Persoons

KATHOLIEKE UNIVERSITEIT LEUVEN
FACULTEIT INGENIEURSWETENSCHAPPEN
DEPARTEMENT WERKTUIGKUNDE
AFDELING TOEGEPASTE MECHANICA
EN ENERGIECONVERSIE
Celestijnenlaan 300A, B-3001 Leuven, België
EXPERIMENTAL FLOW DYNAMICS

IN AUTOMOTIVE EXHAUST SYSTEMS
WITH CLOSE-COUPLED CATALYST
Jury: Proefschrift voorgedragen tot

Prof. dr. ir. A. Haegemans, voorzitter het behalen van het doctoraat
Prof. dr. ir. E. Van den Bulck, promotor in de ingenieurswetenschappen
Prof. dr. ir. M. Baelmans
door
Prof. dr. ir. P. Sas
Prof. dr. ir. R. Sierens (Universiteit Gent, B) Tim PERSOONS
Prof. dr. ir. R. Baert (Technische Universiteit Eindhoven, NL)
Prof. dr. S. Benjamin (Coventry University, UK)
U.D.C. 621.43 Augustus 2006

c Katholieke Universiteit Leuven - Faculteit Ingenieurswetenschappen
Arenbergkasteel, B-3001 Leuven, België
Alle rechten voorbehouden. Niets uit deze uitgave mag worden verveelvoudigd en/of openbaar gemaakt
worden door middel van druk, fotokopie, microfilm, elektronische of op welke andere wijze ook zonder
voorafgaandelijke schriftelijke toestemming van de uitgever.
All rights reserved. No part of this publication may be reproduced in any form by print, photoprint,
microfilm, or any other means without written permission from the publisher.
Wettelijke Depot: D/2006/7515/47

ISBN 90-5682-713-8
iii
Voorwoord
De inwendige verbrandingsmotor “. . . the greatest evil ever visited upon Man-
kind.” volgens J. R. R. Tolkien, maar wát een onderzoeksobject! Met veel en-
thousiasme begon ik in 1999 als onderzoeker bij Prof. Eric Van den Bulck te
werken op de regeling van motoren. Twee jaar later kreeg ik de kans om het fas-
cinerende domein van de stromingsmechanica te combineren met mijn interesse
voor motoren.
Als promotor verdient Eric een serieuze pluim, en niet alleen voor zijn on-
geëvenaarde fysische achtergrondkennis. Hij slaagt erin keer op keer de juiste
balans te vinden tussen onderzoeksvrijheid en de zó noodzakelijke sturende
acties. Altijd is er tijd voor een babbel, een illustratieve uitleg of een intrige-
rende vraag. Eric geeft de ultieme invulling aan het concept promotor, en hoe
toepasselijk is dat voor een thesis rond katalysatoren?
Werken bij TME is altijd een aangename ervaring geweest. De verschillende
collega’s, secretaresses en techniekers vormen een hechte groep. Bedankt alle-
maal, voor de humor, de waardevolle en -loze discussies, en voor de steun op
de momenten dat het minder plezant was. Tegen het einde van dit boek aan,
komt vooral de levenswijsheid aangereikt door een illustere technieker terug
boven “. . . ’t leste komt alles onder ’t dak”.
Een dikke merci aan mijn vrienden buiten het werk – burgies en andere,
aan mijn vriendin Nathalie, mijn broer en zussen, ouders en grootouders om
er altijd te zijn met raad en daad, en de occasionele dosis realiteitszin buiten
de academische wereld. In het bijzonder, bedankt mama en papa, om mij
altijd mijn eigen keuzes te laten maken, hoe absurd ook: “zoudt ge niet beter
industrieel doen, daar is toch meer vraag naar”, en vijf jaar later: “doctoreren?
dat verdient toch niet”. Maar serieus: merci!
Nathalie, mijn allerliefste, gij maakt alles meer dan dubbel de moeite waard.
Uw dankwoord moest het langst worden, maar woorden schieten te kort.
Tim Heverlee, augustus 2006

iv
Abstract
Increasingly stringent vehicle emissions legislation leads to the integration of
the close-coupled catalyst into the automotive exhaust manifold. Obtaining a
uniform catalyst velocity distribution is not straightforward, yet remains crucial
for optimal catalyst operation, in terms of maximizing pollutant conversion
efficiency, minimizing pressure loss and avoiding local catalyst degradation.
This thesis has developed an experimental methodology to obtain time-
resolved bidirectional velocity distributions with high spatial and temporal res-
olution, suitable for validation of computational fluid dynamics.
The charged motored engine flow rig is developed, which generates pulsat-
ing flow in the exhaust system, similar to fired engine operation yet at ambient
temperature. The setup is analyzed in terms of the exhaust stroke flow simi-
larity with fired engine conditions.
A novel oscillating hot-wire anemometer is developed to measure local in-
stantaneous bidirectional velocity, and has been successfully applied to measure
flow reversal in a close-coupled catalyst manifold.
The validity of the addition principle has been statistically established, in
terms of the dimensionless scavenging number S. A critical value Scrit marks
the validity limit of the addition principle, and may be interpreted as the
collector efficiency in terms of catalyst flow uniformity.
The highly transient flow dynamics in the close-coupled catalyst manifold
are studied experimentally, based on time-resolved full catalyst cross-section
velocity distributions, revealing extensive periodic flow reversal and strong res-
onance fluctuations.
A one-dimensional gas dynamic model of the exhaust system has been val-
idated and used to predict the flow dynamics outside the scope of the experi-
mental setup. Numerically determined frequency response functions facilitate
the understanding of the observed gas dynamic resonance phenomena.
v
Korte samenvatting
Steeds strengere emissiewetgeving voor voertuigen maakt de voorkatalysator
tot een geïntegreerd deel van de uitlaatcollector van verbrandingsmotoren.
Stromingsuniformiteit in de katalysator is moeilijk te bekomen, maar cruciaal
voor een optimale katalysatorwerking met het oog op maximaal omzettings-
rendement, minimale drukval en het vermijden van lokale degradatie.
Deze thesis ontwikkelde een experimentele methodologie voor het bekomen
van tijdsafhankelijke bidirectionele snelheidsverdelingen met hoge resolutie in
ruimte en tijd, geschikt voor de validatie van computational fluid dynamics.
De ontwikkelde stromingsopstelling met opgeladen aangedreven motor ge-
nereert in het uitlaatsysteem een pulserende stroming, gelijkaardig aan een
werkende motor, maar bij omgevingstemperatuur. De stroming is geanalyseerd
wat betreft gelijkvormigheid met werkende motorcondities.
Een oscillerende hittedraad anemometer is ontwikkeld om lokale ogenblik-
kelijke bidirectionele snelheid te meten. Deze is gekalibreerd en toegepast om
terugstroming te meten in een uitlaatcollector met voorkatalysator.
De geldigheid van het additieprincipe is statistisch ondersteund in relatie
tot het dimensieloze scavenging getal S. Een kritische waarde Scrit bepaalt de
geldigheidsgrens van het additieprincipe, en kan geïnterpreteerd worden als een
collectorefficiëntie met betrekking tot stromingsuniformiteit in de katalysator.
De stromingsdynamica in de uitlaatcollector is experimenteel onderzocht,
aan de hand van tijdsafhankelijke snelheidsverdelingen in de volledige katalysa-
tordoorsnede. De katalysator is onderhevig aan sterke periodieke terugstroming
en snelheidsfluctuaties, ten gevolge van gasdynamische resonanties.
Een gasdynamisch ééndimensionaal model van het uitlaatsysteem is gevali-
deerd, en voorspelt de stroming buiten het werkingsgebied van de experimen-
tele opstelling. De experimenteel waargenomen gasdynamische resonanties zijn
verklaard door numeriek bepaalde transfer functies.
List of symbols
Symbols
A Cross-sectional area [m2 ]

b Cylinder bore [m]
Cd Discharge coefficient [-], for flow through exhaust valves
c Speed of sound [m/s]
cp , cv Specific heat capacity of air at constant pressure and volume [J/(kg K)]
d Diameter [m]
e, E Index and number of ensembles [-]
f Frequency [Hz]
h Height [m]
i, I Index and number of measurement points [-]
j, J Index and number of crankshaft positions [-]
k Spring stiffness [N/m]
L Length [m]
M Mass flow rate [kg/s]
m Mass [kg]
Ma Mach number [-]
N Engine crankshaft speed [rpm]
ne , n r Number of exhaust valves and runners [-]
P P-value of hypothesis test [-]
p Pressure [Pa]
Q Volumetric flow rate [m3 /s]
Re Reynolds number [-], usually based on runner diameter and mean run-
ner velocity
Rf OHW oscillation frequency, Rf = 120fo /N [-]
r Specific gas constant of air [J/(kg K)]
rS , r M Shape and magnitude similarity measure [-], Eqs. (4.12) and (4.28)
S Scavenging number [-], Eq. (4.42)
s Piston stroke [m]
T Temperature [K]
vii
viii List of symbols
Tp , Ts Flow pulsation and diffuser residence time scale [s], Eqs. (4.46)
and (4.43)
t Time [s]
U Velocity [m/s], along length axis of catalyst
V Volume [m3 ]
x, y Measurement plane coordinates [m]
xo OHW oscillation amplitude [m]
z Lengthwise coordinate [m]
Subscripts
a Ambient conditions
bb Blow-by leakage through piston-cylinder clearance
c Combustion
cyl Cylinder
d Diffuser
e Exhaust, or exhaust valve (opening)
H Helmholtz resonance
i Intake, or intake valve (closing)
m Mean, i.e. spatial area average
o Oscillating hot-wire (OHW)
p Hot-wire probe
r Exhaust runner
ref Reference
rel Relative
s Standard conditions, i.e. 101325 Pa and 273.15 K
Greek symbols
α OHW probe velocity tolerance factor [-]

∆, δ Absolute and relative 95 % uncertainty or deviation
∆θ Exhaust valve opening duration [rad-crankshaft]
η Flow uniformity or conversion efficiency [-]
γ Ratio of specific heats of air, γ = cp /cv [-]
µ, ν Dynamic [Pa·s] and kinematic viscosity of air [m2 /s]
ρ Density [kg/m3 ]
% Volumetric compression ratio [-]
θ Angular position [rad or ◦ ]
ω Angular velocity [rad/s]
Contents
Voorwoord iii
Abstract iv
Korte samenvatting v
List of symbols vii
Contents ix
1 Introduction: Exhaust systems 1

1.1 Legal background . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Aspects of manifold design . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Catalyst . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Deactivation . . . . . . . . . . . . . . . . . . . . . . . . 4
Conversion efficiency . . . . . . . . . . . . . . . . . . . . 6
Pressure loss . . . . . . . . . . . . . . . . . . . . . . . . 8
Design compromise . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Pressure loss . . . . . . . . . . . . . . . . . . . . . . . . 10
Flow dynamics . . . . . . . . . . . . . . . . . . . . . . . 13
Thermal load . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3 Motivation for this thesis . . . . . . . . . . . . . . . . . . . . . 17
1.4 Overview of manifold flow research . . . . . . . . . . . . . . . . 19
1.4.1 Flow distribution . . . . . . . . . . . . . . . . . . . . . . 19
Isothermal flow rigs . . . . . . . . . . . . . . . . . . . . 20
Isochoric flow rigs . . . . . . . . . . . . . . . . . . . . . 22
1.4.2 Flow dynamics . . . . . . . . . . . . . . . . . . . . . . . 23
Isochoric flow rigs: Fired . . . . . . . . . . . . . . . . . 24
Isochoric flow rigs: Motored . . . . . . . . . . . . . . . . 26
1.4.3 Contributions of this thesis . . . . . . . . . . . . . . . . 27
1.5 Goals and scope . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.6 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
ix
x Contents
2 Experimental approach 33
2.1 Exhaust manifolds . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Pulsating flow rigs . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.1 Isothermal flow rig . . . . . . . . . . . . . . . . . . . . . 36
2.2.2 Isochoric flow rig: Charged motored engine (CME) . . . 40
2.3 Exhaust stroke flow similarity . . . . . . . . . . . . . . . . . . . 44
2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.2 Thermodynamic analysis . . . . . . . . . . . . . . . . . 46
2.3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.4 Flow rate measurement . . . . . . . . . . . . . . . . . . . . . . 54
2.4.1 ISO orifice . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.4.2 Laminar flow element (LFE) . . . . . . . . . . . . . . . 56
2.4.3 Cylinder pressure . . . . . . . . . . . . . . . . . . . . . . 59
2.5 Data reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.5.1 Ensemble averaging . . . . . . . . . . . . . . . . . . . . 62
Ensemble-averaged quantities . . . . . . . . . . . . . . . 62
Uncertainty analysis . . . . . . . . . . . . . . . . . . . . 64
2.5.2 Cycle-resolved analysis . . . . . . . . . . . . . . . . . . . 64
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3 Oscillating hot-wire anemometer (OHW) 71

3.1 Introduction: Measuring bidirectional velocity . . . . . . . . . . 72
3.2 Hot-wire anemometer . . . . . . . . . . . . . . . . . . . . . . . 76
3.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.4 Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.5 Hot-wire probes . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.6 Calibration approach . . . . . . . . . . . . . . . . . . . . . . . . 85
3.6.1 Calibration wind tunnel . . . . . . . . . . . . . . . . . . 85
3.6.2 Laser Doppler anemometer . . . . . . . . . . . . . . . . 89
3.6.3 Calibration procedure . . . . . . . . . . . . . . . . . . . 89
3.7 Calibration results . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.7.1 Calibration charts . . . . . . . . . . . . . . . . . . . . . 90
3.7.2 Phase-locked results . . . . . . . . . . . . . . . . . . . . 93
3.7.3 Non-dimensional scaling analysis . . . . . . . . . . . . . 94
3.7.4 Discussion of the calibration results . . . . . . . . . . . 97
3.8 Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4 Addition principle 103

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.2 Validation approach . . . . . . . . . . . . . . . . . . . . . . . . 106
4.3 Data reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.3.1 Flow uniformity measures . . . . . . . . . . . . . . . . . 107
Weltens’ flow uniformity index . . . . . . . . . . . . . . 107
Mean-to-maximum velocity ratio . . . . . . . . . . . . . 109
4.3.2 Distribution similarity . . . . . . . . . . . . . . . . . . . 110
Contents xi
Shape similarity measure . . . . . . . . . . . . . . . . . 110

Spatial autocorrelation . . . . . . . . . . . . . . . . . . . 113
Magnitude similarity measure . . . . . . . . . . . . . . . 115
4.3.3 Flow characteristic . . . . . . . . . . . . . . . . . . . . . 122
4.4 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . 125
Manifold A . . . . . . . . . . . . . . . . . . . . . . . . . 127
Manifold B . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.4.2 Isochoric flow rig . . . . . . . . . . . . . . . . . . . . . . 138
Manifold B . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.4.3 Summary of the results . . . . . . . . . . . . . . . . . . 140
4.5 Interpretation of the results . . . . . . . . . . . . . . . . . . . . 144
4.6 Discussion: A physical interpretation . . . . . . . . . . . . . . . 149
4.6.1 Scalar mixing analogy . . . . . . . . . . . . . . . . . . . 149
4.6.2 Hypothesis: Collector efficiency . . . . . . . . . . . . . . 150
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5 Flow dynamics 155

5.1 Time-resolved flow distributions . . . . . . . . . . . . . . . . . . 156
Manifold A . . . . . . . . . . . . . . . . . . . . . . . . . 156
Manifold B . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.1.2 Isochoric flow rig . . . . . . . . . . . . . . . . . . . . . . 170
Mean velocity . . . . . . . . . . . . . . . . . . . . . . . . 170
Velocity distributions . . . . . . . . . . . . . . . . . . . 175
5.2 Flow reversal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.2.1 OHW validation . . . . . . . . . . . . . . . . . . . . . . 182
5.2.2 Experimental results . . . . . . . . . . . . . . . . . . . . 185
5.2.3 Numerical analysis . . . . . . . . . . . . . . . . . . . . . 187
Model description . . . . . . . . . . . . . . . . . . . . . 187
Results without cold end: Free discharge . . . . . . . . . 193
Results with cold end . . . . . . . . . . . . . . . . . . . 198
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.2.4 Discussion: Physical relevance of catalyst flow reversal . 202
5.3 Helmholtz resonance . . . . . . . . . . . . . . . . . . . . . . . . 204
5.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 204
5.3.2 Analytical model . . . . . . . . . . . . . . . . . . . . . . 205
5.3.3 Experimental results . . . . . . . . . . . . . . . . . . . . 208
5.3.4 Numerical results . . . . . . . . . . . . . . . . . . . . . . 213
Isothermal flow rig . . . . . . . . . . . . . . . . . . . . . 213
Isochoric flow rig: Without cold end . . . . . . . . . . . 214
Isochoric flow rig: With cold end . . . . . . . . . . . . . 218
5.3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 220
5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
xii Contents
6 Design considerations 225

6.1 Addition principle . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.2 Flow dynamics: Helmholtz resonance . . . . . . . . . . . . . . . 226
6.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
7 Conclusion 231
7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
7.2 Suggestions for future research . . . . . . . . . . . . . . . . . . 235
A Catalyst substrate flow 241

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
A.2 Momentum transfer . . . . . . . . . . . . . . . . . . . . . . . . 242
A.2.1 Fully developed laminar flow . . . . . . . . . . . . . . . 243
A.2.2 Developing laminar flow . . . . . . . . . . . . . . . . . . 243
A.2.3 Entrance and exit losses . . . . . . . . . . . . . . . . . . 245
A.2.4 Flow acceleration . . . . . . . . . . . . . . . . . . . . . . 246
A.2.5 Oblique entrance . . . . . . . . . . . . . . . . . . . . . . 247
A.2.6 Overall . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
A.3 Mass transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
A.3.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
A.3.2 Homogeneous reaction kinetics . . . . . . . . . . . . . . 250
A.3.3 Heterogeneous reaction kinetics . . . . . . . . . . . . . . 251
B Exhaust stroke flow similarity 253
C Velocity measurement techniques 259

C.1 Thermal anemometry . . . . . . . . . . . . . . . . . . . . . . . 259
C.2 Optical anemometry . . . . . . . . . . . . . . . . . . . . . . . . 260
D Modeling one-dimensional gas dynamics 267

D.1 Model description . . . . . . . . . . . . . . . . . . . . . . . . . . 267
D.1.1 Decoupling the Euler equations . . . . . . . . . . . . . . 268
D.1.2 Discretization schemes . . . . . . . . . . . . . . . . . . . 270
First order upwind schemes . . . . . . . . . . . . . . . . 270
Second order TVD schemes . . . . . . . . . . . . . . . . 271
D.1.3 Zero-dimensional volumes . . . . . . . . . . . . . . . . . 272
D.1.4 Implementation . . . . . . . . . . . . . . . . . . . . . . . 273
D.2 Model validation . . . . . . . . . . . . . . . . . . . . . . . . . . 273
D.2.1 Sod’s shock tube . . . . . . . . . . . . . . . . . . . . . . 273
Description . . . . . . . . . . . . . . . . . . . . . . . . . 273
Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 275
D.2.2 Frequency response of a pipe . . . . . . . . . . . . . . . 275
Description . . . . . . . . . . . . . . . . . . . . . . . . . 275
Analytical . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 277
D.2.3 Frequency response of a muffler . . . . . . . . . . . . . . 278
Description . . . . . . . . . . . . . . . . . . . . . . . . . 278
Contents xiii
Analytical . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 279
D.3 Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
D.3.1 Spatial resolution . . . . . . . . . . . . . . . . . . . . . . 280
D.3.2 Temporal resolution . . . . . . . . . . . . . . . . . . . . 280
D.4 Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
D.4.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
D.4.2 Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 282
D.4.3 Multisine signals . . . . . . . . . . . . . . . . . . . . . . 282
Nederlandse samenvatting 287

1 Inleiding: Uitlaatsystemen . . . . . . . . . . . . . . . . . . . . . 287
1.1 Achtergrond . . . . . . . . . . . . . . . . . . . . . . . . . 287
1.2 Ontwerpaspecten . . . . . . . . . . . . . . . . . . . . . . 288
1.3 Literatuuroverzicht . . . . . . . . . . . . . . . . . . . . . 289
1.4 Doelstellingen van de thesis . . . . . . . . . . . . . . . . 290
2 Experimentele aanpak . . . . . . . . . . . . . . . . . . . . . . . 291
2.1 Experimentele opstellingen . . . . . . . . . . . . . . . . 291
2.2 Stromingsgelijkvormigheid . . . . . . . . . . . . . . . . . 292
2.3 Datareductie . . . . . . . . . . . . . . . . . . . . . . . . 292
3 Oscillerende hittedraad anemometer (OHW) . . . . . . . . . . . 293
3.1 Inleiding . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
3.2 Methodologie . . . . . . . . . . . . . . . . . . . . . . . . 293
3.3 Kalibratie . . . . . . . . . . . . . . . . . . . . . . . . . . 294
3.4 Toepassing en validatie . . . . . . . . . . . . . . . . . . 295
4 Additieprincipe . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
4.1 Datareductie . . . . . . . . . . . . . . . . . . . . . . . . 295
4.2 Resultaten . . . . . . . . . . . . . . . . . . . . . . . . . 295
4.3 Fysische interpretatie . . . . . . . . . . . . . . . . . . . 297
5 Stromingsdynamica . . . . . . . . . . . . . . . . . . . . . . . . . 297
5.1 Experimentele resultaten . . . . . . . . . . . . . . . . . 297
5.2 Numerieke analyse . . . . . . . . . . . . . . . . . . . . . 298
6 Ontwerpoverwegingen . . . . . . . . . . . . . . . . . . . . . . . 300
7 Conclusie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
Bibliography 303
Curriculum vitae 313
List of publications 314
Index 316
Chapter 1
Introduction: Exhaust
systems
1.1 Legal background

The European Parliament and Council ratified Directive 98/69/EC [1] on 13
October 1998, relating to measures against air pollution by motor vehicle emis-
sions. Amending the original 1970 Directive 70/220/EEC, Directive 98/69/EC
enforces Euro III and Euro IV emission standards on newly registered vehicles,
as of 2000 and 2005 respectively. The Euro V standard will take effect as of
2008, but has yet to be ratified in a Directive. Table 1.1 shows the evolution of
the maximum emission levels for carbon monoxide (CO), hydrocarbons (Cx Hy ),
nitrous oxides (NOx ) and particulate matter (PM). The emissions are measured
during a standardized vehicle homologation procedure performed on a chassis
dynamometer.
Table 1.1 – Tailpipe emission limits for passenger cars below 2500 kg, according to
successive EU standards
Standard Limit values [g/km]
CO Cx Hy +NOx PM
Petrol Diesel Petrol Diesel Petrol Diesel
1992, Euro I 2.72 2.72 0.97 0.97 - - - 0.14
1996, Euro II 2.20 1.00 0.50 0.7/0.91 - - - 0.08/0.11
CO Cx Hy NOx PM
Petrol Diesel Petrol Diesel Petrol Diesel Petrol Diesel
2000, Euro III 2.30 0.64 0.20 - 0.15 0.50 - 0.05
2005, Euro IV 1.00 0.50 0.10 - 0.08 0.25 - 0.025
2008, Euro V 1.00 0.50 0.075 - 0.06 0.20 0.005 0.005
1
2 Chapter 1 Introduction: Exhaust systems
The increasingly stricter emissions legislation has been the main promotor
of technological advances in internal combustion engines over the last years,
and will continue to be so in the face of new regulations. Technological ad-
vances may be noted in different engine aspects. Firstly, limiting the gener-
ation of pollutant emissions is mostly related to (i) engine construction (e.g.
valve timing, combustion chamber layout) and (ii) engine management (e.g.
feedback-controlled fuel injection using lambda-sensors, exhaust gas recircula-
tion). Secondly, and most importantly for this thesis, (iii) the exhaust system
ensures the appropriate aftertreatment of pollutants.
In order to meet emission standards, any present exhaust system contains
at least one catalytic converter. Although this thesis deals with catalysts typi-
cally found in passenger cars and light-duty vehicles, the work is applicable to
motorcycles and heavy-duty vehicles as well. In stoichiometric petrol engines,
a three-way catalyst simultaneously oxidizes CO and Cx Hy to CO2 and water,
and reduces NOx to nitrogen. Diesel engines are fitted with an oxidation cat-
alyst, although NOx constitutes a major fraction of diesel pollutant emissions.
The catalytic material decreases the activation energy of the reactions and
causes adequate reaction rates at typical exhaust system temperatures (400 to
800 ◦ C). The catalyst material temperature must exceed the light-off tempera-
ture (250 to 400 ◦ C) to attain significant reaction rates.
From Euro III onwards, the regulations incorporate the emissions produced
during engine cold start, while the catalyst is still inactive. Rapid catalyst
warm-up is ensured by placing the catalyst in the hot end of the exhaust sys-
tem close to the engine, making it an integral part of the exhaust manifold.
This so-called close-coupled catalyst (CCC) reduces cold start emissions con-
siderably when compared to a traditional underfloor catalyst, mounted in the
downstream tailpipe section or cold end. Figure 1.1 defines the nomenclature
on a schematic diagram of a typical modern four-cylinder gasoline engine.
Tighter limits for NOx and PM in Euro IV and V (Table 1.1) are encour-
aging further developments such as regenerative particulate filters for diesel
engines, and NOx -traps for diesel and lean-burn petrol engines. Both particu-
late filters and NOx -traps contain catalytic material to promote soot oxidation
and NOx reduction respectively.
The present research into the flow in CCC manifolds focuses on three-way
or oxidation catalysts. However, its relevance extends towards NOx traps, PM
(soot) filters or any filtering device or heat exchanger subjected to a complex
pulsating and diverging flow.
Section 1.6 presents the goals and scope of this thesis, with respect to
increasing the knowledge on transient internal flows in manifolds with close-
coupled catalysts.
1 For indirect and direct injection diesel engines, respectively

1.2 Aspects of manifold design 3
manifold
(collector)
runners
(headers)
diffuser
driving
direction
close-coupled
catalyst
exit cone underbody

muffler tailpipe
catalyst
downpipe
hot end cold end
Figure 1.1 – Exhaust system nomenclature
1.2 Aspects of manifold design

This section discusses important aspects in the design of an exhaust manifold
with close-coupled catalyst. In terms of achieving an optimal operation and
a maximum lifetime of the catalyst, Sect. 1.2.1 demonstrates that the cata-
lyst velocity distribution is paramount. The major purpose of the manifold is
to ensure the appropriate distribution of exhaust gases towards the catalyst.
Section 1.2.2 discusses design aspects specific to the manifold’s task of flow
distribution.
Nomenclature In this thesis, the term manifold denotes the combi-

nation of runners and diffuser, excluding the close-coupled catalyst, as
depicted in Fig. 1.1. As such, the design aspects are discussed separately
in Sects. 1.2.1 (catalyst) and 1.2.2 (manifold).
Nevertheless, the catalyst is an integrated part of the overall hot end
structure, comprising the manifold, catalyst, exit cone and downpipe. As
such, in some cases a ‘pars pro toto’ interpretation is used that denotes
the hot end structure as a whole.
1.2.1 Catalyst
An automotive catalytic converter contains noble metal particles, typically
platinum (Pt), palladium (Pd) and rhodium (Rh), that act as catalyst for
the designated reactions. The active metals are distributed as fine particles
in the washcoat, containing mainly porous aluminum oxide (Al2 O3 ) to enlarge
the surface area. The washcoat is typically applied to a ceramic cordierite
(2MgO · 2Al2 O3 · 5SiO2 ) substrate [56], although folded metal foil substrates
Figure 1.2 – Catalyst substrates: (left) metal and (right) ceramic
are also in use. Figure 1.2 shows an example (unwashcoated) substrate of each
type.
The substrate consists of parallel channels with a typical hydraulic diameter
d of 1 mm and wall thickness t of 100 µm. Table 1.2 presents an overview of
typical ceramic substrate cell dimensions. The geometric surface area av is
2
defined as av = 4 d/ (d + t + tw ) [cm2 /cm3 ], where tw /2 is the washcoat layer
thickness, as shown in Fig. 1.3.
During this thesis, only ceramic substrates are used since these feature
tighter dimensional tolerances on the cell distribution. The velocity distribution
in a metal catalyst contains strong gradients due to local differences in the cell
distribution. These are of no particular interest to the design of the manifold
in terms of flow distribution.
The quantity av represents the amount of surface area available for reactions
per given catalyst volume. The dimensionless porosity ε is defined as the
2
relative open frontal area, or symbolically ε = d2 / (d + t + tw ) . The trend
is towards a smaller wall thickness and higher cell density.
Deactivation
Deactivation or degradation of a catalyst reduces the reaction rate. Although
deactivation is inevitable, a good manifold design minimizes its overall magni-
tude and avoids excessive local degradation.
Bartholomew [11] and Forzatti and Lietti [38] review catalyst deactiva-
tion in general due to chemical, thermal, physical and mechanical processes.
2 The substrate cell type is defined as wall thickness/cell density, where the wall thickness
is expressed in mil (1 mil = 1/1000 inch) and the cell density is expressed in cpsi or cells/inch2 .
y
z
L monolith
washcoat
x
½ tw
t
d+t+tw
Figure 1.3 – Ceramic substrate catalyst nomenclature
Table 1.2 – Ceramic catalyst substrate dimensions

Type2 Geometric parameters
d av t ε
mm cm−1 µm -
6.5/400 1.10 27.4 165 0.76
4.3/400 1.16 28.8 109 0.84
4.3/600 0.93 34.5 109 0.80
3.5/600 0.95 35.3 89 0.84
2.5/900 0.78 43.7 64 0.86
Neyestanaki et al. [80] review deactivation in automotive catalysts.

Chemical deactivation is caused by strong chemisorption of certain poisons
on catalytic surfaces, decreasing their reactivity. For automotive catalysts,
contaminants in the engine lubricating oil and fuel (e.g. P, Pb, Zn, S) consti-
tute catalyst poisons. Chemical deactivation is proportional to the amount of
processed exhaust gas. Areas in the catalyst cross-section with a high time-
averaged velocity are more subject to chemical deactivation. A uniform velocity
distribution ensures uniform chemical deactivation, and thereby maximum cat-
alyst lifetime.
Thermal deactivation results from several different processes [11, 38]. The
main process decreases the active surface area through sintering, which involves
crystallite growth in the catalyst phase and blockage of active elements due to
collapsing substrate pores. The sintering rate is usually fitted by a correlation
of the form:
n
d S S Seq
= −ks − (1.1)
dt S0 S0 S0
r
z
∆T
z
Figure 1.4 – Diffuser and catalyst: Basic flow pattern and streamwise temperature
evolution
where S, S0 and Seq are respectively the time-dependent, initial and equilibrium
surface area [m2 ], ks is the sintering rate constant [s−1 ] and n is the sintering
order [-] (typically n = 2). The rate constant ks increases exponentially with
the temperature, and becomes significant above 500 ◦ C [11]. In petrol engines,
the close-coupled catalyst may reach temperatures in excess of 1000 ◦ C. Conse-
quently, any change in temperature significantly affects the degree of thermal
deactivation.
The catalyst substrate temperature depends in a complicated way on the
velocity, temperature and reactant concentration distribution at the catalyst
inlet. The catalyst reactions are exothermic. The gas temperature increases by
an order of magnitude of 100 ◦ C as it passes through the catalyst. A qualitative
evolution of the mass-flow averaged temperature is shown in Fig. 1.4. The
temperature rise ∆T increases for (i) increasing inlet reactant concentration
(e.g. misfire causes a large quantity of hydrocarbons that oxidize in the catalyst)
and (ii) increasing catalyst residence time, which corresponds to decreasing
velocity. Heat conduction through the catalyst substrate in axial and radial
direction also affects the temperature distribution.
Furthermore, this thesis has shown that periodic flow reversal occurs. Fig-
ure 1.4 depicts a typical flow pattern in an axisymmetric diffuser and catalyst,
including backflow near the outer edges of the catalyst. In case of strong flow
reversal, hot processed gas may be recycled through the catalyst, leading to
a further increase in catalyst temperature. The influence of flow reversal on
catalytic conversion, the catalyst temperature distribution or thermal deacti-
vation is unknown. Yet considering the exponential temperature dependence
of thermal deactivation in Eq. (1.1), this effect should not be discarded.
Conversion efficiency
The conversion efficiency ηC [-] is defined as the relative decrease in the overall
pollutant species concentration at the catalyst outlet, compared to the inlet
10
8
ηC = 0.29
Velocity U/Um (-)

6 ηC = 0.56
4
ηC = 0.77
ηC = 0.92
2
ηC = 0.99
1
0
0 0.2 0.4 0.6 0.8 1
Radius r/R (-)
Figure 1.5 – Influence of the flow distribution on the conversion efficiency ηC , based
on a homogeneous catalyst model
concentration. Appendix A.3 discusses the different approaches for modeling

the mass transfer with catalytic reactions. The actual catalytic reaction kinetics
may be very crudely approximated by a set of homogeneous reactions for each
species.
Using this crude approximation, an analytical expression may be derived
for the conversion efficiency ηC (see Eq. (A.24) as a function of the velocity
distribution U (x, y):
I
Ki L
1 − exp − ρ U (x, y) dA
A U (x, y)
ηC,i =
Ki L
1 − exp − ṁ
Um
where the index i refers to the considered species, Ki is a reaction rate con-
stant [1/(m3 s)] and L is the catalyst length [m].
Merely as an example, Fig. 1.5 shows several velocity distributions U (r)
for an axisymmetric catalyst. r/R represents the dimensionless radius of the
substrate, where r/R = 0 is the catalyst centerline and r/R = 1 is the outer
edge, as depicted in Fig. 1.4. For each velocity distribution, the corresponding
value of the conversion efficiency ηC is obtained using Eq. (A.24), where the
reaction rate constant K is estimated arbitrarily.
In spite of the oversimplified model, Fig. 1.5 indicates the sensitivity of
the conversion efficiency to flow non-uniformity. The main external boundary
condition influencing the amount of converted gas is the residence time of the
gas in the catalyst τU ∝ L/U [s], where L is the catalyst length [m] and U is
the channel mean gas velocity [m/s]. Instead of τU , the amount of conversion
is determined by the ratio of τU to the limiting time scale for the chemical
kinetics (see App. A.3). A high ratio of residence time to kinetics time scale
ensures an adequate conversion level. Consequently, conversion is lower in high

velocity regions.
The kinetics time scale is considered to be the transverse diffusion time scale,
defined as τD,t = d2 /Dm [s], where Dm is the molecular diffusivity [m2 /s] of
the species.
In that case, the conversion efficiency ηC is related to τU /τD,t ∝
L d2 ε A/Q , where Q is the volumetric flow rate [m3 /s].
Pressure loss
The catalyst is one of the main contributors to the overall exhaust system
pressure drop or backpressure. Increased backpressure decreases the engine
efficiency in two ways: (i) the pumping losses increase with the exhaust system
backpressure, and (ii) the residual exhaust gas that remains in the cylinders
after the exhaust stroke increases with backpressure. The importance of the
second indirect effect is often neglected, yet an increase in the residual gas mass
decreases proportionally the amount of fresh mixture that can be inducted dur-
ing the following intake stroke, thereby influencing the amount of heat released
during combustion and the indicated efficiency. As such, backpressure directly
and indirectly influences engine efficiency and specific fuel consumption.
Flow in the substrate channels is incompressible (M a < 0.3) and laminar
(Re < 2300). Appendix A.2 describes the different aspects adding to the pres-
sure drop. Neglecting the effects of inlet and exit loss and the development of
the laminar boundary layer, the pressure drop ∆p is proportional to the chan-
nel velocity U/ε, where U is by convention the axial velocity immediately up-
3
2 a fixed volumetric flow rate Q = U A [m /s],
stream of the substrate. Assuming
2
∆p ∼ 4f (L/d ) ρ U /2 ∝ L d Q/(ε A) .
The three-dimensional velocity distribution in the diffuser is influenced in a
complicated way by the presence of the catalyst substrate. Seemingly unimpor-
tant effects such as the pressure drop due to oblique flow entry greatly affect
the flow distribution [15, 43]. Still, this effect is usually neglected in CFD
simulations of flow in CCC manifolds.
For a given catalyst geometry, the minimum pressure drop is obtained for a
uniform, axial inlet velocity profile. In the case of a non-uniform velocity distri-
bution, the high velocity regions create a local high pressure drop. The pressure
drop distribution throughout the remainder of the catalyst cross-section is bal-
anced by other forces, such as radial (outward) flow or swirling (tangential)
flow.
Design compromise
The cost of a catalyst is mainly due to the precious metals incorporated in
the washcoat. As such, the total washcoat volume Vcoat should be minimized.
Assuming an invariable washcoat thickness, Vcoat is proportional to the total
catalytic surface area, or symbolically Vcoat ∝ av A L ∝ ε A L/d.
The flow uniformity ηU influences both the pressure drop and conversion
efficiency. Section 4.3.1 presents several ways of quantifying the flow uniformity.
For now, it is sufficient to keep in mind that ηU varies between zero and unity,
Table 1.3 – Catalyst design criteria

Criterion Parameters
A L d ε
cost%
(i) % % 1 %
∆p %
(ii) 1 % 11 1
ηC 1
(iii) 1 1 %% 1
ηU 1
(iv) 1 %% 1
Optimum 1 ∼ % 11
where ηU = 1 corresponds to a perfectly uniform velocity distribution and when

ηU ' 0, the entire flow passes through a single channel.
The relationship between the flow uniformity and geometric parameters is
difficult to estimate. However, a high flow uniformity is increasingly difficult
to achieve for a larger catalyst cross-sectional area A. On the other hand,
the flow becomes increasingly more uniform as the pressure drop coefficient
K [-] increases. K is defined
as the ratio of pressure drop to dynamic pressure,
or K ∼ 4f (L/d ) ∝ L d2 ε A/Q . Combining both assumptions, the flow
uniformity varies as ηU ∼ L d2 ε/Q .
The influence of the catalyst geometric parameters A, L, d and ε can be
summarized as follows:
(i) Precious metal cost ∝ Vcoat ∝ ε A L/d

(ii) Pressure drop ∆p ∼ L/d2 Q/ (ε A)

(iii) Conversion efficiency ηC ∼ τU ∝ L/d2 ε A/Q

(iv) Flow uniformity ηU ∼ L/d2 ε/Q
Table 1.3 indicates the relationship between the desired change in the four
above stated criteria and the determining geometrical parameters. For instance,
∆p decreases (%) mainly by increasing d (11) and to a lesser extent by increasing
A (1) and ε (1), and decreasing L (%).
Although Table 1.3 is based on crude assumptions, it presents a remarkably
accurate summary of the design compromises involved in automotive catalyst
design.
Based on Table 1.3, the porosity ε should be maximized, which corresponds
to minimizing the substrate wall thickness. This is beneficial for all criteria,
except the precious metal costs. The channel diameter d should be minimal
to maximize conversion and flow uniformity, although this causes a penalty in
pressure drop.
Table 1.2 shows some typical substrate cell structures that are used in au-
tomotive catalysts. The listing is roughly chronological from top to bottom.
The trend is indeed towards thin-walled substrates (t 6 2 mil) with a higher
cell density (> 900 cpsi), which are additionally characterized by a lower ther-
mal mass, thus providing faster warm-up. This is demonstrated by Wiehl and
Vogt [112], describing the recent advances and evolution in ceramic catalyst
substrates.
Table 1.3 indicates that the catalyst cross-sectional area A should be max-
imized to the minimize pressure drop and maximize the conversion efficiency.
However, the diffuser’s performance for obtaining a good flow uniformity should
be taken into account here. Inadequate manifold design leads to a non-uniform
velocity distribution, which is detrimental for the pressure drop and conversion
efficiency. Furthermore, it causes local deactivation which reduces the catalyst
lifetime.
1.2.2 Manifold
As previously indicated, the manifold’s main task is to distribute the exhaust
gases towards the close-coupled catalyst. The short distance available between
engine and catalyst entails particular issues for the manifold’s pressure loss,
flow dynamics and thermal load.
Pressure loss
The pressure drop over the manifold itself (i.e. excluding the catalyst) is of
the same order of magnitude as the catalyst pressure drop. The short distance
between the engine and CCC requires exhaust runners with a small length-
to-diameter ratio, featuring multiple out-of-plane bends with small curvature
radius-to-diameter ratio (see Figs. 1.7, 1.9, 2.1). The runners converge into
a diffuser that aims to distribute the flow across the catalyst cross-section.
Flow in the exhaust manifold is highly pulsating and three-dimensional, fea-
turing strong secondary flows and vortices created by the short curved runners.
Fiedler [36] and Miller [74] provide an explanation for the creation of secondary
swirling flows in interacting bend combinations that are typical for exhaust sys-
tems with close-coupled catalyst.
Furthermore, runners that enter the diffuser at an oblique angle cause strong
mixing in the diffuser. With reference to Fig. 1.6, an oblique entry angle
corresponds e.g. to values of α or β close to 90 ◦ . This is beneficial for a good
flow uniformity, at the cost of an increase in pressure drop. Figure 1.7 shows
two examples of exhaust manifolds with different entry angles. The variant in
Fig. 1.7 (left) exhibits a lower pressure drop yet also a worse flow uniformity
compared to the variant in Fig. 1.7 (right), where the runners enter the diffuser
at a more oblique angle.
Wendland and Matthes [109] and Wendland et al. [110] provide a correlation
between the non-dimensional catalyst pressure drop (denoted ‘smoothing index’
n in [109, 110]) versus a measure of the flow non-uniformity M . This correlation
is given in Fig. 1.8, and is based on experimental data for different types of
underfloor catalysts, obtained on a stationary water flow bench and in fired
engine conditions. M is defined as Umax /Um − 1, where Umax and Um are the
maximum and mean catalyst velocity, respectively. The smoothing index n is
defined as the ratio of the catalyst pressure drop to the dynamic pressure in the
Figure 1.6 – Connection of a runner into the diffuser
Figure 1.7 – Influence of runner layout on the pressure drop (Source: [2])
2
2
inlet pipe upstream of the diffuser, or symbolically n = ∆p ρUm /2 (Ai /A ) ,
where Ai and A are the cross-sectional area of the inlet pipe and the catalyst,
respectively. Figure 1.8 shows a good logarithmic agreement between M and
n, for different geometrical variants and flow conditions.
The numerical example below provides an order-of-magnitude estimation
for the contribution to the total exhaust system pressure drop by individual
components. The relationship between the catalyst pressure drop and the flow
uniformity is quite complex, and is not included in the numerical example.
Numerical example The pressure drop of the exhaust system for a
typical modern mid-range passenger car engine originates mainly from
(i) the runners and diffuser, (ii) the catalyst, (iii) the muffler and (iv) the
exhaust pipe itself (see Fig. 1.1). The calculation below is for a 2000 cm3
4-cylinder gasoline engine. The diameters of the runners, catalyst and
exhaust pipe are estimated at 35, 122 and 50 mm, respectively.
• Runners and diffuser — The runner pressure drop is estimated

Figure 1.8 – Maldistribution versus smoothing index (Source: [109, 110])
based on two closely spaced 90 ◦ bends, forming a 90 ◦ out-of-plane

angle. This is the typical form of an exhaust runner in a compact
close-coupled catalyst exhaust system:
ρUr2
∆p = Krunner
2
ρUr2
∆p = nKb Cf CRe Cb−b (1.2)
2
where Krunner represents the runner pressure drop coefficient [-], n
is the number of bends (= 2) and ρUr2 /2 represents the dynamic
pressure [Pa] based on the mean runner velocity. Kb is the pressure
drop coefficient of a single 60 ◦ bend with infinite inlet and outlet
pipes. According to Miller [74] (Fig. 9.2), Kb ' 0.11 for a radius-
of-curvature = 2D. Cf is a correction for a rough instead of a
smooth pipe. Cf = frough /fsmooth , where f is the friction factor3 ,
obtained from e.g. the implicit Colebrook-White [30] correlation for
fully developed turbulent flow in a rough pipe:
„ «
1 k/D 1.255
√ = −4.06 log10 + √ (1.3)
f 3.71 Re f
where k is the wall roughness (e.g. 0.025 mm for a smooth steel
pipe). CRe is a correction for low Reynolds number, given by
Fig. 9.3 [74] (CRe ' 1.65 for Re = 50 000). Cb−b is the bend
interaction coefficient, given by Fig. 10.3 [74] for a 90 ◦ out-of-plane
combination (Cb−b ' 0.725). In total, the runner pressure drop
coefficient Krunner ' 0.79.
The diffuser pressure drop is estimated based on a sudden expansion
with a diameter ratio of 0.5, which corresponds to Kdiffuser ' 0.5.
3 The Fanning friction factor is used throughout this thesis, defined according to ∆p =
4f (L/D ) ρU 2 /2 . The Fanning friction factor is 1/4 of the Darcy friction factor.
• Catalyst — The pressure drop in the catalyst is estimated assuming

fully developed laminar flow. As shown in App. A.2, the actual
pressure drop is higher, due to contraction and expansion loss, and
the development of the boundary layer. For square channels, the
friction factor is found analytically to be f ' 14.227/Re . The
pressure drop coefficient Kcatalyst = 4f (l/d ), where l and d are
the catalyst length and hydraulic diameter of the catalyst channels.
Due to the laminar flow regime, Kcatalyst is inversely proportional
to the flow rate.
• Muffler — Due to the multitude of muffler designs, the pressure
drop in the muffler is roughly estimated based on the assumption
that the muffler consists of two consecutive expansions with infinite
diameter ratio. In this case, the dynamic pressure is completely lost
twice, which corresponds to a pressure drop coefficient Kmuffler = 2.
• Exhaust pipe — The exhaust pipe is assumed to be 5 m long, 50 mm
in diameter, and made of stainless steel with wall roughness k =
0.025 mm. The friction factor is determined from Eq. (1.3).
The table below shows the contribution to the total pressure drop of each
component, for different engine speeds:
Engine speed N 2000 4000 6000 rpm

∆prunner+diffuser 23 92 207 mbar
∆pcatalyst 32 64 96 mbar
∆pmuffler 12 46 104 mbar
∆pexhaust pipe 15 51 107 mbar
∆ptotal 81 253 514 mbar
The contribution to the total pressure drop of each component is of com-

parable magnitude. An acceptable value for the maximum total pressure
drop is 500 to 800 mbar. Traditionally, the cold end (muffler, exhaust
pipe) contributes 50 to 60 %, and the hot end contributes 40 to 50 %.
Recently, the compactness of the hot end has lead to a relative increase
in hot end pressure drop.
Flow dynamics
An excellent overview of acoustical properties of exhaust systems as a whole is
presented in the doctoral work of Boonen [21]. Further background information
on the design of exhaust silencers and resonators is given in e.g. Davis et al.
[33].
However, this thesis is mainly concerned with the lower frequency flow dy-
namics of the exhaust manifold. As shown in Chap. 4, these dynamics influence
the time-averaged flow distribution and the flow uniformity. Consequently, they
have an effect on the system backpressure, conversion efficiency and catalyst
ageing.
As shown in Chap. 5, the low frequency dynamics in close-coupled cata-
lyst manifolds are governed by a Helmholtz-type resonance phenomenon. In
acoustics and exhaust silencer literature, Helmholtz resonators are typically en-
countered as silencer elements, mounted perpendicular to the main flow duct.
The term Helmholtz resonance used in this thesis differs from that meaning, as
discussed in Sect. 5.3.
The insert below provides some background information on pressure wave
tuning, and its significance with respect to modern compact exhaust systems
and this thesis.
Background: Pressure wave tuning The initial phase of the ex-

haust stroke is known as blowdown, during which the combustion prod-
ucts in the cylinder expand from the residual cylinder pressure to the
exhaust system pressure. The hot pulse of fast moving exhaust gas cre-
ates a positive (compression)√pressure wave traveling down the runner
at the speed of sound c = γrT ' 640 m/s (for T = 800 ◦ C). The
blowdown ends roughly 1 to 5 milliseconds later, when the cylinder pres-
sure and exhaust pressure have equalized. At this point, the momentum
stored in the fast moving blowdown pulse starts a negative (expansion
or rarefaction) pressure wave down the runner.
It is beneficial for the cylinder scavenging if this negative pressure would
persist near the exhaust port during the remainder of the exhaust stroke,
also called the displacement phase. However, the negative pressure grad-
ually disappears as the piston moves upwards and expels the remaining
combustion products.
At some point, the initial compression wave reaches a cross-section en-
largement. The enlargement can be a junction with another runner, a
diffuser (e.g. followed by a catalyst), or the open atmosphere (e.g. in case
of a high powered racing engine). As the compression wave encounters
the enlargement, it starts a reflected expansion wave backwards up the
runner. If the expansion wave reaches the exhaust port before the ex-
haust valve closes, the low pressure helps the cylinder scavenging, similar
to the first expansion wave created after the blowdown.
In a typical four-stroke engine, the intake valve opens before top dead
center (' −10 ◦ ca), whereas the exhaust valve closes after top dead center
(' +10 ◦ ca). If the rarefaction wave reaches the exhaust port during the
valve overlap period, the negative exhaust pressure helps the induction
of fresh mixture from the intake runner.
The exhaust waves only feature this beneficial effect if the length of the
exhaust runners L is such that the time for the wave to travel twice
the runner length 2L/c (i.e. one compression wave traveling down and
one expansion wave traveling up) corresponds to the time between the
beginning of the exhaust stroke and the valve overlap. Therefore, one
specific exhaust system creates a positive scavenging effect for only a
narrow engine speed range.
Increasing the engine performance at speeds above 5000 rpm requires
runners of at least 500 mm long. Preferably, each cylinder should have
an individual runner of equal length without junctions, and all runners
should converge in a collector with a large cross-section.
Such configuration is denoted an ‘N-in-1’ exhaust, where N is the number

of converging runners. Due to the piping size and complexity, this type
of exhaust is rarely used for any purpose other than racing.
For a four-cylinder engine, a 4-in-2-in-1 is a simpler exhaust configura-
tion, consisting of four runners that converge pairwise in two Y-junctions,
followed by a third Y-junction that connects to the cold end. This con-
figuration offers a reduced peak performance increase when compared
to a 4-in-1 exhaust, yet the optimum engine speed range is broader. A
compression wave that reaches the Y-junction reflects back through both
connected runners, resulting in a weaker expansion wave. The typical en-
gine firing order is 1–3–4–2. Two pairs are formed by ‘opposite’ cylinders
that are 180 ◦ ca out of phase, i.e. cylinders 1 and 4 are paired, and 2 and
3 are paired. Pairing opposite cylinders means that the compression wave
generated by the second cylinder’s blowdown does not interfere with the
first cylinder’s displacement phase.
To distinguish between a standard exhaust and a high-performance wave
tuned exhaust, a different terminology is often used in the literature: For
a standard exhaust, the term runner is used in conjunction with manifold,
describing short curved pipes issuing in an arbitrarily shaped manifold
volume with no regard for pressure wave tuning. The manifold’s shape is
determined only by the easiest way of connecting the exhaust ports to the
downpipe. Such a standard manifold is traditionally made of low-cost,
heavy cast iron. By contrast, the term header denotes a long runner made
of smooth steel pipe and gradual bends, which issues into a collector,
rather than a manifold. A collector forms a cross-section enlargement
that produces good rarefaction waves. Flow from the incoming headers
exits the collector straight-through, minimizing losses due to flow turning.
In this thesis, no distinction is made between runners or headers, and
manifold or collector. The manifold denotes the combination of runners
and diffuser, as depicted in Fig. 1.1.
Long runners are required to benefit from the pressure wave tuning effect.
This is generally incompatible with rapid catalyst warm-up. Indeed,
tighter emissions regulations lead to compact manifolds with integrated
close-coupled catalyst. The time scales associated with the pressure wave
travel in the short curved runners are too small to be of any benefit for
cylinder scavenging.
However, engines for high performance passenger cars attempt to com-
bine a N-in-2-in-1 or N-in-1 exhaust with a close-coupled catalyst. Diez
et al. [34] describes the evolution in the exhaust system of two succes-
sive generations of top-end petrol engines. The new generation consists
of a V10-engine with a 5-in-1 manifold (shown in Fig. 1.9) with inte-
grated close-coupled catalyst on both cylinder banks. The old genera-
tion consists of a V8-engine with 4-in-2-in-1 manifold and two underfloor
catalysts.
Missy et al. [75] discuss a numerical study of the chemical kinetics and
light-off behavior in a 4-in-2-in-1 exhaust system. Two close-coupled
catalysts are used following each of the first Y-junctions pairing cylinders
1, 4 and 2, 3.
Figure 1.9 – Example of equal length runners (diffuser not shown) for a V10-engine
(Source: [34])
The above cases [34, 75] are rather exceptional. The typical exhaust
system for a modern engine is more concerned with rapid catalyst warm-
up than with gaining performance in the high engine speed range. State-
of-the-art exhaust systems such as developed by Bosal are made of steel
piping instead of cast iron. This results in a weight reduction and a lower
thermal mass, which accelerates warm-up.
Heywood [47] provides general background information on the gas exchange

process during intake and exhaust stroke. Stone [96] provides an introduction
to gas dynamics related to manifold design. Benson et al. [50, 19] discuss wave
action simulation models and their application to pressure wave tuning, as well
as wave-tuned intake and exhaust manifolds.
Thermal load
The flow inside the runners and diffuser causes a certain distribution of wall
shear stress and heat transfer coefficient. Areas of excessive temperature or hot
spots form on the walls where the (internal) convective heat transfer coefficient
is high. A hot spot which coincides with a vulnerable structural part (e.g.
welds, a protective catalyst mat or lining material) causes differential thermal
expansion and thermal stresses. These stresses add to the existing cyclic ther-
mal stresses induced by successive heating and cooling cycles. The stresses
weaken the component and may result in premature failure.
This situation is aggravated by the ever increasing mean temperature of
the hot end. Indeed, most new exhaust system components (e.g. close-coupled
catalyst, particulate filter, . . . ) require a minimal enthalpy loss in the man-
ifold. Fast warm-up requirements lead to a light-weight manifold, assembled
from thin sheet metal. Furthermore, radiation shields and thermal insulation
protect other components in the engine compartment against external heat
transfer from the exhaust system. This increases the temperatures of the ex-
haust system materials, and raises the importance of accurate prediction of
internal flow and heat transfer in the exhaust manifold.
1.3 Motivation for this thesis 17
1.3 Motivation for this thesis

Exhaust manifold design requires state-of-the-art knowledge of fluid dynam-
ics and heat transfer for this complex flow. Numerical flow simulation using
commercially available computational fluid dynamics (CFD) software does not
necessarily yield reliable results, since the flow is characterized by non-isotropic
turbulence and three-dimensional boundary layers inside the runners and dif-
fuser. These conditions are not ideally suited for RANS simulations (i.e. solving
the Reynolds-averaged Navier-Stokes equations) with conventional turbulence
models.
Interestingly, industrial design procedures rely almost entirely on numerical
studies using CFD. The majority of the calculations involve stationary flow,
due to the high computational cost associated with transient CFD calculations
for such a complex three-dimensional turbulent flow. These stationary flow
predictions are used to estimate the backpressure, the catalyst flow uniformity,
and other flow-related criteria. Furthermore, the stationary flow results are
used as boundary condition in finite element modeling to predict the surface
temperatures and the thermal stresses.
The catalyst velocity distributions in Fig. 1.10 demonstrate the level of
uncertainty associated with CFD predictions of flow in close-coupled catalyst
manifolds. Fig. 1.10a, b and c are CFD results of stationary flow through
runner 1 of manifold A, used in this thesis (see Sect. 2.1). Cases 1.10a, b and c
differ only in terms of the inlet boundary conditions to the runner. Case 1.10a
and b feature a uniform axial velocity distribution, whereas case 1.10c features a
peaked axial velocity distribution into the runner. Swirl is applied to the runner
inlet in cases 1.10b and c, whereas case 1.10a does not feature swirl. One may
appreciate a considerable effect on the catalyst velocity distribution. Case 1.10c
may be seen to correspond best to the experimental result in Fig. 1.10d. Yet
such experimental data are not available in the industrial design environment,
making it impossible to assess the accuracy of the numerical flow predictions
upon which the design is based.
Taking into account the given difficulties regarding accurate numerical sim-
ulations, the principal motivation of this thesis is to develop high-quality ex-
perimental methods for determining the transient velocity distribution in the
catalyst. These data are useful for validation and optimization of the numerical
approach.
Given the strong pulsating nature of the exhaust flow, the question arises
whether using stationary predictions is at all justifiable. This specific issue
forms an important motivation for the experimental approach to this research.
This thesis [88, 86] introduces the addition principle for flow in close-coupled
catalyst manifolds. The addition principle (see Chap. 4) states the following:
Addition principle The time-averaged catalyst velocity distri-

bution in pulsating flow can be predicted by a linear combination
of velocity distributions that results from stationary flow through
each of the exhaust runners, for equal volumetric flow rate.
(a) (b)
Stationary velocity U (-)
3
Runner 1, Qref = 109.3 m3/h
1
30 .75
2.5
0.50 2 2.5
1.5
0.75
0.5
1.5
15 2
0.5
y (mm)
2
0 1.5
1
0.25
1
0.5
3 5
2. 1
-15
2
1.5
0.7
0.2
5
5
5 0.5
0.7
1 0.5
-30
Um = 8.018 m/s, ηm = 0.342, ηw = 0.689
0
-30 -15 0 15 30
(c)
x (mm)
(d)
Figure 1.10 – Example of discrepancies between (a, b, c) numerical and (d) ex-
perimental results for the stationary flow distribution in a close-coupled catalyst
(Source: [2])
If the principle is proved valid, it implies that transient CFD is not required
for designing a manifold with close-coupled catalyst with respect to the catalyst
flow distribution and that steady state CFD simulations suffice.
For the industrial design of these systems within an automotive Tier 1
supplier, the validity of the addition principle carries huge implications, since
the calculation time for transient CFD is at least an order of magnitude higher
compared to stationary CFD. This significant gain in calculation time is directly
reflected in terms of a reduced development time, which is crucial for obtaining
new contracts from original equipment manufacturers (OEMs).
Section 1.4 reviews the available literature, specifically focusing on the in-
fluence of stationary and pulsating flow on the time-averaged catalyst flow uni-
formity. Based on the above motivations and the available literature, Sect. 1.6
defines the goals and scope of this thesis.
1.4 Overview of manifold flow research 19
1.4 Overview of manifold flow research

There are several sources available in literature discussing experimental and
numerical studies of flow in close-coupled catalysts.
Earlier studies (1970 till 2000) (e.g. Howitt and Sekella [51], Lemme and
Givens [64], Kim et al. [60]) examine the pressure drop and velocity distribu-
tion in exhaust systems with catalysts under stationary flow conditions. The
assumption of stationary flow is justifiable for flow through an underbody (un-
derfloor) catalyst, typically located 1 to 2 m downstream from the engine.
Wendland et al. [110] discuss an experimental study of underbody catalysts
with different diffuser and exit cone geometries on a stationary flow rig. The
authors compare theoretical and empirical expressions for the catalyst pressure
drop to their measurements. The authors refer to Wendland and Matthes [109]
for a correlation between a flow non-uniformity measure (based on the ratio of
maximum to mean velocity) and the pressure drop coefficient (defined as the
ratio of the pressure drop to the dynamic pressure upstream of the diffuser).
This correlation, shown in Fig. 1.8, is obtained from experiments on several
different underbody catalysts, with one or two monoliths, with a circular or
oval cross-section, with a centered inlet pipe yet offset or centered outlet pipe.
With the introduction of close-coupled catalysts, the attention has moved
to studying the these systems in pulsating flow conditions (1995 till present).
As such, the literature review in this section focuses on pulsating flow studies.
In this field of research, velocity measurements are generally performed
using hot-wire anemometry (HWA), laser Doppler anemometry (LDA). Other
velocity measurement techniques prove less successful. For example, the pitot-
static probe suffers from an insufficient sensitivity in the low velocity range and
a limited bandwidth. Lambert et al. [61] even propose using a fast response
trace gas detection system to determine the velocity magnitude and direction
in catalyst systems.
The following sections review the relevant literature. A distinction is made
between results concerning the time-averaged flow distribution (Sect. 1.4.1) and
the time-varying flow dynamics (Sect. 1.4.2). Finally, Sect. 1.4.3 reviews the
original contributions of this thesis to the existing literature.
1.4.1 Flow distribution

Several authors (Benjamin et al. [13], Voeltz et al. [104], Breuer et al. [23],
Nagel and Diringer [79]) discuss velocity measurements and CFD calculations
on close-coupled catalyst manifolds in stationary flow conditions. Most of these
studies are aimed at maximizing the catalyst flow uniformity, while minimizing
the pressure drop, often through parametric studies of diffuser and runner
geometry or mechanical flow dispersers.
However, the most interesting studies (see below) are performed in pulsating
flow, either on (i) an isothermal pulsating flow rig, (ii) an isochoric or engine
flow rig. The isochoric or engine flow rig may be motored (i.e. driven at constant
speed, without internal combustion) or fired (i.e. braked at constant speed, with
internal combustion).
Isothermal flow rigs

The easiest way of reproducing the pulsating flow in an exhaust system in a
laboratory setup is by means of incorporating a pulsator device (e.g. rotating
valve, rotating disk or other fast-response flow control device) into a stationary
flow rig. The flow generated in this way differs significantly from the exhaust
flow in a fired engine (see Sect. 2.3). However this type of setup is frequently
used because of its simplicity, and its independently adjustable pulsation fre-
quency and flow rate. Furthermore, stationary flow rigs are commonly used by
exhaust system manufacturers to measure the steady state pressure drop. The
isothermal flow rig used during this thesis is described in Sect. 2.2.1.
Benjamin et al. [17] discuss experimental results on an axisymmetric man-
ifold with catalyst, mounted on an isothermal flow bench with rotating disk
pulsator. The authors perform a parametric study of the influence of the dif-
fuser shape on the catalyst flow distribution. The paper provides time-resolved
velocity data at the inlet and exit of the catalyst brick, for a broad range of
pulsation frequency and flow rate. Although no comparison is made between
pulsating and stationary flow in terms of the addition principle, the authors
define a non-dimensional number as the ratio of pulsation period to diffuser
residence time, based on the runner velocity. A similar number is used to char-
acterize the flow conditions in this thesis. A good correlation exists between a
non-uniformity measure and the non-dimensional number.
Liu et al. [69] use the experimental set up of Benjamin et al. [17], with more
overlapping inlet velocity pulse shapes. Liu et al. [69] do not refer to any non-
dimensional correlation, however the authors report a lower uniformity due to
overlapping inlet flow pulses when compared to the results of Benjamin et al.
[17] featuring non-overlapping pulses. As the pulsation frequency increases,
non-uniformity decreases, i.e. the flow uniformity increases (see Fig. 1.11). This
is in agreement with the findings of this thesis. From Fig. 1.11 also follows that
overlapping inlet flow pulses ( ) result in a lower uniformity when compared
to non-overlapping pulses ( ). In this thesis, a slightly higher uniformity is
observed in the absence of overlap between exhaust valve openings.
Benjamin et al. [16] performed a number of steady and pulsating flow exper-
iments on several types of novel contoured catalyst brick designs, all referenced
to a standard type catalyst brick similar to the one used in this research. The
results for the standard type brick shown in Fig. 1.12 indicate that the flow
uniformity (Note: the plot shows non-uniformity) increases from steady flow
over low pulsation frequencies to high pulsation frequencies, and that flow uni-
formity decreases for increasing flow rate, both in steady and pulsating flow
conditions. This is confirmed by the results of this thesis.
For the CFD predictions, Benjamin et al. [16, 15] model the catalyst flow
resistance using a pressure loss correction for oblique flow entrance into the
monolith. This effect is not included in the porous volume model included in
4 See explanation of valve overlap on page 35.
Figure 1.11 – Comparison of non- Figure 1.12 – Non-uniformity index

uniformity index, with ( ) and with- as a function of flow rate and pulsation
out ( ) inter-cylinder4 exhaust valve frequency (Source: [16])
overlap (Source: [69])
Figure 22 Experimental velocity profiles for the Standard substrate

Figure 1.13 – Experimental versus
compared to CFD simulated
predictions catalyst
with and without entrance effect. velocity distribution, with-
out ( ) and with ( ) oblique flow entrance correction (Source: [16, 15])
commercially available CFD packages. The porous volume model is used to

represent a distributed catalyst monolith flow resistance, without having to
model the small individual monolith channels. By incorporating a correction
for oblique flow entrance in the numerical model, Benjamin et al. [16, 15] obtain
the results shown in Fig. 1.13, revealing a much better correspondence between
the simulated and experimental velocity distribution. The oblique entrance
loss is further discussed in App. A.2. Haimad [43] provides more details on
the practical implementation of the oblique entrance correction in the CFD
software.
Using the same oblique entrance correction approach [16, 15], Benjamin
et al. [18] provides a comparison between experimental and numerical velocity
distributions, obtained on an isothermal flow rig, using a production model CC
exhaust manifold. Figure 1.14 shows that incorporating the oblique entrance
correction significantly improves the correspondence between numerical and
experimental results, although the correspondence is still imperfect.
Kim and Cho [58] used an isothermal flow rig with rotating disk to generate
b (scale as a)
c (scale as a)
Fig. 5 Velocity contours with pulsating flow. 25Hz:(a) port 1 HWA measurements, v,, Re 72000: (b) port 1
CFD case7, without entrance effects, v,, Re 69800; ( c )port I CFD case 8, with entranceeffects,~,,Re
69 800: (d) view angle
Figure 1.14 – (a) Experimental versus (b, c) simulated catalyst velocity distribution,
(b) without and (c) with oblique flow entrance correction (Source: [18])
pulsating flow in the CC exhaust manifold also used by Park et al. [81] and Kim
et al. [59]. The authors conclude that the time-averaged velocity distribution
is more uniform in pulsating conditions compared to stationary flow.
Bressler et al. [22] performed time-averaged velocity measurements of pul-
sating flow in a four-runner manifold with a close-coupled catalyst. Measure-
ments have been performed on an isothermal flow bench with rotating disk to
generate the pulsating flow. The authors used trace gas injection to determine
the load on the catalyst from each cylinder. Bressler et al. [22] provide qual-
itative confirmation of the addition principle’s validity. Results are presented
according to the non-dimensional ratio of exhausted gas volume per cylinder
and per cycle to the diffuser volume. This ratio is actually identical to the scav-
enging number S used in this thesis and a similar number used by Benjamin
et al. [17]. Although no correlation is presented, their results indicate that the
flow uniformity is unaffected by engine speed as long as the non-dimensional
ratio remains constant.
Isochoric flow rigs

A more realistic flow is generated in the exhaust system by using the internal
combustion engine for which the exhaust is designed. To contrast an isothermal
flow rig, the term isochoric (i.e. isovolumetric or constant volume) is chosen,
since the gas is exhausted by a reciprocating volumetric machine. Of course,
the volume changes during the working cycle, yet the term isochoric refers to
the displacement of an equal volume per working cycle from the intake to the
exhaust side.
The engine can be driven (or motored ) without combustion by an electric
motor, resulting in cold flow in the exhaust. The most realistic flow is of course
generated by running the engine with combustion (or fired ), yet this situation
severely complicates the velocity measurements. The motored engine approach
allows measurement of the catalyst velocity distribution with a good spatial
and temporal resolution.
Background The exhaust stroke of an internal combustion engine, ei-

ther motored or fired, consists of two phases: the blowdown and displace-
ment phase.
The blowdown phase starts when the exhaust valve(s) open. Typically,
the exhaust valve opens around 40 ◦ ca before bottom dead center5 . The
blowdown phase ends when the residual cylinder pressure has expanded
to equal the exhaust manifold pressure. The blowdown phase is rela-
tively short (see the discussion on flow similarity in Sect. 2.3.2), and is
characterized by a high peak flow rate and high transients. The end of
the blowdown corresponds roughly to bottom dead center, depending on
engine speed and residual cylinder pressure.
During the subsequent displacement phase, the piston moves upward and
expels the remaining gas from the cylinder. This causes a low peak flow
rate and low transients.
The two-stage exhaust stroke in an isochoric flow rig is quite different from
the single-stage exhaust pulse generated by an isothermal flow rig. Section. 2.3
discusses the exhaust stroke flow similarity of an isothermal and a motored
engine flow rig, compared to true fired engine conditions.
Arias-Garcia et al. [7] performed time-averaged measurements using HWA
of both steady and pulsating flow in a CCC manifold. Both an isothermal ax-
isymmetric flow bench with rotating disk pulse generator (Benjamin et al. [17])
and a motored engine with atmospheric inlet are used for generating the pulsat-
ing flow. Results from the isothermal flow bench do not correlate well with the
motored engine results, probably due to interaction effects caused by exhaust
valve overlap, which is absent in the flow bench set up. The time-averaged
velocity distribution in pulsating flow is more uniform than for steady flow.
Computational fluid dynamic (CFD) calculations underestimate the velocity
magnitude by 50 %, thus overestimating the flow uniformity.
1.4.2 Flow dynamics

This section reviews the existing literature in terms of time-varying flow effects,
in particular the occurrence of (i) gas dynamic resonances and (ii) periodic flow
reversal through the catalyst.
The relatively low frequency (below 500 Hz) gas dynamic resonances are
predominantly encountered in manifolds mounted on an isochoric flow rig. The
fluctuations aggravate flow separation and recirculation in the diffuser, since
periodic flow reversal through the catalyst is reported only on isochoric flow
rigs. Both Hwang et al. [53] and Kim and Cho [58] present phase-locked LDA
5 Crankshaft angle position is expressed in ‘◦ ca’.
Figure 1.16 – Experimental (left)

Figure 1.15 – Experimental ( ) ver- versus simulated (right) velocity distri-
sus simulated ( ) time-resolved mean bution in the diffuser, during exhaust
catalyst velocity (Source: [59]) stroke of cylinder 1 (Source: [81])
results obtained on an isothermal flow rig. Although LDA is capable of mea-

suring bidirectional velocity, no flow reversal is observed.
Isochoric flow rigs: Fired

Park et al. [81] and Kim et al. [59] used phase-locked LDA to measure the
time-resolved local velocity in a CCC manifold on a fired engine. The authors
measured in several points along a straight line, upstream of the catalyst brick.
The study of Park et al. [81] revealed the existence of a distinct high-frequency
velocity fluctuation during the displacement phase. This phenomenon is caused
by Helmholtz resonances, which is not explained by the authors and which
has been observed and analyzed in this thesis (see Sect. 5.3). The authors
conclude that measurement data and results of transient CFD predictions are
in good agreement, although this comparison appears questionable. Figure 1.15
from [59] shows a reasonable comparison between the computed and measured
mean catalyst velocity. However, Fig. 1.16 from [81] shows a poor comparison
between computational and measurement data in terms of the local velocity in
the diffuser.
Adam et al. [4] use a one-dimensional gas dynamic model to provide bound-
ary conditions for a transient three-dimensional CFD simulation of the flow in
a CCC manifold. The simulation results give clear evidence of Helmholtz res-
onances in fired engine conditions.
Liu et al. [70] combine a one- and three-dimensional model in the same way
as Adam et al. [4]. The authors present simulations for a fired and motored
engine with atmospheric intake conditions in Fig. 1.17. For fired engine con-
ditions, the simulated exhaust runner velocity ( ) indicates the presence of
Helmholtz resonances. The motored engine simulations ( ) do not exhibit
similar fluctuations, possibly because the intake system is atmospheric instead
of charged. Based on Fig. 1.17, the flow rate appears somewhat lower for the
motored engine case, compared to the fired engine case. As such, the ‘exci-
tation level’ to trigger the gas dynamic resonances may be quite different for
Figure 1.17 – Simulated runner velocity for fired ( ) versus motored ( ) engine
(Source: [70])
both cases.
Flow reversal was predicted numerically in motored and fired engine con-
ditions. However, measurements using LDA in fired conditions do not show
flow reversal. Perhaps this is due to the limited spatial availability of measure-
ments: the LDA results are only obtained in a few points along a straight line,
downstream of the catalyst.
Regardless of differences in exhaust system geometries, the findings of the
present research are in agreement with those of Adam et al. [4], Liu et al.
[70] and Park et al. [81]. Section 5.3 discusses and explains the Helmholtz
resonances observed in the current study and in the literature.
Tsinoglou and Koltsakis [99] present a numerical study of catalyst hydrocar-
bon conversion efficiency in pulsating flow. The authors non-dimensionalize the
pulsation frequency with the catalyst residence time. This so-called pulsation
index is used to plot the conversion efficiency for different pulse shapes. For a
high pulsation index, the conversion efficiency reaches unity and becomes inde-
pendent of the pulsation index. For a low pulsation index, conversion efficiency
is well predicted by a quasi-steady model. The pulsation index is inversely
proportional to the scavenging number used in this thesis.
Lambert et al. [61] use a fast-response flame ionization detector which is
capable of measuring hydrocarbon trace gas concentrations at up to 1 kHz. The
local velocity is determined based on the time-of-flight of a propane trace gas
‘tuft’. The bandwidth of the overall measurement system (including a propane
injector, based on an inkjet printer valve) is roughly 160 Hz. One propane
pulse is injected every 16 crankshaft revolutions, at a preset delay with respect
to a crankshaft index pulse. This is too low to capture the fast dynamics in a
close-coupled system.
Nevertheless, the authors present time-resolved phase-locked velocity data
obtained in the center of an underbody catalyst, mounted on a fired engine
run at constant speeds between 1500 and 2000 rpm, and engine load6 varying
6 Throughout this thesis, the term engine load denotes the ratio of the engine torque T to
the maximum engine torque Tmax (NOT the ratio of the engine power T ω, relative to the
maximum engine power (T ω)max ). The adjectives zero, part and full load correspond very
Figure 1.18 – Experimental ( ) and simulated ( ) catalyst velocity for fired

engine (Source: [70])
between zero and part load. Remarkably, the authors show the occurrence of
velocity reversal in these conditions, in spite of the position of the underbody
catalyst, far from the engine.
Isochoric flow rigs: Motored

Numerical and experimental results found in the literature [70, 81, 59] on a
fired engine indicate flow reversal in the catalyst following each blowdown. Liu
et al. [70] present fired engine simulations ( ) in Fig. 1.18 that clearly exhibit
flow reversal. Surprisingly, measurements using LDA ( ) do not show flow
reversal. Perhaps this is due to the small number of LDA measurement points
where measurements could be performed.
Flow reversal has also been observed in the current research using a motored
engine flow rig (see Sect. 2.2.2). The occurrence of flow reversal throughout
the engine operating range may be surprising, considering the pressure drop
associated with the close-coupled catalyst (see the numerical example compar-
ing the pressure drop of different exhaust system components in Sect. 1.2.2).
The current research uses an exhaust manifold with free discharge into atmo-
sphere, i.e. without exit cone and cold end. The shape of the exit cone and
the backpressure of the cold end (including muffler and possibly an underfloor
catalyst) might influence the catalyst velocity distribution. However, the ab-
sence of the exit cone and tailpipe is not solely responsible for the occurrence of
flow reversal. Liu et al. [70], Park et al. [81] and Kim et al. [59] present results
showing flow reversal in a close-coupled catalyst in fired conditions, including
the entire exhaust system downstream of the CCC manifold. Flow reversal
consistently occurs after each blowdown. In the absence of blowdown such as
on an isothermal flow rig, flow reversal is not likely to occur. This is confirmed
by Hwang et al. [53] using LDA in a CCC manifold on an isothermal flow rig.
roughly to T /Tmax < 0.25, 0.25 6 T /Tmax < 0.25 and 0.75 6 T /Tmax 6 1 respectively.
Because of the (near) linear relation between the intake manifold pressure pi and the engine
torque, the engine load can also be considered the ratio of actual to maximum intake pressure
pi /pi,max .
1.4.3 Contributions of this thesis

The available literature on experimental and numerical studies of flow in ex-
haust systems is presented above. Due to the inherent variability in the designs
of these systems, the relevance of a number of studies [60, 104, 23, 79, 8] remains
restricted to particular geometries. This contrasts with research of in-cylinder
flows, where the shape of the combustion chamber is more or less identical
throughout different engines.
In spite of the geometric variability and the three-dimensional, pulsating
flow, the influence of geometry and flow conditions on the catalyst flow unifor-
mity [22, 17, 88, 99, 86], the validity of the addition principle (Eq. (4.1)) [88, 86],
catalytic conversion efficiency [99] and the pressure drop [109, 110] can be ex-
pressed in terms of appropriately chosen non-dimensional numbers, such as the
scavenging number (Eq. (1.4)) [88, 86] or similar numbers [22, 17, 99].
The main contribution of Persoons et al. [88] is the experimental validation
of the addition principle for two types of close-coupled catalyst manifolds: one
with and one without exhaust valve overlap. The study uses an isothermal
pulsating flow rig. Pulsating flow is generated using two different pulsators: a
rotating valve and a motored cylinder head. The set up is described in detail
in Sect. 2.2.1.
Two non-dimensional similarity measures are used to quantify the correla-
tion between pulsating and stationary velocity distributions. The selection of
these measures and their statistical characteristics are discussed in Sect. 4.3.2.
Both scalars are correlated to the non-dimensional scavenging number S (see
Sect. 4.3.3). S equals the ratio of apparent flow pulsation period to diffuser
residence time:
apparent flow pulsation period Tp

S= = (1.4)
diffuser residence time Ts
The scavenging number S is large for low engine speed, high flow rate or small
diffuser volume. In that case, exhausted flow pulses from each cylinder inter-
act only slightly, and there is a good correspondence between pulsating and
stationary distributions. When S is sufficiently large, the addition principle
is expected to be valid. S is small for high engine speed, low flow rate or
large diffuser volume. In that case, exhausted flow pulses interact to a higher
degree, likely resulting in a bad correlation between pulsating and stationary
distributions.
The findings in Persoons et al. [88] concerning the addition principle are
confirmed to some extent by Benjamin et al. [17] and Bressler et al. [22]. These
authors use a similar non-dimensional number to characterize the flow condi-
tions, as the ratio of flow pulsation time scale to diffuser residence time scale.
Persoons et al. [86] provide further confirmation of these findings [88], using
a charged motored engine (CME) flow rig. In spite of the different flow gen-
erated by both rigs, the data [88, 86] correlate remarkably well. An improved
definition of the apparent flow pulsation period is introduced, which takes into
account the higher frequency content in the isochoric rig’s exhaust flow. From
the correlation of the velocity distribution similarity measures versus the scav-
enging number S follows a critical scavenging number Scrit . Scrit may be
considered a dimensionless measure of the manifold efficiency with respect to
catalyst flow uniformization.
Chapter 4 discusses the experimental validation of the addition principle.
The relationship between this research and the findings of Benjamin et al. [17]
and Bressler et al. [22] are summarized in Sect. 4.5.
Isothermal flow rig experiments by Persoons et al. [88] revealed fluctua-
tions in the mean (or area-averaged) catalyst velocity, with a frequency that
is independent of engine speed and flow rate. Surprisingly, the strength of
the fluctuations depends on the type of pulsator used. Fluctuations are only
observed when using a cylinder head with poppet valves, not when using a
rotating valve. As explained in Chap. 5, Sect. 5.3, the cylinder head type of
pulsator generates a much stronger excitation for the flow inside the manifold.
The excitation is higher in amplitude and frequency content, compared to a
rotating valve.
Similar yet much stronger fluctuations are observed on a motored engine
flow rig [84]. These fluctuations are explained as Helmholtz resonances in
Sect. 5.3. In fact, a number of authors using isochoric flow rigs [81, 59, 4, 70, 14]
show similar resonances, although explanations as to their origin vary.
In summary, the main contribution of the publications that followed from
this thesis [88, 87, 86, 83, 84, 85] are:
• Experimental approach — The establishment of an experimental approach

that enables high spatial and temporal resolution velocity measurements
in the catalyst cross-section, in pulsating flow similar to fired engine con-
ditions [88, 87, 86, 84, 85].
• Addition principle — The experimental validation of the addition princi-

ple (Eq. (4.1)), based on the correlation between the scavenging number
S (Eq. (1.4)) and two non-dimensional measures that characterize the
similarity between velocity distributions obtained in pulsating and sta-
tionary flow. The validation is supported by statistical hypothesis testing
for these similarity measures [88, 86].
• Resonances — The strong mean velocity fluctuations that are observed

in isochoric flow rigs and numerical simulations of exhaust flow in fired
and motored engines [81, 59, 4, 70, 14] have been explained as Helmholtz
resonances. This assumption is confirmed by means of the frequency
response function of a one-dimensional gas dynamic model of the exhaust
manifold [84, 85].
• Flow reversal — The occurrence of instantaneous local flow reversal is

reported in close-coupled catalyst exhaust systems, subject to fired engine
conditions [70, 81, 59]. Given the inherent inability of HWA to discern
the velocity direction, all sources in the literature use an LDA system
with frequency shifting to obtain bidirectional velocity measurements.
1.6 Goals and scope 29
Problems with optical access and seeding particle concentration result in

a limited spatial and temporal resolution.
Through the introduction of a novel way of measuring bidirectional ve-
locity using oscillating hot-wire anemometry (see Chap. 3) [83], time-
resolved local flow reversal has been detected and quantified in an iso-
choric flow rig [84, 85].
1.5 Goals and scope of this thesis

The goal of this thesis is the experimental study of pulsating flow in modern
compact close-coupled catalyst exhaust manifolds for internal combustion en-
gines. Instead of using a fired engine involving a number of practical problems
due to the hot corrosive exhaust gas environment, the study uses cold pulsating
flow rigs. This should enable the use of velocity measurement techniques that
yield a high spatial and temporal resolution. The accuracy and resolution of the
obtained measurement data should enable detailed validation of computational
fluid dynamic calculations.
The thesis focuses on the most relevant flow-related aspect to the design
of the exhaust manifold: the catalyst velocity distribution. As discussed in
Sect. 1.2, obtaining a uniform catalyst velocity distribution is crucial for an op-
timal manifold design, in terms of minimal local catalyst degradation, minimal
pressure drop and maximal conversion efficiency.
In particular, the thesis investigates the influence of (i) transient flow bound-
ary conditions (similar to fired engine conditions) and (ii) geometrical aspects
(e.g. pre-catalyst diffuser volume, number and length of exhaust runners, ex-
haust valve timing) on the time-resolved catalyst velocity distribution.
Wherever the experimental data have difficulty in explaining the governing
physics, a one-dimensional gas dynamic model of the exhaust system shall be
used to further the understanding of the flow dynamics.
Not within the scope of this thesis are:
• Computational fluid dynamics — Part of the measurement results have

been used for the validation of a CFD approach, during a funded coop-
eration with an industrial partner [2]. The CFD calculations are per-
formed by the industrial partner. This thesis focuses on the experimental
methodology only.
• Catalytic chemistry kinetics — The catalytic reaction chemistry within

the converter under fired engine conditions. Nevertheless, the conversion
efficiency is estimated using a simplified homogeneous and heterogeneous
model of the reaction kinetics, as described in App. A.3.
• Other aspects — Some additional aspects in the design of exhaust man-

ifolds, such as heat transfer, acoustics and vibration are not within the
scope of this thesis.
1.6 Outline of this thesis

Chapter 2 presents the experimental approach, which encompasses (i) the
experimental flow rigs and (ii) the instrumentation and data reduction.
Section 2.2 describes the two flow rigs used during this thesis, each capa-
ble of generating a well-controlled pulsating flow in an exhaust manifold
with close-coupled catalyst.
Section 2.3 discusses the exhaust stroke flow similarity between the pul-
sating flow rigs (using cold flow) and the true conditions in a manifold
mounted on a fired engine. The thermodynamic analysis used in Sect. 2.3
is elaborated in App. B.
Section 2.4 discusses three different flow rate measurement techniques
used in this thesis. An accurate flow rate measurement is crucial to
serve as reference for the velocity distribution measurements. Section 2.4
explains how to overcome the particular issues involved in flow rate mea-
surements in pulsating flow.
Since the thesis deals primarily with measuring periodic phenomena,
Sect. 2.5 presents the basic concepts of data reduction for conditional
sampling and phase-locked averaging. The more advanced cycle-resolved
analysis is briefly introduced and demonstrated in practice in Sect. 2.5.2.
Chapter 3 presents the measurement technique used to determine the time-

resolved velocity distribution in the close-coupled catalyst. The selection
of hot-wire anemometry as the most suitable technique is clarified in
App. C.
Section 3.1 indicates the problems in obtaining accurate bidirectional
velocity measurements. In fact, hot-wire anemometry is intrinsically in-
sensitive to the velocity direction.
The remainder of Chap. 3 presents a novel oscillating hot-wire anemome-
ter (OHW) to overcome this problem. The methodology is presented in
Sect. 3.3. Section 3.4 describes the mechanical oscillator designed for this
purpose.
Section 3.6 discusses the calibration procedure, which allows to use the
OHW on the pulsating flow rig to quantify the time-resolved bidirec-
tional velocity. The OHW is calibrated in a custom-built wind tunnel
(Sect. 3.6.1). Laser Doppler anemometry has been used as a reference
velocity measurement, phase-locked with the oscillating probe’s motion
(Sect. 3.6.2).
Section 3.7 presents and interprets the calibration results. Section 3.7.3
formulates general selection and operation criteria for the OHW, based
on a non-dimensional scaling analysis.
Chapter 4 presents the experimental validation of the addition principle

(Eq. (4.1)). Section 4.3 discusses the elaborate data reduction used in
1.6 Outline 31
this chapter to assess the similarity between a pair of velocity distribu-

tions.
Section 4.4 presents the experimental results for both pulsating flow rigs
and both exhaust manifolds. The combined results are interpreted in
Sect. 4.5. The similarity measures introduced in Sect. 4.3 establish a
remarkable correlation with a single non-dimensional flow characteristic,
the scavenging number S (Eq. (4.42)). The correlations in Fig. 4.29 result
in a critical value for the scavenging number Scrit , which determines the
validity range of the addition principle.
Section 4.6 provides an physical interpretation to the findings of Sect. 4.5.
This complex flow behaves like a zero-dimensional scalar mixing process.
Using this analogy, Sect. 4.6.2 introduces the hypothetical concept of a
collector efficiency ηD , which equals the critical value of the scavenging
number Scrit . As indicated in Chap. 7, this hypothesis warrants further
research.
Chapter 5 discusses the time-varying flow phenomena in exhaust systems
with close-coupled catalyst, whereas Chap. 4 is more concerned with time-
averaged results.
Section 5.1 presents selected results of time-resolved velocity distribu-
tions for different pulsating flow rigs, exhaust manifolds and operating
conditions.
Section 5.2 discusses the occurrence of periodic flow reversal in the close-
coupled catalyst. Section 5.2.1 validates the oscillating hot-wire anemo-
meter (see Chap. 3) in conditions where considerable flow reversal is
known to occur in the catalyst. The validation is performed with re-
spect to integral flow rate measurements. The OHW is used to measure
bidirectional velocity in the isochoric pulsating flow rig.
The experimental data in Sect. 5.2.2 are obtained using the OHW ap-
proach, revealing the time-resolved velocity distribution throughout the
entire catalyst cross-section, including areas of negative velocity.
Section 5.2.3 describes the use of a one-dimensional gas dynamic model
of the exhaust system to simulate the occurrence of flow reversal in terms
of the mean velocity. Appendix D discusses the gas dynamic model in
detail.
Through numerical simulation, Sect. 5.2.3 establishes the influence on
catalyst flow reversal of the presence of the exit cone and cold end.
Section 5.3 discusses the resonance fluctuations observed in the time-
resolved mean catalyst and runner velocity. The same phenomenon is
observed by other authors, yet has never been satisfactorily explained.
Section 5.3.2 provides an analytical explanation of the Helmholtz reso-
nance.
Section 5.3.4 uses the same one-dimensional gas dynamic model (see
App. D) to explain the resonance phenomenon numerically. Gas dynamic
frequency response functions of the exhaust manifold are determined, re-

vealing the nature of the resonating system responsible for these strong
velocity fluctuations in the catalyst.
Chapter 6 formulates some brief comments that interlink the findings of
Chaps. 4 and 5, and explains how the addition principle’s validity is
influenced by the flow dynamics.
Chapter 7 formulates the general conclusions of this thesis, and suggests some
future research opportunities.
Appendix A provides an overview of the mechanisms of wall friction and
other sources of pressure loss, as well as catalytic reaction kinetics in
automotive catalysts.
Appendix B contains the thermodynamic analytical derivation for assessing
the exhaust stroke flow similarity between the pulsating flow rigs using
in this thesis and actual fired engine conditions. These derivations are
used in Sect. 2.3.
Appendix C reviews the advantages and disadvantages of some competing
techniques for measuring time-resolved gas velocity (i.e. thermal and op-
tical anemometry).
Appendix D describes the numerical one-dimensional gas dynamic model
that has been implemented to help understand the flow dynamics (see
Sects. 5.2.3 and 5.3.4). A brief explanation is given on decoupling and
discretizing the Euler equations. The gas dynamics code is validated
using some benchmark problems. Finally, some comments are given on
identification methods by means of multisine signals.
Chapter 2
Experimental approach
“The men of experiment are like the ant; they only collect and use: the
reasoners resemble spiders, who make cobwebs out of their own substance.
But the bee takes a middle course; it gathers its material from the flowers
of the garden and of the field, but transforms and digests it by a power of
its own.”
Francis Bacon (English philosopher, ◦ 1561, †1626)
This chapter presents the experimental approach, comprising two parts: (i) the
experimental setup and (ii) the instrumentation and data reduction.
Sections 2.1 and 2.2 describes the exhaust manifolds and the two pulsat-
ing flow rigs used during this thesis. To facilitate the velocity measurements,
the pulsating flow rigs operate using air at ambient temperature, whereas the
exhaust manifold in a fired engine is subjected to hot, corrosive exhaust gas.
Section 2.3 discusses the exhaust stroke flow similarity between the pulsating
flow rigs and the fired engine conditions.
Section 2.4 discusses the flow rate measurement techniques, which serve as
a reference for the velocity distribution measurements. Section 2.5 presents
the basic concepts of data reduction for conditional sampling and phase-locked
averaging. Section 2.5.2 briefly discusses the effects of cyclic variation, which
is a typical phenomenon occurring in volumetric reciprocating machinery.
2.1 Exhaust manifolds

Two exhaust manifolds with close-coupled catalyst have been used during the
research, denoted manifolds A and B. The specifications are given in Table 2.1.
Each manifold is designed for an indirect injection petrol internal combustion
33
34 Chapter 2 Experimental approach
engine. Both engines are typical recent generation Otto engines, featuring
double overhead camshafts, four valves per cylinder, and a cross-flow pent-roof
type combustion chamber. The engine block and cylinder head are made of cast
aluminium. The bore-to-stroke ratio is approximately 1 : 1. The compression
ratio is 10 : 1. The intake is atmospheric and throttle valve controlled. The fuel
is injected in a multi-point sequential manner, and is electronically controlled.
The specific brake torque is between 90 and 95 Nm/l at 4000 rpm. The specific
brake power is between 45 and 49 kW/l at around 6000 rpm. Both engines
comply with Euro III emissions standards.
Table 2.1 – Exhaust manifold specifications

Manifold A Manifold B
Engine 3175 cm3 V-6 1199 cm3 I-4
Valve timing7 −12 | 242 | −246 | 10 −13 | 220 | −220 | 13
Firing order8 1–4–2–6–3–5 1–3–4–2
Runners 31.5 mm, L = 150, 90, 28.0 mm, L = 160, 80,
120 mm 160, 80 mm
Diffuser volume Vd = 141.4 cm3 Vd = 390.2 cm3
Catalyst ceramic 3/600 monolith, square channels
circular 63 mm, oval 151×101 mm,
L = 52 mm L = 137 mm
Figure 2.1 shows manifolds A and B, indicating the runner numbering that
starts from the engine’s distribution side. The exhaust gas oxygen sensor up-
stream of each catalyst has been replaced by a flush-mounted plug that accom-
modates a static pressure tap and a temperature probe.
The choice of manifolds A and B is not arbitrary: Manifold A is fitted to the
head of one cylinder bank of a V-6 engine. In fact, two very similar manifolds
are used on either side of the engine. Cylinders are numbered starting at
the distribution side from 1 to 3 and 4 to 6, for the first and second bank
respectively. The firing order is 1–4–2–6–3–5. As such, the phase difference
between the valve timing of three cylinders in one bank is 720 ◦ ca/3 = 240 ◦ ca.
This phase difference is comparable to the total exhaust valve opening period
(Table 2.1). Therefore, the exhaust strokes of individual cylinders occur nearly
sequentially. The exhausted flow pulses issue into the exhaust manifold without
overlap between the cylinders.
By contrast, manifold B belongs to an inline four-cylinder engine, with a
phase difference between the valve timings of 720 ◦ ca/4 = 180 ◦ ca. This leads
to an overlap of roughly 60 ◦ ca between the exhaust strokes. The exhaust flow
pulses in manifold B interact to a higher degree when compared to manifold
8 The valve timing is given as four crankshaft positions (◦ ca), relative to top dead center
prior to the intake stroke. IO | IC | EO | EC correspond respectively to intake valve opening

and closing (IO, IC) and exhaust valve opening and closing (EO, EC).
8 The V-6 engine features two exhaust manifolds. Manifold A corresponds to the under-
lined cylinders 1, 2 and 3.

2.1 Exhaust manifolds 35
y x
y
x
1
1
2 2
3
3 4
(a) (b)
Figure 2.1 – Exhaust manifolds (a) A and (b) B
A. This is important with respect to the findings in Chap. 4 concerning the

validity of the addition principle.
Note that this valve overlap should not be confused with the general concept
of valve overlap in four-stroke internal combustion engines. The general intra-
cylinder valve overlap denotes the difference between exhaust valve closing and
intake valve opening of an individual cylinder. The above discussed overlap
between the exhaust strokes of different cylinders shall be denoted the inter -
cylinder valve overlap to avoid confusion.
No exit cone or tailpipe is connected to either manifold, thus allowing access
to measure the velocity distribution in the catalyst. Since only cold pulsating
flow is used in this research, this poses no problems, other than a high noise
level. The velocity distribution could be slightly influenced by the absence of
the exit cone and the backpressure of the tailpipe. However, as indicated in
Sect. 1.4.2, the occurrence of flow reversal in particular is not influenced by the
presence or absence of the exit cone.
The gas exits the catalyst monolith as tiny laminar jets. Each jet experi-
ences a sudden expansion, according to the open frontal area ratio. At a normal
distance to the outlet face z = 0 mm, the velocity distribution contains very
high transverse gradients due to the blockage by the channel walls. Further
downstream as z increases, the neighboring jets are mixing, which averages out
the transverse gradients. This mixing process occurs at a transverse length
scale comparable to the channel diameter d ' 1 mm.
Measurements show that this small-scale mixing process is complete for
p i, t i ∆p θ /2, ω /2
p cat
Roots throttle ISO standardized
compressor valve flow rate orifice
duct section with

mounting plate
U
surge vessel
Figure 2.2 – Isothermal flow rig, with cylinder head mounted
z > 20 mm. All measurements are performed in a measurement plane at

z = 25 mm. Instead of the exit cone, a cylindrical exit sleeve is mounted
downstream of the catalyst, with the same perimeter. Since the exit sleeve
is 40 mm long, the velocity is measured well within the sleeve, thus avoiding
entrainment of surrounding air.
The transverse length scale of the interesting features of the velocity distri-
bution is of the order of 10 to 100 mm (i.e. the order of magnitude of the runner
or catalyst containment hydraulic diameter). The large-scale transverse veloc-
ity gradients are largely unchanged at z = 25 mm, since their characteristic
length scale is one to two orders of magnitude greater than d.
Thus, the measurement position z = 25 mm is a compromise between
(i) avoiding the laminar jet mixing region and (ii) retaining the large-scale ve-
locity distribution. Other researchers choose a similar measurement position.
For instance, Arias-Garcia et al. [7] suggest z = 30 mm, based on doctoral
thesis research by Clarkson [28]. Lemme and Givens [64] show that the jets are
sufficiently mixed at z = 25.4 mm.
2.2 Pulsating flow rigs

2.2.1 Isothermal flow rig
The isothermal dynamic flow bench consists of a surge vessel with removable
duct section and mounting plate. Figure 2.2 shows a schematic diagram of
the set up. The isothermal flow rig mimics flow from infinitely large combus-
2.2 Pulsating flow rigs 37
2
1
Figure 2.3 – Rotating valve ( 1 stator, 2 rotor, 3 rotor bearing block and motor
coupling)
tion chambers. The exhaust stroke flow similarity with respect to true engine
conditions is discussed in Sect. 2.3.
A Roots compressor delivers a maximum air flow rate of 350 Nm3/h at
300 mbar overpressure. The compressor is driven by a variable speed drive.
The compressor feeds the surge vessel via a long pipe section with a standard-
ized flow rate measurement orifice. A throttling valve is inserted between the
compressor and the long pipe section.
The flow rate is determined using an orifice, according to ISO Standard
5167-1991(E) [54]. The standard restricts the use of the orifice to steady or
slowly varying flows. The surge vessel’s volume provides adequate damping of
the pulsations caused by the pulsating flow generator, based on ISO Technical
Report 3313-1998(E) [55]. This is discussed more in detail in Sect. 2.4.1. The
additional uncertainty on the orifice flow rate reading in pulsating flow caused
by using the time-averaged value of the differential pressure is below 0.5 % for
all measurement conditions. However, the orifice reading is only used during
stationary operation to check the flow rate calculated from the catalyst velocity
distribution. The agreement is well within the error bounds of the orifice
measurement (' 5 %).
For manifold A, both a rotating valve and the original cylinder head have
been used to generate the pulsating flow on the isothermal flow rig.
The rotating valve has been chosen since it constitutes a very simple way
of generating pulsating flow. Other authors have used e.g. a rotating disk [22,
7, 17, 58] to obtain the same result. Furthermore, in the open position, the
rotating valve does not obstruct the flow in any way, whereas the poppet valves
of a cylinder head always disturb the flow. This lack of obstruction is partic-
ularly advantageous for comparison of results to numerical simulations, since
the rotating valve requires fairly simple inlet boundary conditions.
For manifold B, only the corresponding cylinder head is used as pulsator.
Figure 2.4 – Isothermal flow rig, with rotating valve and manifold A mounted ( 1
surge vessel, 2 duct section with trace gas injectors, 3 rotating valve)
The same surge vessel is used, with a modified duct section and pulsator mount-
ing plate.
Figure 2.3 shows the components that make up the rotating valve. It con-
sists of a solid cylinder rotor with rectangular holes, that rotates with a tight
tolerance in a stator. O-ring seals are inserted between neighboring exhaust
port sections of the rotating valve, thereby eliminating cross-flow within the
valve body (not shown in Fig. 2.3). The geometry of the holes corresponds
to the original cylinder head’s exhaust port cross-sectional area, and to the
exhaust valve timing.
Figure 2.4 shows the isothermal flow rig, with the rotating valve and man-
ifold A mounted. In the base of the duct section ( 2 in Fig. 2.4), one solenoid
actuated injector is incorporated for each duct, i.e. for each cylinder. These
injectors have been used to inject small amounts of trace gases (e.g. CH4 , NO,
CO). Each injector is controlled to inject sequentially, during the exhaust pe-
riod of its cylinder. The objective of the trace gas injection is to determine the
partial load on the catalyst from each cylinder. These measurements are not
directly relevant within the scope of this thesis.
The rotating valve is driven by an electric motor, which is controlled by a
variable speed drive. The motor is located behind the valve in Fig. 2.4. The
required motor power is low, since only the valve’s internal friction must be
overcome. The valve shaft position is monitored by an angular encoder, which
is not shown in Fig. 2.4. It is mounted on the protruding end of the valve shaft.
The encoder9 is of the sinusoidal incremental type, featuring an angular
accuracy better than 0.1 ◦ . The index pulse of this encoder is used to trig-
ger the hot-wire anemometer, thus phase-locking the measurement to the flow
pulsation.
9 Heidenhain ERN 180, 3600 lines per revolution, sinusoidal incremental interface
Figure 2.5 – Isothermal flow rig,

Figure 2.6 – Isothermal flow rig, with
with cylinder head and manifold A
cylinder head and manifold B mounted
mounted
A more powerful speed-controlled electric motor drives the cylinder head’s

exhaust camshaft using a timing belt. Figure 2.5 shows the isothermal flow
rig set up with manifold A and its cylinder head mounted. The encoder is
mounted on a shaft extending from the camshaft’s timing belt pulley. In the
case of manifold A, the intake camshaft remains stationary, in such a position
that all intake valves are closed. This is possible because of the lack of inter-
cylinder valve overlap (above). To eliminate any leakage due to marginally
opened intake valves, the intake ports are sealed as well. As Fig. 2.5 shows,
the surge vessel is inclined so that the flow exits the catalyst horizontally, with
the sole purpose of facilitating the velocity distribution measurements.
For manifold B, the intake and exhaust camshafts are driven by a timing
chain mechanism, including a chain guide and oil-pressurized chain tensioner.
As Fig. 2.6 shows, the chain mechanism is contained within the original timing
chain cover and a backing plate. The timing chain drives both camshafts. On
the isothermal flow rig, the crankshaft sprocket is not driven by the crankshaft
(since there is no crankshaft). Instead, a customized chain drive shaft is used
that fits the original sprocket. The shaft is supported by two roller bearings,
and sealed using the original crankshaft oil seal in the timing chain cover. A
timing belt is used to connect the electric motor and the chain drive shaft.
The rotary encoder is mounted on the chain drive shaft. The intake ports are
sealed externally in the same way as for manifold A. The motion of the intake
valves could slightly influence the flow inside the flow rig ducts, but this is of
no interest to the present research.
When the cylinder head is used to generate pulsating flow, forced lubrication
∆p p i, t i p cyl
p cat
tcat
screw buffer pressure laminar
compressor vessel regulator flow meter
θ, ω
Figure 2.7 – Isochoric flow rig
is required for the moving parts (e.g. camshaft bearings, followers, tappets,
poppet valves). Also, several components require a sufficiently high oil pressure
for proper operation (e.g. hydraulic valve clearance adjusters, timing chain
tensioner). Therefore, an externally driven pump (shown in the bottom left
of Fig. 2.5) supplies the cylinder head with engine lubricating oil through the
appropriate feed channels, at a pressure of approximately 5 bar. The oil returns
from the cylinder head through the force of gravity, along the original oil drain
channels. The returning oil flows to a reservoir (shown underneath the surge
vessel in Figs. 2.5 and 2.6), from which it is pumped back up through a filter
in a continuous loop.
2.2.2 Isochoric flow rig: Charged motored engine (CME)

The isochoric or charged motored engine (CME) flow rig consists of a four-
cylinder internal combustion engine, mounted on a dynamic engine test stand
with an electric DC motor. The engine corresponds to manifold B. Its specifi-
cations are given in Table 2.1.
Figure 2.7 schematically depicts the CME flow rig. The compressed air is
produced using a screw compressor, which delivers a maximum flow rate of
250 Nm3/h (' 300 kg/h) at 8 atm(10 ). A pressure regulator ( 2 in Fig. 2.9)
maintains a constant pressure in the engine intake system, varying between
10 Throughout the thesis, the standard unit of pressure is atm, defined as 1 atm =
101325 Pa. Unless stated otherwise, all pressures are given in absolute values. Pressures
relative to atmospheric pressure (= 1 atm) are indicated in atm-r.
Experimental approach
Pulsating flow rigs (2)
Introduction
Goal and scope pcyl pcyl
Overview
Fired engine CME flow rig
Experimental approach combustion
Flow rigs
Similarity
Oscillating hot-wire EVO EVO
Approach
Calibration
Validation pres engine pint > 1 engine
Addition principle load pres load
Approach
Results pexh ≅ 1 pexh ≅ 1
Summary pint < 1
Discussion Vcyl Vcyl
Experimental dynamics TDC BDC TDC BDC
Results
Flow reversal
Helmholtz Figure 2.8 – Comparison of the indicator diagram for (a) a fired engine and (b) the
Numerical dynamics
Model
isochoric flow rig
Results
Helmholtz
Conclusion
1.00 atm to 2.25 atm in the current study. The exhaust stroke flow similarity
(Sect. 2.3) shows that this roughly corresponds to the range of zero to full
engine load in true fired engine conditions. The screw compressor’s maximum
flow rate limits the engine speed in the performed measurements to 3000 rpm.
Figure 2.8 shows the difference between a fired gasoline engine and the
isochoric flow rig in terms of their schematic indicator diagram, showing the
evolution of cylinder pressure pcyl versus cylinder volume Vcyl . TDC, BDC and
EVO respectively
Doctoral denote– top
defence Tim Persoons dead
11 May center,
2006 bottom dead center and exhaust valve
(dry run) 1/20
opening. In spite of the differences, the residual cylinder state (immediately
prior to the exhaust valve opening) forms the initial conditions for the exhaust
stroke. Therefore, only the residual state is of importance to obtaining a similar
exhaust stroke in the isochoric flow rig.
Figure 2.8a shows that the intake pressure pint for the fired engine is below
atmospheric pressure, due to the throttled intake manifold. After compression,
combustion and expansion, the residual cylinder pressure pres is indicated. The
exhaust stroke itself is indicated as a bold line.
The indicator diagram for the isochoric flow rig (Fig. 2.8b) features substan-
tial differences. The intake pressure pint is greater than atmospheric pressure,
resulting in a much higher pressure following the compression stroke compared
to the fired engine. Since there is no combustion phase, there is no pressure
rise near top dead center. Theoretically, the pressure would decrease along
the same line during the expansion stroke. However in reality, the pressure is
lower during the expansion stroke, as a result of losses. Gas mass is lost from
the cylinder due to blow-by leakage, and internal energy is lost due to heat
exchange with the walls. Most importantly, the exhaust stroke resembles quite
well to the fired engine case, given the appropriate setting of the intake system
pressure pint .
In a fired engine (Fig. 2.8a), the engine load is varied by adjusting the
throttle, which changes the intake manifold density and thereby the amount of
heat released during combustion. As a result, also the residual pressure varies.
In the isochoric flow rig (Fig. 2.8b), the residual cylinder state is varied
by changing the setting of the intake manifold pressure pint , by means of the
pressure regulator ( 2 in Fig. 2.9).
The engine is motored at a constant speed by means of the test stand DC
motor ( 1 in Fig. 2.9). The aforementioned encoder9 records the crankshaft
position ( 4 in Fig. 2.9). To enable charging the engine with compressed air,
the original intake system has been replaced by a reinforced intake system with
identical manifold volume and runner dimensions. The engine is run without
combustion and fuel injection, to obtain cold clean pulsating flow in the ex-
haust system. The original exhaust valve timing as mentioned in Table 2.1 is
unchanged. However, the intake camshaft is retarded by 30 ◦ ca to avoid unphys-
ical blow-through from the high-pressure intake to the low-pressure exhaust
system during intake/exhaust valve overlap. The valve timing used during the
experiments is therefore 17 | 250 | −220 | 13.
Applying an intra-cylinder or intake/exhaust valve overlap period is com-
mon to all four-stroke internal combustion engines. The valve overlap maxi-
mizes the indicated efficiency. In an engine with a pressure wave tuned exhaust
system, a rarefaction (expansion) wave arrives at the exhaust port during valve
overlap. This negative pressure (i) helps to scavenge the combustion chamber
of exhaust gas, and (ii) helps to start the induction of fresh mixture from the
intake runner. At a low engine speed or in case of excessive valve overlap, back-
flow can occur from exhaust to intake system. Blow-through of fresh mixture
from intake to exhaust rarely occurs for naturally aspirated engines.
For turbo- or supercharged (fired) engines, usually a shorter valve overlap
(or none at all) is used to prevent blow-through of fresh mixture from intake
to exhaust system. The amount of blow-through is determined by the ratio of
intake to exhaust system pressure.
For the CME flow rig, the exhaust pressure is nearly atmospheric, while
the intake system pressure reaches a maximum of 2.5 atm. Using the original
(naturally aspirated) engine’s valve timing would result in a large blow-through
flow rate, which is not present in fired engine conditions. This would cause an
additional peak flow rate near the end of the exhaust stroke. As such, the
valve overlap is removed by retarding the intake camshaft. The effect of the
retarding is indicated in the schematic indicator diagram in Fig. 2.8b.
Figure 2.9 shows how the engine is installed on the test stand. The engine
is mounted without vibration dampers onto the rigid test stand frame. The
velocity measurement probe is mounted on an automated positioning system
( 5 in Fig. 2.9) which is fixed securely onto the lab floor, adjacent to the test
stand. Much care is taken to avoid any relative motion between the engine
exhaust system and the velocity probe.
The intake system flow rate is measured using a laminar flow element meter
or LFE ( 3 in Fig. 2.9). The LFE has been calibrated against the ISO stan-
dardized orifice used on the isothermal flow rig. Section 2.4.2 discussed the
details of this system. Partly because of the altered intake timing, the intake
flow rate is highly pulsatile with periods of extensive backflow.
Although the LFE is an appropriate choice of flow rate measurement for
1
2
4 5
5
3
2
Figure 2.9 – Isochoric (CME) flow rig, with manifold B mounted ( 1 test stand DC
dyno, 2 intake pressure regulator, 3 LFE, 4 encoder, 5 velocity probe positioning
system)
such flows (see Baker [9]), the intake system flow rate is further verified using
a piezo-electric cylinder pressure sensor. The pressure rise during the compres-
sion stroke is used to determine the inducted mass per cylinder per cycle, thus
yielding the mass flow rate. Section 2.4.3 elaborates on the cylinder pressure-
based flow rate measurement. The intake flow rate reading is accurate to within
5 to 10 %, and serves as a reference measurement for the flow rate obtained by
area-averaging the catalyst velocity distribution.
The cold pulsating flow generated by the CME flow rig in the exhaust
system is quite different from the isothermal flow rig. By controlling the intake
system pressure, the residual cylinder pressure at exhaust valve opening (EO)
can be adjusted. This results in a two-stage exhaust stroke with blowdown and
displacement phases, which is typical of fired engine conditions. The CME flow
rig therefore produces cold pulsating flow, which greatly facilitates the velocity
measurements yet still is very similar to fired engine exhaust conditions.
2.3 Exhaust stroke flow similarity

Ultimately, the research presented in this thesis must apply to flow in exhaust
manifolds in actual fired engine conditions. As discussed in the previous sec-
tions, there are some substantial differences with fired engine conditions. Both
isothermal and isochoric flow rigs differ mainly in terms of the temperature of
the flow.
This section aims to quantitatively assess the similarity of the flow gener-
ated by the isothermal and isochoric pulsating flow rigs, with respect to fired
engine conditions. Furthermore, this section explains the proper setting of the
operating parameter pi for the isochoric flow rig to resemble a given engine
load in fired conditions.
2.3.1 Introduction
The exhaust stroke flow similarity between fired engine conditions and the
isothermal and isochoric flow rig are investigated numerically and analytically.
Numerical simulations are performed using a filling-and-emptying internal
combustion engine model written in MATLAB11 . The engine model is based
on Watson and Janota [107] and Heywood [47]. The engine is modeled as zero-
dimensional volumes (e.g. intake and exhaust manifold, cylinders) combined
with quasi one-dimensional pipes for the intake runners (see remark below).
Within each volume, the equations of conservation of mass and energy are
solved, combined with the ideal gas equation. The model uses the appropriate
descriptions for compressible restricted flow over intake and exhaust valves.
The combustion process is modeled using a Wiebe law for single-zone heat re-
lease. Heat loss to the combustion chamber walls is incorporated, based on the
generally accepted correlations by Woschni [113]. Blow-by leakage is taken into
11 MATLABTM and SimulinkTM are products of The MathWorks, Inc., 3 Apple Hill Drive,
Natick (MA) 01760-2098, USA (http://www.mathworks.com)

2.3 Exhaust stroke flow similarity 45
account based on experiments on the CME flow rig (see Fig. 5.21). The model
is solved by means of fixed step fourth order Runge-Kutta time integration.
For reasons of numerical stability, the time step is set to different (yet fixed)
values during the intake and exhaust stroke and during the compression and
expansion stroke.
Background The term quasi one-dimensional may be interpreted as

follows. The runners are not discretized along their length, as is the case
for a true one-dimensional gas dynamic model, as described in App. D.
Instead, a single momentum equation is solved that represents the mo-
mentum of the entire gas mass contained within the runner as a single
(incompressible) plug flow. This equation is used to estimate the stagna-
tion pressure upstream of the intake valves. The gas dynamics are very
crudely simplified in this way. This means a considerable reduction in the
calculation time with respect to a fully one-dimensional model, yet still
estimates the intake ram effect to some extent. A one-dimensional gas
dynamic model (implemented in Simulink11 ) has been used in this thesis
for the investigation of the resonance fluctuations observed in the time-
resolved flow rate through the close-coupled catalyst exhaust manifold
(see Sect. 5.3).
For an engine speed of 1800 rpm and an exhaust flow rate of 100 m3/h
(corresponding to part load conditions), Fig. 2.10 shows the time-resolved non-
dimensional velocity in runner 1 of manifold B. The solid ( ) and dashed line
( ) represent simulations performed for the CME and isothermal flow rig,
respectively. The markers ( ) indicate the runner velocity measured on the
CME flow rig. The non-dimensional exhaust valve lift is plotted in grey.
Figure 2.10 demonstrates that the calculated runner velocity using the
filling-and-emptying model ( ) compares reasonably to the measured veloc-
ity ( ) during the blowdown phase. However, the measured runner velocity
fluctuates substantially during the displacement phase, whereas the simulation
does not. This is due to the oversimplified numerical model, which models none
of the gas dynamics of the exhaust system. The resonance fluctuations are dis-
cussed in detail in Sect. 5.3. In Sect. 5.3.4, a more advanced one-dimensional
gas dynamic model is used which does capture these fluctuations.
Figure 2.10 shows no curve for a fired engine. The runner velocity for a fired
engine is qualitatively the same as for the CME flow rig, except that the ratio
of the peak velocity values during blowdown and displacement is somewhat
different. This is explained in the following section.
The measured runner velocity is only indicative. A hot-film sensor is fixed
flush with the runner wall, at the entrance to the runner. As such, the mea-
surement indicates rather the wall shear stress than the mean runner velocity.
Measuring the actual velocity distribution using a hot-wire probe proved very
difficult in that location, since the wire tends to break due to the highly pul-
sating flow, that reaches peak velocities up to M a = 1. Furthermore, the
temperature in the exhaust system drops as the intake pressure is increased.
This is caused by the reduction in internal energy during the compression and
expansion stroke, due to heat loss and blow-by leakage. As the temperature in
8
CME (measured)
7 CME (simulated)
ISOT (simulated)
Runner velocity U(r) (-)
Exhaust valve lift (-)

5
0
450 540 630 720
Crankshaft angle ω t (°)
Figure 2.10 – Exhaust runner velocity for isothermal and isochoric flow rig
the exhaust system drops below the dew point temperature, water condenses
from the air. These condensate droplets form tiny ice particles as the temper-
ature drops below 0 ◦ C at high intake pressure. The fast moving ice particles
might also contribute to sensor wire breakage. The problem is only encountered
while attempting to measure the velocity in the runner entrance. Downstream
of the catalyst, no such problem arises.
The isothermal flow rig produces a single-stage exhaust pulse, resulting in
nr quasi-sinusoidal pulses per engine cycle, where nr is the number of runners
issuing into the catalyst. The CME flow rig produces a pulsating flow that
strongly resembles fired engine conditions. The two-stage exhaust pulses are
more distorted, resulting in an exhaust flow rate with higher frequency con-
tent. The difference between measured and simulated velocity in Fig. 2.10 is
due to Helmholtz resonances that are most pronounced during the displace-
ment phase. The filling-and-emptying engine model does not incorporate a
sufficiently accurate exhaust system model to capture this effect.
2.3.2 Thermodynamic analysis

The analysis in this section is based on a purely analytical derivation of the sim-
plified thermodynamic evolution of the gas in an reciprocating engine, between
the intake stroke and the exhaust stroke. The derivation aims to provide an
analytical expression for the flow rate during the blowdown and displacement
phase, as a function of the engine load (i.e. intake system pressure) and the
presence of an internal combustion phase.
Some assumptions are required to be able to extract an analytical expres-
sion. The resulting expression needs not be 100 % accurate, since the purpose
of the derivation is foremost to assess the exhaust stroke flow similarity between
the CME flow rig and a fired engine.
The derivation is performed in three consecutive stages:

(1) Evolution between the end of the intake stroke until the residual state,
i.e. immediately before the opening of the exhaust valves.
(2) The blowdown phase, lasting until the residual cylinder pressure has
equalized to the exhaust system pressure.
(3) The displacement phase, where the piston motion determines the flow
rate through the exhaust system.
The complete derivation is elaborated in App. B. Here, a brief overview is
given. Air is taken as working fluid, with thermodynamic properties evaluated
at a fixed mean temperature. During stage (1), the in-cylinder heat loss and
blow-by leakage are neglected. The relation between intake and residual state
follows from the conservation of mass and energy (Eq. (B.1)):
γ γ−1 !
ρe Vi pe Vi ∆Tc V0
= ; = 1+
ρi Ve pi Ve Ti Vi
where ρ is the density [kg/m3 ], p is the pressure [Pa], V is the cylinder vol-
ume [m3 ], T is the temperature [K] and the subscripts i, e, 0 respectively denote
intake valve closing, exhaust valve opening and top dead center. The adiabatic
temperature rise due to combustion equals ∆Tc = φSf / (cv Lf ), where φ is
the product of the equivalence ratio and the combustion efficiency [-], Sf is
the lower heating value of the fuel [J/kg], cv is the specific heat capacity at
constant volume [J/(kg K)] and Lf is the theoretical air-to-fuel ratio [kg/kg].
Stage (2) or the blowdown phase is regarded as the expansion of the resid-
ual cylinder pressure at constant cylinder volume. The mass flow rate over the
exhaust valves is determined assuming compressible restricted flow, with the
appropriate discharge coefficient (see Fig. B.1). The valve lift curve is approx-
imated by a cosine function, which in turn is further approximated at small
lift values by a parabolic function. This yields a non-linear partial differential
equation that gives the evolution of the mass of gas m remaining in the cylinder
(Eq. (B.8)):
√ γ+1 −γ !
ne π 3 de he 2 2 rTe m

d m 2
pa m
= −Cd 2
ω t f
dt me ∆θ Ve me pe me
This equation cannot be solved due to the non-linear function f , which

originates from the equation for compressible flow through the exhaust valve
throat. As such, f is approximated by a function such that Eq. (B.8) can be
solved (see Eq. (B.10)). From the solution of this partial differential equa-
tion follows an Eq. (2.1), describing the maximum mass flow rate during the
blowdown phase ṁ1 [kg/s].
Stage (3) or the displacement phase is regarded as volumetric expulsion of
gas at constant pressure. From the conservation of mass and the assumption
of constant density follows Eq. (2.2), describing the maximum mass flow rate
during the displacement phase ṁ2 [kg/s].
In Eqs. (2.1) and (2.2), term i represents the influence of the intake system
pressure. It varies roughly between 0.25 and 1 for a fired engine and between 1
and 2.5 for the CME. Term ii represents the influence of the temperature rise
during the combustion process. For the CME flow rig, there is no combustion,
reducing term ii to 1. For fired engine conditions, term ii equals roughly 3.5.
√ 1 γ−7
ne π 3 de he rTi 3 Vi
6
ṁ1 = ρi Vi ω 2Cd
∆θ2 ωVi Ve
γ−8
!
γ−1 6γ − 3γ 4
∆Tc V0 pi
· 1+ ·
Ti Vi pa
| {z } | {z }
ii i
 
1
 p γ1 V γ−1 γ !
 i i ∆Tc V0 
· 1+ − 1 (2.1)

 pa Ve Ti Vi 
| {z } | {z }
i ii
− γ1 γ−1 ! γ1
πb2

ω pi ∆Tc V0
ṁ2 = ρi s 1+ (2.2)
4 2 pa Ti Vi
| {z } | {z }
i ii
As the engine load increases, the intake system pressure (or equivalently
term i) increases. In that case, Eqs. (2.1) and (2.2) show that the peak mass
flow rate increases during blowdown (∂ ṁ1 /∂pi > 0) and decreases during dis-
placement (∂ ṁ2 /∂pi < 0). For the CME flow rig, in the absence of combustion,
the intake system pressure should result in peak flow rates comparable to fired
engine conditions. An appropriate change in term i should compensate for the
change in term ii. Figure 2.11 shows the evolution of ṁ1 and ṁ2 according to
Eqs. (2.1) and (2.2) versus the intake pressure, for fired and CME conditions.
ṁ1 and ṁ2 are non-dimensionalized using a reference exhaust flow rate
ṁref [kg/s], assuming a volumetric efficiency of unity:
πb2

ω 720
ṁref = ρi s (2.3)
4 4π 180
The symbols are defined in App. B. ṁref corresponds to a hypothetical exhaust
stroke lasting 180 ◦ ca (hence the factor 720/180 in Eq. (2.3)), where the total
gas mass is exhausted at a constant mass flow rate ṁref .
Figures 2.11 through 2.16 each compare the exhaust stroke of the fired
engine (left) to the CME flow rig (right). Each plot features the intake manifold
pressure pi /pa in abscissa, which corresponds to a range of engine load from
low to high, from left to right. The solid lines ( ) result from the analytical
derivation, whereas the markers ( , , ) result from simulations using the

filling-and-emptying engine model.
For the CME flow rig at low intake pressure pi , ṁ1 is negative because of
the early opening of the exhaust valve. The cylinder volume at intake valve
closing Vi is smaller than the volume at exhaust valve opening Ve . This is due
to the retarding by 30 ◦ ca of the intake camshaft, which is not present for the
fired engine with original intake valve timing.
Retarding the intake camshaft avoids unphysical blow-through during
(intra-cylinder) valve overlap between intake and exhaust valves. Retaining
the original valve overlap would result in an excessively high peak flow rate
crossing the combustion chamber from the high pressure intake system to the
nearly atmospheric pressure in the exhaust system. This does not occur in the
fired engine and would disturb the flow during the final stage of the exhaust
stroke on the CME flow rig. Thus, the intake camshaft timing is retarded.
The high ratio of blowdown to displacement peak flow rate ṁ1 /ṁ2 can only
be achieved on the CME flow rig by increasing the intake pressure pi /pa to
roughly 5. In that case however, without altering the compression ratio, the
maximum cylinder pressure is too high. Furthermore, because of in-cylinder
heat loss, blow-by leakage and the fact that Vi /Ve < 1, the exhaust flow tem-
perature drops below 0 ◦ C roughly when pi /pa > 2.5. In that case, water vapor
condenses from the air and freezes inside the exhaust manifold. As explained
earlier, this inhibits the use of hot-wire probes for measuring the flow inside
the runner. Furthermore, the ice deposits gradually block the small catalyst
channels. Possibilities for extending the operating range include using an air
heater in the intake system or changing the intake camshaft entirely. However,
none of these options are pursued in this research. As such, the intake system
pressure is limited to roughly pi /pa = 2.5.
With respect to flow similarity, not only the mass flow rate-based non-
dimensional groups ṁ1 /ṁref , ṁ2 /ṁref and ṁ1 /ṁ2 should be taken into ac-
count. The Reynolds and Mach number based on mean runner velocity and
diameter are expressed as:
Ur dr
Re = (2.4)
µ/ρ
U
Ma = √ r (2.5)
γrT

where Ur is the runner mean velocity = ṁ/ ρπd2r /4 [m/s], dr is the runner
hydraulic diameter [m] and µ is the dynamic viscosity (Pa·s). Assuming the
exhaust manifold pressure equals atmospheric pressure, the density ρ can be
written as:
12 The solid lines ( ) result from the analytical derivation; the markers ( , , ) result
from the filling-and-emptying engine model
Fired, 1800 rpm CME, 1800 rpm
6 6
5 5
4 4
Mass flow rate (-)
Mass flow rate (-)

3 3
2 2
1 1
0 M1/Mref (-) (blowdown) 0 M1/Mref (-) (blowdown)

M2/Mref (-) (displacement) M2/Mref (-) (displacement)
M1/M2 (-) M1/M2 (-)
-1 -1
0.4 0.5 0.6 0.7 0.8 0.9 1 1 1.5 2 2.5
Intake pressure pi/pa (-) Intake pressure pi/pa (-)
(a) (b)
Figure 2.11 – Mass flow rate12 versus engine load pi /pa , for (a) fired engine and
(b) CME flow rig
250 Re1 (103) (blowdown) 250 Re1 (103) (blowdown)

Re2 (103) (displacement) Re2 (103) (displacement)
200 200
Reynolds number Re (103)
Reynolds number Re (103)
150 150
100 100
50 50
0 0
-50 -50
0.4 0.5 0.6 0.7 0.8 0.9 1 1 1.5 2 2.5

(a) (b)
Figure 2.12 – Reynolds number12 versus engine load pi /pa , for (a) fired engine and
(b) CME flow rig
0.6 Ma1 (-) (blowdown) 0.6 Ma1 (-) (blowdown)

Ma2 (-) (displacement) Ma2 (-) (displacement)
0.5 0.5
Peak Mach number Ma (-)
Peak Mach number Ma (-)

0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
-0.1 -0.1
0.4 0.5 0.6 0.7 0.8 0.9 1 1 1.5 2 2.5
(a) (b)
Figure 2.13 – Mach number12 versus engine load pi /pa , for (a) fired engine and (b)
CME flow rig

3 3
S (-) S (-)
2.5 2.5
Scavenging number S (-)
2 2
1.5 1.5
1 1
0.5 0.5
0 0
0.4 0.5 0.6 0.7 0.8 0.9 1 1 1.5 2 2.5
(a) (b)
Figure 2.14 – Scavenging number12 versus engine load pi /pa , for (a) fired engine
and (b) CME flow rig

5 5
pres/pa (-) pres/pa (-)
4.5 4.5
ρres/ρa (-) ρres/ρa (-)
4 4
3.5 3.5
Residual state
Residual state
3 3
2.5 2.5
2 2
1.5 1.5
1 1
0.5 0.5
0 0
0.4 0.5 0.6 0.7 0.8 0.9 1 1 1.5 2 2.5
(a) (b)
Figure 2.15 – Residual state12 (i.e. prior to exhaust valve opening) versus engine
load pi /pa , for (a) fired engine and (b) CME flow rig

60 60
45 45
Blowdown time scale (°ca)
Blowdown time scale (°ca)
540 °ca 540 °ca
30 30
15 15
θ1 - θEO (°ca) θ1 - θEO (°ca)
0 0
0.4 0.5 0.6 0.7 0.8 0.9 1 1 1.5 2 2.5
(a) (b)
Figure 2.16 – Blowdown time scale12 versus engine load pi /pa , for (a) fired engine
and (b) CME flow rig
− γ1 γ−1 !− γ1
pi ∆Tc V0
ρ = ρi 1+ (2.6)
pa Ti Vi
The Reynolds number differs significantly between fired and CME condi-
tions, more so than the Mach number. The definition of Re and M a is given
by Eq. (2.4). The main temperature dependence is in the density (ρ ∼ 1/T )
and the dynamic viscosity (µ ∼ T , in first approximation). The ratio of specific
heats γ is a much weaker function of T . As such, the following approximate
expressions hold for Re and M a:
Ur dr
Re ∼ (2.7)
T2
Ur
Ma ∼ (2.8)
T 0 .5
Since the absolute temperature varies roughly by a factor 2 between the CME
flow rig (' 20 ◦ C) and fired engine conditions (' 800 ◦ C), Re is expected to
differ by a factor 4 and M a by a factor 1.4 between CME and fired conditions.
However, the velocity Ur should be considered also: the ratio of blowdown
to displacement flow rate is greater in a fired engine compared to the CME
flow rig. Figure 2.12 indicates that ReCM E /Ref ired ' 2.5 during blowdown
and ReCM E /Ref ired ' 10 during the displacement phase. Figure 2.13 shows
that the Mach number is comparable in CME and fired conditions. These
findings agree with the above considerations for the temperature dependence.
Figure 2.16 shows the time scale θ1 − θEO , corresponding to the crank
angle difference between exhaust valve opening (EO) and the occurrence of
the maximum mass flow rate during the blowdown phase. With reference to
App. B, θ1 − θEO equals (360/2π ) ωt1 , where N is the engine speed, and ωt1
is defined according to Eqs. (B.13) and (B.17).
2.3.3 Conclusion
Figures 2.11, 2.12, 2.13 and 2.14 summarize the analytical exhaust stroke flow
similarity. As appears from the figures, the flow conditions in the CME flow
rig are not identical to those in a fired engine. For instance, the maximum
attainable ratio of peak mass flow rates during blowdown and displacement
ṁ1 /ṁ2 is only 2.5 for the CME flow rig, compared to 6 for fired conditions. The
practical limitations are discussed in Sect. 2.3.1. The quantitative assessment
of the similarity level for an isochoric flow rig is valuable in itself, since other
researchers [7, 70] use motored engine flow rigs as well, albeit without charging
of the intake system.
It is a priori evident that a fully similar flow situation can never be ac-
complished, since a 1 : 1 scale set up is used with similar fluid (air, instead of
exhaust gas), yet at a much lower temperature (between 0 and 40 ◦ C, instead
of roughly 800 ◦ C). To guarantee flow similarity, at least the relevant dimen-
sionless numbers M a and Re should be similar in the experiment and reality.
Considering the above approximations for Re and M a in Eq. (2.7), keeping

M a constant requires reducing the characteristic velocity Ur in the experiment
to roughly Ur /1.4 . Taking into account that the geometry is not scaled (i.e. dr
is constant), this causes an increase in Re by a factor 2.8 in the experiment.
Therefore, M a and Re cannot both be kept constant in this type of set up.
In fact, due to the inevitable difference between blowdown and displacement
phase, neither can be kept constant.
Nevertheless, with regard to the flow dynamics in the manifold and espe-
cially for studying the resonance phenomenon, it is important that the fre-
quency spectrum of the ‘excitation’ generated by the flow rig on the gas in
the manifold is similar to fired engine conditions. Since the CME features a
two-stage exhaust stroke, this similarity condition is fulfilled.
The strong mean velocity fluctuations and the occurrence of local flow re-
versal in the catalyst has been observed in a number of other experimental
and numerical studies, on motored or fired engines. The resonance fluctuations
are observed in runner and catalyst velocity [4, 70, 81]. Different geometries
are investigated in these papers, yet the frequency range seems to correspond
to the assumption of a Helmholtz resonance, as suggested in this thesis (see
Sect. 5.3).
Flow reversal in runners and catalyst is measured using LDA or predicted
using transient CFD simulations by other authors [4, 70, 81, 59], both in mo-
tored and fired engine conditions. This shows that the exhaust stroke flow
similarity between CME and fired conditions is good enough, although not
identical.
2.4 Flow rate measurement

This section discusses three flow rate measurement techniques used in this the-
sis. The flow rate serves as a reference to the measured velocity distributions.
The measured mean velocity (i.e. the area integral of the velocity distribution)
should correspond to within the error margin of the flow rate measurement
(typically between 5 and 10 %). Some particular issues are addressed concern-
ing flow rate measurement in pulsating flow conditions.
Section 2.4.1 describes the orifice measurement used for the isothermal flow
rig. For reasons described below, two alternative methods are used for the
isochoric flow rig. Section 2.4.2 describes a laminar flow meter and Sect. 2.4.3
describes a cylinder pressure-based method.
2.4.1 ISO orifice

On the isothermal flow rig (see Sect. 2.2.1), the flow rate is measured using
an orifice, according to ISO Standard 5167-1991(E) [54]. The orifice diameter
is Do = 60.445 mm, and the inner pipe diameter is D = 103.285 mm (4 inch
nominal pipe size). At a pressure of 101325 Pa and a temperature of 293.15 K,
the flow rate measurement range is between 0.0206 kg/s (= 57.4 Nm3/h) and
0.421 kg/s (= 1173 Nm3/h).
2.4 Flow rate measurement 55
Based on the diameter ratio β = Do /D = 0.585, the lengths of the straight

pipe sections upstream and downstream of the orifice are 21D and 10D, re-
spectively. These lengths comply for an installation where the upstream pipe
section is connected to a single bend only. According to the standard, the mea-
surement uncertainty in case of an ideal installation varies between 4.2 % and
0.1 %. An additional uncertainty of 0.5 % is added to these values due to the
limited length of the straight pipe sections.
The standard restricts the use of the orifice to steady or slowly varying
flows. ISO Technical Report 3313-1998(E) [55] describes the error introduced
by flow pulsations on the orifice flow rate reading. The dimensionless Hodgson
number Ho defines a threshold criterion for the relative flow rate fluctuations
at the measurement device with respect to those generated by the pulsator.
Ho [-] is defined as:
Vsurge f ∆pm
Ho = (2.9)
Qm p
where Vsurge is the surge vessel volume [m3 ], f is the pulsation frequency [Hz],
∆pm is the time-averaged pressure difference across the orifice [Pa], Qm is the
time-averaged volumetric flow rate [m3/s], p is the surge vessel pressure [Pa].
For sinusoidal flow rate fluctuations, the flow rate at the pulsator can be
written as Q = Qm (1 + Q e sin ωt). In that case, the following expression relates
the Hodgson number to φ, the maximum expected additional uncertainty [-]
on the mean flow rate measurement:
Ho 0.04 Q
e
> √ (2.10)
γ φ
where γ is the ratio of specific heats [-] and Q e is the relative flow rate fluc-
tuation [-] at the pulsator. For sinusoidal pulsating flow, Eq. (2.10) gives the
minimum value for Ho, that relates to the minimum volume of the surge vessel.
If Eq. (2.10) is fulfilled, the relative error on the mean flow measurement is less
than 100 φ %.
Figure 2.17 represents the worst case encountered in the isothermal flow
rig experiments. The circular ( ) and triangular ( ) markers represent the
dimensionless flow rate at the pulsator and orifice respectively. The relative
fluctuating flow rate at the pulsator Q e = 0.368. The volume of the surge vessel
3
Vsurge = 450 dm . The Hodgson number Ho = 0.399. Based on Eq. (2.10),
the maximum uncertainty introduced by the pulsations φ equals 0.1 %.
Therefore, the surge vessel provides adequate damping of the pulsations
caused by the pulsating flow generator. The additional uncertainty is below
0.2 % for all measurement conditions. Nevertheless, the orifice reading is only
used during steady operation to check the flow rate calculated from the velocity
distribution.
The combined uncertainty on the orifice flow rate measurement is approxi-
mately 5 %. This includes (i) the uncertainty prescribed by the standard for an
ideal installation and steady flow (' 4.2 %), and (ii) the additional uncertainty
for non-ideal installation (' 0.5 %) and (iii) pulsating flow (' 0.2 %).
ISO/TR 3313 (E)

1.5
Flow rate Q/Qm (-)

1
0.5
Pulsator
Orifice
0
0 180 360 540 720
Crankshaft position ω t (ο)
Figure 2.17 – Orifice flow rate measurement in pulsating flow
2.4.2 Laminar flow element (LFE)

Several practical problems inhibit the use of an orifice for measuring the intake
system flow rate on the CME flow rig (see Sect. 2.2.2). Mainly, the CME flow
rig features much stronger flow pulsations in the intake system when compared
to the isothermal flow rig. The pulsations are aggravated by retarding the
intake camshaft. Furthermore, the available space around the set up is limited,
which makes the installation of long straight pipes upstream and downstream
of the orifice as prescribed by ISO Standard 5167-1991(E) [54] difficult.
Based on ISO Technical Report 3313-1998(E) [55], a surge vessel would be
required to damp the pulsations at the location of the orifice. The required
surge vessel volume would be larger than the 450 dm3 for the isothermal flow
rig, and it should withstand a pressure of least 3 bar. Considering these diffi-
culties, a custom built laminar flow element (LFE) meter (Fig. 2.18) is used to
measure the flow rate in the intake system of the CME flow rig.
The principle of measurement of the LFE is based on pressure drop due
to laminar flow through small parallel channels. For fully developed laminar
flow, the pressure drop is proportional to the volumetric flow rate. The main
advantage is the response time: if the flow inside the channels is incompressible
(M a < 0.3), the pressure drop responds quasi instantaneously to flow rate
variations. As such, the overall measurement system response time is solely
determined by the pressure transducer response time. This set up features a
response time of ' 1 ms.
The pressure drop due to friction is irreversibly lost. Therefore, the cross-
section of the element is usually greater than the pipe cross-section, thereby
reducing the velocity and pressure drop through the LFE.
This particular LFE consists of an unwashcoated ceramic catalyst mono-
lith ( 100 mm, 100 mm long, 900 cpsi, 2.5 mil) that is contained within the
cylindrical part 3 in Fig. 2.18. The monolith is kept in position by the shoul-
1 2a 3 2b 4
Figure 2.18 – Laminar flow element (LFE) meter
Figure 2.19 – Laminar flow element (LFE) meter, photograph of version with four
monoliths in series
ders of parts 2a and 2b. Parts 2a and 2b contain static pressure taps and a
thermocouple. The flow enters through a shallow-angle diffuser (part 1) and
exits through a sharp-angle contraction (part 4). The contraction cone angle is
sharper when compared to the diffuser to reduce the overall length of the LFE.
Figure 2.18 shows the diffuser and contraction that connect to 2 inch nominal
pipe diameter. An additional similar diffuser and contraction are available to
connect the LFE to 1 inch nominal pipe diameter, which is used on the CME
flow rig intake system.
On the CME flow rig, the LFE ( 3 in Fig. 2.9) is located downstream of the
pressure regulator ( 2 in Fig. 2.9). As such, the magnitude of the unrecoverable
pressure drop is of no real importance, since the pressure regulator guarantees
a constant pressure in the intake manifold (downstream of the LFE). A piezo-
resistive pressure transducer13 is mounted in the intake manifold, downstream
of the LFE. It measures the static pressure in the intake manifold pi .
13 Kristal 4295A, input range 0 to 3 bar, output range 0 to 10 V, combined non-linearity,
hysteresis and repeatability 6 0.35 % of the full scale reading, response time 0.2 ms.
The LFE operates at an elevated line pressure, equal to the intake manifold
pressure pi , which varies between 1.0 atm and 2.5 atm during the experiments
on the CME flow rig. For the system using a single catalyst element, a pressure
drop of 400 Pa corresponds to a flow rate of 120 m3 /h.
The measurement of the differential pressure between the static pressure
taps in parts 2a and 2b (see Fig. 2.9) forms the basis of the LFE flow rate mea-
surement. This differential pressure measurement has proven very troublesome.
Based on experience during the CME experiments, obtaining an accurate mea-
surement of the differential pressure in the order of 1 kPa at an elevated line
pressure up to 2.5 atm is not straightforward. As many as three differential
pressure transducers were used. Each had a different working principle, yet
each exhibits a certain degree of drift due to line pressure variations or ambi-
ent temperature. This problem proved very hard to overcome even with the
assistance of the sensor manufacturer.
The following differential pressure sensors were used during the measure-
ments on the CME flow rig:
• HBM DP114 — Inductive displacement sensor connected externally to

wheatstone bridge measurement amplifier HBM KWS 506. Very high
acceleration sensitivity, very high sensitivity to line pressure and ambient
temperature.
• Druck LPM 838115 — Eddy current low displacement diaphragm. Sig-

nificant sensitivity to line pressure.
• Druck PMP 417016 — Micro-machined single crystal silicon sensor.
Only the Druck PMP 4170 sensor features an acceptably low line pressure
sensitivity. The pressure sensor has been selected based on the relative insensi-
tivity to line pressure of its solid state silicon sensor. This sensor was however
only available during the final part of the measurement campaign.
Its input pressure range (-7 to 7 kPa) is higher compared to the other
sensors, since for a given line pressure sensitivity, the flow rate measurement
accuracy can be improved by increasing the LFE pressure drop. As such, three
more catalyst elements have been added in series with the first element. For
the meter using four elements, the pressure drop is approximately four times
higher compared to the single element meter. Figure 2.19 shows a photograph
of the version with four catalyst elements.
14 HBM (Hottinger Baldwin Messtechnik) DP1, input range −1 to 1 kPa, output range 0
to 10 V, maximum line pressure 100 bar, combined non-linearity, hysteresis and repeatability
not specified, response time ' 1 ms.
15 Druck LPM 8381, input range −1 to 1 kPa, output range 0 to 5 V, maximum line
pressure 100 bar, combined non-linearity, hysteresis and repeatability 6 0.25 % of the full
scale reading (= 2.5 P a), response time ' 1 ms.
16 Druck PMP 4170, input range -7 to 7 kPa, output range 0 to 2 V, maximum line pressure
70 bar, combined non-linearity, hysteresis and repeatability 6 0.08 % of the full scale reading
(= 5.6 P a), response time ' 1 ms.
LFE calibration chart
1.6 Fit
Measurements
1.4
LFE differential pressure (kPa)

1.2
0.8
0.6
0.4
0.2
0
0 20 40 60 80 100 120
Reference flow rate (m3/h)
Figure 2.20 – Laminar flow element calibration chart
The LFE has been calibrated at atmospheric pressure using the ISO stan-
dardized orifice, described in Sect. 2.4.1. Figure 2.20 presents the LFE calibra-
tion chart. The circular markers ( ) represent the calibration points, and the
solid line ( ) is the fitted calibration curve, based on the predicted relation-
ship between volumetric flow rate and pressure drop in laminar flow channels
discussed in App. A.2. This relationship takes into account the contribution to
the total pressure drop from (i) the fully developed laminar flow region, (ii) the
momentum lost in the entrance length due to the development of the laminar
boundary layer, and (iii) the entrance (i.e. contraction) and exit (i.e. expansion
losses.
The predicted relationship between differential pressure and flow rate is
least square-fitted to the reference flow rate using a single fit parameter, the
hydraulic diameter of the catalyst channels. The fitted value of df it = 0.796 mm
compares well to the geometrical value d = 0.783 mm, obtained from the
substrate density (i.e. 900 cpsi) and wall thickness (i.e. 2.5 mil).
The estimated uncertainty on the LFE flow rate reading is between 5 and
10 %.
2.4.3 Cylinder pressure

The CME flow rig features a second flow rate measurement that serves to
validate the LFE measurement described in Sect. 2.4.2. This second flow rate
measurement is based on the cylinder pressure rise during the compression
stroke. The second measurement proved necessary after the initial differential
pressure sensor used on the LFE exhibited severe drift as a function of the
intake system pressure pi and the ambient temperature.
The motion of the piston is determined by a crank-slider mechanism. The
combustion chamber volume V is given by the following expression:
πb2

s s
V (ωt) = + (1 − cos ωt) +
4 %−1 2
r !#
s 2
cr 1 − 1 − sin ωt (2.11)
2 cr
where b, s and cr are the cylinder bore [m], stroke length [m] and connecting
rod length [m] respectively, and % is the volumetric compression ratio [-]. %
is given by % = Vmax /Vmin , where Vmin is denoted the dead volume [m3 ] and
Vmax − Vmin (= πb2 s/4 ) is the swept or displaced volume per cylinder [m3 ].
The crankshaft position ωt [◦ ca] is determined by an angular encoder.
Based on the reading of a pressure sensor mounted in the combustion cham-
ber, the mass charge m [kg] per cylinder can be determined from the measured
pressure rise during the compression stroke. This calculation is fairly straight-
forward, yet assumes adiabatic compression (i.e. no heat transfer to the walls)
and the absence of blow-by leakage.
During the compression stroke, these conditions are roughly valid. The
major contribution to both blow-by leakage and heat loss is (at least in a fired
engine) expected during the expansion stroke, when cylinder gas pressure and
temperature are maximal.
Blow-by leakage inevitably occurs due to non-ideal sealing of the combustion
pressure by the piston rings. As such, a certain amount of gas escapes from the
high pressure combustion chamber to the low pressure crankcase (or oil sump).
The pressure in the crankcase corresponds to atmospheric or intake pressure,
depending on the particular installation of the crankcase ventilation.
The ideal gas law states:
pV = mrT (2.12)
where p and T are cylinder gas pressure [Pa] and temperature [K], m is the
mass of the inducted gas [kg] and r is the specific gas constant [J/(kg K)], given
by r = R/M , where R is the universal gas constant (= 8.314 J/mol K) and M
is the molecular mass of the gas [mol/kg] (e.g. for air, M = 0.02896 mol/kg,
therefore r = 287 J/(kg K)).
During the compression, the expression below gives the relationship for an
adiabatic change of states:
γ
p (ωt) V (ωt)
= (2.13)
pi Vi
where indices i denote the initial state, which corresponds to the end of the
intake stroke, and γ is the ratio of specific heats (= cp /cv ). pi and Ti are
assumed equal to the time-averaged intake system pressure and temperature.
Thus, the induction ram-effect and heat transfer during the intake stroke are
ignored.
In Eq. (2.13), p (ωt) is measured and V (ωt) is determined from the crank-
shaft position and Eq. (2.11), which means that Eq. (2.13) is overdetermined.
2.5 Data reduction 61
As such, the adiabatic assumption is relaxed to a polytropic relationship, and

the exponent γ is replaced by a polytropic coefficient n:
n
p (ωt) V (ωt)
= (2.14)
pi Vi
where the value of n is determined during the compression stroke, by averaging
the values of n obtained from the measured p (ωt) and calculated V (ωt):

p (ωt) V (ωt)
n = −log log (2.15)
pi Vi
Typically, the polytropic exponent n varies between 1.20 and 1.35, and is
only slightly smaller than the ratio of specific heats κ at the mean tempera-
ture during the compression stroke. Using the obtained averaged value of the
polytropic exponent n, the cylinder gas temperature T (ωt) is determined by
combining Eqs. (2.12) and (2.14):
n−1
T (ωt) p (ωt) n
= (2.16)
Ti pi
Based on the measured pressure p (ωt), the volume V (ωt) known from
Eq. (2.11), and the temperature T (ωt) estimated from Eq. (2.16), the mass
cylinder charge m is determined using Eq. (2.12) as follows:
p (ωt) V (ωt)
m= (2.17)
r T (ωt)
During each compression stroke, the average value of m is retained and used
to determine the mass flow rate ṁ [kg/s] as:
m
ṁ = nc (2.18)
120/N
where nc is the number of cylinders.
Based on the standard deviation of the cylinder pressure-based flow rate
measurement in consecutive engine cycles, the estimated uncertainty on the
averaged value of the mass flow rate ṁ is below 0.5 %, for a confidence level of
95 %.
Of course, this uncertainty estimate neglects the introduction of system-
atic errors e.g. by neglecting the heat transfer and blow-by leakage during the
compression stroke. Taking these approximation into account, the overall un-
certainty is of the same order of magnitude as the LFE flow rate measurement,
i.e. 10 %.
2.5 Data reduction

Identifying periodic flow patterns requires phase-locked (or conditional) sam-
pling. Two data records are phase-locked when their time bases are referenced
to a common starting point, usually defined by a once-per-period trigger signal.

Real-life periodic flows consist of a cycle-resolved and an unresolved component
resulting from (quasi) random phenomena such as turbulence, measurement
system noise, vortex shedding or unsteady flow separation. A data record
spanning a single period is also called an ensemble. Several consecutive data
records are ensemble-averaged to reduce the contribution of the unresolved
component, thus revealing the cycle-resolved flow.
Phase-locked sampling is widely used in all sorts of periodic flow studies,
typically e.g. in turbo machinery and reciprocating machines. Most of the
papers cited in the literature survey in Sect. 1.4 use phase-locked LDA or
HWA measurements.
A reference work on HWA measurement techniques by Bruun [25] contains
an extensive literature survey on conditional sampling and phase-locked aver-
aging applied to HWA and LDA. The survey is subdivided into rotating wake
phenomena (e.g. turbo machinery internal flow) and internal combustion en-
gine flow, both in-cylinder and intake system flow. However, engine exhaust
flow is not included, since only recent evolutions in the exhaust hot end have
lead to an interest in velocity field measurements in the exhaust system.
2.5.1 Ensemble averaging

Throughout this thesis, mostly two-dimensional velocity distributions are dis-
cussed in periodic flow, downstream of a close-coupled catalyst. As such, ve-
locity usually refers to the axial (i.e. along the catalyst axis) component U of
→
−
the velocity vector U .
In the following, an abbreviated subscript notation Ui,j,e = U (xi , yi , ωtj , e)
is used, where i is the measurement point index, j is the crankshaft position
index (i.e. time or sample index) and e is the ensemble or engine cycle index.
Ensemble-averaged quantities
The instantaneous local velocity Ui,j,e consists of an ensemble-averaged com-
ponent hUi,j i = hU (xi , yi , ωtj )i that represents the main periodic flow and
a fluctuating component uF F
i,j,e = u (xi , yi , ωtj , e) caused by unresolved ran-
dom phenomena, including turbulence and any other unresolved effects such as
cycle-by-cycle variations:
Ui,j,e = hUi,j i + uF
i,j,e (2.19)
where the angle brackets h···i denote ensemble-averaging, defined as:
E
1 X
hUi,j i = Ui,j,e (2.20)
E e=1
where E is the number of ensembles.

0
The ensemble-averaged fluctuation intensity uF i,j is a measure of the mag-
nitude of the fluctuating component uF
i,j,e , and is defined as:
v
u
u1 X E
0 2
uF
i,j = t uF
i,j,e
E e=1
v
u
u1 X E
2
= t (Ui,j,e − hUi,j i) (2.21)
E e=1
0
In an ensemble-averaged analysis, this uF i,j is taken as a measure of the
turbulence intensity. However, in the presence of cycle-by-cycle variation, this
measure overestimates actual turbulence, and a better approach is using a
cycle-resolved analysis, as described below in Sect. 2.5.2.
The time-averaged velocity Ui = U (xi , yi ) is defined as:
J Z 4π
1X 1
Ui,j = hUi,j i = Ui (θ) dθ (2.22)
J j=1 4π θ=0
where θ = ωt, J is the number of crankshaft positions per engine cycle (i.e.
two crankshaft revolutions = 720 ◦ ca = 4π rad). For a time-continuous mea-
surement technique such as HWA, J is determined by the sampling frequency
and the engine speed. For instance, a temporal resolution of 2 ◦ ca (J = 360)
at an engine speed of 2000 rpm requires a sampling frequency of J · N /120 =
6000 Hz.
Throughout the thesis, the overbar (···) denoting time-averaging and angle
brackets h···i denoting ensemble-averaging are often omitted for the sake of
clarity, when the nature of the quantity can be derived from the context.
The mean (or spatial averaged) velocity Um,j = Um (ωtj ) is defined as:
I
1 X
Um,j = Ui,j (xi , yi , ωtj ) Ai (xi , yi )
A i=1
Z
1
= U (ωtj ) dA (2.23)
A A
where I is the number of grid points, Ai is the cross-sectional area of mea-
surement grid cell i [m2 ] and A is the total cross-sectional area [m2 ], given by
PI
A = i=1 Ai .
The time-averaged mean velocity Um is defined as:
J
1X Q
Um = Um,j = (2.24)
J j=1 A
where Q is the volumetric flow rate through the catalyst [m3 /s]. In all velocity
distributions figures, the non-dimensional velocity U
e [-] is plotted, defined as
U = U /Um . The tilde (e) is usually omitted from the figures.
e
Uncertainty analysis
This section briefly discusses the statistical inference for ensemble-averaged
quantities.
The absolute uncertainty on Ui,j = U (xi , yi , ωtj ) can be obtained from:
stde∈[1,E] (Ui,j,e )
∆Ui,j = δUi,j · Ui,j = √ (2.25)
E
where ∆ denotes the absolute error and δ denotes the relative error on a vari-
able. The operator ‘std’ denotes the unbiased estimator for the standard devi-
ation, defined as:
v
u
u 1 X E
2
stde∈[1,E] (Ui,j,e ) = t (Ui,j,e − Ui,j ) (2.26)
E − 1 e=1
√
The factor E in the denominator of Eq. (2.25) originates from the
ensemble-averaging process. As explained in Bendat and Piersol [12], the stan-
dard deviation on the average of independent samples (here: Ui,j ) equals the
standard deviation on the samples (here:
√ stde=1...E (Ui,j,e )), divided by the
square root of the number of samples E.
It is generally assumed (see Bendat and Piersol [12]) that for ensemble-
averaging, samples from consecutive ensembles may be considered indepen-
dent. This is not the case for stationary time phenomena, where the temporal
autocorrelation should be examined to verify the independence of the samples.
2.5.2 Cycle-resolved analysis

A cycle-resolved analysis (CRA) attempts to split up the velocity fluctuation
uF C
i,j,e in a cyclic or cycle-by-cycle variation Ui,j,e and the turbulent variation
uTi,j,e :
C
Ui,j,e = hUi,j i + Ui,j,e + uTi,j,e (2.27)
The measure of the turbulence intensity is now based on uTi,j,e . This measure
is called the ensemble-averaged turbulence intensity, which is defined as:
v
u
u1 X E
0 2
uTi,j = t uTi,j,e
E e=1
v
u
u1 X E
C
2
= t Ui,j,e − hUi,j i − Ui,j,e (2.28)
E e=1
C
The key step in CRA is to determine Ui,j,e . In other words, a criterion
should be found to separate the cyclic fluctuations from the ‘turbulence’. Cyclic
variations are likely low frequency in nature, whereas high-frequency fluctua-

tions are presumed to constitute turbulence.
Cycle-resolved analysis is particularly applied to the in-cylinder flow of a
reciprocating machine. The main source of cyclic variation in a reciprocating
machine stems from minor variations in the flow pattern during the intake
stroke. The in-cylinder flow pattern is slightly different in subsequent cycles,
which is denoted cyclic variation. In fired conditions, combustion variables such
as turbulent flame speed are very sensitive to the turbulence intensity. The
presence of combustion tends to increase the level of cyclic variation. Cyclic
variation refers not only to local variables such as velocity, temperature and
gas composition but also to integral quantities such as cylinder pressure, and
the amount of charge.
Cyclic variations increase the rms of the measured quantity. Turbulence
quantities that are determined using ensemble-averaging (i.e. without CRA)
such as e.g. Eq. (2.21) overestimate the actual value.
Heywood [46] and Catania and Mittica [26] provide valuable reviews of
several CRA techniques, for measuring turbulence quantities in the combustion
chamber of internal combustion engines.
Many authors have suggested different methods of performing the sep-
aration between cyclic variation and turbulence. Lancaster [62] uses non-
stationary time averaging (NTA). Rask [89] uses cubic spline fitting (CSF).
Liou and Santavicca [68] use an inverse fast Fourier transform technique (IFT).
Tiederman et al. [98] use a low-pass filter technique, where the cut-off filter fre-
quency is determined similar to the technique by Liou and Santavicca [68].
Many more techniques exist, however each technique basically performs a low
pass filtering on the velocity to determine the cyclic variation contribution. The
cut-off filter frequency may be defined explicitly (e.g. in IFT [68]) or implicitly
(e.g. in NTA [62] and CSF [89]).
This thesis uses IFT [68], because of the appeal of its physical interpreta-
tion. The instantaneous velocity signal Ui,j,e is transformed into the frequency
domain using a fast Fourier transform (Ft ). The high frequency content is
removed above a certain cut-off filter frequency f0 . The inverse fast Fourier
transform (Ff−1 ) results in the cycle-resolved mean velocity:
C
Ui,j,e = Ff−1 (G (f0 ) · Ft (Ui,j,e − hUi,j i))
uTi,j,e C
= Ui,j,e − hUi,j i − Ui,j,e (2.29)
where the filter function G (f0 ) = 1 − H (f0 ). H (f0 ) is the Heaviside step
function, so that the filter function G = 1 for f < f0 and G = 0 for f > f0 .
The current research follows Liou and Santavicca [68] in selecting the cut-off
frequency f0 based upon the maximum frequency contained in the spectrum of
the ensemble-averaged mean velocity, or symbolically:
f0 = f max Ft (hUm (ωt)i) (2.30)

f ∈[0,∞)
Time-resolved mean turbulence intensity Time-averaged turbulence intensity TI (-)

0.05
N = 1200 rpm, Qref = 70.9 m3/h, pi = 1.55 atm N = 1200 rpm, Qref = 70.9 m3/h, pi = 1.55 atm
0.2 TIm(θ), CRA 0.045
FIm(θ), PLA
60
0.04
35
0.
0.03
04
0.02
0.0
2 5
0.0 0.035
0.15 30
25
0.025
0.0
0.03
TIm, FIm (-)
4
0.0
y (mm)
0.03
0.03
0.02
0.025
0.025
0 3 0.035
0.025
0.1 0.0
0.03
15
0.0
0.02
15
02
0.0
0.0
0.
-30
25
25.03 0.015
2
0.0 0
0.0
0.03
0.05 0.01
-60
0.005
Um = 1.563 m/s, ηm = 0.594, ηw = 0.944
0 0
0 180 360 540 720 -60 -30 0 30 60
Crankshaft angle ω t (°) x (mm)
(a) (b)
Figure 2.21 – Time-resolved mean turbulence intensity T Im (ωt) [-] (a) and time-
averaged turbulence intensity distribution T I [-], using cycle-resolved analysis accord-
ing to Liou and Santavicca [68]
In a practical application (see following section), the maximum function in

Eq. (2.30) is weakened by incorporating a small tolerance factor φ. The value
of f0 corresponds to the maximum frequency where the FFT magnitude is
greater than φ times the maximum magnitude. Figure 2.22 illustrates the
value of f0 for the example experiment discussed in the following section.
Application
This section presents an example of the time-resolved fluctuation intensity and
the turbulence intensity, determined using cycle-resolved analysis.
Figure 2.21 shows (a) the time-resolved mean turbulence intensity
T Im (ωt) [-] and (b) the time-averaged turbulence intensity distribution T I [-].
These are obtained on the CME flow rig, using a stationary hot-wire probe.
Figure 2.21a shows the time-resolved mean turbulence intensity T Im ( )
and the fluctuation intensity F Im ( ). Figure 2.21b shows the time-averaged
distribution of turbulence intensity T I. T I and F I are the non-dimensional
0 0
equivalents of the dimensional quantities uTi,j and uF
i,j , defined by Eqs. (2.28)
0 0
and (2.21) in Sect. 2.5. T I equals uTi,j /Ui,j and F I equals uF i,j /Ui,j . The
PI 0
mean quantity T Im is determined as T Im = i=1 uTi,j Ai /(Um A) .
The turbulence intensity T I is obtained using cycle-resolved analysis, ac-
cording to the inverse fast Fourier transform technique by Liou and San-
tavicca [68]. The other approaches discussed in the previous section yield
similar results. The separation of cyclic variations and turbulent variations
is essentially based on a low-pass filtering. The cyclic variation spectrum is
assumed to consist of mainly ‘low’ frequencies, whereas the turbulent spectrum
consists of ‘high’ frequencies. The approach by Liou and Santavicca [68] to
Mean velocity FFT
N = 1200 rpm, Qref = 70.9 m3/h, pi = 1.55 atm
0.3
Tp-1 f0
0.25
FFT of U(ω t) (-)

0.2
0.15
0.1
0.05
0
0 200 400 600 800
Frequency f (Hz)
Figure 2.22 – FFT of the ensemble-averaged mean velocity Um (ωt) [-] for the case
of Fig. 2.21
discern cyclic and turbulent variations is chosen because of its physical inter-
pretation in terms of the frequency spectrum of the ensemble-averaged mean
velocity. The cut-off frequency varies between 200 and 500 Hz from one exper-
iment to the other.
For this particular experiment, the cut-off frequency f0 (Eq. (2.30)) is
440 Hz. Figure 2.22 presents the FFT spectrum of the ensemble-averaged
mean velocity. Since the engine speed is 1200 rpm and the number of runners
nr = 4, the apparent pulsation frequency Tp−1 = 40 Hz.
The velocity measurement position is 25 mm downstream of the catalyst.
The flow inside the catalyst channels is laminar, since Re Recrit ' 2300. As
such, any turbulence at the measurement location is the result of the mixing of
the laminar jets issuing from the catalyst channels. Each jet expands according
to the monolith porosity ε ' 0.85. Due to the transverse velocity gradients,
a complex mixing region exists immediately downstream of the catalyst. It is
found experimentally that after 20 to 25 mm, the jets are sufficiently mixed so
that only the large-scale gradients remain.
In light of the local flow conditions, the turbulence intensity T Im evolution
corresponding to the solid line ( ) in Fig. 2.21a seems a better representa-
tion of the real turbulence intensity, compared to the dashed line ( ) (F Im ).
The mean turbulence intensity is around 2.5 %, whereas the mean fluctuation
intensity is greater than 10 %.
The difference between the dashed ( ) and solid line ( ) is a measure
for the magnitude of the cyclic variations. As shown in Fig. 2.21a, the non-
dimensional magnitude of the cyclic fluctuations are approximately 7.5 %, rel-
ative to the time-averaged mean velocity Um . Such high values are not uncom-
mon for reciprocating engines, although somewhat unexpected since the CME
flow rig operates without combustion. In a fired engine, non-linear phenomena
during the combustion phase increase the cyclic variability (see Heywood [46],
Sect. 9.4).
There is no reference turbulence measurement to verify the values obtained
from the cycle resolved analysis. By ‘tweaking’ the cut-off frequency detection
algorithm, e.g. by changing the tolerance in Eq. (2.30), the difference between
cyclic and turbulent fluctuations may be changed slightly. However, the results
are found to be quite insensitive to small parameter changes.
The flow conditions in Fig. 2.21 correspond to those in Fig. 5.15b. The peak
turbulence intensity occurs following each blowdown phase. Due to the high
velocity and pressure transients following the blowdown, it is not surprising
to find that the turbulence intensity is maximal following these events. The
post-blowdown transients also induce large-scale flow reversal, as discussed in
Sect. 5.2.
As this example demonstrates, cycle-resolved analysis (CRA) is essential for
extracting second order velocity moments or turbulence quantities in periodic
flows with considerable cyclic variation.
Cyclic variation does not directly affect the ensemble-averaged velocity (i.e.
first-order velocity moments). As such, it is of no significant importance to the
validation of the addition principle in Chap. 4, or even to the time-resolved
velocity distributions and the occurrence of flow reversal discussed in Chap. 5.
2.6 Conclusion
Section 2.2 describes two experimental flow rigs. Both rigs generate pulsating
flow in the exhaust system at ambient temperature, thus enabling the use of
any velocity measurement technique.
The isothermal flow rig (Sect. 2.2.1) is commonly used by a number of
authors [88, 58, 18, 69, 17, 16, 15, 43, 22], because of its simplicity to use.
However, the exhaust stroke flow similarity between the isothermal flow rig
and a fired engine is quite poor. The isochoric or charged motored engine
(CME) flow rig (Sect. 2.2.2) is developed to mimic the exhaust system flow
in fired engine conditions as best as possible, while still operating at ambient
temperature. The CME flow rig features blowdown and displacement phase,
typical of the exhaust stroke in a fired engine.
Section 2.3 discusses the exhaust stroke flow similarity between CME and
fired engine conditions, based on the thermodynamic analysis described in de-
tail in App. B. The CME approach and the thermodynamic derivation of the
similarity analysis constitute original contributions of this work. Parts of Sec-
tion 2.3 have been published in an international journal with review:
[84] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Experi-

mental study of flow dynamics in close-coupled catalyst manifolds.
Int. J. Engine Res. (in press).
Section 2.4 describes several reference flow rate measurements, which are
used to continuously verify the measured velocity distributions in the exhaust
2.6 Conclusion 69
system. Problems involving the accurate measurement of strong pulsating flow

rates are adequately addressed.
Section 2.5 describes the data reduction techniques typical for phase-locked
measurements of periodic phenomena. Considering the cyclic variability en-
countered in reciprocating machinery, several techniques for cycle-resolved anal-
ysis (CRA) are discussed. The CRA approach is required for obtaining unbiased
measurements of turbulence quantities.
Chapter 3
Oscillating hot-wire
anemometer (OHW)
“Measure what is measurable, and make measurable what is not so.”

Galileo Galilei (Italian physicist, ◦ 1564, †1642)
This chapter describes the construction, calibration and operation of an oscil-

lating hot-wire anemometer (OHW)17 for the measurement of one-dimensional
bidirectional velocity.
Section 3.1 reviews the literature of other available techniques for measuring
bidirectional velocity. Appendix C clarifies the selection of hot-wire anemome-
try as the most suitable measurement technique, in spite of its intrinsic insen-
sitivity to the velocity direction.
Sections 3.3, 3.4 and 3.5 present respectively the methodology of the ap-
proach, the mechanical oscillator designed for this purpose and the hot-wire
probes used during the calibration.
Section 3.6 discusses the calibration approach, which allows to use the OHW
on the pulsating flow rig to quantify the time-resolved bidirectional velocity.
The OHW is calibrated in a custom-built wind tunnel (Sect. 3.6.1). Laser
Doppler anemometry has been used as a reference velocity measurement, phase-
locked with the oscillating probe’s motion (Sect. 3.6.2).
Section 3.7 presents and interprets the calibration results. Section 3.7.3
formulates general selection and operation criteria for the OHW, based on a
17 The abbreviation ‘OHW’ is used throughout this thesis to denote the measurement tech-
nique (i.e. oscillating hot-wire anemometry) as well as the mechanical oscillator device de-
scribed in Sect. 3.4.
71
72 Chapter 3 Oscillating hot-wire anemometer (OHW)
non-dimensional scaling analysis. Section 3.7.4 discusses the calibration results

and suggests possible future improvements.
Section 3.8 describes the operation of the OHW system and the selection
of the oscillation frequency for measuring the bidirectional velocity on the iso-
choric flow rig with running engine.
3.1 Introduction:
Measuring bidirectional velocity
Obtaining high-quality experimental data that captures instantaneous flow re-
versal is not straightforward. Optical measurement techniques such as laser
Doppler anemometry (LDA) are able to measure bidirectional velocity. How-
ever, these techniques require high quality optical access and adequate seed-
ing18 in the entire measurement region. LDA-based research in CC catalyst
systems is often plagued with spatial or temporal seeding concentration defi-
ciency. This makes it very difficult to obtain a sufficiently high data rate for
measuring the time-resolved catalyst velocity distributions. Most studies using
LDA [59, 81, 70, 53] only measure the velocity in a single point or along a single
straight line in the manifold.
Hot-wire anemometry (HWA) requires neither optical access nor seeding,
although obviously, physical access for the hot-wire probe is required. HWA
features a number of advantages including high bandwidth, continuous output
signal and good spatial resolution. The main disadvantage of HWA is its in-
ability to discern flow reversal. The amount of heat convected from the wire
depends on the velocity magnitude, and not directly on its direction. When
using HWA in a flow featuring flow reversal, rectification or folding errors are
encountered (see Fig. 3.1). As such, the measured velocity is always positive
and overestimates the true velocity, and underestimates the turbulence inten-
sity. Flow reversal occurs in numerous situations, e.g. recirculating or separated
flow, highly swirling flows in combustion chambers, vortex breakdown, or pe-
riodic flow reversal in intake and exhaust piping of reciprocating machinery.
Chapter 8 in Bruun [25] presents an overview of techniques used with thermal
anemometry to resolve the flow direction ambiguity.
Firstly, a hot-wake probe uses several wires to determine the flow direction,
mostly restricted to one-dimensional or near-wall measurements. The probe
consists of a continuously heated or pulse-wise heated central wire with two
temperature sensing wires on either side, operated in constant current mode
(CCA). The velocity magnitude is determined from the time-of-flight of a small
heated amount of fluid. Handford and Bradshaw [45] compiled a review on
pulsed-wire anemometry (PWA), which has been extensively used for both
main flow and near-wall measurements. The primary disadvantage of thermal
wake anemometry is the limited bandwidth, at most around 100 Hz for PWA
18 Appendix C.2 briefly reviews the operation principle of laser Doppler anemometry, in-
cluding the seeding requirement.

3.1 Introduction: Measuring bidirectional velocity 73
Example of velocity folding errors using HWA

4
Velocity U / Um (-)
2
-1
True velocity
HWA velocity
-2
pdf(U)
Time
Figure 3.1 – Example of rectification or folding errors introduced by standard HWA

in reversing flow
probes.
Secondly, flying hot-wire anemometry (FHA) has been in use since the
1960’s. FHA consists of a one- or two-dimensional hot-wire probe that moves
−→
along a trajectory with a probe velocity Up , so that the relative velocity seen
−−→ − → − →
by the probe Urel = U − Up remains within the valid acceptance region of the
probe. For the one-dimensional situation, the probe velocity should be negative
(i.e. counter to the normal flow direction) and larger in magnitude than the
reversing flow velocity. FHA systems are categorized in terms of trajectory. An
important class uses a bean-shaped trajectory generated by a four-bar linkage
(e.g. Thompson and Whitelaw [97], Bruun [25]). Linear FHA systems have
also been used by many authors. Watmuff et al. [106] studied flow separation
and recirculation zones behind bluff bodies. Hussein et al. [52] used a linear
FHA to minimize hot-wire rectification errors in the mixing layer of a highly
turbulent jet . In each flying hot-wire system, a dead time is incorporated to
allow the flow to recover from the probe passage. These systems feature rather
complicated mechanics and are typically restricted to large-scale surroundings
such as wind tunnels.
Recently, high-frequency oscillating hot-wire (OHW) systems have been
described for use in confined spaces and near-wall measurements. Moulin et al.
[78] use a one-dimensional probe mounted onto an inline piezoelectric actuator.
The actuator oscillates the probe at frequencies up to 10 kHz and an amplitude
of a few micrometers, comparable to the wire diameter. The actuator is fixed
to a high resonance frequency support. Li and Naguib [66] use a bending
beam-type piezoelectric actuator with a resonance frequency of 110 Hz. The
actuator is used up to 490 Hz, therefore operating in higher order bending
modes. Prongs are glued to the beam. They protrude through small holes
in an acrylic cover, mounted flush with the wall. Li and Naguib [67] discuss
Figure 3.2 – OHW velocity error versus R∗ (Source: [66])
a further development of this OHW. The oscillation frequency is increased to

3 kHz. This probe measures bidirectional wall shear stress.
A high-frequency OHW differs from a traditional FHA in that the probe
−
→
velocity Up cannot be measured. The velocity direction is detected based on
the phase difference between anemometer bridge output and piezo actuator
signal. As the flow switches direction, the phase difference jumps by 180 ◦ .
The velocity magnitude is found by calibrating the probe while oscillating.
This approach assumes a frozen flow field during one oscillation period, i.e. the
oscillation period 1/fo is smaller than the smallest time scale in the flow.
Li and Naguib [66] provide guidelines for selecting the correct oscillation
parameters. Figure 3.2 shows the error of the high-frequency OHW approach
in terms of R∗ = RU Rf = u0f /u0p (ff /fo ), where u0f is the flow r.m.s. veloc-
ity, is the r.m.s. probe velocity (∝ Up,max ) and ff is the maximum frequency
contained in the flow. The high-frequency OHW measurement is valid for
R∗ < 0.5. Even if the probe oscillation velocity is smaller than typical flow
velocity fluctuations, the method remains valid by sufficiently increasing the
oscillation frequency fo . Unlike for FHA and the OHW used in this thesis, the
measurable negative velocity for a high-frequency OHW is not limited by the
maximum probe velocity, due to the phase detection technique. However, the
selection criteria proposed by Li and Naguib [66] should be kept in mind.
This thesis describes a mechanically driven low-frequency OHW for use in
quasi one-dimensional reversing flow in confined geometries, such as an au-
tomotive exhaust manifold with close-coupled catalyst. The presented OHW
concept combines aspects of traditional FHA and recent OHW systems. Sec-
tions 3.4 and 3.5 describe the OHW and hot-wire probes used in this research.
The calibration results in Sect. 3.7 indicate a maximum measurable negative
velocity of approximately −1.0 m/s.
Unlike the piezoelectric-actuated OHW systems by Moulin et al. [78] and
Li and Naguib [66, 67], the probe velocity is known for the current system.
Therefore, both velocity direction and magnitude can be determined at any
3.1 Introduction: Measuring bidirectional velocity 75
Table 3.1 – Specifications for typical FHA and OHW systems

Authors System Probe Parameters
xo fo Up,max
mm Hz m/s
Thompson and four-bar 2-D ' 130 16 −13.5
Whitelaw [97] linkage
Watmuff et al. [106] linear sled 2-D - - −4
Hussein et al. [52] linear sled 2-D - - −5
Moulin et al. [78] inline piezo 1-D 0.004 6000 −0.15
Li and Naguib [66] bending 1-D 0.160 490 −0.5
piezo
This thesis [83] mechanical 1-D 5.5 40 −1.0
oscillator
given time, as long as the relative velocity seen by the probe is positive. There
is no requirement that the oscillation period 1/fo be smaller than the smallest
time scale in the flow. On the other hand, oscillation frequency is limited to
roughly 50 Hz due to the mechanical drive. The maximum measurable negative
velocity is determined by Up,max ∼ fo,max xo . To achieve sufficient resolution in
the negative velocity range, the amplitude (xo = 5.5 mm) is quite large. This
fact combined with the lack of a dead time in between measurements, means
that the probe is measuring inside its own periodic wake. Phase-locked LDA has
been used during the calibration to examine the influence of this periodic wake
on the measurements. For the same reason, three one-dimensional hot-wire
probes have been tested: (i) a straight probe, (ii) a straight probe with extended
prongs and (iii) a probe with 90 ◦ angled prongs. Section 3.7.3 discusses the
different levels of wake contamination for each probe.
Table 3.1 gives an overview of some typical FHA and OHW systems, in-
cluding the OHW system presented in this thesis.
Oscillating a hot-wire probe may cause vibrations of prongs or sensor wire.
Wire deflection increases its length, thus altering the wire’s resistance. This
strain gauging effect occurs at high oscillation frequencies. The following ex-
pression results from an analytical derivation for the first mode resonance fre-
quency of a cylindrical wire, rigidly clamped onto the prongs without axial
tension:
s
(0) 1 λ2 E d2
fres = (3.1)
2π 4 ρ l4
where E and ρ are the material’s Young’s modulus [Pa] and density [kg/m3 ],
d and l are the diameter and length [m] of the wire. The non-dimensional
constant λ takes the value 4.730 for a double-clamped beam such as the sensor
(0)
wire. For a 5 mm thick, 1.25 mm long tungsten wire results fres = 13.2 kHz.
A similar derivation yields λ = 1.875 for the resonance frequency of a can-
tilever beam representing a single-clamped prong, shaken perpendicular to its

axis. The prongs of the OHW by Li and Naguib [66] are 25 mm long and
(0)
0.55 mm in diameter, resulting in fres = 620 Hz, quite close to the oscillation
frequency of 490 Hz.
Perry and Morrison [82] have experimentally verified the strain gauging ef-
fect by shaking a hot-wire probe in a vacuum chamber, thereby eliminating the
convective heat transfer and isolating the vibration effects in the anemometer
signal. For a 4 µm thick, 1.6 mm long platinum wire, they found that strain
gauging occurs at oscillation frequencies above 4 kHz. Equation (3.1) predicts
a resonance frequency of 3.9 kHz for such a wire.
When applying a high oscillation frequency such as proposed by Moulin
et al. [78] or Li and Naguib [66], wire rigidity should be verified. Li and
Naguib [66] perform a visual check using a microscope and strobe light, re-
vealing no wire deflection for an oscillation frequency of 490 Hz. In this study,
prong vibration is observed for one probe type in the phase-locked results (see
Sect. 3.7).
3.2 Hot-wire anemometer

Hot-wire anemometry is the primary velocity measurement technique used
throughout this thesis. Its main features and advantages are discussed in
App. C.1. The measurement principle and signal analysis are described by
Bruun [25].
Figure 3.3 shows a hot-wire sensor placed in a Wheatstone bridge. Besides
wire sensors, other elements can be used such as film or fiber sensors. RS rep-
resents the resistance of the sensor element (' 3.5 Ω for a wire probe or ' 15 Ω
for a film probe at 20 ◦ C). RL is the combined resistance of the prongs, leads,
connectors and cable (' 1.5 Ω). The sensor element is heated by the current
i1 , which is determined by the bridge top voltage Ebridge and the resistances
R1 + RL + RS and R2 + R3 of each branch of the Wheatstone bridge. The
sensor temperature TS is limited to approximately 300 ◦ C for a wire probe.
Neglecting heat radiation to the environment and conduction to the prongs,
the conservation of energy for the sensor element is:
Nu k
RS i21 = AS (TS − Ta ) (3.2)
l
where N u is the Nusselt number [-], k is the fluid thermal conductivity
[W/(m K)], l is the sensor length [m], AS is the sensor’s surface area exposed
to the flow [m2 ], Ta is the ambient temperature [◦ C]. The sensor resistance RS
and temperature TS are interrelated:
RS = RS,20 (1 + α20 (TS − 20)) (3.3)
where RS,20 is the sensor resistance and α20 is the sensor’s temperature co-
efficient of resistance (' 0.0036 K−1 for a typical wire probe), both at the
3.2 Hot-wire anemometer 77
R1 R2
i1 i2 i
e1 e2
G
E bridge
RL
R3 eoff
RS
Figure 3.3 – Constant temperature hot-wire anemometer bridge (CTA)
reference temperature of 20 ◦ C. The overheat ratio a [-] is defined as the re-

sistance ratio between operating and ambient temperature, or a = RS /RS,20 .
Typically, a = 1.8 for a wire probe.
The empirical relationship between the convective heat transfer and the
fluid velocity U [m/s] is given by King’s law (Chap. 2 in Bruun [25]):
RS i21 = (TS − Ta ) (A + B U n ) (3.4)

where A, B and n are correlation parameters determined during a calibration
with respect to a known velocity.
The left hand side of Eq. (3.4) indicates two possible modes of operation for
the hot-wire anemometer, depending on the bridge configuration. Either RS (∼
TS ) or i1 is kept constant, corresponding respectively to constant temperature
anemometry (CTA) or constant current anemometry (CCA).
The main limitation of the bandwidth of a CCA system is the wire’s ther-
mal inertia ∼ ρS cS l d2 , where ρS and cS are the wire’s density [kg/m3 ] and
specific heat capacity [J/(kg K)] and d is the wire diameter [m]. By keeping
the wire temperature constant, the thermal inertia is excluded from the system
dynamics. As such, the bandwidth of a CTA system is at least an order of mag-
nitude greater compared to CCA. CCA is used for temperature measurements,
whereas CTA is used for velocity measurements. During this thesis, only CTA
is used.
Figure 3.3 shows a basic layout of a CTA electronic circuit. To maximize
the active current i1 , the resistance of the passive branch (right hand side in
Fig. 3.3) is normally larger than the active branch that includes the sensor (left
hand side in Fig. 3.3). The bridge ratio is defined as R2 /R1 (here: R2 /R1 =
20).
Initially, the resistance R3 is set to balance the bridge for a given overheat
ratio and bridge ratio. When the bridge is in equilibrium, the voltage difference
e1 − e2 = 0. The flow conditions are determined by the variables that influence

the right hand side of Eq. (3.4), i.e. mainly the fluid velocity U and ambient
temperature Ta . As U increases, so does N u and the amount of heat convected
from the wire.
An increase in the cooling effect or the convective heat flow rate from the
wire is caused by increasing U or decreasing Ta . This decreases the sensor tem-
perature TS and therefore its resistance RS . As a result, voltage e1 is reduced
and no longer equals e2 . The voltage difference e1 − e2 < 0 is fed to the invert-
ing bridge amplifier, characterized by a large gain G. The resulting increased
output voltage Ebridge is supplied to the Wheatstone bridge top junction. The
sensor current i1 increases, which increases the sensor heat dissipation, and
consequently its temperature and resistance until the bridge is again in equilib-
rium. As such, the bridge voltage Ebridge is correlated to the convective heat
flow rate, and therefore to the fluid velocity.
Peripheral circuitry includes signal conditioning (offset voltage eof f , gain,
low-pass and high-pass filter), square wave frequency testing, a precision re-
sistance measurement, and a compensation circuit for cable capacitance and
inductance.
For this thesis, a Dantec19 StreamlineTM hot-wire anemometer is used with
90C10 constant temperature anemometer modules. The maximum bandwidth
is roughly 30 kHz for air flow at low velocity (U < 30 m/s), increasing up to
200 kHz at high velocity (U > 100 m/s). The HWA is phase-locked with the
hot-wire oscillator shaft encoder (see Sect. 3), with the engine crankshaft or
with the pulsator shaft.
The measurement principle is based on the detection of changes in the
convective heat transfer. HWA is therefore insensitive to the direction of the
flow passing the wire, only to its magnitude. As a first approximation, the
HWA output equals |U |. This phenomenon is denoted rectification or folding
in the literature.
For the same reason, HWA can only be used for flows of low to moderate
turbulence. At a turbulence intensity greater than ' 0.30, instantaneous flow
reversal occurs. In the presence of such strong fluctuations, the time-averaged
velocity measured by the HWA is overestimated due to folding, and the turbu-
lence intensity is underestimated.
Instantaneous flow reversal is known to occur in a close-coupled catalyst
manifold on a fired engine, based on the few sources available that use phase-
locked laser Doppler anemometry [70, 81, 59].
Given the exhaust stroke flow similarity between the CME flow rig and a
fired engine, a method for resolving bidirectional one-dimensional flow was pur-
sued. Oscillating hot-wire anemometry (see Sect. 3) is one such method, that
is above all compact enough for use in the confined geometry of an automotive
exhaust system.
19 Dantec Dynamics A/S, Tonsbakken 16-18, DK-2740 Skovlunde, Denmark

(http://www.dantecmt.com)
3.3 Methodology 79
probe
U >0
Up > 0
Urel > 0
xcr
arccos α
ωo t = 2πfo t xo
Figure 3.4 – OHW nomenclature
3.3 Methodology
This section presents the methodology of the OHW approach. A hot-wire probe
is oscillated back and forth in a direction parallel to the local flow direction.
The flow is assumed virtually one-dimensional (e.g. as is the case downstream
a catalyst).
The hot-wire probe is mounted as with traditional hot-wire anemometry,
facing the positive direction of flow. The positive direction and nomenclature
is indicated in Fig. 3.4.
The true instantaneous flow velocity is denoted U [m/s]. The OHW velocity
U 0 [m/s] is defined as U 0 = Urel + Up , where Urel is the relative velocity as seen
by the probe [m/s] and Up is the velocity of the moving probe [m/s]. The
probe velocity Up is determined from the oscillator device, which is presented
in Sect. 3.4 (Eq. (3.8)). The relative velocity Urel is determined from the
anemometer bridge output voltage Ebridge [V]. The relationship between the
relative velocity Urel and the bridge output voltage Ebridge is determined in
the stationary calibration. The stationary calibration is performed on a fixed
probe in the positive velocity range from 0.05 m/s to 20 m/s, using a Dantec19
type 90H02 free jet automated calibration unit (see Fig. 3.8). This constitutes
the typical calibration required for traditional hot-wire anemometry.
In reverse flow when U < 0, the OHW provides a valid measurement U 0
if the relative velocity Urel > 0, thus if the probe velocity magnitude |Up | is
sufficiently high and counter to the normal flow direction, or Up < U (< 0).
As the probe oscillates, measurements are accepted only in a window around
the maximal negative probe velocity Up ' Up,min , or symbolically when Up 6
−2πfo xo α = −ωo xo α. Approximating the probe motion as purely sinusoidal
(i.e. limxcr →∞ Up = −2πfo xo cos ωo t), this corresponds to the OHW shaft po-
sition interval − arccos α 6 ωo t + 2πn 6 arccos α, where n ∈ Z. The approach
is clarified by the graphical interpretation in Fig. 3.4.

Increasing the tolerance factor α reduces the measurement interval and
increases the interval-averaged probe velocity. α is chosen arbitrarily. During
the calibration, α is taken α = cos (π/8) ' 0.92. During the measurements,
α is taken α = cos (π/4) ' 0.71. The reason for this discrepancy is given in
Sect. 3.8, which describes the practical operation of the OHW on the isochoric
flow rig.
The data reduction uses ensemble averaging (see Sect. 2.5), denoted with
angle brackets h· · · i in Eqs. (3.5) and (3.6):
E
1 X
hU (ωo t)i = U (ωo t, e) (3.5)
E e=1
where e and E are the index and number of ensembles. In the remainder
of the thesis, the brackets denoting ensemble-averaging are mostly omitted.
Except for the discussion of phase-locked results during OHW operation (see
Sect. 3.7), each reference to a velocity denotes the ensemble-averaged velocity,
time-averaged during the interval where Up 6 −2πfo xo α = −ωo xo α. The
velocity U 0 measured by the OHW is defined as:
U0 = Urel + Up
Z ,Z
= U
hUrel (ωo t) + Up (ωo t)idt Up
dt (3.6)
p
ωo xo 6−α ωo xo 6−α
where Up is obtained from Eq. (3.8) and Urel is obtained from the anemometer
bridge voltage Ebridge , converted to velocity using the stationary calibration.
The probe oscillates in its own periodic wake, and the presence of the probe
disturbs the local flow field in a way which is not a priori known. As such, a
customized calibration is required with the aid of a reference velocity measure-
ment technique that allows to determine the influence of the moving probe on
the local flow field. Without such a calibration, no analysis of the accuracy,
dynamic range or signal-to-noise ratio can be performed. The calibration is
described in Sects. 3.6 and 3.7.
3.4 Oscillator
The oscillator uses a slider-crank mechanism to oscillate a hot-wire probe. The
amplitude of oscillation xo is 5.5 mm. For the calibration, a second amplitude
of 2.85 mm is used. The hot-wire probe fits into a probe holder, guided near
its free end by a brass bushing at the end of a rigid support tube (Figs. 3.5
and 3.6). The brass bushing provides tight support close to the probe tip,
avoiding transverse vibrations. The probe holder is clamped to the oscillating
probe holder base, sliding in Teflon guides.
The crankshaft with angular encoder is driven by an electric 70 W DC-
motor. The oscillator shaft speed is read by an optical encoder. The current to
3.4 Oscillator 81
probe probe holder oscillating probe

holder base
U >0
rigid support
Up > 0
Urel > 0 tube
optical
encoder
50 mm
dual balance
shafts
motor
Figure 3.5 – Hot-wire oscillator drawing
the DC-motor is supplied by an amplifier. The input to the amplifier is obtained

from a PI (proportional integrative) feedback controller, that regulates the shaft
rotational speed ωo = 2πfo to within 0.1 rad/s. The OHW speed controller is
implemented on a dSPACE20 DS1103 real-time data-acquisition system.
Ball bearings are used on all shafts and the crank-connecting rod joint. Two
counter-rotating balance shafts are driven using steel gears by the crankshaft.
The balance masses cancel the oscillating mass (i.e. probe holder base, probe
holder and probe). During the calibration, the oscillator is mounted on the cal-
ibration wind tunnel frame (see Sect. 3.6). The maximum oscillation frequency
fo,max = 50 Hz for xo = 5.5 mm (and fo,max = 80 Hz for xo = 2.85 mm), which
is limited by the DC-motor output.
Figure 3.4 shows the symbol nomenclature and sign convention for the OHW
motion. The probe displacement xp and probe velocity Up are:
20 dSPACE GmbH, Technologiepark 25, D-33100 Paderborn, Germany

(http://www.dspace.com)
Figure 3.6 – Hot-wire oscillator

3.5 Hot-wire probes 83
q
sin ωo t 2
xp (t) = −xo sin ωo t − (xcr /xo ) − cos2 ωo t
| sin ωo t|
q
2
− (xcr /xo ) − 1 (3.7)
 
1 sin ω o t sin 2ω o t
Up (t) = −2πfo xo cos ωo t − q  (3.8)
2 | sin ωo t| 2 2
(xcr /xo ) − cos ωo t
| {z }
Up,min
where xo is the crank amplitude (= 5.5 or 2.85 mm), xcr is the connecting
rod length (= 50 mm) and ωo t = 2πfo t is the crankshaft position [rad]. The
minimum velocity Up,min = −2πfo xo occurs at ωo t = 2πn.
Since xcr /xo 1, the motion is nearly sinusoidal. However, due to uneven
friction and backlash in the mechanism, the true probe velocity Up,true differs
slightly from Eq. (3.8). Up,true and Vp,true are measured using laser Doppler
anemometry, by focussing the measuring volume on the blackened side of a
hot-wire probe. The same device is used as reference velocity measurement
during the calibration. The deviation of Up,true and Vp,true (i.e. the transverse
component) are around 1.5 % for the highest oscillation frequency. As such, Vp
is considered negligible and Eq. (3.8) is used to determine Up .
3.5 Hot-wire probes

Figure 3.7 shows the three one-dimensional hot-wire probes used during the
calibration of the OHW: (i) 55P11 and (ii) 55P11L are straight probes with
different prong lengths, and (iii) 55P14 is a probe with 90 ◦ bend prongs. The
55P11 and 55P14 probes are manufactured by Dantec19 . The 55P11L probe is
manufactured in the TME lab21 .
Only one-dimensional probes are used, since the OHW is designed for one-
dimensional reversing flow. Table 3.2 lists the probe dimensions and resonance
frequencies. In determining the prong resonance frequency for probe 55P14,
Eq. (3.1) is used with a value of λ = 1.399 instead of λ = 1.875. This value
results from a derivation that considers a single-clamped beam shaped like
55P14’s prongs.
As indicated in Fig. 3.7, the base of the prongs is covered by an insulating
polymer, that slightly extends from the probe body. The difference between h
and h0 in Fig. 3.7 is roughly 1 mm.
The sensor wire (5 µm thick Platinum-plated tungsten) is spot-welded to
the flattened ends of the prongs. The stainless steel prongs have a circular
cross-section, and slightly tapered toward the end. The flattened end measures
approximately 100 µm in diameter. All probes can be repaired in the TME
21 Katholieke Universiteit Leuven, department of Mechanical Engineering, division TME,
Celestijnenlaan 300A, B-3001 Leuven, Belgium (http://www.mech.kuleuven.be/tme)

l, Ød
l, Ød
l, Ød s
h'
h
h'
h
h'
ØD ØD ØD
(55P11) (55P11L) (55P14)
Figure 3.7 – Hot-wire probes 55P11, 55P11L and 55P14, used for the OHW cali-
bration
3.6 Calibration approach 85
Table 3.2 – Specifications for hot-wire probes

Probe Dimensions Resonance
according to Fig. 3.7 frequency
D d l h s wire prongs
mm µm mm mm mm kHz kHz
55P11 1.9 5 1.25 5.0 - 13.2 -
55P11L 1.9 5 1.50 8.0 - 9.1 -
55P14 1.9 5 1.25 3.0 5.5 13.2 5.2
lab21 , using a custom-made spot welding power generator, a micromanipulator

and a stereo microscope.
3.6 Calibration approach

The term calibration refers to the relation between actual flow velocity U and
the velocity determined by the OHW U 0 = Urel + Up . The OHW is calibrated
for U between −1.5 and 10 m/s. Note the distinction with the stationary
calibration (see Sect. 3.3), referring to the relation between the relative velocity
Urel and the anemometer bridge output voltage Ebridge .
For the negative velocity range (−1.5 6 U < 0 m/s), the wind tunnel
described below is used, in combination with laser Doppler anemometry as
reference velocity measurement.
For the positive velocity range (0 < U 6 10 m/s), the wire is oscillated in
a Dantec19 90H02 free jet calibration unit. Figure 3.8 schematically depicts
this system. A free air jet is formed with highly uniform transverse velocity
profile by means of a settling chamber and contraction nozzle. The reference
velocity U is determined from the isentropic expansion equation, in terms of the
pressure ratio p0 /pambient and temperature T0 . The accuracy on the reference
velocity is 1 %. U 0 does not deviate from U by more than 5 % for the positive
velocity range. LDA measurements are not possible when using the Dantec19
calibration unit, thus no phase-locked results are discussed for U > 0 m/s.
3.6.1 Calibration wind tunnel

Figure 3.9 shows the wind tunnel used for calibrating the OHW in the negative
velocity range. The test section velocity is adjustable between 0 to −10 m/s,
following the coordinate convention defined in Fig. 3.9, using a speed-controlled
suction fan with throttle. The fan is vibration isolated from the wind tunnel.
Wind tunnel and OHW are mounted on a rigid frame. The acrylic test section
has a square cross-section (100×100 mm). The presence of the OHW prevented
the use of a proper settling chamber and contraction, resulting in a radial
flow entry into the test section. A bell-mouth inlet seamlessly connects to
the test section. A settling chamber with perforated screens provides uniform
flow towards the bell-mouth. The OHW probe support tube enters the inlet
p0 , T0 pa
filtered flow settling probe in

air 8 bar controller chamber free jet
Figure 3.8 – Free jet calibrator
LDA lens perforated screens
inlet settling chamber
honeycomb microscope
glass insert
OHW oscillator
bell-mouth (mounted on
U <0
inlet rigid frame)
test section
y, V 50 mm
x, U
Figure 3.9 – OHW calibration wind tunnel

chamber through a sealed hole and is located along the centreline of bell-mouth
and test section.
The hot-wire probe is located 90 mm inside the test section. The calibration
procedure described in Sect. 3.6 requires that the test section is easily removed
from the inlet box. High-quality optical access for the LDA is provided by a
microscope glass insert. Only one window is provided, since the LDA is op-
erated in backward scattering mode. Operation in forward scattering mode
could easily increase the data rate by a factor 10 to 100, and only requires an
additional window on the other side of the test section. Nevertheless, backscat-
tering is preferred because the LDA is also used to measure the longitudinal
and transverse components of the actual probe velocity Up,true and Vp,true .
The sensor wire is oriented along the vertical y-axis. The wire orienta-
tion is important for low velocity behavior since the convective heat trans-
fer differs for a vertical and horizontal wire. Bruun [25] mentions a critical
Reynolds number below which natural convection becomes significant. The
critical Reynolds number Recrit = Gr1/3 . The Grashof number Gr is defined
as Gr = g β d3 (Tw − Ta ) ν 2 , where g is the gravitational acceleration (' 9.81
m/s2 ), β is the thermal expansion coefficient (K−1 ), d is the sensor wire diam-
eter (= 5 µm), Tw and Ta are the wire and ambient temperature (K) and ν is
the kinematic viscosity [m2 /s]. For an ideal gas, the thermal expansion coeffi-
cient β = T −1 . Here, Gr ' 3.8 · 10−6 , yielding Recrit ' 0.016. No influence is
expected since Re varies between 0.51 and 0.08 for −1.50 6 U 6 −0.25 m/s.
Contrary to what Fig. 3.9 depicts, the LDA beams centreline is along the
horizontal z-axis, so that U and V can be measured. In the coordinate sys-
tem in Fig. 3.9, x = 0 corresponds to the center of the OHW oscillation range
and (y, z) = (0, 0) to the center of the wire. The LDA measurement volume
is focussed in the origin (0, 0, 0). As such, the wire passes through the sta-
tionary LDA volume. This shows up in the phase-locked LDA velocity (see
Sect. 3.7). The LDA reading during the passing of the wire is neglected in the
post-processing.
The LDA measurement volume approximates an ellipsoid with long axis
2.2 mm along the z-axis and circular cross-section of diameter 0.165 mm in
the xy-plane.
An axisymmetric boundary layer forms around the probe support tube.
Figure 3.10 shows the spanwise distribution of U (y) and turbulence intensity
u/U [-]. Note that all references to U refer to U (x = 0, y = 0). In the free
stream, turbulence intensity is below 0.8 % and the flow non-uniformity is below
2 % of free stream velocity U∞ . The flow becomes turbulent when |U | > 0.75
m/s, corresponding to Re > 2000, based on the OHW support tube diameter
and the free stream velocity. The turbulent wake features a turbulence intensity
between 5 to 20 %.
Figure 3.11 presents the power spectral density of U for laminar and tur-
bulent flow in the wind tunnel. Spectra are shown (i) in the wake center at
y = 0 mm, (ii) at the location of the 55P14 sensor wire which corresponds to
the edge of the central wake at y = s = 5.5 mm and (iii) in the free stream
at y = 20 mm. The laminar flow spectrum shows evidence of laminar periodic
U(0) = 0.37 m/s (light), U(0) = 1.05 m/s (bold)
1
0.9
U(y)/U∞ and (u/U)(y) 0.8
0.7
y = s (= 5.5 mm)
0.6
0.5
0.4
0.3
0.2
U(y)/U∞ (-)
0.1 (u/U)(y) (-)
0
0 10 20 30 40 50
y (mm)
Figure 3.10 – Wind tunnel velocity ( ) and turbulence intensity ( ) profile for
U = 0.37 m/s (laminar) ( ) and U = 1.05 m/s (turbulent) ( )
-3
U(0) = 0.37 m/s (laminar) -3
U(0) = 1.05 m/s (turbulent)
10 10
y = 0 mm y = 0 mm
-4 y = 5.5 mm -4 y = 5.5 mm
10 y = 20 mm 10 y = 20 mm
Energy spectrum ((m/s)2/Hz)
Energy spectrum ((m/s)2/Hz)
-5 -5
10 10
-6 -6
10 10
-7 -7
10 10
-8 -8
10 10
-9 -9
10 1 2 3
10 1 2 3
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)
(a) (b)
Figure 3.11 – Wind tunnel velocity frequency spectrum for (a) U = 0.37 m/s
(laminar) and (b) U = 1.05 m/s (turbulent), at different spanwise locations
vortex shedding, corresponding to a Strouhal number Sr = f D/U∞ between

0.20 and 0.25, where D equals the diameter of the OHW support tube (=
10 mm), the probe holder (= 4 mm) or the probe body (= 1.9 mm). The
range of 0.20 < Sr < 0.25 corresponds to typical frequencies of periodic vortex
shedding behind bluff bodies.
3.6.2 Laser Doppler anemometer

as reference velocity measurement
The LDA serves as reference velocity measurement during the OHW calibra-
tion in the negative velocity range. The LDA qualifies as a reference velocity
measurement for this application because of the following features:
• Direct measurement — LDA measures the velocity of particles (moving

with the flow) directly, requiring no calibration of its own. Of course, this
assumes optimal operating conditions (e.g. correct setting of validation
and bandwidth, appropriate seeding type and density, correct optical
alignment, . . . )
• High bandwidth — The maximum OHW oscillation frequency fo,max is

80 Hz. As such, a bandwidth of at least 1 kHz is required, which is easily
attained using LDA.
• Non-intrusiveness and directional sensitivity
More details on this measurement technique are given in App. C.2.

The two-component LDA is operated in backward scattering mode, with
Bragg cell frequency shifting on both velocity components, and a Dantec19
58N20 Flow Velocity AnalyzerTM to process the raw signals into velocity values.
Diethylhexylsebacat (DEHS) seeding is introduced into the settling chamber,
taking care not to disturb the flow. Like the HWA, the LDA is phase-locked
with the oscillator shaft encoder.
3.6.3 Calibration procedure

Using laser Doppler anemometry as a reference velocity measurement has one
important drawback: LDA requires that seeding particles are introduced into
the flow (see App. C.2). DEHS is the most practical seeding substance for
this application. However, when a hot-wire is used in a flow contaminated
with DEHS seeding, liquid gradually deposits onto the sensor wire and prongs,
altering its temperature-resistance characteristics. This undesirable effect is
denoted fouling. Fouling increases sensor resistance by 2 % or more, which
yields a velocity error of 10 to 20 % below 5 m/s. Since fouling also introduces
a strong time dependence, it was decided not to use simultaneous HWA and
LDA measurements. Rather, two identical probes are used for each calibration:
(i) a dummy probe during the LDA measurements, performed in DEHS seeded
air flow, and (ii) a reference probe during the HWA measurements, performed
in clean air flow. The calibration procedure involves changing probes in each
set point. For each reference velocity setting, the following actions take place:
(1)
(a) With a stationary dummy probe, the test section velocity Ustat is adjusted
using the LDA measurement.
(b) While the dummy probe is oscillated at each frequency fo , the phase-
locked test section velocity U (ωo t) is measured using LDA.
(2)
(c) With a stationary dummy probe, the test section velocity Ustat is mea-
sured a second time, after which the seeding is turned off.
(d) The test section is removed to replace the dummy with the reference
probe.
(e) While the reference probe is oscillated at each frequency fo , the phase-
locked OHW velocity U 0 (ωo t) = Urel (ωo t) + Up (ωo t) is measured.
(f) The test section is removed to replace the reference with the dummy
probe.
(3)
(g) With a stationary dummy probe, the test section velocity Ustat is mea-
sured a third time.
The fact that LDA and OHW measurements cannot occur simultaneously
is an obvious weakness of the calibration method. The repeatability of the
above procedure is verified by comparing the three reference measurements at
(1) (2) (3)
stationary probe Ustat , Ustat and Ustat during each velocity setting. During a
test incorporating 17 velocity settings, the deviation on Ustat is below 1.5 %.
3.7 Calibration results

3.7.1 Calibration charts
Figure 3.12 presents the calibration charts of OHW velocity22 U 0 versus the
reference velocity U , expressed in non-dimensional form with the maximum
probe velocity Up,max = ωo xo as scaling factor. For each probe type (see
Sect. 3.5), two calibration charts are shown for both oscillation amplitudes
xo = 5.5 mm and xo = 2.85 mm. Each chart contains points corresponding
to different oscillation frequencies. The estimated uncertainty on each point is
below 1 %.
The measurements in the positive velocity range U > 0 m/s and for a
stationary probe (f = 0 Hz) are omitted from Fig. 3.12. For positive velocity,
U 0 deviates from U by less than 5 %. For a stationary probe (f = 0 Hz), the
curves U 0 versus U are asymmetrical about the y-axis, due to the wake formed
by probe body and support in case of negative velocity.
22 The term ‘velocity’ denotes ensemble-averaged velocity, time-averaged during U
p 6
−ωo xo α, as defined in Eq. (3.6)
3.7 Calibration results 91
55P11, xo = 5.5 mm 55P11, xo = 2.85 mm

3 3
10 Hz 20 Hz
2.5 20 Hz 2.5 40 Hz
30 Hz 60 Hz
40 Hz
2 2
OHW velocity U/(ωoxo) (-)

1.5 1.5
1 1
0.5 0.5
0 0
-0.5 -0.5
a = 0.788, b = 0.5 (R2 = 0.813) a = 0.862, b = 0.5 (R2 = 0.921)
-1 -1
-4 -3 -2 -1 0 -4 -3 -2 -1 0
Reference velocity Uo/(ωoxo) (-) Reference velocity Uo/(ωoxo) (-)
(a) (b)
55P11L, xo = 5.5 mm 55P11L, xo = 2.85 mm
3 3
10 Hz 20 Hz
2.5 20 Hz 2.5 40 Hz
30 Hz 60 Hz
40 Hz
2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 -0.5
a = 0.736, b = 0.5 (R2 = 0.949) a = 0.894, b = 0.5 (R2 = 0.903)
-1 -1
-4 -3 -2 -1 0 -4 -3 -2 -1 0
(c) (d)
55P14, xo = 5.5 mm 55P14, xo = 2.85 mm
3 3
10 Hz 20 Hz
2.5 20 Hz 2.5 40 Hz
30 Hz 60 Hz
40 Hz
2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 -0.5
a = 1.145, b = 0.5 (R2 = 0.940) a = 1.274, b = 0.5 (R2 = 0.976)
-1 -1
-4 -3 -2 -1 0 -4 -3 -2 -1 0
(e) (f)
Figure 3.12 – Calibration chart for (a) 55P11, xo = 5.5 mm, (b) 55P11, xo =
2.85 mm, (c) 55P11L, xo = 5.5 mm, (d) 55P11L, xo = 2.85 mm, (e) 55P14, xo =
5.5 mm, (f) 55P14, xo = 2.85 mm (Note: reference velocity U is measured using LDA, OHW
velocity U 0 is obtained using Eq. (3.6))
55P11, xo = 5.5 mm, U = -0.45 m/s, fo = 30 Hz 55P11, xo = 2.85 mm, U = -0.45 m/s, fo = 60 Hz
U (ωot) U (ωot)
3 3
U’ (ωot) U’ (ωot)
2.5 Urel (ωot) 2.5 Urel (ωot)
Phase-locked velocity (m/s)

Up (ωot) Up (ωot)
2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-180 -90 0 90 180 -180 -90 0 90 180

Oscillator shaft angle ωot (ο) Oscillator shaft angle ωot (ο)
(a) (b)
55P11L, xo = 5.5 mm, U = -0.46 m/s, fo = 30 Hz 55P11L, xo = 2.85 mm, U = -0.45 m/s, fo = 60 Hz
U (ωot) U (ωot)
3 3
Up (ωot) Up (ωot)
2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-180 -90 0 90 180 -180 -90 0 90 180

(c) (d)
55P14, xo = 5.5 mm, U = -0.48 m/s, fo = 30 Hz 55P14, xo = 2.85 mm, U = -0.42 m/s, fo = 60 Hz
U (ωot) U (ωot)
3 3
Up (ωot) Up (ωot)
2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-180 -90 0 90 180 -180 -90 0 90 180

(e) (f)
Figure 3.13 – Phase-locked velocity at U ' −0.45 m/s and ωo xo ' 1.04 m/s, for (a)
55P11, xo = 5.5 mm, fo = 30 Hz, (b) 55P11, xo = 2.85 mm, fo = 60 Hz, (c) 55P11L,
xo = 5.5 mm, fo = 30 Hz, (d) 55P11L, xo = 2.85 mm, fo = 60 Hz, (e) 55P14,
xo = 5.5 mm, fo = 30 Hz, (f) 55P14, xo = 2.85 mm, fo = 60 Hz (Note: reference
velocity U (ωo t) is measured using LDA, OHW velocity U 0 (ωo t) equals Urel (ωo t) + Up (ωo t), where
Urel is the velocity relative to the probe, obtained from the anemometer bridge reading, and Up
is the probe velocity)
Table 3.3 – OHW calibration chart fit parameters, according to Eq. (3.9)
Probe xo a b R2 Figure
mm - - - 3.12
55P11 5.5 0.788 0.5 0.813 (a)
55P11 2.85 0.862 0.5 0.921 (b)
55P11L 5.5 0.736 0.5 0.949 (c)
55P11L 2.85 0.894 0.5 0.903 (d)
55P14 5.5 1.145 0.5 0.940 (e)
55P14 2.85 1.274 0.5 0.976 (f)
The following function is used to fit U 0 /(ωo xo ) versus U /(ωo xo ) :
√
 q
2 2 2
u+a (u + 1) + b − u − 1 + b ;u<0

Fa,b (u) = (3.9)
u ;u>0

where u = U /(ωo xo ) [-], a and b are non-dimensional parameters: a determines

the slope (= 1 − 2a) of the function as u → ∞ and b > 0 yields a smooth
transition around U /(ωo xo ) = −1. When b = 0, the function reduces to
Fa,b (u) = u + a (|u + 1| − u − 1). A value of b = 0.5 is chosen arbitrarily for
all curves.
The non-dimensional calibration curves for different oscillation frequencies
collapse reasonably well. The value of a, b and the coefficient of determination
R2 [-] for each correlation fit with the function Fa,b (u) is given in each plot in
Fig. 3.12 and in Table 3.3.
For −1 6 U /(ωo xo ) 6 0, the calibration curve should ideally be U 0 /(ωo xo )
= U /(ωo xo ) . This region is of highest interest to the calibration, since only
here exists an unambiguous relation between U 0 /(ωo xo ) and U /(ωo xo ) .
The behavior is quite different for straight and angled probe types at high
and low amplitude. This unexpected difference is investigated below, using a
non-dimensional scaling analysis.
3.7.2 Phase-locked results

Figure 3.13 shows the phase-locked reference (U ), probe (Up ), wire-relative
(Urel ) and OHW (U 0 = Urel + Up ) velocity. All plots represent different probes
and different amplitudes, for identical values of U and ωo xo . The number of
ensembles taken is sufficiently high to reduce the estimated uncertainty on all
ensemble-averaged velocities to below 1 %.
Vertical dash-dotted lines indicate where Up 6 −ωo xo α. As indicated in
Table 3.2, the 55P11 prong length h is only 5 mm, which is smaller than the
greatest oscillation amplitude (5.5 mm). This causes occasional collapse of the
LDA velocity U to the probe velocity Up , when the probe body passes through
the LDA measurement volume. Due to reflections as the wire passes through
the LDA measurement volume at sin ωo t = 0 (ωo t = πn where n ∈ Z), the

LDA velocity around ωo t = πn is unreliable (e.g. Fig. 3.13c). These erroneous
measurements are excluded from the time averaging in Eq. (3.6).
Figure 3.13e shows high-frequency jitter in the HWA signal (Urel and U 0 )
around ωo t ' π + 2πn. Since the jitter shows up after ensemble averaging, it
is phase-locked with the probe motion. The peak jitter frequency is 4.3 kHz
and is independent of changes in oscillation frequency. The phenomenon is
likely caused by prong vibration. The prong resonance frequency of 5.2 kHz
listed in Table 3.2 is apparently somewhat overestimated. The vibrating prongs
strain the sensor wire, which changes its resistance, thereby affecting the ve-
locity reading. This is only encountered for the 55P14 probe at high oscillation
frequency (f > 50 Hz).
Figure 3.13 provides insight in the periodic flow around the oscillating
probe. Since the LDA velocity U is measured at a stationary point (x, y, z) =
(0, 0, 0) and the OHW velocity U 0 is measured along the moving wire, LDA
and OHW velocity should only be compared at sin ωo t ' 0 or ωo t ' πn. While
0 6 ωo t < π/2 , the LDA velocity magnitude increases due to the momentum
provided by the passing wire and nearby prongs. While π/2 6 ωo t < π, the
probe changes direction and passes through the origin in positive direction,
causing a decrease in the LDA velocity magnitude. While −π 6 ωo t < 0,
the flow around the origin returns to the initial state. The characteristic time
constant for this recovery is simply the convective time constant xo /U .
3.7.3 Non-dimensional scaling analysis

Figure 3.12 shows a discrepancy in the behavior for −1 6 U /(ωo xo ) 6 0.
The calibration points collapse reasonably to U 0 /(ωo xo ) = U /(ωo xo ) for the
55P11 and 55P11L probes at high amplitude, and for the 55P14 probe at low
amplitude. In all other cases, U 0 /(ωo xo ) deviates from U /(ωo xo ) . For each
calibration chart in Fig. 3.12, the influence of the oscillation frequency vanishes
when U 0 and U are scaled with ωo xo . The discrepancy should therefore be
influenced by xo or other geometrical parameters. To quantify the discrepancy,
the velocity deviation ∆U is defined as the difference between the OHW velocity
and the expected velocity, which is given by Eq. (3.9) with b = 0:
U0

∆U U
= − Fa,b (3.10)
ωo xo ωo xo ωo xo
For each probe, a suitable non-dimensional group is selected that corre-
lates the velocity deviation ∆U /(ωo xo ) for both amplitudes in the calibration
region of interest (i.e. −1 6 U /(ωo xo ) 6 0). For the angled 55P14 probe,

Ubangled = (U /(ωo xo ) ) ωo x2o /ν is the suitable non-dimensional group. For

the straight 55P11 and 55P11L probes, U bstraight = (U /(ωo xo ) ) ωo D2 /ν is
the suitable non-dimensional group, where D represents the probe body di-
ameter [m]. The correlations between velocity deviation ∆U /(ωo xo ) and non-
dimensional groups U bstraight and Ubangled are presented in Fig. 3.14. For each
55P11
1
0.8
OHW velocity deviation ∆U/(ωoxo) (-)

0.6
0.4
0.2
-0.2
-0.4 10 Hz, 5.5 mm

20 Hz, 5.5 mm
-0.6 30 Hz, 5.5 mm
20 Hz, 2.85 mm
40 Hz, 2.85 mm
-0.8 60 Hz, 2.85 mm
-1
-120 -100 -80 -60 -40 -20 0
Uo /(ωoxo) . (ωo D2/ν) (-)
(a)
55P11L
1
0.8
0.6
0.4
0.2
-0.2
-0.4 10 Hz, 5.5 mm

20 Hz, 5.5 mm
-0.6 30 Hz, 5.5 mm
20 Hz, 2.85 mm
40 Hz, 2.85 mm
-0.8 60 Hz, 2.85 mm
-1
-120 -100 -80 -60 -40 -20 0
Uo /(ωoxo) . (ωo D2/ν) (-)
(b)
55P14
1
0.8
0.6
0.4
0.2
-0.2
-0.4 10 Hz, 5.5 mm

20 Hz, 5.5 mm
-0.6 30 Hz, 5.5 mm
20 Hz, 2.85 mm
40 Hz, 2.85 mm
-0.8 60 Hz, 2.85 mm
-1
-600 -500 -400 -300 -200 -100 0
Uo /(ωoxo) . (ωo x2o /ν) (-)
(c)
Figure 3.14 – Correlation of velocity deviation ∆U for (a) 55P11 and (b) 55P11L
bstraight (Eq. (3.11)), and for (c) 55P14 versus U
versus U bangled (Eq. (3.12))
probe, ∆U /(ωo xo ) collapses roughly to a linear curve versus the appropriate

non-dimensional group for −1 6 U /(ωo xo ) 6 0:
∆U U ωo D 2
∼ =U
bstraight
ωo xo ωo xo ν
(55P11, 55P11L) m (3.11)
∆U ωo xo D D
∼
U ν xo
∆U U ωo x2o
∼ =U
bangled
ωo xo ωo xo ν
(55P14) m (3.12)
∆U ωo xo D xo
∼
U ν D
The expressions for ∆U /U in Eqs. (3.11) and (3.11) indicate that for a
constant value of ωo xo , the velocity deviation decreases for increasing oscillation
amplitude xo in case of the straight 55P11 and 55P11L probes. Equivalently,
increasing xo yields a better correspondence of U 0 /(ωo xo ) with U /(ωo xo ) for
−1 6 U /(ωo xo ) 6 0. By contrast, for the angled 55P14 probe, the velocity
deviation decreases for decreasing (instead of increasing) amplitude xo .
When the amplitude xo → ∞ for the straight probes, the oscillating motion
tends towards a slow moving linear sled. The amplitude xo = 5.5 mm is a
compromise between compactness and a small velocity deviation ∆U , which
yields a high negative velocity resolution.
For the angled probe, the amplitude xo should rather be decreased, requiring
a higher oscillation frequency to maintain the same value of ωo xo . In this case,
the problem of prong vibration shown in Fig. 3.13f has to be considered. Prong
vibration can be avoided by raising the prong resonance frequency, either by
thickening the prongs or by reducing the prong cantilever length s (Fig. 3.7,
right). Reducing s will likely cause the angled probe’s behavior of ∆U to
resemble that of the straight probes. In other words, the decrease in velocity
deviation ∆U by decreasing xo (and increasing the oscillation frequency) would
be counteracted by reducing the angled prong cantilever length s.
No attempt is made in this thesis to verify the above assumption by de-
creasing the amplitude xo for the angled probe below 2.85 mm. Decreasing
xo to 0.35 mm and increasing the oscillation frequency to 490 Hz results in a
system similar to Li and Naguib [66]. However, such high oscillation frequency
is unattainable with a mechanical drive. Furthermore, severe prong vibration
would occur, unless appropriate measures are taken to raise the prong resonance
frequency. Moulin et al. [78] use a straight probe at low oscillation amplitude
and very high frequency. This choice seems in contradiction with the results
of this study, however extrapolation is difficult. Firstly, the amplitude used by
Moulin et al. [78] (= 4 µm) is of the order of the sensor wire diameter (= 5 µm).
Secondly, as discussed in Sect. 3.1, the high-frequency OHW systems [78, 66]
use a different approach for detecting the flow direction than the low-frequency
OHW system presented in this thesis. Moulin et al. [78] and Li and Naguib [66]
use the phase difference between the anemometer output and the probe motion
signal to detect flow direction.
3.7.4 Discussion of the calibration results

The different behavior of straight and angled probes discussed in Sect. 3.7.3 is
likely due to the local flow around the sensor wire. For the angled 55P14, the
wire oscillates on the edge of the boundary layer formed around probe body
and support (Fig. 3.10). For the straight 55P11 and 55P11L, the wire oscillates
in the wake center.
The effect that the periodic wake caused by the moving probe has on the
calibration is difficult to estimate. It is a clear weakness for the low-frequency
OHW approach. Traditional FHA avoids this problem by incorporating a dead
time in the motion, where the probe remains stationary and located sufficiently
far from the measurement field. However, this seems impossible to achieve in
a confined geometry. Phase-locked LDA has been used in the present study to
measure the time-resolved local flow during wire oscillation.
Wake contamination could be minimized by using a modified version of the
angled 55P14 probe, with longer prongs (i.e. increasing s from Table 3.2). As
Fig. 3.10 shows, an increase in s by only a few millimeters places the wire in the
free stream. Along with the proposed reduction in the oscillation amplitude
xo that follows from the non-dimensional scaling in Sect. 3.7.3, this might be
subject for further research.
Simultaneously increasing the prong length and oscillation frequency re-
quires a sufficient increase in the resonance frequency to minimize prong vibra-
tion. By appropriately shaping the prongs, their stiffness could be increased.
Rather than merely thickening the prongs, perhaps a slender airfoil could be
used to minimize flow disturbance.
The OHW affects not only the local flow field. A periodic thermal wake is
created by the moving hot-wire, subject to convection and diffusion. As the wire
moves through a region heated during a previous passage, the convective heat
transferred from the wire decreases. This effect is equally present in the high-
frequency OHW by Moulin et al. [78] and Li and Naguib [66], perhaps more
so due to the small oscillation amplitude. The effect is difficult to estimate,
however it is completely contained within the calibration. No OHW system
can be used without calibration with oscillating probe.
In summary, the combination of a straight hot-wire probe with long prongs
(55P11L) with high oscillation amplitude seems the best option for the pre-
sented low-frequency OHW. This combination yields the lowest velocity de-
viation ∆U /(ωo xo ) for −1 6 U /(ωo xo ) 6 0 (Fig. 3.14b). Furthermore, this
combination results in a maximum measurable negative velocity of −1.0 m/s,
and the non-dimensional calibration chart is accurately fitted by Eq. (3.9) with
an R2 -value23 of 0.949 (Fig. 3.12c).

Further research is required to verify the feasibility of a low-frequency OHW
with minimal wake contamination, based on a modified version of the angled
55P14 probe with extended prongs.
3.8 Operation
The current section describes the selection of the OHW oscillation frequency
fo for measuring the bidirectional velocity in the periodic flow in the exhaust
system.
Based on the calibration results discussed in Sect. 3.7, the 55P11L probe
is used with an oscillation amplitude xo = 5.5 mm. The oscillation frequency
fo is kept between 30 and 40 Hz to obtain the maximum resolvable negative
velocity. The oscillation frequency fo is also kept proportional to the engine
speed N [rpm]. The non-dimensional oscillation frequency Rf is defined based
on the ratio of engine cycle period (i.e. two crankshaft revolutions) to the
oscillation period:
fo ωo
Rf = = (3.13)
N /120 ω/2
Considering the OHW methodology discussed in Sect. 3.3, the OHW yields
valid measurements when the probe velocity Up is large and counter to the
normal flow direction. In terms of the symbols and sign convention established
in Fig. 3.4, valid measurements are taken when Up 6 −2πfo xo α = −ωo xo α.
The valid measurement window defined by Up 6 −2πfo xo α = −ωo xo α cor-
responds to a time interval (i.e. OHW shaft position interval) of approximately:
− arccos α 6 ωo t + 2πn 6 arccos α (n ∈ Z) (3.14)

The approximation in Eq. (3.14) neglects the non-harmonic distortion intro-
duced by using a crank-connecting rod mechanism instead of a Scotch yoke.
With regard to Eq. (3.8), Up is approximated as limxcr →∞ (Up ) = −2πfo xo ·
cos ωo t.
Equation (3.14) shows that the valid measurement window represents a
fraction of 2 arccos α/(2π) of one oscillation period. To clarify the following
discussion, the fraction of the valid measurement window per period is defined
as 1/A :
1 2 arccos α
= (3.15)
A 2π
m
α = cos (π/A )
23 The R2 -value represents the coefficient of determination, defined as the square of the
correlation coefficient between the values and the fitted curve.

3.8 Operation 99
N = 2400 rpm, fo = 35.0 Hz, Rf = 1.75

1
Up/ωoxo (cycle 1)
Up/ωoxo (cycle 2)
Up/ωoxo (cycle 3)
Up/ωoxo (cycle 4)
Probe velocity Up/ωoxo (-)

Up = - ωoxoα
-α
-1
0 180 360 540 720
Engine crankshaft position ω t (ο)
Figure 3.15 – OHW probe velocity Up , phase-locked with the engine crankshaft
position ωt [◦ ca], for α = cos (π/4 ) ' 0.71 (as used during the measurements on the
CME flow rig)
During the OHW calibration, the non-dimensional tolerance factor α is

chosen as α = cos (π/8 ) ' 0.92. For α = cos (π/8 ), the valid measurement
window corresponds to a fraction of 1/A = 1/8 of one oscillation period.
For the measurements on the CME flow rig, the tolerance factor α is chosen
α = cos (π/4 ) ' 0.71 instead, corresponding to a fraction of 1/A = 1/4 of the
oscillation period. The reason for this arbitrary choice is given in the following
paragraphs.
When the non-dimensional OHW frequency Rf is a whole number (Rf ∈
Z), the OHW moves in synchronization with the engine’s crankshaft. In that
case, the OHW measurements are taken in the same engine crankshaft position
intervals in consecutive engine cycles. In order to cover the entire crankshaft
position range 0 6 ωt 6 720 ◦ ca corresponding to a single engine cycle, the
OHW motion slightly lags or leads the engine rotation. As such, Rf is chosen
as:
Rf = n ± 1/A (n ∈ Z) (3.16)
where 1/A corresponds to the above introduced fraction of valid measurements

per oscillation period, or 1/A = (2 arccos α)/(2π) .
Figure 3.15 shows the OHW probe velocity versus crankshaft position for
a particular engine speed. The whole number n is maximized, limited by the
maximum oscillator frequency fo,max = 40 Hz. For the case of N = 2400 rpm
and taking 1/A = 1/4 , this results in fo = 35 Hz for n = 2 or Rf = 2 − 1/4 =
1.75. In this case, it takes four engine cycles (A = 4) to obtain valid velocity
measurements for the entire range of crankshaft positions 0 6 ωt 6 720 ◦ ca.
Note from Fig. 3.15 that the choice of Rf = n ± 1/A ensures that during
each next engine crankshaft revolution, the valid OHW measurement window
shifts to the next (higher) crankshaft positions. This is an arbitrary choice; the
difference between Rf and the whole number n may be any number between 0
and 1. However, the choice according to Eq. (3.13) ensures the minimum total
measuring time.
The different selection of 1/A and α during the OHW calibration and the
measurements on the CME flow rig has the following implications. From the
calibration to the measurements, the value 1/A has been increased from 1/8
to 1/4 , thereby decreasing the tolerance factor α from cos (π/8 ) ' 0.92 to
cos (π/4 ) ' 0.71). This decreases the mean probe velocity magnitude during
a valid measurement window, thus decreasing the resolution in the negative
velocity range.
The main reason for increasing 1/A is that A engine cycles are required to
complete valid OHW measurements for one engine cycle 0 6 ωt 6 720 ◦ ca. By
increasing 1/A from 1/8 to 1/4 , only four (instead of eight) engine cycles are
required to complete measurements for one engine cycle.
The magnitude of the loss in negative velocity resolution associated with
the decrease in tolerance factor α from cos (π/8 ) ' 0.92 to cos (π/4 ) ' 0.71 is
demonstrated by Fig. 3.16. Figure 3.16 shows two non-dimensional calibration
charts for the 55P11L probe with oscillation amplitude xo = 5.5 mm. The data
reduction for Fig. 3.16a uses α = cos (π/8 ), which corresponds to the value used
during the calibration. The data reduction for Fig. 3.16b uses α = cos (π/4 ),
which corresponds to the value used during the measurements on the CME
flow rig.
The resolution in the negative velocity range slightly decreases for the α =
cos (π/4 ) case in Fig. 3.16b, compared to the α = cos (π/8 ) case in Fig. 3.16a.
This can be noted from the increased value of a, which determines the slope of
the fitted curve according to Eq. (3.9). The parameter a = 0.748 for the case
with low tolerance factor α = cos (π/4 ) (Fig. 3.16b), compared to a = 0.736
for the case with high tolerance factor α = cos (π/8 ) (Fig. 3.16a).
Several hundred ensembles are required to ensure sufficient accuracy after
ensemble-averaging the phase-locked velocity data. A single measurement run
that results in one catalyst velocity distribution takes between two to four
hours when using 1/A = 1/8 . When using 1/A = 1/4 , the slight loss in
negative velocity resolution is acceptable given the reduction by 50 % of the
measurement time.
3.9 Conclusion
A novel low-frequency oscillating hot-wire anemometer (OHW) is presented
to measure bidirectional velocity. The OHW is more compact than traditional
flying hot-wire anemometers reviewed by Bruun [25], and less prone to prong or
wire vibration and strain gauging than high-frequency OHW systems recently
described by Moulin et al. [78] and Li and Naguib [66]. Unlike high-frequency
systems, the presented OHW does not assume a frozen flow field. Therefore, the
oscillation frequency may be well below the maximum frequency contained in
3.9 Conclusion 101
55P11L, xo = 5.5 mm, α = cos(π/8) 55P11L, xo = 5.5 mm, α = cos(π/4)

3 3
10 Hz 10 Hz
2.5 20 Hz 2.5 20 Hz
30 Hz 30 Hz
40 Hz 40 Hz
2 2

1.5 1.5
1 1
0.5 0.5
0 0
-0.5 -0.5
a = 0.736, b = 0.5 (R2 = 0.949) a = 0.748, b = 0.5 (R2 = 0.947)
-1 -1
-4 -3 -2 -1 0 -4 -3 -2 -1 0
(a) (b)
Figure 3.16 – Calibration chart for 55P11L at xo = 5.5 mm, for tolerance factor
(a) α = cos (π/8 ) ' 0.92 (used during the calibration), and (b) α = cos (π/4 ) ' 0.71
(used during the measurements on the CME flow rig)
the flow. An advantage for the high-frequency approach is that the measurable
negative velocity is not limited, although Li and Naguib [66] provide a criterion
relating minimum negative velocity and oscillation frequency.
A small-scale wind tunnel is used to calibrate the OHW in the negative ve-
locity range −1.5 6 U 6 0 m/s. Laser Doppler anemometry is used as reference
velocity measurement. LDA measurements are phase-locked with the OHW.
Three hot-wire probe designs are calibrated, examining the influence of prong
length and shape. Calibrations are performed for two oscillation amplitudes
and several frequencies.
Based on the calibration charts (Fig. 3.12), the best calibration results are
obtained for the 55P11L probe with straight extended prongs, in combination
with an oscillation amplitude xo = 5.5 mm. This choice results in a maximum
resolvable negative velocity of −1.0 m/s.
The non-dimensional analysis indicates in Fig. 3.14 that straight (55P11,
55P11L) and angled (55P14) probes behave differently with regard to the cor-
respondence between the OHW velocity U 0 and the reference velocity U . For
the straight probes, increasing the oscillation amplitude xo (and decreasing
oscillation frequency fo ) reduces the deviation between U 0 and U . For the an-
gled probe, the deviation between U 0 and U is reduced by decreasing xo and
increasing fo .
The presented OHW system can be applied to reversing flows in confined
geometries, such as internal pipe flow. In particular, it has been successfully
applied during this thesis to measure the phase-locked velocity distribution
including instantaneous local flow reversal on the CME flow rig (see Chap. 5).
The contents of this chapter have been published in an international journal
with review:
[83] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Calibra-

tion of an oscillating hot-wire anemometer for bidirectional velocity

measurements. Exp. Fluids, 40(4):555–567, 2006. http://dx.doi.org/
10.1007/s00348-005-0095-4
Chapter 4
Addition principle
“The first principle is that you must not fool yourself and you are the
easiest person to fool.”
Richard Feynman (American physicist, ◦ 1918, †1988)
This chapter presents the experimental validation of the addition principle,

defined by Eq. (4.1). The relevance of this principle is recalled in Sect. 4.1.
Section 4.3 discusses the elaborate data reduction used to assess the simi-
larity between a pair of velocity distributions.
Section 4.4 presents the experimental results for both pulsating flow rigs and
both exhaust manifolds. The combined results are summarized in Sect. 4.4.3
and interpreted in Sect. 4.5. The similarity measures introduced in Sect. 4.3 es-
tablish a remarkable correlation with a single non-dimensional flow character-
istic, the scavenging number S (Eq. (4.42)). The correlations in Fig. 4.29 result
in a critical value for the scavenging number Scrit , which determines the validity
range of the addition principle.
Section 4.6 provides an physical interpretation to the findings of Sect. 4.5
in the form of an analogy to a zero-dimensional scalar mixing process. Using
this analogy, Sect. 4.6.2 introduces the hypothetical concept of a collector effi-
ciency ηD , which equals the critical value of the scavenging number Scrit . This
hypothesis warrants further research.
4.1 Introduction
This chapter discusses the experimental validation of the addition principle.
The relevance of the addition principle has already been briefly introduced in
the literature survey in Sect. 1.4.1.
103
104 Chapter 4 Addition principle
The fluid dynamics aspects of the industrial design of exhaust manifolds

comprises mostly stationary CFD studies, using commercially available CFD
packages. The increase of the computational cost associated with transient
CFD calculations (relative to stationary calculations) is very high. Within the
highly competitive automotive supplier business, achieving short development
times while maintaining design criteria guarantees is critical for obtaining new
contracts. At present, transient CFD is therefore rarely used for exhaust man-
ifold design in industrial practice.
The entire manifold design is based directly and indirectly on the internal
flow field and the catalyst velocity distribution. The stationary flow predic-
tions are used to estimate the backpressure, the catalyst flow uniformity, and
a number of other criteria such as determining the optimal position for the ex-
haust gas oxygen (or lambda) sensors so that they receive equal contributions
of exhaust gas from each cylinder. Furthermore, the stationary flow results are
used as boundary condition in finite element modeling to predict the surface
temperatures, the thermal stresses and the related risk of mechanical failure
due to fatigue.
Given the strong pulsating nature of the exhaust flow, the question arises
whether using stationary predictions is justifiable. For pulsating flow in close-
coupled catalyst manifolds, the addition principle states the following:
Addition principle The time-averaged catalyst velocity distri-

bution in pulsating flow can be predicted by a linear combination of
velocity distributions that result from stationary flow through each
of the exhaust runners, with the exhaust valves in the maximum
open position, and for equal volumetric flow rate.
Or symbolically:
nr
1 X
U
e (x, y)
puls ' U
e (x, y)
stat = U
e (x, y)
stat,r (4.1)
nr r=1
where U e puls denotes the time-averaged non-dimensional velocity distribution

in pulsating flow. U estat is the linear combination of nr stationary flow velocity
distributions U estat,r . An overbar ( ¯ ) denotes a time-averaged quantity. A tilde
( e ) represents a non-dimensional velocity, scaled with the time-averaged mean

velocity U e = U Um .
The comparison between pulsating and stationary flow conditions is per-
formed for the equal volumetric flow rate, as indicated by the definition. Fig-
ure 4.1 illustrates the importance of this aspect. The top plot in Fig. 4.1 rep-
resents the time evolution of the mean catalyst velocity for pulsating flow, for
a four-cylinder engine (r = 1 . . . 4). The time-averaged mean velocity Upuls,m
is indicated by the dashed line ( ).
The middle plot in Fig. 4.1 represents the hypothetical case where all run-
ners are blocked, except for a single one (r = 3). In this case, the overall
time-averaged velocity Upuls,i=3,m ' Upuls,m /nr (nr = 4).
4.1 Introduction 105
U
r=1 r=3 r=4 r=2
Pulsating flow
Upuls, m all runners
Upuls
t
U
Pulsating flow
Upuls, r=3, m single runner
Upuls, r=3
t
U
Stationary flow
Ustat, r=3, m single runner
Ustat, r=3
t
Tp
120 / N
Figure 4.1 – Addition principle: flow rate setting
The addition principle compares the pulsating flow distribution Upuls to

a superposition of flow distributions for stationary flow through each runner
separately. In order to obtain similar flow conditions, stationary flow rate
should equal the time-averaged flow rate in pulsating conditions. Therefore,
the stationary flow conditions are as depicted in the bottom plot of Fig. 4.1.
The preset velocity Ustat,r,m therefore equals Upuls,m .
Since ideally Ustat,r,m = Upuls,m , it is not strictly necessary in Eq. (4.1) to
non-dimensionalize each velocity distribution by its time-averaged mean value.
The reason is rather practical, since there is inevitably some uncertainty in
setting the flow rate in the experiment.
The validity of the addition principle means that the pulsating flow in the
exhaust manifold behaves quasi-stationary, i.e. that there is no influence of
dynamic effects. Indeed, the linear combination of stationary distributions
Ustat may be considered the quasi-stationary limit case for zero engine speed.
Nevertheless, the exhaust manifold of a working engine features a highly
transient, three-dimensional turbulent flow. The flow features a considerable
degree of cycle-by-cycle variability, as illustrated in Sect. 2.5.2. Furthermore,
like all fluid dynamic phenomena, the system is nonlinear, due to e.g. internal
and external friction, and compressibility effects. For example, as shown in
Fig. B.1, the discharge coefficient of the exhaust valves even varies nonlinearly
during the valve lift sequence.
In that sense, it seems a crude approximation to make that any time-varying
effects are irrelevant in this flow, and that the time-averaged pulsating flow
distribution simply equals a combination of velocity distributions obtained for
stationary flow through each runner, with the exhaust valves in the maximum
open position. Therefore, the subject of research of this chapter is to validate
the addition principle and determine its applicability range.

The validity of the principle would imply that transient CFD is not required
for designing a manifold with a CC catalyst with respect to the catalyst flow
distribution and that steady state CFD simulations suffice.
This principle carries important practical implications for an industrial ex-
haust manifold design team. The computation time for steady CFD is at least
one order of magnitude smaller when compared to transient CFD calculations.
This entails a much shorter development time, which is crucial for obtaining
contracts in the fast moving and highly competitive automotive Tier 1 sector.
4.2 Validation approach

A crucial part of the addition principle is the comparison between two spa-
tial distributions of the axial catalyst velocity, (i) one representing the time-
averaged velocity distribution in pulsating flow conditions, and (ii) one rep-
resenting a linear combination of velocity distributions obtained in stationary
flow conditions.
Each distribution is two-dimensional, and given in (x, y) coordinates. The
orthonormal coordinate system (x, y, z) is aligned so that the z-axis coincides
with the catalyst lengthwise axis and the x-axis is along the long axis in case
of an elliptic catalyst. In practice, for a typical four-cylinder engine, the x-axis
is (nearly) parallel to the engine crankshaft.
Throughout the field of engineering, a number of similarity measures are
in use for comparing two-dimensional distributions. The term ‘distribution’
can really denote any two-dimensional collection of scalar data. Much of the
research into similarity measures is concerned with image processing.
For the present application, a good similarity measure should have the fol-
lowing properties:
(a) High sensitivity to the difference in regions of extreme (i.e. high and low)
velocity, since these regions are the most critical in terms of catalyst
degradation, conversion efficiency and pressure drop.
(b) The measure should allow uncertainty analysis, thereby enabling the
quantification of the validity of the addition principle in terms of a rigor-
ous statistical hypothesis test.
(c) Low sensitivity to small spatial shifts, of the order of magnitude of

0.5 mm, due to errors in positioning the velocity probe.
In the image processing literature [71, 63, 90, 31, 65], there exist a number
of similarity measures: the unweighted and weighted Pearson product-moment
correlation coefficient, the Moran’s index [29], the earth mover’s distance [90,
65], and the untrimmed and trimmed Mallows distance [71, 31].
Out of all these, the weighted Pearson product-moment correlation coeffi-
cient corresponds best to the above requirements. This measure corresponds
to the first non-dimensional similarity measure rS used in the remainder of this

chapter to validate the addition principle.
The correlation coefficient is sensitive to the difference in shape between
the distributions, yet insensitive to the difference in velocity magnitude. How-
ever, this latter feature must be included in the comparison, since the velocity
magnitude is related to the flow uniformity.
Therefore, a second similarity measure rM is introduced in Sect. 4.3.2 which
is only sensitive to the difference in velocity magnitude. More specifically, rM
compares the difference in terms of a flow uniformity measure.
4.3 Data reduction

This section describes the particular data reduction used for the validation of
the addition principle.
4.3.1 Flow uniformity measures

In the field of automotive catalysts, several measures are used to quantify the
flow uniformity. The most generally accepted one is the flow uniformity index,
introduced by Weltens et al. [108]. However, this thesis mainly uses a differ-
ent flow uniformity measure, defined as the mean-to-maximum velocity ratio.
This measure is also used with respect to the addition principle validation, to
compare two velocity distributions based on flow uniformity. This is discussed
in the Sect. 4.3.2.
Weltens’ flow uniformity index ηw [108]
The flow uniformity index ηw corresponds to a relative ‘variance’ of the velocity

distribution, where the variance is determined using the absolute value, rather
then the sum of squared deviations:
I
1 1 X
ηw = 1 − |Ui − Um | Ai (4.2)
2 |Um | A i=1
where A is the total cross-sectional area and I is the number of measurement

points. Ui represents the velocity in point (xi , yi ).
Equation (4.2) is actually an adaptation to the original formula proposed
by Weltens et al. [108], which does not account for the cross-sectional area Ai
in each grid point.
If the velocity U remains positive, the value of ηw varies between zero and
unity. For instance, ηw = 1/2 corresponds to a case where half the cross-
section experiences zero flow, and the velocity in the other half is twice the
mean velocity.
The flow uniformity is usually evaluated on the time-averaged24 veloc-

ity distribution, in which case Ui represents the time-averaged local velocity
Ui = U (x, y). However, it may be useful to evaluate the evolution of the flow
uniformity during one engine cycle. In this case, ηw (ωt) is defined based on
the time-resolved velocity U (x, y, ωt) and mean velocity Um (ωt):
I
1 1 X
ηw (ωt) = 1 − |Ui (ωt) − Um (ωt)| Ai (4.3)
2 |Um (ωt)| A i=1
According to the above definitions, the average of the time-resolved uni-

formity index ηw (ωt) (4.3) does not equal the time-averaged uniformity index
ηw (4.2).
The time-resolved definition according to Eq. (4.3) may result in problems
in case of strong flow reversal, as is the case on the CME flow rig (see Sect. 5.2).
The value of ηw (ωt) becomes undefined if Um (ωt) approaches zero.
The statistical inference for Weltens’ flow uniformity index can be deter-
mined from the propagation of the uncertainties on the velocity Ui and the
cross-sectional area Ai . The uncertainty on Ai is assumed negligible compared
to that on Ui . This is justifiable due to the use of a highly accurate automatic
velocity probe positioning system, featuring a positioning accuracy of better
than 0.25 mm.
The propagation of errors is based on the following principle, where the vari-
able Y is a function of n independent variables Xk , or Y = F (X1 , X2 , . . . , Xn ):
v
u n 2
uX ∂Y
∆Y = δY · Y = t ∆Xk (4.4)
∂Xk
k=1
where ∆ denotes the absolute error and δ denotes the relative error on a vari-
2 2
able. This relationship yields the well-known rules ∆ (X1 + X2 ) = (∆X1 ) +
2 2 2 2
(∆X2 ) and δ (X1 · X2 ) = (δX1 ) + (δX2 ) .
Applying the above error propagation principle to ηw according to Eq. (4.2)
results in the following expression for the relative error δηw :
v
u PI 2 2 2
1 − ηw u 2 i=1 (∆Ui + ∆Um ) Ai
δηw = uδU m + 2 (4.5)
ηw t
PI
i=1 |Ui − Um | Ai
where, using the same principle, δUm , ∆Ui and ∆Um can be further ex-
panded in terms of the uncertainties on individual ensemble-averaged velocities
U (xi , yi , ωt). The uncertainty on U (xi , yi , ωt) is defined in Sect. 2.5.1.
24 As mentioned in Sect. 2.5, the overline x (indicating time-averaging) and the tilde x e
(indicating non-dimensionalizing with the time-averaged mean velocity) are usually omitted
for the sake of clarity. Time-resolved quantities are specified with the crankshaft position
(ωt).
Mean-to-maximum velocity ratio ηm

In spite of the general use of the Weltens’ index, a different non-dimensional
flow uniformity measure is used throughout this thesis. ηm is defined as the
ratio of mean to maximum velocity. The following expressions define ηm for
the time-averaged and time-resolved case:
|Um |
ηm = (4.6)
Umax
|Um (ωt)|
ηm (ωt) = (4.7)
Umax (ωt)

where Umax = maxi∈[1,I] (|Ui |) = maxi∈[1,I] U (xi , yi ) and Umax (ωt) =
maxi∈[1,I] (|Ui (ωt)|) = maxi∈[1,I] (|U (xi , yi , ωt)|). Alternatively, ηm can be
defined based on the non-dimensional velocity U e [-], where U
e = U /Um :
1
ηm = (4.8)
Umax
e
1
ηm (ωt) = (4.9)
Umax (ωt)
e
where Ufm = Um /Um = 1, Umax = maxi∈[1,I] (|Ui /Um |) and Umax (ωt) =
e e
maxi∈[1,I] (|Ui (ωt)/Um (ωt) |).
The use of the absolute value ensures that 0 6 ηm 6 1. By contrast
to the definition of ηw (ωt) (4.3), the mean velocity does not appear in the
denominator in the definition of ηm . As such, ηm (ωt) can be used to quantify
the time-resolved flow uniformity in case of strong reversing flows.
Hald [44] discusses the sampling distribution of the largest observation in
a population. This approach is used to determine the sampling distribution of
the non-dimensional maximum velocity U emax :
n o n on
P Uemax = P U
e
n o n on−1 n o
and p Uemax = n P Ue p U
e (4.10)
where p {· · ·} and P {· · ·} represent the probability and the cumulative prob-

ability density function, respectively. Here, n denotes the maximum number
of spatially independent points, determined with the approach using Moran’s
index (see Sect. 4.3.2).
Eq. (4.10) assumes that the n velocity observations are stochastically in-
dependent and have the same probability density function. The probability
density function can be determined directly from the histogram of the non-
dimensional velocity Ue.
4.3.2 Distribution similarity

Shape similarity measure rS
The Pearson product-moment correlation coefficient ρ is an obvious choice
to quantify the comparison between two distributions, in terms of their non-
dimensional shape.
The unbiased estimator25 r for the Pearson product-moment correlation
coefficient ρ is defined as:
n
X
(zi,1 − zm,1 ) (zi,2 − zm,2 )
i=1
r= v (4.11)
u n n
uX 2
X 2
t (zi,1 − zm,1 ) · (zi,2 − zm,2 )
i=1 i=1
1
Pn
where zm,k = n i=1 zi,k , i and n are the index and number of data points.
In light of the validation of the addition principle, the correlation coefficient
is named rS . It forms the principal similarity measure used for quantifying the
shape similarity:
I
X
(Ui,1 − Um,1 ) (Ui,2 − Um,2 )
i=1
rS = v (4.12)
u I I
uX 2
X 2
t (U i,1 − Um,1 ) · (Ui,2 − Um,2 )
i=1 i=1
PI
where Um,k = A1 i=1 Ui,k Ai , i and I are the index and number of measure-
ment points. Ai and A representPIthe local and the total cross-sectional area in
the measurement points (A = i=1 Ai ). In practice, the area weighting using
Ai does not affect the value of rS , since the measurement grid points are a
priori arranged so that each grid point has the same cross-sectional area.
Detailed information on the statistical inference of the Pearson product-
moment correlation coefficient can be found in Hald [44], Spiegel [94], McPher-
son [72] and Guttmann et al. [42].
The sampling distribution of the correlation coefficient ρ is very skewed,
unless for the case ρ = 0. The confidence interval for r can be determined
using Fisher’s Z transformation. The statistic Z defined as [44]:

1 1+r
Z = log (4.13)
2 1−r
features an approximately normal
√ (i.e. normalized Gaussian) distribution with
a standard deviation equal to 1 n − 3 . The normal approximation is accept-
able if n exceeds 20.
25 For the sake of clarity, the statistical terminology such as ‘unbiased estimator’ will be
discarded throughout most of the discussion.

PDF of ρ using Fisher’s Z transformation

12
ρ = 0.9
ρ = 0.75
10 ρ = 0.5
Probability density function (-)

ρ=0
0
0 0.25 0.5 0.75 1
Correlation coefficient ρ (-)
Figure 4.2 – Probability density function of the correlation coefficient ρ, using

Fisher’s Z transformation in Eq. (4.13) with n = 30
Therefore, the confidence interval for a significance level26 α is:

1 1
Z − z1−α/2 √ < Z < Z + z1−α/2 √ (4.14)
n−3 n−3
where z1−α/2 is the critical value for the normal distribution corresponding to
a two-sided confidence interval at a significance level α (e.g. z1−α/2 = 1.96 for
α = 0.05). Transformation from Z to r uses the inverse of Eq. (4.13):
e2Z − 1
r=
(4.15)
e2Z + 1
Fig. 4.2 gives some examples of the probability density function for different
values of ρ, for n = 30.
Spiegel [94] provides the clearest explanation for using a hypothesis test
based on the Pearson product-moment correlation coefficient.
The hypothesis HS,1 tests for totally dissimilar distributions:

S,1 H0 : ρ = 0
H (4.16)
H1 : ρ > 0
For ρ = 0, the sampling distribution of the estimator r is symmetric, and r
may be transformed into a statistic t which features a Student’s t distribution
with n − 2 degrees of freedom (see Hald [44]):
√
r n−2
t= √ (4.17)
1 − r2
On the basis of a one-sided test of Student’s t distribution at a significance
level α, one would reject H0 if t > t1−α;n−2 .
26 The significance level α is equivalent to the confidence level 1 − α. In this thesis, the
confidence level is 95 % or α = 0.05.

The P-value P of an observation with respect to a hypothesis test is defined

as the probability that, given that the null hypothesis H0 is true, the test
variable assumes a more extreme value than the observation. The P-value
quantifies the statistical evidence for rejecting the null hypothesis H0 . If the
P-value is smaller than the significance level α, H0 is rejected.
For the above hypothesis HS,1 , the P-value PS,1 is defined as:
√
r n−2
PS,1 = Pt;n−2 √ <t (4.18)
1 − r2
where Pt;n−2 {t0 < t} is the probability that t exceeds t0 in a Student’s t dis-
tribution with n − 2 degrees of freedom. The value of PS,1 can be obtained
from the corresponding cumulative probability density function.
Recalling the definition of the hypothesis test HS,1 in (4.16), if PS,1 is
smaller than α = 0.05, then H0 is rejected and the correlation coefficient r
between the two distributions is significantly positive. The smaller PS,1 , the
stronger is the evidence for correlation.
The statistical strength of the hypothesis test HS,1 is somewhat limited due
to the fact that the sampling distribution of r becomes increasingly skewed as
r is closer to unity.
Whereas HS,1 tested for dissimilar distributions, the hypothesis HS,2 tests
for highly similar distributions:

S,2 H0 : ρ = ρcrit (> 0)
H (4.19)
H1 : ρ < ρcrit
where ρcrit can be any arbitrary ‘high’ correlation coefficient value. Because
of the exponential correlation between rS and S found during the experiments
(see Sect. 4.5), the value of ρcrit is taken equal to 1 − e−1 ' 0.63. Referring
to Sect. 4.5, the value of ρcrit corresponds to the critical value of rS when the
scavenging number equals the critical scavenging number. It is shown below
that ρcrit should be strictly smaller than unity. The sampling distribution for
the hypothesis test H0 : ρ = ρcrit 6= 0 features a skewed sampling distribution
in r (see Fig. 4.2). Therefore, Fisher’s Z transformation according to Eq. (4.13)
can be applied to transform the skewed r distribution into a symmetric one.
The statistic Z defined by Eq. (4.13) is approximately normal distributed with
√ value µz = 1/2 log ((1 + ρcrit )/(1 − ρcrit ) ) and standard deviation σz =
mean
1 n − 3.
With reference to Hald [44], the mean value µz actually equals µz = 1/2 ·
log ((1 + ρcrit )/(1 − ρcrit ) ) + ρcrit /2 (n − 1) . However, for n > 24, the second
term is more than 10 times smaller then the standard deviation, and as such, it
is omitted. This procedure cannot be performed unless ρcrit is strictly smaller
than unity.
For the hypothesis HS,2 , the P-value PS,2 is defined as:

1 1+r
PS,2 = Pz z < log − µz /σz (4.20)
2 1−r
where Pz {z < z0 } is the probability that z is smaller than z0 in a normal

distribution, and µz and σz are defined above.
If PS,2 is smaller than α = 0.05, then H0 is rejected and the correlation
coefficient r between the two distributions is significantly different from ρcrit .
Since the hypotheses HS,1 (4.16) and HS,2 (4.19) are each other’s inverse, so
are the corresponding P-values. The smaller PS,1 , the stronger is the evidence
for correlation. However, the smaller PS,2 , the stronger is the evidence against
correlation.
The choice of ρcrit is quite arbitrary. The only restriction is that ρcrit is
smaller than unity for using the Fisher transformation.
The number of samples n in the above discussion does not necessarily cor-
respond to the number of velocity measurement points. Instead, n corresponds
to the largest number of spatially independent points of the distributions. Spa-
tial dependence is determined using Moran’s index, which is introduced in the
following section.
Spatial autocorrelation
Moran’s index M is a measure of spatial autocorrelation. It is one of the oldest
available measures, introduced by Moran [77] in 1950, yet remains the standard
spatial autocorrelation measure. M is defined according to Cliff and Ord [29]:
N X
X N
n wij (zi − zm ) (zj − zm )
i=1 j=1
M= N
(4.21)
X 2
W (zi − zm )
i=1
PN PN
where wij are weighting factors, W = i=1 j=1 wij . According to the above
definition, M compares the value of the variable z (here: z = U , the catalyst
velocity) at any one location with the value at all other locations. Like the
correlation coefficient, M varies between −1 and 1. M ' 0 indicates no spatial
autocorrelation, M ' 1 indicates strong positive spatial autocorrelation.
The N -by-N weighting matrix wij can be determined in different ways.
Moran [77] originally introduced the index M in combination with wij = δij ,
where δij is a binary adjacency measure (not the Kronecker δ):

1 if elements i and j are adjacent, yet i 6= j ,
δij = (4.22)
0 otherwise.
Alternatively, instead of a binary (0,1) connection matrix, wij can be a prox-

imity measure, based on the inverse of the Euclidean distance or any other
distance measure. More information on alternative weighting is given by Cliff
and Ord [29]. In this thesis, the above adjacency definition is used according
to Moran [77].
Spatial autocorrelation Spatial autocorrelation

1 1
n = 30 n = 45
Moran’s M (-) Moran’s M (-)
P-value (-) P-value (-)
Moran’s index M (-)
Moran’s index M (-)

0.5 0.5
α = 0.05 α = 0.05
0 0
-0.5 -0.5
0 50 100 150 200 0 50 100 150 200
Number of points N (-) Number of points N (-)
(a) (b)
Figure 4.3 – Determining the number of spatially independent points in the catalyst
velocity distribution, for manifold (a) A and (b) B
If the sample size n is sufficiently large, Moran’s index M is approximately

normally distributed, with a mean value and variance defined in Cliff and
Ord [29] by Eqs. (2.24) and (2.28), respectively. Based on the mean value,
variance and normality assumption, a confidence interval can be established
for M .
Moran’s index can now be applied to determine the number of spatially in-
dependent points in a measured velocity distribution. A spatial autocorrelation
hypothesis test H is constructed based on M :

H0 : M = 0 no spatial autocorrelation
H (4.23)
H1 : M > 0
The P-value P for this hypothesis test is based on the normality assumption
2
for M and the mean value µM and variance σM defined by Eqs. (2.24) and (2.28)
in Cliff and Ord [29]:

M − µM
P = Pz <z (4.24)
σM
For any given velocity distribution, Moran’s index can be determined, along
with P . According to the definition of the hypothesis test H, there is a signif-
icant spatial autocorrelation if P is smaller than the significance level α. The
number of data points in the velocity distribution N is gradually decreased
until P becomes greater than α. Decreasing the number of points is performed
by two-dimensional linear interpolation of the original data on a mesh with
increasingly larger spacing.
Figure 4.3 gives an example of determining the number of spatially inde-
pendent points n for velocity distributions obtained on manifolds A and B. The
values for n are relatively insensitive to changes in flow conditions. As such,

the number of spatially independent points are assumed n = 30 for manifold
A, and n = 45 for manifold B. These values are used e.g. in the statistical
inference for the correlation coefficient, as discussed in the preceding section.
Since the number of spatially independent points is relatively constant for
each of the manifolds, critical limits for rS indicating significant similarity can
be derived based on the hypothesis test HS,1 . For a 95 % confidence level, one
can derive that two velocity distributions are significantly correlated when:

rS > 0.317 for manifold A (n = 30)
(4.25)
rS > 0.254 for manifold B (n = 45)
The difference between the limits is solely due to the different spatial depen-
dence of the distributions of manifolds A and B.
At first sight, these limits appear quite low. Strictly statistically speaking,
the limits apply. However, the discussion on the validation of the addition
principle in Sect. 4.5 yields a more intuitively correct similarity limit, rS >
rS,crit = 1 − e−1 ' 0.63.
Magnitude similarity measure rM

The correlation coefficient rS (4.12) is used to quantify the similarity between
two velocity distributions based on differences in shape. This section explains
the selection procedure for a second similarity measure, based on differences
in magnitude. Two analytical examples are provided to clarify and justify this
selection.
Example 1 (using a cosine velocity distribution) show that rS , by nature
of its definition, is insensitive to differences in magnitude between two velocity
distributions with identical shape. Following Example 1, the magnitude simi-
larity measure rM is introduced, based on the ratio of the flow uniformity in
pulsating and stationary conditions.
Example 2 (using a non-negative velocity distribution) clarifies the choice
of ηm as a flow uniformity measure for the magnitude similarity measure rM .
Example 1: Cosine distribution Consider a simple one-dimensional

case, comparing two velocity distributions U1 (x) and U2 (x), defined as:
U1 = Um,1 (1 + a1 cos (2πx))

U2 = Um,2 (1 + a2 cos (2πx)) (4.26)
where x is the coordinate (−1/2 6 x 6 1/2 ), a1 and a2 are non-

dimensional numbers (a > 0) that determine the magnitude of the veloc-
ity variations in U1 and U2 (see Fig. 4.4a).
The continuous version of the discrete definition of rS (4.12) is:
Z 1/2
(U1 − Um,1 ) · (U2 − Um,2 ) dx
x=−1/2
rS = sZ (4.27)
1/2 Z 1/2
2 2
(U1 − Um,1 ) dx · (U2 − Um,2 ) dx
x=−1/2 x=−1/2
Substituting the expressions for U1 and U2 in Eq. (4.27) yields rS = 1.

Thus, rS is insensitive to changes in mean velocity Um or changes in
velocity magnitude a.
In light of the insensitivity of rS to differences in magnitude of the velocity

distributions, a second similarity measure rM is introduced. Ideally, rM should
complement rS in terms of its sensitivity. In other words, rM should be sensitive
to differences in magnitude, yet insensitive to differences in the shape of the
distribution.
As such, the magnitude similarity measure rM is defined as the ratio of the
flow uniformity measures for each velocity distribution, based on the mean-to-
maximum velocity ratio ηm to quantify the flow uniformity:
ηm,puls
rM = (4.28)
ηm,stat
Substitution of the definition of ηm (4.6) into Eq. (4.28) yields:
(Um /Umax )puls

rM =
(Um /Umax )stat
.
1 Uemax
= . puls
1 Umax
e
stat
U
emax,stat
= (4.29)
U
emax,puls
Unlike the definition of rS (4.12), the above definition of rM is not sym-

metrical (ηm,puls /ηm,stat 6= ηm,stat /ηm,puls ). For the validation of the addi-
tion principle, the subscript puls in Eqs. (4.28) and (4.29) denotes pulsating
flow conditions and stat denotes the linear combination of stationary flow con-
ditions. Therefore, rM quantifies the relative increase in flow uniformity in
pulsating flow, compared to the averaged stationary distribution. If rM > 1,
the pulsating flow has a better flow uniformity than the stationary averaged
flow. As discussed in Sect. 4.4.3, this is actually the case for all flow conditions
encountered.
Example 1: Cosine distribution (cont’d) Continuing the simple
analytical example described above, this example examines the behavior
of the flow uniformity measures ηw and ηm , and the magnitude similarity
measure rM .
The continuous version of the definition of Weltens’ flow uniformity index

ηw (4.2) is:
Z 1/2
1 1
ηw = 1 − |U (x) − Um | dx (4.30)
2 |Um | x=−1/2
R1
where x=−1
dx = 1 vanishes. Substituting Eqs. (4.26) into Eq. (4.30)
yields:
Z 1/2
1 |Um a|
ηw = 1− |cos (2πx)| dx
2 |Um | x=−1/2
a
= 1− (4.31)
π
The mean-to-maximum velocity ratio ηm (4.6) is simply:
Um
ηm =
Umax
Um
=
Um (1 + a)
1
= (4.32)
1+a
The following table summarizes the behavior of the flow uniformity mea-
sures ηw and ηm for the velocity distribution in this simple example, for
different values of the magnitude parameter a (see Fig. 4.4a). Both mea-
sures reach a maximum value of unity for a perfectly uniform distribu-
tion. Figure 4.5a shows the evolution of ηw and ηm versus the magnitude
parameter a.
ηw ηm
Eq. (4.31) Eq. (4.32)
a=0 1 1
a=1 0.682 0.5
a=π 0 0.242
a→∞ −∞ 0
Substituting Eqs. (4.26) into the definition of the magnitude similarity

measure rM (4.28) yields (assuming subscript 1 and 2 correspond to puls
and stat, respectively):
ηm,1
rM =
ηm,2
1 + a2
= (4.33)
1 + a1
Since ηm varies between zero and unity, rM varies between zero and +∞.
Example: Cosine distribution Example: Non-negative distribution

2.5
4 a=0 a=0
a=1 a=1
a=π a=2
3 2 a=4
2
1.5
U / Um (-)
U / Um (-)
1
1
0
-1 0.5
-2
0
-0.5 0 0.5 -0.5 0 0.5
x (-) x (-)
(a) (b)
Figure 4.4 – Velocity distributions, for (a) cosine (see Example 1) and (b) non-
negative (see Example 2) distributions
In Example 1, Eq. (4.33) shows that the definition of rM based on ηm yields

a straightforward and intuitive interpretation, at least for this simple one-
dimensional test case. Example 2 presents a derivation analogous to Example 1,
yet for a non-negative velocity distribution.
Example 2: Non-negative distribution This example discusses
the behavior of rS , ηw , ηm and rM for a simple yet non-negative one-
dimensional velocity distribution, for −1/2 6 x 6 1/2 .
Consider two velocity distributions U1 (x) and U2 (x), defined as:
2
e−(a1 x)
U1 = Um,1 √
π/a1 erf(a1 /2 )
2
e−(a2 x)
U2 = Um,2 √ (4.34)
π/a2 erf(a2 /2 )
where a1 and a2 are non-dimensional numbers (a > 0) that determine the

magnitude and (to a lesser extent) the shape of the velocity variations in
√ Rz 2
U1 and U2 . The error function erf is defined as erf(z) = 2/ π t=0 e−t dt.
These distributions correspond to the normalized Gaussian distribution
function, scaled such that the mean velocity over the interval −1/2 6
x 6 1/2 equals Um .
By contrast to Example 1 using the cosine distribution, the parameters a
affect both the magnitude and the shape of the distributions. As shown
in Fig. 4.4b, the Gaussian bell shape is relatively insensitive to small
variations in a. Since the shape is altered by changes in a, rS is no
longer constant as is the case in Example 1 with the cosine distribution.
R 1/2 2 √
Taking into account that x=−1/2 e−(ax) dx = π/a erf(a/2 ), rS yields:

Flow uniformity measures ηw and ηm Flow uniformity measures ηw and ηm
1 1
ηw ηw
ηm ηm
0.8 0.8
Flow uniformity (-)
Flow uniformity (-)

0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 1 π 6 8 10 0 1 2 4 6 8 10
Magnitude parameter a (-) Magnitude parameter a (-)
(a) (b)
Figure 4.5 – Flow uniformity measures for different velocity magnitude, for (a)
cosine (see Example 1) and (b) non-negative (see Example 2) distributions

Similarity measures rS and rM Similarity measures rS and rM
2.5 2.5
rS rS
rM rM
2 2
Similarity measures (-)
1.5 1.5
1 1
0.5 0.5
Pulsating case (1) is more uniform Pulsating case (1) is more uniform
than stationary case (2) for a2 > a1 than stationary case (2) for a2 > a1
0 0
0 1 2 3 4 0 1 2 3 4
Magnitude parameter ratio a2/a1 (-) Magnitude parameter ratio a2/a1 (-)
(a) (b)
Figure 4.6 – Similarity measures for different velocity magnitude ratios, for (a)
cosine (see Example 1) and (b) non-negative (see Example 2) distributions
2 „√ «2 3 12
a2 2
1 +a2
erf 2
6 2a1 a2
6 7
rS = 6 2 2
“ ” “ ”7 (4.35)
4 a1 + a2 erf √
7
a1 a2 5
2
erf √2
For typical velocity distributions, a varies between zero (i.e. perfectly

uniform) and four (see Fig. 4.4b). Taking a1 as unity and varying a2
between zero and four, yields a variation of rS between 0.83 and unity,
according to Eq. (4.35). This is shown in Fig. 4.6b as well.
No analytical solution can be found for Weltens’ flow uniformity index
ηw (4.2).
The mean-to-maximum velocity ratio ηm (4.6) is:
Um
ηm =
Umax
√
π “a”
= erf (4.36)
a 2
ηw ηm
Eq. (4.36)
a=0 1 1
a=1 0.968 0.923
a=2 0.881 0.747
a=4 0.649 0.441
a→∞ 0 0
The table above summarizes the behavior of the flow uniformity mea-
sures ηw and ηm for the velocity distribution in this simple example.
Both measures reach a maximum value of unity for a perfectly uniform
distribution. Figure 4.5b shows the evolution of ηw and ηm versus the
magnitude parameter a.
Substituting Eqs. (4.34) into the definition of the magnitude similarity
measure rM (4.28) yields (assuming subscript 1 and 2 correspond to puls
and stat, respectively):
ηm,1
rM =
ηm,2
`a ´
a2 erf 21
= (4.37)
a1 erf a22
` ´
The above examples for a cosine velocity distribution and a non-negative veloc-
ity distribution provide some reference values for the flow uniformity measures
ηw and ηm , and the related magnitude similarity measure rM . For a varying
ratio of the magnitude parameter a, Fig. 4.6 presents the variation of rS and
rM for both examples.
In the definition of rM (4.28), the mean-to-maximum velocity ratio ηm (4.6)
is chosen over Weltens’ flow uniformity index ηw (4.2) for the following reasons:
• Boundedness — In case of a not strictly positive velocity distribution (e.g.

see cosine distribution in Example 1), only ηm remains bounded between
zero and unity. For distributions with strong flow reversal, ηw can assume
negative values with an unbounded amplitude (see Fig. 4.5a).
Since rM is to be defined as the ratio of two flow uniformity measures, a
uniformity measure whose value may become zero (such as ηw ) is unde-
sirable.
• Sensitivity — As shown in Fig. 4.5, ∂ηm /∂a (i.e. the derivative of ηm
with respect to the magnitude parameter a) is greater in absolute value
than ∂ηw /∂a . At least, this is true in the region of typical magnitude
parameter values, between zero and four.
In other words, ηm is more sensitive to changes in the magnitude of the
velocity distribution than ηw , yet still ηm is the only measure that remains
bounded.
The statistical significance of rM is based on (i) the sampling distribution of

ηm according to Eq. (4.10), and (ii) on the Mellin convolution. The Mellin
convolution is a technique described e.g. in Springer [95] (Sect. 4.1). Equa-
tion (4.1.7) in [95] presents the probability density function of the quotient
y = z1 /z2 of two non-negative independent random variables with probability
density functions p1 {z1 } and p2 {z2 }, expressed as the Mellin convolution:
Z ∞
z1
p y= = z2 · p1 {y z2 } · p2 {z2 } dz2 (4.38)
z2 z2 =0
Equation (4.38) represents the probability density function of the magnitude

similarity measure rM . Symbolically, p {rM } = p {y = z1 /z2 }, where z1 and
z2 correspond to ηm,puls and ηm,stat , respectively. The probability density
functions p1 and p2 are determined according to Eq. (4.10).
An example of the probability density function for one particular combi-
nation of two velocity distributions is given by Fig. 4.7. The irregular shape
of p {rM } results from the Mellin convolution and the irregular shape of the
distributions for each ηm .
Since this method provides the entire estimated probability density for rM ,
no assumptions are required as to the nature of the distribution, in order to
determine the statistical inference.
The (1 − α) 100 % confidence interval rM,α/2 , rM,1−α/2 is defined as:

PrM rM,α/2 6 rM 6 rM,1−α/2 = 1 − α (4.39)
where the confidence limits rM,α/2 and rM,1−α/2 result directly from the cumu-
lative probability density function PrM {rM }, which is obtained by integrating
p {rM } (4.38).
Similar to rS , a hypothesis test HM is defined as:

H 0 : rM = 1
HM (4.40)
H1 : rM 6= 1
PDF of magnitude similarity measure rM
p{y}
y = rM
Probability density function (-)
0.5 1 rM 1.5 2
Magnitude similarity measure y (-)
Figure 4.7 – Probability density function for rM , based on Eq. (4.10) and the Mellin
convolution (4.38)
The P-value PM for this hypothesis test is based on the probability density for
rM obtained as described above:
PM = PrM {|rM − y| > |rM − 1|} (4.41)

where the right hand side represents the probability of an observation y with
a greater deviation to unity than the actual value rM − 1, in the assumption
that the null hypothesis is true.
Figure 4.7 shows an example of the probability density function of rM ,
obtained using the above described approach. The combined area to either
side of the dotted vertical lines, below the function p (rM ), corresponds to the
P-value PM defined by Eq. (4.41).
4.3.3 Flow characteristic

The pulsating flow in the exhaust manifold with close-coupled catalyst enforces
a periodic scavenging of the diffuser or pre-cat chamber. The scavenging process
is determined by the diffuser volume Vd [m3 ], the exhausted gas volume per
cylinder and per cycle [m3 ], the number and layout of the exhaust runners
issuing into the diffuser, and the flow pulsation period Tp [s].
These quantities are combined into a non-dimensional number S [-] to char-
acterize the scavenging process. The scavenging number S is used in Sect. 4.4
to correlate the similarity measures rS and rM introduced in Sect. 4.3.2. S is
defined as the ratio of two time scales involved in the scavenging process:
Tp
S= (4.42)
Ts
where Ts is the scavenging time scale [s] or the residence time scale of the gas
passing through the diffuser. Ts is defined as the ratio of diffuser volume Vd to

time-averaged volumetric flow rate Q [m3 /s] through the catalyst:
Vd
Ts = (4.43)
Q
The volumetric flow rate Q can be (i) calculated from the time-averaged
velocity distribution or (ii) obtained from a reference flow rate reading. The
reference flow rate reading corresponds to a standardized orifice for the isother-
mal flow rig (see Sect. 2.4.1) and a laminar flow meter for the CME flow rig
(see Sect. 2.4.2).
During the initial experiments on the isothermal flow rig, Tp was based only
on the engine speed N [rpm]:
120/N
Tpold = (4.44)
nr
where nr is the number of runners issuing into the catalyst. The true flow
period is one engine cycle (i.e. two crankshaft revolutions) or 120/N . Each
engine cycle features nr exhaust pulses or exhaust gas passing through the
catalyst. Thus Tp represents the apparent flow pulsation period experienced
by the manifold.
Figure 4.8 indicates the two basic time scales involved in this process: (i) the
apparent flow pulsation period (Tp ), which should not be confused with the
engine cycle period 120/N , and (ii) the scavenging or residence time scale of
the diffuser Ts . The dotted area represents the remaining exhaust gas from
the previous exhaust stroke, which is being scavenged from the diffuser by the
fresh incoming exhaust gas.
The statistical inference for S with the above definition of Tpold is obtained
using the principle of propagation of errors, as defined by Eq. (4.4):
q
2
δS old = δTpold + δTs2
p
' δN 2 + δQ2
' δQ (4.45)
where the uncertainties on geometrical parameters (e.g. δVd ) and the engine
speed are negligible compared to the flow rate error (typically about 5 to 10 %).
The definition of the apparent pulsation period Tp according to Eq. (4.44)
is no longer valid for the pulsating flow generated by the isochoric (CME) flow
rig, or indeed for fired engine conditions.
The exhaust stroke in the CME flow rig consists of blowdown and displace-
ment phases. This two-stage nature in combination with strong Helmholtz
resonances during the displacement phase results in a smaller apparent flow
pulsation period. Therefore, the following new definition of Tp is introduced:
1
Tp = (4.46)
fP SD(Um ),max
U #1 #3 #4 #2
a b c d
Tp
120 / N
(a) (b) (c) (d)
Figure 4.8 – Periodic scavenging of the diffuser
where the frequency fP SD(Um ),max [Hz] corresponds to the maximum in the
power spectral density of the time-resolved mean velocity Um (ωt). The ap-
parent flow pulsation period experienced by the catalyst is better described by
Eq. (4.46) than by the old definition (4.44).
Figure 4.9 gives an example of the mean velocity power spectral density
for both flow rigs, for identical engine speed and a volumetric flow rate corre-
sponding to part load conditions. For the isothermal flow rig, Fig. 4.9a shows
that the peak frequency in the spectrum corresponds to fP SD,max ' nr N /120 .
This is generally true for all experiments on the isothermal flow rig. As such,
the value of Tp according to the new definition in Eq. (4.46) corresponds to the
value of the old definition in Eq. (4.44). The results of Persoons et al. [88] are
therefore not undermined. As shown in Fig. 4.9b, the frequency content of the
flow in the CME flow rig is much higher.
The statistical inference on Tp = fP−1 SD,max is difficult to obtain. It is
assumed that, similar to the old definition of S, the greatest uncertainty is
due to Ts and not Tp . As such, δS is estimated equal to δQ, regardless of the
definition of Tp .
The behavior of S may be summarized as follows. A high scavenging num-
ber S (e.g. low engine speed and/or high flow rate) means that the diffuser
scavenging occurs faster than the flow pulsation period, therefore the catalyst
flow distribution should be relatively unaffected by changes in S, or indeed
changes in engine speed or flow rate. A low scavenging number S (e.g. high
engine speed or low flow rate) results in the opposite, meaning more interfer-
ence of exhaust flow pulses from individual runners. Consequently, the flow
distribution should be more sensitive to changes in S in the lower range of S.
Other researchers use equivalent dimensionless numbers for characterizing
4.4 Experimental results 125
-1 -1
10 10
nr N / 120 = 40.0 Hz nr N / 120 = 40.0 Hz
Energy spectral density of Um ((m/s)2/Hz)

fPSD,max = 40.5 Hz fPSD,max = 118.9 Hz
-2 -2
10 10
-3 -3
10 10
-4 -4
10 10
-5 -5
10 10
-6 -6
10 10
0 100 200 300 400 500 0 100 200 300 400 500
Frequency (Hz) Frequency (Hz)
(a) (b)
Figure 4.9 – Power spectral density of the time-resolved mean velocity Um (ωt) on
(a) isothermal and (b) CME flow rig, for N = 1200 rpm and part load
pulsating flow in a close-coupled catalyst manifold. Benjamin et al. [17] define

J as the ratio of pulsation period to diffuser residence time. Benjamin et al.
[17] use an axisymmetric isothermal flow rig. The residence time is defined
in terms of the length of the diffuser and the mean velocity in the inlet pipe,
which corresponds to a single exhaust runner. J is proportional to S, defined
using Eq. (4.44) as definition for Tp .
Bressler et al. [22] define GEN or ‘gas exchange number’ as the ratio of
exhausted gas volume per cylinder per cycle to the diffuser volume. Both
papers present relationships between flow uniformity in pulsating flow and this
dimensionless number.
In a numerical study on the conversion efficiency of a catalyst subjected to
pulsating flow, Tsinoglou and Koltsakis [99] non-dimensionalize the pulsation
frequency with the catalyst residence time. The pulsation index is inversely
proportional to S.
Regarding the validation of the addition principle, the results of this thesis
are compared to the findings of Benjamin et al. [17] and Bressler et al. [22] in
Sect. 4.5.
4.4 Experimental results

Recalling the definition of the addition principle (4.1), the time-averaged di-
mensionless catalyst velocity distribution in pulsating flow conditions
(= Ue (x, y)
puls ) is compared to a linear combination of nr dimensionless veloc-
ity distributions (= U e (x, y)
stat,r , where r = 1 . . . nr ), obtained for stationary
flow through each of the exhaust runners.
Table 4.1 – Summary of experimental campaign

Flow rig Pulsator Stationary flow Pulsating flow OHW
Isothermal RV A A 2
CH A+B A+B 2
Isochoric (CME) CH B 4
The time-averaged velocity distribution obtained in pulsating conditions

Ue (x, y)
puls is usually abbreviated to pulsating flow distribution or Upuls .
The velocity distributions obtained for stationary flow through each of the
exhaust runners U e (x, y)
stat,r is usually abbreviated to Ustat,r .
The velocity distribution resulting from the linear combination defined by
Eq. (4.1) is usually abbreviated to stationary (averaged) flow distribution or
Ustat .
Two flow rigs are used to generate pulsating flow. The (i) isothermal flow
rig set up is described in Sect. 2.2.1. The (ii) isochoric or charged motored
engine (CME) flow rig is described in Sect. 2.2.2.
The stationary flow distributions Ustat,r cannot be obtained on the iso-
choric flow rig. Therefore, these measurements are always performed on the
isothermal flow rig. For the rotating valve pulsator (see Sect. 2.2.1), the Ustat
measurements are performed with the valve blocked in the maximum flow rate
position. For the cylinder head pulsator (see Sects. 2.2.1 and 2.2.2), the Ustat
measurements are performed with the exhaust camshaft blocked in the position
of maximum valve lift. For each of the nr stationary flow cases, flow is only
allowed through a single exhaust runner.
The addition principle is validated by examining the similarity between the
averaged stationary distribution Ustat defined by Eq. (4.1) and the pulsating ve-
locity distribution, for identical time-averaged volumetric flow rate. Typically,
the flow rate is set to the same value throughout pulsating flow experiments
with different engine speeds. These are then all compared to the same station-
ary flow experiment.
Table 4.1 gives a brief overview of the experimental campaign. ‘RV’ and
‘CH’ denote rotating valve and cylinder head pulsator, respectively. ‘A’ and ‘B’
indicate the use of the three-runner manifold A without exhaust valve overlap,
and the four-runner manifold B with exhaust valve overlap (see Sect. 2.1). The
final column indicates where the oscillating hot-wire anemometer (see Chap. 3)
has been applied for measuring bidirectional velocity.

Each of the velocity distributions are plotted as dimensionless velocity Ue =
U /Um , so as to better compare the distributions at different flow rate. The
actual reference volumetric flow rate Qref [m3 /h] and the mean time-averaged
velocity Um [m/s] are indicated on each plot. For the isothermal flow rig,
the reference flow rate corresponds to the orifice flow rate measurement (see
Sect. 2.4.1). Also indicated are: the exhaust runner (in case of a stationary
velocity distribution), and the flow uniformity measures ηm and ηw , according
to Eqs. (4.6) and (4.2) respectively.
The velocity distribution itself is plotted using contour lines, where the
‘elevation’ U
e of each contour line is indicated in the vertical scale on the right
side of the plot. The unity contour U e = 1 is plotted as a dashed line ( ). The
x- and y-coordinates are presented in mm and correspond to the actual size of
the catalyst cross-section. As mentioned in Sect. 2.1, all velocity measurements
are performed in a plane 25 mm downstream of the catalyst, to avoid the small-
scale mixing region of the laminar jets issuing from individual catalyst substrate
channels.
Below each two-dimensional plot is a one-dimensional cross-sectional plot
of the velocity along the line y = 0 mm.
Manifold A (see Table 2.1)

Figures 4.10 and 4.11 show examples of velocity distributions Ustat,r obtained
on the isothermal flow rig, for stationary flow through each of the three runners
of manifold A (see Table 2.1). The rotating valve is used for the case in Fig. 4.10,
and the cylinder head is used for the case in Fig. 4.11.
The linearly combined velocity distributions Ustat according to Eq. (4.1) are
presented in Fig. 4.12. For Ustat , the values for flow rate and mean velocity are
only indicative. Each corresponds to the mean of the values for the stationary
distributions Ustat,r .
For the isothermal flow rig measurements in pulsating flow conditions, the
term ‘engine speed’ N is used to quantify the flow pulsation frequency, al-
though the flow rig contains no crankshaft. For each runner, the rotating valve
provides two openings per valve revolution. As such, the rotating valve speed
corresponds to N /4 [rpm]. For the cylinder head experiments, the exhaust
camshaft is driven at a speed N /2 [rpm].
Figures 4.13 and 4.14 provide comparisons between time-averaged velocity
distributions Upuls obtained in pulsating flow conditions and stationary aver-
aged distributions Ustat at the same volumetric flow rate. Both are obtained
using the rotating valve, at a comparable flow rate Qref ' 75 m3 /h, yet for
different values of the engine speed N . Figure 4.13 corresponds to a high
scavenging number (S ' 2.0) and Fig. 4.14 to a very high scavenging number
(S ' 4.2). There are no cases available for low scavenging number for mani-
fold A, since the relationship between the scavenging number and the addition
principle’s validity has only been established during the experiments on man-
ifold B. These occurred chronologically much later than the experiments on
manifold A.
Figures 4.15 and 4.16 provide similar comparisons between Upuls and Ustat
for the cylinder head as pulsator. Based on a visual inspection of Figs. 4.13
through 4.16, the similarity is very good. This is due to the high scavenging
number (S > 2). If S is sufficiently greater than unity, the scavenging time
scale is smaller than the flow pulsation period. In that case, one would expect
Stationary velocity U (-) Stationary velocity U (-)

4 4
Runner 1, Qref = 74.5 m3/h Runner 2, Qref = 73.2 m3/h
30 3.5 30 3.5
2.2
.6
0.2
.8
0.4001
1.8
1.4
1.4
3 3
0.8
0.6
0.6
0.4 0.
15 1.8 15 4
0.4
0.4
2.5 0.4 2.5
0.6
1
y (mm)
y (mm)
0.8
0 2 0 2
0.6
2.2
0.2
0.6
1.41
2.6
1.5 1.5
3
0.4
2.2
-15 2.2
-15 3
1
2.6
0.6
.6
01.8
1.8 1 22.2 1
3.4
1.8
1.4 0.8
1.4
0.8 0.6 0.4
1
4
1.4
0.
1 1.8
-30 0.5 -30 0.8 0.6 0.5
0.4
Um = 5.465 m/s, ηm = 0.342, ηw = 0.689 Um = 5.215 m/s, ηm = 0.307, ηw = 0.645
0 0
Uy=0 (-)
Uy=0 (-)
3 3
2 2
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)

4
1 1.
30 4 00.6
.8 3.5
2.2
1.4
3
0.4
1.8
15
1.8
0.6
0.8
2.5
1
2.2
0.4 1.8
y (mm)
0.4
0 2
0.4
6
0.
1
1.4
1.5
0.6
0.8
-15 0.2 0.6
1
0.4
0.4 0.8
-30 0.5
Um = 5.355 m/s, ηm = 0.441, ηw = 0.702
0
Uy=0 (-)
3
2
1
0
-30 -15 0 15 30
x (mm)
(c)
Figure 4.10 – Stationary velocity distributions Ustat,r for flow through each runner
(a, b, c: r = 1 . . . 3) for manifold A on the isothermal flow rig with rotating valve
(Qref ' 75 m3 /h)

4 4
1
0.6 1.4 3.5 3.5
30 0.4
0.8
2.2 30 0.2
0.4
0.4
0.8
0. 3 0.6 3
6
15 15
2.2 1.8
1.8
0.6
0.4
2.5 2.5
1.4
0.6
y (mm)
y (mm)
2.6
0.4 0.6
0.6
2.2
0 2 0 2
1.8
1
0.6
0.2
1
1.4 0.8
1
0.6
0.8 1.8
1.4
1
1.5 1 2.2
1.5
0.4
-15 -15 2.6

0.8
1 2.6 2.2 1
0.4 1. 3
0.8 4 2.2
0.8
1
-30 -30 1.4 1.8 0.6
0.4 0.5 0.4 0.4 0.5
0 0
Uy=0 (-)
Uy=0 (-)
3 3
2 2
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)

4
0.60.4 3.5
30
1
2.2
1.8
1
0.
3
8
1.4
1.4
0.8
15
0.6
1.8
2.5
1.8
y (mm)
0.4
0 2
6
2.2
0.
1.4
0.6
0.4
1.5
4
-15 0.
0.
6
0.81 1
0.2 0.6
-30 0.4
0.5
Um = 4.891 m/s, ηm = 0.459, ηw = 0.723
0
Uy=0 (-)
3
2
1
0
-30 -15 0 15 30
x (mm)
(c)
(a, b, c: r = 1 . . . 3) for manifold A on the isothermal flow rig with cylinder head
(Qref ' 75 m3 /h)
Stationary averaged velocity U (-) Stationary averaged velocity U (-)

2.5 2.5
N = 0 rpm, Qref ~ 73.6 m3/h N = 0 rpm, Qref ~ 75.8 m3/h
0.7
0.8 0. 0.5
0.40.6
30 1.2 1 1.4 9 30 0.7 1.2 1
.6 0.5
0.
6 00..98
2 2
1
1.4
0.04.
0.40.6
5
15 15
0.7
0.9
1.5 1.5
0.8
1.4
0.5
y (mm)
y (mm)
1
0.7
0 1.4
0.8 0
0.5
0.7 9 1.6
1.6 0.6
1.4
1.2
0.
1 1.8 1 1
0.6
1.4
1.2
0.8
1.8
0.4
1.2
0.7
1
-15 -15 1.4
0.6
1.2
1.4
1
1.12
0.8
1.4
0.9
1.9.2 0.5 0.5
0.9
1
1 00.7 0.9
-30 0.8
0.60.5
0.4
-30 0.70.8
0.5
0.6
0.3
Um ~ 5.221 m/s, ηm = 0.613, ηw = 0.873 Um ~ 5.254 m/s, ηm = 0.581, ηw = 0.879
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)
Figure 4.12 – Stationary averaged velocity distributions Ustat according to Eq. (4.1)
for manifold A on the isothermal flow rig with (a) rotating valve and (b) cylinder head
(Qref ' 75 m3 /h)
Time-averaged velocity U (-) Stationary averaged velocity U (-)

2.5 2.5
N = 2810 rpm, Qref = 75.7 m3/h N = 0 rpm, Qref ~ 73.6 m3/h
0.8
0.91 0.7
0.8 0.
30 30 1.2 1 1.4 9
.6 0.
6
2 2
1
1.4
0.04.
5
0.7
15 15
0.6
0.7
1.2
0.7
0.8
1.5 1.5
1.2
0.5
y (mm)
y (mm)
0 0
0.8
1.4
0.8 0.9
0.6
0.7 9
1.6
0.6
1.4
0.
1 1
0.6
1.2
1.4
0.8
1.8
0.4
0.9
1
1.2
-15 -15
1.2
1.4 1
9
0.
1.4
0.9
1 0.5 0.5
1.9.2
1
1 00.7
-30 0.8
0.7 -30 0.8
0.60.5
0.4
0.3
Um = 5.400 m/s, ηm = 0.686, ηw = 0.902 Um ~ 5.221 m/s, ηm = 0.613, ηw = 0.873
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)
Figure 4.13 – Comparison of (a) Upuls and (b) Ustat according to Eq. (4.1), for
manifold A on isothermal flow rig with rotating valve, at high scavenging number
(S = 2.134, rS = 0.960, rM = 1.119)

2.5 2.5
0.6
0.7 0.7
0.8 0.
30
0.8 1.2 30 1.2 1 1.4 9
0.91 1.4 .6 0.
6
2 2
0.04.
5
0.6
15 15
0.5
0.6
0.7
1.4
1.5 1.5
0.5
y (mm)
y (mm)
1
0.9
1
0 0
0.8
1.4
0.8
1.4
0.7 9
0.7
1.6
0.6
1.4
0.
1 1
0.6
1.2
1.2
0.7
0.8
1.8
1.4
0.4
0.8
1.2
0.91
1.2
-15 -15
1
0.9
0.8
1
1.4
0.9
1.2 0.5 1.9.2 0.5
1
1 00.7
-30 0.6 0.7 -30 0.8
0.60.5
0.4
0.3
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)
manifold A on isothermal flow rig with rotating valve, at very high scavenging number
(S = 4.252, rS = 0.965, rM = 1.126)

2.5 2.5
0.8 0.7 0.5
0.40.6
30 0.9 30 0.7 1.2 1 0.5
00..98
2 2
1.2
1.4
0.40.6
1
15 15
0.9
1.5 1.5
0.8
1.4
1
y (mm)
y (mm)
0.7
0.7
0 0
0.5
1.6
1.4
0.9
0.
1.2
1
1.
1 1.8 1
8
1.4
1.2
4
0.8
0.7
0.7
1
-15 -15 1.4

0.6
0.9
1.2
1.4
1.12
0.8
1 1.2 0.5 0.5

0.9
1
-30 -30 0.9

0.70.8 0.6
0.8 0.5

0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)
manifold A on isothermal flow rig with cylinder head, at high scavenging number
(S = 2.158, rS = 0.944, rM = 1.282)

2.5 2.5
7 0.6
0.9 1 0.5
0.40.6
30 0.80. 1.2 30 0.7 1.2 1 0.5
1.4 00..98
2 1.4 2
0.40.6
0.6
15 15
0.9
1.5 1.5
0.7
0.8
1.4
y (mm)
y (mm)
0.7
0.8
0 0
0.5
1.6
0.7
1.6
1.2
1
1.4
1 1.8 1
1.4
1
0.9
1.2
1.4
0.6
0.9
1
0.7
1
-15 -15 1.4
0.6
1.2
1.4
1.2
0.8
2
1.
1.12
0.8
0.9 0.5 0.5
0.9
1 0.8 0.9
-30 0.8 -30 0.70.8
0.5
0.6

0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)
manifold A on isothermal flow rig with cylinder head, at very high scavenging number
(S = 4.334, rS = 0.971, rM = 1.153)
only a minor influence of the flow conditions (engine speed, flow rate) on the
velocity distribution. This is shown by the relatively constant values for the flow
uniformity (0.64 < ηm < 0.75, 0.89 < ηw < 0.92) in Figs. 4.13 through 4.16.
Furthermore, a good agreement is expected between pulsating distributions
and the limit case of zero engine speed, which corresponds to the stationary
averaged distribution. This is confirmed by the high values of rS in Table 4.3 for
manifold A. Table 4.3 quantifies the similarity using the shape and magnitude
similarity measures rS and rM , defined in Sect. 4.3.2. In addition, Table 4.3
provides the 95 % confidence intervals for rS and rM , as well as the P-values
for the hypothesis tests HS,1 (4.16), HS,2 (4.19) and HM (4.40) described in
Sect. 4.3.2.
The velocity distributions show no substantial difference resulting from us-
ing the rotating valve or the cylinder head. This is true for both the time-
averaged velocity distribution in pulsating flow Upuls and the stationary flow
distributions Ustat,r and Ustat . This resemblance is quantified in Table 4.2,
using the correlation coefficient defined as Eq. (4.12).
Manifold B (see Table 2.1)

Figures 4.17 and 4.18 show velocity distributions Ustat,r for manifold B, ob-
tained for stationary flow through each of the four runners, for two different flow
rates Qref . The stationary averaged distributions Ustat according to Eq. (4.1)
are shown in Fig. 4.19. For manifold B, only the cylinder head has been used
as pulsator. The reason for this is twofold. (i) There is a good agreement

6 6
60 60
5 5
30 30
1
4 1 4
0.5
0.5
1
3
y (mm)
y (mm)
1
1
4
2
2
0 3 0 3
1
0.5
2
1
2
1 1
2 1 2
-30 -30
0.5
1
1 1
1
1
1
1 1
-60 -60
0 0
5 5
Uy=0 (-)
Uy=0 (-)
3 3
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)

6 6
60 60
5 5
1
30 4 30 4
1 1
1
y (mm)
y (mm)
1
0 1 3 0 3
2
1
1
0.5
1
1
1
1
0.5
2
1
2 2
1
-30 -30 1
1
1
1 1
-60 -60
0 0
5 5
Uy=0 (-)
Uy=0 (-)
3 3
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(c) (d)
(a, b, c, d: r = 1 . . . 4) for manifold B on the isothermal flow rig with cylinder head
(Qref ' 65 m3 /h)

6 6
60 60
5 5
1
1
2
30 0.5 30
4 1 4
1
0.5
0.5
y (mm)
y (mm)
2
1
4
3
3
0 3 0 3
6
1
3
3
4 5
0.5
1
2
1
1
0.5
-30 2 -30 2
1
2
1
1
1 1
-60 -60
0 0
5 5
Uy=0 (-)
Uy=0 (-)
3 3
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)

6 6
60 60
5 5
1
30 1 4 30 4
2
1
0.5
y (mm)
y (mm)
1
2
0 3 0 3
3
3
4
2
1
1
1
0.5
2 2 2
-30 -30 1
1
5
0.
1
1 1
-60 -60
0 0
5 5
Uy=0 (-)
Uy=0 (-)
3 3
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(c) (d)
(a, b, c, d: r = 1 . . . 4) for manifold B on the isothermal flow rig with cylinder head
(Qref ' 130 m3 /h)
Stationary averaged velocity U (-) Stationary averaged velocity U (-)

2.5 2.5
N = 0 rpm, Qref ~ 64.6 m3/h N = 0 rpm, Qref ~ 131.7 m3/h
60 60
2 2
0. 1
1 8 1.2 0.8
1.2
30 30 0.9
1
0.8
0.9
1
0.8
0.9
0.8
0.9
1
0.8
0.9
1.5 1.5
0.9
0.7
1.4
1.2
1.2
1
.4
1.6
y (mm)
y (mm)
1.6 1
42
11..68 0.7
0.9
1.1.
8
0.9 0.8
0.7
0.
0.8
0.7
0.6
0 0
2.2 2
1.2
1.8
1 0.8
2.4
1.6
0.8
1.6
2.2
1
1
2
0.9
1
1.4
0.9
0.70.8
1 1
1.41.20.9
0.9
1.4
1.2
0.9
2 0.9
0.9
1.
0.8
-30 -30 1.2
0.9
1
1 0.8
0.9
1 1.21
0.5 0.5
-60 -60
Um ~ 1.645 m/s, ηm = 0.445, ηw = 0.919 Um ~ 3.329 m/s, ηm = 0.407, ηw = 0.877
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
Figure 4.19 – Stationary averaged velocity distribution Ustat according to Eq. (4.1)
for manifold B on the isothermal flow rig with cylinder head, for (a) Qref ' 65 m3 /h
and (b) Qref ' 130 m3 /h

2.5 2.5
60 60
2 2
1
0.9 1 0.
8
1
1
1.2
1
30 30
1
0.9
0.9
1
0.9
0.8
0.9
1.5 1.5
1
1
y (mm)
y (mm)
42
0.9
11..68
1.1.
1
0.7
0.8
0 0
2.2 2
1.2
1 0.9
1 0.8
1.2
0.8
1.2
1
0.9
1
1
1.4
0.9
1
1 1
1
1.4
1.2
0.9
1.2
0.9
0.9
2
0.9
-30 -30 1.
0.9
1
1
1
1
0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isothermal flow rig with cylinder head, at very low scavenging number
(S = 0.349, rS = 0.448, rM = 1.547)

2.5 2.5
60 60
0.9 2 2
0.8
1
1
1
0.9
0.
8
30 30 1.2
0.9
0.9 1
1
0.
1
1
9
1.5 1.5
10.9
1.2
1.2
y (mm)
y (mm)
0.9 1
1.2
0 0
0.8
0.9
0.8
1.4
1.2
1.4
0.9
1.6
1
1.2
0.9
1
0.9
1.2
1 1
1
0.9
0.8
0.9
1.2
1
0.9
-30 -30 0.9
0.8
1
0.9
1
1 0.9 1
1 1 0.9
0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isothermal flow rig with cylinder head, at low to moderate scavenging
number (S = 0.806, rS = 0.741, rM = 1.205)

2.5 2.5
60 60
0.8 2 2
1 0.
1.2
1 0.9 8
1.2
30 1 30
1
0.9
1
0.9
1.2
0.9
0.8
0.9
1.5 1.5
0.9
1
y (mm)
y (mm)
42
1.6 11..68
1.1.
0.7
0.8
1.8
0 0
2.2 2
1.4
1.2
1 0.8
0.8
0.9
1.2
1
0.8
0.9
1
1
1.4
0.9
1 1
1.2
0.9
0.9
1.2
1.2
0.9
0.9 .8
2
0.9
-30 -30 1.
0.9
1
0
0.9 1
1
0.9 1
1
0.8 1
0.7 0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isothermal flow rig with cylinder head, at moderate scavenging number
(S = 1.202, rS = 0.782, rM = 1.233)

2.5 2.5
60 60
0.9 2 2
0.9
0.9
0.
0.9
0.9 1 7
1
0.
0.9
8
0.9
30 1 30
0.8
0.9
0.9
0.8
0.8
1
0.7
0.7
0.8
1.5 0.8 1.5
0.8
0.9
0.7
0.10.
1.4
1.4
y (mm)
y (mm)
1.6
91.82
1.4
1.6
1.2
0.8
0.6
2.2
0 0
2 2.2
1.8
1.61.2 1.4 80.9
1
1.2
0.7
2.4
1.2
1.4
2.4
1
1
0.7
1.6 1.4
0.9
0.8 0.8
1
1 1
0.9
1.2
0.9
0.8
0.8
1.8
0.9
0.7
1.4
1
0.
1.2 1.4
1.2
-30 -30
00.8
0.
0.8 8
.9
1
0.9
1
0.9
1
1
0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isothermal flow rig with cylinder head, at high scavenging number
(S = 2.090, rS = 0.965, rM = 1.076)

2.5 2.5
60 60
1 2 2
0.9
0.8
0.9
0.9 0.
0.9
9 1 7
0.
0.9
30 30
0.8
0.9
1
0.70.9
0.9
1
0.8
0.9
1
0.7
0.7
1.5 0.8 1.5

0.8
0.8
1
1.2
0.8
1.2
0.10.
1.4
0.7
y (mm)
y (mm)
1.4
1.6
91.82
1.4
1.6
1.2
1.
0.6
0.7
2.2
4
0 0
1.8
1
0.7
1.2
2.4
1.4
1.61.4
2.4 0.8
1.6 1.4
22
0.8
1.8 2.
0.6 0.7
0.8
1 1
0.9
0.8
1.2
0.8
0.9
0.9
1
0.7
1.
1
4
1.2
0.9
1.4
1.2
-30 -30
00.8
0.
0.9
1 8
.9
1.12
0.
1 0.98 0.9
1
1
1
0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isothermal flow rig with cylinder head, at very high scavenging number
(S = 4.142, rS = 0.974, rM = 1.091)
Table 4.2 – Comparison between velocity distributions for manifold A using rotating
valve (left values) and cylinder head (right values)
Case N ηm ηw rS Figure
rpm - - -
Ustat 0/0 0.61/0.58 0.87/0.88 0.957 4.12
Upuls , S ' 4 1440/1440 0.69/0.67 0.90/0.90 0.948 4.14, 4.16
Upuls , S ' 3 2150/2010 0.69/0.69 0.90/0.90 0.952
Upuls , S ' 2 2810/2810 0.69/0.75 0.90/0.92 0.949 4.13, 4.15
between the velocity distributions obtained for manifold A with rotating valve
and cylinder head (see Table 4.2). This demonstrate that there is little ex-
tra information to be obtained from using the rotating valve. (ii) Using the
cylinder head on the isothermal flow rig poses no appreciable difficulty, other
than the need for forced lubrication and a more powerful electric motor (see
Sect. 2.2.1).
Figures 4.20 through 4.24 show comparisons between the time-averaged ve-
locity distributions Upuls and the corresponding stationary averaged distribu-
tions Ustat at the same volumetric flow rate. Figures 4.20 through 4.24 range
from low to high scavenging number. The scavenging number for Figs. 4.23
(S ' 2) and 4.24 (S ' 4) correspond to the values for Figs. 4.13 through 4.16.
The similarity measure rS exhibits comparable high values. The values for rM
agree not so well. It will be shown in Sect. 4.5 that the statistical evidence
provided by rM is not as strong as rS .
As shown in Table 4.3 summarizing the isothermal flow rig experiments, a
greater variation of the scavenging number S is obtained during the measure-
ments on manifold B: S varies between 0.23 and 4.5, whereas S varies only
between 2 and 4.5 for manifold A. The cases for low (S < 0.5) and moderate
scavenging number (0.75 < S < 1.5) are discussed in Sect. 4.4.2, where these
are compared to measurements on the isochoric flow rig.
4.4.2 Isochoric flow rig

Manifold B (see Table 2.1)
Figures 4.25 through 4.27 present comparisons between time-averaged velocity
distributions obtained on the isochoric (CME) flow rig (see Sect. 2.2.2) and the
stationary averaged distribution for the same volumetric flow rate.
All experiments on the CME flow rig use the oscillating hot-wire anemome-
ter discussed in Chap. 3. This is important due to the occurrence of flow
reversal. Without the use of the OHW, the velocity would be significantly
overestimated using standard hot-wire anemometry. The OHW is validated in
Sect. 5.2.1.
For Figs. 4.25 through 4.27, the scavenging numbers respectively corre-
sponds to the scavenging numbers of Figs. 4.20 through 4.22 obtained on the
isothermal flow rig, using the same manifold B.

2.5 2.5
N = 1200 rpm, Qref = 70.9 m3/h, pi = 1.55 atm N = 0 rpm, Qref ~ 64.6 m3/h
60 60
2 2
1
1 1
0.9 0.
1.2
1 8
1.2
0.8
30 30
1
1.
2
1
0.9
0.9
0.8
0.9
1.4
1.5 1.5
0.9
1
y (mm)
y (mm)
1.2
42
0.8
11..68
1.4
1.4
1.1.
0.9
0.80.9 1
0.7
0.8
1.6
1
1.8
0 0
2.2 2
1.2
0.9
1 0.8
0.8 1 0.9
0.8
1
1
1.2
0.9
1
1.4
0.9
1.2
1 1
0.9
0.7
0.8
0.9
1.2
0.9
2
0.9
-30 -30 1.
0.9
1
1
1 1
0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isochoric (CME) flow rig, at very low scavenging number (S = 0.316,
rS = 0.341, rM = 1.192)

2.5 2.5
60 60
2 2
1
1.2 0.8
0.9
1.2
30 30 0.9
0.8
1
1
0.8
0.8
0.8
0.9
1
0.8
1.4
0.9
1.5 1.5
0.9
0.7
1.4
1.2
1.2
1.2
0.7
1.8
.4
1.6
y (mm)
y (mm)
0.9
1.6 1
1.6
0.7
0.9
1
1.2
0.8
8
0.9 0.8
1.4
2
0.
0.7
0.6
0 0
1.6
1.8
0.9
0.8
2.4
1.6
1.6
1.
0.8
2.2
1
1.21
2
4
0.70.8
0.8
1.4 1 1
1.41.20.9
0.
1 .4
7
1
0.9
0.8
-30 -30 1.2

1 .9
0.9
0.9
0
0.9 0.8
1 1.21
0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isochoric (CME) flow rig, at moderate scavenging number (S = 0.819,
rS = 0.661, rM = 1.206)

2.5 2.5
60 60
2 2
1 1
0.9
1.2
0.8
0.8
0.9
0.9
30 30
1
1.2
0.7 0.7
1.4
1.41.6
0.8
1.21. 1.6
1
1.5 0.7 1.5
1.2
0.8
0.8
y (mm)
y (mm)
1.2
1.4
1
0.8
0.9
4
1.6
1
0.8
0.9
0.8
1.6
1.2
0.7
1.4 1.6
0.7
0 0
0.9
0.6
0.9
2.4
0.6
1.2 0.9
1.8
0.9
1 0.8
1.4
1.6
1.4 11.2 0.9
0.8
0.8
2.2
0.6
0.9
1 1
0.7
1
2.6
1.4
0.8
1.
0.8
0.9
8
-30 0.7 -30 0.8
0.8 1.2
1
1 1
1.2 1 0.9
1.2
0.5 0.5
-60 -60
0 0
2 2
Uy=0 (-)
Uy=0 (-)
1 1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
manifold B on isochoric (CME) flow rig, at moderate to high scavenging number
(S = 1.036, rS = 0.646, rM = 1.252)
In spite of the different flow rigs, the different engine speeds and flow rates,
each pair of figures with comparable scavenging number (S ' 0.3 for Figs. 4.25
and 4.20, S ' 0.7 for Figs. 4.26 and 4.21, S ' 1 for Figs. 4.27 and 4.22)
yield comparable values of the shape similarity measure (rS = 0.341 and 0.450,
rS = 0.661 and 0.781, rS = 0.646 and 0.782). For the magnitude similarity
measure, the agreement is less convincing (rM = 1.192 and 1.549, rM = 1.206
and 1.205, rM = 1.252 and 1.233).
4.4.3 Summary of the results

Tables 4.3 and 4.4 summarize the experiments on the isothermal and isochoric
flow rig, respectively. These experiments are used for the validation of the
addition principle.
Tables 4.3 and 4.4 provide the 95 % confidence limits for rS and rM . The
respective confidence intervals are [rSlow , rShigh ] and [rM
low high
, rM ]. Due to the spe-
cific statistical inference discussed in Sect. 4.3.2, these limits are asymmetrical.
Furthermore, the table gives the P-values for the hypothesis tests described
in Sect. 4.3.2. Recall from the definition of the hypothesis tests (Eqs. (4.16),
(4.19) and (4.40)) that the interpretation for the P-values differs:
• PS,1 < 0.05 (= α) corresponds to a significantly positive correlation co-

efficient.
• PS,2 < 0.05 corresponds to a significantly lower correlation coefficient

than the arbitrary high value ρcrit (= 0.9).
• PM < 0.05 corresponds to a value of rM significantly different from unity.
In summary, significant similarity corresponds to a low value of PS,1 , a high

value of PS,2 and a low value of PM . The P-values in Tables 4.3 and 4.4 that
bear evidence against similarity are printed in italic.
Background: How to interpret P-values For a general statistical

hypothesis test H, the P-value P quantifies the strength of the statistical
argument against the null hypothesis H0 and in favor of the alternative
hypothesis H1 . The smaller P , the stronger the evidence. The P-value is
a continuous measure, and the hypothesis tests should not be regarded
as discrete (similarity versus no similarity) tests. Instead, the following
interpretation may be used:
0.10 < P No evidence against the null hypothesis.
0.05 < P < 0.1 Weak evidence against the null hypothesis, in
favor of the alternative.
0.01 < P < 0.05 Moderate evidence against the null hypothesis, in
0.001 < P < 0.01 Strong evidence against the null hypothesis, in
P < 0.001 Very strong evidence against the null hypothesis,
in favor of the alternative.
The values listed in Tables 4.3 and 4.4 should be regarded with this in
mind.
Tables 4.3 and 4.4 do not explicitly mention the relative uncertainties for the
other variables (N , Qref and S). The uncertainty on the engine speed δN is
negligible. As mentioned in Sect. 4.3.3, δS ' δQref , which is approximately
5 % as derived in Sect. 2.4.1.
Tables 4.3 and 4.4 list both the old and new versions of the scavenging num-
ber (S old and S), using the different definitions of the apparent pulsation period
Tp according to Eq. (4.44) and Eq. (4.46), respectively. Since the isothermal
flow rig lacks a blowdown phase, the new and old definitions of Tp yield nearly
identical results in Table 4.3. The deviation on S is below 0.1, except in one
case featuring strong Helmholtz resonances, as discussed in Sect. 5.3. Since the
CME flow rig features a two-stage (instead of a single pulse) exhaust stroke,
the new definition of Tp according to Eq. (4.46) is typically half of the old value
according to Eq. (4.44), which explains the approximate difference by a factor
two between S old and S in Table 4.3.
Based on Table 4.3, some observations can be made with regard to the
geometrical differences of manifolds A and B. The catalyst cross-section of
manifold B is considerably greater than for manifold A. For the same flow
rate, the mean velocity is inversely proportionally smaller for manifold B. Due
to the lower velocity and the longer substrate (see Table 2.1), the residence
time of the exhaust gas in manifold B’s catalyst is greater. Consequently, so
is its conversion efficiency. On the other hand, the greater cross-section makes
it more difficult to obtain a good flow uniformity. This is clearly shown in
Figs. 4.24 and 4.23. The difference in scavenging number is partly due to the
142
Table 4.3 – Summary of experiments on the isothermal flow rig
low high
Case† N pi Qref S old S rS rSlow rShigh PS,1 PS,2 rM rM rM PM Figure
rpm atm m3 /h - - - - - - - - - -
A, RV 1440 - 77.3 4.217 4.264 0.965 0.926 0.983 0.000 1.000 1.126 0.777 1.176 0.609 4.14
2150 - 78.2 2.953 2.996 0.966 0.930 0.984 " " 1.135 0.763 1.188 0.580
2810 - 75.7 2.118 2.134 0.960 0.917 0.981 " " 1.119 0.803 1.195 0.538 4.13
A, CH 1440 - 77.4 4.279 4.334 0.971 0.939 0.986 " " 1.153 0.760 1.222 0.558 4.16
2010 - 76.8 3.053 3.100 0.957 0.911 0.980 " " 1.190 0.800 1.251 0.445
2810 - 76.6 2.178 2.158 0.944 0.884 0.973 " " 1.282 0.824 1.281 0.353 4.15
B, CH 600 - 43.8 1.561 1.561 0.882 0.793 0.934 " " 1.056 0.643 1.350 0.801
610 - 61.5 2.163 2.158 0.884 0.797 0.935 " " 1.154 0.595 1.595 0.655
600 - 98.3 3.498 3.498 0.965 0.938 0.981 " " 1.055 0.567 1.687 0.866
600 - 116.4 4.142 4.142 0.974 0.953 0.986 " " 1.091 0.590 1.684 0.780 4.24
1200 - 43.3 0.770 0.806 0.741 0.571 0.849 " 0.910 1.205 0.725 1.356 0.476 4.21
1200 - 68.0 1.211 1.202 0.782 0.633 0.875 " 0.976 1.233 0.633 1.630 0.491 4.22
1200 - 99.6 1.772 1.777 0.956 0.920 0.976 " 1.000 1.068 0.569 1.671 0.835
1200 - 117.1 2.085 2.090 0.965 0.937 0.981 " 1.000 1.076 0.586 1.716 0.820 4.23
2000 - 67.2 0.718 0.748 0.724 0.547 0.839 " 0.870 1.444 0.704 1.633 0.368
2810 - 64.1 0.495 0.520 0.628 0.410 0.778 " 0.483 1.600 0.803 1.657 0.236
3600 - 73.3 0.440 0.220 0.463 0.196 0.666 0.001 0.057 1.584 0.814 1.712 0.182
4400 - 72.2 0.351 0.349 0.448 0.178 0.656 0.001 0.044 1.547 0.769 1.679 0.220 4.20
†
Symbols ‘A’ and ‘B’ denote manifolds A and B; ‘RV’ and ‘CH’ denote the use of the rotating valve and cylinder head as pulsators. The
value of pi is not applicable to the isothermal flow rig.
Chapter 4 Addition principle
Table 4.4 – Summary of experiments on the isochoric (CME) flow rig
4.4 Experimental results
low high
Case† N pi Qref S old S rS rSlow rShigh PS,1 PS,2 rM rM rM PM Figure
3
rpm atm m /h - - - - - - - - - -
B, CH 1200 0.98 50.2 0.893 0.183 0.122 -0.178 0.401 0.213 0.000 1.151 0.780 1.301 0.468
1200 1.55 70.9 1.261 0.316 0.341 0.053 0.577 0.011 0.006 1.192 0.613 1.428 0.638 4.25
1800 0.97 79.7 0.946 0.478 0.334 0.045 0.572 0.012 0.005 1.201 0.627 1.504 0.575
1200 2.21 97.3 1.731 0.583 0.554 0.311 0.729 0.000 0.218 1.180 0.571 1.547 0.649
1800 1.56 110.6 1.312 0.439 0.514 0.260 0.702 " 0.127 1.223 0.553 1.525 0.658
1800 1.55 104.1 1.235 0.413 0.525 0.274 0.710 " 0.148 1.228 0.598 1.587 0.585
1800 2.15 137.0 1.625 0.819 0.661 0.456 0.799 " 0.626 1.206 0.668 1.472 0.497
1800 2.20 146.5 1.739 0.879 0.649 0.439 0.792 " 0.574 0.998 0.588 1.378 0.992 4.26
2400 1.53 187.7 1.670 0.830 0.615 0.392 0.769 " 0.426 1.147 0.634 1.466 0.637
2400 2.00 234.6 2.088 1.036 0.646 0.435 0.790 " 0.559 1.252 0.701 1.469 0.410 4.27
3000 1.53 237.4 1.690 0.838 0.598 0.369 0.758 0.000 0.361 1.267 0.714 1.449 0.402
143
large diffuser volume of manifold B. Furthermore, manifold B features four

instead of three runners, which implies a smaller pulsation period Tp for the
same engine speed.
Table 4.3 demonstrates a monotonous behavior of rS versus S in the isother-
mal flow rig experiments. This is not entirely present in the isochoric flow rig
experiments (see Table 4.4). As a particular example, since the scavenging
number is greater for Fig. 4.27 (S = 1.036) than for Fig. 4.26 (S = 0.819), one
would expect the value of rS to be equally greater. This is not the case. How-
ever, as indicated in Table 4.4, the confidence limits for rS become increasingly
wide as the value of rS decreases. This is due to the skewed sampling distribu-
tion of a positive correlation coefficient (see Sect. 4.3.2). Given the fact that
rS decreases for decreasing S, the confidence limits on rS inevitably increase
as S decreases.
Overall, the combined results presented in Sects. 4.4.1 and 4.4.2 indicate
that the scavenging number S is correlated to the degree of similarity between
steady and pulsating distributions, and thus to the addition principle’s validity.
Upon comparing the stationary averaged and pulsating distributions Ustat and
Upuls in Figs. 4.13 through 4.27, it shows that the magnitude similarity measure
rM is consistently greater than unity. Given its definition (4.28), the flow
uniformity is consistently higher for pulsating than for stationary flow. The
quantification and statistical significance of the similarity based on shape (rS )
and magnitude (rM ) are presented in Table 4.3 for the isothermal flow rig and
Table 4.4 for the isochoric (CME) flow rig.
In the CME flow rig experiments (see Table 4.4), the scavenging number
varies roughly between 0.2 and 1, whereas the scavenging number in the isother-
mal flow rig experiments (see Table 4.3) varies roughly between 0.2 and 4.5.
Both for a fired engine and the CME flow rig, the catalyst volumetric flow rate
Q increases linearly with the engine speed N , the intake manifold density ρi and
the exhaust gas temperature Te , at least as a first approximation. Combining
this with Eqs. (4.43), (4.44) and (4.42) shows that S is nearly independent of N
and varies mainly with engine load. S increases for a variation from zero to full
engine load. By contrast, the isothermal flow rig allows the pulsation frequency
(i.e. the engine speed) and flow rate (i.e. the engine load) to be controlled inde-
pendently, thus enabling a wider range of scavenging number than physically
possible in a fired engine. The scavenging number does not differ significantly
between fired engine conditions and the CME flow rig (see Sect. 2.3).
4.5 Interpretation of the results

The data used in Figs. 4.28 and 4.29 correspond to the values in Tables 4.3
and 4.4. Figure 4.28 shows the non-dimensional correlation of both similarity
measures rS and rM versus the old definition of the scavenging number S old ,
using Eq. (4.44). The CME flow rig experiments are plotted as crosses ( ),
whereas the other markers ( , , ) represent experiments on the isother-
mal flow rig. In the figure legends, ‘A’ and ‘B’ denote manifold types A and B,
4.5 Interpretation of the results 145
1
1.6 ISOT, A, RV
0.9 ISOT, A, CH
ISOT, B, CH
Magnitude similarity measure rM (-)

0.8 CME, B, CH
Shape similarity measure rS (-)
1.5 rM’
0.7
rM’ = 1.104 + 0.879 exp(- Sold / 0.621)
0.6 1.4
old
(R2 = 0.82)
rS’ = 1 - exp(- S / 0.621)
0.5 2
(R = 0.96) 1.3
0.4
0.3 1.2
ISOT, A, RV
0.2 ISOT, A, CH
ISOT, B, CH 1.1
CME, B, CH
0.1 rS’
0 1
0 1 2 3 4 5 0 1 2 3 4 5
Scavenging number Sold (-) Scavenging number Sold (-)
(a) (b)
Figure 4.28 – Correlations of similarity measures (a) rS and (b) rM versus scaveng-
ing number S old , using the old definition of Tpold (4.44)
1
1.6 ISOT, A, RV
0.9 ISOT, A, CH
ISOT, B, CH
0.8 CME, B, CH
1.5 rM’’
0.7
rM’’ = 1.118 + 0.337 exp(- S / 0.723)
0.6 1.4
(R2 = 0.30)
rS’’ = 1 - exp(- S / 0.723)
0.5 2
(R = 0.91) 1.3
0.4
0.3 1.2
ISOT, A, RV
0.2 ISOT, A, CH
ISOT, B, CH 1.1
CME, B, CH
0.1 rS’’
0 1
0 1 2 3 4 5 0 1 2 3 4 5
Scavenging number S (-) Scavenging number S (-)
(a) (b)
ing number S, using the new definition of Tp (4.46)
and ‘RV’ and ‘CH’ denote rotating valve or cylinder headas pulsator.
The exponential correlation fits rS0 = 1 − exp −S old Scrit,S
0 0

and rM =
0 0 0
old

rM,∞ + cM exp −S Scrit,M are least square fitted to the isothermal flow
rig experiments, excluding the experiments on the CME flow rig.
Figure 4.29 shows the correlation versus the new definition of the scavenging
number S, using Eq. (4.46). The points corresponding to the isothermal flow
rig experiments remain roughly unchanged with respect to Fig. 4.28. However,
the crosses ( ) representing the CME flow rig experiments now correlate very
well with the isothermal flow rig experiments. Correspondence is remarkably
good for rS yet only moderate for rM . Indeed, the correspondence seems to
even slightly deteriorate for rM .
In Fig. 4.29b, the crosses ( ) representing the CME flow rig cases appear
to deviate from the isothermal flow rig cases. Due to the greater uncertainty
on rM and the limited range of S obtained of the CME flow rig, the deviation
may not be considered very significant. Nevertheless, further experiments are
clearly required to explain this deviation.
The form of the exponential correlation fits is altered into rS00 = 1−
00 00
exp (−S/Scrit ) and rM = rM,∞ + c00M exp (−S/Scrit ). The critical value Scrit
for the rM correlation is set equal to the value obtained from the rS correlation.
The resulting correlations for the similarity measures rS and rM versus the
scavenging number S are:
 00
 rS = 1 − exp (−S/0.723 ) ; R2 = 0.91
(4.47)
 00
rM = 1.118 + 0.337 exp (−S/0.723 ) ; R2 = 0.30
where the critical value of the scavenging number Scrit = 0.723. These cor-
relations combine the CME flow rig experiments with the isothermal flow rig
experiments, obtained for two types of exhaust manifold with and without
exhaust valve overlap, and for a rotating valve and cylinder head as pulsator.
Regarding the statistical significance of rS , the limits in Eq. (4.25) for signif-
icant similarity based on hypothesis test HS,1 (4.16) hold for most experiments
listed in Tables 4.3 and 4.4. Indeed, the P-values PS,1 are smaller than the
significance level α = 0.05, with one exception.
The hypothesis test HS,2 (4.19) tests whether the observed value of rS is
significantly smaller than a ‘high’ correlation value ρcrit . In Sect. 4.3.2, this
critical value is arbitrarily chosen as ρcrit = 1 − e−1 ' 0.63. This choice
becomes clear in light of the remarkable correlation fit between rS and S. ρcrit
corresponds to the expected correlation coefficient when the scavenging number
S equals its critical value Scrit .
The P-values PS,2 in Tables 4.3 and 4.4 are mostly larger than 0.05, indi-
cating weak or no evidence against the null hypothesis, i.e. the values of rS are
not significantly lower than ρcrit . For a very low scavenging number (S . 0.3,
rS . 0.5) PS,2 is smaller than the significance level, which indicates evidence
against similarity.
Concerning the P-values PM of hypothesis test HM (4.40), a value of PM
greater than 0.05 indicates there is no statistically significant deviation between
4.5 Interpretation of the results 147
Table 4.5 – Monte Carlo-based 95 % uncertainty estimate for the fit parameters in
Eq. (4.47) and Figs. 4.29 and 4.28
Parameter Value Error Figure
- -
Using the new definition of Tp (4.46):
Scrit 0.723 ± 0.052 4.29
00
rM,∞ 1.118 ± 0.057
c00M 0.337 ± 0.149
Using the old definition of Tp (4.44):
0
Scrit 0.621 ± 0.069 4.28
0
rM,∞ 1.104 ± 0.061
c0M 0.879 ± 0.284
the value of rM and unity. In other words, if PM > 0.05, similarity cannot be
overruled. In Tables 4.3 and 4.4, values for PM are consistently greater than
0.05. On the other hand, rM is always larger than unity (with one exception),
indicating that the flow uniformity is higher in pulsating than in steady flow
conditions. Nevertheless, the HM (4.40) hypothesis test does not indicate a
significant difference between steady and pulsating flow distributions.
The virtual absence of statistical evidence based on rM is a problem which is
difficult to overcome. Instead of the present definition of rM in Eq. (4.28) which
is based on the mean-to-maximum velocity ratio ηm (4.6), other definitions
have been tried without success. In particular, defining rM as the ratio of two
Weltens’ uniformity indices ηw (4.2) leads to much worse results in terms of
the correlation fit in Fig. 4.29b and the statistical strength.
The evidence based on rM is weaker when compared to rS . Yet strictly
statistically speaking, the addition principle is valid for almost the entire range
of S. Taking into account Eq. (4.47), it seems however more appropriate to
state that the addition principle is valid when S exceeds the critical value
Scrit = 0.723, corresponding roughly to rS > 1 − e−1 = 0.63 and rM < 1.24.
The dashed lines in Figs. 4.29 and 4.28 correspond to the 95 % confidence
bounds on the fitted curve. The estimated uncertainties on the fit parameters
are shown in Table 4.5. These errors are determined using a Monte Carlo-
like simulation. A total of 1024 sets of S, S old , rS , rM are constructed using
a normally distributed random number generator. Each set is a collection of
randomized observations of S + ∆S, S old + ∆S, rS + ∆rS and rM + ∆rM ,
as listed in Tables 4.3 and 4.4. According to Sect. 4.3.3, the uncertainty ∆S
is assumed constant, and equal to 0.05. The uncertainties ∆rS and ∆rM are
determined according to the sampling distribution of each quantity, as discussed
in Sects. 4.3.2 and 4.3.2.
For each of the 1024 randomized sets, the corresponding curves are fitted.
00
This results in a collection of 1024 values for each fit parameter Scrit , rM,∞ , c00M
0 0 0
and Scrit , rM,∞ , cM . Based on these sampling distributions, the uncertainty on
each parameter is estimated, resulting in the values in Table 4.5.
638
Fig. 14a, b. Non-uniformity index against non-dimensional param-

Fig. 13. a Inlet pulse shape from ensemble averaged velocity and eter, J for a 180 diffuser and b 60 diffuser using 152 mm substrate
Figure 4.30 – Non-uniformity index versus J (Note: cases (a) and (b) are for
b velocity profiles at the rear of the substrate at 100 Hz
different diffusers) (Source: [17])

flow path is closed resulting in intermittent wall jets at
3.5
higher frequencies. Non-dimensional correlations
The reasons for the increase in flow uniformity at
The observation of improved flow profiles as Re reduces
higher frequencies are possibly two fold. First, it would
and frequency increases suggests that there is a functional
seem that at high frequency the flow does not have suffi-
relationship between the maldistribution index and these
cient time to establish the inertia dominated steady flow
variables. Clearly the length of the diffuser is also impor-
As mentioned in the tant literature
regimes associated with high Reynolds number flows, that survey
as mentioned above. (Sect.
To examine 1.4.1),
if general relation-the present results are
is separation at the throat and large recirculation zones
ships could be derived two additional 180 diffusers were
confirmed to some extent by
within the diffuser volume. At low frequency, separation at findings from other researchers.
tested. The standard diffuser length (L0) is 61.5 mm long Benjamin et al.
the diffuser inlet may be more likely at the peak of the
and the additional diffusers were of lengths 79.85 (L1) and
[17] define the non-dimensional
pulse as the flow is quasi-steady and hence the flow mal- number J as the ratio
119.15 mm (L2). Velocity profiles for L1 and L2 were taken
of pulsation period to
diffuser residence time.for high
By Recontrast
distribution may be expected to be similar to that under
steady flow conditions as shown in Fig. 6. Second, the
in the range to the definition
of 70,000–110,000 in this thesis (4.43), the
for all the
frequencies. The results have been plotted in Fig. 14a as run
residence time is defined
enhanced mixing within the diffuser at higher frequencies basedindex
non-uniformity onagainst
the themean runnerpa- velocity Um
non-dimensional and the
could also produce flatter profiles. At low values of J there
rameter, J. The results collapse reasonably well when
diffuser length L :
is the possibility of more than one pulse residing ind the
plotted in this form. Figure 14b shows the data for the 60
diffuser volume at any one time leading to pulse interac-
diffuser and 152 mm substrate plotted similarly. Such
tion and increased mixing. Increasing the residence time
correlations could form a useful basis for design engineers
faced with1/f
in the diffuser (say by increasing the diffuser length) the task of reducing flow maldistribution in
J autocatalyst
=
would increase the probability of such an occurrence. systems.
run
Benjamin et al. [17] (4.48)
There is also a possibility that wave dynamics could be Ld /Um
contributing to the increased mixing at higher frequencies.
As such, this definition4Conclusion
However, pressure measurements recorded 30 mm
assumes that a free jet is established in the ‘diffuser’, and
downstream of the rear face of the catalyst (P2 in Fig. 11a)
that no actual diffusion Thistakes
do not indicate any strong reflections from the open end of examined J
study hasplace. the is nevertheless
effect very similar to the scav-
of pulsating flow within
automotive catalyst systems. The flow distribution as ex-
the sleeve and it is therefore concluded that such effects
are probably secondary. enging number S (4.42). hibitedFigure
by velocity 4.30
profiles shows a correlation
at the substrate exit has been of a non-uniformity
measure versus J. Upon inverting the y-axis values from a non-uniformity
into a uniformity measure, Fig. 4.30 compares qualitatively to the evolution
of rM in Fig. 4.29b. Indeed, rM equals by definition (4.28) a flow uniformity
measure (based on the mean-to-maximum velocity ratio ηm ) in pulsating flow,
relative to the corresponding uniformity measure in steady flow. Therefore, the
increasing flow uniformity in pulsating flow for decreasing S is confirmed by
Fig. 4.30.
Bressler et al. [22] define GEN [-] as the ratio of exhausted gas volume per
cylinder per cycle Vexh [m3 ] to the diffuser volume Vd [m3 ]. Upon dividing
numerator and denominator of this ratio by the engine speed N , it becomes
clear that GEN is proportional to S, at least according to the old definition of
4.6 Discussion: A physical interpretation 149
Tpold (4.44):
Vexh 1/N
GEN = = ∼S Bressler et al. [22] (4.49)
Vd Vd /(N Vexh )
Although no correlation is given by Bressler et al. [22], their conclusions indi-
cate that the flow uniformity in pulsating flow is unaffected by engine speed
when GEN remains constant. Also, the authors [22] conclude that the flow
uniformity in pulsating flow is always higher than for steady flow. Further-
more, the deviation between the flow uniformity in pulsating flow and steady
flow is minimal for high values of GEN (i.e. high values of S) and increases for
decreasing GEN (i.e. low values of S).
4.6 Discussion: A physical interpretation

4.6.1 Scalar mixing analogy
Figure 4.29 and the correlations in Eq. (4.47) provide substantial evidence that
the addition principle remains valid under various conditions. The elegance of
Eq. (4.47), at least for rS , further supports that the scavenging number is the
correct choice of non-dimensional group to describe this flow.
A remarkable analogy seems to hold between the current problem and the
scavenging of a volume with a scalar quantity, e.g. a species concentration. For
instance, assume incompressible flow through a perfectly stirred volume V [m3 ]
with one inlet and one outlet, containing a volumetric concentration φ (vol %)
of some compound. The following partial differential equation describes the
concentration evolution in time inside the perfect mixing volume:
∂ (V φ) ∂φ
=V = Qφi − Qφ (4.50)
∂t ∂t
where Q is the volumetric flow rate [m3 /s] and φi is the inlet concentra-
tion (vol %). This corresponds to the following transfer function in the Laplace
domain:
φ 1
(s) = (4.51)
φi τs s + 1
where s is the Laplace variable (s = jω) [s−1 ] and τs is the scavenging time
constant [s], defined as τs = V /Q . The scavenging time τs may be regarded as
the mean residence time of fluid in the mixing volume. Assuming a stepwise
change in the inlet concentration φi at time t = 0, and an initial concentration
φ(t = 0) = 0, the solution to the above partial differential equation is:
φ
(t) = 1 − exp (−t/τs ) (4.52)
φi
Equation (4.52) expresses to what extent the volume is scavenged, as a
function of the non-dimensional time t/τs . If t/τs is sufficiently large, the
solution becomes independent of t/τs . The non-dimensional time corresponds

to the scavenging number S used to characterize the pulsating flow in the
exhaust manifold.
The remarkable correlation fit in Eq. (4.47) for rS suggests that this complex
multi-dimensional flow behaves essentially like a first order zero-dimensional
scavenging process of a scalar quantity.
The flow conditions in the mixing volume may be such that only part of
the volume takes part in the mixing process. For instance, the trace species
introduced at the inlet may not penetrate into recirculation zones that are
formed in corners or near a sudden expansion inlet. In that case, the mean
residence time τs,eff = Veff /Q decreases with respect to the perfect mixing
case, where τs = V /Q . The ratio of the residence time scales τs,eff /τs varies
between zero and unity, and is a measure of the quality of the mixing process.
The ratio τs,eff /τs can also be regarded as the ratio of effective to geometric
volume τs,eff /τs = Veff /V .
4.6.2 Hypothesis: Collector efficiency

In light of the above introduced analogy with a scalar mixing process, the criti-
cal scavenging number Scrit = 0.723 < 1 suggests that only part of the diffuser
volume may be active during the scavenging process. From the introduction of
an alternate scavenging number S 0 :
S Tp Tp
S0 = = = (4.53)
Scrit Scrit Vd /Q Vd,eff /Q
follows an effective diffuser volume Vd,eff = Scrit Vd . The velocity distributions
in Figs. 4.20 through 4.27 indeed indicate that some parts of the catalyst in
manifold B are subject to a very low flow rate, particularly the leftmost and
central areas. Since rM increases for decreasing S and S is inversely propor-
tional to the diffuser volume Vd , the flow uniformity in pulsating flow increases
for an increasing diffuser volume. As such, the ratio of the effective to actual
diffuser volume Vd,eff /Vd could be interpreted as a collector efficiency with
respect to catalyst flow uniformity:
Vd,eff
ηD = = Scrit (4.54)
Vd
The term collector efficiency should be regarded as the effectiveness of the
use of the diffuser volume in distributing the flow across the catalyst cross-
section. It is inappropriate to denote this effectiveness as the ‘diffuser’ efficiency,
since the shape of the exhaust runners plays a major role in the flow distribution
as well. Therefore, the term ‘collector’ is used, comprising the exhaust runners
and diffuser.
The collector efficiency ηD equals the critical scavenging number. In other
words, the higher the critical scavenging number Scrit , the more efficiently the
diffuser distributes the exhaust gas throughout the catalyst cross-section.
4.6 Discussion: A physical interpretation 151
1
ISOT, B, CH
0.9 1.6
CME, B, CH
rM’’

0.8
1.5
0.7
rM’’ = 1.084 + 0.403 exp(- S / 0.722)
0.6 1.4
(R2 = 0.32)
rS’’ = 1 - exp(- S / 0.722)
0.5
(R2 = 0.88) 1.3
0.4
0.3 1.2
0.2
ISOT, B, CH 1.1
0.1 CME, B, CH
rS’’
0 1
0 1 2 3 4 5 0 1 2 3 4 5
(a) (b)
ing number S, for experiments on manifold B
Table 4.6 – Monte Carlo-based 95 % uncertainty estimate for the fit parameters in
Eq. (4.55) and Fig. 4.31
Parameter Value Error Figure
- -
Using the new definition of Tp (4.46):
Scrit 0.722 ± 0.056 4.31
00
rM,∞ 1.084 ± 0.097
c00M 0.403 ± 0.207
The correlation in Eq. (4.47) and the resulting value for the critical scav-
enging number Scrit = 0.723 are obtained for the combined experiments on
manifolds A and B, using two pulsator devices and two types of flow rigs. If
the above assumption is correct, the critical scavenging number is a manifold
geometry-dependent measure, which should differ from manifold A to mani-
fold B.
Unfortunately, the scavenging number range (2 < S < 4.5) for the available
experiments on manifold A is, in retrospect, inappropriately chosen. Based on
the experiments on manifold A, no critical scavenging number can be obtained.
Since the scavenging number range is much wider for manifold B, the exper-
iments on manifold B contribute the most to the value of the critical scavenging
number obtained from the correlation in Eq. (4.47).
Figure 4.31 shows the correlations of the similarity measures rS and rM
versus the scavenging number using the new definition of Tp (4.46), similar to
Fig. 4.29, yet only including the experiments on manifold B. The fitted curves
are:
 00
 rS = 1 − exp (−S/0.722 ) ; R2 = 0.88
(4.55)
00
rM = 1.084 + 0.403 exp (−S/0.722 ) ; R2 = 0.32

Using the same Monte Carlo simulation approach as used previously in

Sect. 4.5, the uncertainty on the fit parameters is estimated and given in Ta-
ble 4.6.
For the experiments performed on manifold B, including the isothermal
and isochoric (CME) flow rig experiments, the collector efficiency ηD (= Scrit )
equals 0.722, with an estimated 95 % uncertainty of 0.056.
Further studies are needed to confirm the hypothesis of the existence of
a collector efficiency. In particular, the effect of the collector geometry (e.g.
shape of the exhaust runners, entrance angle of the runners in the diffuser)
on the critical scavenging number should be investigated using a number of
geometrical variants. This might be the focus of future research.
4.7 Conclusion
Chapter 4 investigates the validity of the addition principle (4.1) for pulsating
flow in two close-coupled catalyst manifolds A and B (see Sect. 2.1). The
addition principle states that the time-averaged catalyst velocity distribution
in pulsating flow Upuls equals a linear combination of velocity distributions
obtained for steady flow through each of the exhaust runners Ustat , according
to Eq. (4.1).
The results obtained on an isothermal and isochoric (CME) flow rig are
in good agreement, in spite of the appreciable difference in the pulsating flow
generated by both flow rigs (see Sect. 2.3.
The CME flow rig generates cold pulsating flow that resembles fired engine
conditions better than the isothermal flow rig, featuring a two-stage exhaust
stroke consisting of blowdown and displacement phases. The exhaust stroke
flow similarity between CME and fired conditions is incomplete. Nevertheless,
combined with Helmholtz resonances (see Sect. 5.3) intrinsic to the manifold,
the pulsating flow features a similar frequency content to fired engines [70, 81,
59]. The increased frequency content compared to the isothermal flow rig leads
to a introduction of a new definition of the apparent flow pulsation period Tp
in Eq. (4.46).
For the experiments on the isochoric flow rig, an oscillating hot-wire anemo-
meter (OHW) is used to measure bidirectional velocity, with a maximum mea-
surable negative velocity of −1 m/s. The effect of using the OHW cannot
be directly observed in the time-averaged velocity distributions shown in this
chapter. Nevertheless, the accuracy of the velocity measurements is greatly
improved over using standard hot-wire anemometry. The beneficial aspects of
the OHW for the isochoric flow rig experiments are discussed in Sect. 5.2.1.
The scavenging number S defined in Eq. (4.42) using the new apparent pul-
sation period Tp definition in Eq. (4.46) forms the appropriate non-dimensional
4.7 Conclusion 153
number to characterize the pulsating flow. The non-dimensional measures

rS (see Sect. 4.3.2) and rM (see Sect. 4.3.2) quantify the similarity between
the Ustat and Upuls distributions based on shape and magnitude, respectively.
These measures are used to quantify the validity of the addition principle.
The results from the entire measurement campaign are combined in Fig. 4.29
and Tables 4.3 and 4.4. Figure 4.29 shows the good correlation between the
similarity measures rS and rM and the scavenging number S. The validity
of the addition principle is quantified in Tables 4.3 and 4.4 in terms of the
statistical significance of rS and rM , as discussed by Sect. 4.3.2.
The correlations in Eq. (4.47) are least-square fitted to the values of rS , rM
and S:
 00
 rS = 1 − exp (−S/0.723 ) ; R2 = 0.91
00
rM = 1.118 + 0.337 exp (−S/0.723 ) ; R2 = 0.30

Strong statistical evidence is given in support of the addition principle, for

nearly the entire range of S. However, no clear validity limit can be derived
based on the statistical significance of rS and rM . Therefore, the practical limit
of the addition principle’s validity is when S exceeds the critical scavenging
number Scrit = 0.723 (± 0.052), corresponding roughly to rS > 1 − e−1 = 0.63
and rM < 1.24.
Other authors [17, 22, 99] have used non-dimensional numbers similar to
S to characterize the pulsating flow in close-coupled catalyst manifolds. How-
ever, the original contribution of this work is to relate S to the flow distribution
similarity between pulsating and stationary flow conditions using rS and rM ,
and furthermore to derive the validity of the addition principle from that rela-
tionship.
Based on the elegance of the rS correlation in Eq. (4.47), this complex
multi-dimensional flow behaves essentially like a zero-dimensional scalar mixing
process. In that respect, the critical scavenging number Scrit may be considered
the ratio of the effective to actual diffuser volume. As such, the hypothesis
is formulated that Scrit corresponds to a collector (i.e. runners and diffuser)
efficiency ηD with respect to catalyst flow uniformity. By maximizing the
collector efficiency ηD , the flow uniformity is optimized, and consequently so is
the catalyst durability, conversion efficiency and exhaust system backpressure.
The correlations in Eq. (4.47) are valid for two different exhaust manifolds. In
Sect. 4.6.2, the correlations in Eq. (4.55) are obtained only for the experiments
on manifold B. Based on these correlations, the critical scavenging number or
hypothesized collector efficiency ηD yields 0.722 (± 0.056).
Further investigations are required to determine (i) whether Scrit indeed
corresponds to a collector efficiency, and (ii) to what extent the collector effi-
ciency depends on the manifold geometry.
The contents of this chapter have been published in two international jour-
nals with review:
[86] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Exper-

imental validation of the addition principle for pulsating flow in
close-coupled catalyst manifolds. J. Fluids Eng.-Trans. ASME,
128(4):656–670, 2006. http://dx.doi.org/10.1115/1.2201646.
[88] T. Persoons, E. Van den Bulck, and S. Fausto. Study of pul-
sating flow in close-coupled catalyst manifolds using phase-locked
hot-wire anemometry. Exp. Fluids, 36(2):217–232, 2004. http://
dx.doi.org/10.1007/s00348-003-0683-0
Chapter 5
Flow dynamics
“A theory is something nobody believes, except the person who made it.
An experiment is something everybody believes, except the person who
made it.”
Albert Einstein (German-born American physicist, ◦ 1879, †1955)
This chapter focuses on the time-resolved 27 aspects of the flow in exhaust sys-
tems with close-coupled catalyst, whereas Chap. 4 is more concerned with the
time-averaged velocity distribution.
Using cold pulsating flow rigs in combination with hot-wire anemometry
yields detailed whole-field time-resolved velocity distributions in the close-
coupled catalyst. Section 5.1 discusses the resulting time-resolved distributions
for different pulsating flow rigs, exhaust manifolds and operating conditions.
Section 5.2 focuses on the spatial and temporal occurrence of periodic flow
reversal in the close-coupled catalyst. The bidirectional velocity measurements
are performed using the oscillating hot-wire anemometer (OHW) (see Chap. 3).
Sect. 5.2.1 validates the OHW in conditions where catalyst flow reversal is
known to occur. The validation is performed with respect to integral flow rate
measurements. Using a bidirectional velocity measurement technique such as
the OHW proves crucial for obtaining accurate measurements in the event of
flow reversal. All velocity measurements on the isochoric pulsating flow rig are
performed using the OHW.
The experimental data in Sect. 5.2.2 reveal the time-resolved velocity distri-
bution throughout the entire catalyst cross-section, including areas of negative
velocity.
27 The term ‘time-resolved’ denotes the (ensemble-averaged) time variation during one en-
gine cycle, or two crankshaft revolutions (i.e. 720 ◦ ca).
155
156 Chapter 5 Flow dynamics
Section 5.2.3 describes the use of a one-dimensional gas dynamic model

of the exhaust system to simulate the occurrence of flow reversal in terms of
the mean velocity. Appendix D discusses the gas dynamic model in detail.
Through numerical simulation, Sect. 5.2.3 establishes the influence on catalyst
flow reversal of the presence of the exit cone and cold end.
Section 5.3 analyzes the ubiquitous resonance phenomenon observed in both
flow rigs. The phenomenon is explained as a Helmholtz-type resonance, using
an analytical explanation in Sect. 5.3.2.
Section 5.3.4 uses the same one-dimensional gas dynamic model (see
App. D) to explain the resonance phenomenon numerically. Gas dynamic fre-
quency response functions of the exhaust manifold are determined, revealing
the nature of the resonating system responsible for these strong velocity fluc-
tuations in the catalyst.
5.1 Time-resolved flow distributions

This section describes the time-evolution of the catalyst velocity distribution.
The difference between the isothermal and isochoric flow rig is demonstrated
by plots of the time-resolved mean velocity Um (ωt), flow uniformity ηm (ωt),
and the time-resolved velocity distribution U (ωt) during the exhaust stroke of
the first cylinder.
The pulsating flow in the isothermal flow rig exhibits a much lower fre-
quency content compared to the isochoric flow rig, because of the single-stage
exhaust stroke. As such, the experiments on the isothermal flow rig reveal
some interesting details of the flow evolution, e.g. the phase lead by the flow
uniformity with respect to the pulsator device.
Resonance fluctuations are observed in the isothermal flow rig, using the
cylinder head as pulsator. Similar yet much stronger fluctuations are observed
on the isochoric flow rig. These are discussed in detail in Sect. 5.3.
The oscillating hot-wire anemometer (OHW) is used for all experiments on
the isochoric flow rig, thus enabling bidirectional velocity measurements. The
phenomenon of flow reversal in the catalyst is discussed in detail in Sect. 5.2.

Manifold A
Figures 5.1 and 5.2 show in the top plots the time-resolved dimensionless mean
catalyst velocity Um (ωt), where ωt is the crankshaft position. As noted earlier,
in case of the isothermal flow rig, this corresponds to the imaginary crankshaft
position corresponding to the position of the pulsator device (i.e. rotating valve
or cylinder head). The bottom plots in Figures 5.1 and 5.2 show the correspond-
ing time-averaged velocity distribution.
The crankshaft position ωt = θ (◦ ca) is defined relative to top dead center
of cylinder 1, prior to the intake stroke. As such, and taking into account the
firing order, the time-resolved mean velocity plots show the consecutive exhaust
5.1 Time-resolved flow distributions 157
Time-resolved Time-resolved
3 3
N = 1440 rpm, Qref = 77.3 m3/h N = 2810 rpm, Qref = 75.7 m3/h
2.5 2.5
Mean velocity, flow uniformity (-)

2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 Um -0.5 Um
ηm ηm
-1 -1
0 180 360 540 720 0 180 360 540 720
Crankshaft angle ω t (°) Crankshaft angle ω t (°)
Time-averaged velocity U (-) Time-averaged velocity U (-)

2.5 2.5
0.8
0.9 0.8
30 1 1.2 30 0.9 1.2
0.7 1 0.7
1.4 2 2
1.4
15 15
0.6
0.8
1.5 1.5
0.6
y (mm)
y (mm)
0 1 0
0.9
1
0.7
0.9
1.4
0.8
1.2
1 1
1.4
1.2
1.4
0.8
0.9
0.9 1
1
0.6
-15 -15
1.2
0.7
1.4 1.2
.9
0.6
0.9
1 .7
80
1.2 0. 0.81
0
0.5 0.5
0.7
-30 -30 0.8
0.7

0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)
Figure 5.1 – Time-resolved mean velocity Um (ωt) [-] (top) and corresponding time-
averaged distribution U [-] (bottom) for manifold A on the isothermal flow rig with
rotating valve, for (a) N = 1440 rpm and (b) N = 2810 rpm
3 3
2.5 2.5

2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 Um -0.5 Um
ηm ηm
-1 -1
0 180 360 540 720 0 180 360 540 720

2.5 2.5
0.60.9
0.7 0.7
30 0.8 1.21 30 .8
00.9
.4 1
2 2
1
0.7
1.2
15 15
0.7
1.5 1.5
y (mm)
y (mm)
0.6
0.8
0.8
0 0
0.7
1.
0.7
0.9
2
1.6 .4
1
1 1.4 1 1
1.4
0.9
1.2
1.4
0.91
0.6
-15 -15
1.2
0.8
0.8
1.2 1.2
0.9
0.9 0.5 1 0.5

1
1
-30 0.8 0.8 -30

0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
averaged distribution U [-] (bottom) for manifold A on the isothermal flow rig with
cylinder head, for (a) N = 1440 rpm and (b) N = 2810 rpm
strokes of cylinders 2, 3 and 1. The same convention is applied throughout the

thesis. The exhaust stroke of cylinder 1 therefore always corresponds to the
rightmost pulse in the time-resolved mean velocity plots, and varies roughly
between 480 and 720 ◦ ca.
In the upper plots, the solid line ( ) is the mean catalyst velocity Um (ωt).
The dashed line ( ) is the time-resolved flow uniformity ηm (ωt), as defined
by Eq. (4.9). The light solid line ( ) represents the dimensional exhaust valve
lift. For the rotating valve, this lift equals the valve open cross-sectional area,
divided by the maximum open area. Close examination of the sawtooth-like
lift curves show a small inter-cylinder28 exhaust valve overlap, corresponding to
the valve timing and the three exhaust runners per catalyst for manifold A. The
same valve overlap is present in the cylinder head (see Fig. 5.2). A significantly
larger overlap can be noted for manifold B (see Fig. 5.8 and 5.9).
Although the vertical scale varies between −1 and 3, no negative velocity
can be measured in this case, since the oscillating hot-wire anemometer (see
Chap. 3) has not been used on the isothermal flow rig, only for the measure-
ments on the isochoric (CME) flow rig (see Sect. 5.1.2).
Figures 5.1a and b are for two different engine speeds. The time-resolved
plots (Fig. 5.1, top) are rather similar for both engine speeds. The same holds
true for the time-averaged plots (Fig. 5.1, bottom), which may be explained
by the high scavenging number S for these flow conditions (S > 2). In both
cases, the mean velocity is quite sinusoidal. This is due to the sawtooth-like
characteristic of the rotating valve open area. This remark is important in light
of the explanation of the resonance phenomenon in Sect. 5.3.
Figures 5.2a and b are for corresponding flow conditions, yet with the cylin-
der head as pulsator. The dimensionless exhaust valve lift is again plotted as
light solid lines ( ), yet now represents the cosine-like evolution of the exhaust
valve lift.
The time-resolved mean velocity Um (ωt) behaves very differently compared
to using the rotating valve (Fig. 5.1). This is due to the combination of (i) a
different cross-sectional area evolution (i.e. sawtooth versus cosine) and (ii) a
different discharge coefficient Cd for both geometries. For compressible flow
through a restriction, Eq. (B.6) gives the mass flow rate for a given restriction
and upstream and downstream flow conditions. The mass flow rate ṁ ∼ Cd A,
where A is the variable cross-sectional area of the restriction, and the discharge
coefficient Cd for exhaust valves is based on Fig. B.1 from Heywood [47]. For a
rotating valve, Cd can be obtained based on the loss coefficient for a ball valve,
as given e.g. by Fig. B.2 from Miller [74].
Figure 5.3 shows the product of discharge coefficient and cross-sectional
area Cd · A, as it evolves during the course of a single exhaust stroke. For the
rotating valve ( ), Cd · A = Amax in the maximum open position, since Cd
becomes unity (see Fig. B.2b). By contrast, for the cylinder head (( ), Cd · A
remains smaller, yet quite constant around the maximum open position, due
28 See the comment on p. 35 concerning the difference between intra-cylinder and inter -
cylinder valve overlap.

1
Rotating valve
Cylinder head
0.8
Cd ⋅ A / Amax (-)
0.6
0.4
0.2
0
θe θe + ∆θ
Crankshaft angle θ
Figure 5.3 – Difference between rotating valve and cylinder head in terms of the
product Cd A (θ)
to the fact that Cd decreases for increasing lift (see Fig. B.1). As such, Fig. 5.3
clarifies the different behavior between both pulsators in Figs. 5.1 and 5.2.
From Fig. 5.2b, the mean velocity Um (ωt) clearly exhibits strong fluctu-
ations. Due to the phase-locked measurement technique and the ensemble-
averaging data reduction (see Sect. 2.5.1), these cannot be caused by random
phenomena, such as noise or turbulence. The fluctuations occur only while
using the cylinder head and not while using the rotating valve, as shown by
comparison of Figs. 5.2b and 5.1b. This phenomenon is discussed in detail in
Sect. 5.3.
Both cases shown in Fig. 5.2a and b are for a high scavenging number
S. The time-averaged distributions appear similar, although the high engine
speed distribution is notably more uniform compared to the low engine speed
distribution (ηm = 0.75 for Fig. 5.2b versus ηm = 0.67 for Fig. 5.2a). The same
is not true for the rotating valve cases in Fig. 5.1a and b. Since the flow rate is
comparable (Qref ' 75 m3 /h), the geometry is identical and the engine speeds
differ by a factor of two, the scavenging numbers differ by the same factor of
two. For Figs. 5.1a and 5.2a, S ' 4, compared to S ' 2 for Figs. 5.1b and 5.2b.
The difference in flow uniformity between Figs. 5.2a and b can be explained
in terms of the different frequency content of the mean velocity. Referring to the
second definition of the apparent pulsation period Tp in Eq. (4.46) based on the
peak frequency in the mean velocity spectral density, and the conclusions in the
preceding chapter concerning the validity of the addition principle, the energy
spectral density of Um in the case of Fig. 5.2b is increased due to the occurrence
of the resonance phenomenon. If the fluctuations were any stronger, the value
of Tp according to Eq. (4.46) would be lower compared to the value based on the
old definition in Eq. (4.44). In that case, the value of the scavenging number
S = Tp /Ts would decrease. In light of the conclusions of the preceding chapter
and in particular Fig. 4.29, a decrease in scavenging number by any means
(e.g. increasing frequency content in Um ) entails an increase in flow uniformity

(see rM versus S in Fig. 4.29b) and a decrease in correlation between pulsating
and stationary flow conditions (see rS versus S in Fig. 4.29a). This is all in
accordance with the observed difference.
Unlike the mean velocity, the flow uniformity ηm (ωt) in Figs. 5.4 and 5.2
( ) is clearly out of phase with (and leading) the rotating valve motion. This
becomes more clear in Figs. 5.4 and 5.5.
Figures 5.4 and 5.5 show some time-resolved velocity distributions U (ωt)
during the exhaust stroke of cylinder 1. The distributions are taken at five
crankshaft positions between 480 ◦ ca and 720 ◦ ca. According to Table 2.1, the
valve timing for the engine corresponding to manifold A is −12 | 242 | −246 | 10.
As such, the exhaust stroke for cylinder 1 occurs between −246 = 474 ◦ ca and
10 ◦ ca. The five crankshaft positions are chosen arbitrarily, for the sake of
simplicity as (a) 480 ◦ ca, (b) 540 ◦ ca, (c) 600 ◦ ca, (d) 660 ◦ ca and (e) 720 ◦ ca.
The middle position (c) corresponds roughly to the maximum lift position,
= (−246 + 10)/2 = −118 = 602 ◦ ca.
Each velocity distribution is obtained using linear interpolation from the J
available distributions for the entire engine cycle. As defined in Sect. 2.5.1, J
corresponds to the number of samples per engine cycle. For the isothermal flow
rig experiments, a constant value of J = 80 is taken. This corresponds to a
temporal resolution of 720/J ' 9 ◦ ca. For the isochoric flow rig measurements,
a higher value of J = 256 is taken to better resolve the higher frequency
content in the time-resolved velocity, corresponding to a temporal resolution of
720/J ' 3 ◦ ca.
Similar to the time-averaged velocity distribution plots in the preceding
chapter, the time-resolved velocity distributions in Figs. 5.4 and 5.5 are plotted
non-dimensionally,
R dividing the velocity by the time-averaged mean velocity
Um = ωt Um (ωt) d (ωt)/(4π) . Contour lines of equal velocity are plotted,
with a dashed contour at unity. A cross-sectional plot is added below each
figure, indicating the dimensionless velocity along the straight line y = 0 mm.
The top left of each figure indicates the engine speed N and time-averaged
reference flow rate Qref , whereas the bottom left indicates the time-resolved
mean velocity Um (θ)/Um and the time-resolved flow uniformity measure ηm (θ)
according to Eq. (4.9). The bottom right of each figure indicates the crankshaft
position θ = ωt (◦ ca) and the time (ms) during the engine cycle. For clarity,
the position of the camshaft29 is indicated by the clockwise rotating marker.
The upright position corresponds to 0 = 720 ◦ crankshaft angle. The numbers
1, 2 and 3 indicate the exhaust strokes of each cylinder, which corresponds to
the firing order for manifold A (1–4–2–6–3–5).
The out-of-phase evolution of the flow uniformity notable in Fig. 5.1a is
observed in Fig. 5.4 as well. During the initial phase of the exhaust stroke
(from Figs. 5.4a to b), the flow distribution becomes more uniform. This may
be explained by the following hypothesis:
29 For the experiments using the rotating valve, this position corresponds to twice the value
of the rotating valve position, since the rotating valve features two openings per revolution.
Time-resolved velocity U (ω t) (-) Time-resolved velocity U (ω t) (-)

6 6
30 1 30
5 1 5
15 4 15 4
1
0.5
1
y (mm)
y (mm)
0.5
0.5
0 3 0 3
1
-15 2 -15 2
1
0.5
0.5 1 1
-30 -30
θ = 480.0 °ca 1 2 θ = 540.0 °ca 1 2
Um(θ)/Um = 0.609, ηm(θ) = 0.447 t = 55.6 ms 3 Um(θ)/Um = 1.043, ηm(θ) = 0.634 t = 62.5 ms 3
0 0
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-30 -15 0 15 30 -30 -15 0 15 30
(a) (b)
x (mm) x (mm)

6 6
1
30 5
30 2 5
1 1
2
1
0.5
15 4 15 4
1
0.5
2
y (mm)
y (mm)
2
5
0.5
0.
0 1 3 0 3
3
0.5
0.5
2
2
-15 2 -15 2
0.5
1
1
0.5
1 1 0.5 1
-30 -30
θ = 600.0 °ca 1 2 θ = 660.0 °ca 1 2
0 0
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-30 -15 0 15 30 -30 -15 0 15 30
(c) (d)
x (mm) x (mm)
Time-resolved velocity U (ω t) (-)

6
N = 1440 rpm, Qref = 77.3 m3/h
30 5
0.5
0.5
15 1 4
1
0.5
1
0.5
y (mm)
0 3
-15 2
1
0.5
0.5 1
-30
θ = 720.0 °ca 1 2
Um(θ)/Um = 0.621, ηm(θ) = 0.377 t = 83.3 ms 3
0
Uy=0 (-)
4
2
0
-30 -15 0 15 30
(e)
x (mm)
Figure 5.4 – Time-resolved velocity distributions U (ωt) [-] for crankshaft positions
(a) 480 ◦ ca through (e) 720 ◦ ca, for manifold A on the isothermal flow rig with rotating
valve, at N = 1440 rpm (see Fig. 5.1a)

6 6
30 0.5 30
1 5 5
1
0.5
15 1 4 15 4
1
0.5
y (mm)
y (mm)
1
0 3 0 3
1
0.5
1
0.5 2 1 2
-15 -15
1
0.5
1 1
-30 0.5 -30 1 1
θ = 480.0 °ca 1 2 θ = 540.0 °ca 1 2
0 0
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-30 -15 0 15 30 -30 -15 0 15 30
(a) (b)
x (mm) x (mm)

6 6
0.5 1 0.5
30 5
30 5
2 2
1
1
2
2
15 15
1
0.5
4 4
1
y (mm)
y (mm)
3
0.5 0.5
0 3 0 3
0.
5
3 3
2
2
0.5
2 2
0.5
-15 -15
5
1 1 0.
0.5
0.5
1 1
0.5
-30 -30
1
θ = 600.0 °ca 1 2 θ = 660.0 °ca 1 2

0 0
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-30 -15 0 15 30 -30 -15 0 15 30
(c) (d)
x (mm) x (mm)

6
N = 1440 rpm, Qref = 77.4 m3/h
0.5
30 5
1
0.5
0.5
1
15 4
0.5
y (mm)
0 3
1
0.5
-15 2
0.5
0.5
1
-30
0.
5
θ = 720.0 °ca 1 2
Um(θ)/Um = 0.641, ηm(θ) = 0.324 t = 84.4 ms 3
0
Uy=0 (-)
4
2
0
-30 -15 0 15 30
(e)
x (mm)
(a) 480 ◦ ca through (e) 720 ◦ ca, for manifold A on the isothermal flow rig with cylinder
head, at N = 1440 rpm (see Fig. 5.2a)

6 6
1 0.5
30 0.5 2
5
30 2
5
1
0.5
1
15 4 15 4
0.5
2
0.5
y (mm)
y (mm)
0 3 0 3
1
3
0.5
1
2
2 2 2
0.5
-15 -15
5
0.
0.5
5
1
0.
1 1 1
-30 -30
0 0
5 5
Uy=0 (-)
Uy=0 (-)
3 3
1 1
0 0
-30 -15 0 15 30 -30 -15 0 15 30
x (mm) x (mm)
(a) (b)
Figure 5.6 – Stationary velocity U [-] for flow through runner 1 on manifold A
mounted on isothermal flow rig, using (a) rotating valve and (b) cylinder head, at
comparable flow rate to Figs. 5.4 and 5.5
Hypothesis: Diffuser flow attachment evolution As the gas

in the open runner and the diffuser starts to move during the ini-
tial stage of the exhaust stroke, the flow in the diffuser appears to
remain attached to the walls. At some point (see Figs. 5.4b to c),
the flow in the diffuser separates and the flow uniformity decreases
significantly.
The peak velocity occurs in the region of the catalyst where the
open runner issues. The time-resolved velocity distribution (c) at
maximum lift (ωt = 600 ◦ ca) is quite similar to (yet slightly more
uniform than) the stationary flow distribution shown in Fig. 5.6a
for steady flow through runner 1, with the rotating valve fixed in
the maximum flow position.
During the remainder of the exhaust stroke (from Figs. 5.4c to e),
the gas decelerates, although the high velocity region remains vis-
ible. The velocity in the rest of the cross-section is very low. Al-
though the value of the time-resolved mean velocity is the same
in Figs. 5.4b and d (Um (ωt) ' 1), one can hypothesize that no
reattachment occurs in the diffuser. As such, the flow uniformity
remains low throughout the remainder of the exhaust stroke, until
the beginning of the exhaust stroke of the following cylinder.
Literature exists on the flow patterns (Fig. 5.7) and pressure recov-
ery performance of diffusers [10, 74, 105]. However, the geometry
of the exhaust manifolds under investigation is quite complex and
(b)
(a)
Figure 5.7 – Diffuser flow patterns for (a) small and (b) wide divergence
(Source: [74])
differs too much from the simple geometries described in literature.

Nevertheless, even simple diffusers exhibit complex separation, de-
pending on the divergence angle and inlet conditions. For a nearly
uniform inlet distribution, Miller [74] identifies five regimes of tran-
sitory separation for an axisymmetric diffuser, ranging from steady
flow (for small angle diffusers) to violent fluctuations in the flow pat-
tern and pressure recovery (for wide angle diffusers, comparable to
those found in close-coupled catalyst manifolds). For a non-uniform
or skewed inlet velocity profile or containing secondary flows (e.g.
in an exhaust manifold), separation is extensive yet stationary [74].
Separation patterns are determined by the location of low energy
fluid, e.g. downstream of the inside of a bend.
Figure 5.7b shows the typical flow pattern in a wide angle diffuser.
When the divergence angle is greater than a few degrees, separation
occurs, causing a distorted outlet velocity distribution with outer
regions of flow reversal. Flow reversal exists up to the reattachment
point in the downstream pipe. However, the influence of the pres-
ence of a catalyst substrate on this flow pattern is unknown. No
literature is available on this subject.
The same observations can be made in Fig. 5.5. The flow uniformity attains
a maximum during the initial phase of the exhaust stroke (Fig. 5.5b), and
remains low throughout the remainder of the exhaust stroke. At maximum lift
(Fig. 5.5c), the velocity distribution is similar to the stationary flow case shown
in Fig. 5.6b.
Manifold B
Figure 5.8 (top) shows the time-resolved dimensionless mean catalyst velocity
Um (ωt), for manifold B mounted on the isothermal flow rig, for a constant
engine speed yet (a) Qref ' 45 m3 /h and (b) Qref ' 115 m3 /h. The bottom
plots show the corresponding time-averaged velocity distributions.
As for manifold A, the crankshaft position ωt is defined relative to top dead
3 3
2.5 2.5

2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 Um -0.5 Um
ηm ηm
-1 -1
0 180 360 540 720 0 180 360 540 720

2.5 2.5
60 60
2 2
0.8 0.9 0.9 0.8
1
1 1 0.8 0.9
0.9
0.9
0.
30 30 0.9 9
1
0.9
0.8 1
1.5 0.8 1.5

1
0.7
0.8
1.2
1.61.4
y (mm)
y (mm)
1.2
0.8
1.621.8
1.2
1.2
0 0
1.2
0.9
2.2
1.4
1.4
0.9 1
1
2.4
1
0.8
0.7 0.8
1 1
1.2
1.4 .81
0.9
1.4
0.9
-30 -30 1.2

0
1
0.9
1 0.9 0.8
1
1
0.5 0.5
-60 -60
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
averaged distribution U [-] (bottom) for manifold B on the isothermal flow rig, for
N = 1200 rpm and (a) Qref ' 45 m3 /h and (b) Qref ' 115 m3 /h
3 3
2.5 2.5

2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 Um -0.5 Um
ηm ηm
-1 -1
0 180 360 540 720 0 180 360 540 720

2.5 2.5
60 60
0.8 2 2
0.9 0.8
1 1
2 1
0.9
1. 1
30 30
0.9
1
0.9
1.2 1.5 1.5

1
0.9
1
1.2
y (mm)
y (mm)
1.6
1.4
1.8
0.8
0 0
1.4
0.9
1
1
1
0.8
1.2
0.9 1 1
1.2
1
0.9
1
1.2
1.2 0.9
-30 -30
0.9
1
0.8
1
0.8
0.9
1
1
1
1
0.7 0.8 0.5 0.9
0.5
-60 -60
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
averaged distribution U [-] (bottom) for manifold B on the isothermal flow rig, for
Qref ' 70 m3 /h and (a) N = 1200 rpm and (b) N = 2810 rpm
center of cylinder 1, prior to the intake stroke. As such, the time-resolved mean
velocity plots show the consecutive exhaust strokes of cylinders 3, 4, 2 and 1.
Some distinct differences may be noted between Figs. 5.8a and b, which dif-
fer only in terms of flow rate while the engine speed remains constant. Firstly,
the peak mean velocities corresponding to each of the maximum valve lift in-
stants do not agree. The maximum valve lifts occur roughly at 75 + n 180 ◦ ca,
where n = 0 . . . 3. For a low flow rate (Fig. 5.8a), the highest peak velocity
occurs during the exhaust stroke of cylinder 4. For a high flow rate (Fig. 5.8b),
the highest peak velocity occurs during the exhaust stroke of cylinder 2. The
difference is likely due to the Reynolds number-dependence of the flow pattern
in the runners, although this cannot be verified experimentally.
Secondly, the flow uniformity ηm (ωt) ( ) also behaves differently depend-
ing on the flow rate. For a high flow rate (Fig. 5.8b), the flow uniformity
exhibits a time evolution which is very similar to that observed on manifold A.
The flow uniformity increases during the initial phase of the exhaust stroke.
At some point prior to the maximum valve lift, the flow in the diffuser de-
taches and the flow uniformity remains low and relative constant throughout
the remainder of the exhaust stroke.
However, this changes dramatically for a low flow rate (Fig. 5.8a). On the
time-averaged level, the flow uniformity is much higher. On the time-resolved
level, the flow uniformity shows no clear sign of flow detachment in the diffuser.
Instead, the flow uniformity ηm increases monotonously during the exhaust
stroke, reaching a maximum after the maximum valve lift event. Possibly,
the flow in the diffuser remains attached at such low flow rate. However, this
hypothesis cannot be supported based on the literature. Miller [74] indicates
that the diffuser pressure loss increases for decreasing Reynolds number.
More likely, the effect is due to interfering effects in the diffuser. Indeed, the
condition in Fig. 5.8a corresponds to low engine load conditions and therefore,
a low scavenging number S. As such, successive exhaust pulses interfere to a
higher degree in the diffuser.
For the same engine speed, an intermediate flow rate of Qref ' 70 m3 /h
is shown in Fig. 5.9a. In this case, the flow detachment can already be noted,
although not as clear as for the high flow rate case of Fig. 5.8b.
Figure 5.9 shows a comparison for the same flow rate yet different engine
speed. The main difference between the two cases is the resonance phenomenon
which was already observed on manifold A using its cylinder head as pulsator.
This phenomenon is discussed in detail in Sect. 5.3.
Figure 5.10 shows some time-resolved velocity distributions U (ωt) during
the exhaust stroke of cylinder 1, for manifold B. The distributions are taken
at five crankshaft positions between 525 ◦ ca and 705 ◦ ca. These positions differ
slightly compared to those for manifold A, due to (i) the later exhaust valve
opening for manifold B (EO equals -220 ◦ ca instead of -246 ◦ ca, see Table 2.1),
and (ii) the significant interference of the exhaust pulses, which results from
the higher inter-cylinder valve overlap.
The overlap ∆θ between consecutive exhaust strokes equals ∆θ =
(EC − EO) − 720/nr , where EO and EC represent the exhaust valve open

6 6
60 60
5 5
0.5 0.
5
30 4 30 4
1
0.5
1
0.5
0.5
1
0.5 0.5
1
y (mm)
y (mm)
2
0 3 0 3
2
0.5
1
0.5
2
2 2
1
-30 -30
2
1 0.5
1
0.5 1 1
1 1
-60 -60
θ = 525.0 °ca 1 3
θ = 570.0 °ca 1 3
4 4
0 0
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
(a) (b)
x (mm) x (mm)

6 6
60 60
5 5
1 1
1
1
1
1
30 1 30
1
4 4
1 1
1
1
y (mm)
y (mm)
1
2
0 3 0 3
1
4
12
2
0.5
1
1
1
1
-30 2 -30 2
1
1
1
1 1
-60 -60
θ = 615.0 °ca 1 3
θ = 660.0 °ca 1 3
4 4
0 0
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
(c) (d)
x (mm) x (mm)

6
N = 1200 rpm, Qref = 68.0 m3/h
60
5
0.
5
1
0.
5
30 0.5 4
1 0.5
y (mm)
0 3
3 2
0.5
0.5
-30 2
1
-60
θ = 705.0 °ca 1 3
Um(θ)/Um = 0.617, ηm(θ) = 0.186 t = 97.9 ms 2 4
0
Uy=0 (-)
4
2
0
-60 -30 0 30 60
(e)
x (mm)
(a) 525 ◦ ca through (e) 705 ◦ ca, for manifold B on the isothermal flow rig, at N =
1200 rpm and Qref ' 70 m3 /h (see Fig. 5.9a)
and close crankshaft angles (see Table 2.1) and nr is the number of exhaust
runners per catalyst (i.e. nr = 3 for manifold A and nr = 4 for manifold B).
This corresponds to an overlap of ∆θA = 16 ◦ ca and ∆θB = 53 ◦ ca. Although
these values differ only 37 ◦ ca, the difference is significant, due to the non-linear
relationship between the crankshaft position and the exhaust valve flow rate
(see App. B, Eq. (B.6)).
In Fig. 5.10, the five crankshaft positions are chosen as (a) 525 ◦ ca,
(b) 570 ◦ ca, (c) 615 ◦ ca, (d) 660 ◦ ca and (e) 705 ◦ ca. Similar to Figs. 5.4 and 5.5,
the middle position (c) 615 ◦ ca corresponds roughly to the maximum lift posi-
tion, = (−220 + 13)/2 ' −104 = 616 ◦ ca. The start position (a) 525 ◦ ca and
the end position (e) 705 ◦ ca are chosen where the exhaust valve lift of cylinder 1
equals that of the preceding and following cylinder (2 and 3, respectively).
The out-of-phase evolution of the flow uniformity notable in Fig. 5.9a is
again observed in Fig. 5.10.
Although the first position (a) in Fig. 5.10a occurs at 525 ◦ ca, whereas the
exhaust valves for cylinder 1 start to open at 500 ◦ ca, the velocity distributions
shows that the gas is still flowing from runner 2. The same observation can be
made for the end position (e), where the gas is still flowing from runner 1 while
the exhaust valves of cylinder 3 are already opening.
The peak velocity occurs in the region of the catalyst where runner 1 issues.
Contrary to the case for manifold A, the time-resolved velocity distribution
at maximum lift in Fig. 5.10c significantly differs from the stationary flow
distribution shown in Fig. 4.17a for steady flow through runner 1, with the
valves blocked in the maximum lift position. The correspondence is better
with the velocity distributions in Figs. 5.10d and e.
5.1.2 Isochoric flow rig

Mean velocity
Only manifold B has been used during the experiments on the isochoric (CME)
flow rig (see Sect. 2.2.2). These measurements are all performed using the
oscillating hot-wire anemometer (OHW) presented in Chap. 3. This system
features a maximum measurable negative velocity of −1 m/s. This limit value
is adequate at low engine speed N < 1500 rpm, yet it proves insufficient to
resolve the strong flow reversal at low engine load and higher engine speed (see
Sect. 5.2).
With reference to Sects. 2.2.2 and 2.3, the isochoric CME flow rig generates
a cold pulsating flow in the exhaust system which closely resembles fired en-
gine flow conditions. In that sense, the flow conditions are denoted somewhat
differently from the isothermal flow rig. For the isothermal flow rig, the engine
speed N is determined by the pulsator frequency, and the flow rate Qref is
determined independently from the engine speed. Thus, a broad scavenging
number range could be tested in Chap. 4. On the other hand, the CME flow
rig and a fired engine are volumetric (hence: isochoric) machines. The flow
rate Qref is in first approximation proportional to the engine speed N , and
is furthermore determined by the engine load. The correct interpretation of
3 3
2.5 2.5

2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 Um -0.5 Um
ηm ηm
-1 -1
0 180 360 540 720 0 180 360 540 720

2.5 2.5
60 60
2 2
1
1 1 1
1.2
00.9.8
1
0.9
8
1.
30 30
0.
2
0.8
1
1
1.2
1.5 1.5
1.2
0.8
0.9
1.2
0.9
y (mm)
y (mm)
0.8
1
9
1.6
1.6
1.4
0.9
0.
4
1.
0.8
1.4
1.8
1
0.9
1.4
1.8 2
0 0
1.2
1
1.6
0.9
0.9
1
0.8
1.2
0.9
0.9
0.9
1 1
1
0.8 1.2
1.2
0.7
0.8
0.7
0.8
0.8
-30 -30
0.8
0.9
0.9
1
1 1
0.5 0.5
-60 -60
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
Figure 5.11 – Time-resolved mean velocity Um (ωt) [-] (top) and corresponding
time-averaged distribution U [-] (bottom) on the CME flow rig, for N = 1200 rpm
and (a) Qref ' 70 m3 /h (part load) and (b) Qref ' 100 m3 /h (high load)
‘engine load’ is given by Footnote 6. Both for the fired engine and for the
CME flow rig, the intake manifold pressure pi (or more precisely: the density
ρi = pi /(r Ti ) ) determines the engine load. As such, the intake system pressure
pi is shown for experiments obtained on the CME flow rig. The reference flow
rate Qref is also shown, to compare with the isothermal flow rig experiments.
Figure 5.11 presents the time-resolved mean velocity Um (ωt) and corre-
sponding time-averaged velocity distributions U obtained on the CME flow
rig, in pulsating flow conditions.
Upon examination of Fig. 5.11, some striking features are noted. Firstly, the
mean velocity exhibits very strong resonance fluctuations. On the isothermal
flow rig (see Sect. 5.1.1), these occur when using the cylinder head as pulsator,
yet only for an engine speed in excess of 2500 rpm. The resonance phenomenon
is further discussed in Sect. 5.3.
Secondly, periodic backflow occurs through the catalyst, in particular at
low engine load (i.e. low intake system pressure pi , low flow rate Qref , low
scavenging number S). Flow reversal is most pronounced immediately follow-
ing the blowdown phase. This is quite remarkable, given that Fig. 5.11 plots
the mean velocity. As such, more extensive local flow reversal is expected.
This is confirmed by the time-resolved velocity distribution plots in Figs. 5.16
through 5.18.
In terms of the time-resolved flow uniformity ηm (ωt), no clear conclusion
can be drawn from Fig. 5.11. The flow uniformity is strongly affected by the
mean velocity fluctuations. Presumably, the flow detachment and reattachment
process in the diffuser is equally affected by the velocity and pressure transients.
In the evolution of the mean velocity in Fig. 5.11, the blowdown phase can be
discerned as the velocity peak immediately following each exhaust valve opening
(e.g. the highest peak in Fig. 5.11b at ωt ' 540 ◦ ca). The displacement phase
follows the blowdown, and is generally characterized by a lower peak flow rate
and lower transients. However, in case of this close-coupled catalyst manifold,
the resonance phenomenon greatly amplifies the velocity fluctuations. This
yields a time-resolved mean velocity during the displacement phase which is
very dissimilar to e.g. the simulated velocity evolution plotted in Fig. 2.10. The
simulation is performed using a zero-dimensional filling-and-emptying model,
which does not take the gas dynamics in the exhaust manifold into account. As
a very crude approximation, the mean velocity evolution for the CME flow rig
shown in Fig. 2.10 can be regarded as a low-pass filtered average of the actual
velocity evolution. However, non-linearities in the gas dynamics ensure that
this comparison does not necessarily hold.
For the part load case (Fig. 5.11a), the magnitude of the blowdown peak
velocity is of the same level as the subsequent peaks during the displacement
phase. Only for the high load case (Fig. 5.11a), the blowdown peaks can be
clearly discerned at approximately 0, 180, 360, 540 ◦ ca for cylinders 3, 4, 2 and
1, respectively.
Figure 5.12 shows a similar comparison as Fig. 5.11, yet for N = 1800 rpm.
The time-resolved mean velocity and flow uniformity are quite comparable to
the N = 1200 rpm case. Since the eigenfrequency of the resonance phenomenon
3 3
2.5 2.5

2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 Um -0.5 Um
ηm ηm
-1 -1
0 180 360 540 720 0 180 360 540 720

2.5 2.5
60 60
2 2
1 1
1
1
30 30
1.2
1.2
0.90.8
0.9
1
0.8
1.41.2
1.
4
0.80.7
1.6
1.5 1.5
1.2
0.8
1.6
1.6
1.4
1.4
0.9
0.9
1.8
y (mm)
y (mm)
1.2 1.4
1.8
1.4
2.4
1.6
0.8
1
0.9
0.9
0 0
1.6
2
0.8
1.6
.8
1.2
1.8
2.2
1.4
0.7 0
2
0.9
1
1 0.9
1.2
1 1
0.8
1
1.4
0.8
2
0.8
0.6
1.
0.8
0.8
1
0.9
1
-30 0.9 -30

0.9
9
0.
0.9
1
1
0.5 0.5
-60 -60
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
time-averaged distribution U [-] (bottom) on the CME flow rig, for N = 1800 rpm
and (a) Qref ' 100 m3 /h (part load) and (b) Qref ' 145 m3 /h (high load)
3 3
2.5 2.5

2 2
1.5 1.5
1 1
0.5 0.5
0 0
-0.5 Um -0.5 Um
ηm ηm
-1 -1
0 180 360 540 720 0 180 360 540 720

2.5 2.5
60 60
2 2
1
1
1
1
30 30
1.2
1.4
1.4
1.6 1.4
0.9
1 0.9
0.8
0.8
1.2
1.6
1.2
1.4
1.5 1.5
1.6
0.7
0.8
1.2
0.8 01.9
0.9
y (mm)
y (mm)
1.2
1.2
0.8
1.41.6
0.8
0.7
1.6
0.9
1.2
1.4
0 0
0.9
0.8
0.7
1
0.8 0.7
1.6
0.8
1.8
0.9
1.2
1.8
1
0.
1
0.8
0.8
1.4
0.8
6 1 1.4 1
0.9
1
-30 -30
1
1
0.9 0.9
1
9
1
0.
1.
1.2
0.5 0.5
-60 -60
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(a) (b)
time-averaged distribution U [-] (bottom) on the CME flow rig, for Qref ' 235 m3 /h
and (a) N = 2400 rpm (high load) and (b) N = 3000 rpm (part load)
is unaffected by engine speed or flow rate (see Sect. 5.3), fewer peaks are shown
during the displacement phase, due to the increased engine speed.
The time-averaged velocity distributions in Fig. 5.12a and 5.11b show a very
good resemblance. This may be attributed to the fact that the flow rates cor-
respond (Qref ' 100 m3 /h) and that the scavenging numbers are comparable
(S = 0.58 for Fig. 5.11b and S = 0.41 for Fig. 5.12a).
Figure 5.13 shows two additional cases at high engine speed. Due to the
higher engine speed, Fig. 5.13a shows only a single blowdown and displacement
peak in the mean catalyst velocity. The same is observed at N = 3000 rpm in
Fig. 5.13b. Since Fig. 5.13b is obtained in part load conditions (pi = 1.55 atm),
the blowdown peak is smaller than the displacement peak.
Velocity distributions
For the experiments on the isochoric flow rig, some selected time-resolved veloc-
ity distributions are shown for a constant engine speed N = 1200 rpm and high,
part and low engine loads. This corresponds to an intake system pressure pi of
approximately 2.2, 1.55 and 1 atm, respectively. As a reference, Fig. 5.14 (left)
provides the corresponding time-averaged velocity distributions, where the en-
gine load varies from high to low from top to bottom.
Figure 5.14 (left) show that the flow uniformity decreases as the engine load
increases (i.e. from bottom to top in Fig. 5.14 left), or equivalently the flow
rate and scavenging number increase.
The plots in Fig. 5.14 (right) show the stationary velocity distribution for
flow through runner 1, at comparable flow rates to each of the engine load cases.
These are obtained on the isothermal flow rig, with the exhaust valves blocked
in the position of maximum lift. The stationary distributions for runner 1 are
shown as reference, since the time-evolution of the velocity distributions are
inspected only during the exhaust stroke of cylinder 1 (see Figs. 5.16, 5.17
and 5.18).
For each engine load case (high to low load from top to bottom), Fig. 5.15
shows the time-resolved mean velocity Um (ωt) and flow uniformity ηm (ωt).
The six vertical lines in each plot indicate the crankshaft positions θi (i =
1 . . . 6) at which the time-resolved velocity distributions are shown.
By contrast to the isothermal flow rig, the crankshaft positions θi are se-
lected differently for each engine load case. This is due to the stronger fluc-
tuations in the mean velocity. Simply selecting a number of fixed crankshaft
positions would complicate a good comparison between the engine load cases.
As such, six consecutive crankshaft positions (a) θ1 through (f) θ6 are selected
during the exhaust stroke of cylinder 1 for each individual case, according to
the following rules:
(a) θ1 corresponds to the maximum Um during the blowdown phase (first

vertical line in Fig. 5.15).
(b) θ2 corresponds to the minimum Um immediately following the blowdown

phase (second vertical line in Fig. 5.15).
(c) θ3 corresponds to the first occurrence of Um = 1 during the displacement

phase (third vertical line in Fig. 5.15).
(d) θ4 corresponds to the first local maximum of Um during the displacement

phase (fourth vertical line in Fig. 5.15).
(e) θ5 corresponds to the second occurrence of Um = 1 during the displace-

ment phase (fifth vertical line in Fig. 5.15).
(f) θ6 corresponds to the first local minimum of Um during the displacement

phase (sixth vertical line in Fig. 5.15).
Time-averaged velocity U (-) Stationary velocity U (-)

2.5 6
N = 1200 rpm, Qref = 97.3 m3/h, pi = 2.20 atm Runner 1, Qref = 94.5 m3/h
60 60
5
2
1
1 0.9
1
8
0.
30 30
1
1
0.8 1. 4
0.5
2
1.2
1
1.4
1.5
0.9 1
0.8
0.9
y (mm)
y (mm)
2
0.8
1.6
1.2 1.4
0.9
1.4
1
3 32
0 0 3
1.6
1.8
45
0.8
1
1.4
0.7
0.9
1.2 1
0.5
2
0.9
0.8
2
0.8
1
-30 0.9
-30
1
1 1
0.5
1
-60 -60
0 0
2 5
Uy=0 (-)
Uy=0 (-)
1 3
1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
(a) (b)
x (mm) x (mm)

2.5 6
60 60
5
2
1
1 1
0.9 1
1.2
0.8
30 30
1
4
1.
2
1
0.5
1.4
1.5
0.9
3
y (mm)
y (mm)
1.2
0.8
1
1.4
1.4
0.9
0.80.9 1
1.6
4
2
1
1.8
0 0 3
0.9
0.5
1 0.9
0.8
1
1.2
1
1.2
1 2
0.9
0.7
0.8
1
-30 -30 2
1
1 1 1
0.5
1
-60 -60
0 0
2 5
Uy=0 (-)
Uy=0 (-)
1 3
1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
(c) (d)
x (mm) x (mm)

2.5 6
60 60
5
1
2
1.2
1.2 1.2
1
1
30 30 4
1
1
1.2
1
1.5 1
0.9
0.9
y (mm)
y (mm)
1
0 0 3
0.9
0.9
1.4
3
2
0.9
0.8
1
1
-30 -30 2
1
1 1 1
0.5
1
-60 -60
0 0
2 5
Uy=0 (-)
Uy=0 (-)
1 3
1
0 0
-60 -30 0 30 60 -60 -30 0 30 60
(e) (f)
x (mm) x (mm)
Figure 5.14 – Time-averaged velocity distributions U [-] (left) and stationary ve-
locity distributions for flow through runner 1 (right), for (a, b) high load, (c, d) part
load, (e, f) low load
Time-resolved
3
2.5

2
1.5
0.5
-0.5 Um
ηm
-1
0 180 360 540 720
(a)
Time-resolved
3
2.5
1.5
0.5
-0.5 Um
ηm
-1
0 180 360 540 720
(b)
Time-resolved
3
2.5
1.5
0.5
-0.5 Um
ηm
-1
0 180 360 540 720
(c)
Figure 5.15 – Time-resolved mean velocity Um (ωt) [-], for (a) high load, (b) part
load, (c) low load; six vertical lines indicate crankshaft positions θi (i = 1 . . . 6) in
Figs. 5.16, 5.17 and 5.18

6 (a) (b) 6
60 5 60 5
3 0
3
3
4 4
3
30 30
0.5
0
1
3
3 3
y (mm)
y (mm)
3 2
3
2
0 0
4
4
2 3 2
0.5
3
1
0
-30 3 -30
3 1 1
-60 0 -60 0
θ = 533.0 °ca 1 3
θ = 565.4 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)

6 6
60 5 60 5
2 2
4 4
30 30
2
2
-0.5
3 -0.5 3
y (mm)
y (mm)
0
-0.5
0 0 -0.5
2
2 2
2
-30 -30
2
2
1 1
2
-60 0 -60 0
θ = 528.8 °ca 1 3
θ = 565.2 °ca 1 3
Um(θ)/Um = 1.920, ηm(θ) = 0.699 t = 73.4 ms 2 Um(θ)/Um = -0.617, ηm(θ) = 0.638 t = 78.5 ms 2
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)

6 6
60 5 60 5
0
0.5
4 4
30 30
0.5
3 3
0.5
y (mm)
y (mm)
0.5
0 0
0.5
2 2
1
-30 -30
0
1 1
0.5
-60 0 -60 0
θ = 537.2 °ca 1 3
θ = 548.4 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
Figure 5.16 – Time-resolved velocity distributions U (ωt) [-], for (top) high load,
(middle) part load, (bottom) low load, at crankshaft positions (a) θ1 and (b) θ2 (other
crankshaft positions: see Figs. 5.17 and 5.18)

6 (c) (d) 6
60 5 60 5
2
1
1 4 4
0.5
30 0.5 30 2
1
32
4
0.5
3 3
1
y (mm)
y (mm)
2
2
0 0
0.5
4
3
2
0.5
2 2
3 4
1
2
0 2
1
0.5 1
-30 -30
1 2
1
1 2
-60 0 -60 0
θ = 590.2 °ca 1 3
θ = 601.8 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)

6 6
60 5 60 5
1 1 2
2
4 4
2
30 30
1
1
2
1
3 3
y (mm)
y (mm)
0 1 0
2
2 2
2
1
1
-30 -30 2
1
1 1
1
2
1 2
-60 0 -60 0
θ = 579.2 °ca 1 3
θ = 590.6 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)

6 6
60 5 60 5
1 4 4
30 30 2
3 3
y (mm)
y (mm)
2
1
0 0 2
2
2
2
1
2 2 2
1
-30 -30
1 1
1
1 2
1
-60 0 -60 0
θ = 555.6 °ca 1 3
θ = 565.2 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(middle) part load, (bottom) low load, at crankshaft positions (c) θ3 and (d) θ4 (other

6 (e) (f) 6
60 5 60 5
4 0.5 4
1 1
30 30
0
1
3 3
y (mm)
y (mm)
0.5
2
0.5
1
0 0
1
2 2
2
3
0
1
0
0.5
-30 -30
1 1
0
1 0
1 0
-60 0 -60 0
θ = 614.1 °ca 1 3
θ = 627.2 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)

6 6
60 5 60 5
0.5
0.5
0.5
4 4
1
0.5
0
30 30
0.5
1
0.5
0.5
3 3
y (mm)
y (mm)
1
1
0 0
1
5
0.
2 2
1
1
0.5
1
-30 -30
1
0.5
1 1 1
0
1
0.5 0
1
-60 0 -60 0
θ = 603.2 °ca 1 3
θ = 616.0 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)

6 6
60 5 60 5
1
4 4
1
30 30
1
0
3 3
1
y (mm)
y (mm)
1
1
0 0
1
0
1
2 2
1
-30 1 -30
1
1 1
1
-60 0 -60 0
θ = 576.4 °ca 1 3
θ = 590.6 °ca 1 3
4 4
-1 -1
Uy=0 (-)
Uy=0 (-)
4 4
2 2
0 0
-60 -30 0 30 60 -60 -30 0 30 60
x (mm) x (mm)
(middle) part load, (bottom) low load, at crankshaft positions (e) θ5 and (f) θ6 (other
Figures 5.16, 5.17 and 5.18 show the time-evolution of the velocity distri-
bution for each engine load case (high to low load from top to bottom), for
crankshaft positions (a) θ1 and (b) θ2 (Fig. 5.16), (c) θ3 and (d) θ4 (Fig. 5.17)
and (e) θ5 and (f) θ6 (Fig. 5.18). As before, the velocity distributions are
plotted as contour lines, where the dashed and dotted contour lines represent
unity and zero velocity respectively. Regions of flow reversal below the dotted
contour line ( ) of zero velocity are shaded gray.
During the first part of the blowdown phase (θ1 : Fig. 5.16a), the velocity
increases while the flow uniformity remains quite high. The second part of the
blowdown (θ2 : Fig. 5.16b) differs according to the engine load. For high engine
load, the distribution is characterized by a sharp velocity peak where runner 1
issues into the catalyst, and extensive backflow throughout the remainder of
the cross-section. Backflow occurs even for high load conditions.
The engine load clearly affects the magnitude and nature of the flow rever-
sal. As discussed in the hypothesis concerning diffuser flow attachment (see
p. 161), for low load conditions, the scavenging number S is low, which means
that successive exhaust pulses interact to a higher degree in the diffuser. For
Fig. 5.16, the three load cases correspond respectively to (top) S = 0.58, (mid-
dle) S = 0.32 and (bottom) S = 0.18.
During the subsequent displacement phase, the mean velocity increases as

the piston starts expelling the cylinder charge (θ3 : Fig. 5.17c). The peak
velocity at high engine load decreases only gradually (Fig. 5.17, top). Due
to the resonance phenomenon in the manifold, the mean velocity continues to
fluctuate throughout the displacement phase. Figures 5.17d and 5.18e,f show
the evolution from the first maximum Um to the first minimum Um during the
displacement phase.
Remarkably, the sixth crankshaft position θ6 corresponding to the first min-

imum Um occurs roughly simultaneously to the point of maximum piston ve-
locity (at 630 ◦ ca). In spite of the high piston velocity, flow reversal still occurs
(θ6 : Fig. 5.18f). It is however less pronounced when compared to the post-
blowdown flow reversal (θ2 : Fig. 5.16b).
Park et al. [81] present phase-locked velocity results obtained using LDA in
a close-coupled catalyst manifold on a fired engine. Figure 5.19 shows periodic
flow reversal in the order of −1 to −2 m/s. The backflow occurs following each
blowdown.
Also using LDA, Liu et al. [70] show flow reversal occurring downstream
of a close-coupled catalyst in motored engine conditions. These conditions
correspond exactly to the CME flow rig with atmospheric intake pressure
pi = 1 atm. Transient simulations show flow reversal in motored and fired
conditions. Equivalently, flow reversal is most pronounced following blowdown.
The maximum backflow varies between −1 and −5 m/s in motored and fired
conditions.
Figure 5.19 – Single-point velocity in the catalyst of a fired engine (Source: [81])
5.2 Flow reversal

This section focuses on the occurrence of flow reversal throughout the catalyst
cross-section. Section 5.2.1 describes the validation of the oscillating hot-wire
anemometer (OHW) on the isochoric flow rig, in conditions where extensive
flow reversal is known to occur. The OHW proves crucial for obtaining accurate
measurements on the isochoric flow rig.
Section 5.2.2 discusses the experimental time-resolved velocity data ob-
tained on the CME flow rig, using the OHW for measuring the bidirectional
velocity (see also Sect. 5.1.2).
Section 5.2.3 discusses numerical results obtained using a one-dimensional
gas dynamic model of the exhaust system. The numerical results are compared
to the measured mean catalyst velocity.
Based on the numerical model, the influence of an exit cone and exhaust pipe
with muffler is investigated on the flow dynamics in the manifold, in particular
on the catalyst flow reversal.
5.2.1 OHW validation

For the experiments on the CME flow rig, the oscillating hot-wire anemometer
(OHW) is used for measuring the bidirectional phase-locked velocity. The
construction and calibration of the OHW are described in Chap. 3.
To assess the effectiveness of the OHW in measuring the bidirectional veloc-
ity, a number of engine operating points that feature flow reversal are selected.
In these operating points, the oscillator frequency Rf [-] defined by Eq. (3.13)
is increased from zero for a stationary probe to the maximum attainable. As
Rf increases, so does the resolution in the negative velocity range, and conse-
quently the correspondence improves between the exhaust flow rate calculated
as the area-averaged OHW velocity Q = Um A and the reference flow rate Qref
measured using the laminar flow meter (see Sect. 2.4.2).
Figure 5.20a shows the non-dimensional flow rate deviation δQ = Q/Qref −
1 [-]. The reference exhaust flow rate Qref [m3 /s] is calculated as Qref =
ρs (Qs,in − Qs,bb )/ρ , where ρs and ρ are the density of air at standard condi-
tions (i.e. 0 ◦ C, 1 atm) and exhaust conditions, respectively. The intake stan-
5.2 Flow reversal 183
Time-resolved
0.6 1.5
N = 600 rpm, pi = 1.00 atm
pi = 1.0 atm, N = 600 rpm Rf = 0
0.5 pi = 1.0 atm, N = 1200 rpm Rf = 6.75
Flow rate error δQ = Q /Qref - 1 (-)
pi = 1.0 atm, N = 1800 rpm 1

0.4 pi = 1.5 atm, N = 600 rpm
Mean velocity Um (m/s)

pi = 1.5 atm, N = 1200 rpm
pi = 1.5 atm, N = 1800 rpm
0.3
pi = 2.2 atm, N = 1200 rpm 0.5
pi = 2.2 atm, N = 1800 rpm
0.2
0
0.1
0
-0.5
-0.1
-0.2 -1
0 1 2 3 4 5 6 7 0 180 360 540 720
OHW frequency Rf (-) Crankshaft angle ωt (°)
(a) (b)
Figure 5.20 – Influence of OHW frequency Rf on (a) the flow rate deviation and
(b) the (dimensional) time-resolved mean velocity Um (ωt) [m/s]
dard flow rate Qs,in (Nm3 /s) is determined by means of a laminar flow meter
in the intake system. This measurement is further verified using a cylinder
pressure sensor, by calculating the cylinder charge per cycle from the pressure
rise during the compression stroke (see Sect. 2.4.3).
The blow-by leakage standard flow rate Qs,bb (Nm3 /s) is estimated based
on a correlation as a function of the engine speed N and the intake manifold
pressure. Figure 5.21 shows the ratio of blow-by standard flow rate Qs,bb to
the intake system standard flow rate Qs,in . These data are obtained on the
isochoric flow rig. The blow-by mass flow rate is determined using low range
rotameters30 connected to the crankcase ventilation outlet. A correlation of
the following form is fitted to the data:
γ+1
2(γ−1)
pi √ 2
Q0s,bb = (Cd δc )bb (4πb) √ γ (5.1)
rTi γ+1
where b is the cylinder bore [m], pi is the intake manifold pressure [Pa], Ti is
the intake manifold temperature [K], γ is the ratio of specific heats [-]. The
factor (Cd δc )bb represents the product of a discharge coefficient and an equiv-
alent clearance width between the cylinder wall and the piston rings. The
factor (Cd δc )bb is fitted to match the experimental data, resulting in a value of
(Cd δc )bb = 0.934 µm.
Equation (5.1) is established based on the assumption that choked flow
occurs within the cylinder-to-piston clearance. This seems a fair assump-
tion, since the averaged cylinder pressure during the compression and expan-
sion stroke pcyl,m (when blow-by leakage is most pronounced) is greater than
30 Kobold low range precision rotameters, ranges 50 to 500 Nl/h and 300 to 3000 Nl/h in
air at 20 ◦ C, 1.2 atm.

Blow-by leakage correlation

0.05
pi / pa = 1.0
pi / pa = 1.5
0.04 pi / pa = 2.0
pi / pa = 2.5
Qs,bb’ / Qs,in
Qs,bb / Qs,in (-)

0.03
0.02
0.01
0
0 1000 2000 3000
Engine speed N (rpm)
Figure 5.21 – Experimental correlation for the blow-by leakage on the isochoric flow
rig
γ/(γ−1)
the critical pressure, or symbolically pcyl,m > pa ((γ + 1)/2 ) (see also
Eq. (B.6)).
The correlation Q0s,bb ( ) in Fig 5.21 is determined according to Eq. (5.1).
Figure 5.21 demonstrates that the blow-by leakage at most amounts to 5 % of
the intake flow rate, at low engine speed.
The data in Fig. 5.20a represent experiments at engine speeds of 600, 1200
and 1800 rpm. At low engine load (pi = 1 atm), strong backflow occurs (see
Sect. 5.2). This situation is not physically possible with fired engine conditions,
as discussed in Sect. 2.3. However, local occasional backflow occurs also at
higher engine load (pi = 1.5 . . . 2.2 atm).
Figure 5.20a shows that for increasing OHW oscillation frequency Rf , the
velocity measurement becomes increasingly more accurate. Traditional HWA
using a stationary probe corresponds in Fig. 5.20a to the points at Rf = 0.
The flow rate error δQ amounts to anywhere between 0 and 50 %. The OHW
reduces δQ to within the uncertainty margins on Qref , which corresponds to
approximately 10 % (see Sect. 2.4).
The maximum attainable dimensionless frequency Rf depends on the engine
speed, since Rf = fo /(N /120 ) and the maximum oscillation frequency fo is
about 50 Hz for xo = 5.5 mm. This explains why the curves in Fig. 5.20a have
a different range, depending on the engine speed N .
Figure 5.20b shows the influence of using the OHW with different frequency
Rf on the time-resolved mean velocity Um (ωt) [m/s]. Contrary to above fig-
ures, the mean velocity is plotted dimensional in m/s, so as to better distinguish
the effect of using the OHW at different oscillation frequencies.
The dashed line ( ) in Fig. 5.20b uses a stationary probe (Rf = 0), whereas
the solid line ( ) uses an oscillating probe at Rf = 6.75. This experiment
corresponds to the rightmost circular marker ( ) in Fig. 5.20a. The mean
velocity using traditional HWA (Rf = 0) exhibits the typical rectification or
Characteristic catalyst regions
60
30
A
y (mm)
D B
0 C
E
-30
-60
-60 -30 0 30 60
x (mm)
Figure 5.22 – Characteristic catalyst regions for manifold B
folding when flow reversal occurs. This is due to the inherent insensitivity of
HWA to the velocity direction (see Sect. 3.2).
The mean velocity fluctuations in Fig. 5.20b during the displacement phases
are due to Helmholtz resonances in the exhaust runners and collector volume.
This is explained in Sect. 5.3. No significant flow reversal occurs during the
latter part of the displacement phase. During those periods, Fig.5.20b demon-
strates that the mean velocity at Rf = 0 and Rf = 6.75 yield identical results.
In conclusion, the OHW seems an appropriate measurement for determining
the time-resolved velocity in case of occasional flow reversal. Further improve-
ments to the OHW approach can be made based on the discussion in Sect. 3.7.4.
The present system features a maximum measurable negative velocity of ap-
proximately −1 m/s, which is sufficient for the CME flow rig at low engine
speed, yet insufficient at higher engine speed (N > 2000 rpm) and low load
conditions. Nevertheless, as demonstrated by Fig. 5.20a the accuracy on the ve-
locity measurements is significantly improved compared to traditional hot-wire
anemometry with a stationary probe.
5.2.2 Experimental results

Upon examination of the time-averaged velocity distributions in Figs. 5.11, 5.12
and 5.13 for manifold B on the CME flow rig, one can discern five characteristic
regions in the catalyst cross-section.
These are indicated in Fig. 5.22. The flow rate in regions A and B is
high, since runners 1 and 2, respectively 3 and 4 issue into the diffuser at those
locations. This is shown in Fig. 2.1. The central region C and the outer regions
D and E are most prone to backflow.
The time evolution of the catalyst flow distribution during the exhaust
stroke of cylinder 1 is examined up-close in Figs. 5.16, 5.17 and 5.18. Based
on the mean velocity evolution in Fig. 5.15, six crankshaft positions θi , i =
6 6
UA(θ) UC(θ)
5 5
UB(θ) UD(θ)
UE(θ)
4 4
Velocity (-)
Velocity (-)
3 3
2 2
1 1
0 0
-1 -1
0 180 360 540 720 0 180 360 540 720
(a) (b)
Figure 5.23 – Time-resolved velocity for 1200 rpm and high load, locally averaged
in (a) regions A and B and (b) regions C, D and E
1 . . . 6 are determined for each engine load case. These positions are defined in
Sect. 5.1.2.
Figs. 5.16, 5.17 and 5.18 show the evolution from crankshaft positions (a) θ1
through (f) θ6 , for (top) high, (middle) part and (bottom) low engine load.
Flow reversal is indicated by a dotted contour at zero (U = 0) and particularly
by the shaded region, representing negative velocity.
During peak blowdown flow rate (Fig. 5.16a), the flow remains relatively
uniform. At the end of the blowdown, the cylinder pressure has dropped to the
level of the exhaust manifold. The inertia of the gas in runners and catalyst
causes it to keep flowing in region A (Fig. 5.16b). The cylinder and manifold
pressure drop below atmospheric pressure and cause backflow in the remainder
of the cross-section, even at high engine load. Backflow is most pronounced in
regions C and E, although at lower engine load, the entire catalyst cross-section
experiences flow reversal.
During the beginning of the displacement phase (Fig. 5.17), the piston is
moving upwards, thereby expelling the remaining expanded gas from the cylin-
der. Due to the resonance phenomenon, the mean velocity fluctuates as shown
in Fig. 5.15. The resonance effect is strong enough to cause further backflow
during the displacement phase (Figs. 5.17 and 5.18). Although the piston veloc-
ity is roughly at a maximum in Fig. 5.18f, backflow is still observed, primarily
in region D. It is increasingly more pronounced at lower engine load.
Figure 5.23 depicts the time evolution of the velocity, locally averaged in
each of the five regions A through E of the catalyst cross-section, defined in
Fig. 5.22.
Regions C, D and E (Fig. 5.23b) are subject to a low time-average flow
rate. The velocity behaves very similar in those areas. The strongest flow
reversal occurs following each cylinder’s blowdown.
The velocity in the high flow rate regions A and B behaves rather differ-
ently (Fig. 5.23a). Peak velocities correspond to the opening of each cylinder.
Runners 1 (540 to 720 ◦ ca) and 2 (360 to 540 ◦ ca) issue near A. Runners 3 (0
to 180 ◦ ca) and 4 (180 to 360 ◦ ca) issue near B. Flow reversal also occurs in
regions A and B following blowdown phases of neighboring cylinders.
Experiments have been performed for several engine load conditions and
engine speeds ranging between 600 and 3000 rpm. The maximum flow rate de-
livered by the lab screw compressor prevented measurements at higher speeds.
Very similar flow patterns are observed throughout the engine operating range.
5.2.3 Numerical analysis

The numerical results discussed in this section are obtained using a one-
dimensional gas dynamic model of the exhaust system. The model uses a sec-
ond order total variation diminishing flux difference splitting scheme and fourth
order Runge-Kutta time integration. The model is implemented in Simulink11 ,
and briefly described below. Appendix D gives more details and validates the
model using benchmark problems.
Subsequently, the numerical results are discussed. Firstly, the results for
the exhaust manifold without cold end are described, i.e. with free discharge to
the atmosphere downstream of the catalyst. This situation is identical to the
experimental set up. Secondly, the results for an exhaust system are described
including a cold end, which consists of a single expansion chamber muffler and
a straight exhaust pipe.
Model description
Model basis: One-dimensional gas dynamics The one-dimensional Eu-
ler equations express the conservation of mass, momentum and energy in the
lengthwise z-direction. For unsteady, one-dimensional flow in a duct with vari-
able cross-section, including wall friction and heat transfer, the Euler equations
in conservative form are (Eq. (D.1)):
∂ (ρA) ∂ (ρU A)
+ = 0
∂t ∂z
2
∂ (ρU A) ∂ ρU A ∂ (pA) ∂A 4A U ρU 2
+ = − +p −f
∂t ∂z ∂z ∂z d |U | 2
∂ (ρEA) ∂ (ρU EA) ∂ (pU A)
+ = − + ρqA
∂t ∂z ∂z
where the quantities ρ [kg/m3 ], U [m/s], E [J/kg], p [Pa] denote average values
across the duct cross-section such that the conservation laws are fulfilled.
The duct is discretized in the streamwise z-direction by a n nodes, such that
the node spacing ∆z = L/(n − 1) (mm). Figure 5.24 shows the discretization
used by the model, which is of the cell-vertex type, i.e. the nodes coincide
with the edges of the control volumes. The Euler equations are solved in each
L
∆z
A
1 2 j j +1 n
Figure 5.24 – One-dimensional gas dynamic model: Discretization
node j, for each time step. Although the implemented model can handle non-
equidistant nodes, the node spacing is kept constant within each duct. Also,
only fixed time stepping is applied for the integration. The local cross-sectional
area A of the duct can vary along the duct length, as is e.g. the case for the
cold end with expansion chamber muffler.
To solve the Euler equations numerically, these coupled partial differential
equations should be transformed into an uncoupled set of scalar differential
equations. Appendix D.1.1 describes this procedure, resulting in the charac-
teristic Euler equations (Eq. (D.14)).
Appendix D.1.2 describes the discretization procedure that is applied to the
characteristic Euler equations. For this model, a second order total variation
diminishing (TVD) discretization scheme is applied, which is developed by
Vandevoorde [101]. Since the model is cell-vertex based, the flux balances are
fulfilled in the nodes and the conservation laws are better satisfied in case of
changes in the cross-sectional area A [101, 103]. The algorithm is given by
Eq. (D.23).
Appendix D.2 validates this model using some benchmark problems, similar
to the approach of Vandevoorde [101]. Appendix D.3 examines the spatial and
temporal resolution of the model in terms of the node spacing ∆z.
Model assembly: Hot end Figure 5.25 shows a diagram of the gas dy-
namic model of the exhaust system. The leftmost dashed box corresponds to
exhaust manifold B as depicted in Fig. 2.1. The model features four exhaust
runners, where runners 1 and 3 are longer than 2 and 4. Table 5.1 gives the
relevant geometrical specifications. Runners 1 and 2, and likewise 3 and 4 join
to form the two ‘joined runners’ in Fig. 5.25, each 60 mm long and with a
cross-sectional area twice that of a single runner. All runners are cell-vertex
discretized, according to the specifications in Table 5.1. The gas dynamics are
solved using the second order TVD model as described above.
As indicated in Table 5.1, the node spacing for each duct (except the ex-
haust pipe) is ∆z = 10 mm. According to each ducts length, this corresponds
31 The runner lengths given in Table 2.1 correspond to the distance between the cylinder
head flange and the diffuser. Each runner extends 100 mm into the cylinder head, up to the
exhaust valves seats. In the numerical model, this length is added to the values in Table 2.1.
exhaust runner 1
joined runner 1-2
exhaust runner 2
diffuser and catalyst exit cone

cold end
exhaust runner 3
cold end (and exit cone)
exhaust runner 4 joined runner 3-4
exhaust manifold B
Figure 5.25 – One-dimensional gas dynamic model of exhaust manifold B including

cold end
respectively to 27, 19 and 7 nodes for runners 1 and 3, runners 2 and 4 and the
joined runners.
As indicated in App. D.3.2, the required number of nodes per minimum
wavelength nλ = 16. Based on Eq. (D.31), the maximum resolved frequency is
fmax ' (c/∆z )/16 ' 2000 Hz for cold flow conditions (i.e. CME flow rig), or
fmax ' 4000 Hz for fired engine conditions.
The calculations are performed with wall friction, yet without heat transfer
or chemical reactions. The internal wall roughness k = 0.05 mm, correspond-
ing to welded steel pipes [74]. The friction factor f is determined using the
Colebrook-White relationship between f , Re and k/d (see Eq. (1.3)) for fully
Table 5.1 – Numerical model geometry

Component Diameter Length31 Nodes ∆z
mm mm - mm
Exhaust runners 1 and 3 28.0 260 27 10
Exhaust runners 2 and 4 28.0 180 19 10
Joined runners 1–2 and 3–4 39.6 60 7 10
Exhaust pipe 40.0 4000 200 20
developed turbulent flow.

Besides the wall friction losses, an additional pressure loss due to the runner
curvature is taken into account. This additional pressure loss K = ∆p/pdyn
is distributed throughout the runner length by increasing the friction factor
feff = f +Kd/(4L) , since this local bend loss is difficult to assign to a particular
area of the runner.
The values for the interacting bend pressure loss are obtained from
Miller [74]. Runners 1 and 3 are considered as two interacting 90 ◦ bends
with radius of curvature rc = 2d at an out-of-plane combination angle of 90 ◦ ,
which yields a total pressure loss coefficient K ' 0.48. Runners 2 and 4 are
considered as single 90 ◦ bends with radius of curvature rc = 1d, which yields
K ' 0.50. The joined runner 1–2 is considered as a single 75 ◦ bend with
radius of curvature rc = 1d, which yields K ' 0.45. The joined runner 3–4
is considered as a single 90 ◦ bend with radius of curvature rc = 0.8d, which
yields K ' 0.78.
The cylinders, diffuser and exit cone are modeled as zero-dimensional com-
pressible volumes with conservation of mass and energy. The working fluid
(air) is assumed an ideal gas. As such, the following equations are solved in
these volumes (Eq. (D.24)):
∂m X X
= ṁin − ṁout
∂t
∂ (mcv T ) X X 1 ∂V
= (ṁH)in − (ṁH)out − mrT
∂t V ∂t
The volume and exhaust valve lift for each cylinder is time-dependent.
Equations (2.11) and (B.7) give the instantaneous cylinder volume Vcyl (ωt)
and exhaust valve lift h0e (ωt), where the crankshaft position for each cylinder
is phased to correspond to the engine firing order (see Table 2.1).
The states in each cylinder are only computed while the exhaust valves are
open. In that case, the inlet boundary condition to the runner corresponds to a
restricted flow end. The discharge coefficient Cd for the exhaust valves is taken
according to Heywood [47] (see Eq. (B.6) and Fig. B.1). When the valves are
closed, the runner is terminated by a closed end. The initial conditions of the
cylinder are set to the residual cylinder pressure and temperature corresponding
to the experimental case.
The diffuser and exit cone were originally modeled as one-dimensional pipes
with a high aspect ratio (d/L > 1), forming tapered cross-sectional area transi-
tions between the catalyst monolith and the connecting pipes. To speed up the
simulation, these blocks are instead modeled as zero-dimensional compressible
volumes, with conservation of mass and energy. This does not noticeably affect
the results.
The flow in the catalyst substrate is modeled as incompressible gas. The
inertia of the gas is taken into account, although this effect is small due to
the large cross-sectional area ratio associated with the catalyst (see Table 2.1).
The wall friction inside the catalyst corresponds to developing laminar flow,
including the entrance and exit losses, as described in App. A.2.
Time integration uses a fourth order Runge-Kutta scheme with fixed time
stepping. The fixed time step ∆t [s] is determined a priori, based on a CFL-
number of 0.5 (see comment below) and assuming a maximum characteristic
eigenvalue λmax = 2c, which results in the expression ∆t = CF L ∆z/(2c) .
After the simulation, the maximum actual CFL number is verified over all the
nodes.
Courant-Friedrichs-Lewy (CFL) number The CFL number is a

dimensionless time step applied to integrate the equations. It follows
from the Courant-Friedrichs-Lewy (1928) stability condition which ap-
plies to finite difference schemes for hyperbolic partial differential equa-
tions. The CFL condition states that the numerical dependence domain
should contain the physical dependence domain. In terms of the eigen-
value discussion in App. D.1.1, the time step ∆t should be smaller than
the time it takes the characteristic of the largest eigenvalue λmax to travel
along one node spacing ∆z. Symbolically, this yields:
∆t ∆t
CF L = max (|λ|) = max (|U ± c|) <1 (5.2)
∆z ∆z
Figure 5.25 shows the exhaust system model including cold end. A second
model has been used that corresponds to the experimental set up. In that case,
the rightmost dashed box in Fig. 5.25 is not modeled, and only the exhaust
manifold is modeled without exit cone and cold end, i.e. with free discharge to
the atmosphere.
Model assembly: Cold end As shown in Fig. 5.25, the entire cold end
is modeled as a single exhaust pipe (4 m long, 40 mm in diameter), with
a single expansion chamber muffler which corresponds to Muffler 2 in Davis
et al. [33] (24 inch long chamber, area ratio m = 16). The muffler is located in
the center of the exhaust pipe. The cold end is modeled as such a simple case
instead of the actual exhaust system, since the purpose is merely to determine
the significance of the cold end’s influence on the catalyst velocity distribution.
The exhaust pipe is terminated with an open end boundary condition at at-
mospheric conditions. No spherical radiation boundary condition is considered
for simplicity, since for low frequencies (f < 500 Hz) the reflection coefficient
of a spherical radiator approximates unity, similar to the open end boundary
condition [20, 33].
The exhaust pipe is modeled using 200 nodes, corresponding to a node
spacing ∆z = 20 mm. Using Eq. (D.31), this amounts to a frequency resolution
fmax = 1000 Hz. For reasons of calculation time, the cold end node spacing is
larger than the spacing in the hot end.
The exhaust pipe wall roughness k = 0.025 mm, corresponding to smooth
steel pipe [74]. The maximum pressure drop in the exhaust pipe (for fired
engine conditions, full load and 6000 rpm) is approximately 100 mbar, which
agrees with typical exhaust systems.
Figure 5.26 shows the numerically determined admittance Y /Y0 of the ex-
haust pipe, (a) with and (b) without muffler. The admittance is defined in
Admittance of exhaust pipe Admittance of exhaust pipe

Pipe: ∅ 40 mm x 4 m, Muffler: m = 16, L = 24" Pipe: ∅ 40 mm x 4 m, No muffler
40 40
Amplitude (dB) of ρcUi/pi
Amplitude (dB) of ρcUi/pi

Theoretical (L1) Theoretical
Numerical Numerical
20 20
0 0
−20 −20
−40 −40
0 80 160 240 320 400 480 560 640 720 0 80 160 240 320 400 480 560 640 720
Frequency f (Hz) Frequency f (Hz)
(a) (b)
Figure 5.26 – Admittance of the cold end (a) with muffler and (b) without muffler
Periodicity
10
Catalyst velocity deviation U − U∞ (m/s)
−2
−4
1st cycle 2nd cycle 3rd cycle
−6
0 720 1440 2160
Figure 5.27 – Periodicity of the numerical solution, based on the catalyst mean
velocity deviation U − U∞
App. D by Eq. (D.28). Figure 5.26b shows the same behavior as observed
in App. D.2.2. Upon introducing the muffler, Fig. 5.26a shows that the ad-
mittance corresponds roughly to the admittance of the pipe upstream of the
muffler, measuring L1 = (2 − 24 · 0.0254)/2 ' 1.7 m in length.
Periodicity Figure 5.27 shows the difference between the calculated mean
catalyst velocity U (ωt) at an increasing simulation time t and the mean catalyst
velocity obtained after several engine cycles U∞ (ωt). Figure 5.27 demonstrates
that a periodic solution is attained after two engine cycles. In the following
sections, simulations have been performed on an exhaust model including a
cold end. For those simulations, a periodic solution is only attained after four
engine cycles, due to the larger residence time scale.
Results without cold end: Free discharge
Isochoric flow rig The results discussed in this section are obtained for a
free atmospheric discharge after the catalyst (i.e. without cold end), as is the
case for the experiments on the CME flow rig. The following section discusses
the influence of the absence of the cold end, based on numerical simulations.
Figures 5.28 and 5.29 compare the numerical results obtained with the
gas dynamic exhaust system model to the experimental results. Each plot
shows the numerical catalyst velocity Um,num (ωt) as a dashed line ( ), and
the corresponding experimental catalyst velocity Um,exp (ωt) as a solid line
( ). These experiments are obtained on the isochoric (CME) flow rig, and
correspond to those discussed in Sect. 5.1.2.
The one-dimensional model seems to give an adequate prediction of the
flow dynamics in the catalyst, even though the flow in the actual manifold (see
Fig. 2.1b) may be expected to be significantly three-dimensional.
The magnitude of the flow reversal following each blowdown is generally
greater than the one observed experimentally. However, one should keep in
mind the limitations of the oscillating hot-wire anemometer (see Chap. 3).
The OHW features a maximum measurable negative velocity of approximately
−1 m/s. If the local time-resolved velocity U (x, y, ωt) drops below −1 m/s,
the resulting measured mean velocity Um (ωt) deviates from the actual value.
These measurement errors are denoted rectification or folding errors. When
using standard hot-wire anemometry, the folding occurs at zero velocity. In
the case of OHW, the folding occurs at −1 m/s. This folding effect may be
noted in Fig. 5.28b, c and d ( ).
Figure 5.30 illustrates this problem for two hypothetical cases with identical
mean velocity Um = −0.5 m/s. For the uniform distribution in Fig. 5.30a, the
OHW obtains the correct mean velocity, whereas for the peak distribution in
Fig. 5.30b, errors occur in the outer region where U < −1 m/s. In the latter
case, the OHW overestimates the actual mean velocity. This causes the local
unphysical increases in the measured mean velocity ( ) in Fig. 5.28b, c and d
when the simulated mean velocity ( ) becomes negative.
The flow uniformity ηm (ωt) in combination with the mean velocity Um (ωt)
in Figs. 5.11a and 5.15b, shows that the OHW errors occur mostly following
the blowdown phase, when the flow uniformity is quite low for high engine load.
This effect also shows up in the velocity distribution in Fig. 5.16b (top), which
exhibits a very low flow uniformity value.
Based on the numerical results, the magnitude of the flow reversal does
not decrease for increasing engine load. In fact, the opposite is true upon
comparison of Figs. 5.28b,c,d for N = 1200 rpm, and Figs. 5.29a,b for N =
1800 rpm. The magnitude of the flow reversal does decrease for increasing
engine speed (compare Figs. 5.28d, 5.29b and c).
The velocity fluctuations during the displacement phase are well captured
by the simulation. Upon close examination of Fig. 5.28a, the fluctuation fre-
quency corresponds very well, yet the magnitude of the fluctuations does not
entirely agree. However, judging by the excellent agreement of the fluctua-
Time−resolved Time−resolved
5 5
Um,exp Um,exp
4 Um,num 4 Um,num
3 3
Mean velocity (−)
Mean velocity (−)

2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(a) (b)
5 5
Um,exp Um,exp
4 Um,num 4 Um,num
3 3
Mean velocity (−)
Mean velocity (−)
2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(c) (d)
Figure 5.28 – Time-resolved numerical and experimental catalyst velocity

Um (ωt) [-], for (a) N = 600 rpm (zero load), (b) N = 1200 rpm (zero load),
(c) N = 1200 rpm (part load), (d) N = 1200 rpm (high load)
5 5
Um,exp Um,exp
4 Um,num 4 Um,num
3 3
Mean velocity (−)
Mean velocity (−)

2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(a) (b)
5 5
Um,exp Um,exp
4 Um,num 4 Um,num
3 3
Mean velocity (−)
Mean velocity (−)
2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(c) (d)
Figure 5.29 – Time-resolved numerical and experimental catalyst velocity

Um (ωt) [-], for (a) N = 1800 rpm (part load), (b) N = 1800 rpm (high load),
(c) N = 2400 rpm (high load), (d) N = 3000 rpm (part load)
Um = -0.5 m/s, Um’ = -0.5 m/s Um = -0.5 m/s, Um’ > -0.5 m/s
5 5
U (true) U (true)
U’ (measured) U’ (measured)
4 4
3 3
2 2
U (m/s)
U (m/s)
1 1
0 0
-1 -1
-2 -2
Cross-section Cross-section
(a) (b)
Figure 5.30 – OHW error for velocity distributions with small area of extensive flow
reversal
tion frequencies, the numerical model may be expected to be quite helpful in

explaining this resonance phenomenon. The same model is therefore used in
Sect. 5.3.
For most conditions in Figs. 5.28 and 5.29, the magnitude of the displace-
ment phase fluctuations in the numerical catalyst velocity is greater than the
measured fluctuations. However, for the particular case of Fig. 5.28b, the re-
verse is true. In the numerical model, the fluctuations may be decreased by
increasing the damping due to wall friction and additional pressure losses. In
particular, the pressure loss due to the runner curvature is difficult to predict.
In the numerical model, the bend losses are distributed throughout the runner
lengths. Furthermore, the loss coefficients are based on correlations [74]. These
incorporate the effect of bend-to-bend interactions, yet they also assume that
the bend combinations are connected to upstream and downstream pipes of
quasi-infinite length. An improved agreement might be obtained between sim-
ulations and measurements by experimentally determining the loss coefficients
of the individual runners using a stationary flow bench. However, this has not
been attempted within the scope of this work.
In conclusion, the correspondence between numerical and experimental re-
sults seems satisfactory, and of similar quality to results obtained by other
researchers using commercially available one-dimensional gas dynamic codes.
Benjamin et al. [14] present LDA measurements downstream of a close-coupled
catalyst, in a fired engine at 2000 rpm and high load. The experimental re-
sults are compared to a transient CFD simulation, coupled with a commercial
one-dimensional gas dynamics code.
Figure 5.31a shows a comparison between measured ( ) and calculated
( ) runner velocity. Figure 5.31b shows the correspondence between numer-
ical and experimental results for the velocity at a single point in the catalyst.
(b)
(a)
Figure 5.31 – Measured and simulated (a) runner and (b) catalyst velocity for fired
engine (Source: [14])
The simulation results predict much stronger backflow than the experimental
observations. Upon assessing the results in Fig. 5.31b, one should take into
account that these are single-point results. Benjamin et al. [14] show no plot
of the time-resolved mean velocity, for which the results are likely in better
agreement.
Fired engine Figure 5.32 shows a comparison of the numerical results for
(a) fired engine conditions and (b) CME flow rig conditions, both with free
discharge to atmosphere after the catalyst (i.e. without cold end). Both cases
are for identical engine speed (N = 1200 rpm) and high engine load. For the
fired engine case, the residual cylinder pressure and temperature is 3 atm and
800 ◦ C, resulting in a temperature of approximately 600 ◦ C in the hot end. For
fired engine conditions, air is used as working fluid instead of exhaust gas. No
heat transfer from the gas to the walls and surroundings is modeled.
There is quite some difference in the flow dynamics between both cases,
which is mainly due to the difference in temperature. Firstly, the ratio of peak
blowdown flow rate to peak displacement flow rate (respectively denoted M1
and M2 in Sect. 2.3.2) is significantly greater for fired engine conditions. This
also resulted from the exhaust stroke flow similarity, analyzed in Sect. 2.3 and
App. B. The limitations of the CME flow rig with regard to the maximum
intake system pressure have been indicated in Sect. 2.3.3.
The magnitude of the flow reversal following blowdown is greater for fired
conditions. This is due to the higher peak blowdown flow rate M1 . The blow-
down exhaust gas pulse contains more momentum compared to cold flow condi-
tions, thereby creating a stronger expansion wave near the end of the blowdown,
which in turn causes the increased flow reversal.
Secondly, the velocity fluctuations during the displacement phase are char-
acterized by a frequency which is approximately twice the value observed on
the CME flow rig. This is simply the temperature effect on the speed of sound.
As explained in Sect. 5.3, the resonance phenomenon is caused by a Helmholtz
resonance in the hot end. The resonance frequency f0 is proportional to the
5 5
N = 1200 rpm, Q = 99.3 m3/h, high load (fired conditions) N = 1200 rpm, Qref = 97.3 m3/h, pi = 2.20 atm
Um,num Um,exp
4 4 Um,num
3 3
Mean velocity (−)
Mean velocity (−)

2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(a) (b)
Figure 5.32 – Time-resolved numerical catalyst velocity Um (ωt) [-] for N =

1200 rpm (high load) without cold end, in (a) fired engine and (b) CME flow rig
conditions
√
speed of sound c ∝ T .pAs such, the ratio of resonance frequencies is expected
to be f0,f ired /f0,cold = Tf ired /Tcold ' 1.72. According to the numerical sim-
ulation, the observed frequency ratio varies between 1.54 and 1.70, which is in
good agreement with the above assumption.
Results with cold end

The purpose of this section is to assess the influence of the absence of the
cold end on the manifold flow dynamics in general, and the occurrence of flow
reversal in the catalyst in particular.
The first paragraph discusses the influence of including a numerical cold end
model into the one-dimensional exhaust model, for the cold flow conditions
experienced on the isochoric (CME) flow rig. The second paragraph shows
the effect of fired engine conditions on the flow dynamics, determined using the
numerical model. No experimental data are available for comparison. However,
the numerical model has proven its validity, based on the good agreement
between numerical and experimental results observed in Sect. 5.2.3.
Isochoric flow rig Figure 5.33 compares the time-resolved catalyst velocity
obtained using the numerical model, with and without cold end. The case
without cold end ( ) corresponds to a free discharge to the atmosphere after
the catalyst, the same situation as for the experiments on the isochoric flow
rig. In the case with cold end ( ), the exhaust pipe described in Sect. 5.2.3
is included in the model.
As shown in Fig. 5.33, the presence of the cold end influences the catalyst
velocity to some extent, albeit rather limited. In terms of the flow reversal
5 5
Um,num Um,num
4 Um,num (with cold end) 4 Um,num (with cold end)
3 3
Mean velocity (−)
Mean velocity (−)

2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(a) (b)
5 5
Um,num Um,num
4 Um,num (with cold end) 4 Um,num (with cold end)
3 3
Mean velocity (−)
Mean velocity (−)
2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(c) (d)
Figure 5.33 – Time-resolved numerical catalyst velocity Um (ωt) [-], with and with-
out cold end, for (a) N = 1200 rpm (part load), (b) N = 1200 rpm (high load),
(c) N = 1800 rpm (part load), (d) N = 1800 rpm (high load)
5 5
Um,num (with cold end) Um,num (with cold end)
4 Um,num (with cold end, increased friction) 4 Um,num (with cold end, without muffler)
3 3
Mean velocity (−)
Mean velocity (−)

2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(a) (b)

1800 rpm (high load) with cold end, (a) with increased (×2.5) wall friction, (b) with
exhaust pipe excluding muffler
observed without cold end, the presence of the cold end does not seem to
reduce the backflow magnitude. In case of Fig. 5.33a and b, the opposite is
even true. During the displacement phase, the velocity fluctuations are slightly
affected by the presence of the cold end, mostly in terms of the fluctuation
frequency and not so much for the amplitude.
In general, the cold end does not significantly affect the flow dynamics in the
catalyst. However, this conclusion is based only on the simulations performed
using the present gas dynamic model of the exhaust system, with the simple
expansion chamber muffler. In truth, a more complex exhaust system will be
present in the real engine configuration, although no attempt was made to
simulate a realistic cold end in this thesis.
Figure 5.34 demonstrates the effect of varying parameters of this simple cold
end model. Figure 5.34a shows the effect on the catalyst velocity of increasing
the wall friction factor f by a factor of 2.5 ( ) with respect to the reference
case ( ), at least for the governing flow conditions at N = 1800 rpm and high
engine load. The increased friction factor is obtained by artificially increasing
the wall roughness from a value k = 0.025 mm to k = 1 mm, corresponding to
a heavily rusted steel pipe [74].
By increasing the wall friction, the stationary (i.e. low frequency, f → 0)
flow resistance of the cold end is increased without changing its reactivity at
higher frequency. In terms of the admittance Y /Y0 , the DC-component is
reduced, while the higher frequency characteristics are unchanged.
Figure 5.34b shows the effect on the catalyst velocity of removing the muffler
from the cold end, resulting in a 4 m long straight exhaust pipe. In this case, the
higher frequency characteristics of the admittance Y /Y0 are changed without
significantly changing the DC-component.
Time-resolved Time−resolved
5 5
N = 1200 rpm, Q = 93.7 m3/h, high load (fired conditions) N = 1200 rpm, Qref = 97.3 m3/h, pi = 2.20 atm
Um,num (with cold end) Um,num
4 4 Um,num (with cold end)
3 3
Mean velocity (−)

Mean velocity (-)
2 2
1 1
0 0
-1 −1
-2 −2
0 180 360 540 720 0 180 360 540 720
(a) (b)

1200 rpm (high load) with cold end, in (a) fired engine and (b) CME flow rig condi-
tions
As demonstrated by Figs. 5.34a and b, the influence of altering the cold end
characteristics is most significant during the displacement phase, affecting both
the fluctuation frequency and amplitude. Neither the backpressure of the cold
end, nor the higher frequency reactivity significantly affect the flow dynamics
in terms of the magnitude of the flow reversal following the blowdown phase.
Fired engine A comparison between Figs. 5.35a and 5.32a reveals an ap-
preciable influence of the presence of the cold end in fired engine conditions,
in particular during the displacement phase. This contrasts with Fig. 5.35b
obtained in cold flow conditions, where only a small influence can be detected
between the dashed line ( ) (with cold end) and the solid line ( ) (without
cold end).
As indicated by Fig. 5.32, in the absence of a cold end, the velocity fluc-
tuations appear at a higher frequency in fired engine conditions. The ratio of
fluctuation frequencies between fired and cold conditions is determined by the
square root of the temperature ratio (see Sect. 5.2.3).
This is not true upon introducing the cold end. In this case, the hot end
interacts with the cold end, resulting in very different gas dynamics and dif-
ferent resonance frequencies. Based on Fig. 5.35b, one might have concluded
that the hot end gas dynamics can be considered decoupled from the cold end
dynamics. However, the difference between Figs. 5.35a and 5.32a shows that
this is not a valid assumption, at least not in fired engine conditions.
Park et al. [81] present phase-locked velocity results obtained using LDA in
a close-coupled catalyst manifold on a fired engine. Figure 5.19 shows periodic
flow reversal in the order of −1 to −2 m/s, that occurs following each blowdown.
Also using LDA, Liu et al. [70] show flow reversal occurring downstream of
a close-coupled catalyst in motored engine conditions, albeit not in fired engine

conditions. Transient simulations do show flow reversal in motored and fired
conditions. Equivalently, flow reversal is most pronounced following blowdown.
The maximum backflow varies between −1 and −5 m/s in motored and fired
conditions.
Conclusion
Reviewing the numerical results discussed in the preceding sections, the gas
dynamic model has improved the understanding of the flow dynamics in the
close-coupled catalyst manifold. Particularly, (i) the effect of the cold end and
(ii) the effect of fired engine conditions have been studied numerically.
In cold flow conditions (corresponding to the isochoric CME flow rig), no
significant influence is noted between the flow dynamics in hot end and cold
end. However, the influence proves much stronger in fired engine conditions.
Based on these numerical results for the time-resolved catalyst velocity, the
experiments appear to be valid in cold flow conditions, regardless of the pres-
ence of the cold end, or the related backpressure. In fired engine conditions, the
temperature and speed of sound is higher, which invariably increases the res-
onance frequencies. Therefore, a straightforward extrapolation to fired engine
conditions cannot be made based on these measurements.
Finally, periodic flow reversal following the blowdown phase is noted in
all operating conditions, (i) fired engine or cold flow conditions, (ii) with and
without cold end, (iii) for two typical variations in the cold end characteristics
(i.e. backpressure and attenuation). For the examined cases, the magnitude
of the flow reversal is minimal corresponding to the experimental conditions
on the CME flow rig. Therefore, in realistic conditions (i.e. fired engine with
cold end), the periodic flow reversal is expected to be stronger than observed
experimentally on the CME flow rig.
5.2.4 Discussion: Physical relevance of catalyst flow re-

versal
Catalytic reactions in automotive catalysts are exothermic. In fired engine
conditions, the temperature of the gas downstream of the catalyst is therefore
higher than upstream. Periodic backflow is observed on the CME flow rig, and
is also reported in fired engine conditions by other authors [4, 70, 81, 59]. In
case of backflow, part of the hot processed gas is recycled through the catalyst,
into the collector-diffuser upstream of the catalyst.
As pointed out in Sect. 1.2.1, thermal degradation of the catalyst is a com-
plex process that mainly involves loss of catalytic surface area by sintering,
due to excessive catalyst temperature [11, 38]. Sintering rate increases expo-
nentially with temperature, so even a small temperature change may signifi-
cantly affect local catalyst degradation. Regions of high flow rate (A and B in
Fig. 5.22) are subject to higher chemical degradation due to poisoning. Flow
reversal in these regions (Fig. 5.23a) may impose an additional thermal load.
The walls of the exhaust manifold are already subject to severe cyclic ther-
mal loads. Manifold design aims to avoid local hot spots, especially on sensitive
areas such as welds. Catalyst flow reversal in near-wall regions (D and E in
Fig. 5.22 and Fig. 5.23b) may raise the time-averaged gas temperature, thereby
adding to the existing thermal load on wall materials immediately upstream of
the catalyst.
Background: Reverse flow catalytic reactors The chemical pro-
cess industry has recently seen the introduction of so-called reverse flow
catalytic reactors (RFCR). These are mainly applied for endothermic or
slightly exothermic reactions (for air purification: e.g. oxidation of hy-
drocarbons, carbon monoxide and sulphur dioxide; for synthesis: partial
oxidation of methane, synthesis of methanol and ammonia).
A standard unidirectional catalytic reactor requires preheating for such
reactions. By applying periodic reversal of inlet and outlet sides, the end
sections of the catalyst act as regenerative heat exchangers. The cata-
lyst itself therefore does part of the preheating. If operating conditions
are well controlled, the catalyst center may obtain much higher temper-
atures compared to unidirectional operation. The RFCR has not been
in widespread use. Its complex dynamics require an advanced control
system to prevent extinction of the reaction.
Mitri et al. [76] discuss experimental results of thermal deactivation in
RFCR operation, for partial oxidation of methane. The average catalyst
temperature is 300 to 400 ◦ C higher compared to unidirectional opera-
tion. A significant increase in thermal deactivation is noted.
Eigenberger and Nieken [35] discuss the influence of several parameters
including the reversal frequency on the maximum catalyst temperature
and reactor stability. The reactor temperature increases for increasing
reversal frequency.
Results from these studies [76, 35] cannot be extrapolated to the present
case. Flow reversal observed in the automotive catalyst is characterized
by (i) higher reversal frequencies (50 to 200 Hz, compared to 0.01 to
1 Hz), (ii) lower reverse flow magnitude. Furthermore, (iii) the processed
gas is recycled yet contains less reactants, whereas an RFCR is fed with
fresh mixture in both forward and reverse flow.
Based on the observed velocity magnitude of the flow reversal in the close-
coupled catalyst manifold, the traveled distance of the hot exhaust gas may be
estimated as it passes through the catalyst. For a negative velocity magnitude
|Urev | ' 1 m/s and a catalyst length of L = 137 mm, the space velocity V =
U /L equals 7.3 s−1 , corresponding to a characteristic travel time of 137 ms.
Thus, this is a rather large timescale compared to the flow pulsation period Tp .
The traveled distance is proportional to the product of the negative velocity
magnitude |Urev | and the flow pulsation period Tp , which yields values in the
order of 10 mm for the traveled distance. This is small compared to the catalyst
monolith length of 137 mm.
Nevertheless, an automotive close-coupled catalyst operates at the outer
edge of permissible material constraints. Exhaust gas temperatures reach in
excess of 1000 ◦ C during high engine load. Any process causing even a minor
increase or decrease in thermal load is worth investigating. Further research is
needed to establish the effect of flow reversal as observed in the current research
on the catalyst and manifold wall temperatures.
5.3 Helmholtz resonance

5.3.1 Introduction
Figure 5.2b shows the first encounter with an unexpected and initially unex-
plained phenomenon. Due to the high bandwidth of the hot-wire anemometer
and the applied phase-locked measurement technique, strong fluctuations are
observed in the time-resolved mean catalyst velocity.
Except for the isothermal flow rig with rotating valve, this phenomenon is
observed for both exhaust manifolds A and B, and for both isothermal and
isochoric flow rigs. For the isochoric (CME) flow rig, the fluctuations are much
stronger compared to the same manifold mounted on the isothermal flow rig.
Upon the first encounter with these fluctuations, the possibility of mechani-
cal vibrations was considered. The hot-wire probe is mounted on an automated
traversing system, which is mounted on the lab floor, adjacent to the flow rig.
Both flow rig and traversing system are mounted as rigidly as possible to the
floor, using thick threaded supports. These threaded supports are also used to
level the flow rig and traversing system perfectly horizontally.
Any significant relative motion between the velocity probe and the exhaust
manifold results in an error velocity ∆U . Using directionally sensitive ac-
celerometers32 , the accelerations of the flow rig, exhaust manifold and travers-
ing system were measured, phase-locked with the flow pulsation. The relative
velocity ∆U is determined
R from the integrated difference of the acceleration
signals, or ∆U = t (a − a0 ) dt, where the probe acceleration a0 is negligible
compared to the exhaust manifold acceleration a a0 ' 0.
For the rotating valve, the operation is virtually vibrationless, due to the
absence of any inertial forces. The cylinder head exhibits some detectable vi-
bration, although completely insignificant compared to an engine with moving
pistons. For both pulsator devices, the resulting error velocity ∆U correspond-
ing to the relative motion between manifold and hot-wire probe is found to be
smaller than 10 mm/s, with peak frequencies not exceeding 25 Hz. As such,
∆U is marginal compared to the observed mean velocity fluctuations, which
are of the order of 1 m/s and higher.
The following sections explore the origin of the observed fluctuations. Sec-
tion 5.3.2 provides an analytical description of the low-frequency character-
istics of a gas dynamic system, introducing the Helmholtz resonance. Sec-
tion 5.3.3 discusses the velocity fluctuations observed experimentally on the
isothermal and isochoric (CME) flow rigs. Section 5.3.4 discusses the frequency
32 PCB Piezotronics type 356B08 three-dimensional accelerometer and ICP amplifier.

5.3 Helmholtz resonance 205
response function of the close-coupled catalyst manifold, by means of the one-

dimensional gas dynamic model used in Sect. 5.2.3 and described in App. D.
5.3.2 Analytical model

A Helmholtz33 resonator [20, 33] consists of a volume connected to a pipe (as
depicted in Fig. 5.36), filled with a compressible fluid. At low frequencies
(f < c/(2L) , where c is the speed of sound and L is the pipe length), this
gas dynamic system behaves approximately as a second order spring-and-mass
mechanical system. The gas in the pipe behaves as an incompressible oscillating
plug, with a mass m [kg] equal to:
m = ρAL (5.3)
where A and L are the cross-sectional area [m2 ] and the length [m] of the pipe.
The compressible gas in the volume V [m3 ] is characterized by a mechanical
spring constant k [N/m] equal to:
k = γpA2 /V (5.4)
where γ is the ratio of specific heats [-]. In Eqs. (5.3) and (5.4), p, ρ and γ
are evaluated at a mean reference condition. This system features an eigenfre-
quency fH [Hz]:
r r
1 k 1 c AL
fH = = (5.5)
2π m 2π L V
√
where c = γrT is the speed of sound [m/s].
In acoustics, the term ‘Helmholtz resonator’ denotes a cavity consisting of
a closed volume and a pipe (or ‘neck’) which is perpendicular to the main duct
(see e.g. Beranek [20], Davis et al. [33], Boonen [21]). While the duct may
experience a net flow, the cavity neck experiences only an oscillating flow with
zero mean. Such resonators are used as narrow-band acoustic dampers (e.g.
in silencers for reciprocating machinery, such as the simple muffler shown in
Fig. 5.37), or as reference tone sound sources in musical instruments (by means
of some aero-acoustic or aero-elastic sound generation).
As noted by Davis et al. [33], the effective oscillating mass in the resonator
neck is increased by a certain amount, due to the entrainment of surrounding
nearby gas on either side of the neck (see Fig. 5.36). This effect is taken into
account by using an effective length in Eq. (5.3), usually defined as:
Leff = L + 2 β d (5.6)
33 Hermann Ludwig Ferdinand von Helmholtz (◦ 31 Aug. 1821 –
† 8 Sept. 1894) was a German physician and physicist. Besides im-

portant contributions to thermodynamics (e.g. Über die Erhaltung der
Kraft, 1847), Helmholtz is famous for his 1863 book Die Lehre von den
Tonempfindungen als physiologische Grundlage für die Theorie der Musik
and of course, for inventing the Helmholtz resonator to show the height
of various tones. (Source: Wikipedia, Universität Heidelberg)
Leff L A
V
p, T
Figure 5.36 – Helmholtz resonator Figure 5.37 – Helmholtz resonator as

nomenclature exhaust muffler (Source: [33])
where d is the diameter of the neck pipe and β is an empirical constant, typically
in the range 0.3 < β < 0.43 [20, 33].
In close-coupled catalyst manifolds, the same resonating behavior is ob-
served, yet with net flow through the resonator. In general, the Helmholtz
frequency fH denotes the zeroth order resonance frequency of a gas dynamic
system. This is the interpretation that should be kept in mind when the term
Helmholtz resonance is used in this thesis.
Considering the complexity of the exhaust system, several resonating com-
binations of masses and springs may be identified. For instance, any volume
with negligible momentum may be considered as a spring (e.g. cylinders, dif-
fuser and exit cone, closed exhaust runners), and any pipe with a significant
momentum may be considered as a mass (open exhaust runners, exhaust pipe).
The combined exhaust system features multiple connections, which are further
complicated by damping components (e.g. exhaust valves, catalyst substrate,
pipe bends and junctions).
Figure 5.38 shows an equivalent mechanical spring and mass system for the
low frequency gas dynamics in the close-coupled catalyst exhaust manifold.
The diagram shows the volumes as springs kcyl , kd , and the pipes as masses
mr , mcat . The factors σd and σe represent the area ratios of the diffuser
and exit cone. Figure 5.38a represents a free discharge, corresponding to the
experimental set up. Figure 5.38b includes the exit cone and the cold end as a
‘black box’ impedance Zcold end (f ).
For each spring, an effective volume should be considered. Typically, kcyl
corresponds to the cylinder volume at mid position in between bottom and top
dead center (see Boonen [21]), or:
πb2

1 1
Vcyl,eff = + s (5.7)
2 %−1 4
where b and s are the bore and stroke [m], and % is the volumetric compres-
sion ratio [-]. The factor 1/(% − 1) is the relative dead volume. The volume
corresponding to kd may be considered the diffuser volume Vd along with part
of the combined volume of the closed runners, or:
mcat
1/ σd
k cyl mr kd
p cyl
Ur pd U
V cyl A r, L r Vd
open runner
closed runner
closed runner
closed runner
(a)
mcat
1/ σd σe
k cyl mr kd ke Z cold end

p cyl
Ur pd U
Z cold end
V cyl A r, L r Vd
open runner
closed runner
closed runner
closed runner
(b)
Figure 5.38 – Equivalent lumped parameter model of the low frequency gas dy-
namics in a close-coupled catalyst exhaust manifold, (a) without and (b) with cold
end
Vd,eff = Vd + α (nr − 1) Ar Lr (5.8)

where nr is the number of runners and 0 6 α 6 1.
Similarly, effective lengths should be used for each mass, by introducing
a correction factor β as described above [33]. Not shown in the mechanical
diagram of Fig. 5.38 are the different damping components corresponding to
flow resistances (e.g. exhaust valves, catalyst substrate friction, pipe bends and
wall friction).
In Fig. 5.38, the quantity of interest is the catalyst velocity U [m/s]. The
driving quantity is the piston displacement, which creates a volume velocity
Qcyl = −dVcyl /dt [m3 /s]. In Sect. 5.3.4, the frequency response function
from ‘input’ Qcyl to ‘output’ U is determined numerically, by means of the
one-dimensional gas dynamic model of the exhaust system previously used in
Sect. 5.2.3.
To establish which system components are in resonance, and cause the ob-
served velocity fluctuations, simultaneous phase-locked velocity and/or pressure
data are required at more than one location, thereby revealing the phase dif-
ference between the signals. In the experimental set up, besides the catalyst
velocity, also the velocity at the entrance to runner 1 has been measured, as
well as the pressure in cylinder 1 and the diffuser pressure.
The following section describes the experimental observations, and the ac-
tions taken to identify the resonating components.
5.3.3 Experimental results

Figure 5.39 gives a comparison at equal engine speed (N = 1200 rpm) between
the time-resolved mean velocity on (a,b,c) the CME flow rig and (d) the isother-
mal flow rig. Only a minor difference can be observed in Fig. 5.39d between
minimum and maximum flow rate (Qref ' 50 . . . 115 m3 /h). This comparison
is presented here since an isothermal flow rig approach is used by numerous
authors [88, 17, 69, 22, 53] for studying pulsating flow in exhaust systems with
close-coupled catalyst, in spite of the appreciably different flow dynamics.
Figure 5.39 shows the time-resolved velocity in runner 1 Ur=1 (ωt) [-] as a
dashed line ( ). Top dead center of cylinder 1 corresponds to 0 ◦ ca. Consid-
ering the engine’s firing order (see Table 2.1), the plots show the consecutive
exhaust strokes of cylinders 3, 4, 2 and 1, respectively. The runner velocity
is measured at the inlet of runner 1 using a hot-film sensor, mounted flush
with the inner wall. Ur=1 (ωt) is measured in a single point, and as such it
is only indicative of the mean runner velocity. It is only used to determine
the phase lag between runner and catalyst velocity, with respect to the reso-
nance phenomenon. The runner velocity measurement is not available for all
experiments, which explains why it is only shown in Fig. 5.39a and b.
Background: Measuring the runner velocity Initially, the runner

velocity has been measured using a hot-wire probe, which could be tra-
versed along a straight line. However, such a probe is very prone to wire
7 7
6 Um(θ) 6 Um(θ)
Ur=1(θ) Ur=1(θ)
5 5
Mean velocity (-)
Mean velocity (-)

4 4
3 3
2 2
1 1
0 0
-1 -1
0 180 360 540 720 0 180 360 540 720
(a) (b)
7 7
N = 1200 rpm, Qref = 50.2 m3/h, pi = 1.00 atm N = 1200 rpm
Qref = 117.1 m3/h
6 Um(θ) 6 Qref = 43.3 m3/h
5 5
Mean velocity (-)
Mean velocity (-)
4 4
3 3
2 2
1 1
0 0
-1 -1
0 180 360 540 720 0 180 360 540 720
(c) (d)
Figure 5.39 – Time-resolved velocity observed on (a,b,c) CME and (d) isothermal
flow rig, for N = 1200 rpm and different flow rates
breakage. According to the manufacturer19 , wire breakage may occur

in strong pulsating flows of near sonic peak velocity. Furthermore, as
explained in Sect. 2.3, the temperature of the exhaust gas stream drops
below 0 ◦ C at high intake pressure pi > 2.5 atm. This is due to the heat
loss from the gas to the combustion chamber walls during the compres-
sion and expansion strokes, and due to the blow-by leakage through the
piston-cylinder clearance. Subzero temperatures in the exhaust system
(in combination with the absence of a dehumidifier in the compressed air
supply) yields tiny ice particles in the exhaust flow. These particles may
impact the wire and further promote breakage. Downstream of the cata-
lyst, this problem is not encountered, likely due to the warming of the air
as it passes through the exhaust runners and catalyst. The metal man-
ifold parts ensure a good heat conduction away from the engine block,
which is heated internally due to viscous friction of the moving parts, to
around 100 ◦ C (depending on the engine speed).
The mean velocity fluctuations during the displacement phases in Figs. 5.39a,
b and c have also been observed with manifold B mounted on the isothermal
flow rig, although to a much lesser extent and only at higher engine speed
(see Fig. 5.9b). For manifold A mounted on the isothermal flow rig, similar
fluctuations have been observed, yet only when using the cylinder head as
pulsator. For the rotating valve, such fluctuations were not observed. The
peak fluctuation frequencies for manifold A are between 200 and 280 Hz (see
Fig. 5.2b).
For manifold B, the fluctuation frequencies observed on CME and isother-
mal flow rig are nearly identical, varying between 140 and 200 Hz. The fre-
quency is independent of engine speed and flow rate, as shown in the summary
in Table 5.2. Table 5.2 shows the peak frequency in the catalyst velocity during
the displacement phase of individual exhaust strokes, for runner r = 1 . . . 4 (see
Table 2.1 and Fig. 2.1b). The peak frequencies are determined from the energy
spectrum, as shown in Fig. 5.40.
Figure 5.40 shows frequency spectra of the time-resolved mean catalyst ve-
locity Um during each cylinder’s exhaust stroke for N = 600 rpm. As Table 5.2
indicates, the peak frequency remains unchanged at higher engine speeds. How-
ever, the spectral resolution decreases as the engine speed increases (due to the
decreasing period), which leads to increasing uncertainty on the peak frequen-
cies.
Table 5.2 demonstrates that the length of the open runner can be detected
from the fluctuation frequency of the catalyst velocity during individual exhaust
strokes. Longer runners (e.g. 1 and 3) result in lower resonance frequencies
than shorter runners (e.g. 2 and 4). The same is observed in measurements
and simulations by other authors (see Sect. 5.3.5).
Table 5.3 demonstrates the prediction of the resonating system, based on
the lumped parameter model in Fig. 5.38. Two possible resonating systems are
considered, that result in resonance frequencies close to the observed values in
Table 5.2. Note that the runner length, cylinder volume and diffuser volume
used to predict the values in Table 5.3 are the effective values, defined by
Eqs. (5.6), (5.7) and (5.8), respectively, with α = 1 and β = 0.43.
-3
10
-4
10
-5
10
fpeak = 160.6 Hz (r = 3)
fpeak = 183.1 Hz (r = 4)
fpeak = 181.9 Hz (r = 2)
fpeak = 157.5 Hz (r = 1)
-6
10
10 20 50 100 200 500
Frequency f (Hz)
Figure 5.40 – Frequency spectra of the catalyst velocity during individual exhaust
strokes for manifold B on the CME flow rig, for N = 600 rpm
Table 5.2 – Catalyst mean velocity peak fluctuation frequencies during individual
exhaust strokes, for manifold B on the CME flow rig
N pi Qref Peak frequency
r=1 r=2 r=3 r=4
rpm atm m3 /h Hz Hz Hz Hz
600 1.00 26.5 158 182 161 183
1200 1.00 46.6 164 181 166 178
1200 1.55 67.7 141 153 126 161
1800 1.00 75.1 148 158 143 144
1200 2.23 93.3 125 173 141 158
1800 1.58 97.4 152 165 141 169
1800 2.23 136.6 143 146 139 144
2400 1.55 192.1 135 193 125 195
2400 2.03 238.8 140 193 153 178
fH,1 , see Table 5.3 166 188 166 188
fH,2 , see Table 5.3 182 205 182 205
The gas in the open exhaust runner oscillates as incompressible plug. Either
the cylinder volume Vcyl act as a single compressible spring (fH,1 ), or the
diffuser volume Vd and cylinder volume Vcyl act as two compressible springs
in series (fH,2 ). The resulting spring constant for two springs in series is the
sum of the individual spring constants kcyl + kd . In terms of Eq. (5.4), the
−1 −1
equivalent volume corresponds to V −1 = Vcyl,eff + Vd,eff , where Vcyl,eff and
Vd,eff are given by Eqs. (5.7) and (5.8).
Up-close examination of Figs. 5.39a and b reveals that the velocity in
runner 1 Ur=1 (ωt) ( ) leads the mean catalyst velocity Um (ωt) ( ) by
∆ϕ = π/2 radians. Although the time-resolved cylinder pressure pcyl (ωt)
is not shown in Fig. 5.39, it also leads the catalyst velocity by π/2 radians.
This phase lead of π/2 rad is quite surprising. Considering the lumped
parameter model in Fig. 5.38, two possible cases may arise in terms of the
phase difference ∆ϕ between runner and catalyst velocity. (i) Firstly, the phase
difference ∆ϕ may be zero, if the diffuser volume acts as an infinitely stiff spring
kd kcyl . (ii) Secondly, the phase difference ∆ϕ may equal π rad. In that
case, the gas mass in the open runner and the catalyst substrate oscillate in
antiphase, and the diffuser volume acts as a spring of finite stiffness kd ∼ kcyl .
However, the observed phase difference of ∆ϕ = π/2 suggests that the
lumped parameter model in Fig. 5.38 is incorrect. The phase lead becomes π/2
if the inertia of the gas contained in the catalyst substrate mcat is negligible.
Taking into account the laminar flow friction in the substrate (see App. A.2),
the catalyst velocity U is then in phase with (and roughly proportional to) the
pressure difference across the catalyst, which reduces to pd for a free discharge.
As such, the runner velocity Ur may lead the catalyst velocity by ∆ϕ = π/2 rad.
This explanation for the resonance frequency seems acceptable for the CME
flow rig conditions. However, when manifold B is mounted on the isothermal
flow rig, nearly the same fluctuation frequencies are observed (between 140
and 200 Hz). Similarly, frequencies between 200 and 280 Hz are observed when
manifold A is mounted on the isothermal flow rig with the cylinder head as
pulsator. As noted earlier, the fluctuations are (i) weaker compared to the
ones observed on the CME flow rig, and (ii) occur only at higher engine speed
(e.g. see Figs. 5.9b, 5.2b).
For the isothermal flow rig, Fig. 2.2 demonstrates that the surge vessel
mimics an infinite cylinder volume at quasi constant pressure. Given the 450 l
surge vessel volume, the spring stiffness associated with the surge vessel is
negligible, according to Eq. (5.4). For a free discharge set up, the only eligible
volume to act as a compressible spring is the diffuser volume Vd .
According to Beranek [20] and Davis et al. [33], there is a considerable
uncertainty on the parameter β used in Eq. (5.6) for correcting the pipe length.
The same holds true for the parameter α in Eq. (5.8). In fact, these parameters
are only known a priori for very basic applications (e.g. a circular orifice in
an infinite plane [33]). In engineering applications, the parameters should be
determined through empirical testing or model fitting.
It is therefore not surprising that by ‘fitting’ these parameters, Eq. (5.5)
may yield the observed fluctuation frequencies not only for the CME flow rig,
FRF FRF
From: Cylinder pressure p , To: Catalyst velocity ρcU From: Cylinder pressure p , To: Catalyst velocity ρcU
cyl cyl
0 0
2nd order TVD 2nd order TVD
Magnitude (dB) of ρcU/pcyl
Magnitude (dB) of ρcU/pcyl

δx = 5 mm δx = 5 mm
-10 -10
-20 -20
-30 -30
-40 -40
L = 150 mm (234 Hz)
-50 L = 90 mm (274 Hz) -50 L = 160 mm (131 Hz)
L = 120 mm (249 Hz) L = 80 mm (181 Hz)
-60 2 3
-60 2 3
10 10 10 10
180 180
90 90
Phase (°)
Phase (°)
0 0
-90 -90
-180 -180
2 3 2 3
10 10 10 10
(a) (b)
Figure 5.41 – Frequency response function between ‘cylinder’ pressure and catalyst
velocity, for (a) manifold A and (b) manifold B on the isothermal flow rig
but also for the isothermal flow rig.

In conclusion, a significant uncertainty remains in identifying the resonating
system by means of the experiments alone. Therefore, the following section
describes the use of the one-dimensional gas dynamic model to estimate the
frequency response function of the exhaust system.
5.3.4 Numerical results

Isothermal flow rig
Figure 5.41 shows the numerically determined frequency response functions for
both manifolds, mounted on the isothermal flow rig. The response function is
defined as:
ρcU
F RF (f ) = (5.9)
pcyl
where the cylinder pressure pcyl is relative to the atmospheric pressure [Pa] and
the characteristic impedance ρc (see App. D.2) is included to non-
dimensionalize F RF (f ).
The numerical model shown in Fig. 5.25 is simplified by replacing the cylin-
ders with constant pressure boundary conditions. Thus, the ‘cylinder’ pressure
is simply the surge vessel pressure. The dynamics of the surge vessel are entirely
neglected. For manifold A, three runners are simulated, without junctions prior
to issuing into the diffuser volume (see Fig. 2.1a).
The usual approach to determine the frequency response function is de-
scribed in App. D.4, and uses a multisine signal as input disturbance. This
approach is used throughout the thesis, except for determining the response
functions in Fig. 5.41. Instead, these responses are determined by linearizing
the model about a steady-state operating point, and subsequently determining
the FRF for the linear model. This is easily accomplished using the built-in
Simulink11 functionality. However, this approach is not used further in this the-
sis, because of its poor resolution in the higher frequency range (f > 500 Hz).
Nevertheless, the numerically determined FRFs in Fig. 5.41 still provides
useful information in the lower frequency range (f < 500 Hz).
The steady state range (f → 0) of the response function behaves as ex-
pected, with zero phase difference between the surge vessel pressure pcyl and
the resulting flow rate U A. The response level (between −25 and −30 dB) is
mainly influenced by the catalyst wall friction and the cross-sectional area. At
increasing frequency, the response magnitude increases and the phase drops. At
the resonance frequency, the magnitude is limited by the damping components
(e.g. exhaust valves, catalyst friction).
The first resonance peaks correspond to the open runner mass oscillating
on the diffuser volume. This resonating system is here denoted Helmholtz
resonances. The frequencies agree with the experimental observations on the
isothermal flow rig. The Helmholtz resonance frequencies range from 234 to
274 Hz for manifold A (from longest to shortest runner), and from 131 to
181 Hz for manifold B (long runners 1 and 3 versus short runners 2 and 4).
The peaks between 500 < f < 1000 Hz are due to the gas mass inside the
catalyst oscillating on the diffuser volume. The higher frequency peaks and
phase changes (f > 1000 Hz) are due to standing waves in the runners.
Isochoric flow rig: Without cold end

For the isochoric flow rig, the numerical model as presented in Fig. 5.25 is
used to numerically determine several frequency response functions presented
in this section. By contrast to Sect. 5.3.4, these functions are obtained using
the approach described in App. D.4, rather than by linearizing the model. The
response function is determined by applying a multisine disturbance to the
cylinder volume and monitoring the response of the model. The mean cylinder
volume corresponds to the mid position between bottom and top dead center,
as discussed in Sect. 5.3.2 for Eq. (5.7).
Figure 5.42a shows the frequency response function between the cylin-
der volumetric velocity Qcyl and the catalyst velocity U . The cylinder vol-
umetric velocity is determined by the piston rate of displacement, or Qcyl =
−dVcyl /dt [m3 /s]. Numerator and denominator both have the dimensions of
a volumetric flow rate by multiplying U with the catalyst cross-sectional area
A. As such, the FRF equals the ratio of two volumetric flow rates at differ-
ent locations in the model, i.e. at the cylinder of the open runner and at the
FRF FRF
From: Cylinder volume velocity Q , To: Catalyst velocity U A From: Runner velocity U A , To: Catalyst velocity U A
cyl r r
Amplitude (dB) of U A/Ur Ar

Amplitude (dB) of U A/Qcyl
f0,r=1 = 146.5 Hz f0,r=2 = 170.9 Hz 40 f1,r=1 = 236.0 Hz f1,r=2 = 236.0 Hz

40
20 20
0 0
−20 −20
Runner 1 Runner 1
−40 Runner 2 Runner 2
−40
2 3 2 3
10 10 10 10
180 180
90 90
Phase (°)
Phase (°)
0 0
−90 −90
−180 −180
2 3 2 3
10 10 10 10
(a) (b)
FRF
From: Catalyst pressure p, To: Catalyst velocity ρcU
40 f0,r=1 = 148.5 Hz f0,r=2 = 172.9 Hz
Amplitude (dB) of ρcU/p
20
−20
Runner 1
Runner 2
−40
2 3
10 10
180
90
Phase (°)
−90
−180
2 3
10 10
Frequency f (Hz)
(c)
Figure 5.42 – Frequency response function between (a) cylinder volume velocity,
(b) runner velocity, (c) downstream catalyst pressure and catalyst velocity, for man-
ifold B on the CME flow rig (free discharge)
catalyst.
The magnitude and phase of the FRF in Fig. 5.42a behaves as expected.
For low frequencies (f → 0), the numerator and denominator have the same
magnitude and zero phase difference. For increasing frequency, the catalyst
velocity magnitude increases and its phase lags with respect to the piston dis-
placement velocity. At the resonance frequency f0 , the phase lag between Qcyl
and U is −π/2 rad.
Figure 5.42b shows the frequency response function between the open run-
ner velocity (near the exhaust port) Ur and the catalyst velocity U . At the
resonance frequency f0 obtained from Fig. 5.42a, the phase difference between
runner and catalyst velocity is still negligible. Therefore, this numerical simu-
lation does not confirm the experimental observation that the runner velocity
leads the catalyst velocity by π/2 rad. The reason for this discrepancy is not
immediately clear.
Corresponding to the two runner lengths, two resonance frequencies are ob-
tained in Fig. 5.42a, f0 = 146.5 and 170.9 Hz (for long runner 1 and short
runner 2). These are in good agreement with the observed fluctuation fre-
quencies in Table 5.2. The resonance peaks for the remaining runners 3 and 4
coincide with these for 1 and 2. The magnitude at resonance is slightly different
for runners 3 and 4. This is due to the different pressure drop coefficient in
joined runner 3–4 compared to joined runner 1–2 (see Fig. 5.25). As shown in
Fig. 2.1b, the joined runner 3–4 features a stronger bend angle.
The next resonance peak in Fig. 5.42a is due to the oscillation of gas inside
the catalyst on the diffuser volume. This is demonstrated in Fig. 5.42b. The
resonance frequency f1 = 236 Hz, which is independent of the open runner
length. At this resonance, the catalyst velocity lags the runner velocity by
π/2 . This phase difference corresponds to the observations in Sect. 5.3.3, yet
at a higher resonance frequency.
As an alternative, Fig. 5.42c shows the transfer function between the pres-
sure downstream of the catalyst and the catalyst velocity. This corresponds
to the admittance of the manifold and engine, at the measurement location
immediately downstream of the catalyst. The admittance Y /Y0 is defined in
App. D.2.2.
The response function in Fig. 5.42c is determined by applying a multisine
disturbance to the pressure downstream of the catalyst and monitoring the
response of the model. Again, the open cylinder volume is set to the mid
position. Figure 5.42c shows approximately the same resonance frequencies as
Fig. 5.42a, corresponding to the runner mass oscillating on the cylinder volume.
The low frequency (f → 0) behavior is characteristic of the combined volume
of cylinder, runners and diffuser. The magnitude slopes at +20 dB/decade and
the velocity lags the pressure by π/2 rad.
Figure 5.42c is shown here, partly because of the comparison to the doc-
toral work of Boonen [21]. Figure 5.43 shows the acoustic impedance of a small
combustion engine, where Fig. 5.43c is obtained for a stationary engine and
Figs. 5.43a and b are obtained for motored conditions with blocked intake ports
−1
(i.e. without net flow). The figures show the load impedance Z/Z0 = (Y /Y0 )
(a) (b)
(c) (d)
Figure 5.43 – Measured (a, b and c) and simulated (d) acoustic impedance of
an engine with exhaust manifold, motored at (a) 1000 rpm, (b) 2000 rpm and (c,
d) stationary (Source: [21])
at the four-in-one junction of the exhaust runners. No catalyst is present. The

impedance functions are obtained through two simultaneous pressure measure-
ments at different locations in a long exhaust pipe, as developed by Boonen [21].
Upon inverting the impedance shown in Fig. 5.43, there is an obvious qual-
itative correspondence between the response functions. The first zero in the
impedance (between 100 and 150 Hz) corresponds to the pole or resonance
peak in the admittance function shown in Fig. 5.42c, and is due to the runner
mass oscillating on the cylinder volume.
Interestingly, there is some discrepancy between the measured impedance
for the stationary engine (Fig. 5.43c) and the motored engine (Fig. 5.43a and b).
Neglecting the low frequency errors in Fig. 5.43c, the magnitude and phase tend
to vary in the vicinity of the first resonance. In particular, a substantial differ-
ence is noted in the frequency range from 150 to 300 Hz. The running engine
seems to exhibit a larger damping in this region compared to the stationary
engine.
The same engine setup is simulated using an electrical equivalent scheme of
the gas dynamics in the exhaust manifold. Figure 5.43d shows the numerically
determined impedance. The model captures the first resonance around 150 Hz
well, yet some deviation is observed for higher frequencies.
Although the setup used by Boonen [21] does not include a close-coupled
FRF FRF
From: Junction pressure p, To: Junction velocity ρcU From: Runner pressure p, To: Runner velocity ρcU
40 f0,r=1 = 170.9 Hz f0,r=2 = 203.5 Hz 40

20 20
0 0
−20 −20
Runner 1 Runner 1
Runner 2 Runner 2
−40 −40
2 3 2 3
10 10 10 10
180
90 0
Phase (°)
Phase (°)
0
−90 −90
−180
2 3 2 3
10 10 10 10
(a) (b)
Figure 5.44 – Frequency response function (a) between junction pressure and junc-
tion velocity, (b) between runner pressure and runner velocity, for manifold B on the
CME flow rig (free discharge)
catalyst, the presented results indicate some uncertainty in the frequency range
around the Helmholtz resonance, comparable to the discrepancy noted in this
thesis between the experimental and numerical results.
Figure 5.44 shows two additional frequency response functions that may
be interpreted as the admittance at two locations: (a) at the inlet to the
joined runner 1–2 and (b) at the inlet to the open exhaust runner. Most of
the manifold dynamics are still visible in the admittance near the runner 1–2
junction (Fig. 5.44a), although the Helmholtz resonance peaks are reduced in
magnitude and slightly shifted to higher frequencies. Figure 5.44b shows the
admittance near the open exhaust ports. As such, only the cylinder volume and
the exhaust valve damping can be discerned in this graph. The +20 dB/decade
slope and −π/2 phase are characteristic for the cylinder volume compressibility.
At very high frequency (f > 1000 Hz), the exhaust valve damping causes the
magnitude to level off and the phase to increase towards zero.
Isochoric flow rig: With cold end

All transfer functions shown in Sect. 5.3.4 are for a free discharge, i.e. with-
out cold end attached to the catalyst, which is the same situation as for the
experiments. Figure 5.45a shows the effect of including the cold end on the
frequency response function between the cylinder volume velocity and catalyst
velocity. The simulated cold end is identical to the one used in Sect. 5.2.3, and
Table 5.3 – Prediction of the resonating system and corresponding frequency fH ,

for manifold B on the CME flow rig
Resonance frequency k m Resonating system,
according to Eq. (5.5) see Fig. 5.38a
fH,1 kcyl mrun Open runner on cylinder
volume
fH,2 kcyl + kd mrun Open runner on cylinder
and diffuser volume
Note: kcyl , kd and mrun are determined using the effective volumes in
Eqs. (5.7) and (5.8) with α = 1 and the effective length in Eq. (5.6) with
β ' 0.43
FRF FRF
From: Cylinder volume velocity Q , To: Catalyst velocity U A From: Cylinder volume velocity Q , To: Catalyst velocity U A
cyl cyl
f0,r=1 = 154.6 Hz f0,r=2 = 179.0 Hz f0,r=1 = 146.5 Hz f0,r=2 = 170.9 Hz

40 40
20 20
0 0
−20 −20
Runner 1 Runner 1
−40 Runner 2 −40 Runner 2
2 3 2 3
10 10 10 10
180 180
90 90
Phase (°)
Phase (°)
0 0
−90 −90
−180 −180
2 3 2 3
10 10 10 10
(a) (b)
Figure 5.45 – Frequency response function between cylinder volume velocity and
catalyst velocity, for manifold B on the CME flow rig, (a) with and (b) without cold
end
Figure 5.46 – Simulated runner velocity for fired engine (Source: [4])
is described in Sect. 5.2.3.

The response function in Fig. 5.45a exhibits little difference compared to
the case without cold end. The Helmholtz resonance peaks are slightly shifted
to higher frequencies, and some additional smaller peaks are introduced by the
coupling between hot end and cold end. Overall, the influence of the cold end
seems limited. This is also observed in the numerical simulation of the time-
resolved catalyst velocity in Fig. 5.33, where the dashed and solid lines are the
simulated catalyst velocity with and without cold end, respectively.
5.3.5 Discussion
As already indicated in the literature survey in Sect. 1.4.2, this type of cat-
alyst velocity fluctuations has been found by other researchers as well, both
experimentally and numerically.
Adam et al. [4] present numerical results for a one-dimensional gas dynamic
model of a close-coupled catalyst exhaust manifold, mounted on a fired engine.
Figure 5.46 shows the velocity in each exhaust runner for 3000 rpm at part
load. The velocity fluctuations during the displacement phases are very similar
to the time-resolved catalyst velocity observed on the CME flow rig. However,
fluctuations in their catalyst velocity are much less pronounced compared to
the CME flow rig.
The fluctuation frequencies during each cylinder’s exhaust stroke differ,
depending on the runner length. Based on visual inspection of the data in
Fig. 5.46, the estimated fluctuation frequency is 450 Hz for the long runners 1
and 4 and 580 Hz for the short √runners 2 and 3. From Eq. (5.5) follows that the
resonance frequency fH ∝ 1/ L. As such, based on these frequency estimates,
the ratio of the length of long to short runners is 1.6, which seems plausible
from their paper [4].
Park et al. [81] present experimental results using LDA for a close-coupled
catalyst exhaust manifold, mounted on a fired engine. Figure 5.47 shows the
velocity in runner 3 for 2000 rpm at part load. Substantial backflow occurs
following blowdown, as is observed on the CME flow rig. The estimated fluctu-
ation frequency is 300 Hz. This frequency is too low to be caused by pressure
Figure 5.47 – Measured runner ve- Figure 5.48 – Simulated runner ve-
locity for fired engine (Source: [81]) locity for fired engine (Source: [70])
waves as explained by the authors [81], yet the value corresponds well with a
Helmholtz resonance of the manifold.
Liu et al. [70] present numerical results for a close-coupled catalyst mani-
fold in fired engine conditions, obtained using a combined one-dimensional and
three-dimensional numerical approach similar to Adam et al. [4]. Figure 5.48
shows the runner velocity at 3000 rpm and full load. The estimated frequency
of the fluctuations during the displacement phase is 310 Hz. Simulation results
by [70] indicate no fluctuations in motored engine conditions. This is rather
unexpected, assuming the Helmholtz resonance explanation stated above is cor-
rect. Perhaps the motored and fired cases do not exhibit the same excitation
required to invoke the resonance effect, although similar flow conditions on the
CME flow rig are found to exhibit the same velocity fluctuations, regardless
of the engine load. Figure 1.18 shows numerical and experimental results us-
ing LDA, downstream of the catalyst. Although no actual positive blowdown
occurs in motored conditions at atmospheric intake pressure, flow reversal is
nonetheless detected in experiments and simulations. For fired conditions, only
the simulations show flow reversal.
For fired engine conditions, the temperature is much higher compared to
the CME
√ flow rig. From Eq. (5.5) follows that the resonance frequency fH ∝
c ∝ T . Since the temperature ratio between cold flow conditions and fired
engine conditions is approximately Tf ired /Tcold ' 1073/273 ' 4, the resonance
frequency (for the same geometry) will be twice as large for a fired engine
compared to the CME flow rig. This seems to correspond in the literature [4, 81]
to the values observed for fired engines, ranging between 300 and 600 Hz.
Benjamin et al. [14] present LDA measurements downstream of a close-
coupled catalyst, in a fired engine at 2000 rpm and high load. Similar to Liu
et al. [70], the experimental results are compared to a transient CFD simulation,
coupled with a commercial one-dimensional gas dynamics code. The simulation
predicts a catalyst velocity between −5 and 22 m/s. The measured velocity
results at the same location exhibit significantly lower amplitude and reduced
backflow magnitude.
Figure 5.31a shows a comparison between measured ( ) and calculated
Table 5.4 – Overview of the resonance frequencies observed in the literature

Author Flow rig Resonance frequency (Hz) Figure
Numerical Experimental
Park et al. [81]Fired engine 300∗ 5.47
∗∗ ∗
Adam et al. [4] Fired engine 450 – 580 5.46
Adam et al. [4] Fired engine 450∗∗ – 580∗ 5.46
Liu et al. [70] Fired engine 310∗∗ 5.48
Motored " no resonance
Benjamin et al. Fired engine 376∗∗ 197∗∗ 5.31a
[14]
This thesis CME flow rig 130∗∗ – 180∗ 160∗∗ – 180∗
Fired engine 330∗∗ – 360∗
∗, ∗∗
These frequencies correspond to a long∗∗ or short∗ runner.
( ) runner velocity. The simulation exhibits strong backflow following the

blowdown, and significant velocity fluctuations during the displacement phase.
The fluctuation frequency is about 376 Hz for the numerical model, which is in
good agreement with the frequencies observed in other fired engine setups [4, 81,
70]. However, the measured runner velocity exhibits a fluctuation frequency of
approximately 197 Hz, which is 1.9 times smaller than the calculated frequency
for the same conditions. This cannot be readily explained based on the details
given in the manuscript [14]. Figure 5.31b shows a qualitative correspondence
between numerical and experimental results for the velocity at a single point in
the catalyst. The simulation results predict much stronger backflow than the
experimental observations.
Table 5.4 summarizes the resonance frequency values observed in the lit-
erature, through numerical and experimental methods. Since most manifolds
feature runners of different lengths, each mentioned frequency is marked by
asterisks, denoting the corresponding runner length compared to the other
runners. Typically for a four-cylinder engine, the outer runners (1, 4) are
longer than the middle runners (2, 3). The resonance frequency values found
in this thesis are quite consistent: the numerical simulations agree with the
experimental measurements for the CME flow rig.
For fired engine conditions, the simulations using the one-dimensional gas
dynamics code yield a resonance frequency that is higher compared to the CME
flow rig, which is readily explained by the ratio of absolute temperatures. The
values found for manifold B in this thesis are consistent with values found
by other authors [81, 4, 70, 14] in fired engine conditions, for similar exhaust
manifolds, taking into account the given variability in manifold geometry and
conditions.
5.4 Conclusion 223
5.4 Conclusion
The characteristic flow dynamics have been investigated in a close-coupled cat-
alyst exhaust manifold. Using phase-locked hot-wire anemometry, Sect. 5.1
comments on the time-resolved catalyst velocity distribution measured in cold
flow conditions, using an (i) isothermal flow rig and an (ii) isochoric or charged
motored engine (CME) flow rig. The exhaust stroke flow similarity with respect
to fired engine conditions has been discussed previously in Sect. 2.3.
The oscillating hot-wire anemometer discussed in Chap. 3 has been suc-
cessfully applied to measure the instantaneous local bidirectional velocity in
the entire catalyst cross-section, yielding velocity data with high spatial and
temporal resolution. The OHW has been validated in Sect. 5.2.1 against a ref-
erence flow rate measurement. It improves the effect of rectification or folding
errors associated with using standard hot-wire anemometry. The high resolu-
tion in space and time, as well as the ability to measure bidirectional velocity
constitute unique features of this experimental approach, making this work an
original contribution to the present state-of-the-art concerning flow in exhaust
systems.
The combination of the OHW approach and the phase-locked measurement
technique has revealed significant periodic flow reversal in the catalyst, under
varying engine operating conditions. This has been previously described in the
literature for close-coupled catalyst manifolds under fired engine conditions.
The spatial and temporal occurrence of backflow is studied in Sect. 5.2.2 on the
isochoric flow rig in cold flow conditions. Strong flow reversal occurs following
each blowdown phase.
A numerical one-dimensional gas dynamic model of the entire exhaust sys-
tem has been implemented. The model uses a second order total variation
diminishing scheme developed by Vandevoorde [101], as described in App. D.
The model is extensively validated in App. D.2, using several benchmark prob-
lems relevant to exhaust systems.
Section 5.2.3 compares the experimental velocity data to the simulation
results of the numerical gas dynamic model. Given the good agreement, the
numerical model is used to predict the influence of the cold end (i.e. exit cone,
exhaust pipe and a reference single expansion chamber muffler) on the flow
dynamics. For cold flow conditions corresponding to the isochoric flow rig,
the presence of the cold end does not appreciably influence the time-resolved
catalyst velocity.
Furthermore, the numerical model has been used to predict the correspond-
ing flow conditions in fired engine operation, with and without cold end. The
exhaust stroke flow similarity is not perfect, as indicated in Sect. 2.3. However,
flow reversal is also observed in fired engine conditions, even more so than in
cold flow conditions.
The catalyst velocity fluctuations observed on the isochoric flow rig have
been analyzed and may be explained as zeroth order gas dynamic resonances
or Helmholtz resonances in Sect. 5.3. By determining the frequency response
function of the numerical exhaust model, the understanding of the phenomenon
is greatly facilitated.
In summary, the relevant flow dynamics in a close-coupled catalyst exhaust
manifold have been studied in cold flow conditions, similar to fired engine
operation. The study is performed using a combination of (i) a high resolution
experimental approach including the ability to quantify bidirectional velocity,
and (ii) a one-dimensional numerical gas dynamic model of the exhaust system.
Parts of this chapter have been published in international journals with
review:

[86] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Exper-
imental validation of the addition principle for pulsating flow in
close-coupled catalyst manifolds. J. Fluids Eng.-Trans. ASME,
128(4):656–670, 2006. http://dx.doi.org/10.1115/1.2201646.
[88] T. Persoons, E. Van den Bulck, and S. Fausto. Study of pul-
sating flow in close-coupled catalyst manifolds using phase-locked
hot-wire anemometry. Exp. Fluids, 36(2):217–232, 2004. http://
dx.doi.org/10.1007/s00348-003-0683-0
Chapter 6
Design considerations
This brief chapter provides some considerations for the design of close-coupled
catalyst exhaust manifolds, based on the experiences gained within this thesis.
Section 6.2 discusses the relationship between the manifold flow dynamics and
the addition principle’s validity, thereby linking the findings in Chap. 4 and 5.
6.1 Addition principle

Some manifold design criteria may be formulated based on Fig. 4.29 and
Eq. (4.47). Based on the correlation for rS (4.47), a good correspondence
between numerical simulations or measurements in stationary and pulsating
flow conditions is obtained for a high scavenging number, S > Scrit . The criti-
cal value equals Scrit = 0.723±0.052 when taking into account the experiments
on manifolds A and B combined, or Scrit = 0.722 ± 0.056 when only taking
into account the experiments on manifold B.
Recall the definition of the scavenging number as S = Tp /Ts , where Tp
is the flow pulsation time scale (4.46), and Ts is the diffuser residence time
scale (4.43). Ts is determined by the diffuser volume Vd and the time-averaged
volumetric flow rate Q through the catalyst, or Ts = Vd /Q .
Figure 6.1 summarizes the findings concerning the validity of the addition
principle. For a given engine geometry and operating range, decreasing the
diffuser volume Vd increases the scavenging number S, thereby improving the
correspondence between stationary and pulsating flow. In the validity region
of the addition principle (S > Scrit ), stationary flow CFD simulations can
be used instead of time-consuming transient simulations for the design of an
exhaust manifold with close-coupled catalyst, resulting in significantly shorter
development times.
Based on the correlation for rM (4.47), the flow uniformity is always higher
in pulsating flow compared to stationary flow. A manifold that is designed
225
226 Chapter 6 Design considerations
Validity of the addition principle

1.5
rS
high Scrit
rM
low Scrit
1
low Scrit
high Scrit
0.5 Q
low engine load S ∼ high engine load
high engine speed
N Vd low engine speed
large diffuser volume small diffuser volume
addition principle not valid addition principle valid

(and high flow uniformity)
0
0 1 2 3 4 5
Figure 6.1 – Validity of the addition principle
based on stationary CFD simulations, and that satisfies the preset criteria for
flow uniformity, will likely feature a better flow uniformity in pulsating flow
conditions.
Flow uniformity in pulsating flow is higher for a small scavenging number,
S < Scrit . Increasing the diffuser volume Vd decreases the scavenging number
S, and consequently increases the flow uniformity. This unsurprising conclusion
will be subject to compromise in terms of geometrical and thermal packaging
constraints within the engine compartment. However, Sect. 6.2 provides an
interesting addition to this conclusion, with regard to the Helmholtz resonance
in the manifold.
6.2 Flow dynamics: Helmholtz resonance

The flow dynamics discussed in Chap. 5 are indirectly related to the addition
principle’s validity derived in Chap. 4. In particular, the Helmholtz resonance
which is intrinsic to the manifold geometry, causes strong fluctuations in the
mean catalyst velocity. These fluctuations are strong enough to cause extensive
flow reversal or otherwise periods of near standstill of the gas in the catalyst.
This has been observed experimentally on the CME flow rig, using the OHW for
6.2 Flow dynamics: Helmholtz resonance 227
measuring bidirectional instantaneous local velocity. Furthermore, numerical

simulations described in Sect. 5.2.3 show that these velocity fluctuations and
the related flow reversal are even stronger in a fired engine. This fact has
been observed experimentally and numerically by some authors in fired engine
conditions [4, 81, 70].
Due to the Helmholtz resonance effect, the frequency spectrum of the mean
catalyst velocity contains more energy in the higher frequency range, e.g. when
compared to the isothermal flow rig experiments. Considering the definition of
the apparent flow pulsation period Tp in Eq. (4.46) used in the expression for
the scavenging number S = Tp /Ts , the higher frequency content of the catalyst
velocity influences Tp and consequently S.
Although S is mainly a function of the collector geometry (diffuser volume
Vd , number of runners nr ) and the engine operating conditions (engine speed N
and engine load, which determines Q), the strong catalyst velocity fluctuations
during the displacement phase may alter the effective value of S as ‘experienced’
by the catalyst. In other words, S is quasi independent of the engine speed N
in the presence of the Helmholtz resonances. As such, the distinction in Fig. 6.1
with respect to engine speed no longer applies and the validity of the addition
principle is governed predominantly by the engine load and diffuser volume.
Section 4.5 gives the summarized results on the validity of the addition
principle. Figure 4.29 and the correlations in Eq. (4.47) demonstrate the strong
relationship between S and (i) the validity of the addition principle (based on rS
and rM ), as well as (ii) the relative increase of the flow uniformity in pulsating
flow compared to stationary flow (based on rM ).
Example As a thought experiment, assume that the exhaust sys-

tem features no gas dynamics at all. In that case, the catalyst veloc-
ity during the exhaust stroke is expected to behaves as indicated by
the solid line ( ) in Fig. 2.10, which is a numerical result obtained
from a zero-dimensional filling-and-emptying engine model.
In that case, a given engine operating point corresponds to a certain
value of S. Assume, with loss of generality, that e.g. S = 1.1.
When the exhaust system gas dynamics are now considered, the
mean catalyst velocity will exhibit the fluctuations as observed dur-
ing the displacement phase, thereby decreasing the flow pulsation
period Tp apparent to the manifold and catalyst, and consequently
the value of S will decrease to e.g. S = 0.4.
Based on Fig. 4.29 and the correlations in Eq. (4.47), the decrease
in S will cause:
• Based on rM : The flow uniformity in pulsating flow conditions

will increase with respect to stationary flow conditions.
• Based on rS : The correspondence in shape between the veloc-
ity distribution in pulsating flow conditions will decrease with
respect to stationary flow conditions.
As indicated in Sect. 5.3.2, the Helmholtz resonance frequency fH corresponds

very well to the observed catalyst velocity fluctuations. fH is defined by
Eq. (5.5):
r
1 c Ar Lr
fH =
2π Lr Vd
As such, fH features the following dependencies:
p
(i) fH ∝ c ∝ Texh
1
(ii) fH ∝ √
Lr
p
(iii) fH ∝ Ar
1
(iv) fH ∝ √ (6.1)
Vd
Section 1.2.1 provided some aspects of manifold design, related to the optimal
use of the catalyst. Catalyst flow uniformity is found to be of major impor-
tance for minimizing local catalyst deactivation, minimizing backpressure and
maximizing the conversion efficiency.
The above example indicates that the observed velocity fluctuations during
the displacement phase are not at all undesirable, in terms of the catalyst
flow uniformity. In fact, based on Fig. 4.29b and the correlation for rM in
Eq. (4.47), the flow uniformity can be maximized in pulsating flow conditions
by decreasing the scavenging number S for a given engine operating range.
As already indicated in Sect. 6.1, increasing the diffuser volume Vd effec-
tively decreases the range of S, yet this opposes packaging constraints. Rather,
S = Tp /Ts can be decreased by decreasing the apparent flow pulsation period
Tp , defined as Eq. (4.46):
1
Tp =
fP SD(Um ),max
Since the frequency spectrum of the catalyst velocity Um (ωt) is governed
largely by the Helmholtz resonance frequency fH :
1 1
S ∝ Tp = ∝ (6.2)
fP SD(Um ),max fH
where the statement fP SD(Um ),max ∝ fH is only an approximation, given the
non-linear nature of the ‘max’ function.
Based on the dependencies in Eq. (6.1) of the Helmholtz resonance fre-
quency, the scavenging number S features the following dependencies:
r r
Tp Q 1 Lr Vd Lr
S= ∝ = (6.3)
Vd Vd Texh Ar Vd Ar Texh
As such, S may be decreased (i.e. flow uniformity may be increased) by:
6.3 Summary 229
(i) higher exhaust gas temperature (Texh 1)
(ii) shorter exhaust runners (Lr %),
(iii) larger diameter exhaust runners (Ar 1)
(iv) larger resonating volume (Vd 1).
Particularly conditions (i) and (ii) are in agreement with the general trends that
are observed in exhaust systems. Rapid catalyst warmup is obtained by using
runners with small length-to-diameter ratio, in combination with heat shields
and thermal insulation to decrease enthalpy losses in the hot end. As such, the
flow dynamics which are governed by the Helmholtz resonance appear to be
favorable in terms of the catalyst flow uniformity. Of course, this derivation
is made based upon the results for only two manifolds. Further research is
required to verify this statement.
Although the Helmholtz resonance phenomenon seems beneficial for the flow
uniformity, the mean velocity fluctuations also causes strong pressure transients
which are favorable for the occurrence of flow reversal. This may be noted in
Figs. 5.16 through 5.18, where the flow reversal occurs mainly following the
blowdown yet also during the displacement phase, when the piston is already
moving fast towards top dead center. The relation between flow reversal and
catalyst ageing is unknown, yet also warrants further research due to the strong
temperature dependence of the ageing process (see Sect. 1.2.1 and Sect. 5.2.4).
6.3 Summary
In summary, the following practical considerations are formulated based on
the findings of this thesis. When faced with the task of designing an exhaust
manifold for a close-coupled catalyst, two stages are discerned:
(1) During the conceptual stage, the optimal manifold geometry is selected
based on the given range of operating conditions that influence the scav-
enging number S (e.g. engine speed, engine displacement, intake system
charging) and the critical value of the scavenging number Scrit :
Firstly, since the flow uniformity in pulsating conditions increases with
respect to the stationary flow uniformity, the scavenging number S de-
creases in case of (see above):
(i) high exhaust gas temperature (Texh 1)

(ii) short exhaust runners (Lr %),
(iii) large diameter exhaust runners (Ar 1)
(iv) large diffuser volume (Vd 1).
Secondly, referring to the hypothetical collector efficiency ηD introduced

in Sect. 4.6.2, the critical scavenging number Scrit should be maximized.
With reference to Eq. (4.54), maximizing ηD = Scrit is equivalent to

maximizing the effectively used diffuser volume in the flow distribution
process (ηD = Scrit = Vd,eff /Vd ).
However, further studies are needed to confirm the collector efficiency
hypothesis, and the influence of the manifold geometry (e.g. the shape
of the exhaust runners, the entrance angle of the runners in the diffuser)
on the critical scavenging number. This might be the focus of future
research.
(2) For a given exhaust manifold geometry, the following rules apply for inter-
preting stationary CFD predictions of the catalyst velocity distribution:
(i) The addition principle is valid for high engine load and low engine
speed (S > Scrit ). The engine speed is of lesser importance given
the inevitable resonance phenomenon which reduces the influence
of N on the flow pulsation period Tp . In these conditions, the sta-
tionary velocity distribution Ustat obtained according to Eq. (4.1)
corresponds to the actual velocity distribution in pulsating condi-
tions.
(ii) The addition principle is not valid for low engine load and high en-
gine speed (S < Scrit ). In these conditions, the stationary velocity
distribution Ustat does not correspond to the pulsating velocity dis-
tribution. However, the flow uniformity in pulsating conditions is
higher than in stationary conditions.
Based on the results in this thesis, one may conclude that an exhaust
manifold for which the stationary flow distribution meets the uniformity
design criterion, will also meet the design criterion in actual pulsating
flow conditions. However, the accuracy of the velocity distribution itself
(e.g. location of maximum velocity and flow reversal, overall shape of the
distribution) is only guaranteed in the region of validity of the addition
principle, i.e. S > Scrit .
Chapter 7
Conclusion
7.1 Conclusion
Recalling the goals specified in Sect. 1.6, this thesis aimed to further the under-
standing of pulsating flow in modern compact close-coupled catalyst exhaust
manifolds for internal combustion engines. Instead of using a fired engine fea-
turing a hot corrosive exhaust gas environment, two cold pulsating flow rigs
have been used. This enables the use of velocity measurement techniques that
yield a high spatial and temporal resolution. Other authors [59, 81, 70, 53, 14]
encountered serious problems in measuring velocity distributions in the exhaust
system of a fired engine, using optical anemometry. Their findings are discussed
in App. C.2.
The thesis has focused on the most relevant flow-related aspect to the de-
sign of the exhaust manifold: the catalyst velocity distribution. As discussed
in Sect. 1.2, obtaining a uniform catalyst velocity distribution is crucial for
an optimal manifold design, in terms of minimal local catalyst degradation,
minimal pressure drop and maximal conversion efficiency.
• Section 2.2 describes two experimental flow rigs. The isothermal flow rig
(Sect. 2.2.1) is commonly used by a number of authors [88, 58, 18, 69,
17, 16, 15, 43, 22], because of its simplicity to use. However, the ex-
haust system flow similarity between the isothermal flow rig and a fired
engine is quite poor. Section 2.2.2 describes the isochoric or charged mo-
tored engine (CME) flow rig, developed within this thesis. The CME
flow rig mimics the exhaust system flow in fired engine conditions as
best as possible, while still operating at ambient temperature. The ex-
haust stroke features blowdown and displacement phase, typical of a fired
engine. Section 2.3 discusses a thermodynamic analytical derivation of
the exhaust stroke flow similarity between CME and fired engine condi-
tions [88, 87, 86, 84, 85].
231
232 Chapter 7 Conclusion
• Chapter 3 presents a novel low-frequency oscillating hot-wire anemome-

ter (OHW) that enables bidirectional velocity measurements in the ex-
haust system. The OHW is more compact than traditional flying hot-wire
anemometers [25], and less prone to prong or wire vibration and strain
gauging than recent high-frequency OHW systems [78, 66].
The OHW is calibrated in a small-scale wind tunnel, in the negative
velocity range −1.5 6 U 6 0 m/s. Laser Doppler anemometry is used as
reference velocity measurement, phase-locked with the OHW. Three hot-
wire probe designs are calibrated, examining the influence of prong length
and shape. Calibrations are performed for two oscillation amplitudes
and several frequencies. The best calibration results are obtained for a
straight probe with extended prongs (55P11L), in combination with an
oscillation amplitude xo = 5.5 mm. This choice results in a maximum
resolvable negative velocity of −1.0 m/s.
A non-dimensional scaling analysis reveals that straight (55P11, 55P11L)
and angled (55P14) probes behave differently with regard to the corre-
spondence between the OHW velocity U 0 and the reference velocity U .
For the straight probes, increasing the oscillation amplitude xo (and de-
creasing oscillation frequency fo ) reduces the deviation between U 0 and
U . For the angled probe, the deviation between U 0 and U is reduced by
decreasing xo and increasing fo .
The presented OHW system has been successfully applied within this
thesis to measure the phase-locked velocity distribution including instan-
taneous local flow reversal on the CME flow rig [83, 84, 85].
• Chapter 4 investigates the validity of the addition principle (4.1) for pul-
sating flow in close-coupled catalyst manifolds. The addition principle
states that the time-averaged catalyst velocity distribution in pulsating
flow Upuls equals a linear combination of velocity distributions obtained
for steady flow through each of the exhaust runners Ustat , according to
Eq. (4.1).
The scavenging number S defined in Eq. (4.42) forms the appropriate
non-dimensional number to characterize the pulsating flow. The non-
dimensional measures rS (see Sect. 4.3.2) and rM (see Sect. 4.3.2) quantify
the similarity between the Ustat and Upuls distributions based on shape
and magnitude, respectively. These measures are used to quantify the
validity of the addition principle.
The results from the entire measurement campaign are combined in
Fig. 4.29 and Tables 4.3 and 4.4. Figure 4.29 shows the correlation be-
tween the similarity measures rS and rM and the scavenging number S,
resulting in the correlation fits (4.47):
 00
 rS = 1 − exp (−S/0.723 ) ; R2 = 0.91
00
rM = 1.118 + 0.337 exp (−S/0.723 ) ; R2 = 0.30

7.1 Conclusion 233
where the correlation is excellent for rS and not quite so convincing for
rM (for reasons explained in Sect. 4.5).
Strong statistical evidence is given in support of the addition principle
in Tables 4.3 and 4.4, for nearly the entire range of S. Since no clear
validity limit can be derived from the statistical significance of rS and rM ,
the practical limit of the addition principle’s validity is when S exceeds
the critical scavenging number Scrit = 0.723 (± 0.052), corresponding
roughly to rS > 1 − e−1 = 0.63 and rM < 1.24.
The validity of the addition principle in terms of S carries two important
consequences for the industrial design of these systems:
(i) Based on the correlation for rM (4.47), the flow uniformity is always
higher in pulsating flow compared to stationary flow. A manifold
that is designed based on stationary CFD simulations, and that
satisfies the preset criteria for flow uniformity, will likely feature a
higher flow uniformity in pulsating flow conditions.
(ii) In the validity region of the addition principle (S > Scrit ), steady-
state CFD simulations can be used instead of time-consuming tran-
sient simulations for the design of an exhaust manifold with close-
coupled catalyst, resulting in a significantly shorter development
time.
Other authors [17, 22, 99] have used non-dimensional numbers similar to
S to characterize the pulsating flow in close-coupled catalyst manifolds.
However, the original contribution of this work is to relate S to the flow
distribution similarity between pulsating and stationary flow conditions
using rS and rM , and furthermore to derive the validity of the addition
principle from that relationship [88, 86].
• Based on the elegance of the rS correlation in Eq. (4.47), this complex
multi-dimensional flow behaves essentially like a zero-dimensional scalar
mixing process. In that respect, the critical scavenging number Scrit
may be considered the ratio of the effective to actual diffuser volume.
As such, the hypothesis may be formulated that Scrit corresponds to a
collector (i.e. runners and diffuser) efficiency ηD with respect to cata-
lyst flow uniformity. By maximizing the collector efficiency ηD , the flow
uniformity is optimized, and consequently so is the catalyst durability,
conversion efficiency and exhaust system backpressure. The correlations
in Eq. (4.47) are valid for two different exhaust manifolds. In Sect. 4.6.2,
the correlations in Eq. (4.55) are obtained only for the experiments on
manifold B. Based on these correlations, the critical scavenging number
or hypothesized collector efficiency ηD yields 0.722 (± 0.056).
• The characteristic flow dynamics are discussed in Chap. 5. Section 5.1
demonstrates the potential of the experimental approach in determining
the time-resolved catalyst velocity distribution in cold flow conditions,
using an (i) isothermal flow rig and an (ii) isochoric or charged motored
engine (CME) flow rig. The oscillating hot-wire anemometer introduced

in Chap. 3 has been successfully applied to measure the instantaneous
local bidirectional velocity in the entire catalyst cross-section, yielding
velocity data with high spatial and temporal resolution. The OHW has
been validated in Sect. 5.2.1 against a reference flow rate measurement.
The OHW reduces the effect of rectification or folding errors associated
with using standard hot-wire anemometry in strong pulsating flows. The
high resolution in space and time, as well as the ability to measure bidirec-
tional velocity constitute unique features of this experimental approach,
making this work an original contribution to the present state-of-the-art
concerning flow in exhaust systems.
Significant periodic flow reversal is observed in the catalyst, under vary-
ing engine operating conditions. This has been previously described in
the literature for close-coupled catalyst manifolds under fired engine con-
ditions. The spatial and temporal occurrence of backflow is studied in
Sect. 5.2.2 on the isochoric flow rig in cold flow conditions. Strong flow
reversal occurs following each blowdown phase [84, 85].
• A numerical one-dimensional gas dynamic model of the entire exhaust sys-

tem has been implemented. The model uses a second order total variation
diminishing scheme [101], as described in App. D. The model is exten-
sively validated in App. D.2, using several benchmark problems relevant
to exhaust systems. Section 5.2.3 compares the experimental velocity
data to the simulation results of the gas dynamic model. Given the good
agreement, the numerical model is used to predict the influence of the cold
end (i.e. exit cone, exhaust pipe and a reference single expansion chamber
muffler) on the flow dynamics. For cold flow conditions corresponding to
the isochoric flow rig, the presence of the cold end does not appreciably
influence the time-resolved catalyst velocity. Furthermore, the numerical
model has been used to predict the corresponding flow conditions in fired
engine operation, with and without cold end. The exhaust stroke flow
similarity is imperfect. However, flow reversal is also observed in fired
engine conditions, even more so than in cold flow conditions [84, 85].
• Strong catalyst velocity fluctuations during the displacement phase are

observed on the isochoric flow rig. These fluctuations have been analyzed
and explained as zeroth order gas dynamic resonances or Helmholtz res-
onances in Sect. 5.3. By determining the frequency response function of
the numerical exhaust model, the understanding of the phenomenon is
greatly improved [84, 85].
• The flow dynamics discussed in Chap. 5 are indirectly related to the addi-
tion principle’s validity derived in Chap. 4. This relationship is illustrated
in Chap. 6.
Equation (6.3) presents the dependencies of the scavenging number S
taking into account the strong Helmholtz resonances that occur in the
7.2 Suggestions for future research 235
manifold. These resonance velocity fluctuations reduce the apparent pul-

sation period and therefore reduce the scavenging number. In terms of
the validity of the addition principle (Fig. 4.29), this causes a higher flow
uniformity in pulsating flow conditions.
Therefore, the observed catalyst velocity fluctuations seem beneficial for
increasing the flow uniformity. Furthermore, the Helmholtz resonance
frequency range increases for shorter exhaust runners and higher exhaust
gas temperature, two factors that agree with the tendency to place the
catalyst close to the engine and reduce the enthalpy loss in the hot end.
Although the Helmholtz resonance phenomenon seems beneficial for the
flow uniformity, it also promotes catalyst flow reversal due to strong pres-
sure transients. The relation between flow reversal and catalyst ageing is
unknown, yet also warrants further research due to the strong tempera-
ture dependence of the ageing process (see Sect. 1.2.1 and Sect. 5.2.4).
In summary, this thesis has improved the understanding of the flow dynamics
in close-coupled catalyst exhaust manifolds in general, and the catalyst velocity
distribution in particular. The experimental approach has yielded high resolu-
tion bidirectional velocity data that would be otherwise very difficult to obtain
in fired engine conditions. The validity of the addition principle in terms of the
scavenging number S carries important consequences for the optimal design of
these systems. And finally, the governing flow dynamics have been analyzed
and explained, with the aid of a one-dimensional gas dynamical model of the
exhaust system.
7.2 Suggestions for future research

• The main subject of future research concerns the hypothesis that has been
formulated in Sect. 4.6.2, stating that the critical value of the scavenging
number Scrit can be interpreted as a collector efficiency with regard to
the catalyst flow uniformity.
This hypothesis has been formulated based on the experimental results
obtained in this thesis, using only two manifold geometries. Further inves-
tigations are required to determine (i) whether Scrit indeed corresponds
to a collector efficiency, and (ii) to what extent the collector efficiency
depends on the manifold geometry.
To determine whether Scrit corresponds to a collector efficiency, the crit-
ical scavenging number should be determined based on the correlations
in Sect. 4.5 for several geometrical variants of a single manifold. Mani-
fold parameters expected to influence the collector efficiency are (i) the
entrance of the runners into the diffuser, (ii) the shape of the exhaust
runner centerlines and (iii) the diffuser volume. During the course of an
IWT project [2] in cooperation with an industrial partner, experience has
been gained with regard to the influence of swirl induced by the runner
curvature on the catalyst velocity distribution. This effect proves to be
of major importance, and is usually not captured by RANS-CFD calcu-

lations.
• Section 4.6.1 presents a remarkable analogy between the pulsating flow
in a close-coupled catalyst manifold and a zero-dimensional scalar mixing
process. Based on this analogy, a fast method might be established to
determine the critical scavenging number (i.e. collector efficiency). The
fast method might use a fast concentration measurement which should
be capable of detecting the time-resolved concentration of a trace gas
species in the diffuser, following a stepwise change in the concentration
upstream of the exhaust manifold. By determining the time constant of
this mixing process, and comparing it to the expected time constant in
the assumption of perfect mixing, the collector efficiency can be obtained.
The time constants in question range between 5 and 50 ms. Therefore, the
trace gas injection and detection should be performed with reaction times
of 1 ms or less. This should be possible e.g. by means of a fast response
flame ionization detector (FID) hydrocarbon measurement device.
• The oscillating hot-wire anemometer (OHW) developed during this thesis
may be further improved according to the recommendations in Sect. 3.7.4.
These recommendations follow from the non-dimensional scaling analysis
performed in Sect. 3.7.3.
Wake contamination by the moving wire could be minimized by using
a modified version of the angled 55P14 probe, with longer prongs (i.e.
increasing s from Table 3.2). As Fig. 3.10 shows, an increase in s by
only a few millimeters places the wire in the free stream. Along with the
proposed reduction in the oscillation amplitude xo that follows from the
non-dimensional scaling in Sect. 3.7.3, this might be subject for further
research.
The challenge lies in simultaneously increasing the prong length and the
oscillation frequency. In order to avoid prong vibration, this requires a
sufficient increase in the prong’s mechanical resonance frequency. By ap-
propriately shaping the prongs, their stiffness could be increased. Rather
than merely thickening the prongs, perhaps a slender airfoil could be used
to minimize flow disturbance.
• This thesis has demonstrated that significant flow reversal occurs in large
portions of the catalyst, for a broad range of operating conditions in fired
engine operation. However, the relevance of flow reversal with respect to
catalyst ageing is unknown. Although some literature exists on reverse
flow catalytic reactors (see Sect. 5.2.4), it does not apply to the condi-
tions of automotive catalysts with periodic flow reversal at a much faster
frequency. Nevertheless, Sect. 1.2.1 indicates the strong temperature de-
pendence of thermal ageing. For future research, an experimental set up
might be conceived to generate a known uniform periodic flow reversal
and verify the gradual deterioration in the catalyst conversion under ac-
celerated ageing conditions. Alternatively, a numerical model based on
7.2 Suggestions for future research 237
the heterogenous catalytic kinetics described in App. A.3 can be used to

estimate the importance of flow reversal.
Appendices
239
Appendix A
Catalyst substrate flow
“The catalytic converter gets hotter than any of the other under-car com-
ponents.”
Jeff Beck (English rock guitarist, ◦ 1944)
A.1 Introduction
An automotive catalyst substrate consists of a large number of small parallel
channels. The channel cross-sectional shape depends on the substrate material,
which can be either (i) ceramic or (ii) metal. Metal substrates are usually made
up of honeycomb-like sheets, wound into a circular shape. The cross-section can
be represented as a sine function. A ceramic substrate (or: monolith, brick) is
extruded and subsequently baked, resulting in a quasi perfect array of channels
with square, circular, or other cross-section.
The ability to produce channels with low tolerances on cross-sectional area
is a clear advantage of ceramic substrates. Furthermore, the washcoat is easier
to apply on ceramic than on metal substrates, because of the porous structure.
Ceramics like cordierite (2MgO · 2Al2 O3 · 5SiO2 ) are stronger than earlier ce-
ramics, making it possible to reduce wall thickness and thus increase geometric
surface area of the catalyst.
However, metals have a higher thermal conductivity and lower specific heat
capacity compared to ceramics, making a metal substrate better suited for
rapid warm-up applications, such as a close-coupled catalyst. On the other
hand, metal also has a larger thermal expansion coefficient, which increases
its sensitivity to thermal shock, and causes packaging problems. The channels
of metal and ceramic substrates differ not only in shape. In metal substrates,
‘bumps’ may be indented in the channels, thereby increasing the mass transfer.
241
242 Appendix A Catalyst substrate flow
The noble metal particles enabling the catalytic reactions are embedded in
the washcoat. The washcoat, a porous metal oxide such as aluminum oxide
(Al2 O3 ), is applied onto the substrate in liquid form. The liquid adheres to the
inner walls of the substrate, leaving a layer of washcoat with active metals as it
dries. The washcoated substrates are heated to dry and harden the washcoat.
From the nature of the washcoating process, it is clear that the washcoat
layer thickness is not uniform throughout the channel. More liquid will accu-
mulate in sharp corners, rounding off the original unwashcoated cross-section.
For instance, where unwashcoated ceramic substrates have square channels,
washcoated ceramic substrate channels are often modeled as a circle-in-square
cross-section.
The porous washcoat enlarges the catalytic surface to reduce the diffusion
resistance and increasing the reaction rate. Therefore, the substrate channel
walls are significantly rough. However, since the flow regime is laminar, this
does not directly affect the pressure drop, nor the heat and mass transfer.
Sect. 1.2.1 gives more details on deactivation of automotive catalysts. The
following sections provide background information on the transfer of momen-
tum (i.e. pressure drop) and mass (i.e. catalytic reaction kinetics).
A.2 Momentum transfer

The Reynolds number Re is based on the catalyst channel hydraulic diameter34
d and the mean channel velocity U = U0 /ε , where U0 is the upstream axial
velocity and ε is the open frontal area ratio or porosity (see Sect. 1.2.1). This
assumes incompressible flow, which is fulfilled in automotive catalysts since the
Mach number is typically M a < 0.1.
For stationary flow and a smooth surface, the flow is laminar for Re < 2300
and turbulent for Re > 4000. According to Çengel [27], the surface roughness
and flow fluctuations have a considerable influence on these limits.
Numerical example The typical velocity in a catalyst substrate

channel varies between 0 and 20 m/s. A ceramic substrate of cell den-
sity 600 cpsi (cells per square inch) and wall thickness of 3 mil (1 mil
= 1/1000 inch), results in a hydraulic diameter d = 0.96 mm (taking into
account a porosity ε = 0.85). Assuming a temperature of 500 ◦ C, the dy-
namic viscosity is approximately 3.5 × 10−5 Pa·s. At 1.2 bar, the density
equals 0.54 kg/m3, so Re varies between 0 and 300. These is the typical
range for Re inside an automotive catalyst.
The cross-sectional area ratio between the catalyst and the upstream
connecting pipe is around 4 : 1 to 10 : 1, making the upstream average
velocity in a pipe of diameter 50 mm roughly 40 to 100 m/s. At the
same conditions of temperature and pressure this yields an upstream
Reynolds number between 30 000 and 80 000.
34 The hydraulic diameter is defined as d = 4S/P , where S and P are the cross-sectional
area and the perimeter of the channel. For square channels, d corresponds to the width. For
circular channels, d corresponds to the diameter.
A.2 Momentum transfer 243
The above numerical example demonstrates that the flow in a channel is

laminar, while the upstream flow is turbulent. This means that the flow will
relaminarize upon entering the channel.
The pressure drop over a channel is defined as the difference in total pressure
between in- and outlet:
∆p0 = p0,in − p0,out

ρU02 ρU02
= pin − pout + − (A.1)
2 in 2 out
where p0 and p represent total and static pressure [Pa], respectively and
U0 [m/s] is the velocity outside of the catalyst.
A.2.1 Fully developed laminar flow

The preferred measure to quantify the dimensionless pressure drop is the fric-
tion factor f , here defined as the Fanning friction factor, which equals 1/4 of
the Darcy friction factor:
L ρU 2
∆p = 4f (A.2)
d 2
where L is the length of the channel [m] and U is the channel velocity (U =
U0 /ε ).
The pressure drop in a fully developed laminar flow is caused by viscous
shear forces, acting on the fluid laminae. For this regime, the Navier-Stokes
equations can be solved analytically for a number of geometries, yielding typ-
ically a parabolic velocity profile, known as the Hagen-Poiseuille profile. The
viscous shear stress at the wall is proportional to the fluid dynamic viscosity µ
and the velocity gradient normal to the wall (y-direction):
∂U (y)
τw = −µ (A.3)
∂y
The shear stress, integrated over the channel length is proportional to the
pressure drop ∆p. For fully developed laminar flow this yields the following
expressions for the friction factor f (Shah and London [92]):

f Re = 16 for a circular cross-section
(A.4)
f Re = 14.227 for a square cross-section
The effect of wall roughness is negligible for laminar flow.
A.2.2 Developing laminar flow

Near the catalyst entrance, a complex flow field is established as the flow con-
tracts due to the finite thickness of the catalyst substrate walls t. The porosity
ε is a measure of the contraction of the flow field. The contribution of this

effect to the overall pressure drop is discussed in Sect. A.2.3.
Immediately following the first entrance region, the velocity profile inside
the channels is roughly uniform. At this point, the hydrodynamic boundary
layer starts to develop. After some distance (denoted the entrance length Le ),
the parabolic velocity profile is established, corresponding to fully developed
flow.
The hydrodynamic entrance length Le is defined as the lengthwise coordi-
nate where the maximum velocity reaches 99 % of the maximum velocity in the
fully developed regime. Shah and London [92] give a correlation by Chen (1973)
based on numerical results by Friedmann (1968):
Le 0.6
= + 0.056 Re ' 0.056 Re (A.5)
d 0.035 Re + 1
which yields Le ' 11 mm for d = 1 mm and Re = 200.
In the near-wall region, viscous shear forces slow down the fluid, causing
the velocity profile to exhibit a drop in velocity near the wall, and a region
of higher velocity in the center of the pipe. This momentum transfer across
the fluid laminae causes the pressure to increase as the fluid flows further
through the pipe. The developing flow thus causes an extra pressure drop.
The total pressure drop caused by a developing laminar flow is characterized
by the apparent friction factor fapp :
L ρU 2
∆p = ∆pfully developed + ∆pdeveloping = 4fapp (A.6)
d 2
where fapp is always greater than the fully developed friction factor f given by
Eq. (A.4).
There is no analytical expression for laminar developing flow. Either numer-
ical or experimental techniques are used to estimate fapp . Shah and London [92]
present numerical results for a circular and a square channel as a function of
the dimensionless lengthwise coordinate z + , defined as z + = z/(d Re) .
Shah and London [92] give a relation for fapp Re, which is fitted to numerical
results by Liu (1974) and Hornbeck (1964). This is an approximate numerical
solution, where as the correlation by Schmidt (1971) is obtained by solving the
Navier-Stokes equations exactly:
.√
+
3.44 f Re + K ∞ /z − 3.44 z+
fapp Re z + = √

+ . (A.7)
z+ 1 + C z+2
For a circular cross-section, f Re = 16, C = 0.000 212 and K∞ = 0.3125 (Shah

and London [92]). For a square cross-section, f Re = 14.227, C = 0.000 290
and K∞ = 0.3575 (Shah [91]).
Equation (A.7) can be used to evaluate the local friction factor along the
length of the channel. To obtain the pressure drop over the entire length,
Eq. (A.7) is evaluated at z + = L+ = L/(d Re) .
The incremental contribution to the pressure drop of the boundary layer

development can be singled out from the fully developed pressure drop by
subtracting f Re from fapp Re given by Eq. (A.7). The resulting pressure drop
is written as a friction factor:
fapp Re − f Re
fdev = (A.8)
Re
Numerical example The contribution of the boundary layer devel-
opment to the overall pressure drop cannot be neglected, in particular
for short catalysts, or at a relatively high Reynolds number. For square
channels, the following table gives the relative increase in pressure drop
due to the boundary layer development, as a function of the dimensionless
catalyst length L+ :
L+ fdev /f
- -
∞ 0
100 0.0003
10 0.0025
1 0.025
0.1 0.24
0.01 1.7
0 ∞
For a 50 mm long catalyst, with square channels measuring d = 1 mm and

at Re = 200, L+ = 0.25. From Eq. (A.7), fapp Re = 15.617. Therefore,
the pressure drop is fdev Re/f Re ' 10 % greater than the pressure drop
for fully developed laminar flow.
A.2.3 Entrance and exit losses

Chapter 5 in Kays and London [57] deals entirely with contraction (i.e. en-
trance) and expansion (i.e. exit) loss in heat exchangers.
The flow entering the channel experiences a cross-sectional change quanti-
fied by the porosity ε. A separation zone appears at the entrance, causing a
static pressure drop ∆pc which can be split up into two parts: a (i) reversible
pressure drop caused by the area change and the resulting flow acceleration
and (ii) an irreversible pressure drop, which is characterized by the entrance
pressure loss coefficient Kc :
ρU 2 ρU 2
∆pc = (1 − ε) + Kc (A.9)
| {z 2 } | {z2 }
i ii
At the channel exit the flow expands, creating a static pressure rise ∆pe
which again consists of a (i) reversible part due to the flow deceleration and an
(ii) irreversible part, characterized by the exit pressure loss coefficient Ke :
ρU 2 ρU 2
∆pe = (1 − ε) − Ke (A.10)
| {z 2 } | {z2 }
i ii
The combined irreversible pressure drop (Kc + Ke ) ρU 2 /2 equals the total

pressure drop. Kc and Ke are a function of the area ratio ε, and to a lesser
extent of the channel Reynolds number Re.
Kays and London [57] provide charts of Kc and Ke indicating the influence
of ε and Re for various geometries. The following correlations have been fitted
based on [57] by Van den Bulck [100] to incorporate the influence of ε and Re
in a functional form. For turbulent flow:
Kc = 0.4 1 − ε2 + 1.16 Re−1/4

2 (A.11)
Ke = (1 − ε) − 1.16 ε Re−1/4
and for laminar flow:

Kc = 0.4 1 − ε2 + 0.78

2 (A.12)
Ke = (1 − ε) − 0.756 ε
The curves for Kc and Ke in Fig. A.1 are plotted using Eqs. (A.11) and (A.12).
Numerical example Assuming laminar flow (Eq. (A.12)) and a typ-
ical porosity of ε = 0.85 (see Table 1.2), the entrance (i.e. contraction)
and exit (i.e. expansion) pressure drop coefficients are Kc ' 0.89 and
Ke ' −0.62. The overall pressure drop of the combined entrance and
exit loss is Kc + Ke ' 0.27.
Wendland et al. [110] estimate the contribution of entrance and exit losses
as a whole to be around 5 % of the total catalyst pressure loss for a typical
400 cpsi monolith.
A.2.4 Flow acceleration

When the temperature of the flow inside the channel changes from inlet to
outlet, the density is not constant. For an operating catalytic converter, heat
is produced by the exothermic catalytic reactions (see Sect. 1.2.1). As such,
the density decreases and the velocity increases along the length of the catalyst
channel. This flow acceleration causes an additional pressure difference between
inlet and outlet.
Kays and London [57] (Eq. 2.26a) present a general formula for static pres-
sure drop in a heat exchanger, incorporating the term accounting for flow accel-
eration. The factor 2 in Eq. 2.26a is an approximation of 1 + ε2 , thus yielding
the correct term to account for flow acceleration:
ρU 2

ρin
∆pacc = 1 + ε2 −1 (A.13)
ρout 2
1.2
1
Laminar
0.8
0.6 Re = 2000
Re = 10 000
Re → ∞
0.4
Kc, Ke (-)
0.2
0
Re → ∞
Re = 10 000
-0.2 Re = 2000
-0.4 Laminar
-0.6 Kc (-)
Ke (-)
-0.8
0 0.5 1
ε (-)
Figure A.1 – Contraction (Kc ) and expansion (Ke ) pressure drop coefficients as a
function of open frontal area ε and Re (adapted from Kays and London [57], using
Eqs. (A.11) and (A.12))
Note that U here represents the velocity inside the channel yet near the en-
trance, where the density is still unchanged. In the expression for total pressure
drop, the factor 1 + ε2 disappears:
ρU 2

ρin
∆p0,acc = −1 (A.14)
ρout 2
A.2.5 Oblique entrance

When the upstream velocity vector is not in line with the channel centerline,
an additional pressure drop occurs. The velocity vector misalignment causes
an enlarged recirculation zone at the channel entrance, reinforcing the entrance
effect discussed in Sect. A.2.3.
The doctoral research of Haimad [43] proves that this effect greatly influ-
ences the CFD-predicted catalyst flow distribution. Since the flow field expands
in the diffuser, one can imagine that only the region of the catalyst inlet face
which is aligned with the upstream runner experience velocity vectors with a
low angle-of-attack (provided the runner is in line to the catalyst axis). Near
the edges of the catalyst, high entrance angles may occur, due to the diffusion
effect, recirculation zones and vortices generated by the curved runners. Ac-
cording to Haimad [43] and Benjamin et al. [15], this may explain why CFD
often overpredicts the flow uniformity compared to experimental data.
Küchemann and Weber [32] propose an upper limit to the pressure loss
caused by oblique entry:
ρUt2
∆pobl,max = (A.15)
2
where Ut is the transverse velocity component.
This upper limit is based on the idea that for an ideal unseparated flow,
the force exerted per unit frontal area by the oblique flow on a series of par-
allel plates equals F = ρUt2 /2 (N/m2 ). For ‘real’ oblique flow over a non-
aerodynamically streamlined body (e.g. a series of parallel plates or a mono-
lith) and assuming that no suction can be sustained without flow separation,
F should equal zero. Equation (A.15) follows directly from this assumption.
Written as a pressure drop coefficient, based on the dynamic pressure,
Eq. (A.15) becomes:
ρU 2

Kobl,max = ∆pobl,max
2
2 2 2
Ut U0 Ut
= = = ε2 tan2 β (A.16)
U U U0
where β is the angle-of-attack, corresponding to the angle between the catalyst

axis and the upstream velocity vector [U0 , Ut ].
A.2.6 Overall
The overall total pressure drop, referenced to the dynamic pressure inside the
channel (yet near the entrance) is:
ρU 2

L ρin ρin ρin
∆p0 = 4fapp + Kc + K e + − 1 + Kobl (A.17)
d ρm ρout ρout 2
where the factors ρin /ρ take the change in density with respect to the en-
trance conditions into account. fapp,m is evaluated at a mean temperature and
RL
ρm represents a mean density, defined so that fapp,m ρin /ρm = z=0 fapp (z)·
(ρin /ρ (z) ) dz/L .
The following expression gives the difference in static pressure from a loca-
tion upstream to a location downstream of the monolith, which incorporates
the reversible static pressure changes and in- and outlet:
A.3 Mass transfer 249

L ρin ρin
+ K c + 1 − ε2 + Ke − 1 − ε 2

∆p = 4fapp
d ρm ρout
2

ρin ρU
+ 1 − ε2 − 1 + Kobl (A.18)
ρout 2
where U still represents the velocity inside the channel yet near the entrance,
where the density equals the upstream density. In the absence of density
changes as is the case on the isochoric flow rig, Eqs. (A.18) and (A.17) re-
duce to the same expression for static and total pressure drop:
ρU 2

L
∆p0 = ∆p = 4fapp + Kc + Ke + Kobl (A.19)
d 2
The expressions described in the above sections have been used to describe
the relation between the flow rate and pressure drop in the laminar flow ele-
ment meter (LFE) (see Sect. 2.4.2). The LFE is used as reference flow rate
measurement during the measurements on the isochoric (CME) flow rig (see
Sect. 2.2.2).
A.3 Mass transfer

A.3.1 General
The equations of conservation of mass and species for a three-dimensional sys-
tem are:
∂ρ
~
= −∇ · ρ U (A.20)
∂t
∂Ci
~ + ∇ · Di ∇

~ (Ci ) +Ri
= −∇ · Ci U (A.21)
∂t | {z } | {z }
convection diff usion
where ρ is the density [kg/m3 ], U~ is the velocity [m/s], Ci is concentration of

species i [kmol/m ], Di is the molecular diffusivity [m2 /s] and Ri is the reaction
3
rate for species i [kmol/m3 s].

Given the strong anisotropic catalyst geometry (§1.2.1), the convection and
diffusion terms in Eq. (A.21) can be split into axial and transversal components.
Three time scales can be defined based on the splitted version of Eq. (A.21):
• convection time or residence time τU = L/U
• axial diffusion time τD,a,i = L2 /Di
• transversal diffusion time τD,t,i = d2 /Di

where d is the channel hydraulic diameter [m]. These three time scales combine
into two dimensionless Peclet numbers:
• axial Peclet number P ea = τU /τD,a,i
• transversal Peclet number P et = τU /τD,t,i
For an automotive catalyst, the axial diffusion time is much larger than
the typical residence time of 10 ms (P ea 1). Axial diffusion is therefore
neglected. Because of the small channel size, the transverse diffusion time is
smaller than the residence time (P et 1). As such, the transverse concentra-
tion distribution within a single channel is considered uniform.
The time scale that limits the chemistry depends on the catalyst tempera-
ture: at a low temperature, transverse diffusion between gas and surface phases
is faster than the time scale of the catalytic reactions occurring in the surface
phase. This process is denoted mass transfer-limited. At a high temperature,
the reactions occur faster than the diffusion time scale, and the process is rate
limited or kinetically limited.
The catalytic reactions can be modeled using (i) a homogeneous approx-
imation, or (ii) a lumped-parameter heterogeneous model. The latter best
describes the reaction kinetics and is generally used in the literature.
A.3.2 Homogeneous reaction kinetics

The homogeneous first order model assumes that all reactions take place in the
gas phase, leading to a reaction rate Ri = −Ki Ci . For the one-dimensional
stationary case, this reduces Eq. (A.21) to:
∂ (Ci U )
= −Ki Ci (A.22)
∂z
where z is the lengthwise coordinate [m]. Assuming U constant35 , this yields
an exponential lengthwise concentration distribution:

Ki z
Ci (z) = Ci,0 exp − (A.23)
U
Based on Eq. (A.23), the conversion efficiency can be defined as the ratio of
total converted amount of species i given a certain (non-uniform) catalyst veloc-
ity distribution, to the converted amount given a uniform velocity distribution
with the same mean velocity:
I
Ki L
1 − exp − ρ U (x, y) dA
A U (x, y)
ηC,i = (A.24)
Ki L
1 − exp − ṁ
Um
35 Catalyst flow is incompressible (M a < 0.3), however because of the generated reaction
heat, the density decreases and the gas accelerates as it passes through the catalyst.
A.3 Mass transfer 251
This equation is used in Sect. 1.2.1 in Fig. 1.5 to demonstrate the influence
of the flow uniformity on the catalyst conversion efficiency.
A.3.3 Heterogeneous reaction kinetics

Alternatively, a catalyst channel can be modeled as a one-dimensional lumped-
parameter system with gas and surface phases. In such heterogeneous reaction
model, all reactions occur in the surface phase. Transverse diffusion between
gas and surface phase is modeled by a mass transfer coefficient km [m/s]. km
represents the ratio of diffusive flux to the surface [kmol/m2 s] to the concentra-
tion difference between bulk and surface phase C − Cs [kmol/m3 ]. Multiplying
this diffusive flux J = km (C − Cs ) with the geometric surface area of the
channel av = 4/d [m−1 ], this yields a reaction rate expression −km av (C − Cs )
[kmol/m3 s]. Upon substitution in Eq. (A.21), the one-
dimensional stationary film model for mass transfer becomes:
∂ (Ci U )
= −km,i av (Ci − Cs,i ) (A.25)
∂z
The Sherwood number is the non-dimensional mass transfer coefficient, de-
fined as Sh = km d/Dm . Sh is the mass transfer analogous of the Nusselt
number N u. Sh and N u can be determined analytically for fully developed
laminar flow [92]. Numerical and experimental correlations exist which take
the developing flow region into account.
The surface phase species concentration is determined by its own conserva-
tion equation. In steady state this equation becomes:
km,i (Ci − Cs,i ) = Rs,i (Ts , Xs,1 . . . Xs,I ) (A.26)

where Ts is the surface temperature [K], X denotes species molar fraction
[kmol/kmol], I is the number of species. The reaction rate expressions Rs,i are
determined experimentally. Catalytic reactions rate expressions are typically
of the following form, including inhibition factors in the denominator:
αr
Q
r Xr
Rs,i (Ts , Xs,1 . . . Xs,I ) = ks,i (Ts ) Q γr (A.27)
r (1 + βr Xr )
and the rate coefficients have an Arrhenius-type temperature dependence:

Ea,i
ks,i (Ts ) = Ai exp − (A.28)
R Ts
Numerical values for the rate equations can found in the literature (e.g.
Mezaki and Inoue [73]). The set of partial differential equations Eqs. (A.25)
and (A.26) for heterogeneous reactions cannot be solved analytically in the
same way as Eq. (A.22) for homogeneous reactions. It should be solved using
numerical integration.
Appendix B
Exhaust stroke flow similarity
The analytical derivation below yields expressions for the peak mass flow rates
during the blowdown (Eq. (2.1)) and the displacement phase (Eq. (2.2)) for a
fired engine, as well as for the CME flow rig used in this thesis. These expres-
sions are used in the discussion of exhaust stroke flow similarity in Sect. 2.3
between the CME flow rig and a fired engine.
For this derivation, in-cylinder heat loss and blow-by leakage are neglected.
Air is taken as working fluid, with thermodynamic properties evaluated at a
fixed mean temperature.
From the conservation of mass and energy (and assuming isentropic com-
pression and expansion, and isochoric combustion) result the following expres-
sions describing the relation between intake and residual state:
γ γ−1 !
ρe Vi pe Vi ∆Tc V0
= ; = 1+ (B.1)
ρi Ve pi Ve Ti Vi
where ρ is the density [kg/m3 ], p is the pressure [Pa], V is the cylinder vol-
ume [m3 ], T is the temperature [K] and the subscripts i, e, 0 respectively denote
intake valve closing, exhaust valve opening and top dead center. The adiabatic
temperature rise due to combustion equals ∆Tc = φSf / (cv Lf ), where φ is
the product of the equivalence ratio and the combustion efficiency [-], Sf is
the lower heating value of the fuel (J/kg), cv is the specific heat capacity at
constant volume (J/(kg K)) and Lf is the theoretical air-to-fuel ratio [kg/kg].
Equation (B.1) can easily be derived in three steps: (i) isentropic compres-
sion between Vi and the dead volume V0 :
γ γ−1
p0 Vi T0 Vi
= ; = (B.2)
pi V0 Ti V0
(ii) isochoric combustion at V0 , assuming no change occurs in the working fluid
composition and incorporating Eq. (B.2):
253
254 Appendix B Exhaust stroke flow similarity
p1 T1
= ; T1 = T0 + ∆Tc
p0 T0
p1 ∆Tc
⇔ = 1+
p0 T0
γ−1
p1 ∆Tc V0
⇔ = 1+ (B.3)
p0 Ti Vi
(iii) isentropic expansion from V0 to Ve , before exhaust valve opening:

γ γ−1
pe V1 Te V1
= ; = (B.4)
p1 Ve T1 Ve
Combining Eqs. (B.2), (B.3) and (B.4) yields Eq. (B.1).
The exhaust stroke is divided into blowdown and displacement phases. The
blowdown phase is regarded as the expansion of the residual cylinder pres-
sure under constant cylinder volume. The displacement phase is regarded as
volumetric expulsion of gas at quasi-constant pressure.
Assuming a constant cylinder volume and adiabatic sudden transition, the
state evolution with respect to the residual state (denoted with subscript e) is
described by:
γ
ρ (θ) m (θ) p (θ) m (θ)
= ; = (B.5)
ρe me pe me
where θ (= ωt) is the crankshaft angle [rad] and m is the gas mass contained
in the cylinder [kg].
Approximating the exhaust manifold pressure with the atmospheric pres-
sure pa , the following expression (see e.g. Heywood [47], App. C) gives the mass
flow rate over the exhaust valves assuming compressible restricted flow:

p (θ) pa dm (θ)
ṁ (θ) = Cd ne πde h0e (θ) p ·f =− (B.6)
rT (θ) p (θ) dt
 s
 pa γ1
γ−1γ
γ
γ−1
2γ pa

pa


p γ−1 1 − p ; ppa > γ+12
(a)
f =
p (θ) γ+1 γ
 √γ 2 2(γ−1)
 γ−1
; ppa 6 γ+12

γ+1 (b)
where conditions (a) and (b) respectively denote subsonic and sonic (choked)
regime. Cd is the exhaust valve discharge coefficient [-], r is the specific gas con-
stant [J/(kg K)], ne and de are the number per cylinder [-] and the diameter [m]
of the exhaust valves. The discharge coefficient Cd for poppet exhaust valves
is based on Fig. B.1, compiled from empirical data by Heywood [47]. Since
the flow pattern depends on the lift height, so does the discharge coefficient.
Figure B.2a shows the pressure loss coefficient K [-] for a rotating valve, based
255
Flow pattern Flow pattern

at low lift at high lift
Figure B.1 – Discharge coefficient Cd for exhaust valves (Source: [47])
0.9
0.8
0.7
0.6
Cd (-)
0.5
0.4
0.3
0.2
0.1
0
0 30 60 90
Valve angle (°)
(a) (b)
Figure B.2 – (a) Pressure loss coefficient K and (b) discharge coefficient Cd for
rotating valve (Source: [74])
on Miller [74]. Figure B.2b shows the discharge coefficient Cd as a function of

2
the valve angle, determined from K using the relation K = Cd−1 − 1 .
The lift height h0e [m] in Eq. (B.6) is approximated by:

1 θ − θe
h0e (θ) = he 1 − cos 2π (B.7)
2 ∆θ
where he is the maximum valve lift height [m], θe is the start of exhaust valve
opening [rad] and ∆θ is the exhaust valve opening duration [rad]. The expres-
sion for h0e (θ) is further approximated using a Taylor expansion for small values
2
of (θ − θe ) /∆θ, and substituting θ−θe = ωt results in h0e (t) = he (π/∆θ) ω 2 t2 ,
where 2πω/60 equals the engine speed N [rpm]. Note that p and T actually
represent total conditions, although this is neglected in the remainder of the
analytical derivation in order to keep the equations solvable.
Substituting Eq. (B.5) in Eq. (B.6) and incorporating the above approxi-
mation for h0e results in:
√ γ+1 −γ !
ne π 3 de he 2 2 rTe m

d m 2
pa m
= −Cd 2
ω t f (B.8)
dt me ∆θ Ve me pe me
For this partial differential equation to be solvable, the function f is ap-

proximated using the following expression:
γ1 !− γ+1
2 γ−3 γ1 !
m pe pe 2γ
m pe
f ' f 0 = cf −1 (B.9)
me pa pa me pa
where cf is a dimensionless fit constant. The exponents −(γ + 1)/2 and

(γ − 3)/(2γ) in Eq. (B.9) have no physical meaning. They are selected with the
sole purpose of rendering Eq. (B.8) solvable. Figure B.3 shows f and f 0 accord-
ing to Eq. (B.9), as a function of the gas mass remaining in the cylinder m/me .
Note that initially, m/me = 1, until the cylinder pressure equalizes with the
1/γ
atmospheric pressure (p = pa ) and m/me = (pa /pe ) (from Eq. (B.5)).
Substituting Eq. (B.9) in Eq. (B.8) yields a solvable partial differential
equation that describes the approximate evolution of m, the gas remaining in
the cylinder:
√ − γ2 γ1 !
ne π 3 de he 2 2 rTe pe

d m m pe
= −Cd ω t −1
dt me ∆θ2 Ve pa me pa

1 m
= −Kt2 −1 (B.10)
A me
The solution to this partial differential equation is:

257
1
1
f (pe/pa= 5) 0.9
0.9 f ’(pe/pa= 5)
f (pe/pa= 2)
0.8
0.8
f ’(pe/pa= 2)
0.7
0.7
0.6
M / M1 (−)
0.6
0.5
f (−)
0.5
0.4
0.4
0.3
0.3
0.2
0.2
analytical, Eq. (20)
0.1 experimental
0.1
0
0 0 1 2 3 4 5 6
0 0.2 0.4 0.6 0.8 1
t / tmax = t / (2A/K)1/3 (−)
m/me (−)
Figure B.4 – Flow rate predicted by

Figure B.3 – Approximation of f ac-
Eq. (B.12) compared to measured run-
cording to Eq. (B.9)
ner flow rate

m K 3
(t) = A + (1 − A) exp − t (B.11)
me 3A
3
√ − γ2 γ1
rTe pe pa
where K = Cd ne π∆θd2e he ω 2 Ve pa [s−3 ] and A = pe [-]. The flow rate
yields:

dm 1 2 K 3
ṁ (t) = − (t) = me K − 1 t exp − t (B.12)
dt A 3A
The maximum mass flow rate during blowdown ṁ1 [kg/s] occurs when the
derivative of Eq. (B.12) is zero. This corresponds to a time after the exhaust
valve opening that is equal to:
1/3
t1 = (2A/K ) (B.13)
t1 [s] represents the characteristic time scale for the blowdown process, where
K [s−3 ] and A [-] are the variables defined above. The peak mass flow rate at
t1 equals:
2/3 1/3
ṁ1 = ṁmax = me (2/e) (K/A) (1 − A) (B.14)
Figure B.4 shows the good correspondence between the analytically pre-
dicted mass flow rate according to Eq. (B.12) and the experimentally deter-
mined flow rate in an exhaust runner on the CME flow rig.
Substituting Eq. (B.1) and choosing the fit constant in Eq. (B.9) cf =
−2
2 (2/e) ' 3.695 yields:
√ 1 γ−7
ne π 3 de he rTi 3 Vi
6
ṁ1 = ρi Vi ω 2Cd 2
∆θ ωVi Ve
γ−8
!
γ−1 6γ − 3γ 4
∆Tc V0 pi
· 1+ ·
Ti Vi pa
| {z } | {z }
ii i
 
1
1 γ−1 γ !
 p γ V ∆Tc V0 
 i i
· 1+ − 1 (B.15)

 pa Ve Ti Vi 
| {z } | {z }
i ii
During the displacement phase, cylinder pressure and density are assumed
constant. From the conservation of mass results the following expression for
the peak mass flow rate ṁ2 [kg/s] during displacement:
− γ1 γ−1 ! γ1
πb2

ω pi ∆Tc V0
ṁ2 = ρi s 1+ (B.16)
4 2 pa Ti Vi
| {z } | {z }
i ii
Equations (B.15) and (B.16) correspond respectively to Eqs. (2.1) and (2.2)
in Sect. 2.3.
Elaborating Eq. (B.13) with substitution of Eqs. (B.1) yields:
√ − 13 1−γ
1 ne π 3 de he rTi

Vi 6
ωt1 = Cd 2
2 ∆θ ωVi Ve
3γ1 γ−1 ! 2−γ 6γ
pi ∆Tc V0
· 1+ (B.17)
pa Ti Vi
| {z } | {z }
i ii
Equation (B.17) is used to plot the analytical values of the blowdown time scale
in Fig. 2.16 in Sect. 2.3.
Parts of this appendix have been published in an international journal with
review:

Appendix C
Velocity measurement
techniques
Several experimental techniques exist for measuring the local instantaneous

fluid velocity. For research in most turbulent gas flows where velocity fluctua-
tions are of interest, two velocity measurement approaches are in competition:
thermal anemometry and optical anemometry.
The following is an overview of the applicability of these techniques, specif-
ically to exhaust system flows.
C.1 Thermal anemometry

Thermal anemometry measures the local fluid velocity indirectly, based on the
convective heat transferred from an electrically heated sensor element (wire,
film, fiber, . . . ). Typically, the sensor’s resistance (and consequently its tem-
perature) is kept constant by incorporating it in a Wheatstone bridge with
feedback amplifier. This mode of operation is referred to as constant tem-
perature anemometry (CTA), and is characterized by a very high bandwidth.
Further details are given in Sect. 3.2.
Any change in the fluid flow that changes the amount of heat transferred
from the sensor is detected by the CTA. As such, not only velocity can be
measured but also temperature, changes in fluid mixture composition or phase
changes in multi-phase flow.
Excellent historical reviews of thermal anemometry are given by Frey-
muth [39] and Fingerson and Freymuth [37]. Freymuth [40] comprises a detailed
bibliography of thermal anemometry. The main reference work by Bruun [25]
provides a detailed background of hot-wire and hot-film anemometry, and selec-
tion criteria for different flow conditions. Data reduction methods for obtaining
259
260 Appendix C Velocity measurement techniques
velocity values from raw anemometer bridge voltages (e.g. velocity calibration,
temperature correction) used in this thesis are all according to Bruun [25].
The most widely used application is hot-wire anemometry (HWA) in con-
stant temperature mode. Advantages of HWA include:
• High bandwidth — A typical hot-wire probe operated in gas using a con-

stant temperature anemometer bridge features a flat frequency response
up to approximately 30 kHz at velocities between 0.5 and 50 m/s. Mea-
surements at several hundred kilohertz are possible for higher velocities.
• Spatial resolution — A standard hot-wire sensor is 5 µm in diameter and

1.25 mm long. Even smaller sensors can be used in boundary layer flows.
The small sensor size yields a good spatial resolution in comparison to
other techniques.
• Signal analysis — By contrast to an optical anemometer, the output

signal of the HWA is a time-continuous signal, which enables the use of
conventional time-domain and frequency-domain analysis.
• Low-velocity sensitivity — The sensitivity of the output voltage versus

velocity is high at low velocity. Given the small sensor dimensions, HWA
is therefore ideally suited for boundary layer research.
• Signal-to-noise ratio — High quality constant temperature anemometer

bridges exhibit a very low noise level (−40 dB or 1/10 000 ), when compared
to LDA (−30 dB or 1/1000 ).
Section 3.2 discusses the hot-wire anemometer used during this thesis.
C.2 Optical anemometry

Optical anemometry comprises mainly laser Doppler anemometry (LDA) and
particle image velocimetry (PIV). A reference book by Albrecht et al. [6] gives
an overview of LDA measurement principles, signal processing and application
issues.
In its most typical form, LDA uses a pair of intersecting laser beams, where
the measuring volume is defined by the intersection of both beams. As shown
in Fig. C.2, an interference pattern of successive light and dark planes (or
fringes) is formed by the intersecting beams, according to regions of positive
and negative interference. The planar fringes are directed along the beam
bisector and the normal to the beam plane. The inter-fringe spacing ∆x =
λb / (2 sin θ), where λb is the laser light wavelength (m) and 2 θ is the intersecting
beam angle.
If the flow is seeded using sufficiently small tracer particles (e.g. oil mist
droplets, fine solid powders,. . . ), a particle traveling through the fringe pat-
→
−
tern at a velocity U (Ux , Uy , Uz ) scatters the laser light, which is picked up by
a photo-detector. The scattered light intensity varies as the seeding particle
Velocity = distance/time
C.2 Optical anemometry 261
Particle-seeded flow
Signal
∆t (measured)
∆x (known)
Time
Figure C.1 – Schematic diagram of LDA working principle and resulting Doppler
signals
travels through the light or dark regions. Assume the beam bisector is along
the y-direction, and the fringe planes are parallel to the y-z plane. The mag-
nitude of the projection of the velocity normal to the fringe plane |Ux | can be
determined based on the known fringe spacing ∆x, the amount of fringe pulses
caught on the detector and the measured time ∆t between individual fringe
pulses (see Fig. C.2).
The seeding particles should be (i) small enough to accurately follow the
fluctuating flow, (ii) smaller than the fringe spacing to yield an optimal detec-
tor signal, yet (iii) not too small to still scatter enough light to the detector.
Furthermore, light is scattered in all directions, yet at a different intensity
depending on the angle to the beam bisector. The intensity of backward scat-
tered light (i.e. towards the laser beam source) is ' 100 times smaller than the
forward scattered light intensity. The highest sensitivity is therefore achieved
using a the detector that picks up the forward scattering light signals.
Introducing seeding into the flow under investigation has some practical
drawbacks. Depending on the type of seeding substance, periodic cleaning of
the test set up is required. Particularly, the viewing windows should be free of
deposits to ensure optimal optical access. In gas flows, oil mist aerosol seeding
is typically preferred over solid seeding.
The direction or the sign of the velocity Ux /|Ux | can only be determined
if the intersecting beams have slightly different wavelengths. Applying a fre-
quency shift fsh to one of the beams causes the fringes themselves to move at
a velocity of ±fsh λb / (2 sin θ). Thus, a stationary particle (Ux = 0) scatters
light at a frequency corresponding to the frequency shift fsh . Particles with
LDA frequency shifting

fmax
U (unshifted)
U (shifted)
Frequency f (= ∆t- 1)
fshift
fmin
Umin 0 Umin Umax Umax

Velocity
Figure C.2 – Principle of frequency shifting for resolving the directional ambiguity
in LDA
negative velocity (Ux < 0) scatter light at a frequency fsh − |Ux | (2 sin θ) /λb .
In conclusion, bidirectional velocity can be measured using frequency shift-
ing, where the minimum measurable velocity is proportional to the frequency
shift, or Umin = −fsh λb / (2 sin θ). For a typical LDA system with a maximum
frequency shift of 40 MHz and λb ' 500 nm, Umin is around −100 m/s.
To measure two or three velocity components, additional intersecting beam
pairs are required. Each beam pair has a different wavelength (i.e. different
color). Note that the difference in wavelength between beam pairs is many
orders of magnitude larger than the wavelength shift corresponding to the
frequency shifting. Using color filters, the scattered light is split into each
component and directed to one photo-detector for each velocity component.
Advantages of a LDA system with frequency shifting include:
• Non-intrusiveness — The main advantage with respect to thermal anemo-

metry is that optical measurement techniques do not disturb the flow.
Nevertheless, the introduction of seeding and the need for high-quality
optical access may require some flow disturbance.
As a consequence of the non-intrusive nature of LDA measurements, there
is no temperature limit to the flow. LDA can be used e.g. in fired burner
research, provided that adequate seeding is used.
• Directional sensitivity — An LDA system with frequency shifting resolves

the velocity with a very large minimum measurable velocity. By contrast,
thermal anemometry as such is incapable of detecting the flow direction.
• High accuracy — Both LDA and HWA systems feature a similar accuracy
up to 0.1. . . 0.2 % in carefully controlled experiments, or most likely 1 %
in practical applications.
• Spatial resolution — The spatial resolution is determined by the size of

the measuring volume. For a typical LDA set up, the volume is 150 µm
by 2 mm, compared to 5 µm by 1.25 mm or less for HWA. Clearly, LDA
is less suited for near-wall flow investigation than HWA.
However, there are some important disadvantages to LDA which make it

difficult to use for measuring the velocity distribution in a catalyst, and in
general any internal flow in a confined geometry with curved boundaries:
• Optical access — In measuring a confined flow, the laser beams require an

optical passage through the boundary. Light travels at a slightly differ-
ent speed through different media (e.g. air, water, glass, plexiglass). The
speed is also a function of the temperature of the medium. As a result,
the laser beam undergoes refraction at interfaces between different me-
dia (and across important temperature gradients). The optical passage
should ensure that the refraction of the beams is such that the beams
intersect in a common measuring volume. In practice, this can only be
achieved using a plane, smooth sheet of transparent material. Further-
more, the quality of the material should be high enough: (i) smooth plane
surfaces, (ii) no gas bubble or other inclusions, (iii) no changes in the re-
fractive index within the material. Using anything other than optical
grade glass results in a drastically reduced data rate or an entire loss of
signal.
Inserting a flat surface for optical access into a curved boundary results
in a local distortion of the flow. As such, the optical passage should be as
small as possible. On the other hand, the laser beam intersecting angle 2 θ
should be as large as possible to minimize the measuring volume, which
results in a good spatial resolution.
• Seeding — Seeding particles are required for LDA measurements. The

amount of acquired velocity measurements per time is called the data
rate. The data rate is roughly proportional to the concentration of seeding
particles crossing the measurement volume.
To obtain a good velocity distribution across several points in a measure-
ment region, the seeding concentration should be as uniform as possible
in all points. The seeding is therefore introduced into the flow sufficiently
far upstream of the measurement region, to obtain a uniform distribu-
tion. However, certain flow phenomena inhibit the seeding distribution.
In particular, compact exhaust manifold flows are characterized by phe-
nomena like flow separation, recirculation and vortex break-down, which
are all prone to cause seeding problems.
Phase-locked measurements are used to resolve periodic flows (see
Sect. 2.5). For HWA, the continuous output signal makes phase-locking
very straightforward. Since the velocity samples for LDA are not equidis-
tant in time, the phase-locked samples are binned in B phase intervals
of equal duration. The N (b) samples acquired in each bin b are av-
→
−
eraged, producing B bin-averaged velocity values U (b) per cycle (e.g.
◦
using B = 36 equally-sized bins from 0 to 360 ca corresponds to one
bin-averaged velocity value every 10 ◦ ca).
Experience learns that the number of samples per bin N (b) can vary
significantly through the cycle. It is often difficult to obtain enough sam-
→
−
ples in each bin to produce a bin-averaged velocity U (b) with sufficient
accuracy. This requires either a longer measuring time, or reducing the
number of bins per cycle B, which results in a loss of temporal resolution.
The phase-locked HWA approach is not hampered by this, and the tem-
poral resolution is only limited by the anemometer bandwidth. For a
moderate bandwidth of 10 kHz, a temporal resolution of 1.8 ◦ ca is ob-
tained at an engine speed of 3000 rpm.
These disadvantages make it very difficult for LDA to measure the time-resolved
velocity distribution throughout the entire catalyst cross-section.
Kim et al. [59] and Park et al. [81] discuss phase-locked LDA measurement
results obtained in fired engine conditions inside the diffuser, upstream of a CC
catalyst. Measurements are taken along a single line. The authors dissolve liq-
uid titanium(iv) isopropoxide (Ti[OCH(CH3 )2 ]4 ) into the fuel, at a volumetric
concentration between 3 to 7 %. The titanium isopropoxide burns along with
the fuel in the combustion chamber and forms titanium dioxide (TiO2 ). This
solid substance is often used as seeding medium in optical anemometry. It is
particularly used in combustion flow research, since TiO2 features a high melt-
ing point of 1850 ◦ C, with excellent light scattering properties. Solid seeding
has the tendency to accumulate and clog small openings, such as the catalyst
channels. However, Kim et al. [59] claim no clogging occurs at the applied
seeding concentrations. This technique has been initially introduced by Zhao
et al. [114].
Liu et al. [70] discuss phase-locked LDA measurement results obtained
30 mm downstream of a close-coupled catalyst in fired engine conditions. Mea-
surements are shown in a few points along a single line. The authors compare
these measurements to CFD results. The correspondence is not very good,
yet the obtained spatial resolution does not allow for a good validation. No
mention is made of the applied seeding.
Benjamin et al. [14] present LDA measurements downstream of a close-
coupled catalyst, in a fired engine. The setup is basically identical to the one
used by Liu et al. [70]. The authors use a fiber-optic single component LDA
with Ar+ laser and a 10 MHz frequency shift. The focal length in 120 mm,
with a 16 mm beam spacing. The measuring volume is about 2.2 mm long.
The authors [14] use the same titanium isopropoxide mixed with the fuel, as
described earlier [59, 81, 114].
Benjamin et al. [14] report no clogging of the catalyst substrate, even after
extended experiments. However, deposits quickly accumulate on the spark
plugs and in the injector nozzles. The authors report extensive difficulties in
measuring close to the optical windows. Also, in some crankshaft position

windows, the seeding density is considerably smaller than in other windows. In
general, an uneven spatial and temporal distribution of seeding concentration,
along with the loss of signal due to wall reflections make this experimental
technique very difficult to use in exhaust systems.
Hwang et al. [53] provide some qualitative results using phase-locked LDA
on an isothermal flow rig. The authors employ liquid aerosol seeding using
dioctyl phthalate (DOP).
Each of these studies [59, 81, 70, 53, 14] shows the difficulties in obtaining
high resolution results using LDA in exhaust systems.
During the calibration of the oscillating hot-wire anemometer (see Sect. 3)
in a custom-built miniature wind tunnel, a two-component LDA system is used
with frequency shifting on both components. Diethylhexylsebacat (DEHS) oil
mist is used as seeding. The LDA system and calibration wind tunnel are
described in Sect. 3.6.
Particle image velocimetry (PIV) is an optical measurement technique that
uses a high-speed camera to record pairs of successive images of a particle seeded
flow field. Each image pair is separated by a very short time interval, as small
as 2 µs. The two images are cross-correlated. This yields a two-dimensional
particle displacement field, which is equivalent to the velocity field. This type
of PIV which enables quantitative velocity measurements is sometimes referred
to as digital particle image velocimetry (DPIV), as opposed to the older method
used only for high speed flow visualization. Adrian [5] and Westerweel [111]
provide excellent reviews on PIV theory and applications.
Applying PIV in exhaust systems will be plagued by similar problems as
LDA: (i) obtaining an adequate seeding density throughout the cross-section
and the engine cycle, and (ii) providing optical access. The quality of the
optical access is less restrictive for PIV when compared to LDA, due to the
different operating principle. No sources are available in the literature on us-
ing PIV in exhaust systems, although the technique has been frequently and
successfully applied for measuring in-cylinder flows. Brucker [24] discusses a
three-dimensional fast scanning approach using PIV with application in com-
bustion chamber flows.
Appendix D
Modeling one-dimensional
gas dynamics
D.1 Model description

The Euler equations describe the motion of a compressible, inviscid fluid,
whereas the motion of viscous fluids is described by the more general Navier-
Stokes equations [48, 49].
In this thesis, only one-dimensional variations of the primitive variables
pressure p, velocity U and density ρ are considered. In particular, the numerical
model of the exhaust system used in Sect. 5.2.3 and Sect. 5.3.4 assumes that
the primitive variables vary only along the lengthwise axis of large length-
to-diameter ratio components (e.g. exhaust runners, exhaust pipe), and that
components with negligible momentum can be described by a single point model
(e.g. cylinders, diffuser, catalyst substrate).
The one-dimensional Euler equations express the conservation of mass, mo-
mentum and energy in the lengthwise z-direction. For unsteady, one-dimensional
flow in a duct with variable cross-section, including wall friction and heat trans-
fer, the Euler equations in conservative form are:
∂ (ρA) ∂ (ρU A)
+ = 0
∂t ∂z
∂ (ρU A) ∂ ρU 2 A ∂ (pA) ∂A 4A U ρU 2
+ = − +p −f
∂t ∂z ∂z ∂z d |U | 2
∂ (ρEA) ∂ (ρU EA) ∂ (pU A)
+ = − + ρqA (D.1)
∂t ∂z ∂z
where the quantities ρ [kg/m3 ], U [m/s], E [J/kg], p [Pa] denote average values
across the duct cross-section such that the conservation laws are fulfilled. The
267
268 Appendix D Modeling one-dimensional gas dynamics
factor U /|U | ensures that the friction acts counter to the local flow direction.
In the perimeter 4A/d [m], d is the hydraulic diameter [m]. Furthermore, for
gas flows, the ideal gas equation p = ρrT is assumed valid, where r is the
specific gas constant [J/(kg K)]. The total internal energy per unit mass E is
defined as E = e + U 2 /2 . For an ideal gas, e = cv T , where cv is the specific
heat capacity [J/(kg K)] at constant volume.
The Euler equations (D.1) can be written in conservative vector notation
as:
∂ξ ∂F
+ =Q (D.2)
∂t ∂z
where the boldface symbols represent vectors. The conservative variable vec-
tor ξ, the flux vector F and the source term Q are defined as:
0
     
ρA ρU A 2
U ρU
ξ =  ρU A  , F =  ρU 2 A + pA  , Q =  −f 4A
d |U | 2
 (D.3)
ρEA ρU HA ρqA
where the total enthalpy per unit mass H [J/kg] is defined as H = h + U 2 /2 .

For an ideal gas, h = cp T , where cp is the specific heat capacity [J/(kg K)] at
constant pressure.
D.1.1 Decoupling the Euler equations

In order to solve the Euler equations numerically, the set of coupled partial
differential equations is to be transformed into an uncoupled set of scalar dif-
ferential equations. In the following derivation, a constant cross-section is as-
sumed for the sake of simplicity; thus the cross-section A vanishes in Eqs. (D.2)
and (D.3).
With the Jacobian matrix J defined as J = ∂F /∂ξ , and since F is homo-
geneous in ξ [49], Eq. (D.2) becomes:
∂ξ ∂ξ
+J =Q (D.4)
∂t ∂z
where for the one-dimensional case, J is given by the Eq. (E16.2.3) in
Hirsch [49]. In order to obtain the eigenvalues of the Euler equations, the
system is written as a function of the primitive variables ρ, U and p instead of
the conservative variables (ρ, ρU and ρE). Similar to ξ, a primitive variable
T
vector ζ is defined as ζ = [ρ, U, p] . Rewriting the Euler equations yields:
∂ζ e ∂ζ = Q
+J e (D.5)
∂t ∂z
The transformation matrix to convert between conservative and primi-
tive variables is M = ∂ξ/∂ζ . M and its inverse M −1 are given by the
Eqs. (E16.2.7) and (E16.2.8) in [49]. The primitive Jacobian J
e is given by
Eq. (E16.2.9) in [49]:
D.1 Model description 269
 
U ρ 0
e = 0
J U 1/ρ  (D.6)
0 ρc2 U
which is a much simpler structure compared to the conservative Jacobian
J (Eq. (E16.2.3) in [49]). Introducing the relation ξ = M ζ, Eq. (D.4) be-
comes:
∂ζ ∂ζ
M + JM =Q (D.7)
∂t ∂z
or after left multiplication with M −1 :
∂ζ ∂ζ
+ M −1 J M = M −1 Q (D.8)
∂t ∂z
e = M −1 J M and Q
Comparing Eqs. (D.8) and (D.4) yields J e = M −1 Q. Now,
J
e can be diagonalized as:
e = RΛR−1
J (D.9)
where Λ is a diagonal matrix with diagonal elements λi , i = 1 . . . 3. Denoting
the columns of R as ri , Eq. (D.9) can be rewritten as:
e ri = λ i r i
J (D.10)
where ri represent the right eigenvectors of Je . Solving the eigenvalue prob-
lem (D.9) or (D.10) yields the following three eigenvalues:
λ1 = U, λ2 = U + c, λ3 = U − c (D.11)
The right and left eigenvector matrices R and R−1 are derived in Sect. 16.4.1
in Hirsch [49]. Inserting Eq. (D.9) into Eq. (D.5) yields, after left multiplication
with R−1 :
∂ζ ∂ζ
R−1 + ΛR−1 = R−1 Q
e (D.12)
∂t ∂z
or by defining the characteristic variable vector ψ as ψ = R−1 ζ:
∂ψ ∂ψ
+Λ = R−1 Qe (D.13)
∂t ∂z
Equation (D.14) represents the characteristic form of the Euler equations,
which corresponds to a set of three decoupled scalar partial differential equa-
tions:
∂ψi ∂ψi
+ λi = qi0 (D.14)
∂t ∂z
where ψi are the components of ψ and λi are the characteristic propagation or
convection speeds in Eq. (D.11).
D.1.2 Discretization schemes

Since the Euler equations can be transformed into a set of decoupled partial
differential equations (D.14), the equations are hyperbolic. This means that
each characteristic variable ψi represents a wave traveling in positive or negative
lengthwise z-direction, depending on the sign of the corresponding eigenvalue
λi . For a positive eigenvalue λi > 0, the solution ψi in point zj can only
be influenced by values in upwind points zj−1 , zj−2 , . . . and not by downwind
points zj+1 , zj+2 , . . ..
For a cell-centered discretization, the following computation scheme results
(neglecting the source term):
∂ξ j 1
=− F j+1/2 − F j−1/2 (D.15)
∂t ∆z
where the time evolution of the conservative variable ξ j results from integrating
this expression in each node j. For a simple central differencing scheme, the
numerical fluxes F j±1/2 in Eq. (D.15) are defined as:
1
F j+1/2 = (F j + F j+1 ) (D.16)
2
and analogous for F j−1/2 . This scheme is not stable, unless some artificial
dissipation is introduced [49]. The reason is the hyperbolic nature of the Euler
equations as noted above.
First order upwind schemes

First-order upwind schemes that do take these physics into account are dis-
cussed in Chap. 20 of Hirsch [49]. The expression for the numerical flux con-
tains an additional flux difference term ∆F j,j+1 that ensures the physical flow
of information along the characteristics:
1
F j+1/2 = (F j + F j+1 − |∆F j,j+1 |) (D.17)
2
The flux difference term is defined as:

∆F j,j+1 = J j,j+1 ξ j+1 − ξ j
−
= J+

j,j+1 + J j,j+1 ξ j+1 − ξ j (D.18)
and the absolute value of the flux difference term is defined as:

|∆F j,j+1 | = |J j,j+1 | ξ j+1 − ξ j
−
= J+

j,j+1 − J j,j+1 ξ j+1 − ξ j (D.19)
In Eqs. (D.18) and. (D.19), the superscripts + and − indicate the positive
and negative parts of the Jacobian, respectively. These matrices are defined
D.1 Model description 271
based on the positive and negative eigenvalues λi . Using the notation as in

e = M −1 J M yields:
Sect. D.1.1, and combining Eqs. (D.9) and J
J = M RΛR−1 M −1 (D.20)
Based on the above eigenproblem expansion, the positive and negative parts of
J are defined as:
J ± = M RΛ± R−1 M −1 (D.21)

+ −
where Λ and Λ contain only
positive+ and negative eigenvalues, respectively.
Symbolically, Λ± = diag λ±
i , where λ i = max (λ i , 0) and λ −
i = min (λi , 0).
Second order TVD schemes

In the above described upwind scheme, the numerical fluxes (D.17) are defined
based on the values in two points j and j + 1. The upwind scheme is only first
order accurate, i.e. the error is proportional to the mesh size ∆z. To improve
the accuracy of the upwind scheme, the fluxes should be calculated based on
the values in more than two points.
A scheme preserves monotonicity if, as a function of time, (i) no new local
extrema can be created and (ii) the magnitude of the value of a local extremum
is non-increasing. According to Godunov’s theorem, a linear scheme for con-
vection equations can only be monotone if it is first order accurate (Chap. 21
in [49]).
However, by introducing the proper non-linearity into a second order
scheme, monotonicity may be maintained. This is done by non-linear lim-
iter functions. In case the scheme tends to create or amplify a local extremum,
the limiter locally reduces the scheme to first order accuracy, thereby ensur-
ing monotonicity. The similar total variation diminishing (TVD) property is
introduced by Harten.
Two such limiter functions that are used in second order TVD schemes are
the minmod and superbee limiter. The minmod limiter takes two arguments,
and returns the argument with smallest absolute value (if both have the same
sign) and zero otherwise. The superbee limiter is used by Vandevoorde [101]
in a second order TVD scheme, with very good results. The same approach is
used for the gas dynamic model in this thesis. The superbee limiter function
is defined as:

x xy xy
Lim (x, y) = max 0, max min 2 |x| , , min |x| , 2 (D.22)
|x| |x| |x|
Most finite difference schemes are cell-centered, i.e. the nodes form the
center of the control volumes. All variables are computed at the nodes, yet
the flux balances are fulfilled at the volume edges or vertices, in between the
nodes. Vandevoorde [101] examined a number of cell-centered upwind and
TVD schemes, and found that the these do not guarantee the conservation
L
∆z
A
1 2 j j +1 n
Figure D.1 – Cell-vertex discretization
of mass near section changes. Vandevoorde [101, 103] introduced a new cell-
vertex second order TVD scheme, where the nodes coincide with the edges of
the control volumes. Thus, the flux balances are fulfilled in the nodes and the
conservation laws are satisfied in case of section changes. Figure D.1 shows the
cell-vertex discretization nomenclature for the model used in this thesis.
For first order time stepping, the cell-vertex TVD algorithm is as follows:
∆t h +
ξ m+1
j = ξm
j − D j−1/2 ∆F j,j−1
∆z
+ 12 Lim D + ∆F j+1,j , D +
∆F j,j−1
j+1/2 j+1/2

+ +
− 2 Lim D j−1/2 ∆F j,j−1 , D j−1/2 ∆F j−1,j−2
1
+D − j+1/2 ∆F j+1,j

− 2 Lim D −
1 −
j+1/2 ∆F j+1,j , D j+1/2 ∆F j+2,j+1
i
+ 12 Lim D −j−1/2 ∆F j,j−1 , D −
j−1/2 ∆F j+1,j
where j and m are the node and time step index, respectively. In analogy to
Eq. (D.21), the positive and negative decision matrices D ± are defined as:
D ± = M RH ± R−1 M −1 (D.23)
where H + and H − are diagonal matrices defined as H+ =

diag[max(sgn(λi ), 0)] and H − = diag[−min(sgn(λi ), 0)].
D.1.3 Zero-dimensional volumes

Components in the model that feature a small length-to-diameter aspect ratio
are not discretized, and rather simply modeled as zero-dimensional volumes,
where all state variables are assumed constant throughout the volume and
stored in a single point. For instance, the cylinders, diffuser and exit cone are
modeled as zero-dimensional compressible volumes with conservation of mass
and energy. The working fluid (air) is assumed an ideal gas. As such, the
following equations are solved in these volumes:
D.2 Model validation 273
∂m X X
= ṁin − ṁout
∂t
∂ (mcv T ) X X 1 ∂V
= (ṁH)in − (ṁH)out − mrT (D.24)
∂t V ∂t
where the volume V can be time-dependent, as is the case for the cylinders.
No heat transfer is modeled.
D.1.4 Implementation
The model used in this thesis is the second order TVD cell-vertex scheme using
the superbee limiter, as described in Sect. 2.8 of Vandevoorde [101] and in
Vandevoorde et al. [103]. The boundary conditions (e.g. open end, restricted
flow through exhaust valves) are implemented according to [101], with valve
discharge coefficients according to Heywood [47].
The model is implemented as a Simulink11 library block, making it easy to
construct a model for the exhaust system comprising several one-dimensional
pipes, junctions and zero-dimensional compressible volumes. Time integration
is performed using Simulink’s built-in fourth order Runge-Kutta scheme.
Figure D.2 shows a diagram of the Simulink model for one-dimensional flow
in a variable cross-section pipe. In the diagram, ‘U’ represents the conserva-
tive state variable ξ, ‘S’ is the cross-sectional area A. Inside the block ‘Truth
matrices i±1/2’, the local eigenvalues and eigenvectors are determined. For il-
lustration, Fig. D.2c shows the calculation of the right eigenvector product M R
(see Eq. (D.20)). The ‘Lim’ blocks implement the superbee limiter function.
In Sect. D.2, this model is validated using some benchmark problems, sim-
ilar to the approach in Vandevoorde [101]. Section D.3 examines the spatial
and temporal resolution of the model in terms of the node spacing.
D.2 Model validation

D.2.1 Sod’s shock tube
Description
The Sod [93] shock tube or Riemann problem is a generally accepted benchmark
for one-dimensional gas dynamic schemes. The shock tube consists of a straight,
frictionless, constant cross-section pipe with a different initial conditions at
either side of a central diaphragm. At time t = 0, the imaginary diaphragm
is removed, thus creating a compression shock wave, an expansion wave and a
contact discontinuity. The solution can be analytically determined.
The problem is discussed in Sect. 16.6.3 in Hirsch [49]. Here, the values
for the initial pressures and temperatures are taken according to Vandevo-
orde [101], instead of [49]. The right boundary is atmospheric (patm and Tatm )
and the left boundary is an infinite reservoir at 1.5 patm and 1.2 Tatm . The
col row
returns eigenvectors R 1
1 U U(E) R11
BC
4
U U(E) R21
1
2 c u2 7
U U(E) R31
T
Initial Conditions
v
type 1
U_IC v 2
U_BC R12
v+c
5
R22
v-c
State
Derivatives 8
type v
R32
par
U 2
U' 1 u
U_BC
x x dU/dt 1 2 v²/2
U 3
xo s U
d d R13
2
Integrate Boundary u
State Conditions 2
6
U 2 R23
u
Q
0 q 3 2 9
R33
Heat flux Source kappa
to gas (W/m²) 2
2
(a)
(c)
x(i+2) ro
State derivative
x(i+2) v U U(k+1)
D-(i+1/2)
A
S(k+1)
x(i+2) T dF(i+3/2) AX
DF(k+1/2) X Lim
U(k)
S(k)
D-(i-1/2)
A
Truth matrices dF(i-1/2) AX
X 2
x(i+1) ro i+1/2
D-(i-1/2)
A Lim
x(i+1) v U U(k+1) D+
dF(i+1/2) AX
X
x(i+1) T U(k) D- 1
U(k+1)
D-(i+1/2)
A
S(k+1)
dF(i+1/2) AX 1
DF(k+1/2) X
U(k)
S(k)
ro
1 U v -1 em
U 1
T
dU/dt
U(k+1)
D+(i-1/2)
A
S(k+1)
dF(i-1/2) AX 1
DF(k+1/2) X
U(k)
S(k)
x(i-1) ro U(k+1) D+
D+(i+1/2) 1
A
x(i-1) v U U(k) D- dF(i-1/2) AX
X Lim
x(i-1) T Truth matrices
i-1/2 D+(i+1/2)
A
dF(i+1/2) AX
X
U(k+1) 2
D+(i-1/2)
x(i-2) ro A Lim
S(k+1)
dF(i-3/2) AX
DF(k+1/2) X
x(i-2) v U U(k)
S(k)
x(i-2) T
x(i+2)
x(i+1)
S
3 u2 pi/4
d x(i-1) 2 dx
x
x(i-2)
(b)
Figure D.2 – One-dimensional gas dynamic pipe model: (a) overview, (b) the state
derivatives according to the second order TVD scheme [101], (c) the right eigenvector
matrix M R (see Eqs. (D.20) and (D.20))
thermodynamic properties are assumed constant (i.e. γ = 1.4). The analytical

time-dependent solution is given in Sect. 16.6.3 in [49] and Sect. 2.2.1 in [101].
Validation
Figure D.3 gives the results for the shock tube, for two discretization schemes.
Case (a) is the scheme used in this thesis. Both cases use the same number
of nodes, fourth order Runge-Kutta integration and a CFL number ' 0.6 (see
the note on the Courant-Friedrichs-Lewy number on p. 191).
Each series of four plots show the following dimensionless quantities as a
function of the dimensionless lengthwise coordinate z/L (from left to right and
top to bottom): the pressure p/patm , the velocity U /catm , the speed of sound
c/catm and the density ρ/ρatm . The solid line ( ) is the analytical solution
and the markers ( ) are the numerical solution.
The deviation between the numerical and analytical solution for the shock
tube problem is used in Sect. D.3.1 to estimate the spatial resolution of four
different schemes.
The cell-vertex TVD scheme features an excellent shock capturing perfor-
mance, which may be noted in Fig. D.3a. Furthermore, Vandevoorde et al.
[102] prove this scheme to be the best solution for modeling the compressible
gas dynamics in intake and exhaust pipes of internal combustion engines.
In this thesis, the gas dynamic model is also used to predict the frequency
response of the exhaust manifold. This is relevant with regard to the Helmholtz
resonances discussed in Sect. 5.3. Therefore, the performance of the model
in terms of its temporal or frequency characteristics is determined using the
following two benchmark problems. Sections D.2.2 and D.2.3 compare the
numerical and analytical frequency response of a straight pipe of constant cross-
section and a single expansion chamber muffler.
D.2.2 Frequency response of a pipe

Description
The considered test pipe is frictionless, of constant cross-section, L = 1 m long,
with open end boundary conditions on either side. The right side boundary is
atmospheric (patm = 1 atm, Tatm = 273.15 K). At the left side, a fluctuating
pressure patm + p is applied, where p represents the acoustic pressure, with zero
mean.
The transfer function of the pipe is defined here as the dimensionless ad-
mittance Y /Y0 at the inlet side. The admittance Y is the inverse of the
impedance Z, defined36 as Z = p/U [Pa/(m/s)]. The characteristic impedance
Z0 [Pa/(m/s)] is defined as the ratio of sound pressure p to the particle veloc-
ity U for a plane wave traveling in open field. This corresponds to Z0 = ρc. The
36 The acoustic impedance defined here as Z = p/U [Pa/(m/s)] is actually the specific
acoustic impedance. The impedance as such is defined in the acoustic literature as ZA =

Z/A [Pa/(m3 /s)], where A [m2 ] is the considered cross-sectional area. [20]
2nd order cell−vertex TVD (superbee), 64 nodes, CFL = 0.59

Theoretical 0.15
Numerical
1.4
U / c0 (−)
p / p0 (−)
0.1
1.2
0.05
1 0
0 0.5 1 0 0.5 1
1.1
1.2
ρ / ρ0 (−)
c / c0 (−)
1.05
1.1
1 1
0 0.5 1 0 0.5 1
z / L (−) z / L (−)
(a)
1st order cell−centered upwind, 64 nodes, CFL = 0.59

Theoretical 0.15
Numerical
1.4
U / c0 (−)
p / p0 (−)
0.1
1.2
0.05
1 0
0 0.5 1 0 0.5 1
1.1
1.2
ρ / ρ0 (−)
c / c0 (−)
1.05
1.1
1 1
0 0.5 1 0 0.5 1
z / L (−) z / L (−)
(b)
Figure D.3 – Sod’s shock tube benchmark problem, using (a) the second order TVD
cell-vertex scheme [101] and (b) a first order upwind cell-centered scheme
dimensionless admittance is therefore Y /Y0 = ρcUi /pi , where the subscript i

denotes the inlet conditions [20].
Analytical
For plane waves37 traveling in a straight pipe of constant cross-section (see
Chap. 2 in Beranek [20]), the solution for the particle displacement ξ of the
one-dimensional wave equation can be written in the following form:
ξ (z, t) = ejωt (A cos kz + jB sin kz) (D.25)

√
where j = −1, k = ω/c is the wave number [m−1 ] and A and B are unknown
coefficients. The velocity U and sound pressure p are related to ξ as U = ∂ξ/∂t
and p = −ρc2 ∂ξ/∂z . Consequently:
U (z, t) = jωejωt (A cos kz + jB sin kz) (D.26)

p (z, t) = −ρc2 ejωt k (−A sin kz + jB cos kz) (D.27)
The transfer function Y /Y0 can be obtained as ρcUz=0 /pz=0 . Filling in
the above expressions for U and p and evaluating at z = 0 yields Y /Y0 =
−A/B . The coefficients A and B are determined by introducing the boundary
conditions (i.e. p = 0 at z = 0 and z = L), yielding A/B = j cos kL/sin kL , or:
Y ρcUi j j
= =− =− (D.28)
Y0 pi tan kL tan (2πf L/c )
From this theoretical admittance (D.28), the standing wave resonance fre-
quencies are tan kL = 0, or f = n · c/(2L) , where n ∈ Z. At low frequency
f c/(2L) , Y ∼ −j/f . Thus, the admittance is unbounded and the velocity
lags the pressure by 90 ◦ .
Validation
The open end boundary conditions are applied to a numerical model of the pipe.
The inlet pressure pi is perturbed using a multisine signal (see Sect. D.4.3). The
transfer function is determined using the approach described in Sect. D.4.
Figure D.4 presents two Bode diagrams of the admittance Y /Y0 = ρcUi /pi ,
for a different number of nodes, yet both using the second order TVD cell-
vertex scheme. The solid line ( ) corresponds to the analytical solution in
Eq. (D.28). The markers ( ) result from the numerical solution, using the
transfer function estimation method described in Sect. D.4.
As expected, Fig. D.4 shows that the spectral resolution increases as the
number of nodes increases. Based on this benchmark problem, Sect. D.3.2
discusses the frequency (or temporal) resolution.
37 The assumption of planar acoustic waves poses some limitations to the pipe diameter
d. (i) d should be small enough to avoid transverse resonances, or d < λ/1.2 [33], where
λ = c/(2πf ) is the wavelength√[m]. (ii) d should be large enough to avoid viscous distortion
of the wavefronts, or d > 0.1/ f [20]. These conditions are fulfilled in exhaust systems for
the frequency range of interest (10 < f < 1000 Hz), except inside the catalyst substrate.
FRF, L = 1 m, c = 343.1 m/s, 16 nodes, CFL = 0.466 FRF, L = 1 m, c = 343.1 m/s, 64 nodes, CFL = 0.490
From: Inlet pressure p, To: Inlet velocity ρcU From: Inlet pressure p, To: Inlet velocity ρcU
40 40

Theoretical Theoretical
Numerical Numerical
20 20
0 0
−20 2 3
−20 2 3
10 10 10 10
2f L/c (−) 1/2 1 3/2 2 3 4 6 8 10 2f L/c (−) 1/2 1 3/2 2 3 4 6 8 10
180 180
90 90
Phase (°)
Phase (°)
0 0
−90 −90
−180 −180
2 3 2 3
10 10 10 10
(a) (b)
Figure D.4 – Frequency response function of a pipe, between inlet pressure and
inlet velocity, using the second order TVD cell-vertex scheme [101], with (a) 16 nodes
and (b) 64 nodes
D.2.3 Frequency response of a muffler

Description
The 1954 NACA Technical Report by Davis et al. [33] provides a theoretical and
experimental approach to determine the attenuation of exhaust mufflers. The
authors provide detailed instructions for an appropriate experimental setup,
which may be used to determine the acoustical characteristics of mufflers with-
out net flow. The appendices in [33] give the theoretical derivation of the
attenuation for various muffler geometries.
The single expansion chamber with infinite connecting pipes is the simplest
type of muffler. Davis et al. [33] define Muffler 2 as a single expansion chamber
with an area ratio m = 16, a chamber length L = 24 inch (' 610 mm) and
connecting pipe diameter of d = 3 inch (' 76 mm). This reference muffler
(see Fig. D.5) is chosen to validate the frequency characteristics of the one-
dimensional gas dynamics code for a case of varying cross-sectional area.
The attenuation of an acoustic element placed in a pipe is defined in deci-
bel [dB] as:

Average incident sound power
Attenuation = 10 log10 (D.29)
Average transmitted sound power
If the attenuation is considered between two points of equal cross-section, the
Attenuation of expansion chamber muffler

Ref: Davis e.a. (1954) NACA Report 1192, Muffler 2
50
Amplitude (dB) of p2i /p2o

Theoretical
40 Numerical
30
20
10
0
0 200 400 600 0
Frequency f (Hz)
Figure D.6 – Attenuation of the

reference muffler in Fig. D.5, us-
ing the second order TVD cell-vertex
Figure D.5 – Reference single ex-
scheme [101]
pansion chamber muffler (Muffler 2
with m = 16 and L = 24 inch)
(Source: [33])
2
attenuation can be determined as 10 log10 |pi /po | , where pi and po are the
acoustic pressures at the inlet and outlet. The attenuation of an acoustic
element is also referred to as the transmission loss (Boonen [21] Chap. 3)
Analytical
Appendix A in [33] describes the analytical derivation of the attenuation for
the single expansion chamber muffler, with an infinite tailpipe. The resulting
expression is given by Eq. (A10):
" 2 #
1 1 2
Attenuation = 10 log10 1+ m− sin kL (D.30)
4 m
where m is the area expansion ratio, k is the wave number and L is the chamber
length. Eq. (D.30) indicates that the attenuation is zero for incident frequencies
equal to the standing wave resonance frequencies kL = nπ or f = n · c/(2L) ,
where n ∈ Z.
Validation
Figure D.6 shows the satisfactory agreement between the numerical result using
the gas dynamic model ( ) and the theoretical attenuation ( ) according to
Eq. (D.30). The deviation that exists between numerical and analytical solution
is likely due to the insufficiency of the non-reflecting boundary conditions used
in determining the numerical result. Further improvements are possible yet not
within the scope of this thesis.
Sod’s shock tube (CFL = 0.5)

Spatial resolution
−1
10
Spatial resolution δz (−)

−2
10
2nd order cell−vertex TVD (superbee)

2nd order cell−vertex TVD (minmod)
2nd order cell−centered TVD (superbee)
1st order cell−centered upwind
−3
10 1 2 3
10 10 10
Number of nodes (−)
Figure D.7 – Spatial resolution based on Sod’s shock tube problem, for different
discretization schemes
D.3 Resolution
D.3.1 Spatial resolution
The spatial resolution of the discretization scheme is defined here based on the
difference between the numerical and analytical solution to Sod’s shock tube
problem introduced in Sect. D.2.1.
More specifically, a spatial difference measure ∆z [m] is defined as the root
mean square lengthwise deviation between the numerical and analytical curves
for the speed of sound c/catm , as shown in Fig. D.3. The dimensionless spatial
difference δz = ∆z /L is plotted in Fig. D.7.
Figure D.7 shows that the cell-vertex TVD scheme with superbee limiter
outperforms all other schemes. For this scheme, the spatial resolution δz is
roughly inversely proportional to the number of nodes.
D.3.2 Temporal resolution

The temporal or frequency resolution of the discretization scheme is defined
here based on the difference between the numerical and analytical solution to
the admittance of a pipe with two open ends, which is the benchmark problem
used in Sect. D.2.2.
The maximum resolved frequency is defined rather arbitrarily as the max-
imum frequency where the magnitude of the admittance 20 log10 (|Y /Y0 |) ex-
ceeds 16 dB. The maximum frequency is plotted in Fig. D.8a versus the number
of nodes. The magnitude resolution δampl [dB] is defined as the root-mean-
square deviation between the magnitude of the theoretical and numerical ad-
mittance. A similar definition is used for the phase resolution δphase ( ◦ ) (see
Fig. D.8b).
D.4 Identification 281
FRF (CFL = 0.5), bandwidth FRF (CFL = 0.5), frequency resolution

2nd order cell−vertex TVD (superbee) 2nd order cell−vertex TVD (superbee)
2
15 10
10
Maximum frequency 2fL/c (−)
8
6
Frequency resolution
4 1
10
3
1
0
10 δampl (dB)
max(2fL/c) (−) δphase (°)
0.5 1 2 3
10 10 10 0 50 100
Number of nodes (−) Number of nodes (−)
(a) (b)
Figure D.8 – Frequency resolution based on the admittance of a pipe with two open
ends, for the second order TVD cell-vertex scheme
Based on Fig. D.8a, the minimum number of nodes can be determined to

resolve a given frequency. The maximum frequency fmax 2L/c roughly equals
1/8 of the number of nodes. As such, fmax ' (c/∆z )/16 . Typically, the
resolution is expressed in terms of the number of nodes per required wavelength
nλ [-]. Introducing ∆z = λmin /nλ in the above expression yields:
c
fmax =
nλ ∆z
L
⇔ n = nλ (D.31)
λmin
where the number of nodes per wavelength nλ ' 16. For instance, if c =
340 m/s and ∆z = 10 mm, fmax ' 2000 Hz.
D.4 Identification
D.4.1 Purpose
The purpose of identification is to retrieve the frequency response function or
transfer function of a particular dynamic system. The transfer function F (f )
refers to the ratio F = Y /U of an output Y (f ) to an input signal U (f ), both
transformed into the Fourier domain.
Identification is usually applied to ‘black box’ systems or complex multi-
dimensional systems, where no analytical derivation can be used to determine
the transfer function.
In the framework of this thesis, identification is used to determine the
frequency response function of the close-coupled catalyst exhaust manifold
in Sect. 5.3.4. In this appendix, identification is used in the same way in

Sects. D.2.2) and D.2.3).
D.4.2 Approach
The system under investigation is excited using an input signal u (t). The input
and resulting output y (t) signals are transformed to the frequency domain,
resulting in U (f ) and Y (f ). Dividing these yields the transfer function F (f ) =
Y /U . The transfer function is usually represented as a Bode diagram, plotting
the magnitude and phase of F as 10 log10 (|F |) [dB] and ∠ (F ) [◦ ]. The Bode
diagram provides only a linearized view of the true system. Gas dynamics in
particular is a non-linear phenomenon, especially for high Mach number flows,
yet also at low M a for restricted flows.
The frequency response function F (f ) is estimated using the MATLAB11
function TFE, which is part of the System Identification Toolbox. The fre-
quency response function is estimated as the quotient of the cross spectrum of
input and output signals Puy (f ) and the power spectrum of the input signal
Puu (f ).
D.4.3 Multisine signals

A multisine is a commonly used input signal for the identification of dynamic
systems. The multisine u is a sum of a i = 1 . . . n sine functions with frequencies
fi and phases φi :
n
X
u (t) = sin (2πfi t − φi ) (D.32)
i=1
The frequencies fi are chosen between a minimum and maximum frequency

fmin and fmax . The phase angles can be set to zero, or randomized. Both
approaches result in a multisine signal with (i) an unconfined amplitude in the
time domain and (ii) a spiky frequency spectrum.
A better approach is to chose the phase angles so that the amplitude remains
bounded and the frequency spectrum is as smooth as possible between fmin
and fmax . This type of signal enables a better identification, since the input
signal spectrum U (f ) is constant in the frequency range of interest, i.e. each
frequency is excited with the same amplitude.
Guillaume et al. [41] introduced the crest factor-minimization algorithm
which determines the optimal phase angles φi to obtain a smooth spectrum.
This algorithm is applied for generating the multisines used in this thesis.
Figure D.9 provides an example of (a) a multisine signal and (b) its Fourier
spectrum. This particular signal is generated at a sample frequency of 1 kHz,
containing frequencies between fmin = 10 Hz and fmax = 100 Hz.
D.4 Identification 283
Multisine signal Multisine signal, Fourier spectrum

Using crest−factor minimization by Guillaume e.a. (1991) f = 1 kHz, ∆t = 1 s, f = 10 Hz, f = 0.1 kHz
s min max
2 0
Amplitude (dB)
1 −50
0 −100
−1 −150
−2 −200 0 1 2
0 0.2 0.4 0.6 0.8 1 10 10 10
2
180
1 90
Phase (°)
0 0
−1 −90
−180
−2 0 1 2
0 0.05 0.1 0.15 0.2 10 10 10
Time t (s) Frequency f (Hz)
(a) (b)
Figure D.9 – Example multisine signal (a) and its Fourier spectrum (b), generated
using the crest factor-minimization algorithm by Guillaume et al. [41]
Nederlandse samenvatting
Experimentele stromingsdynamica
in uitlaatsystemen van voertuigen
met voorkatalysator
285
Experimentele stromingsdynamica
in uitlaatsystemen van voertuigen
met voorkatalysator
1 Inleiding: Uitlaatsystemen
1.1 Achtergrond
Sinds de jaren 1970 gelden er wettelijke bepalingen die de uitstoot van scha-
delijke stoffen bij voertuigen met inwendige verbrandingsmotoren beperken
(Tabel 1.1). De voornaamste schadelijke componenten zijn koolstofmonoxide
(CO), stikstofoxiden (NOx ) en koolwaterstoffen (Cx Hy ) voor benzinemotoren
en NOx en roetdeeltjes (PM) voor dieselmotoren. De steeds strengere emissie-
wetgeving resulteren in talloze technologische verbeteringen aan de constructie
en het management van de verbrandingsmotor. De katalysator werd geïn-
troduceerd in het uitlaatsysteem om de gevormde polluenten om te zetten in
onschadelijke stoffen zoals CO2 en water. De traditionele katalysator is gemon-
teerd in de uitlaatleiding, zodat een uniforme stromingsverdeling eenvoudig te
realiseren is.
De Euro III norm zorgde voor een kentering, door de introductie van een
koude start fase in de homologatiecyclus. Aangezien de katalytische reactie-
snelheid slechts significant is boven een temperatuur van 250 tot 400 ◦ C, is
de opwarmtijd sindsdien een cruciale ontwerpparameter. Een snelle opwar-
ming wordt bekomen door de katalysator vlakbij de motor te plaatsen, als
geïntegreerd deel van de uitlaatcollector. Dit type wordt close-coupled 38 of
voorkatalysator genoemd.
De thesis onderzoekt de stroming in uitlaatcollectoren met voorkatalysator,
en draagt bij tot de state-of-the-art in dit gebied van toegepaste stromingsme-
38 Cursief gedrukte woorden zijn eigen aan het Engelstalige vakjargon. Om ondubbelzin-
nigheden te vermijden worden ze vaak in plaats van hun Nederlandse vertaling gebruikt.
287
288 Nederlandse samenvatting
manifold
(collector)
runners
(headers)
diffuser
driving
direction
close-coupled
catalyst
exit cone underbody

muffler tailpipe
catalyst
downpipe
hot end cold end
Figuur 1.1 – Benamingen in het uitlaatsysteem
chanica. Het onderzoek is relevant voor katalysatoren bij benzine- en diesel-

motoren, alsook roetfilters en NOx -filters.
1.2 Ontwerpaspecten
Een modern uitlaatsysteem ontwerpen is een hoogtechnologische opdracht, die
geavanceerde kennis vergt van transiënte stroming en warmteoverdracht in een
complexe geometrie. Het uitlaatsysteem bestaat uit het hot end en cold end.
Het hot end omvat de uitlaatcollector en voorkatalysator. Het cold end omvat
de geluiddemper(s) en eventuele hoofdkatalysator. De collector bestaat uit
collectorbuizen of runners die samenkomen in de diffusor voor de katalysator
(zie Fig. 1.1).
Pas wanneer de katalysatortemperatuur 250 tot 400 ◦ C overschrijdt gaan
de reacties op. Door de katalysator dicht bij de motor te plaatsen, warmt deze
sneller op (ca. 10 s) dan de traditionele katalysator in het cold end (ca. 2 min).
Dat veroorzaakt een sterke emissiereductie bij koude start. De motornabije
plaatsing, samen met de beperkte ruimte in het motorcompartiment, resulteert
in korte runners en complexe scherpe bochten.
De levensduur hangt af van (i) de thermische belasting en (ii) katalysator-
degradatie. De stroming veroorzaakt een verdeling van warmteoverdrachtsco-
ëfficiënt op de binnenwand. Hot spots zijn zones met hoge wandtemperatuur,
ten gevolge een hoge warmteoverdracht. Differentiële thermische uitzetting en
opeenvolgende thermische cycli induceren spanningen in de wanden, die de kans
op faling (vb. van lasnaden) vergroten.
Door deactivatie of degradatie van een katalysator vermindert de reactie-
snelheid. Deactivatie is onvermijdelijk maar kan beperkt worden door een
degelijk collectorontwerp. Een katalysator voor verbrandingsmotoren bevat
actieve Pt, Pd en Rh partikels, fijn verdeeld in de poreuze washcoat. De was-
hcoat wordt aangebracht op een draagstructuur. Dit is meestal een cordieriet
(2MgO · 2Al2 O3 · 5SiO2 ) matrix [56], bestaande uit parallelle kanaaltjes met
diameter d = 0.8 tot 1.1 mm en wanddikte t = 0.06 tot 0.17 mm (Tabel 1.2).
1 Inleiding: Uitlaatsystemen 289
Bartholomew [11] en Forzatti en Lietti [38] bespreken deactivatie ten gevolge

van chemische, thermische, fysische en mechanische processen. Het overzicht
van Neyestanaki et al. [80] is toegepast op katalysatoren voor verbrandingsmo-
toren.
Chemische deactivatie is het gevolg van chemisorptie van ‘giffen’ op ka-
talytische oppervlakken. Bij katalysatoren voor verbrandingsmotoren zijn dit
contaminanten in motorolie en brandstof (vb. P, Pb, Zn, S). Chemische deacti-
vatie is proportioneel met de hoeveelheid verwerkt uitlaatgas. Katalysatorzones
met een hogere snelheid zijn meer onderhevig aan chemische degradatie.
Thermische deactivatie is het gevolg van verschillende processen [11, 38],
voornamelijk het verminderen van contactoppervlak door sintering. De sin-
tersnelheid verloopt exponentieel met de temperatuur (Vgl. (1.1)). Gezien de
katalysatortemperatuur kan oplopen tot meer dan 1000 ◦ C, betekent een be-
perkte daling van de temperatuur een aanzienlijke verbetering van de levens-
duur. Omdat de katalytische reacties exotherm zijn, is het verwerkte uitlaatgas
warmer dan het onverwerkte. Bovendien blijkt lokale periodieke terugstroming
op te treden in katalysatoren [70, 81, 59, 84, 85], ten gevolge van resonantie-
effecten (Hoofdstuk 5), loshechting en recirculatie in de collector. Periodieke
terugstroming van heet uitlaatgas vormt een extra thermische belasting voor
de katalysator.
Een uniforme stromingsverdeling in de katalysator is vereist om lokale che-
mische en thermische deactivatie te voorkomen. Enerzijds verwerken hoge snel-
heidszones meer gifstoffen en zijn dus meer onderhevig aan chemische deacti-
vatie. Anderzijds is de temperatuur hoger door de reactiewarmte. Bovendien
garandeert een uniforme stromingsverdeling een minimale drukval en optimaal
conversierendement van de katalysator (Fig. 1.5).
De ontwerpeisen zijn tegenstrijdig: een langere katalysator betekent een
hogere drukval maar verbetert de stromingsuniformiteit (door de hogere tegen-
druk) en het conversierendement (door de uniforme reactorverblijftijd). An-
derzijds stijgt de kostprijs van de katalysator met het volume. Tabel 1.3 geeft
een overzicht van ontwerpaspecten (kostprijs, drukval, conversie-efficiëntie en
stromingsuniformiteit) en de relatieve invloed van de belangrijkste parameters.
Volgens Tabel 1.3 is het ontwerp optimaal voor een maximale porositeit ε en
minimale kanaaldiameter d. Deze vaststelling komt volgens Wiehl en Vogt [112]
overeen met de waargenomen evolutie.
1.3 Literatuuroverzicht
Historische studies [51, 64, 109, 110, 60] (1970 tot 2000) onderzochten de druk-
val en snelheidsverdeling in uitlaatsystemen in stationaire condities. Dit is een
aanvaardbare vereenvoudiging voor een traditionele katalysator, typisch 1 tot
2 m van de motor verwijderd. Met de introductie van de motornabije kataly-
sator is de focus van het onderzoek verschoven naar pulserende stroming (1995
tot heden).
Verschillende auteurs (e.g. Benjamin et al. [13], Voeltz et al. [104], Breu-
er et al. [23], Nagel en Diringer [79]) bespreken snelheidsmetingen en CFD
berekeningen in uitlaatcollectoren met voorkatalysator in stationaire stroming.

Deze studies beogen vooral een maximalisatie van de uniformiteit met minimale
drukval.
De meest interessante studies zijn uitgevoerd in pulserende stroming, met
(i) een isotherme pulserende stromingsopstelling [53, 22, 17, 69, 16, 15, 43, 18,
58], of (ii) een isochore motoropstelling, die ofwel aangedreven is (i.e. zonder
verbranding) [7, 86, 84, 85] ofwel werkend (i.e. met verbranding) [81, 59, 4, 70].
Met behulp van een isotherme opstelling is in Persoons et al. [88] de geldig-
heid van het additieprincipe onderzocht (Hoofdstuk 4):
Additieprincipe De tijdsgemiddelde katalysatorsnelheidsverde-

ling in pulserende stroming komt overeen met een lineaire combina-
tie van snelheidsverdelingen bekomen in stationaire stroming door
elk van de uitlaatrunners, bij gelijk volumetrisch debiet (Vgl. (4.1)).
De geldigheid van deze aanname impliceert dat tijdsintensieve transiënte CFD

berekeningen overbodig zijn om een collector te ontwerpen naar stromingsuni-
formiteit en dat stationaire CFD berekeningen volstaan. De bevindingen van
Persoons et al. [88] worden deels bevestigd door Benjamin et al. [17] en Bressler
et al. [22] (Hoofdstuk 4).
Omwille van de inherente variabiliteit in de geometrie van uitlaatsystemen,
is de relevantie van een aantal studies beperkt [60, 104, 23, 79, 8] tot parti-
culiere systemen. Dit staat in contrast met stromingsonderzoek in verbran-
dingskamers, met min of meer identieke vorm. Ondanks de geometrische va-
riabiliteit en de complexiteit van de stroming, kunnen verschillende aspecten
gerelateerd worden aan gepaste dimensieloze getallen, zoals het scavenging ge-
tal S (Vgl. (1.4)) [88, 86] of vergelijkbare getallen [22, 17, 99]. Zo wordt de
invloed nagegaan van stromingsrandvoorwaarden en geometrie op bijvoorbeeld
de geldigheid van het additieprincipe (Vgl. (4.1)) [88, 86], het conversierende-
ment [99] en de drukval [109, 110].
Uit deze thesis resulteerden vier publicaties in internationale tijdschriften
met review [88, 86, 83, 84] en twee bijdragen op internationale conferenties met
review [87, 85].
1.4 Doelstellingen van de thesis

De doelstelling is de experimentele studie van pulserende stroming in moder-
ne compacte uitlaatsystemen met voorkatalysator. Aangezien heet corrosief
uitlaatgas snelheidsmetingen sterk bemoeilijkt, maakt de thesis gebruik van
opstellingen die koude pulserende stroming genereren, waarbij de stromings-
gelijkvormigheid met werkelijke motorcondities onderzocht wordt. De koude
pulserende stroming maakt het gebruik van snelheidsmeettechnieken mogelijk
met hoge resolutie in ruimte en tijd. De nauwkeurigheid en resolutie van de be-
komen data moeten validatie van computational fluid dynamics berekeningen
mogelijk maken.
De thesis richt zich op het meest relevante stromingsgerelateerde aspect in
het ontwerp van moderne uitlaatcollectoren, namelijk de snelheidsverdeling in
2 Experimentele aanpak 291
de katalysator. Een uniforme snelheidsverdeling is cruciaal voor een optimaal

collectorontwerp qua minimale drukval, maximaal conversierendement en het
vermijden van lokale katalysatordegradatie. Meerbepaald onderzoekt de thesis
de invloed van (i) transiënte stromingsrandvoorwaarden en (ii) geometrische
aspecten (vb. diffusor volume, aantal en lengte van de runners, uitlaatklepti-
ming) op de tijdsafhankelijke snelheidsverdeling in de katalysator.
Wanneer de experimentele technieken ontoereikend zijn om de waargenomen
fysische verschijnselen te verklaren, wordt een ééndimensionaal gas dynamisch
numeriek model van het uitlaatsysteem gebruikt. Het numeriek model wordt
tevens gebruikt om de stromingsdynamica te voorspellen buiten de mogelijk-
heden van de experimentele aanpak.
Volgende aspecten vallen buiten de reikwijdte van de thesis:
• Computational fluid dynamics — De bekomen snelheidsgegevens zijn

deels gebruikt voor de validatie van CFD simulaties, in het kader van
onderzoek in samenwerking met een industriële partner [2]. De CFD be-
rekeningen zijn uitgevoerd door de industriële partner. De thesis richt
zicht op de experimentele methodologie.
• Katalytische reactiekinetica — De reactiekinetica in de katalysator in
werkende motorcondities wordt verwaarloosd. De kwalitatieve invloed
van stromingscondities op het conversierendement kan geschat worden
aan de hand van een vereenvoudigd homogeen of heterogeen model van
de kinetica, zoals beschreven in App. A.3.
• Andere aspecten, zoals warmteoverdracht, akoestiek en mechanische tril-
lingen.
2 Experimentele aanpak
2.1 Experimentele opstellingen
Tijdens de thesis is gebruik gemaakt van twee types opstellingen, die beide
een pulserende luchtstroming genereren in het uitlaatsysteem bij omgevings-
temperatuur: een (i) isotherme opstelling en een (ii) isochore of opgeladen
aangedreven motor (CME) opstelling.
Bij de isotherme opstelling (Sect. 2.2.1) is de uitlaatcollector via een pul-
sator gemonteerd op een buffervat (Fig. 2.2). Een roterende klep (Fign. 2.3
en 2.4) en de originele cilinderkop (Fign. 2.5 en 2.6) zijn gebruikt als pulsator.
Het debiet wordt gemeten met een ISO-genormeerde meetflens (Sect. 2.4.1).
Debiet en pulsatiefrequentie worden ingesteld overeenkomstig motorbelasting
en toerental. De stroming is onderzocht in twee types uitlaatcollectoren: met
en zonder uitlaatklepoverlap (Tabel 2.1 en Fig. 2.1).
Omwille van de eenvoud wordt de isotherme opstelling gebruikt in verschil-
lende studies [88, 58, 18, 69, 17, 16, 15, 43, 22]. De stroming verschilt echter
sterk van een werkende motor (Fig. 2.9). De complexe geometrie van uit-
laatcollector en cilinderkop maakt enkel een opstelling met schaalfactor 1 : 1
mogelijk. Om de snelheidsmetingen te vergemakkelijken is gekozen voor een

opstelling met koude stroming. Omwille van het temperatuurverschil is het
slechts mogelijk het tijdsgemiddelde Reynolds- en Machgetal bij benadering in
overeenstemming te brengen met de werkelijke motorcondities.
De isotherme opstelling genereert een ééntrapse ‘uitlaatslag’. Figuur 2.10
toont het verschil met werkelijke motorcondities, met tweetrapse uitlaatslag:
(i) blowdown en (ii) uitdrijvingsfase. Het runnerdebiet vertoont een hogere
frequentie-inhoud bij de werkende motor. Bijgevolg verschilt de schijnbare pul-
satieperiode (Vgl. (4.46)), waardoor de bevindingen op basis van de isotherme
opstelling [88] niet kunnen geëxtrapoleerd worden naar werkelijke motorcondi-
ties.
De isochore CME opstelling genereert koude pulserende stroming gelijkaar-
dig aan werkelijke motorcondities. CME staat voor charged motored engine,
of opgeladen aangedreven motor (Fig. 2.7). De verbrandingsmotor wordt aan-
gedreven met een elektromotor. Op een verstevigde inlaatcollector wordt pers-
lucht met instelbare druk aangelegd. Een laminaire stromingsmeter dient als
referentie debietmeting. Er is geen brandstofinjectie of ontsteking. De inlaat-
druk komt overeen met een bepaalde residuele cilinderdruk, vlak voordat de
uitlaatklep opent. Bij normale motorwerking varieert deze druk tussen 2 en
5 atm. De CME opstelling bekomt een gelijkaardige stroming met een inlaat-
druk tussen 1 en 2.5 atm. De uitlaatslag bij de CME opstelling is gekenmerkt
door een blowdown en uitdrijvingsfase (Fig. 2.10).
2.2 Stromingsgelijkvormigheid
Sectie 2.3 bespreekt de stromingsgelijkvormigheid tussen CME en werkende
motor aan de hand van o.a. een analytische afleiding van het massadebiet
tijdens blowdown en uitdrijvingsfase. In Vgln. (2.1) en (2.2) stellen bevat term i
de invloed van de inlaatdruk en term ii de invloed van de warmtevrijstelling
tijdens de verbranding. Bij afwezigheid van verbranding (i.e. CME) is term ii
= 1. Ter compensatie moet de inlaatdruk (term i) aangepast worden.
Figuur 2.11 toont deze debieten voor werkende motor met verbranding
(links) en CME (rechts) in functie van de motorbelasting (i.e. motorkoppel),
direct gerelateerd aan de inlaatdruk. Bij werkende motor is het maximum van
ṁ1 /ṁ2 ' 6, en slechts 2.5 bij de CME opstelling. Ondanks de onvolledige
gelijkvormigheid, genereert de CME opstelling een tweetrapse uitlaatslag zoals
een werkende motor.
2.3 Datareductie
Het identificeren van periodieke fenomenen vereist een conditionele (of phase-
locked ) meettechniek. De fase (of tijdsbasis) van phase-locked signalen komt
overeen met een vaste referentie, vb. een referentiepuls die eens per perio-
de optreedt. Een werkelijk signaal opgemeten in een periodieke stroming be-
staat uit een cyclusafhankelijke (cycle-resolved ) en onafhankelijke component
3 Oscillerende hittedraad anemometer (OHW) 293
(Vgl. (2.19)) ten gevolge van turbulentie, ruis of onstabiele stromingsverschijn-

selen. Bruun [25] bespreekt de literatuur omtrent phase-locked meettechnieken.
Het bepalen van het ensemble gemiddelde (Vgl. (2.20)) verbetert de nauw-
keurigheid op de cyclusafhankelijke component. De stroming in een volumetri-
sche machine varieert van de ene cyclus tot de andere. Omwille van cyclische
variatie zijn er een groot aantal ensembles nodig om de fout op het cyclusaf-
hankelijke gemiddelde klein genoeg te maken (Vgl. (2.25)).
De meting van turbulentiegrootheden bij cyclische variaties vereist een cy-
clusafhankelijke (of cycle-resolved ) analyse. Deze aanpak splitst de onafhanke-
lijke component op in een bijdrage ten gevolge van cyclische variatie en turbu-
lentie (Sect. 2.5.2).
3 Oscillerende hittedraad anemometer (OHW)

3.1 Inleiding
Het meten van bidirectionele fluïdumsnelheid is geen eenvoudige opdracht.
Nochtans treedt periodieke terugstroming op in de voorkatalysator [70, 81, 59].
Appendix C.2 toont aan dat optische technieken nauwkeurige snelheidsmetin-
gen kunnen opleveren, zelfs in werkende motorcondities. Helaas gaat deze tech-
niek gepaard met aanzienlijke problemen van optische toegang en verdeling van
seeding.
Hittedraad anemometrie (HWA) levert een hoge resolutie in ruimte en tijd,
maar de ongevoeligheid aan stromingsrichting is een inherent nadeel. Sectie 3.1
bespreekt enkele technieken om toch bidirectionele snelheden te meten met
HWA. Elk heeft voor- en nadelen (vb. bandbreedte, verstoring van de stroming,
afmetingen, omslachtigheid).
Een nieuwe laagfrequente oscillerende hittedraad anemometer (OHW) werd
ontwikkeld tijdens deze thesis, in analogie met recente hoogfrequente OHW sys-
temen [78, 66]. De OHW is gebaseerd op een mechanische oscillator (Fig. 3.5).
In tegenstelling tot hoogfrequente OHW systemen is de ogenblikkelijke probe-
snelheid gekend (Vgl. (3.8)). Anderzijds is de oscillatiefrequentie een order-
grootte kleiner (i.e. 50 Hz, tegenover 500 Hz of meer).
3.2 Methodologie
De richting van de stromingssnelheid U ten opzichte van de probe is aangeduid
in Fign. 3.5 en 3.4. De probe beweegt met een snelheid Up (Vgl. (3.8)). Voor
een niet-ambigue meting moet de relatieve snelheid ten opzichte van de probe
Urel = U − Up positief blijven. Wanneer de snelheid U ogenblikkelijk negatief
wordt, blijft de relatieve snelheid Urel positief wanneer de probe voldoende snel
tegen de normale richting van de stroming beweegt, of Urel > 0 ⇔ Up < U < 0.
Tijdens elke beweging van de probe is er een ogenblik waar de probesnel-
heid Up maximaal in grootte is en negatief. De tolerantie α op de maximale
probesnelheid bepaalt het interval waarbinnen metingen genomen worden. De
probe probe holder oscillating probe 55P11L, xo = 5.5 mm

3
holder base
10 Hz
2.5 20 Hz
U >0 30 Hz
rigid support
Up > 0 40 Hz
tube 2

Urel > 0
optical
encoder 1.5
1
50 mm
0.5
dual balance
shafts
0
-0.5
motor a = 0.736, b = 0.5 (R2 = 0.949)
-1
-4 -3 -2 -1 0
Reference velocity Uo/(ωoxo) (-)
Figuur 3.12c – Kalibratiecur-

ve voor 55P11L probe met xo =
Figuur 3.5 – Hittedraad oscillator
5.5 mm
absolute snelheid volgt uit U 0 = Up +Urel (Vgl. (3.6)), waarbij Urel overeenkomt
met het anemometersignaal.
3.3 Kalibratie
De OHW is gekalibreerd bij gekende negatieve snelheid met behulp van een
specifiek ontwikkelde windtunnel (Fig. 3.9). De kalibratie is uitgevoerd voor
verschillende oscillatiefrequenties, twee amplitudes (xo = 5.5 en 2.85 mm) en
voor drie probe types (Fig. 3.7 en Tabel 3.2). Een tweedimensionale laser Dop-
pler anemometer (LDA) met Bragg cel frequentieverschuiving dient als referen-
tie snelheidsmeting. De LDA meting is phase-locked met de oscillatorbeweging,
zodat de evolutie van de stroming rond de probe kan onderzocht worden.
Figuur 3.12 toont de kalibratiecurves voor elk van de onderzochte probes en
amplitudes. Figuur 3.13 toont de tijdsafhankelijke resultaten in functie van de
probepositie, telkens voor dezelfde referentiesnelheid. Op basis van Fig. 3.12c
geeft probe 55P11L met rechte verlengde uiteinden de beste resultaten, in com-
binatie met de grootste amplitude (xo = 5.5 mm). De maximaal meetbare
negatieve snelheid bedraagt ongeveer −1 m/s. Deze waarde is vergelijkbaar
met andere systemen (Tabel 3.1).
Een schaalanalyse werd toegepast om de optimale ontwerpparameters (fre-
quentie, amplitude en probe type) te bepalen. Als performantiemaat dient
∆U , de afwijking tussen gemeten en verwachte snelheid (Vgl. (3.10)). De af-
wijking wordt voor elke probe uitgezet versus een dimensieloze grootheid. Uit
Fig. 3.14 volgt dat voor kleine snelheden ∆U /U vrij goed correleert met twee
dimensieloze groepen: (i) enerzijds (ωo xo D/ν ) (D/xo ) voor de rechte probes
en (ii) anderzijds (ωo xo D/ν ) (xo /D ) voor de 90 ◦ probe.
Bij gelijke probesnelheid ωo xo en voor een rechte probe, neemt de afwijking
∆U af bij toenemende amplitude xo of afnemende frequentie ωo . Het omge-
keerde is waar voor de 90 ◦ probe. Het verschillend gedrag is toe te schrijven
4 Additieprincipe 295
aan de lokale stroming rond de sensordraad.

Beide gevallen zijn onderhevig aan praktische beperkingen. Voor een rechte
probe kan de amplitude niet willekeurig groot worden. Anderzijds treden er bij
een 90 ◦ probe bij stijgende frequentie mechanische vibraties op, die het anemo-
metersignaal verstoren ten gevolge van strain gauging. De verdere optimalisatie
is een mogelijk onderwerp van toekomstig onderzoek.
3.4 Toepassing en validatie

Sectie 3.8 bespreekt de gepaste defasering tussen oscillatorfrequentie en motor-
toerental, waardoor geldige metingen worden genomen in alle krukasposities
(Fig. 3.15). De verhouding tussen beide frequenties wordt aangeduid door Rf
(Vgl. (3.13) en (3.16)).
Sectie 5.2.1 bespreekt de validatie van de OHW op de CME opstelling.
Het debiet bepaald door integratie van de snelheidsverdeling wordt vergele-
ken met het referentiedebiet, gemeten met een laminaire stromingsmeter in het
inlaatsysteem. Figuur 5.20 toont aan dat de OHW met voldoende hoge pro-
besnelheid (Rf 1) de fout op het debiet reduceert. De afwijking is het gevolg
van rectification of gelijkrichtfouten, omdat standaard HWA enkel gevoelig is
aan de absolute waarde van de snelheid. Bij gebruik van standaard HWA is
Rf = 0; volgens Fig. 5.2.1 bedraagt de overschatting van de snelheid dan tot
50 %, afhankelijk van de grootte van de terugstroming.
4 Additieprincipe
4.1 Datareductie
De geldigheid van het additieprincipe (Vgl. (4.1)) is gekwantificeerd met behulp
van twee dimensieloze scalars rS en rM , die de gelijkenis aanduiden tussen
pulserende en stationaire snelheidsverdeling, op basis van vorm (rS , Vgl (4.12))
en magnitude (rM , Vgl. (4.28)).
De geldigheid van het additieprincipe is afhankelijk van een enkele dimensie-
loze grootheid S die de stroming karakteriseert. Het uitdrijving of scavenging
getal S is de verhouding van twee belangrijke tijdschalen: S = Tp /Ts , met Tp
de schijnbare pulsatieperiode [s] (Vgln. (4.44) en (4.46)) en Ts de verblijftijd [s]
(Vgl. (4.43)) in het diffusorvolume voor de katalysator. De gelijkenisgroothe-
den rS en rM vertonen exponentiële correlaties in functie van S. S stijgt met
afnemend toerental, toenemende belasting, en afnemend diffusorvolume.
4.2 Resultaten
Secties 4.4.1 en 4.4.2 tonen aan dat het scavenging getal S gecorreleerd is met
de overeenkomst tussen de stationaire en pulserende stromingsverdeling, en
aldus met de geldigheid van het additieprincipe. In Fign. 4.13 tot 4.27 wijst
een vergelijking van de stationaire en pulserende verdelingen erop dat rM > 1.
1
1.6 ISOT, A, RV
0.9 ISOT, A, CH
ISOT, B, CH

0.8 CME, B, CH
1.5 rM’’
0.7
rM’’ = 1.118 + 0.337 exp(- S / 0.723)
0.6 1.4
(R2 = 0.30)
rS’’ = 1 - exp(- S / 0.723)
0.5
(R2 = 0.91) 1.3
0.4
0.3 1.2
ISOT, A, RV
0.2 ISOT, A, CH
ISOT, B, CH 1.1
CME, B, CH
0.1 rS’’
0 1
0 1 2 3 4 5 0 1 2 3 4 5
(a) (b)
Figuur 4.29 – Correlaties voor gelijkenisgrootheden (a) rS en (b) rM versus scaven-

ging getal S
Gegeven de definitie van rM (Vgl. (4.28)), is de stromingsuniformiteit steeds

hoger in pulserende dan in stationaire stroming.
Uit de gecombineerde resultaten voor de CME en isotherme opstelling volgt
een verbeterde definitie van de pulsatieperiode Tp (Vgl. (4.46)), die rekening
houdt met de hogere frequentie-inhoud van het collectordebiet bij de CME
opstelling. Op basis van de verbeterde definitie komen de resultaten op beide
opstellingen goed overeen (Fig. 4.29), terwijl dat niet zo is voor de originele
definitie van Tp (Fig. 4.28).
Het verband tussen de gelijkenisgrootheden rS en rM en het scavenging
getal S (Fig. 4.29) en de exponentiële correlaties voor rS en rM in Vgl. (4.47)
vormen een belangrijk resultaat. De grafieken bevatten resultaten bekomen op
isotherme en CME opstelling, voor uitlaatcollectoren met en zonder uitlaat-
klepoverlap, en voor een roterende klep en cilinderkop als pulsator. Gezien de
verschillende opstellingen is de overeenkomst opmerkelijk.
Tabellen 4.3 en 4.4 geven de statistische significantie van rS en rM voor de
isotherme en CME opstelling. De P-waarden voor de statistische hypothesetes-
ten leiden niet tot een duidelijke geldigheidslimiet voor het additieprincipe. Op
basis van de exponentiële correlaties voor rS en rM komt de grenswaarde logi-
scherwijs overeen met de kritische waarde voor het scavenging getal Scrit , die
volgt uit de gefitte correlaties. Het additieprincipe is geldig wanneer S > Scrit .
De statistische onzekerheid op Scrit is onderzocht met een Monte-Carlo analyse.
Voor de gecombineerde resultaten van collector A en B is het kritische scaven-
ging getal Scrit = 0.723 ± 0.052. Wanneer enkel de resultaten van collector B
beschouwd worden, is Scrit = 0.722 ± 0.056.
De bekomen resultaten worden deels bevestigd door andere bronnen. Ben-
jamin [17] toont het verband tussen stromingsuniformiteit en een getal over-
eenkomstig aan het scavenging getal S. De resultaten zijn bekomen met een
isotherme opstelling. Net zoals rM stijgt de stromingsuniformiteit bij afnemend
scavenging getal S. Bressler et al. [22] stelt dat de snelheidsverdeling in een
5 Stromingsdynamica 297
katalysator enkel afhankelijk is van een verhouding die overeenkomt met S. De

afwijking tussen pulserende en stationaire stromingsverdeling blijkt minimaal
voor hoge waarde van S, wat overeenkomt met het verband tussen rS en S.
4.3 Fysische interpretatie

De opmerkelijke correlatie tussen rS en S (Fig. 4.29) wijst op een analogie
tussen deze stroming en de nuldimensionale spoelproces van een mengvolume
met een scalaire grootheid, vb. een stofconcentratie. Sectie 4.6 werkt deze
analogie uit, waaruit volgt dat het kritisch scavenging getal Scrit zich gedraagt
als de verhouding van effectief gebruikt tot geometrisch diffusorvolume. Bij
deels stilstaande of circulerende stroming in de diffusor, neemt het effectieve
diffusorvolume dat deelneemt aan het spoelproces af. Dit leidt tot de hypothese
dat het kritisch scavenging getal Scrit een maat is voor de efficiëntie ηD waarmee
de collector (i.e. combinatie van runners en diffusor) de stroming verdeelt over
de katalysator.
Gezien voor collector B Scrit = 0.722 ± 0.056, stelt de hypothese dat slechts
72 % van de diffusor effectief gebruikt wordt voor de verdeling van de stroming.
Dit wordt intuïtief bevestigd door de niet-optimale stromingsuniformiteit. Een
verdere optimalisatie van de collector zou de collectorefficiëntie ηD (= Scrit ,
het kritisch scavenging getal) verhogen en de stromingsverdeling verbeteren.
Verder onderzoek moet uitwijzen of het kritisch scavenging getal inder-
daad overeenkomt met een collectorefficiëntie, wat een objectieve, kwantitatieve
maat zou vormen voor de performantie van de uitlaatcollector qua stromings-
uniformiteit.
5 Stromingsdynamica
5.1 Experimentele resultaten
De hoge bandbreedte van HWA maakt gedetailleerde tijdsafhankelijke metingen
mogelijk. Bij vergelijking van de katalysatorsnelheid op de isotherme (Fig. 5.9a)
en CME opstelling (Fig. 5.11a), blijkt een aanzienlijk verschil in de frequentie-
inhoud. De stromingsdynamica in de CME opstelling komt beter overeen met
werkelijke motorcondities.
Figuren 5.14 tot 5.18 tonen de hoge resolutie in ruimte en tijd bekomen
met de OHW. Figuren 5.16 tot 5.18 tonen de tijdsafhankelijke snelheidsver-
deling in de katalysator tijdens de beginfase van de uitlaatslag van de eerste
cilinder. De grijze zones wijzen op terugstroming. Terugstroming is maximaal
vlak na de blowdown, maar treedt ook op tijdens de uitdrijvingsfase omwille
van de sterke resonantie-effecten. Deze experimenten zijn uitgevoerd op de
isochore CME opstelling, aan 1200 rpm en voor drie motorbelastingen. Perio-
dieke terugstroming is vastgesteld in een breed werkingsbereik. Bepaalde zones
(Fig. 5.22) zijn meer onderhevig aan terugstroming, vooral aan de buitenrand
van de doorsnede, maar evengoed in het centrale deel, overeenkomstig de tijds-
gemiddelde stromingsverdeling. Figuur 5.23 toont dat de lokale snelheid in elk
van de zones in Fig. 5.22 dezelfde fluctuaties vertonen; de stroming in de gehele

doorsnede fluctueert in fase.
De frequentie van deze snelheidsfluctuaties is onafhankelijk van toerental
of debiet (Tabel 5.2), wat wijst op een gasdynamische Helmholtz resonantie.
De Helmholtz-gerelateerde fluctuaties zijn het meest uitgesproken tijdens de
uitdrijvingsfase. Op het eerste zicht beïnvloeden de resonanties de tijdsgemid-
delde snelheidsverdelingen niet. Echter, de fluctuaties verhogen de schijnbare
pulsatiefrequentie 1/Tp en verlagen bijgevolg de waarde van het scavenging
getal S. Gezien het verband tussen rS , rM en S, beïnvloeden de resonan-
ties indirect de stromingsuniformiteit en de geldigheid van het additieprincipe.
Andere auteurs [70, 81, 4] vinden gelijkaardige snelheidsfluctuaties tijdens de
uitdrijvingsfase.
Figuur 5.38 toont een equivalent model van de laagfrequente gasdynamica in
de uitlaatcollector, waarbij componenten met aanzienlijke impuls zich gedragen
als onsamendrukbare massa’s, en de andere als samendrukbare veren. Het gas-
dynamische systeem wordt beschouwd als mechanisch massa-veer-demper equi-
valent, met demping veroorzaakt door vb. de uitlaatkleppen, wandwrijving, en
de katalysatordrukval. De Helmholtz resonantiefrequentie wordt gegeven door
Vgl. (5.5), overeenkomstig de nulde orde gasdynamische resonantiefrequentie.
Hogere orde akoestische resonanties zijn afkomstig van staande golfpatronen in
het leidingensysteem.
Vergelijking (5.5) geeft een goede voorspelling van de geobserveerde fluctu-
atiefrequenties, tussen 140 en 200 Hz voor collector B. Aangezien het uitlaat-
systeem bestaat uit verschillende componenten is het niet duidelijk om welke
resonantie het gaat, temeer omdat de samendrukbaarheid van cilinder en dif-
fusor gelijkaardig zijn. Aan de hand van een numeriek gasdynamisch model
wordt de oorsprong van de resonanties achterhaald.
5.2 Numerieke analyse

Secties 5.2.3 en 5.3.4 maken gebruik van een gasdynamisch ééndimensionaal
model van het uitlaatsysteem. Zoals beschreven in App. D is het model geïm-
plementeerd in Simulink11 , en is het gebaseerd op een tweede orde cell-vertex
total variation diminishing (TVD) discretisatieschema [101] met vierde orde
Runge-Kutta tijdsintegratie. Het model is gevalideerd met relevante testgeval-
len (vb. tijdsevolutie van schokgolven in een leiding (App. D.2.1), geluiddem-
ping in een eenvoudige uitlaatdemper met enkele expansiekamer (App. D.2.3)),
en houdt rekening met de gekende stromingsweerstand van uitlaatkleppen, lo-
kale drukverliezen in gekromde uitlaatrunners en de drukval in de katalysator
(App. A.2). Het inlaatsysteem en de verbranding worden niet beschouwd, maar
de initiële toestand wordt in de cilinders opgelegd vààr elke uitlaatklep opent.
De discretisatie van de uitlaatleidingen is gekozen overeenkomstig App. D.3, zo-
dat het model een frequentieresolutie heeft van ca. 2000 Hz. Bijgevolg kan het
numeriek model dienen om de Helmholtz resonanties (ca. 200 Hz) te verklaren.
De invloed is nagegaan van de aanwezigheid van een uitlaatleiding met
demper, stroomafwaarts van de katalysator. Aangezien alle metingen gebeurd
5 Stromingsdynamica 299
5 5
Um,exp Um,exp
4 Um,num 4 Um,num
3 3
Mean velocity (−)
Mean velocity (−)

2 2
1 1
0 0
−1 −1
−2 −2
0 180 360 540 720 0 180 360 540 720
(a) (d)
Figuur 5.28a,d – Tijdsafhankelijke numerieke en experimentele katalysatorsnelheid

Um (ωt) [-], voor (a) N = 600 rpm (lage last), (d) N = 1200 rpm (hoge last)
zijn met vrije uitstroming, maakt het model het mogelijk de validiteit hiervan
na te gaan.
Figuren 5.28 en 5.29 tonen aan dat de katalysatorsnelheid goed voorspeld
wordt door het numeriek model, ondanks de driedimensionale stroming in de
diffusor. Bij sterke terugstroming (Fig. 5.28b,d en Fig. 5.29b) blijkt de OHW
met de maximaal meetbare −1 m/s niet in staat de correcte snelheid weer te
geven. Op basis van de goede overeenkomst tussen experimentele en nume-
rieke resultaten, is het model betrouwbaar genoeg om de stromingsdynamica
te voorspellen buiten het bereik van de experimentele opstelling, vb. (i) in
aanwezigheid van een uitlaatleiding, en (ii) in werkelijke motorcondities.
Figuur 5.32 toont het verschil tussen een werkende motor en de CME opstel-
ling, voor vergelijkbare condities. De frequentie van de Helmholtz fluctuaties is
hoger, evenredig met de vierkantswortel van de temperatuurverhouding. Verder
is het maximum debiet tijdens de blowdown groter, zoals reeds aangeduid bij
de bespreking van de stromingsgelijkvormigheid (Sect. 2.3). De terugstroming
na de blowdown is sterker in werkende motorcondities.
Figuren 5.33 en 5.34 tonen dat de aanwezigheid van een uitlaatleiding niet
veel invloed heeft op de katalysatorsnelheid, althans in CME condities. Bij
werkende motorcondities is er wel een verschil merkbaar met en zonder uitlaat-
leiding (Fign. 5.35a en 5.32a).
Frequentie respons functies van de uitlaatcollector worden bepaald met het
numeriek model, om het Helmholtz resonantieverschijnsel verder te verklaren.
Voor de CME opstelling (en voor werkende motorcondities) blijkt het gas in de
open runner en katalysator te oscilleren op het samendrukbaar cilindervolume.
Nochtans zijn voor de isotherme opstelling soortgelijke maar zwakkere fluctua-
ties waargenomen, waarbij de diffusor zich samendrukbaar gedraagt, met quasi
dezelfde resonantiefrequentie tot gevolg.
In de CME opstelling is een faseverschuiving van π/2 rad merkbaar tussen
de snelheid in runner en katalysator, wat ook wijst op een samendrukbaar

diffusorvolume. Dit wordt niet bevestigd door het numeriek model. Figuur 5.45
toont dat de uitlaatleiding slechts een kleine invloed heeft op de transfer functie
van de uitlaatcollector.
De transferfunctie in Fig. 5.42c komt overeen met de akoestische admittantie
van de uitlaatcollector. Deze plot komt kwalitatief overeen met het omgekeerde
van de akoestische impedantie opgemeten door Boonen [21] met behulp van een
dubbele microfoon techniek.
Andere auteurs bevestigen de waargenomen stromingsdynamische fenome-
nen. Adam et al. [4], Park et al. [81], Liu et al. [70] en Benjamin et al. [14]
tonen numerieke en experimentele resultaten in uitlaatcollectoren met voor-
katalysator, in werkende motorcondities. Sterke snelheidsfluctuaties worden
geobserveerd tijdens de uitdrijvingsfase, met frequenties tussen 300 en 600 Hz.
Afgezien van de verschillende geometrieën, zijn deze frequenties ongeveer twee-
maal de waarde bij de CME opstelling, omwille van temperatuurafhankelijkheid
van de geluidssnelheid in Vgl. (5.5) en de temperatuurverhouding van ca. 4 : 1.
6 Ontwerpoverwegingen
Voor het industrieel ontwerp van uitlaatcollectoren vormt het additieprinci-
pe besproken in Hoofdstuk 4 een interessante bevinding. Het additieprincipe
houdt in dat het ontwerp kan volstaan met stationaire CFD berekeningen, en
dat tijdsintensieve transiënte CFD berekeningen overbodig zijn. Rekening hou-
dend met de steeds kortere ontwikkelingstijden vormt dit een belangrijk besluit.
Het huidige onderzoek stelt een objectieve grenswaarde vast (i.e. het kritisch
scavenging getal Scrit ) voor de geldigheid van het additieprincipe.
Uitgaande van de correlatie voor rM (Vgl. (4.47)) blijkt de stromingsuni-
formiteit steeds groter voor pulserende dan stationaire stroming. Voor grote
S waarden evolueert rM naar een constante waarde, groter dan één. Bijgevolg
voldoet de stromingsuniformiteit van een collector ontworpen voor stationaire
stroming zeker voor pulserende stroming.
De stromingsuniformiteit in pulserende stroming wordt groter wanneer S <
Scrit . Een grotere diffusor verkleint het scavenging getal S en vergroot aldus
de stromingsuniformiteit. Dit triviale besluit is onderworpen aan geometrische
en thermische beperkingen van het motorcompartiment. Sectie 6.2 maakt een
interessante opmerking bij bovenstaande vaststelling, gerelateerd aan de Helm-
holtz resonanties in de collector, en verbindt op indirecte manier de bevindingen
van Hoofdstukken 4 en 5.
De Helmholtz-gerelateerde snelheidsfluctuaties verhogen de pulsatiefrequen-
tie en verlagen het scavenging getal. Gezien het verband tussen S, rS en rM ,
lijken de resonanties positief voor de stromingsuniformiteit; hoe hoger de reso-
nantiefrequentie fH , hoe beter de √ stromingsuniformiteit.
Volgens Vgl. (5.5) is fH ∝ 1/ L en fH ∝ sqrtT , met L de runnerlengte
en T de temperatuur. Beide aspecten komen overeen met de evolutie van mo-
derne uitlaatsystemen, waarin de katalysator zo dicht mogelijk bij de motor
7 Conclusie 301
wordt geplaatst (L %) en warmteverliezen beperkt worden (T 1). Kortom, de

geobserveerde Helmholtz snelheidsfluctuaties zijn voordelig voor de stromings-
uniformiteit.
7 Conclusie
De thesis betreft een experimentele studie van de pulserende stroming in mo-
derne uitlaatcollectoren met voorkatalysator. Gezien de problemen met CFD
voorspelling van de stroming, en het belang van de stromingsuniformiteit in de
katalysator, richt de thesis zich op het ontwikkelen van een experimentele aan-
pak die nauwkeurige bidirectionele snelheidsmetingen oplevert, met een hoge
resolutie in ruimte en tijd.
• Met het oog op hoge resolutie snelheidsmeettechnieken, maakt de thesis

gebruik van opstellingen die koude pulserende stroming genereren in het
uitlaatsysteem, waarbij de stromingsgelijkvormigheid met een werkende
motor onderzocht wordt [88, 87, 86, 84, 85].
• Een nieuwe oscillerende hittedraad anemometer (OHW) is ontwikkeld

om bidirectionele snelheid te meten met dezelfde resolutie als standaard
hittedraad anemometrie. De OHW is gekalibreerd in een specifieke wind-
tunnel, met laser Doppler anemometrie als referentie snelheidsmeting.
De maximaal meetbare negatieve snelheid bedraagt −1 m/s, vergelijk-
baar met andere systemen. De kalibratie is uitgevoerd voor verschillende
probe types. Uit een schaalanalyse volgt dat probes met rechte en 90ř
uiteinden zich verschillend gedragen, waaruit richtlijnen volgen ter opti-
malisatie van het systeem. De OHW is succesvol toegepast bij meting
van de bidirectionele snelheidsverdeling in uitlaatsystemen met voorka-
talysator [83, 84, 85].
• De experimentele validatie van het additie principe (Vgl. (4.1)) is ge-

baseerd op de correlatie tussen het scavenging getal S (Vgl. (1.4)) en
twee dimensieloze maten voor de gelijkenis tussen snelheidsverdelingen
in pulserende en stationaire stroming. De validatie is ondersteund door
statistische hypothesetesten [88, 86].
De kritische waarde van het scavenging getal Scrit dient als geldigheids-
grens voor het additieprincipe. Het additieprincipe is geldig als S > Scrit .
In dat geval is de correlatiecoëfficiënt tussen de snelheidsverdelingen in
pulserende en stationaire stroming hoger dan 0.63.
Het additieprincipe houdt enkele belangrijke implicaties in voor het in-
dustrieel ontwerp van uitlaatcollectoren met voorkatalysator:
(i) Gebaseerd op de correlatie voor rM (Vgl. (4.47)) is de stromingsuni-

formiteit in pulserende stroming steeds hoger dan voor stationaire
stroming. Een collector ontworpen voor stationaire stroming zal dus
voldoen qua uniformiteit in pulserende stroming.
(ii) In het geldigheidsgebied van het additieprincipe (S > Scrit ), kunnen

stationaire CFD berekeningen de tijdsgemiddelde pulserende stro-
ming accuraat voorspellen, wat resulteert in een kortere product-
ontwikkelingstijd.
• Sterke fluctuaties in de katalysatorsnelheid zijn waargenomen in de CME

opstelling, alsook door andere auteurs in werkelijke motorcondities [81,
59, 4, 70]. Deze fluctuaties worden in Hoofdstuk 5 toegeschreven aan
Helmholtz resonanties. Deze aanname wordt bevestigd door middel van
transfer functies, bepaald met een ééndimensionaal gasdynamisch model
van het uitlaatsysteem [84, 85].
• Het optreden van ogenblikkelijke lokale terugstroming in de katalysator
is onderzocht in de CME opstelling, en is reeds geobserveerd door andere
auteurs [70, 81, 59]. Gezien de inherente ongevoeligheid van hittedraad
anemometrie (HWA) voor de richting van de snelheidsvector, gebruiken
andere onderzoekers optische snelheidsmeettechnieken. Problemen met
optische toegang en seeding beperken de ruimtelijke en tijdsresolutie. Met
de OHW [83] is lokale periodieke terugstroming gemeten in de katalysa-
tordoorsnede, op de CME opstelling [84, 85].
• Een numeriek ééndimensionaal gasdynamisch model van het uitlaatsys-
teem werd ontwikkeld op basis van een tweede orde cell-vertex total va-
riation diminishing discretisatieschema [101]. Het model is gevalideerd
voor relevante testgevallen. Omdat numerieke en experimentele resulta-
ten voor de katalysatorsnelheid goed corresponderen, wordt het model ge-
bruikt om de stromingsdynamica te voorspellen buiten de mogelijkheden
van de experimentele aanpak, vb. de aanwezigheid van een uitlaatleiding
met demper, of de invloed van werkende motorcondities. Op basis van
het gasdynamisch model worden transfer functies van de gasdynamica in
de uitlaatcollector opgesteld, die beter inzicht geven in de geobserveerde
resonantiefenomenen [84, 85].
Deze thesis draagt bij tot het begrip van stromingsdynamica in uitlaatcol-
lectoren met voorkatalysator in het algemeen, en de snelheidsverdeling in de
katalysator in het bijzonder. De experimentele aanpak resulteert in hoge reso-
lutie bidirectionele snelheidsmetingen die moeilijk te bekomen zouden zijn in
werkende motorcondities. De geldigheid van het additieprincipe is vastgesteld,
met belangrijke implicaties voor het optimaal ontwerp van deze systemen. Tot
slot is de stromingsdynamica geanalyseerd met behulp van een ééndimensionaal
gasdynamisch model van het uitlaatsysteem.
Bibliography
[1] Directive 98/69/EC of the European Parliament and of the Council of

13 October 1998 relating to measures to be taken against air pollu-
tion by emissions from motor vehicles and amending Council Directive
70/220/EEC. Off. J. Eur. Commun., 41(L 350):1–57, 1998.
[2] IWT Onderzoek & Ontwikkeling project VLIET/970375/BOSAL, 2003.
[3] IWT Onderzoek & Ontwikkeling project IWT/030582/BOSAL, 2005.
[4] M. Adam, M. Heinrich, M. Hopp, and O. Lang. Optimierung von
Abgasanlagen mit analytischen Methoden. Motortechnische Zeitschrift,
63(10):838–846, 2002.
[5] R. J. Adrian. Particle-imaging techniques for experimental fluid mechan-
ics. Annu. Rev. Fluid Mech., 23:261–304, 1991.
[6] H.-E. Albrecht, M. Borys, N. Damaschke, and C. Tropea. Laser Doppler
and phase Doppler measurement techniques. Springer-Verlag, Berlin, D,
2003.
[7] A. Arias-Garcia, S. F. Benjamin, H. Zhao, and S. Farr. A comparison
of steady, pulsating flow measurements and CFD simulations in close-
coupled catalysts. SAE Paper, 2001–01–3662, 2001.
[8] L. Bai, F. Q. Zhao, Y. Liu, M. C. Lai, and K. S. Lee. Transient flow and
pressure characteristics inside a close-coupled catalytic converter. SAE
Paper, 982548, 1998.
[9] R. C. Baker. Flow measurement handbook: Industrial designs, operating
principles, performance and applications. Cambridge University Press,
Cambridge, UK, 2000.
[10] J. B. Barlow, W. H. Rae, and A. Pope. Low-speed wind tunnel testing.
John Wiley & Sons, New York, 1999.
[11] C. H. Bartholomew. Mechanisms of catalyst deactivation. Appl. Catal.
A: Gen., 212(1–2):17–60, 2001.
303
304 Bibliography
[12] J. S. Bendat and A. G. Piersol. Random data: Analysis and measurement

procedures. John Wiley & Sons, New York, 1971.
[13] S. F. Benjamin, R. J. Clarkson, N. Haimad, and N. S. Girgis. An experi-
mental and predictive study of the flow field in axisymmetric automotive
exhaust catalyst systems. SAE Paper, 961208, 1996.
[14] S. F. Benjamin, W. Disdale, Z. Liu, C. A. Roberts, and H. Zhao. Ve-
locity predictions from a coupled 1D/3D CFD simulation compared with
measurements in the catalyst system of a firing engine. Int. J. Engine
Res., 2006. (in press).
[15] S. F. Benjamin, N. Haimad, C. A. Roberts, and J. Wollin. Modelling the
flow distribution through automotive catalytic converters. Proc. IMechE
C-J. Mech. Eng. Sci., 215(4):379–383, 2001.
[16] S. F. Benjamin, C. A. Roberts, and J. Wollin. A study of the effect of
flow pulsations on the flow distribution within ceramic contoured catalyst
substrates. SAE Paper, 2001-01-1996, 2001.
[17] S. F. Benjamin, C. A. Roberts, and J. Wollin. A study of pulsating flow
in automotive catalyst systems. Exp. Fluids, 33(5):629–639, 2002.
[18] S. F. Benjamin, H. Zhao, and A. Arias-Garcia. Predicting the flow field
inside a close-coupled catalyst - the effect of entrance losses. Proc. IMechE
C-J. Mech. Eng. Sci., 217(3):283–288, 2003.
[19] R. S. Benson, J. H. Horlock, and D. E. Winterbone. The thermodynamics
and the gas dynamics of internal combustion engines, Volume I. Oxford
University Press, Oxford, 1982.
[20] L. L. Beranek. Acoustics. McGraw-Hill, New York, 1954.
[21] R. Boonen. Development of an active exhaust silencer with acoustical
characterization of internal combustion engines. PhD thesis, Katholieke
Universiteit Leuven, Leuven, B, 2003.
[22] H. Bressler, D. Rammoser, H. Neumaier, and F. Terres. Experimental
and predictive investigation of a close-coupled catalyst converter with
pulsating flow. SAE Paper, 960564, 1996.
[23] M. Breuer, C. Schernus, R. Böwing, A. Kuphal, and S. Lieske. Ex-
perimental approach to optimize catalyst flow uniformity. SAE Paper,
2000-01-0865, 2000.
[24] C. Brucker. 3D scanning PIV applied to an air flow in a motored engine
using digital high-speed video. Meas. Sci. Technol., 8(12):1480–1492,
1997.
[25] H. H. Bruun. Hot-wire anemometry: Principles and signal analysis. Ox-
ford University Press, Oxford, UK, 1995.
Bibliography 305
[26] A. E. Catania and A. Mittica. Extraction techniques and analysis of tur-

bulence quantities from in-cylinder velocity data. J. Eng. Gas. Turbines
Power-Trans. ASME, 111:466–478, 1989.
[27] Y. A. Çengel. Introduction to thermodynamics and heat transfer.

McGraw-Hill, New York, 1997.
[28] R. J. Clarkson. A theoretical and experimental study of automotive cat-

alytic converters. PhD thesis, Coventry University, Coventry, UK, 1995.
[29] A. D. Cliff and J. K. Ord. Spatial autocorrelation. Pion, London, 1973.
[30] C. F. Colebrook and C. M. White. Experiments with fluid friction in

roughened pipes. Proc. Roy. Soc. London, A, 161(906):367–381, 1937.
[31] C. Czado and A. Munk. Assessing the similarity of distributions - finite

sample performance of the empirical mallows distance. J. Stat. Comp.
Sim., 60:319–346, 1998.
[32] D. Küchemann and J. Weber. Aerodynamics of propulsion. McGraw-Hill,

London, 1953.
[33] D. D. Davis, G. M. Stokes, D. Moore, and G. L. Stevens. NACA Re-

port 1192: Theoretical and experimental investigation of mufflers with
comments on engine-exhaust muffler design. National Advisory Commit-
tee for Aeronautics (NACA), Langley Aeronautical Laboratory, Langley
Field (VA), USA, 1954.
[34] R. Diez, V. Knipps, and B. Koch. Das Abgassystem des BMW M5.
Motortechnische Zeitschrift, 66(3):178–182, 2005.
[35] G. Eigenberger and U. Nieken. Catalytic combustion with periodic flow

reversal. Chem. Eng. Sci., 43(8):2109–2115, 1988.
[36] H. E. Fiedler. A note on secondary flow in bends and bend combinations.

Exp. Fluids, 23(3):262–264, 1997.
[37] L. M. Fingerson and P. Freymuth. Thermal anemometers. In R. J. Gold-

stein, editor, Fluid Mechanics Measurements, pages 99–154. Hemisphere,
Washington, 1983.
[38] P. Forzatti and L. Lietti. Catalyst deactivation. Catal. Today, 52(2–

3):165–181, 1999.
[39] P. Freymuth. History of thermal anemometry. In N. P. Cheremisinoff

and R. Gupta, editors, Handbook of Fluids in Motion, pages 79–91. Ann
Arbor Science Publishers, Ann Arbor, USA, 1983.
[40] P. Freymuth. Bibliography of thermal anemometry. TSI Inc., St. Paul,

USA, 1992.
306 Bibliography
[41] P. Guillaume, J. Schoukens, R. Pintelon, and I. Kollar. Crest-factor

minimization using nonlinear chebyshev approximation methods. IEEE
Trans. Instr. Meas., 40(6):982–989, 1991.
[42] I. Guttmann, S. S. Wilks, and J. S. Hunter. Introductory engineering

statistics. John Wiley & Sons, New York, 1982.
[43] N. Haimad. A theoretical and experimental investigation of the flow per-

formance of automotive catalytic converters. PhD thesis, Coventry Uni-
versity, Coventry, UK, 1997.
[44] A. Hald. Statistical theory with engineering applications. John Wiley &
Sons, New York, 1952.
[45] P. M. Handford and P. Bradshaw. The pulsed-wire anemometer. Exp.

Fluids, 7(2):125–132, 1989.
[46] J. B. Heywood. Fluid motion within the cylinder of internal combustion

engines - The 1986 Freeman scholar lecture. J. Fluids Eng.-Trans. ASME,
109(1):3–35, 1987.
[47] J. B. Heywood. Internal combustion engine fundamentals. McGraw-Hill,

New York, 1988.
[48] C. Hirsch. Numerical computation of internal and external flows. Vol. 1:

Fundamentals of numerical discretization. John Wiley & Sons, Chich-
ester, UK, 1995.
[49] C. Hirsch. Numerical computation of internal and external flows. Vol.

2: Computational methods for inviscid and viscous flows. John Wiley &
Sons, Chichester, UK, 1995.
[50] J. H. Horlock and D. E. Winterbone. The thermodynamics and the gas

dynamics of internal combustion engines, Volume II. Oxford University
Press, Oxford, 1986.
[51] J. S. Howitt and T. C. Sekella. Flow effects in monolithic honeycomb

automotive catalytic converters. SAE Paper, 740244, 1974.
[52] H. J. Hussein. Evidence of local axisymmetry in the small scales of a

turbulent planar jet. Phys. Fluids, 6(6):2058–2070, 1994.
[53] K. Hwang, K. Lee, J. Mueller, T. Stuecken, H. J. Schock, and J. C. Lee.

Dynamic flow study in a catalytic converter using LDV and high speed
flow visualisation. SAE Paper, 950786, 1995.
[54] ISO 5167-1991. Measurement of fluid flow by means of pressure dif-

ferential devices inserted in circular cross-section conduits running full.
Part 2: Orifice plates. International Organization for Standardization,
Geneva, CH, 1991.
Bibliography 307
[55] ISO/TR 3313-1998. Measurement of fluid flow in closed conduits: Guide-

lines on the effects of flow pulsations on flow-measurement instruments.
International Organization for Standardization, Geneva, CH, 1998.
[56] J. Kaspar, P. Fornasiero, and N. Hickey. Automotive catalytic converters:

current status and some perspectives. Catal. Today, 77(4):419–449, 2003.
[57] W. M. Kays and A. L. London. Compact heat exchangers. McGraw-Hill,

New York, 1984.
[58] D. S. Kim and Y. S. Cho. Ldv measurement, flow visualization and nu-
merical analysis of flow distribution in a close-coupled catalytic converter.
KSME Int J, 18(11):2032–2041, 2004.
[59] H. S. Kim, K. Min, C. L. Myung, and S. Park. A combined experimental

and computational approach to improve catalyst flow uniformity and
light-off behaviour. Proc. IMechE D-J. Automob. Eng., 216(5):413–430,
2002.
[60] J. Y. Kim, M. C. Lai, P. Li, and G. K. Chui. Flow distribution and

pressure drop in diffuser-monolith flows. J. Fluids Eng.-Trans. ASME,
117(3):362–368, 1995.
[61] H. Lambert, S. Simon, N. Collings, and T. Hands. The fast fid as a

velocimeter for flow measurements in an automotive catalyst. SAE Paper,
980879, 1998.
[62] D. R. Lancaster. Effects of engine variables on turbulence in a spark-

ignition engine. SAE Paper, 760159, 1976.
[63] T. Lehmann, A. Sovakar, W. Schmitt, and R. Repges. A comparison of

similarity measures for digital subtraction radiography. Comput. Biol.
Med., 27(2):151–167, 1997.
[64] C. D. Lemme and W. R. Givens. Flow through catalytic converters: An

analytical and experimental treatment. SAE Paper, 740243, 1974.
[65] E. Levina and P. Bickel. The earth moverŠs distance is the Mallows dis-
tance: some insights from statistics. IEEE Int. Conf. Computer. Vision.,
2001.
[66] Y. Li and A. Naguib. An oscillating hot-wire technique for resolving the

magnitude and direction of velocity measurements using single hot-wire
sensors. Exp. Fluids, 34(5):597–606, 2003.
[67] Y. Li and A. Naguib. High-frequency oscillating hot-wire sensor for near-

wall diagnostics in separated flows. AIAA J, 43(3):520–529, 2005.
[68] T. M. Liou and D. A. Santavicca. Cycle resolved LDV measurements in

a motored IC engine. J. Fluids Eng.-Trans. ASME, 107:232–240, 1985.
308 Bibliography
[69] Z. Liu, S. F. Benjamin, and C. A. Roberts. Pulsating flow maldistribution

within an axisymmetric catalytic converter - Flow rig experiment and
transient CFD simulation. SAE Paper, 2003–01–3070, 2003.
[70] Z. Liu, S. F. Benjamin, C. A. Roberts, H. Zhao, and A. Arias-Garcia. A

coupled 1D/3D simulation for the flow behaviour inside a close-coupled
catalytic converter. SAE Paper, 2003–01–1875, 2003.
[71] C. L. Mallows. Note on asymptotic joint normality. Annals Math. Sci.,

43(2):508–515, 1972.
[72] G. McPherson. Statistics in scientific investigation: Its basis, application

and interpretation. Springer-Verlag, New York, 1990.
[73] R. Mezaki and H. Inoue. Rate equations of solid-catalyzed reactions.

University of Tokyo Press, 1991.
[74] D. S. Miller. Internal Flow Systems. BHRA Fluid Engineering, Cranfield,

UK, 1996.
[75] S. Missy, J. Thams, M. Bollig, R. Tatschl, R. Wanker, G. Bachler, A. En-

nemoser, and H. Grantner. Computergestützte Optimierung des Ab-
gasnachbehandlungssystems für den neuen 1.8l-Valvetronic-Motor von
BMW. Motortechnische Zeitschrift, 63(1):18–29, 2002.
[76] A. Mitri, D. Neumann, T. Liu, and G. Veser. Reverse-flow reactor opera-

tion and catalyst deactivation during high-temperature catalytic partial
oxidation. Chem. Eng. Sci., 59(22–23):5527–5534, 2004.
[77] P. A. P. Moran. Notes on continuous stochastic phenomena. Biometrika,

37:17–23, 1950.
[78] A. M. Moulin, M. Gaster, C. N. Woodburn, J. R. Barnes, and M. E.

Welland. A directionally sensitive hot-wire anemometer. Exp. Fluids,
22(6):458–462, 1997.
[79] T. Nagel and J. Diringer. Minimum test requirements for high cell-
density, ultra-thin wall catalyst supports: Part 1. SAE Paper, 2000-
01-0495, 2000.
[80] A. K. Neyestanaki, F. Klingstedt, T. Salmi, and D. Y. Murzin. Deac-

tivation of postcombustion catalysts, a review. Fuel, 83(4–5):395–408,
2004.
[81] S. B. Park, H. S. Kim, K. M. Cho, and W. T. Kim. An experimental

and computational study of flow characteristics in exhaust manifold and
CCC (close-coupled catalyst). SAE Paper, 980128, 1998.
[82] A. E. Perry and G. L. Morrison. Errors caused by hot-wire filament

vibration. J. Phys. E-Sci. Instrum., 5(10):1004–1008, 1972.
Bibliography 309
[83] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Calibration of an

oscillating hot-wire anemometer for bidirectional velocity measurements.
Exp. Fluids, 40(4):555–567, 2006. http://dx.doi.org/10.1007/s00348-005-
0095-4.
[84] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Experimental study

of flow dynamics in close-coupled catalyst manifolds. Int. J. Engine Res.,
2006. (in press).
[85] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Experimental study of
pulsating flow in a close-coupled catalyst manifold on a charged motored
engine using oscillating hot-wire anemometry. SAE Paper, 2006-01-0623,
2006.
[86] T. Persoons, A. Hoefnagels, and E. Van den Bulck. Experimental vali-
dation of the addition principle for pulsating flow in close-coupled cat-
alyst manifolds. J. Fluids Eng.-Trans. ASME, 128(4):656–670, 2006.
http://dx.doi.org/10.1115/1.2201646.
[87] T. Persoons, E. Van den Bulck, and S. Fausto. Study of pulsating flow in a
close-coupled catalyst manifold. In FISITA World Automotive Congress,
23-27 May 2004, Barcelona, Spain, 2004.
[88] T. Persoons, E. Van den Bulck, and S. Fausto. Study of pulsating flow
in close-coupled catalyst manifolds using phase-locked hot-wire anemom-
etry. Exp. Fluids, 36(2):217–232, 2004. http://dx.doi.org/10.1007/s00348-
003-0683-0.
[89] R. B. Rask. Comparison of window, smoothed-ensemble and cycle-by-

cycle data reduction techniques for lda measurements of in-cylinder veloc-
ity. In T. Morel et al., editors, Fluid Mechanics of Combustion Systems,
pages 11–20. ASME, 1981.
[90] Y. Rubner, C. Tomasi, and L. J. Guibas. A metric for distributions
with applications to image databases. In Proc. 1998 IEEE Int. Conf.
Computer Vision, pages 59–66, Bombay, INDIA, 1998.
[91] R. K. Shah. A correlation for laminar hydrodynamic entry length solu-
tions for circular and noncircular ducts. J. Fluids Eng.-Trans. ASME,
100(2):177–179, 1978.
[92] R. K. Shah and A. L. London. Laminar flow forced convection in ducts.
In R. F. Irvine and J. P. Hartnett, editors, Advances in Heat Transfer.
Academic Press, New York, 1978.
[93] G. A. Sod. A survey of several finite-difference methods for systems of
nonlinear hyperbolic conservation laws. J. Comput. Phys., 27(1):1–31,
1978.
[94] M. R. Spiegel. Schaum’s outline series: Theory and problems of statistics.
McGraw-Hill, New York, 1972.
310 Bibliography
[95] M. D. Springer. The algebra of random variables. John Wiley & Sons,
New York, 1979.
[96] R. Stone. Introduction to internal combustion engines. MacMillan Press,

London, 1999.
[97] B. E. Thompson and J. H. Whitelaw. Flying hot-wire anemometry. Exp.

Fluids, 2(1):47–55, 1984.
[98] W. G. Tiederman, R. M. Privette, and W. M. Phillips. Cycle-to-cycle

variation effects on turbulent shear stress measurements in pulsatile flows.
Exp. Fluids, 6(4):265–272, 1988.
[99] D. N. Tsinoglou and G. C. Koltsakis. Influence of pulsating flow on

close-coupled catalyst performance. J. Eng. Gas. Turbines Power-Trans.
ASME, 127(3):676–682, 2005.
[100] E. Van den Bulck. Personal communication: Correlations for the con-
traction and expansion pressure drop coefficients, based on Kays and
London [57]. 2004.
[101] M. Vandevoorde. Model voor de thermodynamische en gasdynamische

cyclus van drukgevulde gasmotoren. PhD thesis, Universiteit Gent, Gent,
B, 1998.
[102] M. Vandevoorde, R. Sierens, and E. Dick. Validation of a new tvd scheme

against measured pressure waves in the inlet- and exhaust system of
a single cylinder engine. J. Eng. Gas. Turbines Power-Trans. ASME,
122(4):533–540, 2000.
[103] M. Vandevoorde, J. Vierendeels, E. Dick, and R. Sierens. A new total

variation diminishing scheme for the calculation of one-dimensional flow
in inlet and exhaust pipes of internal combustion engines. Proc. IMechE
D-J. Automob. Eng., 212(5):437–448, 1998.
[104] S. Voeltz, A. Kuphal, S. Lieske, and A. Fritz. Der Abgaskrümmer-

Vorkatalysator für den neuen 1.0l- und 1.4-l-Motoren von Volkswagen.
Motortechnische Zeitschrift, 60(7):462–468, 1999.
[105] R. A. Wallis. Axial flow ducts and fans. John Wiley & Sons, 1983.
[106] J. H. Watmuff, A. E. Perry, and M. S. Chong. A flying hot-wire system.

Exp. Fluids, 1(2):63–71, 1983.
[107] N. Watson and M. S. Janota. Turbocharging the internal combustion

engine. The MacMillan Press, London, 1982.
[108] H. Weltens, H. Bressler, F. Terres, H. Neumaier, and D. Rammoser. Op-

timisation of catalytic converter gas flow distribution by CFD prediction.
SAE Paper, 930780, 1993.
Bibliography 311
[109] D. W. Wendland and W. R. Matthes. Visualization of automotive cat-

alytic converter internal flows. SAE Paper, 861554, 1986.
[110] D. W. Wendland, P. L. Sorell, and J. E. Kreucher. Sources of monolith
catalytic converter pressure loss. SAE Paper, 912372, 1991.
[111] J. Westerweel. Fundamentals of digital particle image velocimetry. Meas.
Sci. Technol., 8(12):1379–1392, 1997.
[112] J. Wiehl and C. D. Vogt. Keramische Ultradünnwandsubstrate für mod-

erne Katalysatoren. Motortechnische Zeitschrift, 64(2):112–119, 2003.
[113] G. Woschni. A universally applicable equation for instantaneous heat
transfer coefficient in the internal combustion engine. SAE Paper, 670931,
1967.
[114] F. Zhao, L. Li, X. Xie, and M. Lai. An experimental study of the flow
structure inside the catalytic converter of a gasoline engine. SAE Paper,
950784, 1995.
Curriculum vitae
Tim Persoons was born in Leuven (Belgium) on 21 Nov. 1976. He graduated

as Burgerlijk Werktuigkundig-Elektrotechnisch Ingenieur, richting Mechanica
(Mechanical Engineer), at the Katholieke Universiteit Leuven in 1999.
From Oct. 1999 till Sept. 2001, Tim was employed as research assistant
at the dept. Mechanical Engineering (K.U.Leuven), working on model-based
control of internal combustion engines, under supervision of Prof. dr. ir. Eric
Van den Bulck. During that time, he focused on the implementation of a
simulated drive train with manual transmission and virtual driver on a dynamic
engine test stand. The research dealt with the design and optimization of a
feedforward/feedback controller for automated homologation test cycle driving
of a combustion engine in the virtual vehicle environment, created by a dynamic
engine test stand. Currently, Tim is involved in comparing a backstepping
non-linear controller and a proportional integral derivative feedback controller
for combustion engine speed control, in cooperation with Prof. dr. ir. Eric
Van den Bulck and Prof. dr. ir. Jan Swevers (K.U.Leuven, dept. Mechanical
Engineering).
In Oct. 2001, Tim began work on an experimental study of pulsating flow
in automotive exhaust manifolds with close-coupled catalyst, under supervi-
sion of Prof. dr. ir. Eric Van den Bulck. The study was part of a research
project, partially funded by IWT-Vlaanderen [2], in cooperation with the re-
search department of Bosal International, Lummen (B), a Belgian-Dutch based
Tier 1 manufacturer of automotive exhaust systems. From Oct. 2001 to the
present, Tim focuses on experimental fluid dynamics in modern compact au-
tomotive exhaust systems with close-coupled catalysts. An oscillating hot-wire
anemometer has been designed, calibrated and applied for bidirectional velocity
measurements in exhaust systems, featuring sufficiently high spatial and tem-
poral resolution for validation of the computational fluid dynamics techniques
at Bosal Research.
Based partly on the ongoing doctoral work, a new IWT project [3] started in
2003. Tim was involved in the experimental study of flow, wall shear stress and
pressure drop in pipes with interacting bends, typical of compact automotive
exhaust systems. At the same time, work began to investigate the relationship
between inclined, swirling flow and the inlet pressure loss, and its influence on
313
314 Curriculum vitae
the flow distribution in catalyst substrates.

From 1999 to 2005, Tim taught the combustion engine lab sessions for
course H263 Voertuig- en Vliegtuigpropulsie. Since 1999, he has coached eight
Master’s theses of Mechanical Engineering students, in the fields of combustion
engine control and experimental fluid dynamics.
List of publications
International journals with review
(1) T. Persoons, A. Hoefnagels, and E. Van den Bulck. Calibration of an
oscillating hot-wire anemometer for bidirectional velocity measurements.
Exp. Fluids, 40(4):555–567, 2006. http://dx.doi.org/10.1007/s00348-005-0095-4
(2) T. Persoons, A. Hoefnagels, and E. Van den Bulck. Experimental vali-
dation of the addition principle for pulsating flow in close-coupled cat-
alyst manifolds. J. Fluids Eng.-Trans. ASME, 128(4):656–670, 2006.
http://dx.doi.org/10.1115/1.2201646
(3) T. Persoons, A. Hoefnagels, and E. Van den Bulck. Experimental study

of flow dynamics in close-coupled catalyst manifolds. Int. J. Engine Res.
(in press).
(4) T. Persoons, E. Van den Bulck, and S. Fausto. Study of pulsating flow in
close-coupled catalyst manifolds using phase-locked hot-wire anemome-
try. Exp. Fluids, 36(2):217–232, 2004. http://dx.doi.org/10.1007/s00348-003-
0683-0
(5) K. Clement-Nyns, T. Persoons, and E. Van den Bulck. Pressure drop of

pipe sections with single and double 90 degree bends. J. Fluids Eng.-
Trans. ASME (in review)
International conferences with review

(1) T. Persoons, A. Hoefnagels, and E. Van den Bulck. Experimental study
of pulsating flow in a close-coupled catalyst manifold on a charged mo-
tored engine using oscillating hot-wire anemometry. In SAE 2006 World
Congress, 3-6 Apr. 2006, Detroit (MI), USA, 2006.
(2) T. Persoons, E. Van den Bulck, and S. Fausto. Study of pulsating flow in a
close-coupled catalyst manifold. In FISITA World Automotive Congress,
23-27 May 2004, Barcelona, Spain, F2004-F436, 2004.
(3) K. Nevelsteen, T. Persoons, and M. Baelmans. Heat transfer coefficients
of forced convection cooled printed circuit boards. In Proc. 6th Int.
Workshop on Thermal Investigations of IC’s and Systems, Therminic,
Sep 2000, Budapest, Hungary, pp. 177–182, 2000.
“I worry about ridiculous things, you know, how does a guy who drives a
snowplough get to work in the morning. . . That can keep me awake for
days. . . ”
Billy Connolly (Scottish comedian, ◦ 1942)
Index
addition principle, 103 CFL number, 191

definition, 104 discretization, 269
industrial implications, 17, 229 equations, 267
literature on, 147 exhaust system, 188–192
mixing analogy, 149 filling-and-emptying, 45
relation to Helmholtz resonance, flow rate measurement, 54
228 cylinder pressure based, 42, 59,
validation 182
approach, 106 laminar flow element, 42, 56, 182
correlations, 144–147 orifice, 37, 54
interpretation, 144, 225 in pulsating flow, 55
flow uniformity, 107
blow-by leakage, 49, 182, 210 quantifying, 107–109
related to diffuser flow, 161
catalyst, 241 related to pressure drop, 10
chemistry, 6, 203, 249
deactivation, 4, 202 Helmholtz resonance, 204, 213
design criteria, 8 analytical model, 205
geometry, 3, 241 frequency range, 210
pressure loss, 8, 11, 242 relation to addition principle, 228
CFD, problems with, 17 hot-wire anemometry, 259
collector efficiency, 150, 229 directional ambiguity, 72–76
principle, 76
exhaust manifold, 10 problems with, 45, 72, 210
frequency response, 213
literature on flow in, 19–26 laser Doppler anemometry, 260
measurement location, 35 directional ambiguity, 261
pressure loss, 10, 11 principle, 260
specifications, 33 problems with, 263
thermal load, 16 reference for OHW, 89
seeding, 260
flow dynamics, 155
experimental results, 156–181 oscillating hot-wire anemometer, 79
in one exhaust stroke, 162, 175 calibration, 85–97
modeling, 187, 267 flow disturbance, 93, 97
benchmarking, 273 limitations, 184, 193
316
Index 317
operation, 98 effect of cold end, 198, 201

oscillator, 80 effect of fired conditions, 197, 201
prong vibration, 75 physical relevance, 202
validation, 182
scavenging number, 122
phase-locked measurement, 61 critical value, 146, 150
cycle-resolved analysis, 64 similarity
uncertainty, 64 exhaust stroke, 41, 44, 53
pulsating flow rigs, 33, 36–44 thermodynamic analysis, 46,
operation, 41, 170 253
pulsator devices, 37, 159, 204 velocity distributions, 110–122
reverse flow, 182, 193 valve overlap, 35, 42, 49, 168

Persoons

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Persoons

Uploaded by

Copyright:

Available Formats

KATHOLIEKE UNIVERSITEIT LEUVEN

EXPERIMENTAL FLOW DYNAMICS

Jury: Proefschrift voorgedragen tot

U.D.C. 621.43 Augustus 2006

Wettelijke Depot: D/2006/7515/47

Tim Heverlee, augustus 2006

A Cross-sectional area [m2 ]

α OHW probe velocity tolerance factor [-]

List of symbols vii

1 Introduction: Exhaust systems 1

3 Oscillating hot-wire anemometer (OHW) 71

4 Addition principle 103

Shape similarity measure . . . . . . . . . . . . . . . . . 110

5 Flow dynamics 155

6 Design considerations 225

A Catalyst substrate flow 241

B Exhaust stroke flow similarity 253

C Velocity measurement techniques 259

D Modeling one-dimensional gas dynamics 267

Nederlandse samenvatting 287

Curriculum vitae 313

List of publications 314

1.1 Legal background

1 For indirect and direct injection diesel engines, respectively

exit cone underbody

hot end cold end

Figure 1.1 – Exhaust system nomenclature

1.2 Aspects of manifold design

Nomenclature In this thesis, the term manifold denotes the combi-

Figure 1.2 – Catalyst substrates: (left) metal and (right) ceramic

Figure 1.3 – Ceramic substrate catalyst nomenclature

Table 1.2 – Ceramic catalyst substrate dimensions

Neyestanaki et al. [80] review deactivation in automotive catalysts.

Velocity U/Um (-)

concentration. Appendix A.3 discusses the different approaches for modeling

ensures an adequate conversion level. Consequently, conversion is lower in high

Table 1.3 – Catalyst design criteria

where ηU = 1 corresponds to a perfectly uniform velocity distribution and when

(i) Precious metal cost ∝ Vcoat ∝ ε A L/d

Figure 1.6 – Connection of a runner into the diffuser

• Runners and diffuser — The runner pressure drop is estimated

Figure 1.8 – Maldistribution versus smoothing index (Source: [109, 110])

based on two closely spaced 90 ◦ bends, forming a 90 ◦ out-of-plane

• Catalyst — The pressure drop in the catalyst is estimated assuming

Engine speed N 2000 4000 6000 rpm

The contribution to the total pressure drop of each component is of com-

Background: Pressure wave tuning The initial phase of the ex-

Such configuration is denoted an ‘N-in-1’ exhaust, where N is the number

Heywood [47] provides general background information on the gas exchange

1.3 Motivation for this thesis

Addition principle The time-averaged catalyst velocity distri-

1.4 Overview of manifold flow research

1.4.1 Flow distribution

Isothermal flow rigs

Figure 1.11 – Comparison of non- Figure 1.12 – Non-uniformity index

Figure 22 Experimental velocity profiles for the Standard substrate

commercially available CFD packages. The porous volume model is used to

Isochoric flow rigs

Background The exhaust stroke of an internal combustion engine, ei-

1.4.2 Flow dynamics

Figure 1.16 – Experimental (left)

results obtained on an isothermal flow rig. Although LDA is capable of mea-

Isochoric flow rigs: Fired