AE4115:

Experimental Simulations
Reader

Lectures by G. Eitelberg
edited by T. van Pelt & T. Sinnige & E.M. Raijmakers

May 23, 2021

[Page intentionally left blank]
Contents
1 Why experimental simulations and what is it? 1

Why experimental simulations? 1

1.1 Drivers for aeronautical infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Aeronautical industry and development as drivers for the contents of the lecture series 2
1.1.2 Research as driver for experimentation in aeronautics . . . . . . . . . . . . . . . . . . 5
1.2 Course structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 A word of caution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Similarities and scaling 7

2.1 Fractional analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Fractions of forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Fractions of energy fluxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Dimensional analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Independent dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Physical relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Boundary layer: Stokes flat plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Wind tunnels 24
3.1 Different types of wind tunnels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1 Wind-tunnel configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.2 Low-speed wind tunnels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.3 High-speed subsonic and transonic wind tunnels . . . . . . . . . . . . . . . . . . . . . 31
3.1.4 Supersonic & hypersonic wind tunnels . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Wind-tunnel corrections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.1 Blockage effect explained from conservation principles . . . . . . . . . . . . . . . . . . 38
3.2.2 Blockage in open test sections due to pressure effects . . . . . . . . . . . . . . . . . . 40
3.2.3 Potential flow approximation applied to wind tunnel interference corrections . . . . . . 41
3.2.4 Thrust and Drag corrections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4 Reynolds-number effects 57
4.1 Reynolds number effects on airfoils and high aspect ratio wings . . . . . . . . . . . . . . . . . 60
4.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.1.2 Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.1.3 Flow separation as a Reynolds-number effect . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Reynolds-number effect testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.2 Simulating the Reynolds number by the technique of tripping . . . . . . . . . . . . . . 66
4.2.3 Effect of tripping at different locations along the chord of the airfoil . . . . . . . . . . . 67
4.2.4 Conclusion for airfoil and wing testing . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Highly swept wings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Slender bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.5 Pseudo Reynolds-number effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.6 Summary and recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5 Engine integration 78
5.1 Thrust generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

G. Eitelberg 3 Delft University of Technology

 5.1.1 Actuator disk model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.1.2 Correcting for propeller slipstream effects for an isolated propeller . . . . . . . . . . . 82
5.2 Propulsion interference effects on an aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3 Engine simulation requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.4 Simulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.5 Thrust book-keeping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6 Quantifying the TPS performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.7 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

6 Noise 95
6.1 Human hearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2 Acoustics as a wave phenomenon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.3 Scaling of sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.4 Measurement devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.4.1 Presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.5 Beam forming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.5.1 Delay & sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.5.2 Conventional beam forming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.6 Array performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.7 Wind tunnel noise measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.7.1 Moving sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

7 Design of experiments 110

7.1 Factorial designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.1 SWIM model experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

8 Balances 118
1 Why experimental simulations and what
is it?
The relatively brief history of aeronautical testing started with both experiments for phenomenological re-
search (L. Prandtl (1904)) as well as for product development (O. and W. Wright (1903)). The distinction
between the former (scientific curiosity driven experiment) and the latter (product development) provides
the justification for the title of the course : Experimental Simulations. It is about experimentation for science
and about simulation of flight performance in experimental facilities.

Figure 1.1: L. Prandtl with a water tunnel; visualization of the boundary layer.

The straightforward, curiosity driven testing requires the mastery of experimental techniques; it requires
understanding of the signal processing, resolution and measurement capacity.
The testing in a laboratory for the purpose of product development - for the purpose of obtaining from a
laboratory experiment information about the behaviour of a product in the real usage - requires additional
skills of designing and interpreting an experiment. The experimentation performed in aeronautical wind
tunnels for the purpose of predicting aircraft properties and behaviour, falls into this latter category. The
experiments are designed and interpreted as to represent aircraft in a particular flight situation - the ex-
periment simulates flight. This particular attention to the transferabilty of the experiment to interpret flight
performance, the required scaling and the simulation of air vehicle components to represent reality provide
the background to the title of the lecture series about experimental simulations. In this series of lectures the
focus is on aeronautical applications, although an occasional reference to ground vehicles or wind energy
might provide additional illustrations.
The first primitive testing facilities used by L. Prandtl and the brothers Wright were soon followed by more
sophisticated developments in a number of countries aspiring to establish their national capacity to fly.
Especially after the 2nd World War, in the period of striving for national independence in technologies, most
powers established their own research infrastructure for aeronautical testing. The foremost powers were
the US and the SU. The European nations F, UK, D, and NL were also aspiring to maintain independence
in this strategic endeavour.

G. Eitelberg 1 Delft University of Technology

 Figure 1.2: Product development by one of the Wright Bros. (Photo from the Smithsonian Institute)

1.1 Drivers for aeronautical infrastructure

In terms of testing capabilities currently four separate drivers for maintaining experimental simulation capa-
bility can be identified:
• Industry requirements
– Design verification for performance
– Handling qualities
– Future products as a response to

* competitive pressures and

* regulatory requirements
• Government requirements
– Access to space and distance
– Exploration
– Projection of power
• Research institutions
– Fundamental challenges
• Operators of experimental facilities
– Upgrades/improvements
– Skills base
In the following paragraphs some further elaboration of the four items listed above is offered.

1.1.1 Aeronautical industry and development as drivers for the contents of the
lecture series
The challenge faced by the aeronautical industry in the aircraft development phase is achieving the required
performance targets including stability and controllability of aircraft while minimizing the fuel consumption.

G. Eitelberg 2 Delft University of Technology

This requirement is derived from the operational cost consideration as well as from the regulatory require-
ment of minimizing the production of CO2 . The performance is being estimated and pursued already in
the preliminary design phase and verification of the design takes place in scaled testing in experimental
facilities. Even more challenging will be the development of alternative configurations possibly required
for satisfying increased environmental requirements. In these cases no empirical history exists to provide
confidence in the design rules.
For the industry, in the foreseeable future, the most important measurements regarding the aircraft perfor-
mance remain the measurements of forces and moments. In order to characterize the stability and control
of an aircraft, the balance measurements maintain their role as the main mode of aeronautical testing. The
requirements toward wind tunnel operators are for capability of high quality (resolution and accuracy) perfor-
mance measurements with the help of wind tunnel balance technology and good correction methodologies
to account for the influence of the measurement setup. The corrections concern both the measurement
instrument itself (the instrument interferes with the measurement) as well as the difference between the
infinite amounts of free air in flight and the limited amounts of air in a wind tunnel circuit.
Since the models of aircraft in the wind tunnels are scaled down to fit into the tunnels, the scalability of the
results has to be affirmed. This is usually discussed as the accounting for the Reynolds number effects,
even though there are many more scaling phenomena to be considered. Some of the scaling challenges
can be overcome with the help of computational fluid dynamics CFD. Although the numerical calculations
have made great strides in providing information about flow phenomena, they have achieved their current
status by employing sophisticated semi-empirical models to describe the phenomena of turbulence and
separation. A direct numerical calculation of the turbulence is still far from reality for large complicated
objects. Therefore the modelled phenomena are still subject to experimental verification, usually in a wind
tunnel. The uncertainty of predictive capability in relation to high Reynolds number effects is valid for steady
applications but is even more challenging for dynamic situations in high speed, like in maneuvers or for the
rotor blades in open rotor flow.
The exhaust of CO2 and other emissions from aircraft depends on the drag and on the weight of the aircraft
and on the performance of the engines when mounted on the aircraft. While the drag of an aircraft in cruise
can be rather reliably predicted in the design phase already, the aerodynamic limitations are reached in
the phase of the flight taking place in the vicinity of the airports – the approach and take-off phases. The
structural weight of an aircraft is predominantly determined by the aerodynamic loads it is experiencing in
the high lift conditions during take-off at Maximum Take-off-Weight (MTOW) conditions. During landing the
aircraft operates close to stall in high lift conditions. The high lift is generated with the help of the so called
movables: small (relative to the main wing) additional wing elements such as slats, flaps, droop noses etc.,
which are structural parts hidden from the flow during the cruise flight. The prediction of the performance of
those small high lift devices, because they are deployed in difficult flow conditions, is still highly uncertain
in the design phase and relies on extensive verification in experiments, both in wind tunnel (scaled) and in
the much more expensive flight testing. Additionally, the interference between the propulsive devices and
the high lift devices may become particularly critical in those conditions.
Since the high lift devices are small but the prediction of their performance has a large impact on the de-
sign, adequate scaling combined with adequate precision of the shape reproduction is only achieved in
large wind tunnel models. This is due to the manufacturing tolerances. The requirement of monitoring the
shape of the high lift elements in wind tunnel experiments can be satisfied by using non-intrusive optical
techniques, covered in another course. Nevertheless, it is important to recognize that the scaling of the
elastic behavior is an additional challenge in addition to the aerodynamic scaling. Both industrial as well
as research testing will benefit from the capability of high Reynolds number testing, especially since for the
foreseeable future the modeling and thus prediction capability for transition and separation remain depen-
dent upon experimental benchmarks. The Reynolds number effects in aeronautical testing are frequently
called upon to explain any results that do not meet expectations. Therefore it is important to understand
these effects and to differentiate between the direct, the indirect and the pseudo Reynolds number effects,
and realize that not every discrepancy can be explained away as a Reynolds number effect. This will re-
main a requirement despite the rapid development of numerical capacity since only the DNS simulation can
reliably predict turbulent phenomena (Dillmann 2012) like transition and separation.

G. Eitelberg 3 Delft University of Technology

A similar discussion can be led regarding the predictive capability of numerical methods concerning the
unsteady aerodynamics, where the vorticity production in an unsteady case is different from the equivalent
averaged steady condition. The dynamic testing cannot be replaced by quasi-steady simulation, especially
in high speed conditions, since the share of the production of vorticity ω through unsteady motion charac-
terized by an appropriately formulated Strouhal number Str, increases with the square of the Mach number
Ma:
ωunsteady /ωsteady = const × M a2 × Str. (1.1)

This scaling rule can be derived from the appropriate vorticity production equations, see relevant literature
(eg. Lighthill [15]). Here the Strouhal number, Str, represents the unsteadiness of the flow/flight. This
scaling rule shows that at high Mach numbers, much more effort goes into maneuvers than at low low Mach
numbers, even if the Reynolds number stays identical, i.e. the viscous effect is amplified.
An additional component in experimental simulation is the need to consider not only the unsteady aerody-
namics, but also the structural properties of the vehicle or experimental model as well. This is covered by
the term testing for aero-elastics.
From a technological point of view, it is important that the future developments in aeronautical materials
leading to changed behavior of the aircraft structures under wind loads can be properly simulated also in
the cruise conditions. Here the capability again will have to offer a satisfactory compromise between the
aerodynamic scaling of the flow and the strength/elastic scaling of the structures.
Further, there are a range of issues related to the integration of propulsion which will require high quality
experimental studies for clarification. A standard issue for the aircraft designer is the issue of choosing
the proper placing of high by-pass ratio engines for clean undisturbed inlet flow. The comparison of the
performance of the isolated engine with the performance of the installed engine can currently only be
reliably performed in well-designed experiments. The experimental design here is mainly concerned with
providing simulators for the aircraft engines which are representative of the real engine performance. The
requirement for being representative is fulfilled by adequate thrust and power scaling, in addition to the
geometrical scaling.
Proper thrust book-keeping in the experiment requires a well-established calibration procedure, in order to
resolve the interference effects which are significant for flight, but small when compared with the thrust or
drag alone.
The issues of ingestion of the boundary layer, pylon- or wing- wake and of ground or fuselage vortices in
unconventional configurations can be unsteady and highly unpredictable. So will the combination of thrust
generation and ground proximity. In the integration of propulsion with the airframe, conventional definitions
of efficiency and apportioning of installation losses to either the airframe or engine tend to break down.
Therefore the experiments have to be thoroughly designed and understood.
Testing for helicopter or rotary wing aircraft performance can be considered as a subset of propulsion inte-
gration testing with strong unsteady elements in it. Both the performance as well as the acoustic emissions
of the rotating parts are of significant value in the chain of design evaluation.
The acoustics requirements for airframe as well as propulsor noise will continue to accompany the aeronau-
tical experiments for the foreseeable future. Since the noise generation mechanism is a complex process
of interaction of the flow with the submerged solid surface, the predictive capability of numerical methods
encounters its limitations in complex configurations. Aeroacoustic simulation experiments rely on the capa-
bility to design situations in controlled environment which are representative of the real-life situations. This
notwithstanding the need for phenomenological aero-acoustic experimentation for the understanding of the
principles of sound generation and propagation through shear- and boundary layers.
There are only very few experimental facilities available in Europe and world-wide, where properly scaled
aero-acoustic simulation can be achieved. The evidence obtained from comparing the controlled environ-
ment with flight testing points toward the possibility to shift part of the airframe noise certification process
from flight testing into wind tunnels. This shift is justifiable on account of the wind tunnels’ better controllabil-
ity and verifiability of conditions. In the aeroacoustic testing methodology, the measurement of the noise at

G. Eitelberg 4 Delft University of Technology

the location of the microphones, representing the observer, needs to be complemented by the identification
of the sources of the noise on the aircraft (model). The principles of acoustic arrays need to be understood,
in order to judge this capability.
Since the aerodynamic experiments required in the product development process are complex and use
scaled models, a thorough understanding of scaling and similarities is a necessary prerequisite for a good
design of experiments. Only after understanding the scaling laws, will the next step of using good statistics
for experimental design and analysis yield the maximum benefit in transferring results from an experiment
to the real application. This need to understand the scaling laws will be the continuing thread through all of
the chapters of the lecture series that will follow.

1.1.2 Research as driver for experimentation in aeronautics

In general, it can be claimed, that without the research, the development steps will remain constrained
by the existing knowledge. On the other hand, the research testing quite often does not have to simulate
the flight phenomenon; it is directly measuring the phenomenon in the experiment. In the context of the
present course, any experimental technique by itself is not dealt with as a primary topic. The techniques
are only introduced inasmuch they are required to accompany the experiment dedicated to simulation of
flight phenomena.
So, for example the methods to measure sound are introduced, since the sound generation and sound
propagation are important topics in the design, certification and operation of aircraft. Although there are
specialist courses going deeper into aero-acoustics, it is the experience that many of the listeners have not
taken these electives.
On the other hand, the tools that are so general that they have been covered in other parts of the study,
like measurement techniques for pressure, temperature and velocity, widely used in the context of the
phenomenological research, are not discussed in the course of the present lectures. It is assumed that the
participants are familiar with the measurement equipment. Those who need a recap can have a look at the
work of Barlow, Rae & Pope [14].

1.2 Course structure

1. Scaling and similarities
It will be shown, that there are several ways of formulating scaling requirements. The boundaries of
similarities are sought and some universally applied approaches to interpreting experimental results
will be discussed. As part of the exercise the use of dimensional analysis for identifying the rele-
vant parameters for planning experimental studies of phenomena will be illustrated. The dimensional
analysis will also be used to discuss the concept and fundamental characteristics of boundary layers.
Since, in practice, the most widely used aerodynamic similarity is the Reynolds number, a separate
part of the lectures will cover the Reynolds number effects in wind tunnel testing.
2. Testing requirements and design of experiments
The principles of multivariable design of experiments using the fundamentals of statistics are dealt in
this chapter together with the example of calibration of wind tunnel balances.
3. Wind tunnels and wall corrections
In this chapter the governing principles and the scope of wind tunnel testing will be covered. The
limitations of testing within the confines of a wind tunnel are introduced and the vocabulary carefully
specified (speed vs. Mach number,. . . ). The principles of classical correction methodology for lift and
drag will be introduced. The corrections for power effects will be left for discussion of the propulsion
testing.
4. Reynolds-number and other scaling effects
The distinction between direct and indirect Reynolds number effects will be discussed. Additionally the
pitfalls of the pseudo-Reynolds number effects will be identified. A body of evidence from experiments

G. Eitelberg 5 Delft University of Technology

 in wind tunnels will be used to demonstrate the capability of aerodynamic experiments to provide
flight-relevant data despite the shortfall of the Reynolds number duplication capability.
5. Engine integration
The topic starts with the principal discussion of generating thrust force in aeronautics, independent-
ly of the direct method employed. After that the differences between the shrouded and unshrouded
propulsion (i.e. the turbofan and the turboprop in today’s applications) will be discussed when need-
ed. For the turbofan engine simulators the thrust book-keeping and the calibration procedures are
introduced. The approximation of the actuator disc for propulsors will be introduced. The discussion
of wind tunnel correction for power effects will be picked up and discussed as a complementary topic
to the classical corrections introduced in Chapter 2.
6. Testing for acoustic effects
Although this lecture series is too short to provide a comprehensive course in acoustics, some spe-
cific properties of testing in the wind tunnel environment are discussed. In order to describe the
requirements, the characteristics of noise sources in general and sources on aircraft in particular are
analyzed. The specific properties of microphones and arrays of microphones used in wind tunnel
environment will be discussed.
7. Lab exercise
The laboratory practical with a model of a propeller aircraft will form the culmination of the course.

1.3 A word of caution

Even if you go through the lecture series and learn all the tricks of trade for experimental simulation, one
concern remains. It is the concern about the future stability of the development capacity. There is a tendency
of educational institutions to save the cost of education in experimental facilities under the impression that
data manipulation techniques will do with numerical data without the generation of experimental data. This
could become a self-fulfilling-prophecy; after all, what we do not learn at the higher educational institution
will not enter professional practice either. The consequence of such a development will be the narrowing of
future development options, since designs can only be based on reliably verified data.
It would be beneficial if universities could count on a continuous availability of a mechanism of access with-
out detraction from the otherwise available research budget. As elements of retaining the usefulness and a
globally competitive position of wind tunnels (and also of the European industry) the following developments
are required:
• Further deepening of specialization
• Involvement and commitment by the industry
• Continuous research involvement (e.g. through an institutionalized TNA)
• Continuous upgrades
Without this, the skill base will erode and this lack of skill base will become a significant hindrance to the
introduction of novel products.
Last, but not least - this document or chapters hereof are not meant as a textbook. It is not a quotable pub-
lication, and the information herein is not always properly identified either. Some of the material is the result
of an haphazard collection of evidence to support the thought processes gone into the preparation of the
lectures. The reader is recommended to perform his/her own literature review when seriously confronted
with any of the issues raised in this collection of lectures.

G. Eitelberg 6 Delft University of Technology

2 Similarities and scaling
Almost all experiments in laboratories for engineering purposes are based on the assumption that real life
can be simulated in a laboratory. This is usually the case, as the characteristics that are not simulated in
the laboratory can often be ignored, although there are some exceptions in various verification processes
such as:
• Road tests
• Flight tests
• Engines in laboratories
When designs are to be verified in experimental facilities/laboratories, a systematic approach is required
in order to ascertain and even quantify similarities of the verification to the real-life situation. A critical
part of this is the capability to identify governing parameters for the situation being simulated. In our current
methodology, this identification relies on various approaches to the laws of conservation (mass, momentum,
energy) as well as engineering intuition (experience). This chapter won’t be about how to reduce complex
systems to systems described by their dominant parameters as this is dealt with in system theory. This
chapter will be dealing in depth with similarity and scaling, helping to identify the dominant parameters of
the system, without re-formulating the system description. This chapter will provide the tools to answer the
question, when will a laboratory experiment be similar to real life and how an experiment can be scaled in
such a way as to obtain similarity. The possible approaches are:
1. Write down relevant phenomena of a similar character, express those through the relevant scaling
variables and analyse their relevant orders of magnitude.
2. Write down a list of relevant variables, perform dimensional analysis and obtain a reduced set of
governing parameters. When possible, obtain the governing equations.
3. Write down governing equations, non-dimensionalise those and select the right parameters.
Also combinations of the above three methods can be used to answer the questions of similarity and scaling.
The main methods which are used in correspondence with the above approaches are identified as:
1. Method of fractional analysis.
2. Method of dimensional analysis.
3. Method of differential equations.
In our treatment, the method of dimensional analysis will occupy a central role. The method is based on
the theorem, called the Pi-theorem, posted by E.Buckingham [6] and it has been independently proven by
a number of mathematicians. The rigorous proof goes beyond the scope of this reader and it can be found
in the textbooks of J.Zierep [27] and G.I. Barenblatt [3] used as reference for this reader. The Bucking-
ham Pi-theorem will be treated as the essential backbone for exploratory testing planning and analysis of
phenomena. All the other methods will, when the governing parameters are appropriately chosen, lead to
identical results, but might require more up-front knowledge. In the following the methods will be explained
and the approaches demonstrated.

2.1 Fractional analysis

In the method of fractional analysis (fractions are ratios of two governing parameters of the same catego-
ry), ratios of like characteristics of a phenomenon, such as forces, energies etc., are used to obtain non-
dimensional numbers that describe different phenomena to be simulated within the laboratory. The method
of fractional analysis is usually represented in fractions of forces and fractions of energy fluxes. Fractions

G. Eitelberg 7 Delft University of Technology

of any other characteristics are also permissible, as long as they are used in a systematic approach. We
shall start with looking at fractions of forces, as they lend themselves to the most intuitive usage.

2.1.1 Fractions of forces

When scaling forces and defining their ratios for the fractional analysis, it is possible to perform an order of
magnitude analysis of those ratios to obtain the dominant ones. Familiarity with the principles of conserva-
tion of momentum and its differential formulation in at least one dimension is assumed in the discussion.
Further, the existence of a characteristic value for the respective parameter (e.g. U for the velocity and L
for a length etc.) is taken as a given. The characteristic values are denoted in upper case, the flow field
variables are denoted in lower case letters. The material properties which remain as usual, can be distin-
guished by indices. Forces in fluid dynamics can be normalized to those applying per unit mass and in such
a way be reduced to the acceleration of that unit mass. In this way, the dimensions that we shall compare
are those of accelerations. Some of the forces, relevant in dealing with fluid problems, are listed below.
• Inertia
a Steady component of the inertia force.

∂u U2 hmi
FI ∝ u ∝ (2.1)
∂x L s2

b The unsteady component of the inertia force can also be considered.

∂u U hmi
FIunsteady ∝ ∝ (2.2)
∂t T s2

c In case of an axisymmetric problem, the radial inertia force (centrifugal force) can also be con-
sidered separately.

u2t hmi
FIradial ∝ ∝ Ω2 L ∝ f 2 L 2 (2.3)
r s

• Pressure
P P a m3
 
1 ∂p m
Fp ∝ ∝ = 2 (2.4)
ρ ∂x ρL kg m s

• Friction
1 ∂2u µ U hmi
Ff ∝ µ 2 ∝ (2.5)
ρ ∂x ρ L2 s2

• Gravity
hmi
FG ∝ g (2.6)
s2

• Capillarity
1 σ hmi
Fc ∝ (2.7)
ρ L2 s2

Since all scaling factors of the forces per unit mass for the five forces (and their components) shown above
are defined, it is possible to formulate their independent ratios.These ratios are usually then given a name.
This either in honour of somebody who first observed the importance of the ratio or somebody who produced
significant results in the relevant field.

G. Eitelberg 8 Delft University of Technology

 • Pressure and inertia; the Euler number:
When forming the ratio of the pressure force and the inertia force in a flow field, the so called Euler
Number, Eu, is obtained. The definition of the Euler number is shown below.
pressure force Fp p L p
Eu = → = 2
= 2 (2.8)
inertia force FI ρL u ρu
In aeronautics the pressure coefficient is used rather than the Euler number, however they relate to
each other as shown below.
2p
cp = = 2Eu (2.9)
ρu2
When two flow fields exhibit the same pressure coefficient, their ratio of the pressure force to the
inertia force is also the same. This does not express anything about the simultaneously present
viscous forces, for example.
The Euler number is also related to the Mach number through the speed of sound a2 = γ ρp . Including
this definition in eq. (2.8) the following is found.
p 1
Eu = =
ρu2 γM 2
(2.10)
2p 2
cp = 2 =
ρu γM 2
From the above equations it can be seen that the pressure, for a given inertial force, drops with the
square of the Mach number. In this context, it is interesting to discuss the model of incompressible
flow with pressure gradients. The relationship above only shows the scaling factor. In gas dynamics
the exact equations for pressure and Mach number relationship can be obtained, but this is then valid
for the cases where the exact conditions are already known.
• Inertia and viscous forces; the Reynolds number:
When forming the ratio of the (steady) inertia force to the (viscous) friction force in a flow field, the
Reynolds number, Re, is obtained. The expression for the Reynolds number is shown below.
inertia force FI u2 ρL2 ρuL uL
Re = → = = = (2.11)
friction force Ff L µu µ ν
The kinematic viscosity, ν, relates to the mean free path Λ, as follows: ν = aΛ. This relationship can
be used to define the Knudsen number, Kn, which is the ratio of the molecular mean free path length
to a representative physical length scale and is defined as follows.
M u aΛ Λ
Kn = = = (2.12)
Re a uL L
When Kn << 1 it means that the gas can be considered as a continuum and leads to M << Re.
• Inertia and gravity; the Froude number:
As above, when forming the ratio of of the inertial force to that of the force due to the gravitational field,
the Froude number, F r, is obtained. A similar number can be formed with any other force field as well.
This number plays, among others, a significant role in naval architecture as it is used to determine the
resistance of a partially submerged object moving through water.
In eq. (2.13) the relation for the Froude number is shown.
inertia force u2
Fr = → (2.13)
gravity force Lg

Most flight testing is based on Froude-scaled vehicles. This Froude number scaling becomes obvious,
when the need to balance the gravity force with the lift force is considered. This means that the prod-
uct of the Froude number and the lift coefficient has to stay constant. In the following discussion the

G. Eitelberg 9 Delft University of Technology

 similarity of the fraction (ratio) of the lift force to the inertia force can be illustrated, as it is relevant for
the discussion of scaled flight testing. Among the practitioners the ratio is known as the lift coefficient
CL .

An aircraft in level flight is kept in the air by the lift force balancing the weight of the aircraft W. In a
manner analogous to the pressure coefficient a lift coefficient can be formed to describe the similarity
of lift:

W mAC Lg
CL = 2 2 2
=2 · (2.14)
ρf u L ρf L3 u2

This shows that for similarity of level flight the relative density (expressing the hydrodynamic lift force)
of the aircraft and the Froude numbers are similarity parameters. For maneuver simulation additional
similarity parameters are formed for the maneuver accelerations compared with the gravity.
A short ship building example can also be discussed:
Intuition states that the drag of a ship is dominated by wave formation, as large drag generates large
waves in the ship’s wake and in the special case that the ship does not generate drag one would
assume that no bow and wake waves would occur. In the case that viscous forces are omitted, the
parameters to be considered are the speed of the ship, u, dimension of the ship, L, fluid density, ρ
and gravity, g.
As a dimensionless ratio is wanted to describe the phenomenon, the only possibility is that density
u2
drops out as no other terms contain mass. This ratio will result in the Froude number, F r = Lg and if
this number should be kept constant in scaled testing, the ship length should vary with the square of
the velocity.
However, if the viscous forces are to be included, the Reynolds number, Re = uL ν , should be kept
constant as well, which will mean that the length scale should vary inversely with an increase in
velocity. From this it is clear that both the Froude and Reynolds numbers cannot be kept constant
simultaneously. Therefore the product of the two is often kept constant.

u2 uL u3
F r · Re = = (2.15)
Lg ν νg

• Weber number:
The Weber number, W e, is a number often used in cases where there is an interface between two
different fluids, especially for multiphase flows with strongly curved surfaces. The Weber number is
the ratio between the inertial force and the capillarity force and the calculation is shown below.

inertia FI u2 L2 ρ u2
We = → =ρ L (2.16)
capillarity Fc L σ σ

The Weber number plays a role in small scale free surface processing, e.g. crystal growth through
melting/resolidification etc.
Apart from the the ratios above, products of the ratios can be formed that yield additional similarities, such
as the example below:
Fp FI Fp p ρuL pL
Stk = Eu · Re = · = → 2 = (2.17)
FI Ff Ff ρu µ µu

This Stokes number is used for the simulation of lubrication and the characterisation of the behaviour of
particles suspended in a fluid flow.

G. Eitelberg 10 Delft University of Technology

Question: Why do we frequently focus on the Re-number alone?

If we are interested in the ratio: pressure pressure inertia

viscous = inertia · viscous the first term is the Euler number, the second the
Reynolds number. With the results derived in this section, this ratio is then
Re
(2.18)
γM 2

Assuming a constant ratio of specific heats, for aerodynamic similarities, it seems that the Reynolds number
effects have to be corrected by the square of the Mach-number. This combination is used in the hypersonic
flow treatment frequently as the so-called viscous interaction parameter. It will be encountered in later
discussions again.
For similarities of dynamic (characterized by a time scale T , especially periodic with a period of T ) phenom-
ena, the ratio of the two inertia forces, the unsteady and the steady components, can be formed:

U L fL
2
= = Str (2.19)
T U U
where f = 1/T is a characteristic frequency. This ratio is called the Strouhal number, Str, and is abbreviated
with 3 letters in order to distinguish it from the Stanton number St. For maintaining dynamic similarity, the
model length scale has to be maintained in a linear relationship with the convective distance during a period
of oscillation. The Strouhal number similarity is a requirement both for periodic movements of a test object
as well as for simulation of noise effects in aeronautics.
For steady rotation another approach can also be adopted: The unsteady scaling can be compared with
steady angular motion scaling. For a not co-rotating observer, a rotating object might represent a periodic
motion. For a co-rotating observer the motion is steady but coupled to a radial acceleration as under
equation 2.3. Forming a ratio of the radial inertial forces with the axial inertia forces will result in a ratio:

FI,rot f 2 LL
S= = = Str2 (2.20)
FI U2

This suggests that for rotating systems, if we maintain the Strouhal number similarity, we also maintain the
ratio of the radial and axial inertial forces. On the other hand, in terms of momentum balance, it can be
concluded that the significance of the radial momentum flux grows quadratically with the reduced frequency
and thus with the unsteady loading.
In addition, examples of the periodicity of sound waves generated by rotating wings or airplane propellers
can be discussed.

2.1.2 Fractions of energy fluxes

In order to apply the method of fractional analysis on energy fluxes the energy transport equation can be
resorted to and is given below for one of the Cartesian coordinates.
 
dT ∂T ∂T ∂T ∂T 1 1
cp = cp + cp u +v +w = div(λ · ∇T ) + Φ +ėss + ėrd (2.21)
dt | {z∂t} ∂x ∂y ∂z ρ ρ
| {z } | {z } |{z}
ėunsteady ėcv ėcd ėf

The last two terms play a role for example in space re-entry, otherwise radiation and reactions within a
medium can usually be neglected. This leaves us with four steady energy scales and one unsteady scale.
The scales derived below are for 1-dimensional cases.

G. Eitelberg 11 Delft University of Technology

 • conduction

λ ∂2T
 
λ∆T J
ėcd ∝ ∝ (2.22)
ρ ∂x2 ρL2 s

The energy flux through conduction is the part that occurs without motion or exchange of mass, it is
due to the temperature gradient imposed here by the boundary conditions of the problem.
• convection

 
∂T cp u∆T J
ėcv ∝ cp u ∝ (2.23)
∂x L s

The convective part of energy flux, although requiring a temperature gradient like the conduction
as well, represents the part of energy transport achieved with the help of mass transport with a
characteristic velocity U.
• friction

2
νu2 J
  
∂u
ėf ∝ ν ∝ (2.24)
∂y L2 s

The friction energy flux is the energy dissipated due to friction. Although this occurs in all viscous
flows, it is neglected in the approximation of isothermal flows.
• sources & sinks

 
J
ėss ∝ Q̇ (2.25)
s

The sources and sinks energy flux is governed by reactions within the medium.
• unsteady

 
cp ∆T J
ėunsteady: ∝ (2.26)
t s

From these energy fluxes independent ratios can be formed and are shown below.
• Péclet number:
The Péclet number is obtained when the ratio between the conductive and convective energy transport
terms is formed. It is of importance for the scaling of energy transport within a flow field. It illustrates
the similarity (or lack thereof) of the convective and diffusive processes of transport for momentum
and internal energy.

convection ėcv cp u∆T ρL2 cp ρuL uL uL

Pe = → = = = λ = (2.27)
conduction ėcd L λ∆T λ ρc
k
p

Where k is the thermal conductivity and defined as: k = ρcλp . The Péclet number can also be calcu-
lated as the product of the Reynolds Re and Prandtl, P r, numbers.
uL ν ν
Pe = = Re = Re · P r (2.28)
k ν k

G. Eitelberg 12 Delft University of Technology

 The Prandtl number describes the ratio of the momentum transfer by viscous diffusion to heat transfer
by thermal diffusion. In other words it is the ratio between the thickness of the velocity boundary layer
and the thermal boundary layer.
Question: in a hypothetical experiment, where the space between two horizontal plates is filled
with a fluid and the upper wall is instrumented with a temperature sensor and a shear stress sensor,
we set the lower wall in motion and increase its temperature simultaneously. Which sensor provides
information first if the space is filled with liquid metal, oil, air?
• Eckert number:
The Eckert number, Ec, is obtained in the process of forming the ratio between the dissipative energy
flux and the convective energy flux.

friction ėf νu2 L 1 u2

→ = 2 = (2.29)
convection ėcv L cp u∆T Re cp ∆T

In this ratio of energy fluxes, two terms can be recognized. The first component is the inverse of the
already identified Reynolds number. The second component is a ratio of the kinetic energy in the flow
to that of a characteristic enthalpy (difference) in the flow. The latter component is called the Eckert
number.
u2 kinetic energy
Ec = = (2.30)
cp ∆T enthalpy

This Eckert number can also be represented by the Mach number:

u2 u2 T T
Ec = = = (γ − 1)M 2 (2.31)
cp ∆T cp T ∆T ∆T

When looking back to the ratio of the dissipation and convective flux it can also be represented in a
different way.

ėf Ec M2 T
= = (γ − 1) (2.32)
ėcv Re Re ∆T
2
Where M Re is referred to as the viscous interaction parameters and is often used in hypersonic flow.
We encountered the same parameter earlier when discussing the ratios of forces (momentum fluxes)
as well.
• Fourier number:
The Fourier number is obtained as the ratio between the unsteady convective energy flux and con-
ductive transport of energy.

unsteady ėunsteady cp ∆T ρL2 cp ρL2 L2

Fo = → = = = (2.33)
conduction ėcd t λ∆T λt kt

The time scale t in the Fourier number can also be represented by the frequency, f = 1t , which leads
to the following relation for the Fourier number.

L2 f
Fo = (2.34)
k
It can be seen that for similarity, the frequency needs to be scaled with L2 , when unsteady conductive
processes are considered. This similarity requirement is missing in the ratios of forces, but when just
the steady and unsteady convective transport components are considered, then the Strouhal number
will be the same. Thus the conductive and convective unsteady similarities cannot simultaneously be
fulfilled in a scaled experiment.

G. Eitelberg 13 Delft University of Technology

 • Damköhler number:
The Damköhler number relates the chemical reaction timescale (reaction rate) to the energy trans-
port rate occurring in a system and can be calculated as shown below. In other words it describes
similarities of flows with chemical reactions.

sources/sinks ėss Q̇t Q

Da = → = = (2.35)
unsteady ėunsteady cp ∆T cp ∆T

An example for the use of the Damköhler number is the distance between the shock and the object in
hypersonic flows as shown in figure 2.1.

Figure 2.1: The shock stand-off distance, D, vs. the Da-number.

2.2 Dimensional analysis

Dimensional analysis is the analysis of the relationships between different physical quantities by identify-
ing their base units, such as length, mass, time for mechanical phenomena. For thermal phenomena the
measure of temperature (Kelvin) has to be added as another base unit. Other, different base units can also
be used, such as force length and time (FLT) or whatever combination is consistent with in the physical do-
main. In the following, preference is given to the SI system, where the mechanical quantities are expressed
through the MLT combination.
L unit of length m
M unit of mass kg
T unit of time s
With these base units other units can be derived.
[v] velocity LT −1
[a] acceleration LT −2
[ρ] density M L−3
As dimensions of physical relations should be the same both on the right and left side of the equal sign it is
possible to express more difficult parameters in base units.

F = ma
(2.36)
[F ] = M LT −2

G. Eitelberg 14 Delft University of Technology

2.2.1 Independent dimensions
Quantities a1 ...ak are said to have independent dimensions if none of them has a dimension function that
can be represented as a product of the dimensions of the remaining quantities. For instance, density,
velocity and force have have independent dimensions. To prove this it is needed to express force as a
function of density and velocity as done below.
[F ] = [ρ]x [v]y

x
y (2.37)
M LT −2 = M L−3 LT −1

M :1=x
L : 1 = −3x + y (2.38)
T : −2 = −y
As can be seen these three equations are incompatible, thus the properties have independent dimensions.

2.2.2 Physical relationships

In physical studies, it is attempted to obtain relationships among the quantities that characterise the phe-
nomenon under study. This relationship will be of the form.
a = f (a1 , . . . , ak , b1 , . . . , bm ) (2.39)
Where.
• a1 , . . . , ak have independent dimensions (dimension functions)
• b1 , . . . , bm have dependent (upon ai ) dimensions (dimension functions)
• n number of governing parameters (n = k + m)
In that case.
[b1 ] = [a1 ]p1 . . . [ak ]r1
(2.40)
[bm ] = [a1 ]pm . . . [ak ]rm
And two special cases arise.
• m = 0: all governing parameters are of independent dimensions.
• k = 0: all governing parameters are dimensionless.
The dimension of phenomenon a, equation 2.39, will be in the form as seen below.
[a] = [a1 ]p . . . [ak ]r (2.41)
With the terms above, equation 2.39 can be transformed into dimensionless parameters.
a
Π= p
a1 . . . ark
(2.42)
b1 bi bi
Π1 = p 1 r 1 , . . . Πi = p i r i , . . . Πm = p m
a1 . . . ak ai . . . ak am . . . arkm
Here the exponents of the governing parameters with the independent dimensions ai are chosen such that
all the parameters Π, Πi . . . Πm are dimensionless. With this equation 2.39 can be reformulated.
f (a1 , . . . , ak , b1 , . . . , bm )
Π=
ap1 · . . . · ark
1 (2.43)
Π= p f (a1 · . . . · ak , Π1 · ap11 · . . . · apk1 , . . . , Πm · ap1m · . . . · arkm )
a1 · . . . · ark
Π = F (a1 , . . . , ak , Π1 , . . . , Πm )

G. Eitelberg 15 Delft University of Technology

In the case of parameters with independent dimensions, changing of the value of one parameter does not
affect the values of the other parameters with independent dimensions, nor does it affect the value of the
function. Then, obviously, the relationship in equation 2.43 can be expressed in terms of a function Φ of m
parameters instead of m = k + m parameters.

Π = Φ(Π1 , . . . , Πm ) (2.44)

This, non-dimensional, expression confirms the intuitive observation, that physical laws cannot depend on
the choice of units. With the derivation above the Π-theorem has just been explained. This Π-theorem
states the following:
‘A physical relationship between some dimensional quantity and several dimensional governing parameters
can be expressed as a relationship between a dimensionless parameter and several dimensionless prod-
ucts of the governing parameters. The number of dimensionless products is equal to the total number of
governing parameters less the number of governing parameters with independent dimensions’.

In other words, the Π-theorem can result in orders of magnitude reduction of evaluation effort.

2.2.3 Examples
Pendulum
Determine the time of the oscillation period, τ , of the pendulum from figure 2.2. [τ ] = T It can be seen that

Figure 2.2: Pendulum.

the governing parameters are:

• length l: [l] = L
• mass m: [m] = M
• gravity g: [g] = LT −2
These governing parameters all have independent dimensions, n = k = 3 & m = 0, which leads to.
τ
Π= 1 1 = const. (2.45)
g− 2 l 2
q
Thus τ = const. gl , where the constant can be determined from experiments. Mass is not a governing
parameter.

Shock/boundary layer interaction in transonic flow over an airfoil

Question: Will the pressure rise behind the recompression shock lead to separation?

In figure 2.3 a representation of an airfoil submersed in transonic flow is shown. The governing parameters
are the speed of sound in the boundary layer, aδ , and the flow speed at the boundary layer, uδ . Furthermore

G. Eitelberg 16 Delft University of Technology

 M >1 ← shock
δ

Figure 2.3: Boundary layer and shock interact at transonic flight conditions.

the boundary layer thickness, δ, the radius of curvature of the airfoil, R, the density of the medium, ρ, and
the dynamic viscosity of the medium, µ are also of importance. It is possible to play with the independent
variables. In this case both δ and R can be chosen as the independent parameter for the length scale and
both aδ and uδ can be chosen as the independent parameter for the speed. However once one is chosen
the other one cannot be described as an independent variable, since the dimension of the other parameter
can be expressed as a function of the dimension of the chosen independent parameter. At separation the
wall shear stress is 0, τw and can therefore be described as:

τw = f (uδ , aδ , δ, R, µ, ρ) = 0 (2.46)

The chosen independent parameters are:

• speed of sound aδ : [aδ ] = LT −1
• boundary layer thickness δ: [δ] = L
• dynamic viscosity µ: [µ] = M L−1 T −1
This leads to the following:

τw δ
Π=
µaδ
uδ
Π1 = = Mδ
aδ
(2.47)
R
Π2 =
δ
ρuδ δ
Π3 = = Reδ
µ

At separation Π = 0 & Π1 = f (Π2 , Π3 ) so Mδ = f (Reδ , Rδ ). Results of further analysis can be presented in

a Mδ vs. Reδ - diagram, figure 2.4, with Rδ as a parameter of family of curves (Zierep [27]). This means that
there are only 3 parameters instead of the original 7 to be considered.
Note, that for a given fixed Mδ and Reδ a reduction in curvature increases the tendency of the flow to
separate. This is because the strength of expansion behind the shock determines the favourable pressure
gradient necessary for flow to remain attached.

Solid sphere falling through a fluid

Question: What is the terminal velocity of a solid sphere falling through a fluid?

v = f (d, ρs , ρf l , µ, g) (2.48)

G. Eitelberg 17 Delft University of Technology

 separated

Mδ R
→∞
attached δ
R
δ
<∞
R
δ
<< ∞

Reδ
Figure 2.4: Parametric (M, Re)-curve, with the non-dimensional radius of curvature R/δ as the parametric
variable.

V
Π = √ = f1 (d, ρs , ρf l , µ, g)
gd
ρs
Π1 =
ρf l
√ (2.49)
dρf l gd
Π2 =
µ
 √ 
V ρs dρf l gd
→ √ = f2 ,
gd ρf l µ
√
dρf l gd
As Re → 0 then µ → 0, which will result in.
 
V ρs
√ = f3 (2.50)
gd ρf l

Where f3 is an odd function of density ratio.

Drag as a function of angle of attack

D ∝ D0 · (1 + αn) . Since the drag will vary independently of the direction of changing α from 0, it has to be
an even function: D = D0 + cα2 + higherorderterms.

2.3 Differential equations

An universally valid method is to use similarity parameters obtained from differential equations. These simi-
larity parameters are obtainable when the relevant equations are known. In most experiments, the analysis
is supported by theory which needs to be deployed also in the preparation of the experiments. Since a lot of
aerodynamic experimentation revolves around description and characterization of the boundary layers, it is
considered worthwhile to dedicate extra sub-chapters to the specific characteristics of boundary layers and
their similarity characteristics. The approximations underlying the BL theory can serve as powerful tools in
the interpretation of the results obtained in experimental simulation of flight (and other) phenomena. This
chapter will discuss the example from the boundary layer theory in the case that the flow can be assumed
to be 2D, steady state and incompressible. First the Navier-Stokes equations for the conservation of mass
and momentum need to be applied.

G. Eitelberg 18 Delft University of Technology

Mass conservation (continuity equation):
∂u ∂v
+ =0 (2.51)
∂x ∂y

Momentum conservation:
∂u ∂u 1 ∂p
u +v =− + ν∇2 u (2.52a)
∂x ∂y ρ ∂x

∂v ∂v 1 ∂p
u +v =− + ν∇2 v (2.52b)
∂x ∂y ρ ∂y

And the boundary conditions are:

• At the solid wall (y = 0): u = v = 0
• At y = ∞: u = u∞ & v = 0
For further development, it is important to realise that there are two length scales involved, L in the x
direction and δ in the y direction. With the above it is possible to non-dimensionalise the parameters and
then perform an order of magnitude analysis consequently.
x
x0 = → O(1)
L
y
y0 = → O(1)
δ
u
u0 = → O(1) (2.53)
u∞
v
v0 = → O(?)
u∞
p
p0 = → O(1)
p∞
With the above non-dimensionalised parameters it is possible to obtain the order of magnitude for v 0 . This
is found from the non-dimensionalised continuity equation.
u∞ ∂u0 u∞ ∂v 0
+ =0 (2.54)
L ∂x0 δ ∂y 0
Which can be rewritten to:
∂u0 v0 L
 
∂
0
+ 0 =0 (2.55)
∂x ∂y δ
Assuming smooth and slowly changing functions (this is a restriction of the so far generally valid description
0
excluding discontinuities), then O(u0 ) = O(x0 ) = O(1) = O( ∂u∂x0 ). Since the conservation of mass should
0
be maintained, the order of vδL should also be O(1). This results in O(v 0 ) = O( Lδ ), where Lδ << 1 for a
boundary layer (Note, that this approximation does not fulfill the far field boundary condition exactly; there
the velocity in the y-direction was required 0.)
By non-dimensionalising and rearranging the equation for momentum conservation the following equations
are found.
0
v 0 L ∂u0 p∞ ∂p0
 2 0
L2 ∂ 2 u0

0 ∂u ν ∂ u
u + =− + + 2 (2.56a)
∂x0 δ ∂y 0 u∞ ρ ∂x0 u∞ L ∂x02 δ ∂y 02

∂v 0 v 0 L ∂v 0 p∞ L ∂p0
 2 0
L2 ∂ 2 v 0
 
ν ∂ v L
u0 0 + 0
= − 0
+ 02
+ 2 02
· (2.56b)
∂x δ ∂y u∞ ρ δ ∂x u∞ L ∂x δ ∂y δ

G. Eitelberg 19 Delft University of Technology

The above conservation of momentum in the y-direction is expanded by multiplying both sides with the ratio
L
δ

L 0

!
v 0 L ∂ Lδ v 0 ∂ 2 Lδ v 0 L2 ∂ 2 Lδ v 0




0∂ δv p∞ L2 ∂p0 ν
u + = − + + 2 (2.56c)
∂x0 δ ∂y 0 2
u∞ ρ δ ∂x0 u∞ L ∂x02 δ ∂y 02

With boundary conditions:

• At y 0 = 0: u0 = v 0 = 0
• At y 0 = ∞: u0 = 1 & v 0 = 0
And similarity parameters:
δ
• Π1 = L
u∞ L
• Π2 = ν , which is identified as the the Reynolds number.
p∞
• Π3 = ρu2∞ , which is identified as the Euler number.

The equations 2.56a and 2.56c will yield identical solutions for the non-dimensional velocities and pressures,
as long as the three similarity parameters for different flow fields are same.
From equations 2.56a and 2.56c, non-dimensional forms of conservation of momentum can be further
analysed for special cases, where orders of magnitude of each term can be discussed. In the following the
Reynolds number dependency of further approximations will be discussed.
u∞ L
1. Re = ν << 1 (creeping flow)
 
∂u0 v 0 L ∂u0 ∂p0 1  2 0 2
∂ u + L ∂ u 
2 0
u0 0 + 0
= −Eu 0
+ 02
(2.57)
∂x} δ ∂y | {z } ∂x Re |∂x
 δ ∂y 02 
2
| {z | {z } ? |{z} |{z} {z } |{z} | {z }
∼1 ∼1 ∼1 >>1 ∼1 ? ∼1

 
∂ Lδ v 0
v 0 L ∂ Lδ v 0 2 L 2 L 0 0





L2 ∂p0 1  2
∂ δ v + L ∂ δ v
u0

+ = − Eu +  (2.58)
∂x0 } δ ∂y 0 |{z} δ 2 ∂x0 Re | ∂x 02 δ2 ∂y 02 
| {z | {z } ? |{z} |{z} |{z} {z } |{z} | {z }
∼1 ∼1 ? ∼1 >>1 ∼1 ? ∼1

Since the left-hand side is of O(1), the terms of order greater 1 on the right have to be compensated
by the unknown terms:

∂p0 ∂ 2 u0 L2 ∂ 2 u0
Eu · Re = + (2.59a)
∂x0 ∂x02 δ 2 ∂y 02

L2 ∂ 2 Lδ v 0 L2 ∂ 2 Lδ v 0



∂p0
Eu · Re = + (2.59b)
∂y 0 δ 2 ∂x02 δ 2 ∂y 02
L 2 ∂p0 ∂p0


This will hold when δ = O(1) and Re · Eu = O(1). This is the case since ∂x0 ≈ ∂y 0 ≈ O(1).
2. Re >> 1 From equation 2.57 it can be seen that the left hand side is of O(1). This means that the sum
of all terms on the right hand side should be of the same order. In order to satisfy the no-slip condition
1 L2 L √1 . Furthermore
the wall friction should be finite. This results in Re δ 2 = O(1), which results in δ ∼ Re
p∞
the Eu = ρu ∞
≤ O(1).

G. Eitelberg 20 Delft University of Technology

 2 ∂p0
From equation 2.58 Eu Lδ2 ∂y 0 ≤ O(1). Substituting the order of magnitude found earlier this results in
∂p0 1 ∂p
∂y 0 ∼ Re → ∆py << 1. With this result, setting ∂y = 0, the BL equations can be formulated.
Mass conservation:
∂u ∂v
+ =0 (2.60)
∂x ∂y
Momentum conservation:
∂u ∂u 1 ∂p ∂2u
u +v = +ν 2 (2.61)
∂x ∂y ρ ∂x ∂y
With the wall friction coefficient being (Re >> 1).
   0
τwall µ ∂u ν ∂u 1 L
ξ= 2
= 2 = 0
∝ (2.62)
ρu∞ ρu∞ ∂y w δu∞ ∂y w Re δ
√
Since Lδ ∼ Re the wall friction coefficient is ξ ∼ √1Re , which is valid for laminar flow.
These boundary layer equations and characteristics are obtained through non-dimensionalising conserva-
tion equations and analysing the orders of magnitude of the terms in the non-dimensional equations. In this
case the understanding of phenomena is achieved without resorting to a large number of calculations/mea-
surements.

2.4 Boundary layer: Stokes flat plate

In this section the special case of the development of a boundary layer over a suddenly accelerated flat
plate will be discussed. In figure 2.5 a graphical representation of the problem is shown. Lets assume that
this flat plate is at rest in a medium that is also at rest. In other words the velocity of the flow as well as
the velocity of the plate at t < 0 is 0. At t ≥ 0 the velocity of the flat plate is u∞ and because of no slip
conditions the velocity at the wall should be u∞ as well. The governing equations for this problem are the
y

Uplate = U∞ →

U∞

Figure 2.5: Stokes flat plate sudden acceleration to u∞ for t ≥ 0.

continuity equation and the momentum conservation equation.

div ~u = 0 (2.63)

D~u
ρ = −∇p + µ∇2 ~u (2.64)
Dt
The initial conditions for this problem at t = 0 are: u = 0 in the whole field.
The boundary conditions for t > 0 are: at y = 0 → u = u∞ & at y = ∞ → u = 0. The velocity of the
flow field is dependent on the distance from the plate and the time, u = u(y, t), it does not depend on the
horizontal coordinate.

G. Eitelberg 21 Delft University of Technology

With these conditions the conservation of momentum reduces to the following.

∂u ∂2u
=ν 2 (2.65)
∂t ∂y
| {z }
diffusion

∂p u
since ∂y = 0. If the velocity is non-dimensionalised with the velocity of the flat plate, u0 = u∞ the momentum
equation can be further simplified to:

∂u0 ∂ 2 u0
=ν 2 (2.66)
∂t ∂y

Including the boundary and initial conditions the non-dimensional velocity follows u0 (0, t) = 1 & u0 (y, 0) = 0.
2
The remaining parameters t, ν, y 2 can be non-dimensionalised as yνt or √yνt . As a conclusion u0 = uu∞ =
 
f √yνt and the solution to the momentum balance is:
 
0 y
u = 1 − erf √ (2.67)
2 νt

describing an continuously falling velocity with the distance from the wall and at the same time continuously
increasing velocity (to its maximum value at the wall) in the field with time.

√δ
νt
t
νt
y

√y

U∞ u
= u′ 1
U∞

(a) Velocity profile original dimensions (b) Dimensionless velocity profile.

Figure 2.6: The dimensional and dimensionless velocity profiles of the Stokes flat plate boundary layer.

In addition to the velocity profile, also a vorticity profile can be used describe the flow. The vorticity in the
above described boundary layer flow can be expressed by

∂u u∞ −y2
rot ~u = =√ e 4νt (2.68)
∂y πνt

From here a few observations can be made. One is that at all times ∂u ∂y
→ 0 as y → ∞. Another one is that
∂u ∞ ∂u
∂y → 0 when t → ∞ and y > 0, but finite. The integral of the vorticity 0 ∂y
dy = const. The fact that the
integral stays constant shows that vorticity is produced at the boundary at the instant of the acceleration of
the flat plate, and consequently diffused into the fluid. For this case no additional vorticity is generated at
the wall after the initial acceleration in this zero pressure gradient boundary layer.

G. Eitelberg 22 Delft University of Technology

Another interesting observation in connection with the production of vorticity by accelerated objects. As
shown above, the vorticity is produced on the solid boundary and then diffused into the fluid. The vorticity
flux from the wall in a general case can be shown to be caused by the pressure gradient along the surface
and by the acceleration of the surface [15].

DU0
 
1
J · n̂ = −n̂ × + (∇p)y=0 (2.69)
Dt ρ

By non-dimensionalising this equation for the vorticity flux, the relative importance of the two terms can be
analysed. The ratio of the vorticity production by the accelerated surface to that of the pressure gradient
can be expressed as a product of a Strouhal number with the square of the mach number:

f L ρu2∞
= γStrM 2 (2.70)
u∞ p

The Strouhal number could be formed with a frequency, as in the case of flapping wings, or with the inverse
of a characteristic time required for a manoeuvre. This is an ongoing a topic of research, such as e.g. in
the case of the shear layers rolling off of a rocket during launch in the transonic and supersonic phase.

There are many more tools to approach the simulation of real life in a laboratory frame - the ones described
could also be taken to further levels of refinement. The above chapter should, however, be sufficient as a
starting point for most of the frequently encountered challenges. The material provided should be sufficient
to make intelligent choices of where to start and what to look for.
In the aerodynamic testing, the laboratory frame is represented by wind tunnels. These are usually char-
acterized by their capability to simulate some velocity or force ratios, the other similarities are customarily
neglected - sometimes with good justification, sometimes without - the reader should by now be able to
judge that. The next chapter will deal with the wind tunnels as the laboratory framework of aeronautical
testing.

G. Eitelberg 23 Delft University of Technology

3 Wind tunnels
Wind tunnels are required to simulate aspects of vehicles moving through the air or for aspects of air mov-
ing over objects. The wind tunnels have been around since the 1870’s and have since proven their use
as an effective way to find the aerodynamic characteristics of all kinds of vehicles. Even with the rise of
computational fluid dynamics wind tunnels remains of importance. The importance resides in the fact that
the flow in the wind tunnel is not a model. The turbulence in the boundary layer is what it is, it is not repre-
sented by an approximation like in the RANS calculations. Another aspect is that for configuration studies,
the productivity of the wind tunnels is still unmatched by anything else. Even non-symmetrical polars can
be performed in a continuous manner. A third aspect is that the environment in the wind tunnel, although
far from perfect, is controllable to a much higher degree of accuracy than in any flight test - scaled or not.
Nevertheless, it remains impossible for these wind tunnels to simulate all aspects of the aerodynamics of a
vehicle simultaneously. In aeronautics, we usually try to either simulate some of the Reynolds-number sim-
ilarity related effects or some of the Mach number similarity related effects. Sometimes it is more effective
to simulate the combinations of similarity parameters like the Ma and Re numbers as it remains impossible
to simulate all relevant similarities simultaneously. This may be the case even if a phenomenon is studied,
which looks identical on the surface, but the responsible mechanism is different. This is illustrated in the
following example of generation of noise in a wind tunnel.
Example: Aero-acoustic noise can be generated by external and internal effects. From the literature it
is known that external protrusions produce wake oscillations at a relatively constant Strouhal number in
incompressible flows:

fl
Str = ≈ 0.2. (3.1)
u
Thus for a scaled object, the frequency scales with the scaling factor (the inverse of the length scale) of a
geometrically similar object. This holds for most noise sources in an aircraft, like landing gear, wing and
flap tips, even for the angular speed of a rotating engine.
Another noise source can be a cavity in the structure. The frequency of the (acoustic) pressure oscillations
scale with the following expression for a Helmholtz resonator:
r
a A
fH = (3.2)
2π V0 L
Where a is the speed of sound, A is the surface area of the cavity neck, V0 is the cavity volume and L is the
length of the neck. This frequency is invariant with the flow velocity, thus does not scale with the Strouhal
number, but it does still depend on the geometry in a similar proportionality as the external noise. This
invariance of frequency with the velocity lends itself to the identification of noise sources - if the Strouhal
number stays constant, it is a protrusion; if the frequency stays constant (i.e. the Strouhal number varies
with the velocity), it is a cavity.

In the context of scaling reality, the Helmholtz resonator cannot be left as a dimensional quantity - it too has
to be scalable. the following scaling of Helmholtz resonators enables one to simulate them in scaled wind
tunnel models:

r
3 D
f D = const. D V (3.3)
a L

G. Eitelberg 24 Delft University of Technology

The non-dimensional Helmholz resonator frequency is a function of the ratio of the volumes of the neck/to
the reservoir and the neck aspect ratio.

There are yet other mechanisms for generating noise in the flow, like periodic shear layer impingement etc.
which follow yet other scaling laws. This means, that while it is possible to study the mechanism of noise
generation in an experimental facility, it is not always possible to generate a similar noise to that observed
in real life.

In this example, part of the reasoning behind not scaling automobiles in wind tunnels can be found. We
shall look at the Ma and Re-number considerations separately.

The principles of the behaviour of wind tunnels can be described by the isentropic relations given below.
These equations are valid for both the subsonic as well as the supersonic behaviour of the flow in the wind
tunnel circuit, as long as there are no significant source or sink terms involved. The sources sinks could
be heat exchangers or flow straighteners in some sections of wind tunnels, for example. Some times there
are also actual sources, like engine simulators, or sinks like windmills in the tunnels. These objects will be
treated separately, when simulation of thrust or rotor propulsion is discussed. Here we deal with the wind
tunnels themselves.

T 1
= γ−1 (3.4a)
T0


1 + 2 M2

ρ 1
= 1 (3.4b)
ρ0 1+ γ−1 2

γ−1
2 M

p 1
= γ (3.4c)
p0 1+ γ−1 2

γ−1
2 M

Nevertheless, in the case for subsonic testing, the Bernoulli equation together with the mass conservation
is often sufficient to estimate the relationship between the nozzle contraction and the flow velocity. How to
use these equations to find the behaviour and limits of wind tunnels will be described in more detail later in
this chapter.

3.1 Different types of wind tunnels

Wind tunnels appear in all forms and sizes and there are many ways to categorise them. The two most
common categorisations will be discussed in this section, namely.
• Based on the configuration of the wind tunnel.
– Open return
– Closed return
– Open test section
– Closed test section
• Based on the flow regime of the wind tunnel, see figure 3.1.
– Subsonic (M << 1)
– Transsonic (M ≈ 1)

G. Eitelberg 25 Delft University of Technology

 – Supersonic (M > 1)
– Hypersonic (M >> 1)

Figure 3.1: Wind tunnel classification in terms of flow regime. The drive system is only indicative; many
tunnels do not have a continuous drive. [7]

3.1.1 Wind-tunnel configuration

The four different configurations (closed/open for circuit and for test section) were already introduced above.
Apart from the main lay-out of the wind tunnel there are also two options for the lay-out of the test section
as it can be an open or closed test section. Thus there are four main configurations within wind tunnels.
They will be discussed in some more detail in the following.

Open-return wind tunnel

The simplest configuration is the open-return wind tunnel, which is commonly referred to as the Eiffel-type
wind tunnel, figure 3.2. The name is that of the original designer of this type of wind tunnel, Mr. Eiffel,
of the tower fame. This type of wind tunnel accelerates the stagnant air from the environment and later
‘dumps’ this air into its surroundings. Though this wind tunnel is easy and cheap to build it is energetically
inefficient to operate as this type of wind tunnel does not re-circulate the air. Since the accelerated air is not
reused the air needs to be constantly accelerated from zero velocity and the energy gets fully dissipated
downstream of the wind tunnel again. This results in more work being done than in the case of keeping the
once accelerated air in continuous motion. A measure of energy efficiency is the ratio of power P supplied
to the drive fan to the power 12 ρAU 3 of the flow in the cross section. Even when assuming that the fan
efficiency is 100%, this ratio will be no less than 1 for the open return wind tunnel. On the other hand, for a
closed return wind tunnel this value can be as low as 0.1, meaning that the tunnel wall losses are of order
of 1/10th of the total power.
Another major drawback is the quality of the flow. If any disturbances occur in the air upstream of the tunnel
intake, these disturbances can travel to the wind tunnel and into the test section. These disturbances are
hard to account for as they are independent on the wind tunnel. As a summary the main pros and cons of
this type of wind tunnel are given below.
• Pro
– Low construction effort
– No cleaning

G. Eitelberg 26 Delft University of Technology

 • Con
– Frequency quality a function of environment prior to inlet
– High energy requirement, flow accelerated from u = 0
– Noise radiation

Test section
Flow field about a model
simulates conditions of flight Model

Controlled
airstream Motor

Balance

Test instrumentation

Figure 3.2: Open-return (Eiffel type) wind tunnel. [23]

Closed-return wind tunnel

Rather than continuously accelerating the air it is more energy efficient to let this accelerated air circulate,
such that less energy is lost. By doing this the wind tunnel will require less power to operate.
The type of wind tunnel that uses this principle is called a closed-return type wind tunnel, however it is
also referred to as a Göttingen type wind tunnel, figure 3.3. Though this wind tunnel is far more energy
efficient than an open-return type wind tunnel the construction effort is much higher and therefore the costs
as well, as measures to keep the flow optimal, such as air cooling and turning vanes, are needed. However
when a closed-return type wind tunnel is well designed, the quality of the flow in it is better than that in an
open-return type tunnel. The main pros and cons of closed-return type wind tunnels are shown below.
• Pro
– Flow quality determined by internal construction
– Energy efficiency
– No environmental impact of operating noise
• Con
– High cost of construction
– Contamination settles in the circuit
– Air cooling required
As mentioned before, the closed-return wind tunnel requires measures for good and constant flow proper-
ties. One method to obtain constant properties over the duration of testing is achieved by cooling the air.
Even though the amount of dissipated energy compared to the total flow power can be managed at a low
level, the viscous dissipation is not negligible. The cooling is necessary as work is constantly being done
on the air and this work increases not only the kinetic energy but also the internal energy of the air. The
internal energy is expressed by the temperature of the air. As air starts to heat up its properties change,
which is unwanted. On the other hand, an air cooler is an obstruction in the flow path, so the design has to
consider the effect of the cooling to the instantaneous and local flow quality in addition to the benefit of the
temporal distribution.
Apart from cooling the air also turning vanes to aid the redirecting of the air flow around the wind tunnel
circuit corners are needed. These vanes are required for good flow properties but also play a significant
role in the power consumption of the wind tunnel. Since the cross section of the wind tunnel circuit is larger

G. Eitelberg 27 Delft University of Technology

 Guide vanes Model supporting frame Actuator fan
Nozzle

Return duct

Figure 3.3: Closed-return (Göttingen type) wind tunnel. [23]

in the corner along the diagonal than it is in the perpendicularly joining sections of the tube, infinitely thin
turning vanes will maintain this variation in the stream tube cross-sectional area. This change in cross-
section causes the flow to either expand or compress at the corners, both of which cause energy losses
and affect the uniformity of the flow. This phenomenon can be seen in figure 3.4. One can see that the
turning vanes depicted on the right keep the cross-section constant and therefore keep the pressure in the
flow constant.

Figure 3.4: The importance of turning vane design. In this case the gap width is held constant. Illustration
from an internal project in DNW. [12]

Example: As mentioned before, operating an open-return wind tunnel requires more energy than a closed-
return type as the flow needs to be constantly accelerated. But how much power does an open-return wind
tunnel actually need to operate? In order to find out how much this is the energy balance of the flow is
needed, which is shown below.
 
1
P = Ė = e + u2 ṁ
2
  (3.5)
1
= e + u2 ρuA
2

Assuming that the internal energy of the flow does not vary, which is an assumption of reasonable accuracy
for low speed tunnels, the power required per square meter of wind tunnel cross-section is found to be
Preq = 12 ρu3 .
In case of the DNW-LLF wind tunnel a speed of 100 m/s can be achieved in a 6x6 m2 test section. If this
tunnel would have been open circuit tunnel it would require 22 M W of power to be delivered to the flow,
even more to the fan when accounting for its limited efficiency.
Fortunately, this wind tunnel is of a closed-return type, thus part of the kinetic energy of the flow is recovered
and about half that power is actually needed to drive the fan.

G. Eitelberg 28 Delft University of Technology

Closed test section
As previously mentioned, there are two main kinds of test sections, namely an open test section and a
closed test section. Both have their advantages and disadvantages, however for most aerospace applica-
tions closed test sections are preferred. An example of a closed test section is shown in figure 3.5. The
main characteristic of a closed test section is that the flow is not free to move as it would have in real life
applications. This constraint affects the quantitative measurements of aircraft model behaviour in flow. This
means that the simulation of the aircraft flight properties is not perfect. Section 3.2 will discuss in detail how
to correct for effects that occur in closed test sections.

Figure 3.5: Closed test section. [20]

Another characteristic that needs attention, when utilizing the closed test section configuration is the rise of
a boundary layer along the wind tunnel walls. As the flow is constrained at the walls, a boundary layer is
created that not only decreases the effective cross-section but it also interferes with wall mounted objects,
such as testing a 2D wing in a test section where the wing covers the entire width of the tunnel. Ways to
improve this is by introducing boundary layer blowing (or suction, although this means additional removal of
momentum, which the the BL does already) in which ‘new’ air is introduced at the walls of the wind tunnel
which will make the flow more uniform as shown in figure 3.6.

Figure 3.6: Boundary layer blowing to reduce the wall effect. [1]

Open test section

Open test sections, such as shown in figure 3.7, do in certain situations allow for the flow to behave more
like it would in real life applications as the flow is free to move around the model as there are no walls
constraining the flow. The main problem though is prediction of the boundary conditions as these vary for
every model configuration. In general the blunt body models can be tested relatively well as circulation of
the flow created by lifting surfaces is not present. Lifting surfaces disturb the flow in such a way that it is

G. Eitelberg 29 Delft University of Technology

hard to predict the real behaviour of the model. That is why these kind of tunnels are used mainly in the
automotive sector as the lift and circulation for car models is often small enough to be neglected. In section
3.2 methods to correct for the limitations/constraints of the wind tunnel flow are discussed in more detail.

Figure 3.7: Open test section. [16]

3.1.2 Low-speed wind tunnels

Low Speed wind tunnels are used for low Mach numbers (M < 0.3). In general the air is moved by a large
axial fan that increases the dynamic pressure to overcome the viscous losses. This working principle is
based on the continuity and Bernoulli equation and is explained below. Note that the flow is assumed to be
incompressible, which is a reasonable approximation for low wind speeds. In the following, the limitations
of the above assumption and also the finite size of models assumed to have infinite span (2D-models) will
be discussed.
Starting with the continuity equation we have the following.

dA du
Au = constant ⇒ =− (3.6)
A u

When introducing Bernoulli’s equation the following is found.

C 2 psettl − pm ∆p
u2m = 2 ≈2 (3.7)
C2−1 ρ ρ

In which the contraction ratio is introduced as C = AAsettlm

. From eq. (3.7) one can see that for large
contraction ratios the velocity achieved in the test section is directly proportional to the applied pressure
difference.
For subsonic wind tunnels it is often assumed that compressibility effects are negligible, ∆ρ
ρ < 0.01 or < 1%,
and therefore can be assumed to not influence the experiments performed in the tunnel. For what range
of Mach numbers is this assumption valid? In order to find this out a series expansion of equation 3.4b is
performed, as given below.

ρ 1 M2
≈ 2 + ... ≈ 1 − + ... (3.8)
ρ0 1 + M2 2

G. Eitelberg 30 Delft University of Technology

From this equation it can be seen that the change in density is proportional to the square of the Mach
number.
∆ρ M2
∼ (3.9)
ρ 2
√
From the above relation it can be seen that the assumption of incompressible is valid for M < 0.01 · 2 ≈
  most ‘low speed’ tunnels are operated up to M ≈ 0.3, which results in a density change of
0.14. However,
about, ∆ρρ ≈ 0.045 ≈ 5%.

Apart from compressibility affecting the measured aerodynamic parameters in a subsonic (low speed) wind
tunnel, there are other effects that need to be considered as well. This could be the span when testing 2-
dimensional airfoils. In the subsonic regime both the effect of the span width and the effect of compressibility
can be accounted for with the help of linearised potential theory and they show up together as a sort of
compound effect. For the limiting case of zero compressibility and infinite span,
span b
M∞ → 0, σ = chord = c → ∞,
the relation for the lift coefficient is known to be Cl = 2πα, where α is the angle of attack in radians. The
effect of the span in the wind tunnel for an elliptic lift distribution changes the above relation for the lift
coefficient to be:
2πα
Cl = (3.10)
1 + σ2

For Mach numbers smaller than 1 another linearised correction for the effect of the Mach number on the lift
coefficient needs to be taken into account as shown below.
2πα 1
Cl = p · (3.11)
1−M 2 1+ √ 2 2
| {z ∞} | σ 1−M∞
Mach effect
{z }
span effect

The first term describes the influence of the Mach number upon the lift coefficient at infinite aspect ratio, the
second term corrects for finite aspect ratio. This means that for profile testing the Mach number influence is
not negligible, even at very low compressibility! Increase in Mach number is equivalent to reduction in span
width. Thus it is preferable to obtain Reynolds number variation independently of Mach number variation.
Therefore for fixed wing testing, M ≈ 0.3 should be seen as a limit.

3.1.3 High-speed subsonic and transonic wind tunnels

High-speed subsonic wind tunnels (0.4 < M < 0.75) and transonic wind tunnels (0.75 < M < 1.2) are
designed for higher flow speeds. However as the velocity of the fluids becomes larger compressibility
effects come into play. This makes the previously useful incompressible Bernoulli equations inadequate
and isentropic expansion relations are needed to predict the wind tunnel behaviour in this compressible
regime. It is helpful to recall the definition of the critical values at the sonic throat when flow is accelerated
to supersonic speeds where a = a? = V .
The critical values obtained from isentropic relations are given below:

T? 2
= ≈ 0.833 (3.12a)
T0 γ+1

γ
p?
  γ−1
2
= ≈ 0.528 (3.12b)
p0 γ+1

G. Eitelberg 31 Delft University of Technology

 1
ρ?
  γ−1
2
= ≈ 0.634 (3.12c)
ρ0 γ+1

Using the area relations and substituting the isentropic relations provided in the beginning of the chapter
one can find the relationship between the nozzle area ratio and the Mach number in the nozzle, of which a
crude derivation is given below.

ρAu = const = ρ? A? u? = ρ? A? a?
A ρ? a? ρ? ρ0 a? (3.13)
?
= =
A ρ u ρ0 ρ u

All the three ratios in the product of the right hand side of the lower equation in 3.13 are known from the
isentropic relationships given before. The multiplication results in the following expression:

2  γ+1
γ − 1 2 γ−1
  
A 1 2
= 2 1+ M (3.14)
A? M γ+1 2

Expression eq. (3.14) means, that the Mach number in the expanding part of the nozzle, when the sonic
conditions where obtained at the throat, is only a function of the area ratio as can be seen from fig. 3.8.

Figure 3.8: Relationship between the Mach number and the area ratio in an expanding flow channel. [2]

To accelerate the flow to supersonic conditions a convergent-divergent nozzle is required. The convergent
part is required to reach the sonic conditions. To reach the sonic conditions, an overpressure of approx-
imately 2 is required. After the sonic conditions are reached, additional expansion to yet higher Mach
numbers can only be reached in a diverging flow channel called a diverging nozzle. In consequence, the
Mach number achieved in a test section in a convergent-divergent wind tunnel is solely related to the ex-
pansion ratio ( AAthroat
test
), so the effect of pressure won’t influence the Mach number, provided the pressure to
achieve sonic conditions in the throat is provided.

In transonic wind tunnels the walls of the test section are often slotted or perforated. Since the transonic
testing is usually required for high speed cruise condition testing at free stream Mach numbers just slightly
below 1, the model may represent such a restriction in the test section as to create sonic conditions between
the wind tunnel wall and the model. This is called blocking or chocking the flow in the transonic wind

G. Eitelberg 32 Delft University of Technology

tunnel. Therefore, in wind tunnels with close to sonic flow, the effect of the model blocking up the cross-
section of the test section can give rise to local supersonic flows. Increasing the pressure in the settling
chamber will not change that. It only can raise the flow rate, but cannot change the position of the sonic
transition, and the wind tunnel remains chocked. In order to reduce the blockage effect, the wind tunnel
walls are made permeable by perforation or longitudinal slots. This enables the flow to escape the straight
wall section, and reenter the test section downstream of the model again. A schematic representation is
shown in figure fig. 3.9. [26]. The slotted/perforated walls are said to adapt to the flow conditions. Their
drawback is that their influence upon the flow field is difficult to quantify. Closed wall boundary conditions
enable a straightforward formulation for the wall interference issue, the semi permeable walls are much less
accessible to the quantitative analytical treatment.

Figure 3.9: Perforated/slotted wind tunnel. Note that the reentry of the flow into the tunnel circuit is not
shown, but must occur in a closed loop configuration. [26]

Another way to deal with shocks reflected from the wind tunnel walls, is to design the experiment in such
a way that the shocks generated by the model do not interfere with the tests. In the transonic testing for
civil transport aircraft, the shocks closing the local supersonic regions above the wings stay far from the
walls. In supersonic or hypersonic testing, the testing is frequently performed in free jet conditions and the
reflections of the shocks return as expansion waves far downstream of the test articles, so the reflections
sketched here for solid walls, do not influence the test results, figure 3.10.

Wall-reflected
Model bow shock
shockwave wave

M = 1.5
M = 1.3

M = 1.1

Figure 3.10: Experimental rhombus variation with the Mach number. [23]

G. Eitelberg 33 Delft University of Technology

3.1.4 Supersonic & hypersonic wind tunnels
It requires a lot of effort to achieve supersonic and hypersonic speeds (approximately in the range 1 < M <
3 & 3 < M respectively). These speeds can be achieved with an appropriate design of the convergent-
divergent nozzle and the Mach number and flow quality greatly depend on the nozzle geometry. In order
to change the Reynolds number of the flow one can vary the density level or the pressure in the settling
chamber.
The main operational challenges for the design of a supersonic wind tunnel are:
• Pressure ratio to achieve sonic conditions in throat
• Air drying system to avoid condensation of water vapour residues
• Pre-heating of the gas to avoid condensation of air (M > 4)
• High power demand for continuous operation; this can be avoided by intermittent blow-down operation.
The relationship between the Mach number and temperature in an adiabatic expansion was already provid-
ed before:
T 1
= γ−1 (3.15)
T0 1 + 2 M2

From this relationship it can be seen that the static temperature in the wind tunnel drops with the square
of the Mach number for high Mach numbers. This explains why the reservoirs of hypersonic wind tunnels
are often heated. The same behaviour observation can be explained without the exact equation; just by
re-formulating the square of the Mach number as a dimensionless ratio between the kinetic energy and the
thermal/internal energy of the fluid.

u2 Ėkin
M2 = ∝ (3.16)
a2 Ėinternal
Here, too, the internal energy (expressed as a product of temperature with specific constants or alternatively
as the square of the speed of sound) is inversely proportional to the Mach number. Looking at the equation
above one can see that for high Mach numbers the kinetic energy dominates the thermal energy of the fluid
and this makes it clear why the temperature drops significantly for high Mach numbers. In other words, in
order to achieve hypersonic flow conditions in the test section the internal energy in the reservoir needs to
be very high as all this internal energy will be converted into kinetic energy by the time it reaches the test
section. The high internal energy is provided by heating the fluid in the reservoir. However, this causes a
technical problem as it requires a lot of energy to simulate some flows. For instance, if one would simulate
a reentry vehicle entering the atmosphere one would require to simulate a flow at 100 km altitude with
a velocity of 8 km/s. This would result in tunnel temperatures of 9000 K, which is above the melting
temperature of most materials used to contain the gas. [8].

Power requirements for supersonic wind tunnels

The power required to run supersonic and hypersonic wind tunnels is enormous and can reach up to 50 MW
per square meter of test section. For this reason most tunnels operate intermittently using energy stored
in high-pressure tanks as shown in fig. 3.11 for blow-down operation, or vacuum storage tanks for suction
type supersonic wind tunnels; however the latter is seldom used as those wind tunnels are restricted to low
Reynolds numbers.
Other problems include, enough supply of dry air, wall interference effects, and the need to use fast instru-
mentation for intermittent measurements.

Conditions for steady-state operation

To understand how a blow-down tunnel produces its flow it is helpful to use the following relationship for the
speed of sound a.

G. Eitelberg 34 Delft University of Technology

 Supersonic nozzle Test section

Settling chamber Model

Sonic throat
Exit throat

Pressure control
High-pressure tank

Exit

Support
Spreader Screens

Figure 3.11: Blow down wind tunnel. [24]

√
a∼ T (3.17)

So one can relate the different speeds of sound for the reservoir and throat by using the following relation
obtained from the fundamental relationships for adiabatic expansion.

a?
r
T?
= ≈ 0.91 (3.18)
a0 T0

For a homogeneous (i.e. ignoring the wave propagation), but low rate of change in reservoir conditions
we require that the information about the outflow be spread effectively and instantaneously overall in the
reservoir vessel. In this approximation the pressure and temperature (p, T ) remain uniformly distributed in
the vessel. The characteristic time of spreading information is aD0 , in which D is the diameter of the reser-
voir. This is the time required to equilibrate the conditions at one part of the reservoir with another, e.g. the
conditions everywhere in the reservoir with those at the exhaust location.

The characteristic time in the discharge scales with the inverse of the mass flux ρ? A1? a? ( kg s
) out of the
reservoir through the sonic contraction. This time is also proportional to the mass of gas contained in the
reservoir ρ0 D3 (kg). Dividing the total mass in the reservoir by the mass flow rate provides an estimate for
the time it takes to drain the tank. Thus, in order to be able to assume a reservoir condition for the outflow, it
remains important that the time scale of the spreading of information is much smaller than the time it takes
to drain the tank, which results in the following:

D
a0 A
ρ0 D 3
∝ << 1 (3.19)
D2
ρ? A? a?

If this condition, which means that the reservoir is large compared to the nozzle throat, is met it means that
the time it takes to equilibrate the reservoir condition is small compared to the time required to empty the
vessel.
It can be assumed that under these conditions and a sufficient pressure ratio that the exit Mach number will
remain constant and determined by the area ratio of the discharge nozzle. Still, the Reynolds number will
vary according to the pressure and temperature drop in the reservoir.

In the case that the above condition is not met, or if the size of the reservoir cannot be described by the
one diameter only, i.e. there are multiple length scales to be considered, wave propagation has to be

G. Eitelberg 35 Delft University of Technology

considered. In a case where the exhaust nozzle throat is comparable to the diameter of a reservoir, but the
reservoir is shaped as a long tube, a special type of an intermittently operating blow-down super/hypersonic
wind tunnel will result. In such a case the opening of the exhaust nozzle has to be carefully designed to be
practically instantaneous, achievable by either very fast acting valves or even diaphragms with a controllable
burst conditions. The original invention was called a "tube wind tunnel", but the conventional name these
days is a "Ludwieg tube", named after the inventor, Prof. Hubert Ludwieg.
In such a tunnel, after the valve is opened an unsteady expansion will propagate into the storage tube.
Simultaneously, a steady expansion into the test section downstream of the sonic throat takes place (after
the passing of the unsteady starting process). This steady expansion has a lower effective total pressure
and temperature than the storage tube charge conditions, depending on the nozzle throat contraction and
thus the strength of the expansion wave. The figure 3.12 shows the principles in the case of a straightforward
constant cross section shock tube.

t [s]

p4 p [s]

p1
x [m] x [m]
(a) (x, t)-diagram showing the moving shock, contact surface (b) (p, x)-diagram of a shock tube.
and expansion waves.

Figure 3.12: Wave propagation in a shock tube.

By providing for a controlled contraction and expansion through a well designed nozzle, the shock tube
becomes a Ludwieg tube, visualised in figure 3.13. In the case for a Ludwieg tube, the tunnel is designed in

Ludwieg tube test section

T 4 , p4 T 1 , p1

Figure 3.13: Ludwieg tube.

such a way to make sure to maintain the sonic conditions at the throat of the nozzle. That means that there
is a standing characteristic, vertical in the t-x plane at that location. This is the tail of the expansion wave
in the above diagram. Because the speed of a wave in the centred expansion fan is given by the following
relation.
γ+1
c = a4 + u (3.20)
2
In which c is the speed of the shock wave (0 m/s at the throat) and a4 is the speed of sound at location 4 in
fig. 3.13. Since the shockwave at the throat is a standing wave it follows that.
2
u= a4 (3.21)
γ+1

Which is different from the steady expansion.

G. Eitelberg 36 Delft University of Technology

It further follows that pressure at that location can be expressed as
 γ
 γ−1  2γ
 γ−1
p T a
= = (3.22)
p4 T4 a4

It is this combination of the steady and unsteady processes, which leads to the term ‘effective total pres-
sure’. In the case when the nozzle throat is equal to the charge tube diameter the last expansion char-
acteristic will remain in the throat, i.e. the steady expansion sets in from M = 1. At that condition, the
pressure has dropped to 0.28p4 , corresponding to an effective total pressure for quasi-steady experiments
0.28
of 0.528 = 0.53p4 .

Note that the main reason for the implementation of the Ludwieg tube is because all characteristics in the
nozzle are steady. Thus, even though the discharge time is in the order of seconds, the conditions are
favourable as they don’t change in time, at least not up to the point that the reflection of the expansion wave
reaches the nozzle again. This is why Ludwieg tubes are quite long as this increases the testing time. In
the case of a steady expansion, the Mach number is steady throughout time, but other factors, such as
the Reynolds number are not, however these types of wind tunnels can operate for longer periods of time,
which has its advantages as well.

Length of a hypersonic nozzle

Under the conditions that DA2 << 1 and a sufficient pressure ratio are present, the Mach number at the
exit plane of the supersonic/hypersonic nozzle will remain constant and determined by the area ratio of
the discharge nozzle. In order to generate homogeneous and parallel flow from a reservoir, not only a
strong expansion according to the equation 3.14, illustrated in figure 3.8, is required. In addition to that the
expansion has to be controlled by a well designed wall contour of the nozzle. In such a case, in addition
to the area ratio, the required nozzle length can be easily estimated from basic principles and model size
requirements. Assuming that for geometrical resolution a model characteristic size of Lmodel = Dnozzle 3 is
required, then it is prudent to require a test section size of at least Dnozzle .
When designing the nozzle one needs to keep the total nozzle expansion angle, α, see figure 3.14 below the
value determined by the Mach number in such a way, as to obtain parallel (non-expanding) flow at the exit
plane. Assuming that the Mach angle is small, valid for large Mach numbers, the following approximation

Dnozzle ≈
α 3Lmodel

Lnozzle

Figure 3.14: Graphical representation hypersonic nozzle.

sin α ≈ tan α ≈ α can be assumed to hold. The nozzle expansion angle can then be related to the Mach
1
number as α ≈ M and with these relations it is possible to find the minimum required nozzle length, Lnozzle ,
to be,
3Lmodel
Lnozzle ≈ M (3.23)
2
At hypersonic Mach numbers this will have to be complemented by the boundary layer displacement effect.
An approximate estimation of the BL displacement effect can either be obtained from the so-called Mirel’s

G. Eitelberg 37 Delft University of Technology

correlation or estimated from the viscous interaction as:
δ M2
∼√ e (3.24)
x Re
Where Me is the Mach number in the free stream (BL edge). The reason for a very thick boundary layer
is, that the density in the hypersonic expansion drops with the 5th power of the Mach number and will thus
significantly reduce the Reynolds number. As explained in the previous chapter, low Reynolds numbers
show that viscous forces play a larger role and will thus increase the boundary layer thickness.

The limit of a supersonic wind tunnel

Now that the requirements for steady state operation have been stated it is possible to ask the final question,
namely, up to what limit can one push a blow-down wind tunnel?
A good estimation can be obtained by using the compressible formulation of the Bernoulli equation, which
represents the conservation in momentum in a stream tube, and by neglecting gravity for simplicity.

u2 γ p u2 a2
+ · = + = const (3.25)
2 γ−1 ρ 2 γ−1

In a blow-down wind tunnel the fluid is accelerated from the rest, thus the first term of the above equation
will be zero since this the initial velocity is zero. This leads to the following.

a20 u2 a2
= + (3.26)
γ−1 2 γ−1

The maximum possible velocity in the nozzle exit plane is then obtained by setting the speed of sound in
that plane to zero:
r
2
umax = a0 (3.27)
γ−1

In the case for air (diatomic gas with constant γ ) this will lead to a maximum velocity of umax ≈ 2.2a0 .
The above observation again illustrates the need not to mix non-dimensional and dimensional parameters
in a single discussion. Although there is a limit to the velocity of the flow at the exit of the expansion, there
is no limit to the achievable Mach number. The Mach number is therefore no expression of velocity. For
simulation of hypersonic flight in atmospheric conditions (such as re-entry from space orbit) this means
that the realistic supersonic conditions can be generated for the ratio of velocities to the local speed of
sound, and the correct ratios of pressure viscous forces are obtainable. The correct energy relationships
for the ratios of the kinetic energy to the molecular excitation energies may however be restricted due to
the restriction of the obtainable velocities. In order to obtain sufficient kinetic energy in the flow, the level
of internal energy in the reservoir has to be raised. The special techniques available for that are subject to
specialist literature.
Question: Explain the plausibility of the above observation in terms of the Mach number as representing
the fractional analysis of energy (fluxes) as done in equation 3.16. Assume that the Mach number is zero
in the reservoir and infinity at the end of the expansion.

3.2 Wind-tunnel corrections

3.2.1 Blockage effect explained from conservation principles
As discussed previously the walls of wind tunnel test sections influence the flow in such a way that differ-
ences are found when comparing the flow field in a wind tunnel to that of the real free flight case. In this
section the responsible phenomena for those differences will be discussed.

G. Eitelberg 38 Delft University of Technology

One of the most obvious issues for testing in closed test sections is the phenomenon called blockage. This
blockage is the fact that a model is obstructing the flow and will therefore change the pressure and velocity
field in a wind tunnel causing inaccuracies. The issue was briefly mentioned in connection with the potential
of forming a sonic throat between the model and the wind tunnel wall in testing at close to sonic conditions.
Even when no viscous effects or aerodynamic characteristics are directly resolved and the test section is a
long straight-walled conduit an additional drag will be measured on a solid body with a wake (the wake itself
is a viscous phenomenon) in the wind tunnel. The explanation for this is straightforward in principle and can
be provided with the help of the conservation laws for mass and momentum, here neglecting compressibility
effects.
In fig. 3.15 a simple test section with a model and wake representation is depicted. In this figure, the
conservation of mass can be used to determine the velocity u2 in the tunnel at the location of the wake
or the tail mount of the model. In this case the cross-sectional area of the wake or mount is represented
by d2 and the cross-sectional area of the tunnel is L2 . This means that the velocity at location 2 can be
determined to be the following.

u ∞ , p∞ u 2 , p2
d

Figure 3.15: Blockage due to model with wake and sting.

L2
u2 = u∞ (3.28)
L2 − d2

Once the velocity is determined, the Bernoulli equation can be used to determine the reduction of the back
pressure due to acceleration of the flow at the location of the wake blockage:

2 !
L2

ρ
∆p = u2∞ −1 (3.29)
2 L − d2
2

This pressure is assumed to be uniform over the whole downstream cross-section of the wind tunnel,
including the wake.
The conservation of momentum will, in turn, deliver the drag of a displacement body in the wind tunnel,
where the sum of the momentum flux and pressure forces is equal to the drag D of the model.

D = ρu2∞ L2 − ρu22 (L2 − d2 ) + ∆pL2 (3.30)

Which leads to a drag coefficient CD ,

G. Eitelberg 39 Delft University of Technology

 d2
L2
CD = 2
2
(3.31)
1 − Ld 2

This shows, that even in the inviscid and incompressible approximation, a displacement body produces
drag in a closed wall test section. For small blockage ratios, this drag coefficient approaches the blockage
ratio. This blockage drag is the correction that needs to be applied to the measured drag in a closed test
section wind tunnel. In the free air case, where the blockage ratio goes to zero, we obtain the potential flow
solution of zero drag at this level of approximation.
Similar expressions can be obtained adopting different approaches. In the literature (see e.g AGARDograph
336) it is customary to discuss the difference between the measured and the corrected coefficients. For
incompressible flow in a closed test section this may take the following form.

CDunc − CDcorr d2
= 2.5CDunc 2 (3.32)
CDcorr L

1
In the case that the requirement for tunnel accuracy for development tests of 10000 in drag coefficient, it can
be concluded that the correction is non-negligible.
Yet, another formulation for the correction in drag coefficient, now accounting for the compressibility effect
is given in literature as eq. (3.33), where  is the blockage ratio, which often is defined in terms of velocities.

CD = −(1 − (γ − 1)M 2 )CDunc  (3.33)

3.2.2 Blockage in open test sections due to pressure effects

In the case of open test sections there are no walls that force the flow into a certain path, however there
are other factors that make relating the measurements to the real behaviour of the desired model difficult
to do. Despite the missing test section walls, geometrical constraints are still provided by the size of the
exhaust nozzle and the size of the collector at the entrance to the diffusor. The main effects considered in
the automotive world with 3/4 the open test section are shown in figure fig. 3.16 and are summarised below
together with more generic observations:
• Nozzle Blockage: this is the classical correction for an open jet wind tunnel. It causes a flow velocity
reduction at the model because of an over expanded jet when compared to the free-air flow. This
effect can be increased when a blunt object is moved closer to the nozzle which causes an increase
in angle of attack and an even stronger expansion.
• Solid Blockage & Jet Expansion: One of the largest interference effects is the fact that the pressure
distribution of a free jet with no model is changed when a blunt object is introduced in the flow. The
pressure gradients caused by this interference are large and extend over the model. These gradients
are not constant along the model and therefore are cause for inaccuracies.
• Empty-Tunnel Pressure Gradients: Due to the empty tunnel pressure gradient changes of drag occur
at the model. The presence of these pressure gradients will also influence the measurements by
adding or subtracting to the undisturbed free stream drag, depending upon the model location. This
pressure gradient will in turn be affected by the presence of a model. This is particularly disturbing
in the automotive testing context. In figure 3.17 the pressure distribution along the test section is
shown for two different tunnel nozzle exit plane configurations at the same overall geometry. In the
case described as with "Sefertfluegel", the nozzle exit boundary layer is disturbed in such a way as
to produce longitudinal vorticity in the shear layer in addition to the transverse vorticity. Although the
stability of the shear layers can be improved in this way, there is a penalty in the pressure distribution.

G. Eitelberg 40 Delft University of Technology

 Figure 3.16: Schematic view of a open jet wind tunnel with the effects on the model [9].

Figure 3.17: Static pressure distribution along the tunnel axis. (Figure from DNW)

• Collector Blockage Effects: There is a effect on a bluff model when its wake enters the collector as
the flow is transformed from an open jet boundary condition to a closed wall boundary condition.
The entry of the bluff-body wake into the collector may result in a closed-wall, wake-induced velocity
increment at the model due to the changed constraint on the wake.
• Wake-Induced Effects: When compared to the other effects this is the odd one out. This effect causes
a incremental increase in drag on the model as the solid wall of the collector has its effect on the
flow, which causes a additional drag on the model. Till this time there is no correction found for this
phenomenon.
In the case of an open test section there is no ‘direct’ blockage such as in a closed test section as the air
is free to move, however there is a similar correction as found in equation 3.32. In this case the correction
has to be subtracted rather than added and leads to the following equation [9],

CDunc − CDcorr d2
= −1.3CDunc 2 (3.34)
CDcorr L

3.2.3 Potential flow approximation applied to wind tunnel interference corrections

This chapter follows the approach described in AGARDograph 336 [9], Chapter 2 by Kryzcinski. It is pos-
sible to correct for blockage in terms of the equations of conservation of mass and momentum in an in-
compressible approximation. For that we illustrated the effect of a solid body with its wake (itself a viscous
phenomenon) as an obstruction in the path of air in the test section. A more conventional approach is to

G. Eitelberg 41 Delft University of Technology

deal with the far field effects of the walls at the location of the model in terms of the potential flow theory.
In the potential flow approximation in which a similar correction could have been achieved, if the solid body
was replaced by a source term with an infinite wake into the free stream with velocity u∞ :
Q
φ= lnr (3.35)
2π
where Q is the strength of the singularity and r the distance from the origin, as long as the body was small
enough. For a finite extended body, a sink (thus negative Q) separated by the appropriate distance could
be added. For an extended body a combination of a source and sink have to be used as seen in figure 3.18
A special case is the doublet of a source and sink. The expression for a 2D dipole is given by:

Figure 3.18: Extended body represented by a source and sink.

mx
φ= (3.36)
x2 + y 2

representing a circular cylinder with infinite span. Without a wake, this represents no loss of momentum
and thus also no drag, though.

In order to explain the lift induced correction, it is useful to recall the mechanism of lift generation. In the
thin airfoil approximation it is the distribution of circulation γ along an airfoil, in the simplest form the lift
generating infinite span airfoil can be approximated by a potential vortex with circulation Γ which generates
the lift L per unit span:

L = ρu∞ Γ (3.37)

This lift can be normalised with the dynamic pressure and a length scale l (area = l · 1), yielding a lift
coefficient
Γ
Cl = (3.38)
u∞ l
No drag is generated in this model (paradox of d’Alambert).
The essence of the lift generation mechanism is the pressure distribution as a consequence of the curvature
of the streamlines introduced by the superposition of the free stream with the circulation. The interference
of the wind tunnel walls/shear layer is then obvious, since the walls stay straight, as a rule. The above
simple observation is valid for a lifting wing of infinite span; an airplane can be approximated by a so called
horseshoe vortex, where the wing tip vortices extend to infinity downstream of the aircraft model.
There are a lot of misconceptions circulating around the generation of lift. One of the most persistent ones
is that a downdraft is required to produce lift. This misconception leads to difficulties in the practice and
interpretation of wind tunnel correction applications.
First, it is important to realize that in the case of level flight, there is no net vertical displacement of mass.

G. Eitelberg 42 Delft University of Technology

Neither of the airplane, nor of air. Since for the aircraft the expression mav ertical = 0 holds, meaning that
the left-hand side of the momentum balance equation is zero for the vertical component; the mass force
mg has to be balanced by the pressure force in the vertical direction. Again, for the case of level flight the
viscous force component in the vertical direction is exactly zero.
Since there is no mass displacement taking place in the vertical direction for the airplane, the same holds
integrally for the air in the control volume. Flying does not compress the atmosphere. Although the lift
is produced by local streamline curvature, the net (integral) effect is that the left hand side of the vertical
component of the momentum balance remains zero. The lift force is the pressure force mentioned above.
In order to generate that pressure force the circulation is required (the mechanism by which the circulation
is obtained is not relevant in the present context).
The circulation has the property that the net downdraft downstream of the 2D wing is exactly equal to the
net updraft upstream of the wing. For the horseshoe configuration the analogous observation is valid for the
two streamwise legs of the model – there is no net displacement of mass. This, although it can be observed
that the mutual induction makes the horseshoe legs move downwards over the downstream distance. Still
there is no net displacement of mass in the vertical direction.
All of the above is the required introduction to wind tunnel boundary corrections with the help of potential
flow theory, also called classical corrections. It is important to note that there is no separate downdraft
correction without an updraft correction – both are necessary. There is a separate downdraft correction for
tandem arrangements, e.g. when the main wing is in the wake of a canard, or the empennage is embedded
in the downdraft of the main wing. There is no splitting for airfoil testing, infinite or finite span.
In order to obtain a (potential) flow field with the straight streamlines at the location of the wind tunnel walls,
the circulation of the airfoil is reflected on the walls as indicated in figure 3.19. This reflection has to be
performed an infinite number of times, in order to account for the interference with the far walls and the
required correction as well. In the formulation of the potential flow, the flow field can be expressed as a

Figure 3.19: Reflection of the circulation on the floor and ceiling.(I. Philipsen, DNW)

superposition of the unperturbed free stream with the model induced and wall induced flow as in equation
3.39.

Φ = u∞ x + φmodel (x, y, z) + φwall (x, y, z) (3.39)

The boundary conditions then are formulated for the sum of the two perturbations. For the solid, imper-
meable walls, where there is no flow perpendicular to the wall at the wall and thus the wall-perpendicular
gradient of the potential has to remain zero:

∂φ
=0 (3.40)
∂n

G. Eitelberg 43 Delft University of Technology

For the open jet configuration use is made of the assumption that there is no mechanism for the open jet to
drive the flow outside the jet and thus on the jet boundary the flow remains unaccelerated:

∂φ
=0 (3.41)
∂x
This means that there is no pressure gradient along the free jet boundary. The pressure in the plenum is
assumed to remain unaltered along the undisturbed boundary of the free jet. This boundary condition has
to be satisfied at the location of the undisturbed free jet. This can be somewhat misleading in practice. It is
easy to understand the plausibility of the solid wall boundary condition, but it is more difficult to do so with
the free shear layer. In principle, this BC is assumed to hold on the undisturbed boundary, while it actually
allows for the boundary to be displaced.

The boundary conditions can be satisfied by introducing additional potential flow elements outside the flow
field in a symmetric manner for the solid walls and in an anti-symmetric manner for the free jet boundaries.
The reflection of the circulation at the walls can be illustrated by simple schematic drawings, as shown in
figure 3.24. Rigorous solutions exist for only simple cases; the rest will have to be calculated from the
known information.

(a) Curved free stream path due to a lifting surface. (b) (CL − α)-curves for the experiment, the CFD model and
the free air case.

Figure 3.20: Illustration of the deviation of the free stream in an open test section and the resultant difference
from (numerical) free air lift coefficient. [21]

The open test section testing, although in the strict sense of determining the boundary conditions, is inferior
to the testing in closed test sections, is still quite popular for practical reasons. these are ease of access,
lower blockage, acoustic transparency etc. For lifting bodies it might not be sufficient to satisfy the far field
boundary conditions at the location of the undisturbed velocity jump as described above, since the displace-
ment might not be negligible, as shown in the figure 3.20.

In figure 3.20, we also note that the upstream and downstream deflection of the flow are not symmetrical, as
one would expect to be the case of potential flow approximation where, in order to satisfy the constant pres-
sure boundary condition, an updraft of equal value is generated upstream as the downdraft is downstream.
This is illustrated in the following Figure, and can also be deduced from Figure 3.24 b.
The reason for the lack of symmetry in Figure 3.20 can be found in the upstream boundary condition that
is present in the open-jet wind tunnel, which differs from the model used in the previous discussion. The
flow field produced in the wind tunnel is a jet of finite extent and thus does not extend from −∞ to +∞ as
assumed in the previous description. In the open jet wind tunnel case, the flow potential and its gradients

G. Eitelberg 44 Delft University of Technology

Figure 3.21: Circulation superposed on horizontal homogeneous free stream results in a symmetry between
the down- and updraft. (Figure from http://galileo.phys.virginia.edu/classes/311/notes/aero/node2.html)

(i.e. velocity components) are prescribed (determined) at the exit plane of the nozzle contraction.

Approximating a 2D lifting surface in those conditions as a potential flow singularity, expressed through
circulation Γ, requires consideration of the distance between that lifting surface and the exit plane of the
nozzle contraction. This gives rise to a correction on the effective angle of attack of the lifting surface, as is
explained in the following sequence of approximative treatment.
We us the standard expression for sectional lift in 2D potential flow, as
L
= ρu∞ Γ (3.42)
s

with s denoting the spanwise distance, and obtain the standard description for the induced flow ui at any
th
distance r from the singularity; in our case from the 14 chord location of the airfoil.

Γ
ui = (3.43)
2πr

On the other hand, one can find the relationship between the value of the circulation and the lift coefficient
cL of a thin airfoil (not accounting for separation):

L cl ρu2∞ sc 1
Γ= = = cl cu∞ , (3.44)
sρu∞ 2sρu∞ 2

Where c denotes the chord of the airfoil,

In order to relate the lifting performance of the lifting surface in the open-jet wind tunnel to that in the ideal
situation in an unbounded flow field, the change in inflow conditions imposed by the boundary condition at
the nozzle exit plane needs to be determined. The situation is sketched in Figure 3.22.

In an unbounded flowfield, the velocity induced by the lifting surface at the location of the nozzle exit plane is
obtained by combining equations 3.43 and 3.44, making use of the thin airfoil approximation in 2D, cl = 2πα:

cL c α c
ui = · · u∞ ∼
= · · u∞ (3.45)
4π r 2 r

The distance r denotes here the distance between the nozzle exit plane and the lifting object in the open
test section and the angle of attack α is the one in the potential flow frame, in unbounded flow the geometric
angle of attack.

G. Eitelberg 45 Delft University of Technology

Figure 3.22: Schematic drawing of lifting surface in open-jet wind tunnel and induced velocity at nozzle
exit plane by that lifting surface. The prescription of a straight inflow at the nozzle exit plane requires
compensation of this induced velocity through a modification of the effective angle of attack. Lifting surface
not drawn to scale. The streamline angularity which would account for the induced component is noted as
αi .

In the open-jet wind tunnel, however, the flow is straight at the nozzle exit plane. Accounting for this nozzle
exit plane condition as a requirement for the entrance plane of the control volume, and matching that to the
potential flow model of the lifting surface, requires a rotation of the effective inflow that the lifting body is
exposed to. This rotation is equal to the induced angle of attack that follows directly from equation 3.45:

ui α c
αcorr ≈ = · (3.46)
u∞ 2 r

This last expression allows one to determine the effective angle of attack of the airfoil – it is the geometrical
angle of attack minus the induced angle of attack imposed by the boundary condition at the nozzle entrance:

αef f = αgeom − αcorr (3.47)

The correction for the induced angle of attack (equation 3.46) is directly proportional to the geometrical
angle of attack; hence the linear deviation of the two cl curves in Figure 3.20 it is also inversely propor-
tional with the distance of the airfoil from the nozzle exit plane. In the case of the nozzle exit plane being
at infinite distance upstream, the standard potential flow solution will be obtained. The last observation re-
sults in a clear recommendation not to test too close to the nozzle exit in open jet test section configurations.

As an illustration of the principle of the correction, the above Figure 3.21 has to be split into two. The first
part represents the nozzle exit plane where the inflow condition to the control volume is still horizontal and
homogenous. The second part adapts the potential flow field to that boundary condition.
The above approximation was simplified for a lifting body with zero lift at zero angle of attack. For cambered
airfoils like the one illustrated in the Figure 3.20 above the zero lift angle of attack will have to serve as the
reference instead of α = 0. Furthermore, the modelling relies on the small perturbation approximation, like
all corrections. This means that the rotation angles remain small, that the sin(αcorr ) = αcorr and tan(αcorr )
= αcorr and that cos(αcorr ) = 1.
In practice, it is customary to express the interference through the non-dimensional parameters of
ui
ε= (3.48)
u∞

G. Eitelberg 46 Delft University of Technology

Figure 3.23: The potential flow field when the control volume inflow plane in the proximity of the circulation
coincides with the nozzle exit plane. The rotation of the field or the induced vertical component depends on
the distance r between the exit plane and the lifting object as given in eq. 3.46 above.

(a) Reflection of a lifting vortex on the wind tunnel floor with a resultant horizontal
velocity component.

(b) Anti-symmetric reflection of the lifting vortex on the free shear layer with a
resultant vertical velocity on the undisturbed boundary.

Figure 3.24: Lift corrections for closed and open test section. [9]

where ui is the axial flow component induced by the wall/shear layer boundary condition, and
wi
∆α = (3.49)
u∞
where wi is the wall perpendicular (vertical) velocity component induced by the boundary conditions. Or

G. Eitelberg 47 Delft University of Technology

even the derived quantities such as the upwash interference parameter
C
δ = ∆α (3.50)
SCL
where C is the cross sectional area of the wind tunnel, S is a wing surface area and CL is the measured lift
coefficient. In this way change of angle of attack, the ‘upwash’ is normalised with the relative size and the
lift coefficient of the model. The streamwise gradient of it is defined as

∂δ
δ1 =   (3.51)
x
∂ βL

√
where β is the compressibility factor β = 1 − M 2 . Further, the same potential flow approximation can
be applied to the displacement effect of the model, where the solid displacement can be obtained by a
judicious application of source/sink doublets as shown in the illustrations in figure 3.25 below.

(a) Displacement effect correction for the closed test section.

(b) Displacement effect correction for the open test section.

Figure 3.25: Image reflection to satisfy the boundary condition for displacement effect. [9]

For the 2D lift interference, when the wing is approximated as a line vortex with the potential
 
Γ βz
φ= atan (3.52)
2π x

the integration over all reflections and from infinity upstream to infinity downstream can be performed to
obtain the results for the upwash interference as graphically presented in figure 3.26. Note, that the upwash
is negative upstream of the singularity and positive downstream. This is of importance, when not only the
lift coefficient, but also the pitching moment needs to be corrected, because the lift correction introduces
another error in the moments. Note the importance of positioning of the model representation. In the 2D

G. Eitelberg 48 Delft University of Technology

 Figure 3.26: Interference in flow angularity normalised with the lift coefficient. [9]

case, the area ratios are replaced by the ratio of the wind tunnel height and the chord length. For the
streamwise interference a remarkable result is obtained as there is no interference along the centreline of
the tunnel, if that is the location of the lifting surface! The strongest interference occurs at the location of
the tunnel walls. This can be explained by the opposing signs of the induced velocities of the reflected
circulations, see figure 3.27. Further cases can be obtained from the literature. All these rely on the

Figure 3.27: Interference in free stream velocity normalised with the lift coefficient. [9]

principles of linear superposition, which means that only small perturbations are permissible. An example
of the importance of corrections for a 2-D measurement, where a more than 50% error is shown to be
correctable is illustrated by Barlow, Ray and Pope in figure 3.28.
Question: what happens to the stability mechanism that is represented by the tail plane?
   
1 h Γ Γ
uinterf erence ≈ − (3.53)
2π l l l

The total interference behaviour as a function of the tunnel aspect ratio can be plotted as in figure 3.29. The
same principal approach as applied to the 2D corrections can also be applied to finite width lifting bodies.
The line singularities are then replaced by horseshoe vortices. The handling of those requires a more
complex mathematical approach, but the principles of reflections are maintained. Corrections for massively
separated wakes rely on the quadratic relationship between drag and lift (See Barlow Rae and Pope) and
make use of the relationship

CD ∼ CD0 + CL2 (3.54)

G. Eitelberg 49 Delft University of Technology

(a) Lift characteristics for NACA 0012 airfoil section uncorrected (b) Lift characteristics for NACA 0012 airfoil section corrected for
for tunnel interference. tunnel interference.

Figure 3.28: correction of the lift curve in 2D testing. [14]

Figure 3.29: Interference factors for classical correction for a small model in the centre of the wind tunnel. [9]

which can be plotted as a straight line against the quadratic lift coefficient. Where the measured relationship
deviates from the above relationship, separation (stall) is assumed to be responsible and a strong wake
ensues. This wake can be modelled as one caused by a combination of sources and sinks. The velocity
Q
increment in a tunnel of width B and height h can be expressed as ∆u = 2Bh , where Q is the source
strength. This can be related to the drag and the wing aspect ratio Span/Chord (S/C):

∆u S
ε= = CDuncorrected (3.55)
u 4C
In the literature, this is further split into the components due to the constant viscous drag, the induced drag
quadratic in lift, and the wake blockage:

S 5S
εwb = CD0 + (CDuncorrected − CDi − CD0 ) (3.56)
4C 4C
The term in the brackets vanishes, when there is no separation, leaving us with one of the standard wake
blockage corrections. There are other similar approaches, but this, known as the Maskell correction, is
widely used.

G. Eitelberg 50 Delft University of Technology

3.2.4 Thrust and Drag corrections
A special case of wind tunnel corrections is testing when real momentum and/or mass sources and sinks
are encountered in a wind tunnel, such as engine simulators or propellers. The corrections of these sources
and sinks are often derived from the ‘actuator disk model’, in which a propeller or turbine is assumed to be
an infinitesimal thin plate that either adds or subtracts momentum from the flow. In the case of a propeller
a graphical interpretation of this model is shown in figure 3.30. The actuator disk model maintains mass

(a) Control volume for analysis of a propeller. (b) (u, x) and (p, x)-diagrams.

Figure 3.30: Actuator disk model for propeller. [22]

conservation (while adding momentum) and therefore the continuity equation remains valid, thus:

S∞ u∞ = Sp up = Se ue (3.57)

If a momentum source or sink is introduced in the wind tunnel measurement inconsistencies will occur, as in
normal operation there are no sources and sinks within the wind tunnel measurement volume. Because of
these the flow within the tunnel is either increased, in the case of a momentum source, or decreased, in the
case of a momentum sink. There are two models that are often used for these thrust and drag corrections.
The first goes back to Glauert, and relies on obtaining an equivalent free stream velocity for the given thrust,
where the thrust is estimated from the momentum equation for the free stream on the one hand and the
combination of pressure and momentum consideration in the closed wall wind tunnel on the other. For the
conventional actuator disk model, the ratio of the corrected velocity over the tunnel velocity is expressed as:
S
ucorr Tc p
= 1 − √ Swt (3.58)
u 2 1 + 2Tc
The last expression is derived for the conventional actuator disk model, whereas in the further modelling the
contraction behind the propeller will be neglected. For large Tc ’s the above expression yields a correction to
the free stream velocity that is proportional to the product of the area ratio and the square root of the thrust
coefficient.

The other method, somewhat simpler but yielding the same result views the propeller as a singularity. For
the far field, the propeller acts as a sink with strength Q:
Q
φ= ln(r2 ) (3.59)
4π
Resulting in velocity correction
Q
∆u∞ = (3.60)
4πr · r

G. Eitelberg 51 Delft University of Technology

In the above equation the second r is to obtain a 3D axisymmetric description instead of the usual 2D line
sink/source description. The strength of the sink is obtained from the contraction of the inflow stream tube:
r !
8
S∞ − Sp = Sp 1 + Tc − 1 (3.61)
π

r !
8
Q = u∞ (S∞ − Sp ) = u∞ Sp 1 + Tc − 1 (3.62)
π

From the last expression, a correction velocity to the far upstream field can be calculated as
q 
u∞ Sp 1 + π8 Tc − 1
∆u∞ = (3.63)
4πB · H
Here the velocity correction is assumed to be an additional component to the free stream velocity in a
constrained cross section of dimensions H and B, and the free stream is assumed to be uniform over that
cross section. Again, as in the Glauert method, for large Tc the product of the area ratio and the square root
of the thrust coefficient determine the correction. The above approach means, that in a constrained wind
tunnel, the upstream velocity at a given thrust is higher than it would be in an unconstrained inflow, thus the
corrected velocity for the measured thrust for unconstrained space is

ucorr = u − ∆u
ucorr ∆u (3.64)
=1−
u u
It is customary to plot this relationship in the inverted way, as shown in figure 3.31. This corresponds to
the perception that for the measured thrust will be lower than that in an unconstrained space. This can be

u
( ucorr )

thrust coefficient (Tc )

Figure 3.31: Thrust correction.

evaluated to provide a corrected free stream velocity, which for a given value of thrust results in the same
axial velocity at the propeller location as that observed in the wind tunnel. Here, the corrected velocity ucorr
is the sum of the tunnel velocity and the area ratio is a parameter. Yet another alternative approach can
be taken to compare the pressure effect upon the back plane of the propeller. Due to the acceleration of
the flow in the slipstream, the conservation laws require the flow outside of the slipstream to slow down
to maintain continuity between the tunnel walls. Again, since the slipstream velocity is a function of thrust
(effectively, it is a consequence of the inflow stream tube contraction), a relationship between the additional

G. Eitelberg 52 Delft University of Technology

back pressure and the additional thrust can be formulated. In both cases, for large Tc the correction term
remains proportional to the surface area ratio of the propeller to the tunnel cross section and also to the
square root of the thrust coefficient. The conventional model sets restrictions to the contraction in the in-
compressible approximation.

The inverse approach can be taken to correct for a windmill test in a closed test section. In a windmill case
one can assume that the inflow is unadulterated; the windmill acts as a source for the downstream part of
the actuator disk, leading to an expanding wake. If we adapt the symmetrical approach, then the additional
(positive) velocity far downstream of the windmill will be the same as in the inlet of the propeller:
q 
u∞ Sp 1 + π8 Dc − 1
∆u∞ = (3.65)
4πB · H
On the other hand, we know that the drag of the windmill can be expressed as the product of mass flux with
the velocity increment, here deceleration.

D = ṁ∆u (3.66)

For a given windmill drag, with a fixed inflow condition determined from the free stream parameters, one
can express the velocity increment as

∆u = u∞ − ue (D) + ∆u∞ (D) (3.67)

where both the last terms are functions of the (measured) drag. The second term is the slipstream velocity
that would establish itself in the unconstrained space; the third term is that correction of the slipstream
which is due to the constrained space. For a given drag, this would also mean that the measured drag
corresponds to a free stream with a velocity

ucorr = u − ∆u∞ (D) (3.68)

All of the derivations above have been performed for a closed test section, however it can also be done for
an open wind test section. For a propeller, with a contraction in the stream tube upstream of the propeller
plane, this means that ambient air has to be entrained from the plenum area, where the air is assumed to
remain still. The amount of air is determined from the strength of the sink, as was described in the case of
the closed wind tunnel:
r !
8
Q = u∞ (S∞ − Sp ) = u∞ Sp 1 + Tc − 1 (3.69)
π

We now assume, that the work done to entrain that fluid can be ascribed to the inflow stream tube of the
propeller actuator disk, and thereby the stream tube averages between the entrainment form the ambient
still air and the air that exits the wind tunnel contraction. In reality, there is also a layer of wind tunnel air
between the inflow stream tube and the propeller. The simplest approximation accounting for that is to
assume that the inflow velocity condition in the stream tube is a weighted average:

Stunnel
uinf low = u∞ q 
Stunnel + Sp 1 + π8 Tc − 1
(3.70)
1
= u∞ q 
Sp
1+ Stunnel 1 + π8 Tc − 1

The measured thrust would then be

T = ṁ∆u (3.71)

G. Eitelberg 53 Delft University of Technology

where the mass flux ṁ is viewed from the unadulterated slipstream

ṁ = ρu∞ Sp (3.72)

And the thrust then

 
1
T = ṁ∆u ue − u∞ (3.73)
 
q 
Sp
1+ Stunnel 1 + π8 Tc − 1

This thrust can be compared with the thrust that would correspond to the unbounded domain, where ∆u =
ue − u∞ . From this comparison, the velocity correction or the thrust correction can be obtained: for a given
thrust, the tunnel inflow velocity has to be compensated for the sink effect of the propeller. We measure
thrust that corresponds to a lower inbound velocity than provided by the tunnel. Or, inversely, the measured
thrust in an unbounded free jet would be smaller than it is obtained in a bounded free jet wind tunnel, see
figure 3.32. Tunbounded < Tbounded

u∞
Figure 3.32: (Tc , J)-diagram for a propeller. (J = ωD )

Next, the question can be turned around for a windmill. Here the assumption is then made of an unadulter-
ated inflow and the source effect in the wind turbine wake slipstream. For the windmill case in a bounded
free jet, we return to the stream tube approximation again. In this approximation we ignore the upstream
influence of the wind turbine and assume that all the work is again done by the suction side. From experi-
mental evidence we conclude that this approximation is about as far from reality as is the approximation of
even distribution of the deceleration before and after the actuator disk, see figure from MEXICO project [5],
th th
where approx. 13 of the deceleration is done before the turbine and 23 is done downstream of it. Impor-
tant is that the assumption of velocity continuity across the disk is maintained and the drag is given by the
pressure jump across the turbine plane as actuator disk. In this case, we can apply the Bernoulli equation
from the rotor downstream, assuming that in the far wake the ambient pressure prevails.

u2∞ u2
ρ + p2 = ρ e + p∞ (3.74)
2 2
Together with the incompressible mass flux conservation, we can express the slipstream area far down-
stream as
1
Se = √ Sp (3.75)
1 − Dc
u2
Where the drag coefficients are defined as Dc = SpDq∞ , with q∞ = ρ 2∞ . Note that the thus defined drag
coefficient Dc cannot exceed 1! (For similarity with the propeller, where the expanded part is the reference

G. Eitelberg 54 Delft University of Technology

Figure 3.33: Deceleration of the flow due to wind turbine. PIV measurements vs. AD model. (Figure from
DNW)

p u2
Se = 1 − Dce Sp , where qe = ρ 2e .) The second expression is just to demonstrate the similarity between
the propeller producing thrust and the turbine producing drag; it is not customary to normalise drag with
the velocity in the slipstream. With the above equation, the source strength for the wind turbine can be
determined:

Q = u∞ (Se − Sp ) (3.76)

This is the flow that is forced out of the bounds of the open jet configuration into the nominally still air at
ambient conditions. It is obvious that this is more work than forcing the air into co-flowing stream. When
we again, form a weighted average of the slipstream and the ambient flow (actually just assume that the
stream-tube has been widened by the influence of the source) then we can define a corrected free stream
velocity
1
ucorr = u∞   (3.77)
Sp
1+ √ 1 −1
H·B 1−Dc

This corrected velocity can be used for determining the corrected tip speed ratio as well as the corrected
drag coefficient belonging to the corrected tip speed ratio. This corrected free stream velocity accounts for
the fact, that we require the wind tunnel circulation to conserve mass. In the end, all the flow exiting the
wind tunnel has to again enter it. We account for that by reducing the flow rate out of the contraction to
match the flow rate entering the collector. In reality there is also a change in the plenum pressure due to
the spreading (source) effect of the wind turbine that has to be accounted for in the flow reference system,
but for a first approximation the proposed methodology will do. The new tip speed ratio will then defined as

πωD
λ= (3.78)
ucorr
and the new drag coefficient will be

D
Dc = u2corr
(3.79)
Sp ρ 2

This means, that the measured drag appears lower than it would be in an unbounded flow field, Dunbounded <
Dbounded , see figure 3.34.

G. Eitelberg 55 Delft University of Technology

 measured
corrected

CD ∼ D

1
λ∼ u

Figure 3.34: (CD , λ)-diagram for a wind turbine.

G. Eitelberg 56 Delft University of Technology

4 Reynolds-number effects
This lecture draws heavily on the material prepared by Bram Elsenaar for DNW, which again is based on
the work done by himself and his co-authors and reported in: A. Elsenaar, T. Binion, E. Stanewsky and H.
Hornung for the AGARDograph 303.
In aerodynamic testing the Reynolds number is frequently considered of utmost importance. Since almost
all aerodynamic tests are scaled tests with the Reynolds number not identical to the real application, it
is assumed that this number can explain a lot of the differences between the flow over objects in tests
and in free flight. The effects that accompany variation of the Reynolds numbers are called the Reynolds
number effects; in fact, the correct expression would be the effect of varying the Reynolds number. Note
that there is no such thing as the Reynolds effect; this would be the case if there was only a singular effect
associated with the researcher Reynolds. Here we are discussing the effect that varying the Reynolds
number has on the flow around airfoils. Though many differences between scaled experiments and free
flight can be identified, there are quite a few phenomena that are not necessarily related to the change in
flow development with Reynolds number for a particular configuration in free air. One could refer to these
effects as ‘pseudo Reynolds number effects’, which will be discussed later in this chapter.
As mentioned above, Reynolds number effects are phenomena that are related to the change in flow
development with Reynolds number for a particular configuration in free air. In chapter 2 the Reynolds
number was defined as the ratio of the inertial force with the viscous force in a flow field. In the case that
a pressure gradient is absent the Reynolds number determines the characteristics of the boundary layer.
The relative boundary layer thickness is an inverse function of the Reynolds number and the shape of the
function depends upon the BL character, being either laminar or turbulent. Generally the transition from
laminar to turbulent flow in a BL occurs at a certain Reynolds number; however what the value of that
Reynolds number is, depends on measures taken to make sure that transition is occurring earlier or later.
In the following discussion, three groups of aeronautical simulation situations are addressed that have to be
treated with particular care when discussing the influence of the Reynolds number. These groups are:
• Airfoils and high aspect ratio wings with flows that are essentially two-dimensional or very weakly
three-dimensional, such that local strip theory still applies.
• Low aspect ratio wings with a highly three-dimensional flow development characterised by free vortex
flow, highly skewed shock waves and/or three-dimensional separated regions
• Slender bodies with free-vortex flow development and/or base flow interacting with a jet or plume.
The list of Reynolds number effects is substantial and some of these effects are listed below:
• Transition to turbulence (Tollmien-Schlichting waves)
• 3D transition mechanisms (e.g. cross flow instabilities)
• Leading edge and trailing edge separation
• Vortex separation
• Shock wave-boundary layer interaction
• Interaction between the above mentioned effects.
One of the rational reasons for dealing with those issues is the prediction accuracies for practical application-
s. While the fundamental research is quite often satisfied with identifying and understanding phenomena,
in the engineering applications of predicting flight performance, better than 1% accuracy is required for de-
termining the competitiveness of a product/aircraft. Since the verification of an aircraft design takes place
in a wind tunnel, the remaining uncertainty to be dealt with, is the effect of scaling. Current airplanes fly
in the speed domain, where the compressibility and unsteady effects cannot be neglected. They provide

G. Eitelberg 57 Delft University of Technology

additional influence parameters in addition to the Reynolds number. See the flight domain description in
figure 4.1, where the domain limits are described in parameters other than the Reynolds number. In most of
the possible flight domain, the limit is given by the maximum lift coefficient CLmax , but also other phenomena
are significant.
It is important to recognize that in terms of fractional analysis, the lift coefficient CL is another ratio of
forces, just like the Reynolds number is. In an analogous manner the, Mach number could, for the sake of
consistency of approach, also be approached with the help of fractional analysis. It could then be squared
and viewed as a ratio of inertia force and the pressure force (see also the discussion of the Euler num-
ber). Since, however, this only works for compressible media, denoting the Mach number as a measure of
compressibility is also a valid approach.

Figure 4.1: (L, M )-diagram for transonic testing. Dashed line represents the operating conditions and the
solid line represents the limits.

It can be seen that near the cruise condition the drag rise and the buffeting (irregular oscillations of part of
an aircraft due to periodic variation of the aerodynamic forces) phenomena need to be considered. Both
phenomena are related to the local supersonic flow being terminated by a shock. On the other hand the
maximum lift coefficient, CLmax , is the limiting factor for lower velocities such as during take-off and landing.
As the behaviour of a aircraft in free flight is different from the model in the wind tunnel it is necessary to
clarify how these limiting boundaries translate from wind tunnel test conditions to free flight. To do this an
understanding of scaling effects is required.

When testing aircraft models in a wind tunnel it is important to identify different wing regions and profile
characteristics as different regions react differently to the flow as proposed in figure 4.2.
The Figure 4.2 separates the wing and the wing root into regions that may have different dependencies upon
the Reynolds number. Whereas perts of the wing can be treated almost as 2-dimensional flow cases where
the transition Reynolds number dominates the flow behaviour, the tip and root regions are dominated by the
formation of root and tip vortices responding to different Reynolds number dependencies. So extrapolation
to flight conditions is different for different regions. Different components of the aircraft may exhibit different
sensitivities to the variation of the Reynolds number. This sensitivity analysis is most important for the
determination of the CLmax , the cruise drag and the drag rise that occurs when the aircraft is flying close to
the speed of sound.
In case of the determination of the CLmax one can see that the unsteadiness of the flow to increases
when measurements are performed close to the maximum lift coefficient as illustrated in figure 4.3. This
unsteadiness can be observed at low speed as well as in the transonic regime when it occurs as buffeting.

G. Eitelberg 58 Delft University of Technology

 (a) Various wing regions. (b) Profile characteristics.

Figure 4.2: Isolation of wing in various regions with varying profile characteristics.

Which detailed phenomena are involved and how this is affected by the global Reynolds number, will be
discussed later, here a characteristic illustration of the unsteadiness of the stress measured by a strain
gauge at the root of the wing is used as an illustration of the boundaries that need to be considered.

Figure 4.3: Unsteadiness of lift coefficient near CLmax

The second of those limiting conditions is the drag rise or drag creep. This is the limiting effect for cruise
performance and thus indirectly also for the fuel efficiency of an aircraft. The determination in an experiment
of the drag creep is not always straightforward and is usually based on a number of indicators. The indicators
used to determine the drag creep boundary is illustrated in the figure below for three indicators. These are
listed below and visualised in figure 4.4.
• The lift divergence observed at constant angle of attack for increasing Mach number.
• The angle of attack divergence at constant lift coefficient.
• The change in the trailing edge pressure indicating separation.
The divergences of the angle of attack and of the lift have to do with the transonic conditions and the
change of the shock position as a result of the boundary layer characteristics. Some of the ways to avoid
misinterpreting the experimental results due to the scaled Reynolds numbers will be covered in the following
sections. First the limits and the governing phenomena for the 2D airfoils and the large aspect ratio wing

G. Eitelberg 59 Delft University of Technology

 (a) Lift divergence. (b) Drag creep.

Figure 4.4: Occurrence of lift divergence and drag creep close to sonic conditions.

configurations will be treated. These airfoils and high aspect ratio wings behave in a manner analogous to
the 2D phenomenology. Later on low aspect ratio phenomenology will be introduced.

4.1 Reynolds number effects on airfoils and high aspect ratio wings
4.1.1 Background
The Reynolds number effects of airfoils and high aspect ratio wings are grouped together as in both cases
the flow experienced over the airfoil or wing is practically two-dimensional or very weakly three-dimensional
(such that local strip theory, essentially the negligibility of the 3rd dimension, still applies). In figure 4.5 the
influence of the boundary layer character, and thus also the Reynolds number, upon the flow over an airfoil
in transonic speed range is schematically illustrated.
The Reynolds number affects the relative location of the transition of the boundary layer from laminar to
turbulent and consequently the thickness of it under the shock wave as well as at the trailing edge of the
airfoil. The dependence of these three geometrical parameters upon the Reynolds number are termed as
direct Reynolds number effects. The consequent changes in the flow field, like the position of the shock
as a consequence of the change in the circulation due to the thickness change of the trailing edge, are
referred to as indirect Reynolds number effects. In this figure separation and stall are not shown, nor are
the different mechanisms depicted in detail. In all of these direct Reynolds number effects the Mach number
is another governing parameter as it determines the pressure effects. See the discussion of the viscous
interaction parameter and the SWBL interaction in the chapter about scaling and similarities. Some of these
more detailed issues will be dealt with in the following sections.

G. Eitelberg 60 Delft University of Technology

 Figure 4.5: Graphical interpretation of various Reynolds number effects.

4.1.2 Transition
When designing aircraft, knowing where and how transition occurs is important because it affects the drag,
and thus fuel consumption, of an aircraft significantly. In figure 4.6 the skin friction coefficient as a function of
Reynolds number is described for both laminar and turbulent flows. In this figure the effect of the Reynolds
number on the skin friction coefficient has been shown for a flat plate without longitudinal pressure gradients.
From the graph it can be seen that there is a range of Reynolds numbers with a simultaneous possibility of
having both laminar and turbulent flow. Factors other than the Reynolds number determine whether either
laminar or turbulent flow will occur. It can be seen that the turbulent friction coefficient is significantly higher
than the laminar friction coefficient which means that if possible, laminarity should be maintained for the
longest stretch possible in order to reduce the viscous drag over an object. Since it is inherently difficult
to maintain laminar flow for large Reynolds numbers and thus for a long stretch, only high quality testing
conditions allow testing the full extent of the laminar BL on an airfoil.
Typically a laminar boundary layer up to 60% of chord is possible with appropriate design of the airfoil for
the cruise Mach numbers. In good quality wind tunnels it is possible to demonstrate the laminarity of the
boundary layer, on specially designed laminar wing profiles, up to the shock induced transition.
The presence of the domain of simultaneous flow regimes is indicative of the importance to obtain and
quantify information about the turbulent characteristics of the free stream in a wind tunnel; in the flight
conditions the free stream is assumed to be disturbance free.
The above classical illustration was obtained in the 2D parallel straight flow conditions. The transition to
turbulence occurs via the Tollmien-Schlichting mechanism of instability. In the swept wing condition, the
flow can be much more complicated, as illustrated in figure 4.7. The component of the velocity parallel to
the leading edge of the wing for the swept wing is not always contributing to the pressure change, although
it can significantly contribute to the transition, through the cross-flow mechanism as will be discussed later.
In figure 4.7 the external ‘potential flow’ streamline is shown as smooth curve with no strong gradients in
it. In the boundary layer adjacent to the airfoil surface a completely different situation can be observed.
The streamlines can be ‘attached’ or ‘separated’ and their direction can strongly differ from the mean flow.
Not only is the flow direction affected by the sweep of the airfoil, also the transition to turbulence can
occur via different dominant mechanisms. A number of mechanisms is illustrated in figure 4.8. There is
enough evidence, though, that with modern civil transport aircraft, with wings at 37◦ sweep, the transition
is dominated by cross-flow instabilities or by the stagnation line contamination. The theoretical limit for the
Tollmien-Schlichting instability as a transition mechanism for non-zero pressure gradient boundary layers
(the Granville criterion) is preceded by other mechanisms in the case of sufficiently high Reynolds numbers.
Even for the Tollmien-Schlichting mechanism the Reynolds number is not the only transition criterion, also

G. Eitelberg 61 Delft University of Technology

 Figure 4.6: Laminar vs. turbulent friction coefficients for various Reynolds numbers.

the leading edge curvature, describing the rate of acceleration at a given Reynolds number, influences the
transition mechanism.
In figure 4.8 four Reynolds numbers are shown (Rx ) each of which signify different transition mechanisms
briefly described below.
R1 : At R1 the transition follows directly from the Tollmien-Schlichting instability.
R2 : At R2 leading-edge contamination followed possibly by re-laminarisation and then transition through
Tollmien-Schlichting instability.
R3 : At R3 leading-edge contamination followed possibly by re-laminarisation and then transition through
cross-low instability.
R4 : At R4 transition follows from cross-flow instability if leading-edge contamination were absent.
In realistic cases of current generation of airfoils, however, the transition in free flight occurs at approximately
5% of the chord length. Since the transition determines the characteristics of the boundary layer over the
rest of the chord length it is important to treat that part of experimental simulation with particular care. It
also affects the maximum lift performance, the stall characteristics, the buffeting and the separation at the
transonic shock location.

4.1.3 Flow separation as a Reynolds-number effect

It is possible to distinguish different Reynolds-number dependent mechanisms for separation. An illustrative
differentiation is between the separation that occurs in the low speed domain and the separation under the
influence of a shock wave. The separation can also occur without a shock wave at transonic conditions, but
then its mechanism is similar to the low speed one.
The boundary layer separation leads to loss of lift in the case of an airfoil. This loss of lift is called stall,
and there are a number of different avenues for stall to occur as illustrated for the low speed 2D cases (i.e.

G. Eitelberg 62 Delft University of Technology

 Figure 4.7: Schematic representation of streamlines at separation.

Figure 4.8: Various types of leading edge separation.

G. Eitelberg 63 Delft University of Technology

cases without shocks) in figure 4.9. Usually, stall behaviour is one of the dominant questions in landing and
take-off simulation. Few aircraft fly at the vicinity of stall in cruise conditions, whereas landing and take-off
almost always take place close to CLmax .

(a) Graphical interpretation of various types of stall for (b) Experimental data of various types of stall for the
different airfoils. same airfoil.

Figure 4.9: Different types of stall behaviour.

In the figure a clear distinction emerges between a thin and a thick airfoil behaviour. Thin airfoils tend to
stall at the leading edge first, though the definition of thin and thick itself is a Reynolds number dependent
category, as shown in figure 4.10. At low Reynolds number the chosen airfoil behaves like a thin one and
also displays a rather low value of maximum lift. As the Reynolds number is increased by an order of
magnitude from the baseline, the same airfoil displays all the characteristics of a thick airfoil with the stall
being initiated at the trailing edge. At the intermediate Reynolds numbers the CLmax increases continually
and the separation characteristics vary.
Furthermore, it can be observed, that for the different stall mechanisms, not only the Reynolds number
along the airfoil, but also the characteristic radius of the leading edge of an airfoil plays a role. Although
the radius of curvature can be used as the relevant length scale in the formulation of a second Reynolds
number, the ratio of the radial acceleration with the linear inertia force might be a more appropriate scaling
factor. The similarity of that is automatically given for geometrical similarity of models. In the figure below,
this similarity (nose bluntness) is expressed in terms of the ratio of the airfoil thickness at 1.25% of the chord
to the chord itself. One can observe that not only the pure forms occur, but also combinations thereof can
occur as well.
In figure 4.10 the domains of different separation regimes are indicated in a plot of leading edge curvature
vs. free stream Reynolds number. This figure is an impressive illustration of the fact that it is not the global
Reynolds number alone that determines the behaviour of the flow field.
The separation will become even more of an intricate interaction when the flow becomes transonic. The
shock wave-boundary layer interaction phenomenon was discussed in the lectures for dimensional analysis.
The dimensional analysis provided information and delineation about the separated domain boundary, but

G. Eitelberg 64 Delft University of Technology

 Figure 4.10: Various separation regimes.

not about the reattachment, i.e. the possible existence and size of a separation bubble. The shock wave-
boundary layer interaction is illustrated further in figure 4.11. The pressure rise downstream of the shock is
communicated upstream via the boundary layer, leading to an advance compression of the free stream and
a thickening of the boundary layer upstream of the shock. Thus it is a combination of viscous (Reynolds
number) and inviscid (Mach number) effects.

(a) Type A at expanding bubble stage. (b) Type B at early stage with localised (c) Type B at later stage with one sep-
shock bubble and rear separation. aration extending from shock to trailing
edge.

Figure 4.11: Shock wave-boundary layer interaction.

In the figure about the shock induced speration, the separation is split up into two classes. The type A
and the type B, both eventually leading to flow breakdown. A type A is one where the separation starts as
a bubble under the shock and eventually grows to a length exceeding the trailing edge distance from the
shock. This type separation start is a weak function of the Reynolds number and more dependent on the
Mach number.
In type B separation the shock induced separation starts with a bubble either simultaneously with a trailing
edge separation or is wholly dominated by a trailing edge separation which reaches upstream to the shock
position. Many rules exist for predicting the start of the flow breakdown. A rather simple one correlates the
pressure gradient at the trailing edge with the momentum thickness of the boundary layer at that location.

G. Eitelberg 65 Delft University of Technology

This criterion requires the product of these two characteristics to remain above a certain constant value:
 
Θ ∂cp
· > 0.007 (4.1)
c T E ∂ xc

where Θ is the momentum thickness of the boundary layer and thus Reynolds number dependent. This flow
breakdown criterion indicates that a given profile will experience an earlier breakdown at lower Reynolds
numbers. This means that the wind tunnel test will tend to over-predict the separation. The position of the
shock itself can react in a very sensitive manner to variations in boundary conditions both upstream as well
as downstream of it. Whereas for standard supercritical airfoils the shock position can be expected in the
vicinity of the 60% chord, in low Reynolds number (thick boundary layer) experiments the shock position
tends to move forward.

4.2 Reynolds-number effect testing

4.2.1 Requirements
The requirements toward testing vary dependent on the research or development target. While in the
phenomenological testing quite often a simple binary result will be satisfactory (either the phenomenon
does occur or it does not), in the aircraft performance predictions a very high accuracy is required in order
to determine the competitiveness of a product/aircraft. When testing the aerodynamic performance of a
product it is customary to talk about counts rather than a direct change in lift or drag coefficient. The reason
is that when fine tuning the aerodynamic performance the change in lift and drag coefficients are often quite
small and it is inconvenient to compare these small numbers.
One count in drag equals a change in drag coefficient of 0.0001 (∆CD = 0.0001 = 1 count). On the other
hand one count in the lift coefficient equals a change in lift coefficient of 0.001 (∆CL = 0.001 = 1 count). The
resolution in determining the angle of attack by inclinometers is ∆α = 0.1deg. When testing is performed at
a characteristic lift coefficient CL = 0.5, then this resolution contributes half a count in the drag resolution.
Example: If thanks to better design an aircraft has increased its maximum lift coefficient by three counts
and decreased its zero lift drag coefficient by 2 counts it would mean that ∆CLmax = 3 · 0.001 = 0.003 and
∆CD0 = −2 · 0.0001 = −0.0002.
In aerospace accuracies in the order of several counts are required to be able to determine wether or not
an aircraft will be competitive.
The high Reynolds numbers encountered in free flight cannot be duplicated in scaled experiments in atmo-
spheric wind tunnels. The Reynolds numbers are a function of scaling. The scaling effect can be at least
partially compensated in pressurised wind tunnels affecting the density of air and even more so in cryogenic
wind tunnels affecting also the viscosity in addition to the density. In the following subsections, methods and
techniques to compensate for the lack of the Reynolds number capability for airfoil and high aspect ratio
wing testing will be discussed, in addition to the pitfalls of the pseudo Reynolds number effects.

4.2.2 Simulating the Reynolds number by the technique of tripping

In general, the transition location can be brought to resemble the free flight case by ‘tripping’ the boundary
layer at an appropriate location. Tripping a boundary layer causes the boundary layer to transition from
laminar to turbulent flow if the ‘trip’ is designed and applied sufficiently well. A graphical interpretation of
tripping is shown in figure 4.12. In this figure it is indicated that adding roughness to the object the boundary
layer is forced to transition into turbulence and this results in faster growth of the BL thickness .
The effectiveness of this tripping needs to be validated by some observation technique, usually this is
performed by a visualisation technique like infrared thermography, but sometimes even simple tufts will do
the trick. In low speed atmospheric testing also just listening to the boundary layer noise can serve as a
good observation technique. The effectiveness of the application of boundary layer tripping depends upon
the flow conditions. Since in general only one tripping mechanism for the whole test entry is available, it has

G. Eitelberg 66 Delft University of Technology

 Figure 4.12: Boundary layer tripping.

to be checked for the lowest Reynolds number and usually at max lift conditions. A well optimised transition
strip should trigger the transition from a laminar to a turbulent boundary layer without additional detrimental
effects like:
• An extra increase in boundary layer thickness (an unwanted deterioration of the boundary layer).
• The generation of streamwise vortices similar to the effect of vortex generators that re-energise the
boundary layer
A rule of thumb for trip application is to use the trip Reynolds number, Rek .
kulocaledge
Rek1 = (4.2)
ν∞
Where k is the size of the grid used. In general good tripping conditions occur when the trip Reynolds
number is approximately 600, Rek ≈ 600. A more refined and also more physical version of the grid size
requirement is given in equation 4.3. In this definition of the trip Reynolds number the skin friction coefficient,
Cf plays a role.
The velocity has to be the free stream velocity. Note that the criterion using the square root of the skin
friction coefficient has an approximate value of the square root of the criterion 4.2. This is further explained
by Barslow [4].

q
Cf
kuedge local 2
Rek2 = ≈ 25 (4.3)
νlocal
An example of the effectiveness of the grit size is shown in figure 4.13, where the measured drag is com-
pared with a theoretical estimate for a fully turbulent boundary layer. In this figure, it is obvious that at higher
Reynolds numbers for the free stream, smaller trips become efficient. Large transition strips on the other
hand generate their own drag, which should be accounted for. It should be noted, that in addition to its ben-
eficial effect, the transition due to tripping can affect the result of a drag measurement rather significantly
as shown in figure 4.14.

4.2.3 Effect of tripping at different locations along the chord of the airfoil
The previous examples were used to demonstrate the effect of tripping the airfoil boundary layer near the
leading edge of an airfoil, around the location where it would occur in free flight Reynolds number condition-
s. An interesting alternative to tripping the boundary layer at the location where it occurs in free flight is the
possibility to trip at such a location as to obtain the relative boundary layer thickness corresponding to free
flight at the trailing edge of the model airfoil. The principle of this method, also termed aft fixation, is illus-
trated in figure 4.15. This approach can be recommended, when the trailing edge-dominated phenomena
are of significance, and when the Reynolds numbers are too low to trust the effectiveness of the leading
edge tripping in the experiment. We have applied that approach in the small tunnels of the TUD (M-Tunnel),
to avoid trailing edge separation.

G. Eitelberg 67 Delft University of Technology

Figure 4.13: Curve showing the effectiveness of the trip for various grid sizes. Dashed line represents the
turbulent approximation without tripping.

Figure 4.14: Variation of subsonic airfoil drag with Reynolds number.

The positioning of the strip forcing transition can lead to apparent shifting from one type of separation to the
other one as illustrated in figure 4.16, where the dependency of the maximum lift as well as the characteristic
shape of the lift curve upon the tripping location is shown for three different Mach numbers. If the maximum
lift is also influenced by the occurrence of the buffet phenomenon, then a more complicated development of
the maximum lift can be observed, see figure 4.17. For the lowest Mach number the shock is closer to the
leading edge and the turbulent BL has a detrimental effect.
Apart from changing the maximum lift coefficient of wings, tripping also affects the shock location. In figure
4.18 the sensitivity of the shock location to the location of tripping is also displayed. There are two different
locations obtained for the free transition case and three locations for forced transition case. In the forced
transition it is shown that tripping too far upstream will move the shock wave location upstream as well.
The mechanism responsible for the forward movement of a shock is the thickening of a boundary layer. If

G. Eitelberg 68 Delft University of Technology

 (a) Graphical interpretation of the method of aft fixation. (b) Experimental results by using the aft fixation princi-
ple.

Figure 4.15: Principle of simulation by aft fixation.

Figure 4.16: Effect of transition fixation on lift development on a supercritical airfoil for various Mach num-
bers.

the boundary layer is tripped the thickness of the boundary layer increases as turbulent boundary layers
are thicker, with respect to laminar boundary layers. Apart from being more prone to separation, these
thicker boundary layers cause the effective shape of the airfoil to change which reduces the circulation.
This phenomenon is often referred to as decambering.
However, one should alway pay attention to the set up, as shown in figure 4.19. It shows a comparison
between free flight data at Reynolds numbers of 20 · 106 in comparison with transition fixing at 30% of chord
when tested at Reynolds numbers an order of magnitude smaller. In this comparison two different angles
of attack are shown. One, where the transition tripping produces a pressure distribution very much like the

G. Eitelberg 69 Delft University of Technology

 Figure 4.17: Effect of transition fixation on maximum lift.

Figure 4.18: Effect of transition point changes on the pressure distribution of a supercritical airfoil.

one from the corresponding free flight test, whereas at a higher angle of attack (and thus more circulation)
the agreement significantly deteriorates.
The domain, where aft fixing might be recommended is given in figure 4.20, where the domain is separated
into 3 regions for forward, middle and aft fixing of transition. This sort of graph should only be used as an
indication and not a prescription. The experimenter will always have to verify the strategy and compare with
alternatives.

G. Eitelberg 70 Delft University of Technology

 (a) For low angles of attack the simulation holds. (b) At higher angles of attack the simulation varies
with free flight.

Figure 4.19: Comparison free flight data with simulations where transition fixing at 30% of the chord was
applied.

Figure 4.20: Forward, middle and aft-fixing domains.

4.2.4 Conclusion for airfoil and wing testing

The flow over an airfoil is influenced by the characteristics of the boundary layer in a number of ways. To
obtain behaviour characteristic to a turbulent boundary layer where the boundary layer in the experiment
would remain laminar, the tripping technique is a powerful tool. For transonic flow, just tripping near the
leading edge can occasionally be too much of a good thing in the low Reynolds number test. Since the
boundary layer thickness at the trailing edge relative to the airfoil chord will be thicker in the tripped experi-
ment than in the free flight, the circulation may be reduced and a significant deviation in the shock position
may result. A judicious application of tripping is required. The potential for overcorrecting is shown in figure
4.21, where the tripping at the fixed location in the wind tunnel tests led to a large under-prediction of the
performance of a transport aircraft.

G. Eitelberg 71 Delft University of Technology

 Figure 4.21: Under-prediction of real aircraft due to overcorrecting the transition location.

4.3 Highly swept wings

A separate class of testing is the testing of objects with highly swept wings, generally highly manoeuvrable
vehicles capable of going supersonic. A generic configuration is given in figure 4.22. In terms of aerody-

Figure 4.22: Delta wing model used in the following experiments.

namics delta wings represent a class of flying vehicles which can be described as having low aspect ratio
wings with a highly three-dimensional flow development characterised by free vortex flow, highly skewed
shock waves and/or three-dimensional separated regions. In this section experiments will be discussed in

G. Eitelberg 72 Delft University of Technology

which a pressure distribution measurement at the mid-chord location performed. In this experiment the flow
has not been tripped. From this experiment the importance of the Reynolds number for the leading edge
vortex development can be obtained. These tests were performed on the model shown in figure 4.22 at
two different wind tunnels, namely the High Speed Tunnel (HST) and the Transonic Wind tunnel Göttingen
(TWG). In order to test the effects of the Reynolds number on a delta wing a wide range of Reynolds num-
bers spanning one order of magnitude has been tested of which the results are shown in figure 4.23. in

(a) (Cp − η)-curve, α kept constant. (b) (Cp − α)-curve, η kept constant.

Figure 4.23: Reynolds number effect on the vortex generation of delta wings.

figure 4.23a that the peak in −Cp increases for higher Reynolds numbers. This decrease in Cp is caused
by vortex generation at the wing’s leading edge. As the vortex introduces and additional velocity compo-
nent the static pressure decreases as, according to Bernoulli the sum of the static pressure and dynamic
pressure must be constant. Though the vortex position is mainly determined by the free stream conditions
and the separation location by the curvature of the leading edge, a Reynolds number dependency is still
found in the size of the vortex. The size of the vortex is evidenced by the change in the pressure distribution
characteristics on the upper surface of the delta wing.
However, not all measured effects can be explained with the term Reynolds number effect as not all of
the highly complicated phenomena scale directly with the Reynolds number. In the case of delta wing
testing at transonic speeds this is shown in figure 4.24 It can be seen that doubling the Reynolds number
leads to fundamental changes in the flow features in the opposite direction that was expected according
to the previous figures. It was anticipated that an increase in Reynolds number would cause a suction
peak, whereas in this case suction almost entirely disappears as the Reynolds number increases, with the
exception of the suction peak at the leading edge of the wing.

G. Eitelberg 73 Delft University of Technology

 Figure 4.24: Vortex formation on fighter model with delta wings.

4.4 Slender bodies

Another object that experiences Reynolds number effects are slender missile shaped bodies, such as rock-
ets. These bodies are heavily influenced by cross-flow and depending on how these objects are moving
through the air different kinds of vortex flows occur as seen in figure 4.25.

Figure 4.25: Typical flow conditions visualised by the hydrogen bubble technique.

It can be seen that under high angles of attack the flow around a slender body behaves similarly to a
cylinder. This means that if a slender body is assumed to behave like a cylinder the several flow regimes
can be discerned that are shown in figure 4.26, where RD is the Reynolds number based on the diameter of
the cylinder. In figure 4.27 other flow characteristics about a circular cylinder are shown. In addition to the
cross flow effect occurring on all cylinders, the aspect of vortex separation on the side edges resembles that
of the highly swept wings. The same sort of transverse pressure gradient on both sides of the symmetry
plane produces a component of downstream directed vorticity on the upper surface with the opposite sign
to the equatorial plane vorticity, leading to converging streamlines between those two lines on the body.
The radius/diameter of the relevant geometry is a suitable length scale for the Reynolds number scaling
in experiments, if at all. The ratio of the angular to linear acceleration might have more influence on the
performance.

G. Eitelberg 74 Delft University of Technology

 Figure 4.26: Identification of flow regimes on a circular cylinder.

(a) Lift fluctuations (r.m.s.). (b) Strouhal number of the lift fluctuations.

(c) Drag coefficient.

Figure 4.27: Characteristics of the flow about a circular cylinder.

4.5 Pseudo Reynolds-number effects

There are other effects occurring in a wind tunnel simulation that can be erroneously ascribed to the
Reynolds number effect. These usually are due to the insufficient quality of the simulation like disturbances
in the free stream, model quality or even wind tunnel wall effects. Figure 4.28 illustrates the pitfalls of 2D
profile testing in a wind tunnel which inherently limits the span. The shift in shock location in this case not
ascribed to the inadequate Reynolds number but to the effect of the presence of wind tunnel side walls.
This, as well as the a priory variation of the presence of disturbances in the flow can influence the transonic
shock location. The previously encountered example of a swept wing aircraft or delta wing aircraft losing
some of the suction with increasing Reynolds number can also be included in the discussion of the pseudo-
Reynolds number effects. Care should also be taken in classifying the effects when comparing laminar

G. Eitelberg 75 Delft University of Technology

 Figure 4.28: Effect of sidewall boundary layer removal on two-dimensional airfoil pressure distribution.

transonic behaviour with turbulent transonic boundary layer behaviour, as the following two examples illus-
trate, see figure 4.29. From figures 4.29b and 4.29c one can see that the shock-boundary layer interaction
for a laminar boundary layer results in different flow conditions than for a turbulent boundary layer. As the
pressure distribution of the laminar profile is heavily dependent on Reynolds number, whilst for the turbulent
boundary layer a change in Reynolds number hardly has an effect on this pressure distribution.

4.6 Summary and recommendations

Not all variations in measurement results occurring with variation of the free stream conditions can be
ascribed to the Reynolds number effect. Many interactions are more complicated than that. Nevertheless,
some simple guidelines can be formulated when concerned with the applicability of low Reynolds number
measurement results to high Reynolds number free flight conditions. For a low speed test program it can be
recommended to test at constant Mach number since maximum lift conditions are Mach number dependent.
If that is not an option, apply a good Mach number correction to the results. For testing at transonic
conditions it can be concluded that the resulting data are very sensitive for Mach number variations. It is
illustrative to ‘map’ the flow characteristics in the CL −M number plane. Although testing is most often done
at specified incidence (α) conditions, it might help to analyse the results at constant lift, as this corresponds
with a constant circulation. For delta wing type of configurations, tripping can cause more unresolved issues
than it resolves. The flow is extremely sensitive to the angle of attack and no ideal tripping location can be
found. The only tripping that can be recommended is that on the pressure side, to maintain a constant
transition point there. Chord Reynolds numbers above 3 · 106 are recommended. Furthermore:
• Always check the transition strip at the most critical condition (e.g. lowest Reynolds number, near
maximum lift condition, etc.).
• Know the details of the flow at the strip location (possible shock wave formation, possible bubble
separation ahead of the trip, etc.).
• Remember that strips deteriorate in time.

G. Eitelberg 76 Delft University of Technology

 (a) Early example of the effect of flow quality on the pressure distribution (pseudo-Reynolds
number effect).

(b) Shock-laminar boundary layer interaction (Reynolds number (c) Shock-turbulent boundary layer interaction (Reynolds number
effect). effect).

Figure 4.29: Examples of (pseudo) Reynolds number effects.

G. Eitelberg 77 Delft University of Technology

5 Engine integration
In chapter 3 wind tunnel theory is treated together with various methods that correct for inaccuracies when
performing measurements in wind tunnels. Some of these corrections involved the actuator disk model, in
which a propeller or wind turbine is described as an infinitesimal thin disk that adds or subtracts momentum
from the flow. This chapter will go slightly deeper into the actuator disk theory and will also treat how one
simulates propulsion integration effects.

5.1 Thrust generation

Aircraft move forward because of the thrust generated by the engines. In wind tunnel testing thrust is also
a momentum source that adds momentum to the flow. Thrust is a force that is directed in the opposite
direction of the drag. In order to analyse thrust the momentum balance equation is used:

F~ext
X
ρ~u (~u · ~n) dA = (5.1)
A

This equation states that the net momentum flux across the boundaries of a control volume has to be
balanced by the sum of the external forces, see figure 5.1. The external forces in general are both the
forces applied to the mass as well as the forces applied to the surface. The latter include the pressure
force, which is sometimes forgotten.

ṁ

u∞

u∞ usl

u∞

ṁ
Figure 5.1: Illustration to the application of the balance of momentum across a control volume including a
propulsor.

The illustration above can be used to derive the expression for thrust, without any specific knowledge about
the thrust generation mechanism inside the control volume. The inflow plane is described by uniform flow;
the outflow plane includes the accelerated slipstream emanating from the propulsor. It is assumed that
the entrance plane and the exit plane are in sufficiently far field, such that the static pressure in those
planes is that of the ambient air. The pressure forces on the top and bottom (symmetrical) planes of the
sketched control surface can be ignored, because of the antisymmetric geometry. In the thus specified

G. Eitelberg 78 Delft University of Technology

case, all pressure forces acting as external forces can be neglected. However, the mass continuity of the
flow requires a sidewise inflow of fluid into the rectangular control volume as the velocity in the slipstream
is accelerated and it is assumed that the flow outside the slipstream remains u∞ .

ṁ = ρ∞ u∞ (S∞ − Ssl ) (5.2)

Here S∞ and Ssl are the cross-sectional areas of the stream tube through the propulsor upstream and
downstream respectively. This mass flux produces a momentum force according to the above integral
equation. The component parallel to the free stream, the x-component, of this force is

FMx = u∞ ṁ = ρ∞ u2∞ (S∞ − Ssl ) (5.3)

By summing all the momentum forces acting upon the control volume in the x-direction and including the
thrust force T , which compensates for holding the engine in place, it is possible to write:

ρ∞ u2∞ Sp + ρ∞ u2∞ (S∞ − Sp ) − ρsl u2sl Ssl − ρ∞ u2∞ (Sp − Ssl ) = −T

ρ∞ u2∞ S∞ − ρsl u2sl Ssl = −T (5.4)

Thanks to the continuity equation it is known that ρ∞ u∞ S∞ = ρsl usl Ssl = ṁT as the mass flow through the
stream tube must remain constant, which leads to:

T = ṁT (usl − u∞ ) (5.5)

Although the above equation was obtained from the momentum balance across all surfaces of the control
volume, the result contains only inflow and outflow conditions. It is customary to define the inflow part,
ṁT u∞ , as the ram drag and the outflow part, ṁT usl , as the gross thrust. This is a somewhat arbitrary
separation of the two components, because one would not exist without the other. In the case that a
turbofan is generating the thrust, the controlling parameter is the Fan Pressure Ratio (FPR), which will be
defined later.
If it can be assumed that the above description of thrust remains valid even when the engine is mounted
on an aircraft, i.e. the interaction between the propulsor and the airframe is negligible. In such a case, and
only in such a case, the propulsion efficiency η can be defined. This efficiency is the ratio of the work done
to overcome the drag of the aircraft without the engine and the amount of work done for to accelerate the
free stream flow to the slipstream velocity in the engine. If it is assumed that the thrust generated by the
propulsor is equal to the drag of the airframe, then the efficiency becomes:

T · u∞ 2u∞ 2
η= = = (5.6)
ṁT
2 (u2sl − u2∞ ) u∞ + usl 1 + uu∞
sl

Note that by applying this formulation to any other situation, such as to a Boundary Layer Ingestion (BLI)
or also wake ingesting configuration, will lead to confusing and even wrong interpretations of the engine
performance. The deviation from this idealised behaviour is usually defined as the engine interference and
is included in the drag bookkeeping for the aircraft.
When the control volume, as described above, is shrunk to an extent where the static pressure cannot be
considered equal to the free stream pressure along its boundaries everywhere, then the pressure forces
have to be included in the thrust calculations. This may be the case when the engine is mounted in a
accelerated/decelerated flow field or when the engine is viewed as an infinitely thin disk alone. The latter
case, the actuator disk model deserves a separate explanation.

G. Eitelberg 79 Delft University of Technology

5.1.1 Actuator disk model
Propulsion by a propeller can be considered as a limiting case of the above thrust consideration. In this
limiting case, the control volume is shrunk to an infinitely thin layer, called an actuator disk. The theory
behind the actuator disk approximation is treated in this subsection.
In the actuator disk model the conservation of flow parameters such as velocity and mass flux passing
through the propeller plane is assumed. However, the pressure is allowed to jump across the propeller
plane. This results in obtaining the stream tube going through the propeller plane for both the upstream
and downstream parts of the disk in such a way as to have an equal contraction of the stream tube on both
sides of the disk. The velocity of the fluid at the propeller plane is obtained as the average between the
free stream velocity and the slipstream in the far wake. The final velocity increment obtained is expressed
through the so called axial inflow factor a:
1 usl − u∞
a= (5.7)
2 u∞
However, the axial inflow factor a can also be expressed through the thrust coefficient Tc as done below.
r !
1 8
a= −1 + 1 + Tc (5.8)
2 π

Where the thrust coefficient Tc can be expressed as:

T
Tc = (5.9)
ρu2∞ D2
This thrust coefficient should not be confused with another representation of the thrust coefficient, cT . This
one, the Tc , is valid for all devices producing thrust, independent of the method to do it. The other one,
defined below, is specific for a rotor as a propulsor.
T
cT = (5.10)
ρn2 D4
This particular thrust coefficient can be related to the previous one by introducing the advance ratio, J = unD
∞
.
rev
In this case n is the angular velocity measured in s , or rps. The advance ratio is a similarity parameter
relating the translational forward motion of the rotor to the rotating motion of the rotor blade tip. Thus
cT = Tc · J 2 .
The actuator disk model makes no use of the rotating motion of the propulsor, only axial momentum is
considered.
The slipstream contraction can be found by applying the following equation obtained from the conventional
actuator disk approximation:
v
Rsl u 1+a
=t (5.11)
u  
R 1+a 1+ √ x R2 +x2

For an isolated propeller with no or negligible forward motion, J = 0, and constant cT , the Tc will approach
infinity, which results in a large axial inflow factor, a >> 1. When this is combined with x >> R,q which
means that the measurement is performed far away from the disk the contraction ratio is found to be 12 as
shown below.
r r
Rsl 1+a 1
= ≈ ≈ 0.7 (5.12)
R 1 + 2a 2
This requires a rapid contraction close behind the actuator disk. This approximation can be used to de-
termine the wing area wetted by the propeller slipstream when correcting for the “superlift” in wind tunnel
testing for propeller aircraft.

G. Eitelberg 80 Delft University of Technology

As mentioned previously, the mass flow and velocity remain continuous in the actuator disk model, but
a pressure discontinuity is allowed across the propeller plane. If this pressure discontinuity is permitted
as is compatible with the previous analysis of axial inflow conditions, then the assumption of the velocity
reaching its slipstream final value immediately after the actuator disk is a further permissible approximation.
This approximation is also plausible since there is no further acceleration mechanism behind the actuator
disk. (There is no vorticity generated in this approximation.) This is equivalent to shrinking the control
volume to the location of its exit plane in Figure 5.1.
As the velocity and mass flow are continuous, there is no increase in linear momentum across the infinitely
thin disk. Thus, the only way to generate thrust is through a pressure jump as shown in the equation below.

T = Sp (p2 − p1 ) (5.13)

Where Sp = π4 D2 is the cross-section of the actuator disk, p1 is the pressure upstream and p2 is the pressure
downstream of the actuator disk. Note that with this representation a positive thrust is against the direction
of the flow. For determining the thrust, no assumption about the distribution of the pressure jump across the
disk needs to be made. In the classical actuator disk model half of the pressure change occurs upstream
of the disk, also known as the suction side. The other half of the pressure jump occurs downstream of the
disk, or the pressure side. However, equally well, the same thrust can be ascribed to a suction side only, so
that behind the propeller the free stream static pressure is recovered immediately.
This approximation corresponds to the case where the pressure directly behind the actuator disk equals the
free stream pressure. Under such condition, neglecting the contraction of the stream tube downstream of
the actuator disk can be assumed a reasonable approximation. With these assumptions the jump conditions
at the location of the propeller could be applied.

p2 = p∞ (5.14)

Thus the pressure immediately ahead of the actuator disk can be formulated as below:
T
p1 = p∞ − (5.15)
Sp
When the incompressible Bernoulli equation is applied to the stream tube from far upstream to the plane
immediately in front of the actuator disk, we obtain :
1 1 T
p∞ + ρu2∞ = p∞ + ρu21 − (5.16)
2 2 Sp
or simply
1 2 1 T
ρu∞ = ρu21 − (5.17)
2 2 Sp
Further, assuming continuous incompressible flow up to the actuator disk, one can obtain the tube contrac-
tion ratio ahead of the propeller plane from the continuity equation.
S∞ u∞ = Sp u1
S∞ (5.18)
u1 = u∞
Sp
Inserting this result in equation 5.17 will result in:
 2
1 2 1 S∞ T
ρu∞ = ρu2∞ − (5.19)
2 2 Sp Sp
From which the contraction ratio in terms of areas can be obtained.
 2
S∞ T
=1+ (5.20)
Sp Sp ρu2∞

G. Eitelberg 81 Delft University of Technology

 π 2
And when introducing the actuator disk cross-section Sp = 4D .
s
S∞ 8 T
= 1+ (5.21)
Sp π ρu2∞ D2

It should be noted that the last term represents the thrust coefficient Tc , thus:
r r
S∞ 8 8 CT
= 1 + Tc = 1 + (5.22)
Sp π π J2

When comparing the above equations one can see that the area ratio is obviously related to the axial inflow
factor discussed earlier in this chapter in equation 5.8. For small advance ratios, the contraction ahead
of the propeller plane becomes large. In the case of u∞ = 0, S∞ → ∞. This trend is in agreement with
empirical observations. The limiting value for the contraction ratio could be obtained from the observation
that there is a limit to the thrust by achievable aerodynamic performance of the propeller blades. Equations
5.21 and 5.22 could be used for a scaling exercise for the elevation of the propeller above ground to such a
position that the outer limit of the stream tube just reaches the ground at ∞ upstream.
r r
h∞ 4 8 4 8 CT
= 1 + Tc = 1 + (5.23)
rp π π J2

5.1.2 Correcting for propeller slipstream effects for an isolated propeller

In a closed test section, the slipstream velocity will be higher than that of the free stream velocity, if positive
thrust is generated. Due to the conservation of mass requirement, this will mean that the flow outside of
the slipstream will have to be slower than the free stream upstream of the actuator disk. This effect would
not occur in an unbounded condition. Along with the reduced velocity, the static pressure increases along
the walls. This translates into increased static pressure also inside the slipstream. For a given propeller
setting, this translates into increased thrust, or thrust corresponding to a corrected free stream velocity. This
corrected velocity can be expressed as:

u0∞ 2Tc α1
=1− q (5.24)
u∞ π 1 + π8 Tc

Sp
Where α is the ratio of the of the disk area to the tunnel cross section area, α = St . A more extensive
discussion of this is included in the chapter about wind tunnels and wall corrections.
The actuator disk approach above is kept generic enough to be applicable to compressible and incompress-
ible flow situations. In the section dealing with the particular testing technique of turbofan engine effects,
the discussion will be refined to account for compressibility effects as well.

5.2 Propulsion interference effects on an aircraft

This section discusses the general interference effects between the engine installation and the aircraft, or
more specifically the wing. An illustration of the interference effects for a wing mounted arrangement of a
turbofan engine is shown in figure 5.2 and listed below:
• Geometrical elements such as cowl, pylon (inlet spillage) and jet interference.
• Wing pylon junction leading to loss of lift.
• Junction flow.
• Channel flow and fan exhaust interactions.
• Jet induced flow/separation.

G. Eitelberg 82 Delft University of Technology

 Figure 5.2: Interactions of wing mounted engine with the wing aerodynamics. [13]

A number of the interference effects also remain present when the aft fuselage mounted configuration is
chosen. Additionally the slipstream interaction with the tail-plane and control surfaces will then need to be
considered.
One way of splitting up the aerodynamic interference effects is looking at the geometry effects and the
power setting effects separately. Both of them affect the flow around the aircraft, and may amplify or
dampen each other. But they are easier to understand, when viewed separately. As geometry effects, the
channel geometry, the surface area and the cross section area of the engine might be classified.
The channel flow effect is present in all versions of podded engine configurations; both on the wing as well
as on the fuselage. An example for the fuselage mounted configuration is shown in figure 5.3. Here the
‘surface streamline’ pattern of the pylon flow is made visible with the help of a fluorescent dye visualization,
where viscous fluid embedding fluorescent paricles is spread in a thin layer on the solid surface. The oil
spreads according to the wall shear stress vector field forming a so-called surface streamline pattern. In the
current example the channel flow is generated with the help of the simplified engine simulator, a through
flow nacelle (TFN). This particular image has been obtained in DNW-HST. Part of the pattern could be
interpreted as indication of a separation bubble in the conditions representing flight idle settings.

Figure 5.3: Surface flow visualisation in the channel between the engine cowling and the fuselage. Possi-
bility of a separation bubble on the pylon fuselage junction.(Figure from DNW)

The ‘surface area effect’ can be estimated by calculating the engine cowling and pylon surface areas, and
multiplying those with an appropriate local skin friction coefficient Cf . For a flat plate in turbulent flow this

G. Eitelberg 83 Delft University of Technology

would be:
τw (x) −1
Cf = 2
≈ −0.03Rex 5 (5.25)
ρu∞
The cross section increase of the aircraft through engines can be accounted for by considering the influence
of the ram drag, which is proportional to the product of free stream velocity u∞ and the engine intake mass
flux ṁengine . At any given free stream velocity, the mass flux is then proportional to the intake cross sectional
area and thus the ram drag component increases in a straightforward manner. Note that in the wind tunnel,
when simulators are used, only the fan intake mass flux ṁF is considered. The sizes of the engines are
mainly determined by their bypass ratios. The core diameter of the turbine does not vary much in a given
thrust category.
The power effect can also be viewed from its two aspects. One, the more obvious effect is the one that can
occur in the configuration, where a large high bypass ratio engine is mounted close to the wing. Then the
fan exhaust can directly interact with the high lift elements, such as flaps, affecting the aircraft behaviour. In
figure 5.4 seen below, three different bypass ratio engine simulators are mounted at an identical distance
below the wing. In the case of the simulator for the Ultra High Bypass Ratio (UHBR) engine, the fan exhaust
flow is directly interacting with the wing.

(a) Engine simulators at DNW. (b) Effect of the bypass ratio on the en-
gine/wing interaction in case of constant verti-
cal distance between the wing and engine axis.

Figure 5.4: Engine simulators at DNW and their wing interference. (Figure from DNW)

The second, also fundamental power effect is the loss of lift as an interaction between the wing-mounted
engine and the wing circulation. This is shown in figure 5.5. In this figure the different configurations of
engines (or simulators of the engines) are shown with their effect upon the vertical wing loading, i.e. the lift
and compared with the clean wing, which is the dashed black line.
A special case of interference occurs in the case of wing mounted propeller engines in a tractor or puller
configuration. In that case part of the wing is submerged in the propeller slipstream, figure 5.6, experiencing
flow with a total pressure higher than that of the free stream. This exposure to the increased total and
therefore also the increased dynamic pressure affects the behaviour of the propeller aircraft and needs to
be accounted for in the wind tunnel testing. As the propeller is mounted upstream of the wing, the wing will
experience accelerated flow in the slipstream of the propeller. This causes the lift to increase locally. This
can both be beneficial or disadvantageous depending on the application.

G. Eitelberg 84 Delft University of Technology

 Figure 5.5: Effect of engine installation on span loading.

Figure 5.6: Effect of propeller slipstream on wing and tail. [9]

The propeller does not only cause the flow in the slipstream to locally accelerate, it also introduces swirl
to the flow which affects the wing lift distribution. The general rule of thumb is that the downward stroke
reduces the wing lift in the slipstream and the upward stroke increases it. The two propeller interference
effects are shown in figure 5.7. For the correction of lift values, it is usually sufficient to account for the
increased dynamic pressure in the slipstream only. The swirl effects on the downstream and upstream
stroke are approximately equal (when the wing is operating in the linear range of the lift curve) and can be
neglected in the accounting for required corrections.
Figure 5.8 gives a general estimation of the ’degree of difficulties’ when choosing the position oft the engine
relative to the wing. Here the difficulty is mainly determined by the desire to avoid detrimental interaction
between the airframe and the engine.
All the interactions in their sum can be counterproductive to its original purpose of increasing propulsive
efficiency by increasing the bypass ratio. The general rule is that higher the bypass ratio, the higher the effi-
ciency of the isolated engine. The installation interference effects can cancel this benefit. The contradicted

G. Eitelberg 85 Delft University of Technology

 (a) Propeller induced swirl. (b) Asymmetrical lift distribution due to swirl in
slipstream.

(c) Increase in lift due to increased velocity in

slipstream.

Figure 5.7: Effect of swirl in propeller slipstream on lift distribution. [9]

(a) Engine mounting parameters. (b) Plot showing the challenge of certain engine configura-
tions.

Figure 5.8: Degree of difficulties in the engine positioning challenges.

tendencies are illustrated in figure 5.9 in a qualitative manner. In this graph, the qualitative amount of fuel
consumption of an engine is plotted against its bypass ratio, and for an isolated engine this is the monoton-
ically falling curve. At the same time, the installation drag increases at such a rate that at a certain point
the total fuel consumption can start increasing again. This consideration is valid not only when designing
a new aircraft where detailed measures can be applied to remedy the situation, but also when re-engining
an existing aircraft, when in general fewer design degrees of freedom are present. In both cases, a careful
evaluation of the quantitative amount of interference can be obtained from suitably designed experiments.
Overall, the increase in efficiency of propulsion achievable through increasing the mass flow through the
fan and reducing the pressure jump across it can become counterproductive when the interactions are not
properly dealt with, as indicated in figure 5.9. Note that the graph is only qualitative since manufacturers
jealously guard their data.
Note: Flow non-uniformities on the entrance plane affecting the performance of the engines are usually not
considered in civil aircraft applications. It is assumed that the engines are placed in undisturbed flow. This
is not the case in military aircraft engine integration, where a lot of effort is dedicated to reducing the losses
in the inlet ducts.

G. Eitelberg 86 Delft University of Technology

 Figure 5.9: A schematic representation of fuel consumption reduction as a function of bypass ratio.

5.3 Engine simulation requirements

For the experimental study of the quantitative effects caused by interference the appropriate simulation
tools have to be defined. The appropriateness of the tool is determined by the desire to simulate realistic
relationships. The aircraft engines are also called the propulsion units because their main task is to propel
the aircraft forward. This propelling forward is due to the thrust developed by the engines according to the
principles as discussed in the first section of this chapter. Since the propulsive devices change the flow
field around an aircraft, their geometrical extensions should be scaled in a manner similar to the rest of the
aircraft parts. This means that the diameter of the propulsive device should retain the same relative size in
the experiment as in the free flight.
In order for the aircraft to fly horizontally at a constant velocity, the thrust has to equal the drag. Since the
drag force is scaled in such a way that the drag coefficient of the scaled model and that of the free flying
aircraft are equal the following must hold:

Cdf f = Cdwt (5.26)

Where f f stand for the free flight case and wt stands for the wind tunnel model. The thrust coefficient of
the engine simulator will also have to be equal to that of the real motor, thus:
T
Tc = 1 2
(5.27)
2 ρu Sp

It can be seen that the thrust coefficient is computed in the same manner as the lift and drag coefficient.
with the exception of the reference surface. When computing thrust for isolated propulsors the most used
reference area is the cross-sectional area of the propulsors as shown above. Though airframe designers
quite often choose to reference the thrust to the wing surface area as well, for convenience sake. It is
important to remember this definition when different sizes or types of a propulsor are compared with each
other. It is clear from the definition of the thrust coefficient, that the thrust requirement for the simulators
scales linearly with density and Sp and quadratically with velocity. From this follows that, when assuming
that the testing is performed at the same velocity and the same atmospheric pressure, then the power
requirement scales with the square of the size of the model, Dp .
 3  2
ρwt uwt Dwt
Pwt = Pf f (5.28)
ρf f uf f Df f
This expression will have a slightly modified form, when we discuss the open rotors in a more specific
manner. While the requirement for power decreases with the area of the propulsor, the power density of

G. Eitelberg 87 Delft University of Technology

a storage medium generally remains constant. This means that the available power decreases with the
volume of a simulator. This can be a real experimental challenge.
In the discussion of interference effects, the operating conditions of the engines, like idle and high power
conditions are usually given as well. The following three separate approaches to interference can be used
to introduce three categories of interference testing:
1. Installation effect, as the difference between the forces of the powered aircraft or model at a given
power setting and the aircraft without the power units.

Finst = Fpowered model − Fclean configuration (5.29)

Here the powered model data has to be ‘cleaned’ of the thrust values themselves in terms of the thrust
book-keeping, as will be discussed later.
2. Power effect, where the power setting dependency of interference is evaluated.

Fpower = Fpowered at thrust − Fpowered at ground idle (5.30)

In the power effect testing, the question of behaviour of the aircraft at different power setting is de-
scribed, without a direct look at the detail of the interactions. The installation effect is already account-
ed for.
3. Jet interference effect.

Fjet = Fpowered at thrust − Fpowered at flight idle (5.31)

This deals mainly with the effect of the slipstream on the control surfaces in flight conditions. it is not
dissimilar to the previous category, except that a more detailed analysis of the interaction is aimed at.
The categories are arbitrary, any other categorisation that is well structured, can be introduced.

5.4 Simulators
The simplest form of engine simulation in the wind tunnel is a through flow nacelle (TFN); an example has
already been shown in figure 5.3. This is an aerodynamic shape corresponding to the outer shell of the
engine cowling and providing acceleration inside the hollow inner contour. The use of a TFN corresponds
approximately to the simulation of flight with engines set on idle. In order to be able to simulate the effects
of different power settings, powered engine simulators are required. The most advanced of those are the
air turbine powered simulators (TPS), shown in figure 5.10. In this figure the shaded areas represent the air

Figure 5.10: Turbine powered simulator with the performance monitoring planes identified.

G. Eitelberg 88 Delft University of Technology

driven turbine and the engine fan simulator. The air supply line is visible as a dashed contour. The instru-
mentation required for monitoring the TPS performance is shown as well. For the operational monitoring
additional instrumentation is required. Since these TPSs operate at very high rotational speeds, up to 75
000 rpm, not only its aerodynamic but also its mechanical operation has to be continuously monitored. The
evaluation of the aerodynamic performance relies both on online real time monitoring and the use of the
calibration data of the TPS, to be more specific on its fan performance. There is one significant difference
between the real engine and the simulator aerodynamics, as illustrated in figure 5.11 below. In a real turbo-
fan engine, the propulsion is generated by an ‘air breathing engine’. In the simulation with the TPS the free
stream flow does not enter the turbine. This has to be accounted for in the design of the inner contour of
the TPS.

(a) Engine: all air enters through the fan-intake. (b) TPS: turbine air enters through pipe-line.

Figure 5.11: Differences in air mass flow through turbine and TPS (M.Laban et al.).

Other effects to be considered are the shaping of the spinner and the outer contour. The outer contour
performance is affected by the character of the boundary layer developing on the nacelle and the propensity
of the lower Reynolds number boundary layer in the wind tunnel to separate earlier at high angles of attack.
The same AIAA 2005-3703 paper [10] describes in some detail the methodologies of reshaping engine
nacelles for TPS testing in wind tunnels.
In the case of simulating propulsion by propeller engines, actual scaled propellers are used in wind tunnels.
These can be driven either by electric, hydraulic or, in a manner similar to the TPS, by air motors. The
drawback of the current electric motors is their lower power density compared to pneumatic and hydraulic
motors. This often leads to larger nacelle diameters for electrically powered propulsors, Furthermore, the
larger heat generation complexifies measurements with internal load cells, and the high current supplied
to the motor generates an unfavorable testing environment in terms of electromagnetic interference. On
the other hand, compared to the traditional pneumatic and hydraulic motors, the use of electric motors has
advantages in terms of flexibility and simplicity of installation and operation of the engines, and improved
quality of the control of the operational conditions of the engine (like rotational speed, phase angle).

5.5 Thrust book-keeping

Regarding the quantification of effects in wind tunnel testing, an elaborate procedure considering measured
and calculated values has to be followed. In the case of propeller-propulsion simulation, there is an existing
technology available for measuring the direct forces between the propeller and its mount (the nacelle) on the
aircraft. This technique, called the rotating shaft balance (RSB), benefits from the fact that the connection
between the rotor and the aircraft is colinear with the rotor axis. This possibility does not exist for the pylons
of turbofan engines due to the non-axisymmetric mounting. In addition, the complication of the channel flow
developing between the engine cowling, the pylon and the wing or the fuselage of the aircraft are not always
individually quantifiable. The situation would be extremely complicated especially in the sideslip conditions.
The standard practice is that the total forces and moments acting on an aircraft model are measured with the
help of an internal six-component strain-gauge balance. Neglecting the thrust component in the direction
of lift, or rather just including it in the lift interference term, the drag contribution of the engine has to be
evaluated. The schematic approach for how this is done is given in the figure 5.12.

G. Eitelberg 89 Delft University of Technology

 Calculation

Fan thrust: F19 Turbine thrust: F9

Gross thrust: FG

Ram drag: F0 Nett thrust: FN

Drag Balance force: Fx

Measured

Figure 5.12: Force balance for determining the interference drag. The calculation is based on ideal expan-
sion and calibrated correction values.

The only directly measured parameter is the total force (in the direction of the thrust, Fx ) measured by
the balance. For determining the interference, the drag of the aircraft with installed power units has to
be measured and compared with that of the clean configuration. For that the thrust produced by the air
turbine and the thrust produced by the fan have to be determined. The drag of the TPS inlet has to be
subtracted from the gross thrust to yield the net thrust. The difference between the net thrust and the
balance measurement is the drag including the interference effects.

D = FT hrust,net − FBalance (5.32)

The parameters ram drag (also known as inlet momentum), fan thrust and turbine thrust are determined
indirectly with the help of measured thermodynamic parameters pressure and temperature in addition to the
calibration results from the engine calibration facility. For the discussion of the thermodynamic parameters
and their use, the definition of different reference planes is given in the figures 5.13 and 5.14. The most
important reference planes are the upstream far field entrance plane 0 and the fan exit plane 19. The turbine
exit plane is denoted as the plane 9.
In the following, the procedure of the determination of single components of force bookkeeping as was
illustrated in figure 5.12 is described in some more detail with the help of figure 5.14. With the given free
stream conditions in the upstream reference plane the outflow conditions have to be determined from the
measured parameters. Instead of the more global evaluation of the thrust generation in the beginning of
this chapter, here also the compressibility effects will be accounted for. The basic accounting uses the
isentropic relationships.

T 1
= γ−1 (5.33)
T0


1 + 2 M2

G. Eitelberg 90 Delft University of Technology

 Figure 5.13: Definition of the planes of evaluation and measurement in a TPS calculation.

Figure 5.14: Schematic representation of the performance of a TPS.

p 1
= γ (5.34)
p0 1+ γ−1 2

γ−1
2 M

The deviations from the isentropic expansions will be accounted for in the calibration process.

5.6 Quantifying the TPS performance

If a more detailed description of the TPS performance is wanted, its performance quantified for its main
components such as the fan and the turbine needs to be determined separately. In order to do so, mea-
surements at several locations within the flow field of the TPS are necessary. These locations are given in
the previously shown figures 5.13 and 5.14.
Although there is no thrust possible without inflow into the engine, it is customary to separate the inflow
momentum flux force from the exhaust momentum flux force. The inflow component is then termed the
ram drag, and the outflow component is then termed as the fan thrust or the turbine thrust, or when added
together the gross thrust. This is not strictly correct, as the thrust is the sum of the inflow and the outflow
momentum flux as shown at the beginning of the chapter.
The gross thrust is generated by the fan exhaust and the turbine exhaust, thus in order to find this thrust the

G. Eitelberg 91 Delft University of Technology

mass flow and the flow velocity of both the fan and turbine are required. In case of a TPS there is no turbine
inlet, thus for the ram drag determination it suffices to find the mass flux and velocity through the fan.

p
u19 = M19 a19 = M19 γRT19 (5.35)

From the isentropic expansion relationship for pressure above, the Mach number for the fan exit plane can
be expressed as:
v "  γ−1 #
u
u 2 p0 γ
M19 = t −1 (5.36)
γ−1 p

The relation between temperature and pressure can be described by:

  γ−1
T p0 γ
= (5.37)
T0 p

With this the fan velocity can be expressed as a function of free stream paprameters and parameters
obtained in the slipstream measurements:
v
u
u 2γRT0
"   γ−1 #
p19 γ
u19 (p, T0 , p0 ) = t 19
1− (5.38)
γ−1 p019

It is assumed that the flow is perfectly expanded, such that p19 corresponds to the free stream static pressure
p∞ , and the total pressure and temperature are conserved from the measurement station 15 to the exit plane
19 of the control volume, p015 = p019 and T015 = T019 .
In the case that the jet in the fan stream expands into a flow field with a static pressure different from the
free stream static pressure, extra correction for the pressure difference is required. This could be the case
when the fan exhaust is under the wing and the pressure side has a non-zero pressure coefficient Cp . This
would result in a different mass flow through the channel.
In the above expressions no allowance is made for the variation of the ratio of specific heats in the flow field.
In the practice, for the benefit of accuracy, those values are also determined according to local conditions
and include humidity in the ambient air.
For the pressure this is only satisfied when local supersonic conditions are not reached anywhere in the
expansion, which would have to be decelerated by shocks. That would be the case if p∞15 ≥ 1.89p∞19 , but
is not generally encountered in our simulations of engines with fan pressure ratios (FPR) of 1.3 to 1.5.
Even with the lack of shock induced losses, the ideal isentropic expansion results have to be corrected for
‘imperfections’ by calibration, as will be described later.
The fan mass flow is a product of the velocity with the density and the cross sectional area of the stream
tube:

ṁ19 = u19 A19 ρ19 (5.39)

In this case A19 is determined by the geometry and ρ19 needs to be determined at station 15 by using
p
ρ = RT . Since the total temperature is conserved, the temperature at the location 19 is determined by the
isentropic expansion. The area A19 is taken to be equal to the area at the fan exit plane. Maintaining this
approach consistently in the calibration as well as in the test evaluation assures that any error introduced
by this interpretation is absorbed in the calibrated values.

G. Eitelberg 92 Delft University of Technology

For the other thrust generating flow, that of the turbine exhaust, the turbine jet velocity can be obtained in a
manner identical to the determination of the fan velocity:
v
u
u 2γRT0
"   γ−1 #
p9 γ
u9 (p, T0 , p0 ) = t 5
1− (5.40)
γ−1 p05

It is assumed that p8 = p9 = p∞ . For the turbine mass flow, there is an additional option available for
determining it from the supply line evaluation instead of measurements at the location 15. Usually sonic
venturis are used to determine the mass flow. These are chocked devices that determine the maximum
mass flow through a contraction.

5.7 Calibration
In order to account for any imperfections and deviations from the isentropic, inviscid expansion relation-
ships, a calibration procedure has to be applied. In order to obtain an independent measurement for those
parameters that have to be determined in the wind tunnel test in an indirect way, a calibration facility and a
corresponding procedure is required. A schematic representation of the DNW Engine Simulator Calibration
facility is given in figure 5.15. It is essentially a vacuum tank with controlled through-flow. The inlet of this
tank is formed by the TPS or components thereof; the outlet is formed by a selection of sonic venturi noz-
zles. These allow a precise determination of the mass flow, required for the thrust calculations. The precise
mass flux determination is also required in order to calibrate the calculations of the ideal flow through the
TPS to the real values.

Figure 5.15: Calibration facility featuring a bellmouth intake, a TPS mounting mechanism equipped with
load cells and a set of venturi nozzles for mass flux determination.

In order to obtain controlled mass flow through the fan, a bellmouth shape contraction is attached to the
inlet plane of the TPS in order to avoid possible separation effects caused by the suction around the intake
lip.
As mentioned, the evaluation of test results relies on calibration of the TPS-s, because the drag, as well
as the lift and other forces and moments, are measured by one central balance integrated into the airplane
fuselage. This is the case even in half-model testing with a balance external to the wind tunnel as it is still
integrated with the body of the aircraft. In the calibration procedure the net thrust has to be subtracted from
the measured drag and comparing that result with the drag of a clean configuration will yield the installation
drag, as discussed earlier.

G. Eitelberg 93 Delft University of Technology

Since the calibration tank is equipped with known and calibrated mass flux meters, the fan mass flow can
be independently determined in order to obtain a calibration coefficient to correct for the non-isentropy of
the flow. In order to obtain an accurate determination of the so-called ram drag, the real, rather than the
ideal mass flow has to be used in the thrust book-keeping procedure. For the correction of the idealised
values, the fan discharge coefficient can be given as

ṁ19 ṁBM
Cd = = (5.41)
ṁ19isentropic ṁ19insentropic

ṁtankventuri − ṁBM
CdBM = (5.42)
ṁBMinsentropic

The mass flow in the bellmouth is given by:

ṁBM = ṁ19 = CdBM ṁBMinsentropic (5.43)

Where

ṁBMinsentropic = uBM ABM ρBM (5.44)

In a similar approach, in order to correct for the fan exit velocity inaccuracy obtained from using the isentropic
approximation, needed to determine the gross thrust of the fan, a velocity coefficient has to be determined
from the calibration procedure. The required calibration for the fan velocity coefficient is defined in an
analogous way:
u19 ṁ19 u19
Cu19 = = (5.45)
u19insentropic ṁ19 u19insentropic

The numerator on the right, which is the term for the fan thrust, can be obtained from the balance measure-
ment in combination with the known turbine conditions:
Fbalance − Fturbine Fbalance − ṁ9 u9insentropic
Cu19 = = (5.46)
ṁ19 u19insentropic ṁ19 u19insentropic

The discharge coefficients typically have values between 0.9 and 0.95 accounting for the losses in the
boundary layers and elsewhere.

G. Eitelberg 94 Delft University of Technology

6 Noise
Due to the increasing population and global trends in urbanisation cities keep increasing in size. This is
an issue as airports are currently struggling with the increasing demand of flights whilst cities move closer
and demand quieter operation of airports. This is one of the main reasons why a lot of research is put in
reducing aircraft noise.
This chapter will treat the basics of noise measurements, but before that the basics of human hearing is
discussed.

6.1 Human hearing

As with all mammals humans are gifted by the sense of sound. Though the complexity of the human hearing
system, see figure 6.1 goes beyond this course. Here one needs to know what is sound and how it is scaled.
Sound is generally defined as a pressure variation, ∆p, that is experienced by the tympanic membrane,
which is transferred to several systems within the ear before a nerve sends a pulse to the brain.

Stapes
attached to
val window)
Semicircular
Canals
ncus
Vestibular
Nerve

Cochlear
Nerve
Cochlea
External
Auditory Canal Ty
Ca

Tympanic Eustachian Tube

Membrane
Round
Window

Figure 6.1: Schematic representation of the human hearing system.

When discussing sound the frequency domain is often used. Depending on the age and hearing quality of
a person the human hearing system can detect pressure fluctuations between 20 Hz-20 kHz, although the
upper limit decreases with age. The sensitivity of the frequency is not uniform as seen from figure 6.2. In
general it is most sensitive for frequencies around 3-4 kHz, which is also approximately the frequency of
the human voice.
As seen from figure 6.2 the sensitivity of the human hearing system is expressed in dB; the dB is the unit
(actually, it is not a unit at all as shown below, although it is usually used as such) generally used to express
the Sound Pressure Level (SPL), which is one of the methods to describe the ‘strength’ of the sound. The
other measure is the Sound Power Level (SWL). Both the Sound Pressure and Sound Power levels can
provide useful information about sound but what they represent is different. the Sound Power Level is used
to describe the strength of the sound source, whilst Sound Pressure Level describes the effect it has on its
surroundings, as illustrated in figure 6.3. Equation 6.1 shows how the SWL is computed. It can be seen that
the sound power level, LW , is logarithmically dependent on a power ratio of which P0 = 1pW = 1 · 10−12 W
is the reference power and P is the source power.
 
P
LW = 10 · log10 [dB] (6.1)
P0

G. Eitelberg 95 Delft University of Technology

 Figure 6.2: Perceived human hearing sensitivity.

The sound pressure level, Lp , is computed similarly to the sound power level, however in this case a
pressure ratio, rather than a power ratio is used.
 
prms
Lp = 20 · log10 [dB] (6.2)
p0

In the above equation p is the root mean square sound pressure and p0 = 20µP a is the reference pressure.
As both SWL and SPL are logarithmic functions it can be deduced that the sum of two equal noise sources
results in an increase of noise strength by 3dB independently of their individual strength; so 80dB + 80dB =
83dB as is 40dB + 40dB = 43dB .

Figure 6.3: Difference SWL and SPL.

To give a better understanding as to how sound scales typical sound pressure and power levels that relate
to well-known sound sources are shown in figure 6.4.

6.2 Acoustics as a wave phenomenon

In the previous section the human hearing system is discussed as well as the definition of SPL and SWL.
In this section these will be discussed in more detail.
The propagation of sound is often described by waves, which in one dimensional form can be described by
the following equation.

∂2p 1 ∂2p
− =0 (6.3)
∂x2 c2 ∂t2

G. Eitelberg 96 Delft University of Technology

 200 dB 200 dB Saturn V rocket (1 · 109 W )

180 dB 180 dB
Jet airliner (500, 000W )

160 dB Immediate hearing damage 160 dB

Propeller airliner (500W )

140 dB Pain threshold 140 dB

Small private airplane (5W )

120 dB Jet airplane take-off at 500 m 120 dB

dB
dB
(relative to 20µP a) 100 dB (relative to 10pW )
100 dB

80 dB Truck at 20 m 80 dB
Small office machine (20 · 10−6 W )

60 dB 60 dB

Conversation at 1 m

40 dB 40 dB

Rustling of leaves at 20 m Whisper (10−9 W )

20 dB 20 dB

0 dB Hearing threshold 0 dB

(a) Typical sound pressure levels, Lp . (b) Typical sound power levels, LW .

Figure 6.4: Typical sound levels.

The above form is valid when there is no ambient flow as in that case the pressure, p, is only a function
of x and t. c in this case is the speed of sound, which can be expressed as c2 = γRT . A solution to the
differential equation shown above is of the following form:
p = f1 (ct − x) + f2 (ct + x) (6.4)
This solution describes characteristics moving left and right with the speed of sound. Note that c describes
the phase velocity and not a transport velocity. In the mean, no material is being transported. When
vibrating sources are used rather than impulses the above solution takes another form, which is shown
below.
p = A1 sin(ωt − kx + φ1 ) + A2 sin(ωt + kx + φ2 ) (6.5)
ω 2π
Where ω is the frequency of the harmonic oscillation, and k = c = λ is the wave number and λ is the
wavelength.
The sound pressure is a simple harmonic in time oscillating around a steady value. It is the linear superpo-
sition of two waves travelling in opposite directions at a constant speed. In order to find this sound pressure
the characteristic impedance of a medium is required, which is computed as: R = ρc = up . This means that
kg kg
for air, with c = 340 m
s and ρ = 1.2 m3 , the characteristic impedance is R = 415 m2 s

The intensity is defined as time averaged power transported across a surface. Power is the product of force
and velocity, P = F · v = p · A · v. Taking the time averaged value of p results in prms and normalising it with

G. Eitelberg 97 Delft University of Technology

 P
W 
the characteristic impedance of a material, sound intensity, I = A m2 , then becomes:

p2rms
I= (6.6)
R
In the case that the signal is a sine wave the above equation can be written in terms of the amplitude of the
pressure signal as:
p2
I= (6.7)
2R
Sound power is the surface integral over a closed surface including the source within it whilst the radiat-
ed power is a function of its boundary conditions. A solid wall doubles the power of a dipole, a corner
quadruples it.
The sound power and sound pressure levels are expressed as shown in equations 6.1 and 6.2 respectively.
W
When it comes to sound power the power can also be interchanged with sound intensity in m2.

6.3 Scaling of sound

When compact solid bodies are submerged in a flow the sound generated is due to the pressure fluctuations
of the solid boundary. The compactness of the body should comply with ωL c << 1, since Strouhal scaling
ignores non-compactness effects, where Str = ωLU = 0.2 and is valid when M = Uc << 1.
Furthermore, the force F on a solid body scales with ρU 2 L2 , which results in an acoustic dipole of strength
< F 2 > radiated by the body. In this case the acoustic power would scale with the 6th power of the flow
velocity, which results in the following:
hp2 i
∝ M6 (6.8)
(ρ0 c2 )2

In the case for compact subsonic jets the characteristic wavelength is of the order of λ = cDU , where D is
the jet diameter and U the flow velocity. The acoustic power scales with the 8th power of the flow velocity:
hp2 i
∝ M8 (6.9)
(ρ0 c2 )2
Since free turbulence is a less efficient sound generator than the wall bounded turbulence, the amplitude
decay with the distance squared, rD2 , needs to be considered for measurement purposes. This is due to the
fact that the power needs to be distributed over an increasing area, which scales with r2 .
Another case is sound scattering from an edge where the solid body ends. In this case the acoustic power
depends on the direction as seen in the equation below.
hp2 i
 
5 2 Θ
∝ M · sin (6.10)
(ρ0 c2 )2 2
When looking at the Mach dependence of the acoustic power one can see that the above case is the most
efficient sound generator in the flow field as it scales with M 5 rather than the 6th or 8th power.
When measuring sound in aerodynamic situations, it has to be remembered, that sound waves can be
convected with the flow. In the measurements this manifests itself as a Doppler effect. The Doppler effect
describes the perceived change in frequency or wavelength. When a wave from a single source carried
with velocity w with its emitted frequency ω, or corresponding period T = 2π
ω , then the front end of the wave
travels a distance cT during that period, whereas the tail end has already travelled the distance wT in that
time. The perceived wavelength is then shortened by wT . The Doppler-shifted frequency is then:
ω
ωshifted = (6.11)
1 − wc

G. Eitelberg 98 Delft University of Technology

Note that there are other sources of Doppler shifts associated with the macroscopic changes of sources,
but these will be left for more specialist studies.
It is also important to remember that a sonic boom is not really an acoustic phenomenon as it is a non-
isentropic N-wave in the far field of a supersonic vehicle.

6.4 Measurement devices

When measuring noise phenomena good quality microphones are required to gain reliable data. In general
there are two types of microphones namely capacitive and resistive. Capacitive microphones utilise the
capacity variation caused by the deformation of the pressure sensitive membrane isolated from a fixed
metal electrode, whilst for resistive microphones the strain on the connection between the membrane and
a fixed structure is used as indication of the sound pressure levels.
When performing noise measurements many microphones distributed in rakes or microphone arrays are
used. However, what are the required specifications of the microphones for wind tunnel testing? In table 6.1
these requirements are shown. Only high quality microphones are capable of handeling the requirements

Table 6.1: Microphone requirements for wind tunnel testing.

Maximum sound pressure level SP Lmax ≥ 130dB due to boundary layer noise
Low self-noise level Lnoise < 20dB for high frequency signals
High sensitivity S > 10 mV
Pa sufficient signal to noise ratio
Broad frequency range frange ≤ 100Hz and ≥ 40kHz for scale models

shown in table 6.1. In most cases B&K 1/2”, 1/4” or 1/8” are used, which are very expensive. When including
the cable and the pre-amplifier the cost of the set is about e2000. Due to the cost of these microphones
these are only used for very accurate measurements where fewer microphones in an array are used, for
large arrays cheaper microphones, such as the WM-61 B102A electret microphone seen in table 6.2, are
usually opted for. These microphones are commonly used in arrays with ∼ 500 installed microphones.

Table 6.2: WM-61 B102A electret microphone specs.

SP Lmax ≥ 120dB
Lnoise 20dB
45 mV dBV
S 94 dB or −27 P a
frange 20Hz ≤ frange ≤ 40kHz
e ∼3

When acquiring noise data the acquisition parameters shown in table 6.3 are typical for noise measure-
ments. This results to ∼ 1Gb raw time data for 1 data point (dpn).

Table 6.3: Typical acquisition parameters.

Microphone channels 100 − 144

Sample frequency 120kHz
Acquisition time 30 seconds

Sound levels are quantified with the help of frequency spectra, power density as a function of the frequency,
similar to the transfer function in the control theory. The difference being that SPL’s are integrated over
octaves in order to get a result. Octave filters are bandpass filters, with a bandwidth of 71% of the center
frequency.

G. Eitelberg 99 Delft University of Technology

6.4.1 Presentation
As discussed above, the noise is customarily presented as SPL vs. frequency. The SPL is a numerical
value without a dimension. The frequency has a dimension of 1/sec or Hz. Even though the SPL has no
dimension, it is not a dimensionless similarity parameter in the sense of the Buckinghams’s Pi-theorem.
The prms is made dimensionless in order to be able to take a logarithm of that value. The factor 20microP a
is not a governing parameter describing the sound generation or propagation. It only describes a property
of the instrument “human ear”, which on the average does not register signals below that value.

6.5 Beam forming

6.5.1 Delay & sum
When it comes to noise measurements beam forming is referred to as a signal processing technique that
is used in sensor arrays for directional signal reception. This is achieved by having a microphone array and
a post processing algorithm that is able to dissect the different noise sources and find the origin of these
sources.
One of the first algorithms that was able to perform this task was the delay and sum algorithm. Though
relatively simple compared to the newer algorithms this algorithm is less precise and takes more processing
time, which is why it is not used as often anymore. In order to perform the delay and sum algorithm a scan
plane with monopole sources at locations ξ~ as well as a microphone array with microphone locations ~x are
required, see figure 6.5. As the distance between the source and microphones 1 and 2 are different the

!x ξ!
2

Microphone array Scan plane

Figure 6.5: Microphone array and scan plane with monopole source.

noise experienced are in different phases. In the delay and sum algorithm these signals are corrected such
that identical phases are obtained and these corrected signals are summed. This procedure of delaying
and summing the signals is done for every scan point and the source location is found by having the highest
sum as this must be the shortest distance to the noise source. A graphical representation how this is done
is shown in figure 6.6.
In order to compute the array output as a function of the source position several steps need to be taken.
First of all Green’s function, which is an expression describing the response to an acoustic monopole, needs
to be determined.
2πif
− c ||~ ~
x−ξ||
~ =e
G(~x, ξ) (6.12)
~
4π||~x − ξ||

G. Eitelberg 100 Delft University of Technology

 Noise source Measured Delay & Sum Source plot
time signals Processing

" !

! "

Difference in
amplitude
and phase
" !

! "

Microphones

Figure 6.6: Microphone array and scan plane with monopole source.

When Green’s function is determined the array output can be found as shown below.
N
~ = 1 X pn
a(ξ)
~
N n=1 G(~x2 , ξ)
N (6.13)
1 X h 2πif ||~x−ξ||
~
i
~
= pn e c 4π||~x − ξ||
N n=1 | {z | {z }
} Amplitude correction
Phase correction

In the above equation pn is the pressure of the nth microphone in the array

6.5.2 Conventional beam forming

Another way of looking at beam forming is computing the array output in the frequency domain. As it is
easier to think of noise in the frequency domain and it provides with a better
 endresult of the array output.
p1 (f )
In order to find the array output in this way the pressure vector p =  ...  and the steering vector
 

pN (f )
~
 
G(~x1 , ξ1 )
..
g=  are required and the minimisation problem is posed: J = ||p − ag||2 The solution of this
 
.
G(~xN , ξ~N )
minimisation problem is shown below.

~ = g? p
a(ξ)
||g||2
PN
gn? pn
= Pn=1
N (6.14)
n=1 |gn |2
N 2πif ~
1 X e− c ||~x−ξ||
= p n
||g||2 n=1 ~
4π||~x − ξ||
Comparing equations 6.13 and 6.14 one can see that for the array output for the delay and sum algorithm
~ term, whilst for the conventional beam forming case it is inversely
is linearly dependent on the 4π||~x − ξ||

G. Eitelberg 101 Delft University of Technology

 ~ This means that the conventional beam forming has a better signal to noise ratio
dependent on 4π||~x − ξ||.
as at large distances the function receives less weight. With Delay and sum on the other hand there is more
bias noise as there is a high weight for large distances.

6.6 Array performance

When designing microphone arrays for the detection of noise sources for various objects a good resolu-
tion and gain are required such that it can detect the noise sources with enough accuracy and precision.
But what are the array gain and resolution exactly? As shown in the previous section each microphone
obtains noise data, which is later processed to obtain an acoustic image. However since noise travels in
all directions the microphones that are not directly aligned with the source will still experience noise, which
will show up in the acoustic image. The difference of the SPL of the main peak and the second highest
peak is referred to as the array gain. This is visualised in figure 6.7. Resolution on the other hand refers
to the amount of detail that can be recovered from the acoustic image. High resolution means that very
small noise sources can easily be identified, whilst low resolution will show only the general noise trends.
The resolution of the array is the characteristic length that corresponds to a difference of 3 dB, as shown in
figure 6.7.

(a) Graphical interpretation of array gain. (b) Graphical interpretation of

array resolution.

Figure 6.7: Definitions of array gain and resolution.

As a rule of thumb:
425 ds
resolution = · (6.15)
f DA

Where f is the source frequency, ds is the distance from the source to the array and DA is the array
diameter.
When first designing microphone arrays for the mapping of sound the microphones were placed randomly
as one thought this would prevent aliasing of the signal and result in better array gains. This theory was put
to the test at DNW where two different arrays were tested, one being completely random whilst the other
array still had some order in the placement of microphones, see figure 6.8. Both arrays consisted of 50
microphones and were fitted to an array of 2x2 m. In order to test the performance of these arrays a noise
source with a frequency of 2000 Hz was located 6 m from the centre of the array (0, 0, 6). In the perfect
case only one peak at should be visualised in the acoustic image located at the centre of the array, however
when looking at figure 6.9 one can see this is not the case. Especially the acoustic image of array 1 does
not perform that well. One can see that apart from the main lobe, located at (0, 0) there are several side
lobes. These side lobes are not physical and are due to the aliasing of the noise. From this figure it can
be seen that random arrays do not necessarily outperform arrays with some order as the acoustic image of
array 2 performs a lot better as the array gain has be increased from 8.5 dB to 12.5 dB.

G. Eitelberg 102 Delft University of Technology

 (a) Random microphone array 1. (b) Random microphone array 2.

Figure 6.8: Two different microphone arrays with varying randomness of microphone placement.

(a) Acoustic image of random microphone array 1. (b) Acoustic image of random microphone array 2.

Figure 6.9: Acoustic images of two different microphone arrays.

Nowadays the arrays used at DNW are mostly ordered and the microphones are distributed as seen in
figure 6.10. Note that when using an ordered microphone distribution that different length scales between
microphones should be taken into account as without this aliasing can occur. When looking at the radial
distribution used by DNW one can see that the microphones are placed in circles with a higher microphone
density at the center. This allows for a high frequency range and a dynamic range of 12 − 13 dB

6.7 Wind tunnel noise measurements

When doing noise measurements there are two setups that are often used, both with their own advantages
and disadvantages. For open jet noise measurements the shear layer is often the problem and for closed
test sections the boundary layer noise can be a real nuisance. Below the advantages and disadvantages

G. Eitelberg 103 Delft University of Technology

 Figure 6.10: Microphone array used by DNW.

of the different setups are discussed as well as how to tackle some of the problems when testing in these
conditions.

Open jet noise measurements Open jet wind tunnels are favourable for noise testing wind tunnel models
as the testing hall around the Open Jet provides a semi anechoic environment that results in better quality
of the measurements. Having enough room in the hall also allows for far-field microphone measurements
to be used in parallel with the microphone array, however there are some drawbacks too. One of the most
problematic drawbacks of using open jets for noise measurements is the fact that the shear layer between
the hall and the jet causes serious disturbances. Not only does it refract noise as a lens does, see figure
6.11, it also causes a strong loss of coherence for frequencies > 20 kHz. This coherence loss is dependent
on several factors such as the frequency of the noise, the flow speed and the distance from the nozzle.
To correct for the refraction of the noise as it passes through the shear layer an Amiet correction needs to
be performed.
The shear layer is not the only factor that makes noise testing hard. In most cases for open jet facilities
the arrays are placed further away from the model with respect to closed test sections, which limits the
resolution for frequencies > 15 kHz.
The last problem with testing in open jet sections is the lesser aerodynamic quality when comparing tests
in closed sections. Thus it is harder to obtain the correct flight condition when testing aircraft link the noise
measurements to the final flight condition.

Closed test section noise measurements One of the advantages of doing noise testing in closed sec-
tions are that significant data up to ≈ 50 kHz can be acquired without too much loss of coherence for these
high frequencies. This is mainly due to the fact that the microphone arrays can be placed closer to the
model and that refraction of noise through the shear waves becomes less of an issue. Also, by placing
the array closer to the model higher resolutions can be achieved and noise sources can be determined
more accurately. Furthermore, there is no loss of aerodynamic simulation quality, which means that the
noise measurements and balance measurements can be performed simultaneously, which is cheaper than
performing two tests.

G. Eitelberg 104 Delft University of Technology

 sound
source
Identical
signals

Wind
tunnel
Shear layer

Distorted
signals

Array

Figure 6.11: Open jet shear layer noise distortion.

However, there are some drawbacks to this type of noise testing too. Since testing in a closed section
typically means testing in a hard walled environment issues like standing waves, acoustic reflections are a
real issue and have to be accounted for. Not only that, but the noise generated by the turbulent boundary
layer can be a real nuisance. This is due to the fact that microphones measure pressure differences and the
highly turbulent flow continually causes pressure differences that are recorded by the microphones as seen
in figure 6.12. Since the arrays have to be placed in the closed section they are often put in the walls of the
test section. Having these arrays inside the test section means that the echoing of the noise has to be dealt
with. And even though a closed jet wind tunnel does not have a shear layer such as the open jet, it does
cope with other problems causing loss of coherence. The main issue with this kind of noise testing is the
noise generated with the turbulent boundary layer. This boundary layer noise can be seen as generating

Flow

Turbulent boundary layer

Array

Figure 6.12: Boundary layer noise in close test section.

random white-noise that distorts the real signal. When looking at only one microphone one could say that
the measured noise of the nth microphone is the sum of the signal noise and the boundary layer noise as
shown below.

pn = qn + εn (6.16)
|{z} |{z} |{z}
measured noise signal noise boundary layer noise

G. Eitelberg 105 Delft University of Technology

When comparing the boundary layer noise with the signal noise one would expect that the variance of the
boundary layer noise is much higher as this is a random signal, whilst the signal noise is not expected to
change over time. Therefore the following is valid:

|εn |2 >> |qn |2 (6.17)

One of the solutions to tackle this problem is to exclude the diagonal of the main cross-power matrix as this
provides clean source plots with improved resolution. This effect can be seen in figure 6.13. One drawback
of removing the diagonal of the main cross-power matrix is that it may lead to non-physical negative ‘source
powers’ in the acoustic source plots.

(a) Fokker-100 half model in wind tunnel.

(b) Noise map including diagonal terms. (c) Noise map without diagonal terms.

Figure 6.13: Acoustic maps of the F-100 model with and without boundary layer noise.

G. Eitelberg 106 Delft University of Technology

6.7.1 Moving sources
Though the noise characteristics of static models can easily be determined from ordinary wind tunnel tests,
moving models come with an additional dimension called time that makes determining the noise character-
istics more complicated. When reception and emission time is not taken into account when conventional
beam forming is used the noise is spread of an entire area and not just on the object itself, figure 6.14a.
This is why reception and emission time is of utmost importance when performing noise measurements
on moving objects. To find the noise behaviour of the moving object the following relation needs to be
introduced.

χn (tn ) = σ(τe )F (τe ) (6.18)

Where χ(tn ) is the microphone signal received at time tn , σ(τe ) is the source signal emitted at time τe and
F (τe ) is the transfer function for a moving monopole that relates the two signals. Using one microphone the
source signal can be determined if the transfer function is known as shown below:
χn (tn )
σ(τe ) = ≡ σ(τe ) (6.19)
F (τe )
When an array with a multitude of microphones is used the delay and sum algorithm can be used to solve
for the source signals as:
N N
1 X 1 X χn (tn )
σ(τe ) = σn (τe ) = ≡ σ(τe ) (6.20)
N n=1 N n=1 F (τe )

In figure 6.14 it can be seen that using this moving source processing technique delivers much better results
than ignoring this phenomenon.

(a) Conventional beam forming on moving source. (b) Moving source processing.

Figure 6.14: Different processing techniques on moving sound sources.

Another good example on moving source processing is measuring the noise of a helicopter. Tests by DNW
have shown that both the frequency of the rotors as well as the flight condition affect the measured noise.
When looking at figure 6.15 one can see that increasing the rotor frequency increases the rotor noise as
higher frequencies lead to higher tip velocities of the main rotor that at some point will become supersonic
causing loud shocks to occur. This is especially visible for the 4kHz case where the shock location can be
deduced by the relatively high noise region. As the frequency increases the relative noise of the tail rotor
goes down as the main rotor becomes the more dominant noise source. Not only does the frequency of the
rotor have an effect on the noise characteristics, the flight condition proves to be an important factor as well.
It can be seen that when descending the main rotor contributes the most noise. This is due to the fact that

G. Eitelberg 107 Delft University of Technology

 (a) Effect of rotor frequency on helicopter noise.

(b) Effect of flight condition on helicopter noise.

Figure 6.15: Most dominant sound sources for different frequencies and flight scenarios.

the vortices created by the rotor travel downwards and when descending the main rotor interferes with the
previously generated vortices. This is also why relatively no noise is generated by the rotor during climb.
In figures 6.14 and 6.15 the noise measurements for the moving turbine and helicopter are still performed
inside a wind tunnel, however many moving source measurements are performed outside. One of these
tests are fly-over tests in which an aircraft flies over a microphone array in order to find out what aircraft
components cause the most amount of noise. Some typical results are shown in figure 6.16.

Flap side edge

Engine exhaust Landing gear

Nose gear
Engine inlet Slats

(a) Typical noise sources on MD82. (b) Fokker 100.

G. Eitelberg 108 Delft University of Technology

 (c) Airbus A340. (d) Boeing 777.

Figure 6.16: typical acoustic maps of various aircraft.

G. Eitelberg 109 Delft University of Technology

7 Design of experiments
When it comes to aircraft performance wind tunnel measurements and various other experiments are an
unwanted necessity as experiments are both time consuming and very costly. That is why lots of time is put
into the design of such experiments, such that fewer tests are required to practically get the same results.
This is where the use of statisticians comes in as these can design the optimal way of experimenting. After
all, according to Ronald A. Fisher:
“To call in the statistician after the experiment is done may be no more than asking him to perform a post-
mortem examination: he may be able to say what the experiment died of.”
To showcase the importance of a good design of experiment is the following: Say that there are two models
that need to be accurately weighed, one being an aircraft, A, and the other a helicopter, H. Since a precision
σ
of √σ2 is wanted the models need to be weighed twice as σy = √yn , where n are the number of experiments.
As the weight is determined by a very accurate scale each measurement takes approximately 1 hour and
costs e1000, thus limiting the amount of experiments saves both time and money.
conventional approach: In this approach the conventional One Factor At a Time, OFAT, is opted for. This
means that each parameter is changed on its own, which is why in this case A should be weighed twice
as well as H in order to obtain the required precision. This results in 4 experiments as shown in figure 7.1.
Knowing that each experiment takes 1 hour to measure and e1000, the costs of these would be e4000 and
4 hours of one’s time.

(a) Measure aircraft weight 2 times. (b) Measure helicopter weight 2 times.

Figure 7.1: OFAT

Designed experiment: When one carefully designs an experiment different less measurements are re-
quired with often the same amount of precision. In this example the solution would be to measure the
added weight of the models and the difference, see figure 7.2, as this way the weight of the models is still
measured twice, but omits the need for two more tests saving money and time. Due to the fact that each
model is still weighed twice the precision of √σ2 remains unchanged, with the proof being: Assume S is the
sum of the aircraft and helicopter models and D is the difference

S +D S −D

S =A+H
→A= & H= (7.1)
D =A−H 2 2

G. Eitelberg 110 Delft University of Technology

 (a) Measure sum, S, of A and H. (b) Measure weight difference, D, of A and H.

Figure 7.2: Modern Design of Experiment (MDOE).

It can be seen from above that A averages S + D and H averages S − D so in both cases n = 2.
From this example it can clearly be seen that designing experiments can reduce the costs and testing time
involved in testing.

7.1 Factorial designs

Factorial designs are an often used design method when it comes to experiments with lots of variables.
Full factorial designs feature “all combinations of every level of every independent variable or factor” and
this method is often used as it provides good means to examine both main effects, Response change due
to change in the level of an individual variable, as well as interaction effects, Change in a main effect due
to a change in some other variable, but also happens to be more efficient than OFAT. Factorial designs
are especially of use for screening experiments, where the most important variables are required amongst
a large amount of variables. In this way it can be seen as a building block to see where more complex
experiments are needed.
A trivial example of the difference between factorial designs and one factor at a time is given below: Assume
that there are two factors,A and B, that need to be examined. Both can experience a low and a high
value. Since the wanted precision is √σ2 at least two measurements of the low and high values need to be
measured.
Two-level factorial designs: When opting for the factorial designs the average of the main effects of factors
A and B are computed as below:

EA1 = 50 − 30 = 20
EA2 = 62 − 40 = 22 (7.2)
50 + 62 30 + 40
EA = − = 21
2 2

EB1 = 40 − 30 = 10
EB2 = 62 − 50 = 12 (7.3)
40 + 62 30 + 50
EB = − = 11
2 2

G. Eitelberg 111 Delft University of Technology

 40 62 38,42
High + High +
Factor B

Factor B
Low - Low -
30 50 30,32 50,54

- + - +
Low High Low High
Factor A Factor A
(a) Two-level factorial. (b) One factor at a time.

Figure 7.3: Difference between factorial designs and OFAT.

It can be seen that with only four data points the required precision is found. However, it can be seen from
the figure that one measurement takes place when the other factor has been varied as well and it is likely
that this has an effect on the main effect. In other words, there is an interference effect of factor B on factor
A. In this example the effect can be computed as follows:
EA2 − EA1 22 − 20
EAB = = =1 (7.4)
2 2

OFAT: When the same precision is wanted with OFAT all points need to be measured twice, which leads to
six data points.

EA1 = 50 − 30 = 20
EA2 = 54 − 32 = 22 (7.5)
54 − 32 50 − 30
EA = − = 21
2 2

EB1 = 38 − 30 = 8
EB2 = 42 − 32 = 10 (7.6)
38 − 30 30 + 50
EB = − =9
2 2
Since there is no interaction between factors A and B there are no interaction effects to be deduced when
using OFAT.
From the above example it can be seen that the main and interaction effects take the form of:
±y1 ± y2 ± · · · ± yn
ζ=
n/2
      (7.7)
2 2 2
=± y1 ± y2 ± · · · ± yn
n n n

G. Eitelberg 112 Delft University of Technology

Recall that the variance of such a function takes the form of:
n  2
2
X ∂f
σf = σi2 (7.8)
i=1
∂xi

The partial derivative term from the above equation can also be found for the main and interaction effects
as shown below.
 2 
∂ζ n 2 4
= ± = 2 (7.9)
∂yi 2 n

Since all effects y1 . . . yn takes the same form it can be seen that all standard deviations are the same so
σyi = σy . substituting this into the variance formula the following is found.
     
4 4 4
σζ2 = σ 2
y + σ 2
y + · · · + σy2
n2 n2 n2
 
4
=n σy2 (7.10)
n2
4
= σy2
n
Thus the variance formula can be finally written as:
2
σζ = √ σy (7.11)
n

Note that the same variance formula applies to all factorial effects, including main effects and interactions.

7.1.1 SWIM model experiment

To see the effects of various types of factorial designs an experiment is proposed in which a generic wind
tunnel model, see figure 7.3, changes the following 6 variables:
• A: Angle of attack (0◦ /10◦ )
• B: Airbrake width (small/large)
• C: Airbrake height (low/high)
• D: Airbrake location (close/far)
• E: Flapsize (small/large)
• F: Winglets (on/off)
Since there are two options for every variable there are 26 = 64 combinations possible. The matrix showing
the options can be seen in table 7.1. In this matrix the - denotes the small/low/close options whilst the
+ denotes the large/high/far options of the discussed variables. Since 64 measurements are taken it is
possible to discern 2k − 1 = 63 phenomena, either being main or interaction effects, but is it possible to
quantify these 63 effects with only half the effort?
When only 32 points are measured the matrix in table 7.2 can be found, but as one can see several of these
columns seem to be duplicated, which is denoted with similar colours. This duplicated columns are due to
aliases, but what are these aliases exactly?
To be able to reduce the costs of experiments the amount of measurements need to be reduced. In the case
of a half factorial experiment the amount of measurements is halved, but this does have negative effects.
In order to be able to reduce the amount of measurements a column with only + signs, such as the column
containing ABCDEF, is defined as the unit vector I. When looking at the derivation below one can see that
by doing this one column is duplicated, or aliased.

G. Eitelberg 113 Delft University of Technology

 Figure 7.4: SWIM model used in experiment.

Table 7.1: Full factorial design

Run Variable
Number A B C D E F ABCDEF Block
1 - - - - - - + 1
2 - - - - - + - 2
3 - - - - + - - 2
4 - - - - + + + 1
5 - - - + - - - 2
6 - - - + - + + 1
7 - - + + - + 1
8 - - - + + + - 2
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
63 + + + + + - - 2
64 + + + + + + + 1

Say F is the following vector.

F = + − + + −···+ (7.12)
(7.13)

When multiplying with itself a unit vector arises as shown.

F · F = F2 = I = + + + + +···+ (7.14)

It can be seen in table 7.2 the ABCDEF column is such a unit vector, thus:

I = ABCDEF (7.15)

G. Eitelberg 114 Delft University of Technology

 Table 7.2: Half factorial design

Run Main effects Two-factor interactions Five-factor interactions Six-factor

Number A B C D E F . . . BF CF DF EF ... ABCDE ABCDF ABCEF ABCDEF
1 - - - - - - ... + + + + ... - - - +
4 - - - - + + ... - - - + ... + + - +
6 - - - + - + ... - - + - ... + - + +
7 - - - + + - ... + + - - ... - + + +
10 - - + - - + ... - - - - ... + - - +
11 - - + - + - ... + - + - ... - + - +
13 - + + - - ... + - - + ... - - + +
16 - - + + + + ... - + + + ... + + + +
.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
. . . . . . . . . . . . . . . . .
32 + + + + + + ... + + + + ... + + + +

F =F ·I
= F · ABCDEF
= ABCDEF 2 (7.16)
= ABCDEI
= ABCDE

The duplicate columns can be explained since F = ABCDE as proven above. This shows that the amount
of effects are reduced from 2k − 1 = 63 to 2k−1 − 1 = 31. Other aliases are found in a similar fashion such
as AB · I = AB · ABCDEF → AB = A2 B 2 CDEF = I 2 CDEF = CDEF . In general, n-way interactions
are aliased with m-way interactions, where n + m = nvariables . So in the case of a 6 variable experiment the
main effects are aliased with the 5-way interactions, and the 2-way interactions with the 4-way interactions,
or, in extreme cases 3-way interactions will be aliased with 3-way interactions.
Thus it is not possible to find all 63 effect by only measuring 32 times. Though not all effects can be
determined the most dominant ones are, so a half factorial design might be of use to discern the 31 most
dominant effects. These dominant terms are shown in the Pareto chart in figure 7.5. From both the pareto
chart and the nominal plots from figure 7.6 one can see that the most dominant effects remain visible,
however the less dominant effects are filtered out when reducing the amount of measurements. When
using only 16 measurements, or a quarter factorial experiment, one can still see the most dominant effects,
such as the angle of attack, A, or the flap size, E.

G. Eitelberg 115 Delft University of Technology

 Figure 7.5: Pareto chart with the most dominant effects.

G. Eitelberg 116 Delft University of Technology

 (a) Full factorial design. (b) Half factorial design.

(c) Half factorial design.

Figure 7.6: Normal plots for full factorial, half factorial and quarter factorial designs.

G. Eitelberg 117 Delft University of Technology

8 Balances
Although the balances are the main measurement device in aeronautical development testing, and have
been so for decades, they still pose challenges in the engineering practice. That is why we still need to
discuss the basic principles and the limitations and uncertainties that need to be dealt with.
There are a number of different balance applications. We classify them as main balances and as component
balances. In wind tunnel testing, the main balances can be either internal or external to the model.
The component balances are in practice either for steady loads, like flap or slat balances, or for a dynamic
application, like a rotating shaft balance used to isolate the propeller (or rotor) forces from the airframe
forces.
All of the applications make use of the linear elasticity of the balance material and the strain gauge applica-
tion. The linear elasticity means that there is a linear relationship between the stress σ and a strain ε of the
material, described by a single constant E called modulus of elasticity: σ = E · ε.
The strain gauge is an electric device, the output voltage of which is a function of the deformation of the
material of the sensitive element. This is achieved by connecting the strain gauge in a Wheatstone-bridge
arrangement [25].
In the first approximation, when all forces and corresponding strain gauge signals are linearly independent,
then a six-component balance needs six independent strain gauge signals to determine the six forces and
moments in Cartesian coordinates:
   
Fx U1
 Fy  
kF 1 · · · kFx 6 U2 
 
 Fz   .x
 
.. U3 
..   
 = . . (8.1)
Mx  . .  U4 
k · · · k
   
 My  Mz 1 Mz 6 U5 
Mz U6

In this representation the forces are represented by F and the moments by M . The measured strain gauge
voltage is denoted as U . This matrix representation corresponds to the schematic representation as shown
in figure 8.1.
The coefficients ki in the balance matrix are the balance coefficients and are obtained from a calibration
procedure. In this procedure, the loads are applied a single one at a time, and it is assumed that the
combined loads do not affect the coefficients in the application. When a more refined approach is opted for,
the combination loads can not be neglected. In this case the balance equation is including the quadratic
(interaction) terms as well:
 
R1
 .. 
 
Fx
 . 
 
 Fy  
  a11 a12 · · · a16 a111 · · · a166  R6 
 Fz   .  2 

 = .
 Mx  .  R1 
 (8.2)
R1 R2 
a61 a62 · · · a66 a611 · · · a666 
 
 My  
 . 
Mz  .. 
R62

In this quadratic equation the balance coefficients are denoted as aij and the balance signals as R. (This is
probably a historical notation because the Wheatstone-bridge output voltage U is a function of the change
of the resistance R of the sensitive element).

G. Eitelberg 118 Delft University of Technology

 Figure 8.1: Schematic representation of the balance forces and moments.

It is obvious, that the 2nd order calibration to obtain the balance coefficients is more time consuming than
the first order one. There are a number of recent developments represented in the literature about different
strategies of balance calibration in order to have an efficient procedure and still maintain a low uncertainty
level. The balance uncertainty is usually given by the expression:
 
 
6
K  X Fj 
UCALn = ± Fnmax an + bn (8.3)
 
1000 Fjmax

 
 j=1 
j 6= n

This expression is used to link the inaccuracy of the individual measured balance signal to its full scale
range.
How the balance uncertainty depends on the choice of the calibration procedure is discussed in the pro-
vided literature [11, 17–19]. However procedures as one factor at a time (OFAT) or a more statistics based
approach, such as modern design of experiment (MDOE), do have different effects on the uncertainty, see
figure 8.2.

G. Eitelberg 119 Delft University of Technology

 Figure 8.2: Load table designs.

G. Eitelberg 120 Delft University of Technology

Bibliography
[1] Franky Aerodynamics. Flow control techniques – from aerospace to motorsport. https://tianyizf1.
wordpress.com/tag/s-duct/, January 2015.
[2] J.D. Anderson. Modern Compressible Flow; With Historical Perspective. McGraw Hill, 2004. ISBN
007-124136-1.
[3] G.I. Barenblatt. Ähnlichkeitsgesetze und Modellregeln der Strömungslehre. Cambridge University
Press, 1991. https://doi-org.tudelft.idm.oclc.org/10.1017/CBO9780511814921.
[4] A.L. Barslow. Review of the effect of distibuted surface roughness on boundary layer transition. Langley
Research Center, Hampton, 1960.
[5] G. Boyet. Eswirp: European strategic wind tunnels improved research potential program overview.
page 249–268, 2018. DOI: 10.1007/s13272-018-0305-4.
[6] E. Buckingham. On physically similar systems illustrations of the use of dimensional equations. Phys-
ical Review, 4:345–376, 1914.
[7] W. R. Corliss D. D. Baals. Wind Tunnels of NASA. 1981.
[8] G. Eitelberg. Aiaa 94-2525 first results of calibration and use of the heg. 6 1994.
[9] B.F.R. Ewald, editor. AGARDograph 336: Wind Tunnel Wall Corrections. North Atlantic Treaty Organi-
zation, 1998.
[10] C.C. Rossow H. von Geyr. A correct thrust determination method for turbine powered simulators in
wind tunnel testing. 7 2005. AIAA 2005-3707.
[11] J. Zhai I. Philipsen. Comparative study of strain-gauge balance calibration procedures using the bal-
ance calibration machine. 1 2007. AIAA 2007-143-923.
[12] H. Holthusen J. Pereira Gomes, A. Bergmann. Aeroacoustic wind tunnel design. 2019.
https://doi.org/10.1007/s13272-019-00372-7.
[13] F. Jaarsma. Report of the AGARD Ad Hoc Committee on Engine-Airplane Interference and Wall
Corrections in Transonic Wind Tunnel Tests. AGARD-AR-36-71, 1971.
[14] A. Pope J.B. Barlow, W.H. Rae. Low-Speed Wind Tunnel Testing. 1999.
[15] M.J. Lighthill. Introduction: Boundary layer theory. 1963.
[16] NASA. Full-scale tunnel story told in new nasa book. https://nasa.gov/aero/cave-of-the-winds.
html, September 2014.
[17] I. Philipsen R. Bergmann. An experimental comparison of different load tables for balance calibration.
6 2010. AIAA 2010-4544-548.
[18] I. Philipsen R. Bergmann. Some contemplations on a proposed definition of uncertainty for balances.
6 2010. AIAA 2010-4546-329.
[19] I. Philipsen R. DeLoach. Stepwise regression analysis of mdoe balance calibration data acquired at
dnw. 1 2007. AIAA 2007-144-278.
[20] D. Reckzeh. Aerodynamic design of airbus high-lift wings in a multidisciplinary environment. 7 2004.
[21] M. Pott-Pollenske V. Ciobaca, S. Melber-Wilkending. A cfd process chain for simulating open windtun-
nel test sections. 2009.
[22] L.L.M. Veldhuis. Propeller wing aerodynamic interference. 6 2005. ISBN 90-9019537-8.

G. Eitelberg 121 Delft University of Technology

[23] Wikipedia. Subsonic and transonic wind tunnel. https://en.wikipedia.org/wiki/Subsonic_and_
transonic_wind_tunnel, October 2015.
[24] Wikipedia. Supersonic wind tunnel. https://en.wikipedia.org/wiki/Supersonic_wind_tunnel, 09
2015.
[25] Wikipedia. Wheatstone bridge. https://en.wikipedia.org/wiki/Wheatstone_bridge, 05 2021.
[26] Y. V. Polezhaev. Wind tunnels. http://www.thermopedia.com/content/1263/, 02 2011.
[27] J. Zierep. Ähnlichkeitsgesetze und Modellregeln der Strömungslehre. 1991.

G. Eitelberg 122 Delft University of Technology