Sie sind auf Seite 1von 102

DISSERTATION

Internal Flow and Valve Dynamics


in a Reciprocating Compressor

ausgeführt zum Zwecke der Erlangung des akademischen Grades eines Doktors der
technischen Wissenschaften unter der Leitung von

Ao.Univ.Prof. Dipl.-Ing. Dr. techn. Herbert Steinrück


am Institut für Strömungsmechanik und Wärmeübertragung (E 322)

eingereicht an der Technischen Universität Wien


Fakultät für Maschinenwesen und Betriebswissenschaften

von

Dipl.-Ing. Roland Aigner


Matrikelnummer 9625691
1230 Wien, Triester Straße 221/3/9

Wien, im Juni 2007


Kurzfassung

Die Strömung in einem Koblenkompressor wird durch die Bewegung des Kolbens und
durch das Öffnen und Schließen der Ventile hervorgerufen. Vorallem die Wechselwirkung
von Druckwellen mit der Ventilbewegung bestimmt die auftretenden Strömungsformen.
Die vorliegende Dissertation liefert ein genaues aber einfaches numerisches Verfahren
um die Innenströmung und die Ventildynamik zu berechnen. Dabei erstreckt sich das
betrachtete Rechengebiet von den Ventilen der Saugseite bis zur Druckkammer des Kom-
pressors auf der Auslassseite. Für die mathematische Beschreibung des Strömungsfeldes
werden quasi eindimensionale Eulergleichungen verwendet. Bei Kompressoren mit mehr
als zwei Ventilen wird hingegen ein zweidimensionales Modell eingeführt um den Ein-
fluss mehrerer Ventiltaschen zu berücksichtigen. Obwohl die Energiegleichungen bei den
Eulergleichungen durch die Isentropenbeziehung ersetzt wird, werden Druckverluste an
Sprüngen des Strömungsquerschnittes mit Hilfe von Verlustkoeffizienten eingerechnet. Ver-
schiedene Randbedingungen vervollständigen das mathematische Strömungsmodell. Für
die Ventilbewegung wird das Modell von Costagliola dahingehend modifiziert, dass auch
Druckwellen im Zylinder berücksichtigt werden können.
Die resultierenden Gleichungssysteme werden mit Hilfe finiter Volumenverfahren von
Le Veque, Mac-Cormack, Lax, Wendroff und Friedrichs gelöst. Ein besonderes Augenmerk
wird dabei auf die Vor- und Nachteile der einzelnen Verfahren gerichtet.
Schließlich wurden Messungen an einem Testkompressor mit zwei Ventilen von Burck-
hardt Compression (Schweiz) durchgeführt. Vergleiche mit den numerischen Lösungen
zeigen, dass die wichtigsten physikalischen Effekte genau beschrieben werden und daher
sowohl die Strömungsform als auch die Ventilbewegung gut bestimmbar sind. Zusätzlich
wurden die Messergebnisse eines Kompressors mit acht Ventilen von Ariel (USA) mit der
Lösung des zweidimensionalen Modells verglichen. Hier werden die von der Simulation
vorhergesagten komplizierten Wellensystem von der Messung bestätigt.
Abstract

The gas flow inside a reciprocating compressor is the result of the oscillating piston motion
and the opening and closing of the valves. Above all, the interaction of pressure waves
with the valve motion determines the present flow patterns.
This thesis provides an accurate but simple numerical model capable of calculating
internal flows and valve dynamics in a reciprocating compressor. The considered compu-
tational domain extends from the valves of the suction side to the pressure chamber of
the discharge side. Quasi one-dimensional Euler equations are employed to describe the
gas flow. However, in case of cylinders with more than two valves two-dimensional models
are introduced to account for the effects of multiple valve pockets. Although the energy
equation is replaced by the isentropic condition, pressure losses at sudden changes of flow
cross-section are taken into account by means of loss coefficients. Different kinds of bound-
ary conditions complete the mathematical model of the gas flow. The valve dynamics are
described by a modified Costagliola-model which takes pressure waves in the cylinder into
account.
In order to solve the governing systems of equations finite volume methods from LeV-
eque, Mac-Cormack, Lax, Wendroff and Friedrich are employed and their advantages
highlighted.
Finally measurements have been carried out on a test compressor with two valves
from Burckhardt Compression (Switzerland). Comparisons with the results of the quasi-
one-dimensional model show that main physical effects are described accurately and thus
both, gas flow in the compressor and valve motion can be predicted well. In addition the
measurement results of a compressor from Ariel (USA) with eight valves are compared to
the solution of the quasi two-dimensional model. Here the complicated two-dimensional
wave pattern inside the cylinder and valve motion agrees well with the the measurement
data.
Acknowledgement / Danksagung

Herrn Ao. Prof. Dr. Herbert Steinrück sei an dieser Stelle mein besonderer Dank ausge-
sprochen. Nur durch seine Unterstützung, Betreuung, und Hilfe wurde diese Dissertation
möglich.

Ich bedanke mich sehr herzlich bei Prof. Gotthard Will für die Übernahme des Koreferats.

Herrn Dr. Peter Steinrück, Herrn Dr. Georg Samland und Herrn Dr. Gunther Machu
danke ich sehr herzlich für die zahlreichen fachlichen Diskussionen.

I am very grateful to Fred Newman who provided me with valuable suggestions and mea-
surement data from Ariel Corporation.

Dank gebührt auch Dr. Daniel Sauter für die reibungslos verlaufenden Messungen bei
Burckhardt Compression AG.

Herzlich bedanken möchte ich mich bei meiner Freundin Inés, meinen Freunden und meiner
Familie.

Des Weiteren möchte ich mich bei den Mitarbeitern des Instituts bedanken. Vor allem
Richard, Uli, Guido und Thomas haben den Arbeitsalltag erleichtert und in so manchen
unproduktiven Stunden für Ausgleich gesorgt. Besonders dankbar bin ich Herrn Prof. Wil-
helm Schneider für die zahlreichen wissenschaftlichen und fächerübergreifenden Ratschläge.

I acknowledge the European Forum for Reciprocating Compressors that has financed this
work.

1
Contents

Content . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . III
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VII
List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IX

1 Introduction 1

2 Modelling of the Reciprocating Compressor 5


2.1 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Simulation domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 One-dimensional model . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.3 Two-dimensional model . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Characteristic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 Gas Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 Heat Transfer and Dissipation . . . . . . . . . . . . . . . . . . . . . . 9
2.2.3 Classification of Compressors . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Piston Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Gas Flow inside the Compressor . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.1 Quasi one-dimensional Model . . . . . . . . . . . . . . . . . . . . . . 10
2.4.2 Quasi two-dimensional Model . . . . . . . . . . . . . . . . . . . . . . 13
2.4.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.4 T-piece . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.5 Interface Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.6 Pressure Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Valve Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5.1 Valve Dynamic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5.2 Flow through the valve . . . . . . . . . . . . . . . . . . . . . . . . . 19

I
2.5.3 Asymmetric Flow in Valve Pockets, Valve Masking and Slotted Valve
Pockets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3 Numerical Method 22
3.1 Finite Volume Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.1 One-dimensional Finite Volume Schemes . . . . . . . . . . . . . . . . 22
3.1.2 Two-dimensional Finite Volume Schemes . . . . . . . . . . . . . . . 30
3.2 Numerical Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Walls one-dimensional . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.2 Mass-flow through the Boundary . . . . . . . . . . . . . . . . . . . . 35
3.2.3 Pressure Outlet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.4 Boundary Problem at Sudden Changes of Cross-section . . . . . . . 37
3.2.5 Boundary Conditions at the Valve . . . . . . . . . . . . . . . . . . . 38
3.2.6 Boundary Conditions at the T-piece . . . . . . . . . . . . . . . . . . 39
3.2.7 Wall two-dimensional . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.8 Interface Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Equation of motion for the valve plate . . . . . . . . . . . . . . . . . . . . . 42

4 Measurements 44
4.1 Compressor with two valves . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.1.1 Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.1.2 Results of Measurement . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Compressor with eight valves . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5 Comparison of Measurement and Numerical data 59


5.1 One-dimensional Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1.1 Comparison with Measurement Results of the Burckhardt test com-
pressor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1.2 Super elevation of the pressure due to initial sticking . . . . . . . . . 63
5.1.3 Pressure loss at sudden changes of cross section . . . . . . . . . . . . 63
5.1.4 Different locations of outflow boundary . . . . . . . . . . . . . . . . 64
5.1.5 Impact velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.1.6 Valve losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2 Two-dimensional model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2.1 Comparison with measurement results of the Ariel test compressor . 65
5.2.2 Variation of discharge pressure . . . . . . . . . . . . . . . . . . . . . 69

II
5.2.3 Pressure distribution inside the cylinder . . . . . . . . . . . . . . . . 70
5.2.4 Moment onto the piston . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2.5 Valve Opening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2.6 Valve Masking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

6 Remarks on Numerical Schemes 75


6.1 Comparison of full three-dimensional, quasi one-dimensional and quasi two-
dimensional numerical methods . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.1.1 Pressure waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.1.2 Computational Time . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.2 Limitations of the Numerical Schemes . . . . . . . . . . . . . . . . . . . . . 78
6.2.1 Limitations of the one-dimensional model . . . . . . . . . . . . . . . 78
6.2.2 Limitations of the quasi two-dimensional model . . . . . . . . . . . . 79

7 Summary 80

A Tables 82
A.1 Sensors of Burckhardt Test Compressor . . . . . . . . . . . . . . . . . . . . 83
A.2 Pressure loss coefficients for the T-piece . . . . . . . . . . . . . . . . . . . . 83

Reference 85

Curriculum Vitae 87

III
List of Figures

1.1 Double-acting compressor of barrel-design. . . . . . . . . . . . . . . . . . . . 1


1.2 Results of a full three-dimensional simulation: pressure distribution at the
cylinder head and in the symmetry plane [22]. . . . . . . . . . . . . . . . . 2

2.1 Flow geometry inside the compressor. . . . . . . . . . . . . . . . . . . . . . 6


2.2 Schema of the one-dimensional model. . . . . . . . . . . . . . . . . . . . . . 6
2.3 Cross-section areas of the one-dimensional model. . . . . . . . . . . . . . . . 6
2.4 Model of discharge system VR1: Pressure chamber is modelled at the end
of the valve retainer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 Model of discharge system VR2: Pressure chamber is connected to the valve
retainer by a T-piece. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.6 Model with discharge pipe and damper. Crank end working chamber is
optional (dashed lines). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7 Schema of the two-dimensional model. . . . . . . . . . . . . . . . . . . . . . 9
2.8 Schema of the two-dimensional model with eight valves. . . . . . . . . . . . 9
2.9 Right boundary of computational domain: Characteristics variables enter-
ing and leaving at the right end of the computational domain. . . . . . . . . 14
2.10 T-piece consisting of tube 1, tube 2 and tube 3: The shaded square displays
the junction area. Its appropriate in-going and out-going characteristics w1,i
and w2,i are indexed by the adjacent tube i = 1, 2, 3. . . . . . . . . . . . . . 15
2.11 Interface between one-dimensional and two-dimensional domain. . . . . . . 16
2.12 Modelling pressure losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.13 Sudden expansion of cross-section . . . . . . . . . . . . . . . . . . . . . . . . 17
2.14 Pressure losses at the T-piece . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.15 Schema of a plate valve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.16 Pressure distribution and velocity vectors in the symmetry plane during
discharge (full three-dimensional simulation). . . . . . . . . . . . . . . . . . 21
2.17 Valve Masking. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.18 Slotted Valve Pockets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

IV
3.1 Subdivision of the computational domain into grid cells Ci . . . . . . . . . . 23
3.2 Riemann problems at each cell boundary. . . . . . . . . . . . . . . . . . . . 25
3.3 Riemann problem at the left boundary xi− 1 of the ith cell . . . . . . . . . . 26
2

3.4 Riemann problem of a non-autonomous system at the left boundary xi− 1


2
of the i-th cell. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 Riemann problem of a non-autonomous system at the left boundary xi− 1
2
of the i-th cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.6 Block-structured grid. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.7 Typical control volume in a non-orthogonal grid. . . . . . . . . . . . . . . . 30
3.8 Ghost cell Ck+1 at the boundary . . . . . . . . . . . . . . . . . . . . . . . . 35
3.9 Mass flow through the boundary . . . . . . . . . . . . . . . . . . . . . . . . 35
3.10 Numerical treatment of sudden changes in flow cross-sections: Computa-
tional domain is separated and pressure loss pv is introduced . . . . . . . . 37
3.11 Valve region and the adjacent finite volume cells: Ghost cells Ck+1,1 and
C0,2 are introduced for the numerical boundary conditions. . . . . . . . . . . 38
3.12 Finite Cells at the T-piece. Junction area is shaded . . . . . . . . . . . . . . 39
3.13 Boundary cell in the two-dimensional domain . . . . . . . . . . . . . . . . . 40
3.14 Interface between two-dimensional and one-dimensional domain. . . . . . . 41
3.15 Mass correction: Mass inside the artifical cells and mass fluxes during one
time step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.1 Shema of the Burckhardt experimental compressor . . . . . . . . . . . . . . 45


4.2 Location of the sensors inside the cylinder 1, the suction and the discharge
valve retainer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Comparison of relative pressure at pc5 and valve lift at different times ta ,
tb and tc . Case: pout = 3 bar, cs =48N/mm , xv,max =1.35 mm. . . . . . . . 47
4.4 Comparison of relative pressure at pc5 and valve lift with 2 cylinders,
1 cylinder and 1 chamber working. Case: pout = 3 bar, cs =48 N/mm,
xv,max =1.35 mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.5 Comparison of relative pressure in the valve retainer: 2 cylinders, 1 cylinder
and 1 working chamber. Case: pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm. 48
4.6 Relative Pressure at pressure sensors pc5, pc6 and pout and valve lift. Case:
pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm. . . . . . . . . . . . . . . . . . 49
4.7 Pressure difference pc5-pc1. Case: pout = 4 bar, cs =48 N/mm, xv,max =1.35
mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.8 Relative Pressure at pressure sensors pc1, pc3, pc4, pc5 and pc6. Case:
pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm. . . . . . . . . . . . . . . . . . 51

V
4.9 Relative pressure in front (pc5) and behind pressure valve (pout) . Cases:
1 chamber, pout = 4 bar, cs =48 N/mm, xv,max = 1.35 mm and 1 chamber,
pout = 1 bar, cs =48 N/mm, xvmax =1.35 mm. . . . . . . . . . . . . . . . . . 51
4.10 Difference pressure pc5-pout over valve lift for different discharge pressures
. Cases: 1 chamber, pout = 1/2/3/4 bar, cs = 48 N/mm, xv,max = 1.35 mm. 52
4.11 Comparison of valve plate motion at sensor VP1, VP2 and VP3. Case: 1
chamber, pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm. . . . . . . . . . . . 53
4.12 Measurements of valve plate motion with sensor VP1 and VP2. . . . . . . . 54
4.13 Askew valve plate: Different contributions. Case: 1 chamber, pout = 4 bar,
cs =48 N/mm, xv,max =1.35 mm. . . . . . . . . . . . . . . . . . . . . . . . . 54
4.14 Comparison of relative pressure at pc5 and valve lift. Cases: pout = 4 bar,
cs =48 N/mm and cs =80 N/mm, xv,max =1.35 mm. . . . . . . . . . . . . . . 55
4.15 Comparison of relative pressure at pc5 and valve lift. Cases: pout = 4 bar,
cs =48 N/mm, xv,max = 1.05 / 1.35 / 1.65 mm . . . . . . . . . . . . . . . . 56
4.16 Location of sensors inside cylinder . . . . . . . . . . . . . . . . . . . . . . . 57
4.17 Measurement data of absolute pressure at different sensors. . . . . . . . . . 58

5.1 Burckhardt test compressor: Comparison of absolute pressure at pc5 and


valve lift. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.2 Burckhardt test compressor: Comparison of absolute pressure at pc6 and
valve lift. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 Burckhardt test compressor: Comparison of absolute pressure at pout. . . . 61
5.4 Burckhardt test compressor: Comparison of difference pressure pc5-pc1. . . 61
5.5 Initiation of waves: Pressure distribution inside the cylinder and in the
valve retainer at the beginning of the discharge . . . . . . . . . . . . . . . . 63
5.6 Delay of valve opening due to viscosity in valve gap . . . . . . . . . . . . . . 64
5.7 Effect of pressure loss at sudden change of cross-section . . . . . . . . . . . 64
5.8 Comparison of absolute pressure at pout for different models of the pressure
chamber. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.9 Comparison of absolute pressure at pc5 and valve lift for different models
of the pressure chamber. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.10 Valve Losses at discharge valve . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.11 Impact velocity of discharge valve depending on discharge pressure . . . . . 66
5.12 Ariel test compressor: Comparison of pressure distribution at pc2 and pout 67
5.13 Ariel test compressor: Comparison of pressure distribution at pc4 and pin . 68
5.14 Ariel test compressor: Comparison of difference pressure pc2-pc4. . . . . . . 68
5.15 Variation of discharge pressure: pout = 640000 / 610000 Pa . . . . . . . . . 70
5.16 Ariel test compressor: Pressure distribution in the cylinder at various times. 72

VI
5.17 Ariel test compressor: Velocity magnitudes in the cylinder at various times. 73
5.18 Moment onto Piston . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.19 Calculating the moment onto the piston: Pressure distribution inside cylinder 73
5.20 Valve motion for different discharge valves . . . . . . . . . . . . . . . . . . . 74
5.21 Mass flux through different discharge valves . . . . . . . . . . . . . . . . . . 74
5.22 Effect of Valve Masking: Comaparison of pressure at suction side pc2 . . . 74

6.1 Comparison of different finite volume schemes: Pressure pc5 inside the
cylinder and valve motion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.2 Comparison of different finite volume schemes: Pressure difference pc6-pc5. 77
6.3 Comparison of absolute pressure at pc5 and valve lift with discharge pres-
sure of 2 bar. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

VII
List of Tables

3.1 Different one-dimensional finite volume schemes . . . . . . . . . . . . . . . . 30

4.1 Main Specifications of the Burckhardt test compressor. . . . . . . . . . . . . 45


4.2 Specifications of the valves. . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 Pressure difference at closed valve (valve starts to open) for different dis-
charge pressures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.4 Impact velocities for different maximum valve plate lifts. . . . . . . . . . . . 55
4.5 Main Specifications of the Ariel test compressor . . . . . . . . . . . . . . . . 56

5.1 Specifications of the test case . . . . . . . . . . . . . . . . . . . . . . . . . . 59

6.1 Calculation Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

A.1 Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A.2 Pressure loss coefficients for the T-piece [14] . . . . . . . . . . . . . . . . . . 84
A.3 Values for fbr [14] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

VIII
List of Symbols

A area
Av force area of the valve
c speed of sound
cF force coefficient
cm1 valve masking constant
cp specific heat capacity at constant pressure
cv specific heat capacity at constant volume
cs spring constant
C area of two-dimensional cell
d diameter of cylinder
fe1mm valve flow area when xv = 1mm
F numerical flux vector
FLF Lax-Friedrich’s numerical flux vector
FR Richtmyer numerical flux vector
f flux vector in x-direction
Fadh valve plate force due to adhesion effects
Fv total valve plate force
F second order correction of F-wave
g flux vector in y-direction
h specific enthalpy
h tensor
ht total specific enthalpy
i unit vector in x-direction
j unit vector in y-direction
J Jacobian matrix
K pressure loss coefficient
l length of piston rod
l1 initial deflection of valve spring
m mass
ṁ mass flux

IX
mv mass of valve plate
M mass per area
n speed of the crankshaft
n unit normal vector
p pressure
pin suction pressure
pout discharge pressure
pt total pressure
p0 reference pressure
Q̇ heat flux
r length of the cranklever
r eigenvector of the Jacobian matrix
R mass per length
Re Reynolds number
s source vector
S surface
T temperature
Ty moment onto piston
T matrix of right eigenvectors of the Jacobian matrix
t time
t̄ dimonsionless time composed with wave speed and diameter of cylinder
t̂ dimonsionless time composed with piston motion
u state vector
U averaged state vector
u velocity in x-direction
v velocity in y-direction
vv valve plate velocity
V volume
Vch volume of cylinder head recess
w characteristic variable
ws width of valve pocket opening
W shock-wave discontinuities
x coordinate in lateral direction, coordinate in plane parallel to cylinder head
xm0 characteristic length of valve masking
xm1 characteristic length of valve masking
xv distance between valve plate and valve seating
xv,min minimum distance between valve plate and valve seating
xv,max maximum distance between valve plate and valve seating
y coordinate in plane parallel to cylinder head
z coordinate in direction of cylinder axis
Z distance between cylinder head and piston

X
Z0 cylinder clearance, Z at dead centner
Z F-wave, numerical flux increments

Greek symbols

α coefficient for the wave decomposition


αnum numerical viscosity in x-direction
αv valve constant describing effective cross-section
β coefficient for the F-wave decomposition
βnum numerical viscosity in y-direction
βv valve constant describing effective cross-section
γ ratio of specific heats
∆t time step
∆x width of cell in x-direction
∆y width of cell in y-direction
∆pmax maximum pressure difference inside cylinder
ǫ compressor classification number
Θ quotient of pressure gradients
λ eigenvalue of the Jacobian matrix
µ dynamic viscosity
Ψ flux limiter
ρ density
ρ0 reference density
σ eigenvalue of the Jacobian matrix
φ effective flow area of the valve
ϕ angle of the crankshaft

XI
Chapter 1

Introduction

The task of a reciprocating compressor is to increase the pressure of a gas. The main
parts are a cylinder, a driven piston and passive unidirectional restrictor valves (figure
1.1). During expansion the piston movement increases the volume of the compression
chamber. When the pressure inside the cylinder drops below the suction pressure gas
enters the cylinder via the suction valves. The piston reaches the lower dead centre,
reverses its movement and therefore starts to compress the fresh gas. Now the suction
valves close and the gas undergoes an almost adiabatic compression. When the piston
approaches the top dead centre the pressure inside the compression chamber forces the
discharge valves to open and the gas is delivered to the discharge pipe. Finally, the piston
reaches the top dead centre, discharge valves close and the cycle starts again.
Smaller reciprocating compressors can be found in everyday life applications such as
refrigerators and trucks whereas this thesis addresses rather large barrel design compressors
used in the gas industry, steel industry, chemical industry, oil refineries and low density
polyethylene productions. In a barrel design compressor the valves are located alongside
the cylinder (figure 1.1).

Figure 1.1: Double-acting compressor of barrel-design.

Reciprocating compressors are often crucial and expensive systems in this fields of
application and reliability is very important. Since most of the machines are custom

1
designed or produced in small batches designing by means of prototypes can not be used.
Therefore accurate design tools play an important role. Not only power consumption and
overall flow rate are important design criteria but also valve motion and moments on the
piston, which influence the life-cycle of the the machines considerably, must be determined
accurately.
Full three-dimensional simulations of reciprocating compressors with present commer-
cial CFD-programs are the state of the art. They produce accurate and reliable results
regarding valve motion, gas flow and heat transfer [22]. However, they cannot meet the
demands of engineers for usability and short computation times [22].
Common simplified approaches with short computation times are based on the works
done by Costagliola [5]. In this model the change of state inside the cylinder is assumed
to be quasi static and isentropic. Thus the thermodynamic state (pressure, temperature,
density) is taken to be uniform within the cylinder. The valve motion is determined by
the pressure difference across the valve. Since the valve is responsible for most of the
breakdowns during a lifecylce of a compressor [25] special attention has to be paid to the
valve motion. Moreover valve manufacturers have identified the impact velocity of the
valve plate onto the catcher as one of the most important design criteria. Due to the
fact that wave phenomena within the cylinder are neglected the predicted valve motion
deviates considerably from the measured one and thus these simplified models fail to be a
suitable base of a design tool. In addition, moments on the piston which are caused by the
pressure waves inside the cylinder cannot be determined. Therefore critical excitations of
the piston rod cannot be predicted.
Calculating the gas flow in a compressor taking waves into account seems to be a
complicated task and a full three-dimensional simulation unavoidable. However, the results
of [22] for the cylinder show that plane waves in lateral direction dominate (figure 1.2).
Near the centre line the curvatures of the pressure surfaces are small. It turns out that a
quasi one-dimensional model of the gas flow in the compressor, coupled to a simple valve
model is sufficient to describe the main effects in a compressor with two valves. Namely
the interaction of the plane waves with the valve dynamics. Compressors with more than
two valves require a quasi two-dimensional approach in order to account for the multiple
valve pockets, where the gas enters and leaves the cylinder.

Figure 1.2: Results of a full three-dimensional simulation: pressure distribution at the


cylinder head and in the symmetry plane [22].

Studies of E. Machu [19] give a first insight of the effect of waves and unsteady flow

2
inside the cylinder. He calculates simple waves by using the method of characteristics.
In his paper it is shown that calculated impact speed is considerably reduced, compared
to Costagliola models, by taking the waves into account. However, only the initial stages
of the outflow phase can be captured well. Moreover this model does not account for
geometrical effects. Later G. Machu [20], [21] introduced a model based on the one di-
mensional Euler equations, which is capable of describing laterally running waves in the
compressor and their interaction with the valve dynamics. This model provides the basis
for the present work.
In Chapter 2 different models for the interaction of the gas flow and the valve dy-
namics are described. Depending on the description of the geometry three-, two- and
one-dimensional models are considered. It will be discussed how a complicated three-
dimensional geometry is mapped onto a quasi one- or two-dimensional geometry. Special
care is taken selecting the computational domain and appropriate boundary conditions.
The main advantage of quasi one- and two- dimensional models is that the computational
domains can be kept constant although the flow domain inside the cylinder changes due to
the piston motion. Furthermore we derive the governing equations for the gas flow from
the Euler equations by integrating over a cross-section or cell. The resulting system of
hyperbolic equations with source terms can be written in the form

∂u ∂f (u, x, y) ∂g(u, x, y)
+ + = s(u, x, y) . (1.1)
∂t ∂x ∂y
where u stands for the state vector. The flux vector f and g and the source vector
s depend not only on u, but also on the space coordinates x and y. In case of one-
dimensional models the dependencies on y and the source terms s can be eliminated. The
system of equations (1.1) is supplemented by adequate boundary conditions at the valves,
the walls, the inlets and the outlets. T-pieces and interfaces between one-dimensional and
two-dimensional computational domains are treated as special boundaries with additional
equations. Finally the description of the valve dynamics and the flow through the valves
complete the mathematical model.
In Chapter 3 we address the numerical solution of the governing equations. The focus
lies on the discussion of finite volume schemes for the one-dimensional and two-dimensional
gas flow equations. Here difficulties arise from the space varying flux vectors and due to
the fact that in (1.1) the source terms s become infinite in case of jumps in cross-sections or
cell heights. Different approaches for this problem and their advantages and disadvantages
are discussed. In addition the boundary conditions as well as the T-piece and interface
conditions are implemented into the numerical scheme. Finally the numerical treatment
of the motion equation for the valve plate is presented.
Chapter 4 is concerned with the measurements of two test compressors at Burckhardt
Compression AG, Switzerland and Ariel Corporation, USA. Both compressors are double-
acting barrel design compressors. However, the Burckhardt compressor has two valves
and is about three times smaller than the compressor from Ariel with eight valves. The
results of these measurements give good hints about the interaction of pressure waves and
valve motion in a reciprocating compressor.
In Chapter 5 the quasi one-dimensional and two-dimensional simulations are compared
to selected test cases of the measurements. It will be shown that the models are capable of
capturing the most important physical effects. Moreover they predict accurately not only
gas flow and pressure distribution inside the cylinder but also important design criteria of

3
the valves such as impact velocity of the valve plate and valve losses.
Chapter 6 is concerned with comparison of different numerical schemes and their so-
lution. Computational time, limitations and accuracy of the different approaches are
discussed.

4
Chapter 2

Modelling of the Reciprocating


Compressor

In this chapter the mathematical model of a reciprocating compressor is discussed. Start-


ing from a complicated three-dimensional geometry of a compressor the one-dimensional
model is derived. In order to include the effects of multiple discharge and suction valves
the gas flow inside the cylinder is also modelled quasi two-dimensionally. Integrated Euler
equations form the governing equations for the gas flow and different boundary condi-
tions complete the mathematical model. A major part of this chapter is dedicated to the
valve description which follows and extends the ideas of Costagliola [5]. In addition piston
motion and heat transfer are addressed.

2.1 Geometry

2.1.1 Simulation domain

The simulation domain covers not only the working chamber (consitsting of the cylinder,
the valve pockets and the cylinder head recesses), but also the valve retainers and pressure
chambers (figure 2.1). Since measurements show that the pressure variations in the suction
valve retainers are negligible small and thus pressure and temperature are constant, the
inlet boundary of the simulation domain is set right before the suction valve. However, at
the discharge side the conditions are much more complicated. Due to high mass outflow
rates during discharge, pressure waves with high amplitudes are running back and forth
through the valve retainer, the pressure chamber and the discharge pipes and thus influence
the valve motion. Therefore the outlet boundary must be set at a sufficient distance to
the valve retainer, where the pressure is almost constant. This distance depends strongly
on the geometry of the pressure chamber and the discharge pipe.

2.1.2 One-dimensional model

In the quasi one-dimensional model the wave propagation along the diameter (x-axis) of
the cylinder from the suction to the pressure side is considered (figure 2.2). The Euler
equations are integrated over a cross section A(x, t) perpendicular to the x-axis. The

5
Valve Cylinder Head Boundary
Pocket Recess Outlet

Boundary
Valve
Inlet Retainer

z
y
Cylinder x

Piston
Suction
Valve Discharge Pressure
Valve Chamber

Figure 2.1: Flow geometry inside the compressor.

effective cross sections of the quasi one-dimensional model are displayed schematically
in figure 2.2. Starting from left we have the suction valve followed by suction valve
pocket, cylinder, discharge valve pocket, discharge valve and valve retainer. In the case of
compressors with more than two valves the suction valves and discharge valves are replaced
by one adequate suction and discharge valve, respectively. Due to the piston motion the
cross-section A(x, t) inside the cylinder varies with time. Figure 2.3 shows the appropriate
area functions depending on x at different times t1 and t2 . At t2 the piston masks the
valve pocket in a way that the effective cross section area A(x, t) is discontinuous there.
A(x, t)

cylinder t1
valve pocket
A(x, t)
valve
retainer
Z

t2
x
0
suction discharge suction cylinder discharge valve
valve valve valve valve retainer

Figure 2.2: Schema of the one-dimensional Figure 2.3: Cross-section areas of the one-
model. dimensional model.

Several approaches are used to model the discharge side of the compressor.

6
• The first model has been used by Machu [20] (figure 2.4). Here, the pressure chamber
is added to the subdomain of the valve retainer and the outlet boundary is set at the
end of the valve retainer. A pressure loss at the outlet is specified to simulate the
hindered flow of the valve retainer to the pressure chamber. Moreover, a reduced
outflow area causes reflections of the pressure waves at the end of the valve retainer.
• In the second approach the complicated geometry of the pressure chamber is replaced
by a cylinder connected to the valve retainer using a T-piece (figure 2.5). Again the
pressure is assumed to be constant at the end of the pressure chamber.
• The third approach is used when effects of the discharge pipe and damper are taken
into account. Here the outlet boundary is located at the end of the damper (figure
2.6). In case of double-acting compressors the pressure chamber links the valve
retainer of the crank end (CE) to the one of the head end (HE). Again we use a
T-piece to connect the discharge pipe to the pressure chamber.

It must be pointed out that reducing the pressure chamber to a simple cylinder may be
an oversimplification in some cases.

Valve Valve Retainer


Valve Pocket
Valve Outlet Boundary
Valve Pocket T-piece

Cylinder

Cylinder
Piston

Piston Valve Retainer

Pressure Chamber Outlet Boundary

Figure 2.4: Model of discharge system Figure 2.5: Model of discharge system
VR1: Pressure chamber is modelled at the VR2: Pressure chamber is connected to
end of the valve retainer. the valve retainer by a T-piece.

2.1.3 Two-dimensional model

The two-dimensional model takes the wave propagation in a plane parallel to the cylinder
head (x,y-plane) into account. The equations of motion (Euler equations) are integrated
over the height z(x, y, t) of the cylinder. In terms of the compressor the height is the
distance between the piston and the cylinder head. The two-dimensional computational
domain is displayed schematically in figure 2.7.
In order to keep the model as simple as possible and hence least time consuming (in
terms of computational time) a one-dimensional approach is used for the valve pockets
and valve retainers. The coupling between the quasi one-dimensional computational sub-
domain and the quasi two-dimensional one is described by the interface conditions, see

7
Valve Valve Retainer
Valve Pocket

T-piece

Cylinder
head end
Pressure
Chamber Damper
Piston Outlet
T-piece Discharge Pipe Boundary

Cylinder
crank end
T-piece

Figure 2.6: Model with discharge pipe and damper. Crank end working chamber is op-
tional (dashed lines).

section 2.4.5. A typical arrangement for a compressor with eight valves is shown in figure
2.8.

2.2 Characteristic Data

In this section the basic assumptions about the gas flow and heat transfer in a compressor
are stated. They are based on measurements, full three-dimensional CFD-simulations and
experiences of compressor manufacturers.

2.2.1 Gas Flow

For each phase of the compression cycle a Reynolds number characterises the gas flow.
During discharge and intake the maximum gas velocity, the actual distance between cylin-
der head and piston and the actual viscosity are used as the reference values. According
to full three-dimensional CFD-simulations [3], [22] the Reynolds numbers range from 104
to 105 . During compression and expansion the mean piston velocity, the diameter of the
cylinder and the actual viscosity yields a Reynolds number in the order of magnitude of
104 . Comparisons with critical Reynolds numbers leads to the assumption that in the
cylinder turbulent flow occurs during the whole cycle. Furthermore the high Reynolds
numbers allow the use of inviscid gas flow models. Although these results are true for the
test cases calculated (see tables 4.1 and 4.5, the assumptions can be extended to most of
the gases compressed.

8
valve pocket
one-dimensional Valve
Cylinder Interface

cell valve pocket


Valve
retainer
y

Z
x
Suction Discharge Cylinder
Valve Valve two-dimensional

Figure 2.7: Schema of the two- Figure 2.8: Schema of the two-
dimensional model. dimensional model with eight valves.

2.2.2 Heat Transfer and Dissipation

In turbulent flow the heat transfer from the gas to the surrounding structure is determined
by local flow properties only [4], [9]. However, if we assume inviscid flow and Prandtl num-
bers in the order of 1 then an almost adiabatic change of condition can be expected. In
other words, the gas flow is not influenced by the heat transfer from and to the surround-
ing material. In addition for small Eckert numbers compared to the Reynolds number
(EC ≪ Re) the dissipation can be neglected [4]. During compression and expansion the
mean piston velocity, the specific heat capacity for constant pressure and the temperature
difference between suction pipe and discharge pipe yields a Eckert number in the order of
magnitude of 10− 4. During discharge and intake the maximum gas velocity, the specific
heat capacity for constant pressure and the temperature difference across the cylinder are
used as the reference values. According to full three-dimensional CFD-simulations [3], [22]
the Eckert numbers of the test compressors are in the order of magnitude of 1. Hence, heat
transfer and dissipation are neglected. In fact the essential part of temperature changes
of the gas are due to compression and expansion.
When the dissipation and heat transfer are very small, the Euler equations [9] can be
used. However, at sudden changes of flow cross-sections pressure losses occur and this
problem is addressed in section 2.4.6.

2.2.3 Classification of Compressors

The sudden opening of the discharge valves excites a rarefaction wave. Considering differ-
ent time scales allows to estimate the response to this excitation. Moreover, this allows a
classification of compressors whether or not pressure waves play an important role. Dur-
ing discharge two different characteristic time scales can be identified. On one hand we
have the time t̄ a wave needs to travel through the cylinder, defined by the diameter of
the cylinder D and the wave speed at reference state c0 . On the other hand we have the
time t̃ specified by the parameters of the piston motion, more precisely the distance of the

9
piston to the cylinder head Z0 and the acceleration of the piston Z̈0 in dead centre.
t t
t̄ = D
, t̃ = q (2.1)
Z0
c0
Z̈0

Now for the non-dimensional characteristic parameter (classification number) ǫ = t̃/t̄/ we


can distinguish three different orders of magnitude. If ǫ ≪ 1 then the speed of the waves
is much faster than the piston motion and an incompressible outflow of the compressor
occurs. However, ǫ of typical compressors are of order 1. Therefore compressibility effects
must be taken into account. ǫ ≫ 1 holds true in the case of heavy gases and very fast
running compressors.

2.3 Piston Motion

The distance between the piston and the cylinder head Z depending on time t can be
determined using the properties of the crank mechanism. We use the length of the crank-
lever r, the length of the piston rod l, the smallest distance between piston and cylinder
top Z0 and the present angle of the crankshaft ϕ to obtain
r  r 2
Z(t) = r + l + Z0 − r cos ϕ(t) − l 1 − sin2 ϕ(t) , (2.2)
l
where φ(t) = 2πnt. Here, n denotes the speed of the crankshaft. As a first approximation
for rl ≪ 1 (2.2) reduces to
Z(t) = r + Z0 − r cos φ(t) . (2.3)

2.4 Gas Flow inside the Compressor

2.4.1 Quasi one-dimensional Model

The governing equations for the gas flow depending on time t and the space coordinate
x are obtained by taking the mass, momentum and energy balance over a cross section.
The system of equations is written as
∂ ∂
[ρA] + [ρuA] = 0 , (2.4)
∂t ∂x
∂ ∂  2  ∂A
[ρuA] + ρu A + pA = p , (2.5)
∂t ∂x ∂x
u2 u2
      
∂ ∂
ρA cv T + + ρuA cv T + + uAp = 0 , (2.6)
∂t 2 ∂x 2

where the variables p, ρ, T and u have their usual meaning, pressure, density, temperature
and velocity in x-direction, respectively. The specific heat capacity for constant volume
and for constant pressure are denoted by cv and cp respectively. In case of smooth solutions
the energy equations (2.6) can be replaced by

∂ ∂
(ρs) + (ρus) = 0 , (2.7)
∂t ∂x

10
thus allowing isentropic solutions (s = const.). Assuming isentropic solutions we can
express the pressure p in terms of the density and entropy

p = p(ρ, s) . (2.8)

In case of constant heat capacities following relation is obtained

pρ−γ = const. , (2.9)

with γ = cp /cv . Smooth solutions are the case if no shocks or flow separations occur.
Measurements show that the first condition is satisfied. However the second one is violated
at sudden changes of cross-sections, sharp turns of the flow direction, and T-pieces. At
this point we accept this violation and assume that the associtated entropy productions
are sufficiently small. In section 2.4.6 we present a correction of this problem.
With an adequate reference state, an associated reference pressure p0 and reference
density ρ0 , (2.9) can be rewritten as

pρ−γ = p0 ρ0 −γ . (2.10)

We observe that (2.5) is not a homogenous conservation law [28]. The right hand side
of (2.5) constitutes a momentum source of the flow due to a variation of the cross section.
Since the cross section A may vary rapidly at the transition from the cylinder to the valve
pocket, the derivative dA/dx may become large causing numerical problems. For smooth
solutions these problems can be avoided by transforming the momentum equation (2.5) to

∂ u2
 
∂u 1 ∂p
+ + =0. (2.11)
∂t ∂x 2 ρ ∂x

Furthermore, using the definition (2.8) and assuming constant entropy yields

∂ u2
 
∂u 1 ∂p ∂ρ
+ + =0. (2.12)
∂t ∂x 2 ρ ∂ρ ∂x
q
With the definition of the speed of sound c = ( ∂p
∂ρ )s we obtain

Zρ 2
 
∂u ∂  u2 c (ρ̃, s0 ) 
+ + dρ̃ = 0 . (2.13)
∂t ∂x 2 ρ̃
ρ0

In case of constant specific heat capacities equation (2.13) reduces to


 
∂u ∂ 1 2 γ p
+ u + =0. (2.14)
∂t ∂x 2 γ−1ρ

From now on we will consider constant specific heat capacities only, since the extension
to the general case with heat capacities depending on the temperature of the gas causes
no difficulties.
Finally the system of governing equations can be written in conservative form:

∂u ∂f (u, x)
+ =0. (2.15)
∂t ∂x

11
Here the state vector u and the flux function f (u) can be given by
! !
R Ru
u= , f (u) = u2 p0 R γ−1
 , (2.16)
u 2 + ρ0 γ A

where R = ρA. Note that 2.16 does not correspond to a physical conservation law since u
is no conservation variable.

Local Properties of the system equations In case of constant cross-sections A the


flux vector in (2.15) depends on the state vector only and the system of equations reduces
to
∂u ∂f (u)
+ =0. (2.17)
∂t ∂x
Furthermore, the linearisation around an arbitrary fixed state u0 yields
∂u ∂u
+ J(u0 ) =0, (2.18)
∂t ∂x
where J(u0 ) is the Jacobian matrix of the flux function f (u) and given by
!
u R
J(u) = c2 . (2.19)
R u
Here, c denotes the speed of sound. The eigenvalues λ1 , λ2 of J are
λ1 = u − c ,
λ2 = u + c . (2.20)
The matrix T of right eigenvectors of the Jacobian J is
!
− Rc Rc
T = (r1 , r2 ) = . (2.21)
1 1

We introduce the characteristic variables w = (w1 (u(x, t)), w2 (u(x, t)))T defined by
w := T−1
0 u, T−1 −1
0 = T (u0 ) . (2.22)
Here w1 and w2 are given by
1
w1 = (u − c) , (2.23)
2
1
w2 = (u + c) . (2.24)
2
Finally 2.18 can be transformed to a decoupled system of linear partial differential equa-
tions in terms of the characteristic variables
∂w1 ∂w1
+ λ1 (u0 ) =0, (2.25)
∂t ∂x
∂w2 ∂w2
+ λ2 (u0 ) =0. (2.26)
∂t ∂x
In section 3.2 we will use these results, especially the characteristic variables and their
corresponding partial differential equations, in order to derive the numerical formulations
of different boundary conditions. Moreover the wave propagation structure described by
the characteristics will be incorporated in the finite volume scheme F-wave (see section
3.1.1).

12
2.4.2 Quasi two-dimensional Model

The governing equations for the flow of gas are obtained by taking the mass and momentum
balance over a cell. In addition to the one-dimensional approach v denotes the velocity
in y-direction and the momentum equation is written for the x-direction and y-direction,
respectively.
∂ ∂ ∂
[ρZ] + [ρZu] + [ρZv] = 0 , (2.27)
∂t ∂x ∂y
∂ ∂  ∂ ∂Z
ρZu2 + pZ +

[ρZu] + [ρZuv] = p , (2.28)
∂t ∂x ∂y ∂x
∂ ∂ ∂ ∂Z
[ρZv] + [ρZvv + pZ] + [ρZuv] = p , (2.29)
∂t ∂y ∂x ∂y
where Z denotes the distance between the piston and the cylinder head.
Assuming isentropic flow conditions we replace the energy equation by the isetropic
equation of state (2.9). Note that equations (2.28) and (2.29) are not of conservation form.
The right hand terms constitute a momentum source of the flow due to the variation of the
height inside the cylinder. However, only smooth cylinder head recesses can be considered
since the gradients of Z become very large in case of sharp changes.
The governing equations (2.27)-(2.29) reduce to the one-dimensional model equations
by replacing the height Z with the cross-section A and setting the velocity in y-direction
v to zero.
For the numerical analysis it is useful to write the continuity equation (2.27) and both
equations of motion (2.28) and (2.29) in the following form:

∂u ∂f (u) ∂g(u)
+ + =s. (2.30)
∂t ∂x ∂y
Here the the state vector u and the analytical flux vectors f and g are given by
     
M Ix Iy
 Ix 2 p0 M γ   Ix Iy 
u =  Ix  ,  M + ργ0 Z γ−1  ,
f (u) =  g(u) = 
    , (2.31)
 2 M 
Ix Iy Iy p0 M γ
Iy M M + γ
ρ Z γ−1
0

where M = ρZ, Ix = ρZu and Iy = ρZv. The source vector s can be written as
 
0
s = p ∂Z . (2.32)
 
∂x 
p ∂Z
∂y

When Z is constant 2.32 becomes s = (0, 0, 0)T . This is true for compressors with plane
cylinder heads and pistons.

2.4.3 Boundary Conditions

The computational domain of the compressor is divided into the subdomains of the cylin-
der, the valve retainers, and the pressure chamber. In addition, if the flow cross section

13
in the cylinder changes abruptly the cylinder is divided into two further computational
subdomains. On each subdomain the governing equations for the gas flow have to be
supplemented with boundary conditions or interface conditions to adjacent subdomains.
In general we can distinguish two types of boundary conditions needed: on the one hand
impermeable walls such as closed valves, or the cylinder wall in the two-dimensional ap-
proach and on the other hand permeable boundaries where mass flows enter and leave the
computational domain (flow boundaries).

Walls On walls we require that the velocity and the momentum normal to the boundary
is zero.

Flow Boundaries In order to obtain adequate boundary conditions we must consider


characteristics leaving and entering the computational domain. A necessary condition
for a well posed problem is that the boundary condition in terms of physical variables
do not lead to prescription of a characteristic variable associated with a characteristic
leaving the computational subdomain under consideration. In other words, the number
of boundary conditions should be equal to the number of characteristics pointing into the
region at a boundary. Considering linearised system (2.18) and assuming subsonic flow the
signs of the eigenvalues λ1 , λ2 defined in (2.20) remain constant. Hence the characteristic
corresponding to the negative eigenvalue λ1 always points to the left whereas the second
one always points to the right. Figure 2.9 shows the boundary-value problem on the right
side of a computational domain. Wibmer [31] has shown, that a well posed problem is
obtained, if one and only one of the variables pressure, density, mass flow or velocity is
prescribed (due to one entering characteristic).

right boundary

w1
x
w2

computational domain

Figure 2.9: Right boundary of computational domain: Characteristics variables entering


and leaving at the right end of the computational domain.

2.4.4 T-piece

In the valve retainer the mass flow coming from the cylinder is divided in the one that
runs further downstream and the one that enters the pressure chamber of the compressor.
This junction, or so called T-piece, can be described by a simple one-dimensional model.
However at this point we have to neglect effects from different branching angles, pressure
losses, and flow separations. These issues are subject to section 2.4.6 where pressure losses

14
for T-pieces are introduced. Furthermore, in case of double-acting cylinders this model
can be applied to the T-piece in the pressure chamber, where the mass flows from the
crank end side and head end side merge and enter the discharge pipe.

w1,1 w2,2
x1 x2
w2,1 w1,2

1 w1,3 2
w2,3
h1 = h2 = h3
x3 ṁ1 + ṁ2 + ṁ3 = 0
3

Figure 2.10: T-piece consisting of tube 1, tube 2 and tube 3: The shaded square displays
the junction area. Its appropriate in-going and out-going characteristics w1,i and w2,i are
indexed by the adjacent tube i = 1, 2, 3.

Now the T-piece itself is reduced to a boundary condition problem for the three pipes.
Similar to section 2.4.3 the number of conditions must be equal to the number of char-
acteristics pointing into the computational domains of the pipes. Figure 2.10 shows the
characteristics entering and leaving the shaded junction area. We assume that the junction
itself has no volume, therefore the net mass flow ṁ entering and leaving the junction is
zero. Furthermore we require that the net enthalpy flux at the T-piece is also zero. Now
using the mass balance for the T-piece and assuming that the entropy is constant in all
three pipes yield that the total specific enthalpy ht on each boundary must be equal. The
three analytical boundary conditions can be written as

ṁ1 + ṁ2 + ṁ3 = 0 , (2.33)


ht1 = ht2 = ht3 , (2.34)

where
u2
ht = h +. (2.35)
2
The numerical treatment of the T-piece is subject of section 3.2.

2.4.5 Interface Conditions

The connection between one-dimensional and two-dimensional areas are called interfaces
(figure 2.4.5). We assume that the interface has no volume and thus the mass flux leaving
the cylinder via the interface must enter the computational domain of the valve pocket and
vice versa. Furthermore conservation of energy across the interface is required resulting
in a parity of enthalpies on each face. Similar to a T-piece with only two branches the
equations for the interfaces can be written as

ṁ1d + ṁ2d = 0 , (2.36)


ht1 = ht2 , (2.37)

15
where ṁ1d and ṁ2d denotes the mass flux from cylinder and valve pocket to the interface,
respectively.
Interface

Cylinder Valve pocket


Governing equations (2.15) Governing equations (2.30)

Face 2d Face 1d

Figure 2.11: Interface between one-dimensional and two-dimensional domain.

2.4.6 Pressure Losses

At sudden changes in cross-section and in T-pieces pressure losses occur. Although in


the quasi one-dimensional gas flow model the energy equation is not solved and isentropic
flow is assumed, the pressure losses can be taken into account by means of simplified
models. However, the governing equations for the gas flow (2.15) are not valid any more.
Moreover at this locations the computational domain must be split up and new conditions
connecting the resulting subdomains must be chosen. Similar to section 2.4.3 the number
of connecting conditions must be equal to the number of characteristics pointing into the
computational subdomains.
Usually pressure losses in adiabatic flows are associated with conservation of energy
and an entropy production. However in the present model constant entropy is assumed.
Therefore we have to drop the requirement for adiabatic flow and compensate the entropy
production by an artificial heat flux Q̇12 across the side walls.
For the transition from state 1 in front of the pressure loss (p1 , ρ1 , u1 ) to state 2 (p2 ,
ρ2 , u2 ) we require conservation of mass and entropy (figure 2.12):

ṁ = ṁ1 = ṁ2 , (2.38)


ṡ = ṁ1 s1 = ṁ2 s2 , (2.39)

and thus 2.39 can be replaced by s1 = s2 . In addition the losses pv in total pressure pt are
given by
pt1 = pt2 + pv . (2.40)
The total pressure is the pressure at the thermodynamic state that would exist if the gas
was brought to zero velocity at constant entropy. For a compressible gas with constant
specific heat capacities using the isentropic equation of state (2.9) we can write
γ
γ − 1 ρ u2
  γ−1
pt = p 1 + . (2.41)
γ p 2

16
In case of incompressible flows losses in total pressure are described by loss coefficients
K, which can be written as
pv
K= 2 , (2.42)
ρu /2
where ρu2 /2 denotes the velocity pressure. According to Miller [23] the incompressible
loss coefficients are suitable for compressible flows as well if the losses are related to the
dynamic pressure pt − p instead of the velocity pressure. For compressible flows equation
(2.42) takes the form
pt − pt2
K= 1 , (2.43)
pt1 − p1
However, Miller points out that in case of separated flows the application of these coeffi-
cients may not be very satisfactorily, since choked flow is neglected.

Sudden Expansions in Cross-section In consequence of a sudden expansion in of the


flow cross-section A(x, t) separation zones arise (figure 2.13). In this area the velocities
are rather small and the pressure is equal to p1 . Further downstream the inflowing stream
mixes with the surrounding gas. The incompressible pressure loss coefficient for a sudden
expansions is given by
K = (1 − A1 /A2 )2 , (2.44)
where A1 is the inlet area and A2 the outlet area. In case of infinite outlet areas K becomes
1 and thus (2.43) reduces to
p1 = pt2 = p2 . (2.45)

Pressure Loss
p1 K p2
Dead water zones
u1 u2
ρ1 ρ2 A2
A1 p2
p1
u1 u2
ρ1
Q̇12 ρ2

Figure 2.12: Modelling pressure losses Figure 2.13: Sudden expansion of cross-
section

In order to implement the pressure losses at sudden expansions in cross-section a model


as described above can be employed resulting in computing the quasi-stationary system
of equations (2.38) and (2.40), see section 3.2.4.

Pressure losses at T-pieces Starting from the loss-free T-piece (section 2.4.4) each
pipe is provided with an pressure loss (figure 2.14). The thermodynamic states adjacent to
the junction are marked with stars. The pressure loss coefficients K1 , K2 and K3 depend
on the flow direction of the single branches, the cross-section areas A and the mass fluxes.
They are derived from the incompressible loss coefficients of Idelchik [14] and given in the

17
appendix for important cases. Even the effects of branching angles and separation zones
can be easily incorporated by adjusting the loss coefficients.
We employ the equations of the T-piece (2.33) and (2.34) to connect each branch and
the pressure loss equations (2.38) and (2.43) to obtain

3
X pti − p∗ti
ṁ∗i = 0 , Ki = ,
pti − pi
i
h∗t1 = h∗t2 , ui ρi = u∗i ρ∗i ,
h∗t2 = h∗t3 , pi ρ−γ
i = p∗i ρ∗i −γ . (2.46)

where i is the subscript of the branch running from 1 to 3.

p1 p∗1 p∗2 p2
u1 u∗1 u∗2 u2
ρ1 K1 ρ∗1 ρ∗2 K2 ρ2
A1 A2

p∗3 , u∗3 , ρ∗3

K3

p 3 , u 3 , ρ3

A3

Figure 2.14: Pressure losses at the T-piece

2.5 Valve Model

2.5.1 Valve Dynamic

The discharge and the suction valves are located adjacent to the circular lateral surfaces
of the valve pockets. Both valves are unidirectional restrictor valves, plate valves or ring
valves are usually used. In this work we concentrate on plate valves but most of the
equations can be easily adopted to any kind of restrictor valve. Figure (2.15) shows the
basic function of a plate valve. The state of a valve is specified by the distance between
the valve plate and the seating xv . The motion of the valve plate is determined by the
forces acting on it. We consider the following three contributions to the resulting force:
the pressure difference across the valve acting on an effective force area Av of the valve

18
Catcher
p2
xv Valve Springing

Valve Plate

Seating
sv
p1

Figure 2.15: Schema of a plate valve.

plate, the springing and thirdly a contribution due to viscous forces in the initial stages
of valve opening. Denoting the pressure in front of the valve p1 and behind the valve p2
we obtain the equation of motion for the valve plate

mv ẍv = (p1 − p2 )Av − cs (xv + l1 ) − Fadh , (2.47)

Here, mv stands for the mass of the valve plate. The constants cs and l1 denote spring
constant and initial deflection of the spring, respectively. An initial sticking effect is
modelled by the force Fadh . It is caused by the viscosity µ of the gas or oil in the valve gap
resulting in a small time delay when the valve is opening. We follow the ideas of Reynolds
lubrication theory to obtain
1 dxv
Fadh = µsv 3 dv 3 . (2.48)
xv dt
Here sv and dv stand for the length and the total depth of the valve gap. A detailed
description and derivation can be found in [7] and [8]. The motion of the valve plate
is limited by the seating on one side and by the catcher on the other side. Therefore
the deflection of the plate xv ranges from xv,min to xv,max . Because of the roughness of
the seating and valve plate it is necessary that xv,min > 0 [8]. Due to the fact that the
flow within the valve changes with xv , and thus the effective force area Av changes, we
introduce a force coefficient cF . The force area Av can be written as

Av = Av,0 cF , (2.49)

where Av,0 stands for the force area at closed valve. For common plate valves the relation
between the force coefficient and valve lift takes the form
 
xv
cF = 1 − 0.2 . (2.50)
xv,max

2.5.2 Flow through the valve

The flow through the valves is considered as (quasi stationary) outflow of gas from a
pressurised vessel through a convergent nozzle. In case of constant heat capacities it is

19
given by St.Venant and Wantzell [32]
v
 1 u   γ−1 !
p2 γ u t 2γ pt 1 p2 γ
ṁ = φρ1 1− , (2.51)
pt 1 γ − 1 ρ1 pt 1

where pt 1 is the total pressure before and p2 is the pressures after the valve, respectively.
The effective flow cross section φ of the valve is assumed to be a function of the position
of the valve plate xv only. This is not true if effects of asymmetric flow conditions in the
valve pocket are taken into account (section 2.5.3). For common valves the equation for
the effective cross section takes the form
s
(fe1mm xv )2
φ(xv ) = . (2.52)
αv + βv xv 2

The constant fe1mm denotes the valve flow cross-section when the valve plate is 1 mm
above the valve seating. The constants αv and βv describe non-linear dependencies of the
valve plate lift and have to be determined empirically.

2.5.3 Asymmetric Flow in Valve Pockets, Valve Masking and Slotted


Valve Pockets

Valve masking occurs when the piston approaches top dead centre (the distance between
cylinder head and piston becomes very small) and a part of the valve pocket is masked
by the piston itself, see figure 2.17. Two effects take place. Firstly, the sudden change of
cross-section areas produce pressure losses due to separation. This problem is discussed in
detail in section 2.4.6. Secondly, if the distance between piston edge and valve is so small
that the separation zone extends across the whole valve pocket length then the outflowing
gas will use only a part of the valve flow cross-section. Similar effects take place in slotted
valve pocket entries where the effective flow cross-section of the valve is reduced due to
the separation zones (figure 2.18). However, full three-dimensional simulations show that
even in open valve pockets without valve masking asymmetric flow and separation zones
can occur. In figure 2.16 the simulation results of an academic compressor are displayed
[24]. In order to take above effects into account we assume that the valve flow area is not
only a function of valve plate lift xv but also of the piston position Z,
s
(fe1mm xv )2
φ(xv , h) = fmask (h) . (2.53)
αv + βv xv 2

Here fmask denotes a empirical function dependent on Z. In case of valve masking the
following approximation is used:
−(Z−xm0 )
fmask (h) = 1 − cm1 e xm1
. (2.54)

where the constants cm1 , xm0 , and xm1 must be matched to measurement data.

20
5.80e+06
5.79e+06
5.78e+06
valve pocket
5.76e+06
5.75e+06
5.74e+06
5.72e+06 cylinder
5.71e+06
5.70e+06
5.69e+06
5.68e+06
5.66e+06
5.65e+06
5.64e+06
5.62e+06
5.61e+06
5.60e+06
discharge
5.59e+06 piston valve
5.58e+06
5.56e+06 Y X
5.55e+06
Z

Contours of Total Pressure (pascal) (Time=7.8333e-02) Mar 14, 2007


Crank Angle=654.00(deg) FLUENT 6.2 (3d, dp, segregated, dynamesh, ske, unsteady)

Figure 2.16: Pressure distribution and velocity vectors in the symmetry plane during
discharge (full three-dimensional simulation).

cylinder valve pocket valve pocket


cylinder

valve valve
retainer retainer
piston
piston
valve valve
separation separation
zone zone

Figure 2.17: Valve Masking. Figure 2.18: Slotted Valve Pockets.

21
Chapter 3

Numerical Method

In the first part of this chapter we present the treatment of the hyperbolic Euler equa-
tions by finite volume schemes, following [16], [17], [13], [29], [6] and [30] . Here attention
has to be turned to the fact that the flux functions in the governing systems of equa-
tions (2.15) and (2.30) are not only a function of the state vector but also a function of
space coordinates x and y. Therefore not every finite volume method is suitable. For
the one-dimensional conservation law the finite volume methods by LeVeque [16],[17] and
the approach of MacCormack [13],[17], [29] are employed. Both methods are second order
accurate in time and space. However LeVeques flux splitting scheme has the advantage
that boundary conditions, where calculating characteristics is required, can be easily in-
corporated in the scheme. The two-dimensional Euler equations in the cylinder are solved
by Lax-Friedrich’s [13] finite volume scheme of first order and the second order method of
Lax-Wendroff [15]. In case of discontinuities the performance of the second order approach
is rather poor and thus a flux limiter is employed.
The boundary conditions are subject of the second part of this chapter. We distinguish
between physical boundary conditions derived in section 2.4.3 and numerical boundary
conditions required for computation. The physical boundary conditions almost never fully
determine the set of dependent variables of the underlying system of differential equations,
whereas the numerical methods used require all dependent variables at the boundaries.
In this work two different approaches are discussed and employed. The first method
extrapolates the interior state to an additional cell (ghost cell) outside the computational
domain. The second method adds the characteristic equation (2.25) corresponding to the
outgoing characteristic to the imposed physical boundary conditions.
Finally, the integration of the motion equation of the valve plate (2.47) is discussed.

3.1 Finite Volume Schemes

3.1.1 One-dimensional Finite Volume Schemes

Each spatial subdomain is divided into intervals Ci (finite volumes, cells) of constant length
∆x with midpoint xi (figure 3.1). Denote the i-th cell by
 
Ci = xi− 1 , xi+ 1 , (3.1)
2 2

22
Cell
Cell Cell Cell
Ci−1 Cell
C0 Ci Ck
Ci+1
x0 xi−1 xi xi+1 xk

∆x

Figure 3.1: Subdivision of the computational domain into grid cells Ci .

1
where the cell boundary on the left side is located at xi− 1 = 2 (xi + xi−1 ) and on the
2
1
right side at xi+ 1 = (xi + xi+1 ). Note that in a quasi one-dimensional problem, the
2
2
cross-section areas and thus the ”physical” volume of the cells can vary along x.
The actual derivation of the governing equations (quasi one-dimensional Euler equa-
tions) is based on balances over control volumes. It seems natural to return to this formu-
lation, since subdomains are divided into finite control volumes. Now the Euler equations
are satisfied in each cell, above all the conservation of the state quantities is ensured.
Moreover, the integral formulation allows discontinuous solutions (shocks), whereas the
differential formulation requires smooth solutions. Integrating the differential form of the
quasi one-dimensional Euler equation

∂u ∂f (u, A)
+ =0,
∂t ∂x
h i
over the cell Ci and one time step, xi− 1 , xi+ 1 × [tn , tn+1 ], yields the integral formulation:
2 2

xi+1/2 xi+1/2 tZn+1 tZn+1


Z Z
u(x, tn+1 )dx− u(x, tn )dx+ f (u(xi+1/2 , t), A)dt− f (u(xi−1/2 , t), A)dt = 0 .
xi−1/2 xi−1/2 tn tn
(3.2)
Note that the cross-section area A, and thus the analytical flux f , are a function of time
t and coordinate x. The first two terms in (3.2) correspond to the total change of the
state quantities inside the cell during the time step ∆t = tn+1 − tn . The third and fourth
integral represent the total fluxes of the state quantities across the boundaries xi+ 1 and
2
xi− 1 . Therefore the system of equations (3.2) simply states that every gain or loss of
2
state quantities inside the cell is due to the fluxes across the boundaries only. Moreover
the conservation of the state quantities within the whole computational subdomain will
be ensured [16].
Let Uni denote the cell average of the state quantities over the ith cell at time tn :
xi+1/2
1
Z
Uni ∼ u(x, tn )dx . (3.3)
∆x
xi−1/2

23
The flux integral of (3.2) at the boundary xi− 1 is approximated by the numerical flux
2
Fni− 1 :
2
tZn+1
1
Fni− 1 (Ui−q1 ..Ui+q2 , A) ∼ f (u(xi−1/2 , A, t))dt . (3.4)
2 ∆t
tn
The numerical flux depends on the cell averages of the state quantities Ui−q1 ..Ui+q2 and
the cross-section area A. Whereas the actual number of involved cell averages given by q1
and q2 is determined by the numerical scheme used. Finally, when employing the above
approximations the system of equations for Un+1
i of cell Ci at the time tn+1 can be written
as:
∆t  n 
Un+1
i = U n
i − F 1
i+ 2
− F n
1
i− 2
(3.5)
∆x
A numerical scheme which has the form (3.5) is called conservative since it mimics the
properties of (3.2).
Generally we require that a numerical scheme is convergent which means that the
numerical solution converges to the true solution of the differential equations as the grid
is refined. According to Leveque [16] this is true if following two conditions are satisfied:

• Firstly the method must be stable, meaning that small errors made in each time
step do not grow too fast in later time steps.
• Secondly the method must be consistent, meaning that the numerical flux ap-
proximates the analytical flux well. More precisely, for constant state quantities
U = Ū = const. and constant cross-sections A = Ā the numerical flux must be
equal to the analytical flux:
F(Ū, .., Ū, Ā) = f (Ū, Ā) . (3.6)

A detailed description and mathematical formulation of these conditions can be found in


[16], [13], [29], [6] and [30]. Later we will state the conditions under which the numerical
schemes used in this work are convergent. Now we introduce different numerical flux
functions of (3.4) resulting in different finite volume methods.

Godunov method First we consider autonomous systems, where the flux vector f is
a function of the state vector u only. In case of quasi onedimensional Euler equations
this means that the cross-sections are constant through the entire computational domain
(A = A(t)). The resulting system of equations can be written as
∂u ∂f (u)
+ =0. (3.7)
∂t ∂x
Following the ideas of Godunov [16] we decompose the original initial value problem into
a set of Riemann problems, by replacing the initial data by the cell average on every cell.
Thus the resulting initial data has discontinuities at every cell boundary and we have to
solve a Riemann problem for every cell boundary. In particular the Riemann problem
at the left boundary xi− 1 of the i-th cell consists of the conservation law (3.7) and the
2
piecewise constant initial data at the time tn
(
Uni−1 : x < xi− 1
u(x, tn ) = 2 (3.8)
Uni : x > xi− 1
2

24
with the similarity solution
x − x 
i− 12
ū(x, t) = Ûi− 1 . (3.9)
2 t
We note that ū(x, t) = Û(0) is independent of the time t. Hence the numerical flux at the
left cell boundary takes the form
 
Fni− 1 = f Ûni− 1 (0) , (3.10)
2 2

and the balance over the control volumes 3.5 can be written as
∆t h  n   i
Un+1
i = Uni − f Ûi+ 1 − f Ûni− 1 . (3.11)
∆x 2 2

Thus we only have to determine for every Riemann problem the value of the solution
along the line xi− 1 . If we choose the time step ∆t sufficient small, so that the propagating
2
discontinuities of the neighbouring Riemann problems do not reach the cell edge xi− 1
2
(figure 3.2) the intermediate state Ûi− 1 is constant along the cell edge xi− 1 in the time
2 2
step. Note that the intermediate state Ûi− 1 and thus the numerical flux Fi− 1 depends
2 2

tn+1

Uni−1 Uni Uni+1


tn x
xi− 3 xi− 1 xi+ 1 xi+ 3
2 2 2 2

Figure 3.2: Riemann problems at each cell boundary.


on the state quantities Ui and Ui−1 only. Finally the solution for the state quantities
Un+1 at time tn+1 is obtained by piecing together the Riemann problems of each cell
interface. However, this method involves the solution of nonlinear problems and can be
very time consuming. Therefore the Riemann problems will be replaced by appropriate
linear Riemann problems in the following method.

Roe method At every cell boundary xi− 1 the associated Riemann problem is replaced
2
by a linear Riemann problem
∂ Û ∂ Û
+ Âni− 1 =0, (3.12)
∂t 2 ∂x

where the locally defined flux matrix (so called Roe matrix) Âi− 1 (Ui−1 , Ui ) of the cell
2
edge xi− 1 must satisfy following conditions in order to obtain a conservative, consistent
2
finite volume method:
Âi− 1 is diagonalisable with real eigenvalues , (3.13)
2

Âi− 1 (Ui−1 − Ui ) = f (Ui−1 ) − f (Ui ) , (3.14)


2

Âi− 1 (U, U) = J(U) . (3.15)


2

25
Here J(U) is the Jacobian matrix of the flux function f (U) of (3.7). Autonomous systems
(3.7) have been studied extensively and a variety of Roe-type linearisations have been
developed. For an overview we refer to [16], [10].

1
t 2
Wi− 1 Wi− 1
2 2

Ui− 1
2

Ui−1 Ui

xi− 1 x
2

Figure 3.3: Riemann problem at the left boundary xi− 1 of the ith cell
2

For a linear system consisting of two equations (conservation of mass and equation of
motion) the discontinuous initial condition at time tn decomposes in two shock disconti-
k
nuities (so called waves) of size Wi− 1 for k = 1, 2, propagating along the characteristics
2
k
with speed σi− k
1 (figure 3.3). The discontinuities W separate three distinct regions. On
2
i− 1 2
the one hand two undisturbed zones, where the uniform initial conditions Uni and Uni−1
are present, and on the other hand an intermediate region, where a constant state Uni− 1
2
can be found. Now the discontinuities can be expressed in terms of the state quantities:
1 n
Wi− 1 = U
i− 1
− Uni−1 , 2
Wi− n
1 = Ui − U
n
i− 1
, (3.16)
2 2 2 2

and the sum of the equations (3.16) yields

Uni − Uni−1 = Wi−


1
1 +W
2
i− 1
. (3.17)
2 2

The two discontinuities are proportional to the eigenvectors r̂ki− 1 of Âi− 1 and propagate
2 2
k
with speeds σi− k
1 = λ̂ given by the eigenvalues of the Roe-matrix. More precisely the
2
i− 1 2
solution is obtained by solving the linear system
2
X
Ui − Ui−1 = αki− 1 r̂ki− 1 (3.18)
2 2
k=1

for the coefficients αki− 1 and then setting for the left- and rightgoing discontinuities
2

1 1
Wi− 1 = α r̂1 ,
i− 1 i− 1
2
Wi− 1 = α
2
r̂2 .
i− 1 i− 1
(3.19)
2 2 2 2 2 2

Using (3.16) yields the intermediate state vector Ui− 1 at the cell boundary interface xi− 1 :
2 2

Ui− 1 = Ui−1 + α1i− 1 r̂1i− 1 = Ui − α2i− 1 r̂2i− 1 (3.20)


2 2 2 2 2

Furthermore employing (3.10) we can write for the numerical flux at the cell edge xi− 1
2

Fi− 1 = f (Ui−1 ) + λ̂1i− 1 α1i− 1 r̂1i− 1 = f (Ui ) − λ̂2i− 1 α2i− 1 r̂2i− 1 . (3.21)
2 2 2 2 2 2 2

26
1
Wi− 2
Wi−
1 1
2 2

Uli− 1 Uri− 1
2 2

Ui−1 Ui

xi− 1 x
2

Figure 3.4: Riemann problem of a non-autonomous system at the left boundary xi− 1 of
2
the i-th cell.

F-wave algorithm Our aim is to solve quasi one-dimensional Euler equations which are
non-autonomous conservation laws since the flux function f depends not only on the state
vector u but also on the space-varying cross-section A(x, t). Here we extend the ideas
of the approximated Riemann solvers. First the space dependent flux function f (u, x) is
discretised to yield a flux function fi (u) that holds throughout the i-th cell. This approach
is called cell-centred flux function and in case of quasi one-dimensional Euler equations it
simply means that the cross section areas are constant in each cell. We denote the cross-
section area of the i-th cell at time tn by Ani . The resulting Riemann problem is much
more complicated since the intermediate state Ui− 1 is split up into the intermediate state
2
Uri− 1 at the right side of the cell boundary and the one on the left side Uli− 1 (figure 3.4).
2 2
Note that attempting to solve this Riemann problem by a decomposition of the form (3.17)
would fail in this case, since the jump in U at xi− 1 would have been neglected. However
2
taking this jump into account and computing the two intermediate states require solving
a nonlinear system of equations in each time step. Moreover a Roe-type linearisation
described above cannot be found in this case. Therefore the F-wave algorithm developed
by Leveque [17] is employed. In order to avoid these difficulties this scheme does not
attempt to perform a classical decomposition of the jump in U. Instead it uses a flux-
based wave decomposition, in which the flux difference fi (Ui ) − fi−1 (Ui−1 ) is decomposed.

1
t 2
Zi− 1 Zi− 1
2 2

Fi− 1
2

fi−1 (Ui−1 , Ai−1 ) fi (Ui , Ai )

xi− 1 x
2

Figure 3.5: Riemann problem of a non-autonomous system at the left boundary xi− 1 of
2
the i-th cell

Again we consider the the Riemann problem at the cell edge xi− 1 (figure 3.5). Here,
2

27
linearisation of the flux vector about the state vector Ui−1 on the left side of the cell edge
yields
fi−1 (U) = fi−1 (Ui−1 , Ai−1 ) + Ji−1 (U − Ui−1 ) . (3.22)
Similar, on the right side of the cell edge we obtain

fi (U) = fi (Ui , Ai ) + Ji (U − Ui ) . (3.23)

The Jacobians Ji−1 and Ji of the flux function are evaluated with the quantities of the left
side Ui−1 , Ai−1 and the quantities of the right side Ui , Ai , respectively. The appropriate
k
eigenvectors rki− 1 and eigenvalues σi− 1 are given by
2 2

1 n n 2 n n
σi− 1 = λ1 (Ui−1 , Ai−1 ) σi− 1 = λ2 (Ui , Ai ) (3.24)
2 2

and
r1i− 1 = r1 (Uni−1 , Ani−1 ) r2i− 1 = r2 (Uni , Ani ) . (3.25)
2 2

In order to ensure a conservative finite volume method the fluxes at the cell edge xi− 1
2
must be constant. So we set
   
Fi− 1 = fi−1 (Ui−1 , Ai−1 ) + Ji−1 UL
i− 1 − Ui−1 = f i (Ui , Ai ) + J i UR
i− 1 − Ui , (3.26)
2 2 2

where the resulting intermediate flux is denoted by Fi− 1 . Similar to the decomposition
2
k
(3.16) two flux discontinuities of size Zi− 1 for k = 1, 2, propagate along the characteristics
2
k
with speed σi− 1 (figure 3.5):
2

1 n n 2 n n
Zi− 1 = F
i− 1
− fi−1 , Zi− 1 = fi − F
i− 1
, (3.27)
2 2 2 2

and the sum of the equations (3.27) yields

fin − fi−1
n 1
= Zi− 1 + Z
2
i− 1
. (3.28)
2 2

Here Z is called F-wave which bears analogy to waves W of (3.19) but carries flux incre-
ments instead of increments in U. Now instead of solving the system (3.18), we solve
1 1 2 2
fi (Ui , Ai ) − fi−1 (Ui−1 , Ai−1 ) = βi− 1r
i− 1
+ βi− 1r
i− 1
(3.29)
2 2 2 2

k
for the coefficients βi− 1 and then set
2

k k
Zi− 1 = β rk .
i− 1 i− 1
(3.30)
2 2 2

k
Note that we can recover the waves Wi− 1 by setting
2

k 1 k
Wi− 1 =
k
Zi− 1 . (3.31)
2 σi− 1 2
2

Finally we can write for the numerical flux at the cell interface xi− 1
2

1 2
Fi− 1 = f (Ui−1 , Ai−1 ) + Zi− 1 = f (Ui , Ai ) − Z
i− 1
(3.32)
2 2 2

28
The finite volume scheme for Un+1
i of cell Ci at the time tn+1 is given by:

∆t  1 
Un+1
j = Unj − 2
Zi+ 1 + Zi− 1 (3.33)
∆x 2 2

It can be easily extended to second order for smooth solutions by adding a correction
term
∆t  1  ∆t  
Un+1
j = U n
j − Z i+ 21
+ Z 2
i− 12
− F n
i+ 12
− F n
i− 12
(3.34)
∆x ∆x
where
2  
n 1X p ∆t p p
Fi− 1 = |σ 1 | 1 − |σ 1 | Zi− 1 (3.35)
2 2 p=1 2 ∆x 2 2

Mac Cormack Scheme It is worth noting that there are finite volume methods which
do not require the computation of the wave structure. One of the most popular is the
Mac Cormack scheme [16], [29], [13], which is based on Lax-Wendroff methods. Here
we employ the two-step approach which can handle the problems arising from the space-
varying flux functions. The numerical flux Fi− 1 at the cell boundary xi− 1 for the quasi
2 2
one-dimensional Euler equations is given by
1
Fni− 1 = f (Uni+1 , Ani+1 ) + f (U∗i , A∗i ) ,

(3.36)
2 2
where
∆t
U∗i = Uni − f (Uni+1 , Ani+1 ) − f (Uni , Ani ) .

(3.37)
∆x
This finite volume method is also second order accurate.

Stability conditions of the finite volume schemes The time step ∆t of the numer-
ical integration has to be chosen in a way that the stability conditions of the numerical
schemes are satisfied. For all finite volume methods mentioned above the CFL condition
provides a necessary and sufficient condition [13], [16], [29]. It simply states that the ana-
lytical domain of influence lies within the numerical domain of influence. In case of quasi
one-dimensional Euler equations the CFL-condition can be written as
∆x
∆t ≤ . (3.38)
c + |u|

Comparison of schemes Both finite volume schemes for non-autonomous systems are
listed in table 3.1. The advantage of the Mac Cormack approach is that it is very easy to
implement since it does not require the computation of the wave structure of the system
of differential equations. On the other side employing the F-wave algorithm makes the
implementation of boundary conditions, where the characteristics are calculated, easier.

29
 
Un+1 1
Unj + U∗i − ∆t
f (U∗i , A∗i ) − f (U∗i−1 , A∗i−1 ) with
 
MacCormack: j = 2 2∆x
∆t
U∗i = Uni − f (Uni+1 , Ani+1 ) − f (Uni , Ani )

∆x

   
f-wave : Un+1
j = Unj − ∆t
∆x
1
Zi+ 1 +Z
2
i− 1 − ∆t
∆x
n
Fi+ 1 − F
n
i− 1
2 2 2 2

Table 3.1: Different one-dimensional finite volume schemes

3.1.2 Two-dimensional Finite Volume Schemes

N
Ci,j+1 ne
nw
E
P
W
Ci+1,j
Ci−1,j Ci,j
se
sw C n
y, j

i,j−1
S

x, i

Figure 3.6: Block-structured grid. Figure 3.7: Typical control volume in a


non-orthogonal grid.

The domain of the cylinder is divided into a finite number of small control volumes
(areas). To simplify matters we only employ non-orthogonal, block-structured grids (figure
(3.6). A finite volume Ci,j is defined by its corners sw, se, nw and ne (see figure 3.7). The
computational node and therefore all dependent variables are assigned to the centre P of
the control volume. At this point we assume that the distance between piston and cylinder
head is only a function of time t and later on we will discuss the effects of space-varying
heights Z(x, y, t). A detailed derivation of the finite volume methods for two-dimensional
grids can be found in [29]. Similar to the one-dimensional methods we transform the
governing differential equations into integral formulation. Integrating

∂u ∂f (u) ∂g(u)
+ + =0 (3.39)
∂t ∂x ∂y
over the cell Ci,j gives Z  
∂u ∂f (u) ∂g(u)
+ + dV = 0 . (3.40)
∂t ∂x ∂y
V

30
Using Green’s theorem , this equation becomes

Z Z
udV + h · ndS = 0. (3.41)
∂t
V S

Here, n denotes the unit vector normal to the surface S of the finite volume and the tensor
h can be expressed in Cartesian coordinates as

h = f (u)i + g(u)j ,

where i and j denotes the unit vector in x and y direction, respectively. In case of two-
dimensional non-orthogonal grids one obtains

h · n dS = (f (u)dy − g(u)dx) , (3.42)

which can be substituted in 3.41 to yield



Z Z
u dx dy + (f (u)dy − g(u)dx) = 0. (3.43)
∂t
V S

The surface S consists of the four sections nw − sw, sw − se, se − ne and ne − nw. Now
integrating over one time step tn+1 − tn = ∆t one obtains

Z Z tZn+1Z

u(x, y, tn+1 ) dx dy − u(x, y, tn ) dx dy + (f (u)dy − g(u)dx) dt = 0. (3.44)


V V tn S

Again, the first two terms in (3.44) correspond to the total change of the state quantities
inside the cell during the time step ∆t = tn+1 − tn . The third integral represent the total
fluxes of the state quantities across the cell boundaries. Let Uni,j denote the cell average
of the state quantities over the cell (i, j) at time tn :

1
Z
n
Ui,j ∼ u(x, y, tn ) dx dy . (3.45)
|Ci,j |
V

where the cell area |Ci,j | can be written as

1
|Ci,j | = |(ne − sw) × (nw − se)| , (3.46)
2
The flux integral of (3.44) at the section sw − se is approximated by the numerical fluxes
Fni,j− 1 and Gni,j− 1 :
2 2

tZn+1Z
1
Fni,j− 1 ∼ f (u)dy , (3.47)
2 ∆t
tn S
tZn+1Z
1
Gni,j− 1 ∼ g(u)dy . (3.48)
2 ∆t
tn S

31
Finally, when employing the above approximations the system of equations for Un+1 i,j of
cell Ci,j at the time tn+1 can be written as:
h
Un+1
i,j = U n
i,j − ∆t
|Ci,j | Fi,j− 1 (yse − ysw ) + Fi+ 1 ,j (yne − yse ) +
2 2

Fi,j+ 1 (ynw − yne ) + Fi− 1 ,j (ysw − ynw )
2 2

− Gi,j− 1 (xse − xsw ) + Gi+ 1 ,j (xne − xse ) +
2 2
i
Gi,j+ 1 (xnw − xne ) + Gi− 1 ,j (xsw − xnw ) , (3.49)
2 2

Similar to the one-dimensional finite volume schemes, a numerical method written in the
form (3.49) is called conservative. The boundary problems in the two-dimensional domain
(cylinder) do not require the calculation of characteristics and therefore we employ simple
and robust finite volume methods, where calculating the wave structure is not necessary.

Lax-Friedrich’s scheme The space centred explicit Lax-Friedrich’s scheme was one of
the first schemes used to solve Euler equations [13]. Due to the poor first order accuracy the
solutions must be questioned. However it can be used as intermediate step in high order
schemes, as in the two-step Lax-Wendroff method (Richtmyer). A detailed description
of this method can be found in [13] and [29]. For the numerical flux Fi+ 1 ,j of the face
2
ne − nw we can write
1 1 |Cij | + |Ci,j+1 |
Uni,j − Uni,j+1 ,

FLF i,j+ 1 = (fi,j+1 + fi,j ) + (3.50)
2 2 16 ∆t (ynw − yne )

where
fi,j+1 = f Uni,j+1 .

(3.51)
The numerical fluxes for the other faces can be easily derived. The second term on the
right side of equation (3.50) is used to stabilise the method. It can be interpreted as an
dissipative term proportional to the second derivative of u. Therefore the discretisation
scheme (3.50) is consistent with

∂u ∂f (u) ∂g(u) ∂2u ∂2u


+ + = αd 2 + βd 2 , (3.52)
∂t ∂x ∂y ∂x ∂y
which is a dissipative system of equations with the numerical viscosities αnum and βnum
of order O(∆x). Again, the time step ∆t of the numerical integration has to be chosen
such that the stability conditions of the numerical schemes are satisfied. Unfortunately,
a sufficient stability criterion can not be given for this finite volume scheme on non-
orthoganal quadrilateral grids. However, it turns out that using following conditions lead
to stable methods in the tested cases:
∆x ∆y
∆t ≤ , ∆t ≤ . (3.53)
2 (c + |u|) 2 (c + |v|)

Lax-Wendroff schemes, Richtmyer scheme In this work we discuss the two-step


Lax-Wendroff schemes introduced by Richtmyer [13], [29]. There are also one-step Lax-
Wendroff schemes but they suffer from the difficulty of requiring calculations of Jacobian
matrices. The Richmyer scheme consits of two steps: The predictor step, in which an

32
estimated solution is calculated, and a corrector step, where the actual solution is found.
The numerical fluxes from the Lax-Friedrich’s scheme are used as first guess. In the second
step the same fluxes but without the dissipative terms and evaluated at the intermediate
step are employed. As an example we approximate the analytical flux fi+ 1 ,j of the face
2
1
ne/nw by the numerical flux FR i+ 1 ,j . First the predictor flux at the time n + 2 is given
2
by
n+ 1 1 1 Ai,j + Ai,j+1
uni,j − uni,j+1 .

Fi,j+2 1 = (fi,j+1 + fi,j ) + (3.54)
2 2 16 ∆t (ynw − yne )
Similarly, we obtain the numerical fluxes for the other faces. Thus we can calculate the
1
predictor state vectors Un+ 2 by means of equation (3.49). Second the corrector step can
be written as  
1 n+ 12 n+ 1
FR i,j+ 1 = fi,j+1 + fi,j 2 , (3.55)
2 2
where  
n+ 21 n+ 12
fi,j+1 =f Ui,j+1 . (3.56)

The necessary and sufficient stability condition for the case of orthogonal grids where
∆x = ∆y was given by Richtmyer [13] and can be written as

∆x
∆t ≤ √  √ . (3.57)
2 c + u2 + v 2

Although, in case of non-orthogonal grids this relation is not sufficient, satisfying results
have been obtained using it.

Flux Limiter In the vicinity of discontinuities second order methods performs rather
poor. Precisely, non-physical oscillations can occur. Now we want a method which uses
an accurate second order scheme in regions of smooth solutions and non-oscillating first
order methods in all other regions. Therefore the numerical flux is computed by a linear
combination of first order and second order fluxes. A detailed description can be found in
[16]. Here we use the first order Lax-Friedrich’s flux FLF and the second order Richtmyer
flux FR to obtain

F(Un ) ≈ FLF (Un ) + Ψ(Un ) [FR (Un ) − FLF (Un )] , (3.58)

where the flux limiter Ψ(Un ) describes the smoothness of the solution. The quotient of
the pressure gradients in two adjacent cells Θ represents a measure of smoothness. For
instance, at the face between nw − ne it takes following form:
pi,j+2 − pi,j+1
Θi,j+ 1 = , (3.59)
2 pi,j+1 − pi,j

In order to avoid biased weighting 3.59 is replaced by the following formula in every other
time step:
pi,j+1 − pi,j
Θi,j+ 1 = , (3.60)
2 pi,j − pi,j−1
Note that for smooth solutions Θ becomes 1. A wide variety of flux limiters have been
developed and we refer to [13] for an overview. In this work we employ the Minmod limiter

33
where the flux limiter Ψ as function of Θ is given by

 0
 : Θ<0
Ψ(Θ) = Θ : 0≤ Θ≤1 (3.61)

1 : Θ>1

Treatment of Source Terms Cylinder head recesses are often used in order to channel
the gas flow into the valve pockets. Thus the distance between cylinder head and piston
Z varies not only with time but also with the space coordinates x and y, resulting in a
source vector s in the two-dimensional Euler equations

∂U ∂f (u, Z) ∂g(u, Z)
+ + =s. (3.62)
∂t ∂x ∂y
Here, the source vector takes the form
 
0
s = p ∂Z
∂x  . (3.63)
 

p ∂Z
∂y

In order to solve this problem we employ a fractional step method in which we split the
general problem (3.62) into a homogeneous conservation law and an ODE given by

∂u
=s. (3.64)
∂t
Now for both subproblems standard methods can be employed. First we use one of the
∂f (u,Z)
finite volume methods above to solve ∂u∂t + ∂x + ∂g(u,Z)
∂y = 0 and second (3.64) is
integrated by a forward Euler approach which is written as
 
0
Un+1 ∗  ∗ ∂h 
i,j = Ui,j + ∆t pi,j ∂x  . (3.65)
p∗i,j ∂h
∂y

The vector U∗i,j denotes the intermediate state vector derived from a finite volume scheme.
In case of second order two-step methods, in the predictor step and in the corrector step
(3.65) applies.

3.2 Numerical Boundary Conditions

In terms of numerical treatment of the boundary conditions, we have to distinguish be-


tween the physical boundary conditions that follow from Section 2.4.3, 2.4.5, 2.4.6 and 2.4.4
and the numerical boundary values required for computation. The numerical algorithms
require the information of all state quantities at the boundaries. The physical boundary
conditions only provide the boundary value corresponding to the incoming characteristic.
In this section we discuss the various possibilities to determine the missing conditions for
different boundary problems.

34
3.2.1 Walls one-dimensional

The mass flux ṁ through the wall is equal to zero and thus the velocity u at the boundary
xk+ 1 is also equal to zero. This boundary problem can be modelled by simply using a
2
ghost cell (boundary cell) Ck+1 as mirror element of Ck (figure 3.8. The state quantities
Uk+1 of the ghost cell are set to the ones in the adjacent cell, except that the sign of
velocity is changed.

Wall Boundary

Cell Cell Cell Cell Cell ṁ Cell


Ck−1 Ck Ck+1 Ck−1 Ck Ck+1

xk−1 xk xk+1 xk−1 xk xk+1

Figure 3.8: Ghost cell Ck+1 at the bound- Figure 3.9: Mass flow through the bound-
ary ary

3.2.2 Mass-flow through the Boundary

Again, the ghost cell Ck+1 outside the subdomain is added (figure 3.9). The mass flux ṁ
through the boundary provides the necessary first equation for the two state quantities of
the ghost cell Ck+1 . Now a wide variety of methods for the missing condition have been
found and we refer to [13] for an overview. Here we introduce three different approaches:

Leaving characteristic The additional relationship for the missing state quantity is
given by the linear partial differential equations (2.25) for the characteristic leaving the
subdomain and entering the ghost cell. These linear PDEs are discretised by a simple
upwind scheme and the outgoing characteristic can be calculated at the desired boundary
at the new time. According to (2.25) the equation for the leaving characteristic on the
right boundary takes the form:
∂w2 ∂w2
+ λ2 =0. (3.66)
∂t ∂x
We assume, that the flow situation at time tn is known. Then first order upwinding of
3.66 yields
n+1 n ∆t
+ λ2 (Unk+1 , Ank+1 ) n n

w2,k+1 = w2,k+1 w2,k − w2,k+1 (3.67)
∆x
Together with the definition of the characteristic variables (2.23) we are able to calculate
the full boundary data of Ck+1 to apply to the numerical scheme. Equation (3.67) and the
prescription of ṁ form a system of equations which is solved by means of a well-known
numerical Newton-Raphson iteration [26]. Similarly, solutions for this boundary problem

35
at the left side of the subdomain can be found. Let C0 denote the ghost cell on the left
side. Here the equation for the leaving characteristic takes the form

n+1 n ∆t
+ λ1 (Un0 , An0 ) n n

w1,0 = w1,0 w1,0 − w1,1 . (3.68)
∆x

Note that the ideas of this approach can be used for all kind of boundary problems
(e.g. pressure, mass flow, velocity, density).

Extrapolation The second method to determine the missing equation is extrapolation.


Instead of considering characteristics leaving the state quantities for the ghost cell Ck+1
are extrapolated from the interior points. According to Hirsch [13] the extrapolation
used can be one order lower than the finite volume scheme without altering the accuracy
and stability of the interior numerical method. For this boundary problem one possible
assumption is, that the mass per unit length of the boundary cell Rk+1 can be extrapolated
from the adjacent ones. It is given by

Rk+1 = 2Rk − Rk−1 . (3.69)

Together with the given mass flux ṁ one is able to determine both state quantities. The
state vector for the ghost cell Ck+1 takes following form
! !
Rk+1 2Rk − Rk−1
Uk+1 = = ṁ . (3.70)
uk+1 Rk+1

Source Term The third approach also models this boundary problem without consid-
ering the leaving characteristics. As in the case of walls the ghost cell is used as a mirror
element. The loss or gain of mass and impulse inside the computational domain due to
mass flux ṁ through the valve are included by a source term in the cells adjacent to the
mirror element. The state vector Uk of cell Ck at the time tn+1 is given by:
∆t  n 
Un+1
k = U n
k + F 1
k+ 2
) − F n
1
k− 2
) − ∆tS , (3.71)
∆x
where T
ṁ2

1
S= ṁ, . (3.72)
∆x 2R1 dnk 2

3.2.3 Pressure Outlet

At the end of the valve retainer or at the end of the pressure chamber a pressure outlet is
defined. Either the characteristic method or the extrapolation method, both described in
section 3.2.2, can be employed. However, instead of the mass flux the pressure pout is pre-
scribed at the boundary. In addition, the pressure losses due to sudden changes pv , which
have been derived in section 2.4.6, can be easily incorporated. Using the extrapolation
method the state vector Uk+1 of the ghost cell Ck+1 is given by
!  1

A p γ ρ0
R1dk+1 k+1 res 1
Uk+1 = = p0 γ  , (3.73)
uk+1 2un − un k k−1

36
where the resulting pressure pres is written as

pres = pout + pv . (3.74)

Note that pv and thus pres depend on the state quantities Uk at time tn .

3.2.4 Boundary Problem at Sudden Changes of Cross-section

Cell Cell
Cell Cell
Ck Ck+1
Ck−2 Ck−1

xk−2 xk−1
xk xk+1 Separation

pv
Cell Cell Cell Cell
Ck−2 Ck−1 Ck Ck+1

xk−2 xk−1

xk xk+1

Figure 3.10: Numerical treatment of sudden changes in flow cross-sections: Computational


domain is separated and pressure loss pv is introduced

At the location of sudden changes in flow cross-section the computational domain is


split up (figure 3.10. Then the reconnection of the resulting subdomains is considered
as as a boundary problem for the two cells Ck and Ck−1 adjacent to the splitting. As
mentioned above the finite volume methods used require the full set of state quantities at
these cells. First we make use of the equations of the pressure losses (2.40),(2.38) derived
in section 2.4.6. Second, the characteristics pointing into the cells Ck and Ck−1 provide
the equations for the missing conditions. Similar to the mass flux boundary conditions we
obtain equations for the characteristic variables by solving the partial differential equations
for the characteristic variables (2.25) with and Upwind scheme. Finally the system of
equations for the state quantities of the cells Ck and Ck−1 are given by

ṁn+1 n+1
k−1 = ṁk
pt n+1 n+1
k−1 = pt k + pv
n+1 n ∆t
+ λ2 (unk−1 ) n n

w2,k−1 = w2,k−1 w2,k−2 − w2,k−1 (3.75)
∆x
n+1 n n ∆t n n

w1,k = w1,k + λ1 (uk ) w1,k − w1,k+1 .
∆x

37
A drawback of this method is that it imposes unwanted mass sinks or mass sources on
the location of the separation. However, this problem can be easily solved using following
procedure. First the state vectors of the boundary cells un+1
k−1 and uk
n+1
are calculated
without the pressure losses by means of finite volume methods. Then the system of
equations (3.75) is solved by means of Newton-Raphson iterations. Finally the lost mass
∆m is added to the cells Ck and Ck−1 according to their volumes. We obtain

∆m = Rk − 1n+1 |WL +Rkn+1 |WL −(Rk−1


n+1
+ Rkn+1 ) (3.76)

and

n+1 n+1 ∆m An+1


k−1
Rk−1 = Rk−1 |WL − (3.77)
∆x An+1
k−1 + An+1
k
∆m An+1
Rkn+1 = Rkn+1 |WL − k
, (3.78)
∆x An+1
k−1 + An+1
k

where |WL denotes solutions of (3.75).

3.2.5 Boundary Conditions at the Valve

Cylinder Valve Chamber


1 2
Cell Cell Cell Cell

Ck−1,1 Ck,1 C1,2 C2,2

xk−1 xk,1 x1,2 x2,2 analytical

numerical

Cell Cell Cell


Ck−1,1 Ck,1 Ck+1,1

xk−1 xk,1 xk+1,1

Cell Cell Cell


C0,2 C1,2 C2,2

x0,2 x1,2 x2,2

Figure 3.11: Valve region and the adjacent finite volume cells: Ghost cells Ck+1,1 and C0,2
are introduced for the numerical boundary conditions.

A valve separates the computational subdomain of the cylinder from the adjacent valve
retainer subdomain. Figure 3.11 shows the valve region and its adjacent boundary cells.
Again we introduce two ghost cells Ck+1,1 and C0,2 and define the boundary conditions for
these cells in a way that the equation of the valve connecting this subdomains is satisfied.
Now we distinguish two different states of the valve:

38
• Closed Valve (xv = xv,min ): Here the mass flux ṁ through the valve is equal to zero.
Therefore the method for the one-dimensional wall (section 3.2.1) can be applied to
the ghost cells Ck+1,1 and C0,2 .

• Open Valve (xv,min < xv ≤ xv,max ): The mass flux ṁ through the valve is given by
(2.51). Thus the methods for the mass-flow boundary in section 3.2.2 can be applied
to the ghost cells Ck+1,1 and C0,2 .

3.2.6 Boundary Conditions at the T-piece

Following the solution of the boundary problem at sudden changes of cross-section we


expand the method of characteristics to model the T-piece. Now there are three compu-
tational domains with their appropriate boundary elements Ck,1 , Ck,2 and C1,3 . Again the
ghost cells Ck+1,1 , Ck+1,2 and C0,3 were added. Figure 3.12 shows the T-piece region and
adjacent cells.

Ck−1,1 Ck,1 Ck+1,1 Ck+1,2 Ck,2 Ck−1,2

C0,3

junction area
C1,3

C2,3

Figure 3.12: Finite Cells at the T-piece. Junction area is shaded

The equations for the leaving characteristics and the equations for the T-piece (2.33),
(2.34) form a system of equations for the required state quantities in the boundary cells:

ṁn+1 n+1 n+1


k+1,1 + ṁk+1,2 = ṁ0,3 (3.79)
htot n+1 n+1 n+1
k+1,1 = htot k+1,2 = htot 0,3 (3.80)
n+1 n ∆t
+ λ2 (unk+1 ) n n

w2,k+1 = w2,k+1 w2,k − w2,k+1 |1 (3.81)
∆x
n+1 n ∆t
+ λ2 (unk+1 ) n n

w2,k+1 = w2,k+1 w2,k − w2,k+1 |2 (3.82)
∆x
n+1 n ∆t
+ λ1 (un0 ) n n

w1,0 = w1,0 w1,0 − w1,1 |3 . (3.83)
∆x
Again, a numerical Newton-Raphson iteration [26] is employed to solve this system. If
the pressure losses in the T-piece are taken into account the above system of equations is
extended by the six pressure loss equations (2.47).

39
3.2.7 Wall two-dimensional

At walls we require that the velocity normal to the boundary vn vanishes (figure 3.13).
Therefore we introduce a ghost cell C ∗ and set its computational variables to the ones of
the adjacent cell C except for the velocity v∗ . Simple vector analysis yields

v∗ = v − 2 |nb v| nb , (3.84)

where nb is the normal vector to the boundary.

ghost ∗
cell C

v∗
nb
vn
v
wall
vt
C

Figure 3.13: Boundary cell in the two-dimensional domain

3.2.8 Interface Condition

The interface connects the two-dimensional computational domain representing the cylin-
der with the one-dimensional domain representing the valve pockets (figure 3.14). Since
we are linking quantities of different dimensions we have to reduce the two-dimensional
ones to suitable one-dimensional information by introducing artificial cells C0,2d and C1,2d .
These cells are two-dimensional cells but extend across the whole cross-section. Similarly,
at the side of the one-dimensional domain two artifical cells, C0,1d and C1,1d , are introduced.
The volume of each artificial cell is the sum of all volumes of the cells inside the cylinder
adjacent to the interface under consideration. Moreover we require, that the artificial cells
C0,2d and C1,2d are identified with C0,1d and C1,1d , respectively. Without loosing generality
we assume that the valve pocket is parallel to the x-axis.
Now we employ the following calculation procedure in each timestep: First a two-
dimensional finite volume scheme is applied to determine the state vector U0,2d at the time
tn+1 . In doing so the velocity normal to the valve pocket axis (x-axis) is set to zero and
thus impulse Iy 0,2d is zero. Similarly, the state quantities Un+1
1,1d of cell C1,1d are calculated
using one-dimensional finite volume methods. However we still have to determine the
boundary cells C0,1d and C1,2d . Here we follow a different approach compared to the
previous boundary condition problems. The state quantities of the cell C0,1d are obtained
by projecting the state vector Un+1 0,2d of the two-dimensional computational domain on

40
Cylinder

Interface

C0,2d C1,2d
Artificial
ws

Cells

Extension Two-dimensional
Projection One-dimensional

C0,1d C1,1d C2,1d


Valve pocket
Artificial
Cells
Interface

Figure 3.14: Interface between two-dimensional and one-dimensional domain.

n+1
the state vector U0,1d of the one-dimensional domain. Furthermore, we extend the state
vector U1,1d of C1,1d in order to derive the state vector Un+1
n+1
1,2d . So we can write

R1,1d
   
M1,2d ws
Ext. 
u1,1d R1,1d  ,

U1,2d =  Ix1,2d  := (3.85)
  
 ws 
Iy 1,2d 0
 
! M0,2d ws
R0,1d Proj.  Ix0,2d 
U0,1d = :=   . (3.86)
u0,1d  M0,2d 
0

Here ws denotes the width of the interface.


The drawback of this procedure is that the requirement for conservation of state quan-
tities can not be satisfied, which is seen from following considerations (figure 3.15). Let
mn0 and mn1 denote the mass inside the artificial cells C0,2d , C0,1d and C1,2d , C1,1d , respec-
tively. In addition, ∆mc and ∆mv stand for the mass fluxes through the interface at the
cylinder and valve pocket side during one time step. The mass fluxes between the artifical
cells in the one-dimensional and two-dimensional domain are denoted by ∆m1 and ∆m2 ,
respectively. All these mass fluxes can be easily derived from the numerical fluxes of the
finite volume methods used. Now the masses mn+1 0 and mn+11 inside the cells C0,2d and

41
C1,1d at time tn+1 are given by
mn+1
0 = mn0 + ∆mc − ∆m2 , (3.87)
mn+1
1 = mn1 + ∆m1 − ∆mv . (3.88)
(3.89)
Then the projection and extension described above assigns the new masses to the appro-
priate cells in the other computational domain. Finally the mass inside the artifical cells
in the one-dimensional domain and two-dimensional domain at the time tn+1 is given by:
(mn+1
0 +mn+1
1 )1d = (mn+1
0 +mn+1
1 )2d = (mn0 +mn1 )+(∆mc −∆mv )+(∆m1 −∆m2 ) . (3.90)
If the last term of (3.90) is zero the method used would be conservative, since every gain
or loss of mass inside the artifical cells is due to the fluxes across the interfaces. However,
the numerical fluxes and thus the mass fluxes of the one-dimensional scheme differ from
the fluxes of the two-dimensional scheme, leading to unwanted mass sources and sinks.
Since we can quantify the mass gains produced by the procedure described above the
errors made can be easily corrected. In detail, the gained mass ∆m1 − ∆m2 is subtracted
from the artifical cells Ci,1d and Ci,2d for i = 0, 1. We obtain
n+1 n+1 ∆m1 − ∆m2 n+1 n+1 ∆m1 − ∆m2
Ri,1d = Ri,1d |B − , Mi,2d = Mi,2d |B − , (3.91)
2∆x 2∆xws
where |B denotes the state quantities before the correction.

C0,2d C1,2d
∆mc ∆m2

Interface
m0 m1
to
Cylinder
Projection Extension
Interface
to
C0,1d C1,1d Valve pocket
∆m1 ∆mv

m0 m1

Figure 3.15: Mass correction: Mass inside the artifical cells and mass fluxes during one
time step

3.3 Equation of motion for the valve plate

The second order ODE of the valve motion (2.47) can be solved analytically if a constant
valve force Fv is assumed during the time step ∆t. The solution for the valve lift xv and
the valve plate velocity vv at the time tn+1 takes the form

42
r r  r 
n+1 n cs 1 n cs
vv = vv cos ∆t + Fv − cs (l1 + xv ) sin ∆t (3.92)
mv mv cs mv
  r r r
Fv Fv cs mv n cs
xv n+1 = − l1 + l1 + xv n − cos ∆t + vv sin ∆t , (3.93)
cs cs mv cs mv

where
Fv = (p1 − p2 )Av − Fadh . (3.94)

43
Chapter 4

Measurements

In this chapter measurements for two test compressors from Burckhardt Compression,
Switzerland and Ariel, USA are discussed. Both compressors are double-acting barrel
design compressors. However, Burckhardt’s compressor has two valves and is about three
times smaller than Ariel’s compressor with eight valves.

4.1 Compressor with two valves

A double-acting, 2 cylinder, barrel design reciprocating compressor was tested. The exper-
iment was carried out by Daniel Sauter, Markus Lehmann and Roland Aigner in November
2004 at Burckhardt Compression, Switzerland. The main specifications of the compressor
can be found in table 4.1. Various sensors were used to measure the

• pressure inside the cylinder and inside the suction and discharge chamber,
• temperature inside the cylinder and in the pipes,
• valve motion of the discharge valve,
• volume flow,
• speed of the crankshaft.

Figure 4.1 shows the experimental compressor and its main components. Figure 4.2 shows
the location of each sensor in cylinder 1. A detailed description of the sensors used can
be found in the appendix A.1.

4.1.1 Test Cases

In order to get a wide variety of test cases the valve configuration (see table 4.2) of the
pressure valve of the first cylinder and the discharge pressure have been varied. In addition,
the influence of the number of working cylinders has been examined. For the comparison
of the measurements with numerical simulations three of four working chambers have been
sealed off during measurements by covering the valve pockets of these chambers. All test
configurations were mesasured at 4 different relative discharge pressures: 1, 2, 3 and 4 bar.
The ambient pressure at the days of measurement was 0.97 bar. Therefore in this section

44
Type of compressor: Burckhardt compression 2K90-1A
2 cylinders, double-acting, barrel design
Diameter of cylinder: 0.22 m
Stroke: 0.09 m
Speed of crankshaft: 980-990 rpm
Suction pressure: ambient pressure
Maximum pressure ratio: 1/5
Clearance: 1.6-2 mm
Gas: Air
Length of connecting rod: 0.25 m
Type of valves: 110K12
Classification ǫ: 0.22

Table 4.1: Main Specifications of the Burckhardt test compressor.

pressure and temperature sensors

figure 4.2 discharge pipe


suction pipe

111111
000000
000000
111111
cylinder 2 cylinder 1 000000
111111
000000
111111
1111111111
0000000000 engine
000000
111111
0000000000
1111111111
0000000000
1111111111 000000
111111
000000
111111
damper
000000
111111
000000
111111
outlet restriction valve heat exchanger

orifice flywheel with trigger

Figure 4.1: Shema of the Burckhardt experimental compressor

45
pressure is displayed relative to this ambient pressure. Six different valve configurations
have been tested. First, the maximum valve lift has been set to 1.05, 1.35 and 1.65 mm.
Second, a total spring constant of the valve of 48 and 80 N/mm has been used. The signals
of the sensors were recorded with a sampling rate of 10000 Hz. For every test case three
independent measurements were taken for a complete compressor cycle at three different
times ta , tb and tc , which were approximately one minute apart.

Discharge valve mass of valve plate 0.076 kg


Measurement valve number of springs 8
spring deflection 0.0009 m
spring constant cs 48/80 N/mm
maximum valve lift xv,max 1.05/1.35/1.65 mm
valve force area 0.0056 m2
parameter of valve αv /βv 2.042 [1]/ 80000 1/m2
parameter of valve fe1mm 1.86 m2 /m
Discharge valve number of springs 12
spring deflection 0.9 mm
spring constant cs 48 N/mm
maximum valve lift xvmax 1.35 mm
Suction valve number of springs 12
spring deflection 0.594 mm
spring constant cs 21.3 N/mm
maximum valve lift xv,max 1.35 mm

Table 4.2: Specifications of the valves.

4.1.2 Results of Measurement

Periodic Process After a warm-up of several minutes the process inside the compressor
is periodic (figure 4.3). Then comparisons of valve motion and pressure distribution over
one period at different times show only small deviations.

Variation of number of Working Chambers The difference between one cylinder


and two cylinders working are very small. Whereas, if only one working chamber is active,
pressure distribution and valve motion show slightly different behaviour (figure 4.4 and
figure 4.5). This is due to the fact that in case of double-acting cylinders pressure waves
from the crank end side and the head end side interact, since both sides are connected
by the pressure chamber. Hence the outflow condition for the discharge valve changes,
resulting in different pressure distributions inside the working chamber. We observe that
the pressure in the pressure chamber has one or two peaks depending on whether the
compressor is double-acting or not. However, the differences in the valve retainer during
discharge (between 130◦ CA and 180◦ CA) are rather small and therefore differences of
the pressure inside the cylinder and differences of the valve motion are small.

46
suction valve discharge valve
4

8 6 1 5 9

7 2

1 . . . . . . pc1 4 . . . . . . pc4 7 . . . . . . pc7


2 . . . . . . pc2 5 . . . . . . pc5 8 . . . . . . pin and Tsc
3 . . . . . . pc3 6 . . . . . . pc6 9 . . . . . . pout and Tpc

Figure 4.2: Location of the sensors inside the cylinder 1, the suction and the discharge
valve retainer.

400000 0.0016
pressure, time a
pressure, time b
350000 pressure, time c 0.0014
valve lift, time a
Relative pressure at pc5 [Pa]

300000 valve lift, time b 0.0012


valve lift, time c

250000 0.001

Valve lift [m]


200000 0.0008

150000 0.0006

100000 0.0004

50000 0.0002

0 0

−50000 −0.0002
0 50 100 150 200 250 300 350

Crank angle CA[◦ ]

Figure 4.3: Comparison of relative pressure at pc5 and valve lift at different times ta , tb
and tc . Case: pout = 3 bar, cs =48N/mm , xv,max =1.35 mm.

47
400000 0.0016
pressure, 2 cylinder
350000 pressure, 1 cylinder 0.0014
pressure, 1 chamber
valve lift, 2 cylinder
Relative pressure at pc5 [Pa]
300000 valve lift, 1 cylinder 0.0012
valve lift, 1 chamber
250000 0.001

Valve lift [m]


200000 0.0008

150000 0.0006

100000 0.0004

50000 0.0002

0 0

−50000 −0.0002
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 4.4: Comparison of relative pressure at pc5 and valve lift with 2 cylinders, 1 cylinder
and 1 chamber working. Case: pout = 3 bar, cs =48 N/mm, xv,max =1.35 mm.

340000
pout, 2 cylinder
pout, 1 cylinder
330000 pout, 1 chamber
Relative pressure at pout [Pa]

320000

310000

300000

290000

280000

270000

260000
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 4.5: Comparison of relative pressure in the valve retainer: 2 cylinders, 1 cylinder
and 1 working chamber. Case: pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm.

48
Initiation of Waves In figure 4.6 the pressure readings at pc5 and pc6 close to the
discharge valve and the suction valve, respectively, and the position of the valve plate as
a function of the crank angle CA are shown. During compression, when both valves are
closed the pressure is uniform in the cylinder (CA = 140◦ ). When the pressure in the
cylinder exceeds the pressure in the valve retainer plus the springing the valve starts to
open. The pressure near the outflow valve ceases to increase and reaches a maximum,
while the pressure close to the suction valve still increases. The opening of the valve has
initiated a wave which reaches the opposite side of the cylinder at a crank angle CA=146◦ ,
just before the valve has completely opened. Then the rarefaction wave is reflected at the
suction side. Thus the pressure at the suction side (pc6) attains also a maximum. But it is
markedly larger than that on the discharge side (pc5). When the valve plate hits the valve
seat the pressure in front of the pressure valve is large enough to keep the valve open until
CA = 155◦ . Then the pressure has dropped and the valve begins to close again reducing
the mass outflow. Again this leads to an increase of the pressure inside the cylinder. Now
increasing of pressure forces the valve to open and a complicated interaction of the valve
motion with the pressure waves in the cylinder takes place. Shortly after the lowest volume
point the valve closes. At that time a complicated system of waves travelling back and
forth is left in the cylinder. In figure 4.7 the difference of the pressure readings between
suction side and cylinder centre (pc5 - pc1) is shown. We observe that during expansion
the amplitudes of the waves are proportional to the mean pressure inside the cylinder
and their periodic time tends to the travelling time of a wave across the cylinder 2d/c
corresponding to the lowest eigen mode in lateral direction. We want to point out that
at the beginning of the expansion phase a mix of modes with several eigen frequencies
is present. These modes are initiated by the valve motion. As said above the zeroth
eigenmode has the smallest damping and remains visible until the suction phase starts.

600000 0.003
pc5
pc6
pout
500000 valve lift 0.0025
Relative pressure [Pa]

400000 0.002 Valve lift [m]

300000 0.0015

200000 0.001

100000 0.0005

0 0

100 120 140 160 180 200 220 240


Crank angle [◦ ]

Figure 4.6: Relative Pressure at pressure sensors pc5, pc6 and pout and valve lift. Case:
pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm.

49
60000
diff. pressure pc5−pc1

Pressure difference pc5-pc1 [Pa]


40000

20000

−20000

−40000

−60000
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 4.7: Pressure difference pc5-pc1. Case: pout = 4 bar, cs =48 N/mm, xv,max =1.35
mm.

Plane Waves As described above, pressure waves run through the cylinder in lateral
direction (x-direction). For example the opening of the valve initiates a wave. This wave
runs at the speed of sound and reaches the sensor next to the discharge valve (pc5) first
(figure 4.8). Then sensors pc1, pc2, pc3, pc4 at the y-axis notice the rarefraction wave.
Since the differences between the pressure readings at these sensors are rather small, a
plane wave can be assumed. Finally, the information of the opening of the valve has
travelled to the sensors pc6 and pc7.

Pressure differences across the discharge valve When the discharge valve starts
opening, the gas velocity is almost equal to zero and the pressure is constant inside the
cylinder except from very small disturbances. Therefore the pressure difference (pressure
loss) across the valve can be determined exactly from the pressure readings at pc5 and
pout (figure 4.9). Both, Figure 4.10 and table 4.1.2 show that higher pressure differences
occur at higher discharge pressures. Since pressure forces due to the valve springing are
constant and flow inside the cylinder is negligible small we assume that faster increasing
pressure inside the cylinder at higher discharge pressures, valve plate sticking, and the
inertia of the valve plate contribute to this behaviour. Note that in case of open valves,
the time, waves need to cover the distance between the valve and the sensors, must be
taken into account and thus the difference of the pressure readings at pc5 and pout do not
give the exact pressure difference across the valve.

Valve Plate Motion The motion of the discharge valve plate has been recorded using
three distance sensors. They were mounted on the circumference of a mean diameter of the
valve each with a distance of 90◦ to the next one. While at the first opening of the valve the
plate keeps plane, at the end of the discharge phase the valve plate becomes slightly askew
(figure 4.11). We have identified that asymmetric flow and variations in the springing

50
pc1
pc5
pc6
500000 pc4
pc3
Relative pressure [Pa]

450000

400000

350000

300000
120 130 140 150 160 170 180 190 200
Crank angle [◦ ]

Figure 4.8: Relative Pressure at pressure sensors pc1, pc3, pc4, pc5 and pc6. Case:
pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm.

500000
pout, 4 bar
450000 pout, 1 bar
pc5, 4 bar
pc5, 1 bar
400000
Relative pressure [Pa]

350000

300000

250000

200000

150000

100000

50000

0
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 4.9: Relative pressure in front (pc5) and behind pressure valve (pout) . Cases: 1
chamber, pout = 4 bar, cs =48 N/mm, xv,max = 1.35 mm and 1 chamber, pout = 1 bar,
cs =48 N/mm, xvmax =1.35 mm.

51
120000
pc5−pout, 4 bar
pc5−pout, 3 bar
pc5−pout, 2 bar
Difference pressure pc5-pout [Pa]

100000 pc5−pout, 1 bar

80000

60000

40000

20000

0
0 0.0002 0.0004 0.0006 0.0008 0.001 0.0012 0.0014
Valve Lift [m]

Figure 4.10: Difference pressure pc5-pout over valve lift for different discharge pressures .
Cases: 1 chamber, pout = 1/2/3/4 bar, cs = 48 N/mm, xv,max = 1.35 mm.

Discharge Pressure Pressure Difference (pc5-pout) Pressure Gradient


×105 [Pa] ×105 [Pa] ×105 [Pa/s]
2 0.08 260
3 0.16 480
4 0.30 690
5 0.40 842

Table 4.3: Pressure difference at closed valve (valve starts to open) for different discharge
pressures.

52
are responsible for this. In order to quantify the contribution of both effects, following
measurement procedure has been conducted (figure 4.12). In the first measurement the
compressor was tested with a certain position of the valve plate. In detail, the sensors
VP1 and VP2 were located at Position A and Position B, respectively. Then the discharge
valve was rotated at 180◦ and again the valve motion recorded (sensor VP1 at Postion B
and sensor VP2 at position A). The influence of the asymmetric springing is represented
by the line AS which is given by

AS = xv,VP1, Pos. B − xv,VP2, Pos. B , (4.1)

and the line for the influence of the asymmetric flow AF is obtained by
   
AF = 0.5 xv,VP1, Pos.B + xv,VP2, Pos.B − xv,VP1, Pos.A + xv,VP2, Pos.A . (4.2)

It turns out that the effect of springing dominates (figure 4.13).

0.0014 VP1
VP2
VP3
VP average
0.0012

0.001
Valve lift [m]

0.0008

0.0006

0.0004

0.0002

0
100 120 140 160 180 200
Crank angle [◦ ]

Figure 4.11: Comparison of valve plate motion at sensor VP1, VP2 and VP3. Case: 1
chamber, pout = 4 bar, cs =48 N/mm, xv,max =1.35 mm.

Variation of springing In figure 4.14 the comparison of measurement data with dif-
ferent springing configurations (cs =48 N/mm and cs =80 N/mm) is shown. It turns out
that stiffer springing leads in this case to following results:

• Lower impact velocity of the valve plate: In this case the impact velocity is reduced
from 1.67 m/s for cs = 48 N/mm to 1.48 m/s in case of cs = 80 N/mm. This is true
for most compressors.
• The valve closes faster
• The valve plate tends to open and close more often and therefore excite more often
pressure waves.

53
1. Measurement 2. Measurement
discharge valve

Pos. A 1 2

Pos. B 2 1

piston

1 . . . VP1 (distance sensor)


2 . . . VP2 (distance sensor)

Figure 4.12: Measurements of valve plate motion with sensor VP1 and VP2.

1.5
VP1 Pos. A
VP2 Pos. B
VP1 Pos. B
VP2 Pos. A
asymetric springing
1 asymetric flow
Valve lift [m]

0.5

−0.5
120 140 160 180 200 220
Crank angle [◦ ]

Figure 4.13: Askew valve plate: Different contributions. Case: 1 chamber, pout = 4 bar,
cs =48 N/mm, xv,max =1.35 mm.

54
500000 0.0025
pc5 c=48 N/mm
pc5 c=80 N/mm
valve lift, c=48 N/mm
400000 valve lift, c=80 N/mm 0.002
Relative pressure at pc5 [Pa]
300000 0.0015

Valve lift [m]


200000 0.001

100000 0.0005

0 0

−100000 −0.0005
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 4.14: Comparison of relative pressure at pc5 and valve lift. Cases: pout = 4 bar,
cs =48 N/mm and cs =80 N/mm, xv,max =1.35 mm.

Variation of maximum valve plate lift xv,max The variation of the valve plate lift
xv,max from 1.05 mm to 1.65 mm has very little effect on the pressure distribution inside the
cylinder (figure 4.15). Whereas the impact velocity of the valve plate increases markedly
with higher maximum valve plate lifts xv,max since the plate is accelerated over a larger
distance (see table 4.1.2).

Maximum valve plate lift xv,max [m] Impact velocity [m/s]


1.05 1.36
1.35 1.67
1.65 1.91

Table 4.4: Impact velocities for different maximum valve plate lifts.

55
500000 0.0025
pc5 xvmax=1.05 mm
pc5 xvmax=1.35 mm
pc5 xvmax=1.65 mm
400000 valve lift, 1.05 0.002
Relative pressure at pc5 [Pa]
valve lift, 1.35
valve lift, 1.65
300000 0.0015

Valve lift [m]


200000 0.001

100000 0.0005

0 0

−100000 −0.0005
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 4.15: Comparison of relative pressure at pc5 and valve lift. Cases: pout = 4 bar,
cs =48 N/mm, xv,max = 1.05 / 1.35 / 1.65 mm

4.2 Compressor with eight valves

A double-acting, 1 cylinder, barrel design reciprocating compressor with eight valves was
tested. The experiment was carried out by Ariel Compression Ltd. The main specifications
of the compressor can be found in table 4.5. Various sensors were used to measure pressure
inside the cylinder and at the suction and discharge flange (figure 4.16).

Type of compressor: Ariel Compression JGD26.5


1 cylinder, double-acting, barrel design
Diameter of cylinder: 0.6731 m
Stroke: 0.1397 m
Speed of crankshaft: 1182 rpm
Gas: Nitrogen
Suction pressure: 2.2 · 105 Pa
Discharge pressure: 6.4 · 105 Pa
Valves: 8
Classification ǫ: 0.65

Table 4.5: Main Specifications of the Ariel test compressor

The pressure readings of the various sensors can be found in figure 4.17. Note that
the locations of the pressure sensors differ from the Burckhardt test compressor. Most
of the statements made above for the Burckhardt test compressor can be applied to this
compressor as well. However, since this compressor is almost three times bigger and has
higher overall mass flow rates some significant differences compared to the Burckhardt

56
suction valves discharge valves
5 and
valve retainers

1 . . . . . . pc1
2 . . . . . . pc2
2 4 3 . . . . . . pc3
4 . . . . . . pc4
5 . . . . . . pc5
pout at discharge flange
pin at suction flange
3

Figure 4.16: Location of sensors inside cylinder

compressor can be identified. Therefore the measurements of the Ariel compressor are
discussed in the following. Although the gas can leave the cylinder through four valves,
the maximum measured pressure inside the cylinder exceeds the discharge pressure by more
than 1.4 bar. This is due to the fact that high mass fluxes through the valves increase the
pressure in the valve retainers and discharge pipe markedly resulting in hindered outflow of
gas. In addition, the pressure differences between suction and discharge side in the cylinder
are very high since the diameter of the cylinder and thus the distance from the suction
side to the discharge valves is large. Note that these differences of the test compressors
can be well described by the classification number ǫ. In case of the Ariel test compressor
ǫ is 0.65, which is almost three times bigger than the ǫ of the Burckhardt test compressor.
Although the pressure sensor pc1 is located in the vicinity of pc3, the pressure read-
ings differ during discharge. We assume that the sensor pc1 in the port records total
pressure whereas the sensor pc3 registers the static pressure only. Therefore the difference
corresponds to the dynamic pressure.
Another difference to the Burckhardt test compressor is that the pressure waves in the
discharge pipe do not fade away. This behaviour can not be described by modelling the
valve retainer and pressure chamber only. Instead the whole discharge system, consisting
of discharge valve retainer, pressure chamber, discharge pipe, and damper, must be con-
sidered. However for calculating the gas flow in the cylinder it is sufficient to give a good
approximation of the pressure at the downstream side of the discharge valves. This issue
is discussed in detail in section 5.2.

57
900000
pin pc1
pout pc2
800000 pc3
pc4
700000
Absolute pressure [Pa]

600000

500000

400000

300000

200000

100000

0
−150 −100 −50 0 50 100 150

Crank angle [ ]

Figure 4.17: Measurement data of absolute pressure at different sensors.

58
Chapter 5

Comparison of Measurement and


Numerical data

In this chapter the quasi one-dimensional and two-dimensional simulation results are com-
pared to measurement data of the test compressors. The deviations are discussed and
important physical effects highlighted.

5.1 One-dimensional Model

5.1.1 Comparison with Measurement Results of the Burckhardt test


compressor

The quasi-onedimensional model is employed in order to obtain the numerical results.


For modelling the pressure chamber it is assumed that the pressure variations inside the
discharge pipe are negligible small and thus the second approach VR2 is chosen where the
pressure chamber is connected to the valve retainer by a T-piece and the outflow boundary
is located at the end of the pressure chamber. The specifications of the test case calculated
can be found in table 5.1.
Note that from now on all pressure results are displayed in absolute pressure. The
dead centre with the smallest cylinder volume is at 180◦ CA.

Number of active Discharge Pressure Valve springing Maximum valve lift


Working Chambers po ut [Pa] cs [N/m] xv,max [m]
1 5 · 105 48 · 103 1.35 · 10−3

Table 5.1: Specifications of the test case

In Figure 5.1, 5.2 and 5.3 the comparisons of the measured pressures at the discharge
side (pc5), the suction side (pc6) and in the valve retainer (pout) with the numerical
solution of the quasi one dimensional model are given. In addition, the motion of the
discharge valve obtained by numerical simulation and measurements can be found in the
same figures. The pressure difference between discharge side (pc5) and cylinder centre
(pc1) is displayed in figure 5.4. In the following we will discuss the solutions in detail.

59
600000 0.003
pressure pc5, Num.
pressure pc5, Exp.
valve lift, Num.
500000 valve lift, Exp. 0.0025
Absolute pressure at pc5 [Pa]

400000 0.002

Valve lift [m]


300000 0.0015

200000 0.001

100000 0.0005

0 0
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 5.1: Burckhardt test compressor: Comparison of absolute pressure at pc5 and valve
lift.

700000 0.0035
pressure pc6, Num.
pressure pc6, Exp.
600000 valve lift, Num. 0.003
valve lift, Exp.
Absolute pressure at pc6 [Pa]

500000 0.0025
Valve lift [m]
400000 0.002

300000 0.0015

200000 0.001

100000 0.0005

0 0
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 5.2: Burckhardt test compressor: Comparison of absolute pressure at pc6 and valve
lift.

60
pout numerical
540000 pout experimental
Absolute pressure at pout [Pa]

520000

500000

480000

460000

0 50 100 150 200 250 300 350


Crank angle [◦ ]

Figure 5.3: Burckhardt test compressor: Comparison of absolute pressure at pout.

60000
diff. pressure pc5−pc1, num.
diff. pressure pc5−pc1, meas.
40000

20000
Valve lift [m]

−20000

−40000

−60000
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 5.4: Burckhardt test compressor: Comparison of difference pressure pc5-pc1.

61
Compression The gas velocities in the cylinder are very small and an isentropic com-
pression takes places. The almost uniform pressure inside the cylinder (figure 5.5) is a
function of the actual cylinder volume and the initial pressure and volume at the begin-
ning of the compression only. In other words differences in pressure between simulation
and measurement during compression can be traced back to the fact that the compression
volume is not specified correctly.

Discharge When the pressure in the cylinder exceeds the pressure in the valve retainer
plus the springing the valve starts to open and a wave is initiated (figure 5.5). The
rarefaction wave, which is initiated at CA = 142◦ , reaches the suction side at CA = 146◦ .
Note that in undisturbed regions the isentropic compression still takes place. However,
the following complicated interaction of the valve motion and the pressure waves is well
represented by the numerical model. Most notably the vale motion is given correctly. Also
the calculated pressure differences in the cylinder, which are the cause for the moment
onto the piston, matches the measured ones.
Due to the outflow of gas through the valve, the pressure in the valve retainer rises.
Again this increase is represented well by the model.
We observe that at the end of the discharge the calculated amplitudes of the pressure
waves are bigger than the measured ones. Resulting in higher pressure differences across
the cylinder in the simulation. In other words, the one-dimensional model overestimates
the pressure waves and thus the moment on the piston cannot be determined accurately
in this stage of the discharge phase.

Expansion We observe that the calculated amplitudes of the pressure waves are higher
than the measured ones. Whereas the frequencies agree. Furthermore, an agreement of
the decay of the amplitudes, due to the increase of volume, can be observed.
The decrease of mean pressure inside the cylinder is not well represented by the numer-
ical model. More precisely, the pressure decreases slower in the model. In order to find the
cause of the deviations full three dimensional simulations have been carried out. It turns
out that the remaining gas runs through the cylinder at higher velocities as compared to
the compression phase. However heat transfer from the gas to the surrounding and dissi-
pation can still be neglected. Hence, similar to the compression phase the mean pressure
is only a function of volume (for given initial pressure and volume at the beginning of
the expansion), and thus the mass inside the cylinder must be higher at the time when
the discharge valve closes. Two different causes can be identified. Firstly the minimum
clearance Z0 is smaller than specified for the calculation resulting in smaller volumes at
the start of the expansion. Secondly higher amplitudes of the pressure waves lead to more
mass inside the cylinder. Since the calculated pressure during compression matches the
measured one perfectly the second cause is most likely responsible for different slopes of
the pressure during expansion.

Suction All statements made for the discharge phase concerning pressure waves are true
for the suction phase as well. However, since the overall pressure level is low compared
to the discharge phase, the resulting pressure waves have very low amplitudes. Therefore
problems arising from the pressure difference across the cylinder cannot be found during
the suction phase. The gas velocities inside the cylinder are rather small during the suction

62
phase.

cylinder head recess


valve pocket
cylinder valve retainer

Absolute pressure ×105 [Pa]


6.5
140° CA
142° CA
144° CA
146° CA
6

5.5

5
0 0.1 0.2 0.3
x [m]

Figure 5.5: Initiation of waves: Pressure distribution inside the cylinder and in the valve
retainer at the beginning of the discharge

5.1.2 Super elevation of the pressure due to initial sticking

Comparing the pressure reading pc5 close to the discharge valve with the corresponding
numerical solution, where the initial sticking is not taken into account, shows that the
opening of the valve is predicted too early (figure 5.6). As a consequence the pressure
maximum at pc5 is about 0.3 bar to small. With other words in reality the opening of
the valve does not start when the pressure in the cylinder balances the pressure in the
retainer and the springing. An additional delay occurs. In [8] the dynamic behaviour of a
valve plate is analysed. If the initial gap is very small, say 1-5 µm, viscosity hinders the
gas to flow into the valve gap resulting in a local pressure drop in the valve gap. Applying
Reynolds lubrication theory, an additional force Fadh holding back the valve plate can be
identified. It turns out that the result depends on the initial gap width, which cannot be
measured easily. In this case a initial gap width of 2 µm has been chosen to fit the data.

5.1.3 Pressure loss at sudden changes of cross section

The pressure at pc5 predicted by the numerical solution, wherein the pressure loss has been
neglected, drops after the first maximum at a crank angle of 144 degrees to a minimum at
CA 156◦ (figure 5.7). This minimum is about 5 · 104 Pa smaller than that of the measured
data. At the second minimum at a crank angle of 180◦ (dead centre) the difference is even
worse (8 · 104 Pa). An explanation for this difference is the following. During outflow the
gas is compressed under the piston. There it flows through a narrow channel towards the
valve pocket. At the entrance of the valve pocket the gas faces a sudden increase of the

63
Numerical without adhesion Numerical without pressure losses
Numerical with adhesion Numerical with pressure losses
Experiment Experiment
600000 600000

550000 550000
Pressure pc5 [Pa]

Pressure pc5 [Pa]


500000 500000

450000 450000

400000 400000
120 130 140 150 160 170 180 190 200 120 130 140 150 160 170 180 190 200
Crank angle [◦ ] Crank angle [◦ ]

Figure 5.6: Delay of valve opening due to Figure 5.7: Effect of pressure loss at sud-
viscosity in valve gap den change of cross-section

effective cross section of the flow channel when the piston is close to the dead centre. This
sudden increase of the cross section is physically associated with a pressure loss.
Although the modelling of the pressure losses at sudden changes of cross-section in-
creases the quality of the numerical solution in the vicinity of the dead centre (180◦ )
markedly, the deviations of pressure at the first minimum remain.

5.1.4 Different locations of outflow boundary

Different approaches for the valve retainer and pressure chamber have been introduced
in section 2.1.2. The comparison of the measured pressure in the valve retainer with the
solutions of the two different numerical models is given in figure 5.8. In the following we
will discuss the differences in detail.

• First approach VR1 (figure 2.4): The pressure chamber is added to the subdomain
of the valve retainer and the outlet boundary is set at the end of the valve retainer.
This model describes the first peak of the pressure in the valve retainer (pout) well,
although the amplitude is too high. However, the gas leaves the valve retainer almost
immediately and thus the pressure oscillates about the discharge pressure. During
expansion the pressure is damped due to the pressure loss at the outflow.

• Second approach VR2 (figure 2.5): Here the pressure chamber is replaced by a
cylinder connected to the valve retainer using a T-piece. The calculated pressure in
the valve retainer follows the measured one since the gas is hindered to flow out of the
valve retainer. However, the small oscillations of the pressure cannot be described
well by the simulation. These small variations have only a minor influence on the
pressure distribution inside the cylinder.

Observing the pressure at the discharge side of the cylinder (pc5) shows that both
approaches give good results and differences are rather small (figure 5.9). Whereas the

64
valve motion shows different behaviour at the second opening of the valve. However in
terms of pressure distribution inside the cylinder these differences play a minor role, since
during this stage of the discharge phase only a small amount of gas flows through the
valve.

pout, Num., VR2


540000 pout, Num., VR1
pout experimental
Absolute pressure at pout [Pa]

520000

500000

480000

460000

0 50 100 150 200 250 300 350


Crank angle [◦ ]

Figure 5.8: Comparison of absolute pressure at pout for different models of the pressure
chamber.

5.1.5 Impact velocity

The impact velocity of the valve plate onto the valve seat can be extracted from the valve
motion. In figure 5.11 the calculated and measured impact velocities for different discharge
pressures are compared. The difference between measurement and simulation is less than
15%.

5.1.6 Valve losses

In figure 5.10 the computed mean pressure p is shown in the p,V -diagram. The shaded
area corresponds to losses at the discharge valve. In the standard case defined above the
valve losses amount 5.3% of the total input energy.

5.2 Two-dimensional model

5.2.1 Comparison with measurement results of the Ariel test compressor

The two-dimensional model is employed to simulate the Ariel test compressor specified in
section 4.2. The pressure chamber of this compressor, where the gas flows from four valve

65
600000 0.003
pc5, Num., VR2
pc5, Num., VR1
pc5, Experimental
500000 valve lift, Num., VR2 0.0025
valve lift, Num., VR1
Absolute pressure at pc5 [Pa]

valve lift, Exp.


400000 0.002

Valve lift [m]


300000 0.0015

200000 0.001

100000 0.0005

0 0
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 5.9: Comparison of absolute pressure at pc5 and valve lift for different models of
the pressure chamber.

7 2.5
averaged pressure ×105 [Pa]

valve losses
6
impact velocity [m/s]

2
5

4 1.5

3 1
2
0.5
1 numerical
measurement
0 0
0 1 2 3 4 5 0 1 2 3 4 5
volume [dm3 ] relative discharge pressure ×105 [Pa]
Figure 5.10: Valve Losses at discharge Figure 5.11: Impact velocity of discharge
valve valve depending on discharge pressure

66
retainers merge, is far too complicated to be modelled as a simple cylinder. Therefore,
the first approach VR1 for the pressure chamber is chosen. The measured pressures at
different locations inside the cylinder are compared to simulated ones (figures 5.12 and
5.13). Furthermore the pressure difference between suction and discharge side (pc2-pc4)
is displayed in figure 5.14. Note that in the following diagrams the dead centre with the
smallest volume is at 0◦ CA.
900000
pressure pc4, Num.
pressure pc4, Exp.
800000 pressure pin, Num.
pressure pin, Exp.
700000

600000
Pressure [Pa]

500000

400000

300000

200000

100000

0
−150 −100 −50 0 50 100 150
Crank angle [◦ ]

Figure 5.12: Ariel test compressor: Comparison of pressure distribution at pc2 and pout

In the following the four stages of the compression cycle will be discussed:

Compression The computed pressure inside the cylinder increases slightly faster than
the measured one. As pointed out in section 5.1.1 this can be attributed to errors in
the specification of the volume. Since the calculated slope of the pressure is steeper, the
clearance volume, which has been derived from drawings provided by Ariel and specified
in the numerical simulation, must be smaller than the real one.

Discharge The discharge valves start to open simultaneously at −73◦ CA (figure 5.20).
During discharge the calculated pressures inside the cylinder are about 0.7 bar higher
than the measured ones. In addition we observe that this difference varies only slightly.
This issue will be discussed in detail in section 5.2.2.
The pressure difference between suction side and discharge side is well represented by
the model. Not only the frequency but also the amplitudes are calculated almost correctly.
However at the end of the discharge phase the calculated second peak is lower than the
measured one. Moreover from this point on the calculated pressure difference is about 8◦
CA ahead. The moment onto piston is derived from the pressure acting on the piston.
Due to the good approximation of the pressure difference we expect the computed moment

67
900000
pressure pc2, Num.
pressure pc2, Exp.
800000 pressure pout, Num.
pressure pout, Exp.
700000

600000
Pressure [Pa]

500000

400000

300000

200000

100000

0
−150 −100 −50 0 50 100 150
Crank angle [◦ ]

Figure 5.13: Ariel test compressor: Comparison of pressure distribution at pc4 and pin

200000
diff. pressure pc2−pc4, Num.
diff. pressure pc2−pc4, Exp.
150000

100000
Pressure difference [Pa]

50000

−50000

−100000

−150000

−200000
−150 −100 −50 0 50 100 150
crank angle [◦ ]

Figure 5.14: Ariel test compressor: Comparison of difference pressure pc2-pc4.

68
to approximate the real one also well (section 5.2.4).

Expansion The computed pressures inside the cylinder decrease faster than the mea-
sured ones. Similar to the compression phase we identify deviations in the clearance
volume as cause for the different slopes. However in this case different initial pressure
distribution at the beginning of the expansion phase leads to deviations in the pressure
readings during expansion too.
The calculated pressure difference between discharge side and suction side matches the
measured one with respect to frequency, amplitude, and damping. However, it is still 8◦
CA ahead.

Suction During suction the small pressure variations inside the cylinder are underes-
timated by the simulation model. It turns out that at the end of the suction phase the
measured mean pressure is slightly higher (0.1 bar) than the computed one. Again, the
pressure oscillations inside the suction pipe, which are neglected in the model, are respon-
sible for this.

5.2.2 Variation of discharge pressure

The results of the Ariel test compressor show that during discharge the calculated pressure
inside the cylinder is too high compared to the measured one. This can be explained as
follows. Firstly, the computed pressure inside the cylinder increases faster at the beginning
of the discharge phase than the measured one resulting in an overestimation of the pressure
during discharge, and secondly the actual discharge pressure is lower than the specified one.
In order to determine the influence of the first effect simulations have been performed where
the clearance volume has been adjusted in a way so that the computed pressure increase
during compression fits the measured one. However, it turns out that the difference in the
pressure rise at the beginning of the discharge can be neglected. In order to evaluate the
second explanation we take a closer look at the discharge system and its numerical model.
From the measured pressure readings at pout we can see that in this case the amplitudes
of the pressure waves inside the discharge system do not decay. Moreover the difference
between pressure maximum and minimum at the discharge flange is 1 bar. It can be
assumed that the pressure variations are of the same order in the valve retainer. So if it
happens that the discharge valves open during a pressure valley at the downstream side
of the valves then the actual discharge pressure would be lower than the nominal one. On
the other hand the numerical simulation uses a model of the discharge system, where the
pressure waves are damped. Thus at the beginning of the discharge phase the numerical
simulation cannot provide a good approximation of the pressure at the downstream side
of the valve. However, by reducing the discharge pressure specified for the numerical
simulation, the model can mimic lower pressures at the downstream side of the valves. In
figure 5.15 the computed pressure for a reduced discharge pressure of 6.1 bar is compared
to measurement data and numerical results where the discharge pressure is 6.4 bar. It
turns out that the pressure inside the cylinder is reduced and the trend of the pressure
distribution is not changed. Therefore following conclusions may be drawn:

• The pressure variations in the discharge system, which cannot be represented by the
simplified numerical model, are responsible for the overestimation of pressure inside

69
the cylinder during discharge.

• The trend of the pressure distributions inside the cylinder is not influenced by small
variations of the discharge pressure. In other words the resulting pressure level in
the cylinder is only shifted by the variation of the discharge pressure.

• The simplified one-dimensional models of the valve retainer and pressure chamber
are not sufficient to compute the pressure distribution inside the discharge system
for the whole working period. However they are sufficient to describe the pressure
rise in the valve retainer during discharge.

• During discharge the pressure level inside the cylinder is determined by the pressure
at the downstream side of the cylinder.

In order to provide a good approximation of the pressure at the downstream side of the
valve the whole discharge system must be modelled. Furthermore the model must be valid
throughout the whole working periode. However the gas flow in the pressure chamber is
too complicated to be modelled by an one-dimensional approach, at least two-dimensional
approaches must be employed.

900000
press. pc4, Num. pout=6.4 bar
press. pc4, Num. pout=6.1 bar
800000 pressure pc4, Exp.
700000

600000
Pressure pc4 [Pa]

500000

400000

300000

200000

100000

0
−100 −50 0 50 100
Crank angle [◦ ]

Figure 5.15: Variation of discharge pressure: pout = 640000 / 610000 Pa

5.2.3 Pressure distribution inside the cylinder

The Ariel test compressor has 8 valves located at the circumference of the cylinder. The en-
tering and leaving of gas through the valves initiates pressure disturbances which strongly
deviate from plane waves. In fact a two-dimensional gas flow in the x, y-plane of the cylin-
der can be observed (figure 5.16). At four selected crank angles the pressure distributions
inside the cylinder are discussed in the following.

70
At -64◦ CA the discharge valves are already open and the pressure in the vicinity of
the discharge valves decreases due to the outflow of gas. On the other hand the gas in the
suction side and in the centre of the cylinder still undergoes isentropic compression. In
this regions the velocity is zero (figure 5.17). The pressure difference between suction and
discharge side is about 0.9 bar resulting in a turning moment onto the piston relative to
the cylinder centre of 2350 Nm (figure 5.18).
At -32◦ CA a maximum velocity of about 130 m/s occurs at the valve pocket entries
leading to the discharge valves. We observe that due to these high gas velocities the
pressure in the vicinity of the discharge valve pockets drop to 6.95 bar. Whereas the
highest pressure can be found at the stagnation point between the two rightmost discharge
valves.
Although the piston has reached the dead centre and reversed its movement, the valves
are still open at 3◦ CA. This is due to the fact that at this time the pressure peak is at
the discharge side of the cylinder. Moreover the expansion of the gas has not yet resulted
in a sufficient pressure drop.
At 26◦ CA the valves have already closed and the remaining pressure difference across
the cylinder of 0.8 bar drives the pressure waves back and forth in the cylinder. However
the amplitudes will be damped due to the decrease of mean pressure in the cylinder.

5.2.4 Moment onto the piston

The moment onto the piston is derived from the pressure distribution inside the cylinder.
The pressure difference results in a moment onto the piston, which is given by
Z
Ty = p(x, y)x dS . (5.1)
S

Here S denotes the area of the piston. In figure 5.18 the y-component of the turning
moment relative to the centre of the cylinder is displayed. There the trend of the moment
must follow the difference pressure. In case of special pressure distributions inside the
cylinder a relationship between turning moment Ty and maximum pressure difference
∆pmax can be found. For example, if the pressure is constant along y and linear along x
(figure 5.19), then the equation 5.1 yields
 3
1 d
Ty = π∆pmax . (5.2)
8 2
Note that all other components of the moment must vanish due to symmetry.

5.2.5 Valve Opening

Let us denote the discharge valve which is located closer to the suction side with Vp1
and one of the rightmost discharge valves with Vp2. Figures 5.20 and 5.21 show the valve
motion and mass flux of Vp1 and Vp2. Since the pressure in the cylinder is almost uniform
during compression, the discharge valves start to open simultaneously at −73◦ CA. They
stay open until the piston reaches the dead centre. It turns out that the mass flux through
the different valves differ only slightly. However, the discharge valve Vp1 located closer to
the suction side closes earlier due to the pressure distribution inside the cylinder.

71
(a) Shortly after opening of discharge valves (b) During discharge (CA = -32◦ )
(CA = -64◦ )

(c) Closing of the discharge valves (CA = 3◦ ) (d) During expansion (CA = 26◦ )

Figure 5.16: Ariel test compressor: Pressure distribution in the cylinder at various times.

72
(a) Shortly after opening of discharge valves (b) During discharge (CA = -32◦ )
(CA = -64◦ )

Figure 5.17: Ariel test compressor: Velocity magnitudes in the cylinder at various times.

4000 2
Difference pressure ×105 [Pa]

2000 1
Moment [Nm]

p
∆pmax

0 0

−2000 −1
Moment onto piston
Pressure difference pc2−pc4
−4000 −2
−60 −30 0 30 60 90
Crank angle [◦ ] − d2 d
2 x

Figure 5.18: Moment onto Piston Figure 5.19: Calculating the moment onto
the piston: Pressure distribution inside
cylinder

73
0.004 6
Valve lift Vp1 mass flux through Vp1
0.0035 Valve lift Vp2 mass flux through Vp2
5
0.003

mass flux ṁ [kg/s]


4
valve lift [m]

0.0025
0.002 3
0.0015
2
0.001
1
0.0005
0 0
−100 −80 −60 −40 −20 0 20 40 −100 −80 −60 −40 −20 0 20 40
crank angle [◦ ] crank angle [◦ ]
Figure 5.20: Valve motion for different Figure 5.21: Mass flux through different
discharge valves discharge valves

5.2.6 Valve Masking

Measurements at the Ariel test compressor have shown that valve masking leads to sep-
aration zones resulting in reduced effective flow cross-sections of the valve and pressure
losses at the valve pocket entries. In the two-dimensional model the pressure losses are
neglected. Only the valve flow area is reduced according to the distance between piston
and cylinder head. We observe that taking this effect into account hinders the gas to flow
out. Therefore the calculated pressure inside the cylinder is higher (figure 5.22).

900000
pc2, Num. with valve mask.
pc2, Num. without valve mask.
800000 pressure pc2, Exp.
700000

600000
Pressure [Pa]

500000

400000

300000

200000

100000

0
−100 −50 0 50 100
Crank angle [◦ ]

Figure 5.22: Effect of Valve Masking: Comaparison of pressure at suction side pc2

74
Chapter 6

Remarks on Numerical Schemes

This chapter is concerned with comparison of different numerical schemes and their so-
lution. Not only computational time but also accuracy of the different approaches are
discussed. In addition the limitations of the one-dimensional and two-dimensional ap-
proach is considered.

6.1 Comparison of full three-dimensional, quasi one-dimensional


and quasi two-dimensional numerical methods

In order to evaluate the simplified models the test case of section 5.1.1 has been simulated
using full three-dimensional, quasi one-dimensional and quasi two-dimensional numerical
methods. However to simplify matters the outflow boundary has been placed at the
downstream side of the discharge valve. That means that the effects of the discharge
system have been neglected.
For the three dimensional simulation the commercial CFD-code of Fluent Inc. is
employed. Here the three-dimensional Reynolds-averaged Navier-Stokes equations, the
continuity equation and the energy equation are solved by means of first order methods
in the flow domain of the cylinder. In addition an appropriate turbulence model must
be chosen. The simple valve models presented in section 2.5 have formed the in- and
outflow boundaries. The piston motion and thus the variation of the computational do-
main with time has been modelled using dynamic meshing. A detailed description of the
implementation can be found in [22].
In figure 6.1 the pressure inside the cylinder and the valve motion are compared. It
turns out that the simple models are capable of capturing the most important physical
effects, resulting in a very good agreement with the three dimensional model. Most notably
the valve motion is predicted accurately by the simplified models although differences can
be observed at the end of the discharge phase. However the differences of the models are
discussed in the following.

75
pressure pc5, Num. 1D
600000 pressure pc5, Num. 2D 0.003
pressure pc5, Num. 3D
valve lift, Num. 1D
Absolute pressure at pc5 [Pa]
500000 valve lift, Num. 2D 0.0025
valve lift, Num.3D

400000 0.002

Valve lift [m]


300000 0.0015

200000 0.001

100000 0.0005

0 0
−150 −100 −50 0 50 100 150
Crank angle [◦ ]

Figure 6.1: Comparison of different finite volume schemes: Pressure pc5 inside the cylinder
and valve motion.

6.1.1 Pressure waves

During discharge the interaction of the pressure waves and the valve motion is the domi-
nant physical effect inside the cylinder. We use the pressure difference across the cylinder
(pc6-pc5) to describe the properties of the pressure waves (figure 6.1). It turns out that
both simple models are capable of describing the first excitation of the rarefraction wave
due to the opening of the discharge valve. Most notably the amplitudes and the frequen-
cies match. However in the course of the discharge small deviations can be observed. Most
notably, at the end of the discharge phase the one-dimensional and two-dimensional model
differ from the three-dimensional model. This is due to the fact, that the distance between
cylinder head and piston is so small, that the resulting flow in the discharge valve pocket
and suction valve pocket is three-dimensional. This behaviour cannot be described by the
simple models. In case of the one-dimensional model the differences at the dead centre
(0◦ CA) causes the discharge valve to open a third time, resulting in another wave exci-
tation with high amplitudes. Finally, when the discharge valve stays close, the remaining
pressure waves run back and forth in the cylinder. However the amplitudes of the waves
decay on the one hand due to dissipation and on the other hand due to the increase of
the volume, whereas the effects of the dissipation are rather small. In numerical simula-
tions another effect takes place. The numerical dissipation of the used method leads to
an artificial damping. However the magnitude of the numerical dissipation decreases with
decreasing cell size ∆x. In other words with a sufficient fine grid the numerical dissipation
can be neglected. For a detailed description of this issue we refer to [13] and [30]. In table
6.1 the number of cells per diameter are listed for each numerical model. Since a very
fine grid has been used for the one-dimensional simulation the numerical dissipation can
be neglected and the decrease of the amplitudes correspond to the increase of the cylin-
der volume only. Although the two-dimensional mesh is finer than the three-dimensional

76
one, the numerical dissipation is higher. Hence the amplitudes of the waves decay faster.
The phase shift between the different models during expansion can be attributed to the
different behaviour at the end of the discharge phase.

100000
pc6−pc5, Num. 1D
pc6−pc5, Num. 2D
diff. press., Num.3D
Pressure Difference pc6-pc5 [Pa]

50000

−50000

−100000
−100 −50 0 50 100
Crank angle [◦ ]

Figure 6.2: Comparison of different finite volume schemes: Pressure difference pc6-pc5.

6.1.2 Computational Time

The computational time for the one-dimensional and two-dimensional simulation was mea-
sured on a Pentium 4 Mobile processor with 512 MB RAM. The simulation of one crank
shaft revolution took about two minutes and about four hours for the one-dimensional and
two-dimensional model, respectively. The three-dimensional calculations were performed
on an Athlon 64 X2 Dual Core 4800+ processor with 2 GB RAM. Here the computational
time of one crank shaft revolution was 2.5 days. In table 6.1 the calculation time and the
grid size are listed for each model.

Method Total number of cells Cells per Diameter Calculation Time per periode
1D 400 330 2 min
2D 6272 104 4 hours
3D 22035-102297 70 2.5 days

Table 6.1: Calculation Time

77
6.2 Limitations of the Numerical Schemes

6.2.1 Limitations of the one-dimensional model

The quasi one-dimensional model can not be applied on all of the test cases success-
fully. Although, for most of the compressor specifications a solution can be obtained,
the results must be questioned. In the following we will discuss different cases where the
one-dimensional model is only limited applicable.

Two-dimensional flow in the x, y-plane One main assumption of the quasi one-
dimensional model is that the thermodynamic state in each flow cross-section is uniform.
This is violated in case of compressors with more than two valves since multiple valve pock-
ets leads to a two-dimensional flow in the x, y-plane (figure 5.16). Hence the results of
the one-dimensional model are not satisfying. Also another problem arises when multiple
valve pockets have to be calculated: The two equivalent valve pockets which are derived
by adding up all discharge and suction valve pockets, have very large cross-sections. Es-
pecially, when the piston approaches dead centre the cross-sections of the cylinder volume
are rather small compared to the valve pockets. This may lead to supersonic flow in
the cylinder cross-sections and furthermore in connection with large jumps in flow areas
to the breakdown of the numerical calculation. It must be pointed out that supersonic
flow is not likely to occur in common compressors. For example, when employing the
one-dimensional model to the Ariel test compressor with eight valves the results show
supersonic flow before the simulation breaks down. Whereas the gas velocity computed
by the two-dimensional model never exceeds sonic speed.

Two-dimensional flow in the x, z-plane Again the thermodynamic state is not uni-
form in the cross-section at a time t instead it is a function of x and z. This flow can be
found in compressors with low compression ratios and in cylinders, where the flow cannot
follow the form of the cross-sections and separation zones are present (figure 2.16). In the
first case the discharge valve opens when the piston is far away from the dead centre and
the pressure disturbance coming from the valve pocket can expand in x and z direction,
resulting in a complicated wave pattern. In the second case the height Z may be small
during discharge but the geometry of the cylinder head recesses or the valve pockets leads
to separation zones or at least to cross-section areas where the thermodynamic state and
the gas velocity is not uniform. Here a two-dimensional model in the x,z-plane must
be employed. We refer to Meyer who used a commercial CFD code to modell this two-
dimensional problem [22]. It is worth mentioning that with above finite volume schemes
a two dimensional model in the x,z plane can be easily derived.
Figure 6.3 shows the comparison of the measured pressure and valve lift of the Burck-
hardt test compressor with simulation results. Here a discharge pressure of 2 bar is chosen
and thus the compression ratio is about 2. Although the initial stage of the discharge is
calculated correctly, later the calculated pressure distribution and valve motion show large
deviations from the measured one. Again the one-dimensional model fails to represent the
wave structure and at least two-dimensional approaches must be employed in order to
calculate this case.

78
300000 0.003
pressure pc5, Num.
pressure pc5, Exp.
valve lift, Num.
Absolute pressure at pc5 [Pa] 250000 valve lift, Exp. 0.0025

200000 0.002

Valve lift [m]


150000 0.0015

100000 0.001

50000 0.0005

0 0
0 50 100 150 200 250 300 350
Crank angle [◦ ]

Figure 6.3: Comparison of absolute pressure at pc5 and valve lift with discharge pressure
of 2 bar.

Three dimensional flow If a compressor features more than one different characteristic
stated above the cylinder flow will be three dimensional. Therefore the thermodynamic
state at a time t depends on x, y, z and thus three-dimensional models must be employed.

6.2.2 Limitations of the quasi two-dimensional model

Also the quasi two-dimensional model can not be applied to all compressor specifications.
In the following most problem cases will be described.

Gas flow depending on z If the thermodynamic state and gas velocity depends on
the z-coordinate, one main requirement for the two-dimensional model is violated and
three-dimensional model must be employed in order to calculate this flow accurately.

Sudden changes of height inside the cylinder Certain cylinder head recess lead to
sudden changes of height inside the cylinder and following problems may occur. Firstly the
thermodynamic state and the gas velocity depends on z and three-dimensional gas flows
are present. Secondly, as pointed out in section 2.4.2, the source terms in the momentum
equations (2.28) and (2.29) become very large, resulting in the break-down of the numerical
simulation. However, in this case following simplification of the cylinder geometry can be
used. The volume Vch of the cylinder head recess is included in the cylinder clearance.
The resulting cylinder clearance Z¯0 = Z0 + Vch /(d2 π/4) is only a function of t.

79
Chapter 7

Summary

Measurements conducted on two different test compressors showed that special attention
has to be paid for the interaction of the pressure waves and the valve motion in order to
simulate the gas flow and valve motion of barrel design reciprocating compressors accu-
rately. Most notably important design criteria such as impact velocity of the valve, the
valve losses, and the moments onto the piston are governed by these interactions.
In the case of compressors with two valves it turned out that considering plane waves
only is sufficient to capture the wave structure inside the cylinder. The resulting governing
equations were solved by means of the finite volume scheme F-wave which offers two ben-
efits: Firstly this numerical scheme directly incorporates the underlying wave propagation
structure and secondly the boundary conditions can be easily implemented. Finally sim-
ple quasi-static valve models complete the numerical model of a reciprocating compressor.
Since the model is one-dimensional, only very short computational times are required, even
with very fine grids. The comparison of the measurements with the simulation show, that
the one-dimensional model is capable of capturing the most important physical effects,
resulting in a very good agreement. Most notably the valve motion and the pressure dis-
tribution inside the cylinder and the valve retainer can be computed accurately. However,
the amplitudes of the pressure waves at the end of the discharge phase and during the
expansion are overestimated, since three-dimensional effects at the end of the discharge
cannot be described by this simplified model.
In the case of compressors with more than two valves a two-dimensional approach
was chosen in order to account for the multiple valve pocket entries, which lead to two-
dimensional flows in the cylinder. The valve pockets and the valve retainers were mod-
elled one-dimensionally. The resulting governing equations for the gas flow were solved
by means of the robust and simple finite volume scheme of Richtmyer. Again the com-
parison of the measurements with the simulation show a good agreement in terms of gas
flow and valve motion. Even the pressure difference across the cylinder is computed accu-
rately, resulting in an accurate solution for the moment onto the piston. However during
discharge the pressure inside the cylinder is overestimated, which is attributed to the pres-
sure variations in the valve retainers. In fact the simplified one-dimensional models of the
valve retainer and pressure chamber are not sufficient to compute the pressure distribu-
tion inside the discharge system throughout the whole working period. However they are
sufficient to describe the pressure rise in the valve retainer during discharge. In order
to correct this shortcoming we suggest to calculate the gas flow in the pressure chamber

80
two-dimensionally.
Due to the low computation time and accurate results the simplified models presented
in this work, namely the one-dimensional and two-dimensional model, can form the basis
for a engineering design tool, which predicts gas flow and valve motion in a reciprocating
compressor.

81
Appendix A

Tables

82
A.1 Sensors of Burckhardt Test Compressor

The Table A.1 lists all sensors and their appropirate coordinates in the Burckhardt test
compressor. The point of origin of the coordinate system is the top dead center of the
first cylinder.

Identifier Type of sensors Name Coordinates or Position


[mm/mm/mm]
TRIG light trigger flywheel
VP1 distance valve plate clearance 1 valve plate
VP2 distance valve plate clearance 2 valve plate
VP3 distance valve plate clearance 3 valve plate
Ta temperature ambient air temperature -
Tst temperature suction tube 30 cm behind flange
Tsc temperature suction chamber -265/-30/0
Tpc temperature pressure chamber 265/0/-30
Tpt temperature pressure tube 30 cm behind flange
pc1 pressure cylinder pos. 1 0/0/0
pc2 pressure cylinder pos. 2 0/-33/0
pc3 pressure cylinder pos. 3 0/-66/0
pc4 pressure cylinder pos. 4 0/66/0
pc5 pressure cylinder pos. 5 75/0/0
pc6 pressure cylinder pos. 6 -75/0/0
pc7 pressure cylinder pos. 7 -75/-33/0
pin pressure suction chamber -265/0/-20
pout pressure pressure chamber 265/-30/0
po pressure pressure orifice orifice
dpo pressure pressure difference orifice orifice
To temperature temperature orifice orifice

Table A.1: Sensors

A.2 Pressure loss coefficients for the T-piece

83
( 2   2 " #)
ṁ1 2
  
A3 A3 ṁ1
ṁ1 ṁ2 K1 = 1+ +3 − ∗
A1 A1 ṁ3 ṁ3
A1 A2 pt3 − p3
∗fbr
pt1 − p1
ṁ3 K2 = K1 but subscript 1 is to be replaced by 2
K3 = 0
A3
fbr see table A.3

ṁ1 ṁ2
K1 = 0
A1 A2 pt1 − p1
K2 = −0.6
pt2 − p2
ṁ3  !
ṁ3 2 pt1 − p1

ṁ3
K3 = − 1 − 0.5 +2
ṁ1 ṁ1 pt3 − p3
A3

ṁ3 2 pt2 − p2
 
ṁ1 ṁ2 ṁ3
K1 = 1.55 −
ṁ2 ṁ2 pt1 − p1
A1 A2
K2 = 0
" #
ṁ3 A2 2
  
ṁ3 ṁ3 pt − p2
K3 = 1+ −2 1− fbr 2
ṁ2 A3 ṁ2 pt3 − p3
A3 fbr see table A.3 but subscript 3 is to be replaced by 2

"  2 #
ṁ1 ṁ2 u1 pt3 − p3
K1 = − 1 + 1.5
A1 A2 u3 pt1 − p1
K2 = K1 but subscript 1 is to be replaced by 2
ṁ3
K3 = 0

A3

Table A.2: Pressure loss coefficients for the T-piece [14]

fbr
A1 /A3 ≤ 0.35 1
A1 /A3 > 0.35 ṁ1 /ṁ3 ≤ 0.4 0.9 (1 − ṁ1 /ṁ3 )
ṁ1 /ṁ3 > 0.4 0.55

Table A.3: Values for fbr [14]

84
Bibliography

[1] Aigner, R., Steinrück, H. (2004): Waves in reciprocating compressors. PAMM Proceedings in Ap-
plied Mathematics and Mechanics, 4(1), 516 - 517.
[2] Aigner, R.,Meyer G.,Steinrück H. (2005): Valve Dynamics and Internal Waves in a Reciprocating
Compressor. Proceedings of 4th EFRC Conference, 169 - 178.
[3] Aigner, R.,Steinrück H. (2007): Modelling Fluid Dynamics, Heat Transfer and Valve Dynamics in
a Reciprocating Compressor. Proceedings of 5th EFRC Conference, 169 - 178.
[4] Baehr, H. D., Stephan, K. (2004): Wärme- und Stoffübertragung. Springer.
[5] Costagliola M. (1950): The Theory for Spring Loaded Valves for Reciprocating Compressors. J.
Appl. Mech., 415-420.
[6] Ferziger, J. H., Perić, M. (2002): Computational Methods for Fluid Dynamics. Springer Berlin
Heidelberg.
[7] Flade, G., Steinrück, H. (2004): Anf”angliches ”Offnungsverhalten von Kompressorventilen. PAMM
Proceedings in Applied Mathematics and Mechanics, 4, 450-451.
[8] Flade, G. (2006): Weiterentwicklung der Berechnungsmethoden für Kobenverdichterventile auf der
Basis zweidimensionaler Strömungssimulation. Doctoral thesis, Technical University Dresden.
[9] Gersten, K., Herwig, H. (1992): Str”omungsmechanik. Verlag Vieweg.
[10] Godlewski, E., Raviart, P. A. (1996): Numerical Approximation of hyperbolic systems of conserva-
tion laws. Volume 118 of Applied Mathemaical Scienses, Springer.
[11] Grau, U. (1996): Haftkräfte und deren Einfluss auf die Kinematik selbsttätiger Ventile oszilierender
Verdrängerpumpen. Doctoral thesis, Technical University Erlangen-Nürnberg.
[12] Habing, R. A., Peters, M. C. A. M. (2006): An experimental method for validating compressor valve
vibration theory. Journal of Fluids and Structures, Vol. 22, 683-697.
[13] Hirsch, C. (1992): Numerical Computation of internal and external flows. Computational methods
for inviscid and viscous flows. John Wiley & Sons
[14] Idelchik, I.E. (1994): Handbook of Hydraulic Resistance. CRC Press
[15] Lax, P. D., Wendroff, B. (1960): System of Conservation Laws. Comp. Pure and Appl. Math., 13,
217-237.
[16] Leveque, R. (2004): Finite Volume Methods for Hyperbolic Problems. Cambridge University Press.
[17] Leveque, R., Bale, D., Mitran, S., Rossmanith, J. (2002): A Wave Propagation Method for Con-
servation Laws and Balance Laws with spatially varying flux functions. SIAM J. Sci. Comput., Vol.
24, No. 3, 955-978.
[18] Langseth, J. O., LeVeque, R. J. (1997): A wave propagation method for multi-dimesnional hyper-
bolic conservation laws. Journal of computational Physics, Vol. 131, 327-353
[19] Machu, E. (1998): Problems with modern high speed short stroke reciprocating compressors: In-
creased power requirement due to pocket losses, piston masking and gas inertia, eccentric gas load
on the piston. Proceedings of Gas machinery conference USA.
[20] Machu, G. (2004): Calculating reliable impact valve velocity by mapping instantaneous flow in a
reciprocating compressor. Proceedings of gas machinery conference GMRC USA.
[21] Machu, G. (2005): Pusationen im Verdichtungsraum - eine potentielle Schadensursache. Indus-
triepunpen und Kompressoren Heft 2/2005 Vulkann Verlag Essen, Mai 2005.

85
[22] Meyer, G. (2004): Simulation der Strömung in einem Kolbenverdichter. Diploma thesis, Technical
University Vienna.
[23] Miller, D. S. (1990): Internal Flow Systems. BHRA (Information Services)
[24] Müllner, T. (2007): Wärmeübertragung in einem Kolbenkompressor. Unpublished Report, Technical
University Vienna.
[25] Nickol, J. (2000): 24000 Laufstunden Verfügbarkeit von Kobenkompressoren in Prozessanlagen -
Utopie oder Realität? Industriepunpen und Kompressoren Heft 2/2000 Vulkann Verlag Essen, Mai
2000
[26] Press, W. H., Flannery, B. P., Teukolsky, S. A., Vetterling W. T. (1986): Numerical Recipes. The
art of scientific computing. Cambridge University Press.
[27] Roe, P.L.E. (1981): Approximate Riemann Solvers. J. Comp. Phys. 43, 357-372.
[28] Serre, D. (1999): Systems of conservation laws 1. Hyperbolicity, Entropies, Shock Waves. Cambridge
University Press.
[29] Tannehill, J. C., Anderson, D. A., Pletcher, R. H. (1997): Computational fluid mechanics and heat
transfer. Taylor and Francis.
[30] Toro, E. F., (1999): Riemann Solvers and Numerical Methods for Fluid Dynamics. Springer.
[31] Wibmer, M. (2004): One-dimensional Simulation for Arterial Blood Flow with Applications. Doc-
toral thesis, Technical University Vienna.
[32] Zierep, J. (1997): Grundzüge der Strömungslehre. Springer Berlin Heidelberg.

86
Curriculum Vitae

Personal Data

Name: Roland Aigner


Nationality: Austria
Date of Birth: November 13th , 1976
E-mail: roland aigner@gmx.net

Education

06/2004 - present Vienna University of Technology, Austria


Institut of Fluid Mechanics and Heat Transfer
• Doctorate in Mechanical Engineering
10/1996 - 01/2003 Vienna University of Technology, Austria
Department of Mechanical Engineering
• Diploma Thesis ”Formation of the strand for continuous
casting of steel”
• Graduation with honours as Master of Science (Diplomingenieur)
09/2000 - 01/2001 University of Salford, Salford (Greater Manchester), UK
Department of Mechanical Engineering
• Exchange semester
09/1991 - 06/1996 Höhere Technische Bundeslehranstalt Wels, Austria
Department of Mechanical Engineering
• Graduation with distinction (Matura)

Professional Experience

06/2003 - 06/2007 Vienna University of Technology, Austria


Institut of Fluid Dynamics and Heat Transfer
Teaching, Supervisor of exercise courses:
• Heat Transfer
• Numerical Methods for Fluid Mechanics and Heat Transfer
Research / Industrial Projects:
• Hoerbiger Valve Tec GmbH: ”Numerical Simulation of
a pipe branching wye”
• Berndorf AG: ”Heat Transfer of a rotating cylinder”
• European Forum for Reciprocating Compressors: ”Heat
Transfer in a Reciprocating Compressor”
08/1999 - 09/1999 General Motors Austria GmbH, Vienna, Austria
• Intern
07/1995 - 08/1995 BRP-Rotax GmbH & Co. KG, Gunskirchen, Austria
• Intern

87

Das könnte Ihnen auch gefallen