtained results from the MATLAB with use of 3rd order quadrature at phases 1 and 2 and
with NBC applied, differ from, approximately, 0℃ to 1e-03℃ in comparison to the values
provided by MSC.Patran. In case of the 2nd order quadrature those differences are much
higher and they are from about 1e-03℃ to 0.9℃. They reach maximum values at the nodes
to which NBC is applied and the smaller they are the further from the BC of the second
kind. In case of pure DBC those differences are negligible and do not depend on the used
quadrature's order.
In comparison to Ansys Transient Thermal (direct method without mass matrix lumping),
differences of the results are much higher and their maximum values are, again, at the nodes
to which NBC is applied. Global maximum differences are reached in the last step when NBC
is applied and amount to even about 12℃ while maximum temperature is about 5500℃. After
NBC removal, those differences drastically decreased to about 1e-02℃ or even less. The more
distant node is (from the NBC) the lower difference is between corresponding to each other
nodal temperatures provided by the Ansys' direct solver and factorization in MATLAB.
All of the Ansys' solvers (direct and iterative) were accurate with respect to each other
regardless of whether lumping was used or not. Most likely some additional unknown
procedure is performed, however manual does not refer to it.
Factorizations (direct method) have the same accuracy but they strongly differ from each
other in case of performance. The selection of the type of the factorization should be preceded
by the analysis of their efficiency for each particular case (number of nodes in the mesh,
number of iterations, number of time steps, available memory). As an example, in short
variant LU factorization was the best but in long variant Cholesky was better.
Mass matrix lumping in case of implicit scheme (BDM) has no major sense due to its
negligible impact on the performance. On the other hand, it strongly affects the results
damaging the accuracy, especially in early iterations. Lumping becomes justifiable in case of
dealing with huge geometry with use of explicit scheme and when computation's time is more
important than the accuracy.
194
Direct methods are faster and more accurate than iterative solvers but simultaneously they
require far more available memory.
Neumann's Boundary Condition, which simulates the laser, generates errors that grow with
time until it is turned off. The results from iterations after, approximately, 0.4 s from the
beginning of the numerical analysis, have very good accuracy and can be successfully used in
further computations.
Nature of the applying the NBC requires some amount of time to model temperature
distribution in a physical way.
During this work the program has been developed for the purposes of the project. This
program not only solves NBC – based transient heat transfer problems in orthotropic
materials, but also imports geometry data, exports solution data, has log module, error
reporting module and above all, is easy to manage, cheap (so to speak – free) and fully –
customizable. Furthermore, its performance as well as accuracy is not worse than Ansys'.
More information about the program features and capabilities can be found in Appendix A.
In the future, incorporation of the program into main algorithm for determining the TC will
have to be done. This will require the implementation of the calling (this program), data
managing and sharing(between the program and the main algorithm) procedures. In addition
to that, radiative heat losses will have to be implemented in order to make the obtained
temperature field more reliable. Last thing to do is to make density and specific heat
temperature – dependent.
References
[1] Thermal
Conductivity
Measurements
Methods,
http://tpm.fsv.cvut.cz/student/documents/files/BUM1/Chapter16.pdf, access 14 June
2016.
[2] W.P. Adamczyk et al., Retrieving thermal conductivities of isotropic and orthotropic
materials,
Applied
Mathematical
Modelling
(2015),
http://dx.doi.org/10.1016/j.apm.2015.10.028.
[3] J. Parker, R. Jenkins, C. Butler, G. Abbott, Flash method of determining thermal
diffusivity, heat capacity and thermal conductivity, J. Appl. Phys. 32 (9) (1961) 1679–
1684.
[4] Description
of
MATLAB,
website
of
the
MathWorks,
http://www.mathworks.com/products/matlab/, access 2 June 2016.
[5] Website
of
the
FileExchange
provided
by
the
MathWorks,
http://www.mathworks.com/matlabcentral/fileexchange/, access 2 June 2016.
[6] E. Kostowski, Przepływ Ciepła, Wydawnictwo Politechniki Śląskiej, Gliwice 2006, pp.
28.
[7] A. J. Nowak, ed, Numerical Methods in Heat Transfer, International Studies in Science
and Engineering, Gliwice, 2009, pp.77-96.
[8] J. Szargut, Modelowanie numeryczne pól temperatury, Wydawnictwo Naukowo Techniczne, Warszawa, 1992, pp. 113-115.
[9] O. C. Zienkiewicz, Metoda elementów skończonych, Warszawa, 1972, pp. 123-125.
195
[10] Finite Element Method – shape functions, website of the University of Colorado
Boulder,
http://www.colorado.edu/engineering/CAS/courses.d/AFEM.d/AFEM.Ch11.d/AFEM.
Ch11.pdf, access 2 June 2016.
[11] R.W. Lewis, K. Morgan, H.R. Thomas, K.N. Seetharamu, The Finite Element Method
in Heat Transfer Analysis, Wiley, 1996.
[12] Serendipity family of finite elements, website of the Cornell University Library,
http://arxiv.org/pdf/1101.0645.pdf, access 2 June 2016.
[13] Gaussian quadrature, website of the California State University Fullerton, Department
of
Mathematics,
http://mathfaculty.fullerton.edu/mathews/n2003/GaussianQuadMod.html, access 2
June 2016.
[14] Jacobian
matrix,
website
of
the
StachExchange,
Mathematics,
http://math.stackexchange.com/questions/14952/what-is-jacobian-matrix, access 2
June 2016.
[15] Sarru's rule for calculating 3x3 matrix determinant, website of the Wikipedia,
https://en.wikipedia.org/wiki/Rule_of_Sarrus, access 2 June 2016.
[16] Lumped and consistent mass matrices, website of the Kielce University of Technology,
Department
of
Informatics,
http://kis.tu.kielce.pl//mo/COLORADO_FEM/colorado/IFEM.Ch31.pdf, access 2 June
2016.
[17] Left
division,
website
of
the
MathWorks,
Support,
http://www.mathworks.com/help/matlab/ref/mldivide.html?refresh=true, access 2 June
2016.
[18] P. Tarvydas, A. Noreika, Usability Evaluation of Finite Element Method Equation
Solvers, Electronics and Electrical Engineering, 2007, No. 2(74).
[19] Factorization
written
by
Timothy
Davis,
FileExchange,
http://www.mathworks.com/matlabcentral/fileexchange/24119-don-t-let-that-inv-gopast-your-eyes--to-solve-that-system--factorize-, access 2 June 2016.
[20] T. A. Davis, Algorithm 9xx, Factorize: an object – oriented linear system solver for
MATLAB, ACM Transactions on Mathematical Software, Vol. V, No. N, M 20YY, pp.
1-20.
[21] Preconditioned Conjugate Gradient, website of the Mathworks, Support,
http://www.mathworks.com/help/matlab/ref/pcg.html, access 2 June 2016.
196
Acknowledgements
I would like to express my immense gratitude to my supervisor
dr ing. Arkadiusz Ryfa for his infinite patience and all his time he devoted to me
and for that, he taught me basics of the Finite Element Method.
Necessary hardware and software have been provided by
the Institute of Thermal Technology for which I cordially thank.
I would like also wholeheartedly thank my dear mother
Ewa Tokarska for all the support she shown to me during my studies.
Thank you, Mum.
197
Analiza numeryczna pola temperatury w ortotropowej próbce
Mieszko Tokarski2
e-mail: [email protected]
Słowa kluczowe: MATLAB, Ansys, FEM, Warunek Brzegowy Neumann'a, Faktoryzacja
Streszczenie
W pracy tej opisano proces tworzenia pogramu opartego na Metodzie Elementów
Skończonych a służącego do obliczania pola temperatury w ortotropowej próbce z
wykorzystaniem warunku brzegowego Neumann'a. Program ten został stworzony na potrzeby
projektu prowadzonego w Instytucie Techniki Cieplnej w Gliwicach. Stanowi on ważną część
zautomatyzowanej procedury mającej wyznaczać współczynnik przewodzenia ciepła badanej
próbki poprzez dopasowanie rozkładu temperatury otrzymanego na drodze obliczeń
numerycznych z rozkładem temperatury otrzymanym z pomiarów. Z racji specyfiki pomiarów
oraz samej procedury sterującej całą analizą, stworzony program charakteryzuje się dużą
wydajnością porównywalną do wydajności Ansysa (dla tego samego przypadku),
wystarczającą dokładnością oraz przygotowaniem do sprzęgnięcia z głównym algorytmem.
Najważniejszymi cechami tego programu są: moduł do importu geometrii z Ansysa, moduł do
eksportu danych dla algorytmu sterującego, dwa pliki tekstowe umożliwiające łatwe
zarządzanie tym kodem (zwłaszcza w przypadku sterowania przez zewnętrzny program),
logowanie działania programu oraz raportowanie o błędach.
2
Autor przygotował niniejszy rozdział podczas pracy nad projektem dyplomowym
magisterskim wykonywanym przez autora w Instytucie Techniki Cieplnej na Wydziale
Inżynierii Środowiska i Energetyki Politechniki Śląskiej pod opieką dr inż. Arkadiusza Ryfy.
Prace były prowadzone w ramach projektu naukowego realizowanego w Instytucie Techniki
Cieplnej.
198
Appendix A
Features of the developed program
One can distinguish three main characteristic features of the code:



Consistency – the code written in MATLAB is executed line by line, so it is logical
and without any unnecessary commands, for example clearing variable just to load it
again in another line. The same applies to the variables and external functions – they
are well named to make sense without thinking too much. Repetition of the lines of the
code is also eliminated, because this makes editing and debugging the code easier due
to lack of necessity of modifying additional lines.
Simplicity and transparency – program has been written in such a way that it can be
easily debugged and modified. This has been achieved by the use of basic functions,
for instance 'For' loops instead of more sophisticated nested functions like 'cellfun' or
'bsxfun'. In case of MATLAB, 'For' loop has better performance than most of the other
functions. Another thing that makes debugging easier is modular design. Grouping of
the code into smaller pieces responsible for strictly specified tasks is a good habit and
it helps one finding procedural errors.
Forethought – this program can be easily extended with new functions and procedures
without major changes into already existing code.
The program is started with use of batch file to call MATLAB's command line and to declare
all necessary directories automatically. It consists of:

startPanel_0 - main procedure which calls and coordinates all others,
o globalVar – structural variable storing all global variables (program's memory),
o var_GDefine_1 – import module, uses UserSettings.txt,
 import_NCoords_2 – subprogram for nodal coordinates import from
Ansys' geometry output files,
 import_Connectivity_3 – subprogram for connectivity matrix import
import from Ansys' geometry output files,
 import_MConstants_4 – subprogram for material properties import
from Ansys' geometry output files,
 import_FENodes_5 – subprogram for face and emission nodes import
from Ansys' geometry output files,
 import_USetup_6 – subprogram for user setup data import from
UserSetup.txt file,
o un_varDelete_0 – subprogram responsible for clearing all of unnecessary
variables from memory,
o MatCon - main subprogram for building system of equations,
 Ansys_N_order – function responsible for proper nodal reordering,
 Ansys_N_order_8n – function responsible for building connectivity
matrix in order to apply NBC,
 LaserRadius – function controlling laser's diameter,
 Ni – function with shape functions 20 – node hexahedron, PATRAN's
order,
199

Shape_functions_Patran_8n – function with shape function 8 – node
quadrilateral, PATRAN's order,
 dNdksi_Patran_20n – function with shape functions' 3d derivatives,
 dNdksi_PAtran_8n – function with shape functions' 2d derivatives,
o Solver – main subprogram for solving linear systems of equations,
 Factorize – algorithm written by Timothy Davis,
 MemoryCheck – function computing memory usage by variables in the
workspace.
Whole program is controlled with use the of two text files (Figure 31 and Figure 32):


UsetSettings.txt – defining geometry file and its components to import:
o nodal coordinates,
o connectivity matrix,
o material properties (disabled by default),
o face and emission nodes,
UserSetup.txt – defining setup of the analysis:
o density,
o specific heat,
o TC components x, y, z,
o laser beam radius,
o laser heat flux,
o initial temperature,
o laser emission time,
o camera recording start time,
o camera recording times (in fact time increments)– this vector can be almost
freely long and that is why it is at the very end of the file.
The program checks for last modification date of UserSettings and UserSetup and compares
them with ones already stored as variables in structural array globalVar. If they differ from
each other then program overwrites them and executes proper procedures of import (when
UserSettings date variable is overwritten) and/or overwrites setup variables (when UserSetup
date variable is overwritten). In special case while such date variable or variables do not exist,
they are created as new ones and again all necessary procedures are executed.
200
Figure 31: UserSettings.txt input file.
Figure 32: UserSetup.txt input file.
All subprograms and functions are vectorized and optimized in order to achieve as high
performance as possible. Moreover, the program is fully automated – besides input text files
(UserSettings.txt and UserSetup.txt) does not require any other user interference. It should be
stressed that, with use of input files (that will be prepared by the main algorithm or user) it
can be easily managed by the external procedures.
Results are saving into specially created folder in working directory named
"Results_YYYY.MM.DD__hh.mm.ss", where " YYYY.MM.DD__hh.mm.ss" is current date
and time:






YYYY – year,
MM – month,
DD – day,
hh – hours,
mm – minutes,
ss – seconds.
Results are stored as .dat files named "T(s).dat", where "(s)" is step number.
Figure 33 and Figure 34 show exemplary folder with exemplary files with results.
Additionally to these files, analysis' configuration file is attached.
201
Figure 33: Folder containing analysis' results.
Figure 34: Files .dat with results together with UserSetup.txt which stores analysis'
configuration.
202
Each T(s).dat file contains: all nodes of heating surface, nodal coordinates, nodal temperatures
and, additionally, current analysis' time, current time step and current iteration number.
Exemplary T(s).dat file is showed in Figure 35.
Figure 35: Files .dat with results together with UserSetup.txt which stores analysis'
configuration.
Logfile.txt (Figure 36) is created in the main directory. All statements about executed or
skipped procedures are stored in that file. Also errors and warning reports as soon any occurs.
Memory usage and analysis time are also reported. Furthermore, logfile.txt and UserSetup.txt
are copied into results' folder to provide additional information about the process and its
configuration each time the analysis is performed.
203
Figure 36: Exemplary part of the logfile.txt. It provides information about executed or skipped
procedures, analysis' time and memory usage.
204
Indeks autorów
Bandola D., 5
Gracka M., 29
Knopek A., 53
Orłowski A., 81
Simla T., 113
Tokarski M., 153
205
Indeks słów kluczowych
Świnoujście LNG terminal, 113
Ansys, 153
ANSYS Fluent, 53
aorta, 5, 29, 53
arterial pressure, 5
blood flow, 5, 29, 53
boundary conditions, 5
promienniki podczerwieni, 81
przepływ krwi, 5, 29, 53
przepływ wielofazowy, 29
radiant heater, 81
radiation, 81
regasification, 113
regazyfikacja, 113
skroplony gaz ziemny, 113
CFD, 29, 53
ciśnienie t˛etnicze, 5
cryogenic exergy, 113
DPM, 53
egzergia kriogeniczna, 113
energy recovery, 113
Euler-Euler, 53
Euler-Euler model, 29
Euler-Lagrange, 53
Factorization, 153
Faktoryzacja, 153
FEM, 153
hala przemysłowa, 81
heating, 81
industrial hall, 81
LNG, 113
MATLAB, 153
model Euler-Euler , 29
multi-fluid, 29
multiphase, 53
Neumann’s Boundary Condition, 153
odzysk energii, 113
ogrzewanie, 81
promieniowanie, 81
206
terminal LNG w Świnoujściu, 113
Warunek Brzegowy Neumann’a, 153
warunki brzegowe, 5
wielofazowy, 53
Windkessel Model, 5