Académique Documents
Professionnel Documents
Culture Documents
Fluid Mechanics 3
Flow Measurement Methods
Tom Bruce1
February 2006
Summary
This short course aims to generate an awareness of the range of contemporary flow measurement
devices and methods available for application to both industrial and research flow problems in
Mechanical Engineering. Well-established mass and volume flow rate measuring devices are
reviewed, and the strengths and weaknesses of various meters and classes of meter are discussed.
Modern non-invasive methods - magnetic and ultrasonic - are also discussed.
Velocimetry (or anemometry) methods are then discussed, with a distinction drawn between
point measurement methods and 2-D methods. In the former category, Laser-Doppler
anemometry is described in detail. Under 2-D methods, Particle Image Velocimetry is described
in detail and a range of applications presented.
1
Lecturer, School of Engineering and Electronics, University of Edinburgh, King’s Buildings, Edinburgh,
EH9 3JL, Scotland. Tel: +44 (0)131 650 8701, email: Tom.Bruce@ed.ac.uk
1
flow meas notes 0506 v04.doc
1. Introduction
1.1 Rationale, Aims and Objectives
This short course was introduced in 93/94 to reflect a major research interest of the Fluid
Mechanics research group within the School.
Fluid flow measurements are performed across the breadth of engineering, eg flows of oil, gas,
petrol, water, process chemicals, effluent are all necessarily and routinely measured. In the
research laboratory, advanced flow measurements are providing new insights into a wide range of
engineering flow problems in hydrodynamics (eg wave impact loading on coastal defences,
beach erosion) combustion (eg low NOx burners, IC engines), aerodynamics (eg wind turbine
optimisation and performance prediction) to list but a few.
The course aims to generate an awareness and understanding of the range of contemporary flow
measurement techniques available with the emphasis on devices and techniques with wide
application in Mechanical Engineering. It is the objective of the course that by its end, the
participant should be able
• to describe the principles of operation of differential pressure, positive displacement, rotary
inferential, fluid oscillatory, electromagnetic and ultrasonic flow meters.
• to discuss advantages and disadvantages of the above meters for different applications.
• to design systems incorporating differential pressure meters.
• to describe the principle of operation of hot-wire anemometry.
• to describe the principles of Laser-Doppler Anemometry (LDA).
• to discuss the strengths, weaknesses and limitations of LDA.
• to design LDA systems to suit given experimental flow problems.
• to describe the principles of Particle Image Velocimetry (PIV).
• to discuss the strengths, weaknesses and limitations of PIV.
• to design PIV systems to suit given experimental flow problems.
The most common industrial flow measurement requirement is a measure of the volume or mass
of fluid flowing per second through a given cross-section of a pipe. A wide range of devices exist
for these purposes reflecting the wide range of conditions which may prevail – liquid flow, gas
flow, fluid temperature, pressure, viscosity, conductivity, the cleanliness of the fluid, the presence
of flow disturbance… A selection of the most common and useful devices are presented in
Section 2.
In many applications, particularly in the research laboratory, it is the actual local flow velocity
that is of interest rather than the total flow rate. Velocimetry methods fall into two broad
categories: those which measure the flow velocity at a single point and those which offer velocity
data over a 2-D plane or even a 3-D volume. Point measurement methods may provide very high-
resolution time histories of the velocity at a point. However, if a flow is neither steady nor
precisely repeatable or periodic, then single point methods cannot be used to build up a 2-D
velocity map in a point by point manner. 2-D methods are a recent addition to the armoury of
methods of flow measurement. Point measurement methods are described in Section 3. Section 4
is devoted to the description of 2-D methods.
Differential pressure flow meters all infer the flowrate from a pressure drop across a restriction in
the pipe. For many years, they were the only reliable methods available, and they remain popular
despite the development of higher performance modern devices, mostly on account of
exceptionally well researched and documented standards.
The analysis of the flow through a restriction (Figure 2.1) begins with assuming straight, parallel
stream lines at cross sections 1 and 2, and the absence of energy losses along the streamline from
point 1 to point 2.
D1 Dt D2
& . By continuity,
The objective is to measure the mass flow rate, m
m& = !u1 A1 = !u2 A2 [2.1]
Bernoulli’s equation may now be applied to a streamline down the centre of the pipe from a point
1 well upstream of the restriction to point 2 in the vena contracta of the jet immediately
downstream of the restriction where the streamlines are parallel and the pressure across the duct
may therefore be taken to be uniform:
u12 p1 u2 2 p2
+ = + [2.2]
2 ! 2 !
assuming that the duct is horizontal. Combining with [2.1] gives
A2
m& = 2 " ( p1 ! p 2 ) [2.3]
A22
1!
A12
For a real flow through a restriction, the assumptions above do not hold completely. Further, we
cannot easily measure the cross-sectional area of the jet at the vena contracta at cross-section 2
where the streamlines are parallel. These errors in the idealised analysis are accounted for by
introducing a single, cover all correction factor, the discharge coefficient, C, such that
CAt Dt
m& actual = 2#( p1 ! p2 ) where ! " . [2.4]
1! "
4 D1
Dt and At are the diameter and area of the throat of the restriction respectively.
The three main restriction devices are the orifice plate, Venturi meter and flow nozzle.
Orifice Plate
Orifice plates vary in the profile of the orifice and the location of the pressure taps. Figure 2.2
shows the tapping arrangements covered by BS EN ISO 5167: flange taps, pipe taps (both D and
D/2) or corner taps. Figure 2.3 shows a typical orifice, with a sharp, square edge on the upstream
side. Installation of the plate back-to-front can introduce an error ∼20% in m&!
Older standards (BS 1042) give C as a (relatively) simple function of pipe Reynolds number Re
and diameter ratio β, eg for corner taps,
" 2 .5
C = 0. 5959 + 0. 0312" 2.1 ! 0.184"8 + 91. 71
Re 2.5
±0.6% for 0.2 < β < 0.75 and 104 < Re < 107 .
C
The graph in Figure 2.4 shows this relation in terms of the flow coefficient, K ! .
4
1" #
The current standard BS EN ISO 5167-2 gives more accurate (but much longer) equations for C.
Venturi Meter
The Venturi meter (after Giovanni Venturi, 1746−1822) is designed to cause minimal head loss
as the flow passes the restriction. Figure 2.5 shows a typical arrangement. Like the orifice plate,
the Venturi is dealt with by a British / ISO standard (BS EN ISO 5167-4)
For a Venturi, C = 0.99 for 105 < Re < 107 is a useful approximation. BS EN ISO 5167-4 gives
more accurate Figures for particular details of an installation (ranges of Re, β etc).
The main advantages of the Venturi over the orifice plate are
• low head loss
• less affected by upstream flow disturbance
• good performance at higher β
• even more robust
• self-cleaning
• less affected by erosion
Flow Nozzle
In many respects, the flow nozzle is a compromise between the compact orifice plate and the
efficient Venturi. There are two standardised designs - Figures 2.6 and 2.7. Flow nozzles have
proved particularly suited to high velocity applications, eg steam metering.
Figure 2.6: The ISA flow nozzle. Figure 2.7: The ASME long radius nozzle
(Furness, 1989). (Furness, 1989)
For the ASME nozzle (see BS EN ISO 5127-3)
6. 53" 0.5
C = 0. 9995 ! ± 2% for 0.25 < β < 0.75 and 104 < Re < 107.
Re 0.5
Figure 2.8 shows this relation graphically. Figure 2.9 shows a comparison of permanent ie
unrecovered head losses caused by the three designs of restriction flow meter that have been
considered.
Advantages:
• better head loss characteristics than orifice plate
• self-cleaning
Drawbacks:
• higher cost than orifice plate
• more sensitive to upstream disturbance than Venturi.
2.2 Constant Pressure Flow Meters
This class of meter measure the flow by dividing the fluid into “packets”, each of a precisely
known volume. The number of such packets counted in a known time gives a precise measure of
the volume flow rate. They are also known as PD Meters or simply Displacement Meters. Under
suitable conditions, this class of device offers the highest performance of any mechanical meter,
achieved through careful manufacture to high tolerances.
The Sliding Vane meter, Figure 2.11, is among the most accurate PD meters - uncertainties in the
volume flow rate Q may be less than ∼0.2%. The dynamic range is also quite high, typically
∼20:1. The liquid is channelled smoothly into the measuring crescent, minimising head losses.
The vanes and channel are carefully machined to give smooth operation with very low leakage.
The number of rotations is usually counted mechanically.
Another PD meter in widespread use is the Oval Gear meter - Figure 2.12. Again, close
tolerances ensure minimal leakage. It is unlikely that an oval gear meter can approach the
accuracy of the sliding vane meter. They may also introduce a much greater pulsing into the
outlet stream, which may or may not be a concern.
A further variation on this theme is the Nutating Disc meter (Figure 2.13). The incoming fluid
fills the chamber, alternately above and below the disc, driving the disc in a rocking, circular
motion - nutation (cf spinning top ?). There is a greater area over which leakage could occur than
for sliding vane or oval gear meters, so the accuracy is not in general so good, although meter life
is potentially longer.
By far the most common gas PD meter is the Diaphragm meter, used worldwide to meter
domestic gas consumption (Figure 2.15).
The two chambers are filled and emptied alternatively, controlled by a sliding valve as shown.
The motion of the diaphragm is connected mechanically to a counting mechanism and readout.
Mechanical reliability is outstanding, but the design is really suited only to low flow rate, low
pressure gas flows. Mass production means that these are inexpensive devices.
At lowest flow rates, the friction of the moving parts may become significant and lead to the
meter running too slowly. Leakage is also most likely to be significant at lowest flows, again
leading to under-reading.
This class of meter all rely on moving parts with close tolerances, so sustained operation over
long periods is not their strength. A further consequence of their close tolerances is that they may
be sensitive to temperature changes, and almost certainly will not operate well is the flow is not
clean, ie has particulate matter entrained at all.
This class of meter are all basically small hydraulic turbines running at zero load. The rotary
element rotates at an angular velocity which is proportional to flow rate. This rotation speed is
monitored by mechanical means, or better, by magnetic or optical methods.
The basic axial turbine flow meter is shown in exploded view in Figure 2.17.
The axial flow turbine has a bladed rotor running on bearings. The assembly is mounted on a
central shaft, which is itself held by hanger assemblies up and downstream. A magnetic pick up
senses the turbine blades as they pass, and the pulse frequency is the measure the flow rate. The
total number of pulses recorded measures total volume passed.
The driving torque of the fluid is resisted by mechanical bearing friction, fluid drag and magnetic
drag, all of which effects vary with flow rate. The relation between flow rate and the rotation
speed may be written
Q = K. !
where K is the meter factor which must be calibrated for a given meter.
Simplification of the axial turbine meter is possible by using a mechanical pick up to drive a
counting system. Such meters are clearly less accurate, and so may be manufactured to lower
tolerances. The cost is lower, and they are also more suited to dirty flows.
Figure 2.20: Strouhal number vs. Reynolds number for circular and triangular section
bluff bodies..
Figure 2.20 shows how the Strouhal number varies with Reynolds number for vortex shedding
from bluff bodies of triangular and circular cross-sections. Clearly, for the circular body, S ≈ 0.2
over the range ∼500 < Re < 2x105. The principle of the vortex shedding meter is to measure this
shedding frequency, so such a meter has a potential to measure flows over an enormous dynamic
range, ∼400:1. A similar situation pertains for the triangular body, for which S ≈ 0.15 over ∼200
< Re < 105.
Figure 2.21: The vortex shedding flowmeter Figure 2.22: Bluff body shapes...
Figure 2.21 illustrates the principle of a flowmeter based upon vortex shedding from a bluff
The flowrate in a channel of diameter D, for a bluff body of “characteristic dimension” d is given
by
(. D4 f " 4 kd %
Q= $1 ! '
4S. d # (. D&
where S is the Strouhal number, f is the vortex shedding frequency and k is a shape factor for the
bluff body (taking account of width and aspect ratio). Figure 2.22 illustrates a range of shapes
proffered by manufacturers of these flowmeters.
Relatively recent arrivals, these meters are now competing with DP meters in many areas, eg
water, steam and air. Linearity may be as good as ±0.5%, and achieved dynamic ranges ∼15:1 (cf
Orifice meter, 4:1 at best).
• No moving parts, crevices or seals - suitable for applications where high flow cleanliness is
important (eg semiconductor manufacture)
• May be used with liquids or gases.
• Insensitive to fluctuations in temperature, pressure or density.
• Digital output.
When a conductor is moved through a magnetic field (Figure 2.23), a voltage is induced across
the conductor. This voltage (emf) e is proportional to the field strength B, the velocity u of the
conductor and the length l of the conductor:
e = Blu
This is the basis of the electromagnetic flowmeter, or magmeter. For the magmeter, the conductor
is the flowing fluid, a field is applied across the pipe, and the induced voltage is therefore a
measure of the velocity with which the conductor is moving. Clearly, non-conducting fluids such
The ultrasonic “time of flight” meter is in its relative infancy. It is probably the only type of
meter capable of high performance (eg 1%) with bores of >3m. The basis of operation is the
measurement of the difference in the “time of flight” of sound waves propagated in opposite
directions, with and opposing the flow. Figure 2.25 illustrates the arrangement.
d 1 d 1
t1" 2 = . and t2 !1 = .
sin è (c ! u cos è ) sin è (c + u cos è )
These meters have been used successfully on water flows, clean process fluid flows and on
natural gas pipelines. As with electromagnetic meters, ultrasonic meters cover the whole
spectrum of sizes from mm up to an 11m bore application on a hydro scheme.
The basis of operation of this class of flowmeter is that if sound of a given frequency is reflected
from a moving object, the frequency of the reflected sound is shifted by an amount proportional
to the speed of the moving object (cf passing ambulance, driving past players of bagpipes in
Glencoe laybys...). In these maters, ultrasound is transmitted into a flow which contains scatterers
travelling with the flow (eg dirt particles, bubbles), and the scattered sound wave detected by a
receiver. The frequency shift is then a measure of the flow speed. A possible arrangement is
shown in Figure 2.26.
Electromagnetic meters
• Faraday’s Law: Voltage induced across conductor (conducting fluid) moving in magnetic
field ∝ speed of conductor, ie flow rate ∝ induced voltage across duct.
• Long-term reliability.
• Can be used only with conductive liquids - no gases.
• Moderate to good accuracy; better than 1% when installed correctly.
• Moderate dynamic range; ∼10:1
• Application possible in extreme conditions of temperature, pressure and flow rate.
• Demanding applications in dirty flows, corrosive liquids, non-Newtonian liquids possible.
• Wide range of pipe sizes.
• Bi-directional flows OK.
• Non-invasive - zero head loss.
• Insensitive to upstream disturbances.
• Sound propagated with and against the flow. Difference in time of propagation ∝ flow rate.
• Long-term reliability.
• Clean gases and liquids.
• Very accurate.
• Large dynamic range.
• Bi-directional flows OK.
• Extremely large pipe sizes possible.
• Non-invasive - may be just clamped to pipe.
• Sensitive to pressure and temperature changes.
• Ultrasound scattered from moving scatterers in flow. Shift of frequency ∝ speed of scatterer.
• Non-invasive - zero head loss.
• Bi-directional applications possible.
• Liquids or gases.
• Wide range of sizes.
• May simply be clamped to pipe.
• Works with dirty or aerated flows.
• Sensitive to velocity profiles.
• Flow must contain ultrasound scatterers.
• Long-term stability unproven.
• Low accuracy.
Section 2 has reviewed a wide range of devices for measuring the total flow rate in pipes. In this
Section, we concentrate on methods for measuring the actual fluid flow velocity at a single point
in the flow. Methods suitable for velocimetry over a two-dimensional field of points are held
back until Section 4.
Whereas most applications of the flowmeters of Section 2 are clearly in the industrial arena, the
focus tends to shift to the engineering research laboratory as attention is turned to anemometry
(UK term) or velocimetry (US-imported term).
The humblest of devices for measuring flow velocity directly is the Pitot-static tube. Figure 3.1
shows the principle of operation.
p p
1 2
1 2
We apply Bernoulli’s equation to a streamline which meets the tip of the tube. The flow is steady,
so there is no flow in the tube. Thus there is a stagnation point, so u2 = 0. The pressure difference
p2 - p1 is the difference between the impact or stagnation pressure at the tip of the tube, p2 , and
the static pressure in the body of the fluid, p1 . From Bernoulli,
2(p2 ! p1 )
u1 =
"
The most common practical design based upon the above is shown in Figure 3.2. A pair of
concentric tubes is used: the inner tube measured the impact pressure, the outer tube has a
number of tiny tappings, flush with the tube, to measure the static pressure.
Accuracy is crude, but these devices do provide a very simple and fast estimate of flow velocity.
They are clearly not well suited to dirty flows in which their tappings may become blocked.
In most designs, the current through the wire is kept constant, and the change in resistance is the
measure of the local flow velocity. It is possible to record a time history of the flow velocity at a
particular point, and very high time resolutions -up to 50 kHz - are possible - though clearly such
rates require highly sophisticated electronics to track wire resistance changes. Two or three wires
may be arranged orthogonally to give an estimate of two or all three velocity components.
Hot wire anemometry is fundamentally a single point method, so finds most applications in flows
whose structure is well known a priori, and where the interest is in the time variation of velocity
at a point, eg in wind tunnel studies of vortex shedding, or in measurements of turbulent
intensities.
Introduction
The early development Laser Doppler Anemometry (LDA) dates back to the very end of the 60s,
when low power continuous wave (CW) lasers began to become available at costs which were
not astronomical. Since then it has developed into a sophisticated and robust tool suited both to
research laboratories and industrial applications. Developments in electronic and computer
processing have improved data gathering and reduction beyond measure, and optical
developments, notably in fibre-optics, have opened up many new application possibilities.
The basis of LDA is not complicated, but a short digression into some properties of light and of
laser beams will prove useful.
Laser Basics
All objects emit thermal radiation - a continuous range of frequencies with the peak wavelength
of the spectrum depending upon the temperature of the body. Hotter bodies emit a distribution of
radiation peaked at shorter wavelengths = higher energy radiation. The radiation from the Sun
peaks at about 500 nm wavelength - green light - corresponding to a temperature of around 6000
K. Photons from thermal sources are emitted over a wide range of wavelengths and in all
directions.
The radiation from a laser (Light Amplification by Stimulated Emission of Radiation) is quite
different in character. Photons emerge with identical energies, ie identical wavelengths, identical
phases, and all travelling in the same direction.
N1
! E = hf
E=0
thermal source laser source
Figure 3.3 illustrates the fundamental difference between thermal and laser radiation. Quantum
mechanics dictates that electrons “orbiting” the nuclei of atoms cannot have any arbitrary energy,
but must be in one of a number of discrete energy states. If the electron absorbs a “quantum” of
radiation - just the right amount of energy - then it moves up to the next discrete energy level.
Equally, it may emit radiation, and in doing so, lose energy and fall back one or more levels. The
energy associated with a photon of light of frequency f is given by
E = h. f
where h is Planck’s constant: h = 6.63 x 10-34 Js.
In a collection of atoms in a normal state, electrons are continuously jumping between a large
number of different states, separated by a range of different energies - they continuously absorb
and emit over a range of energies and therefore frequencies. In a laser, it is arranged by some
means that a large number of atoms have a higher proportion of electrons in a particular, raised
state - a population inversion. There is a tendency for these electrons to drop back to their lower
energy state, and in doing so, they all emit photons of exactly the same energy corresponding to
the energy difference between the levels.
Lasers have a unique ability to form light beams with high energy concentrations. However, the
quantum nature of photons means that a small divergence of the beam is always present. The
beam cannot be focused to a point, but only to a beam waist of thickness 2wo , Figure 3.4.
2w0
!
lens
Figure 3.4: Laser beam waist.
The beam waist thickness is given by
"
!=
#. w0
Thus efforts to bring the beam to sharper focus result in increased divergence.
Example: λ = 514nm, w0 = 12mm, θ = 0.014 mrad ⇒ w = 25mm at ∼1000m from waist.
λ = 514nm, w0 = 1mm, θ = 0.164 mrad ⇒ w = 25mm at ∼6m from waist.
Optics Basics
Speed of light in a vacuum: c0 = 3.0 x 108 ms-1
Speed of light in a medium of refractive index n, c = c0 / n .
c
f = : f = frequency (Hz) independent of medium, λ = wavelength (m)
!
2!
k= : k = wavenumber (m-1)
"
Interference:
Figure 3.5 shows how the wavefronts associated with two laser beams cross. When two peaks
coincide, there is constructive interference and a local maximum of intensity results. When a
peak and a trough coincide, they cancel and a locally dark area results - destructive interference.
bright
dark
!/2
bright
dark
bright
s !
u
!/2
laser
photodetector
beam splitter
!/2
u
"
Practical Implementation
LDA systems are generally bought “off the shelf”. Dantec (based in Copenhagen) and TSI Inc.
(USA) are the market leaders. The beam splitting and convergence optics are generally packaged
into one black box, the detector plus its optics into another, and the signal processing carried out
by sophisticated electronics in a third black box, usually linked to a PC or workstation.
The seeding of the flow for LDA is not usually a problem: only in the cleanest conditions in
water flow experiments are there no suitable scatterers present. In gas flows, corn oil droplets
have been used to good effect.
Most systems are laid out as sketched in Figure 3.8 - a forward scatter arrangement. This requires
optical access to both sides of the test volume. An alternative is to work in back scatter mode,
detecting light scattered back in the direction of the incoming beams. This set up has the
disadvantage that the much more light is scattered forward than back, so for a given laser power,
there is much less light scattered to the detector. However, optical access is required from only
one side, making it well suited to use with fibre optics - two fibres are used - one to carry the
laser light to the measurement volume, and the second to carry the back-scattered light to a
detector. A tiny optical head on the end of the fibre does the beam splitting and alignment.
Multi-Component LDA
This chapter reviews the Particle Image Velocimetry (PIV) flow measurement method. All
stages in the process of obtaining PIV flow maps are described. The underlying principles are
introduced in Section 4.3. The acquisition phase of PIV is divided into illumination (Section 4.5)
and image acquisition (Section 4.6). The analysis phase - extracting the velocity data from the
PIV flow record - is discussed in Sections 4.4 and 4.5. The chapter concludes with a review of
applications and brief look at the future of the PIV technique.
4.2 Motivation
The basic principle of PIV is very simple: tiny tracer particles in the flow are illuminated and
their positions recorded photographically or by video at two successive instants. The
photographic or video “flow record” of the whole 2-d field is then divided into a grid of cells and,
in each cell, the distance moved by the tracer particles from the time of the first image being
recorded to the second is determined. Knowing this distance travelled and the time taken, the
velocity of the flow in that cell is therefore measured. This is done for every cell, giving a 2-d
map of the “instantaneous” velocity field at the time of the recording.
Thus it can be seen that implicit in the method are two basic assumptions; that the flow field as a
whole does not change over the time between first and second exposures, and that the velocity in
each cell is uniform. In any real flow of interest, neither assumption will hold fully and the extent
to which they hold is always an important consideration.
In the early years of PIV (1980s), the usual approach was to illuminate the flow stroboscopically
and to record a double- or multiple-exposure photograph of the flow, ie holding the camera
shutter open long enough such that each tracer particle is illuminated twice and therefore gives a
pair of images on the flow record. This is PIV using the autocorrelation method.
With developments in still image video technology, the complication of arranging for the
stroboscopic illumination of a plane within the flow can be avoided by exposing the first and
second images as successive frames recorded by a CCD2 camera or on two separate CCD arrays.
The movement of the tracer particle images over the time between the exposure of the first frame
and the second can then be analysed to give the velocity map. This is the cross-correlation
method. Cross-correlation has a number of significant advantages over autocorrelation and is
now almost universally the preferred approach.
2
charge-coupled device - a video array chip
General objectives for any analysis system can be laid down quite simply. We have a flow
record or successive records containing velocity information over a large field. We wish to be
able to define a grid of points over this area, and at each point, interrogate a small area of the
photograph to establish the most common (most correlated) particle image separation.
Additionally, we require that this process be highly accurate, fast (ie hours are not taken to
analyse one flow record), and, apart from initialising a run, completely automated.
The basis for cross-correlation analysis are two PIV flow records taken of exactly the same field
in a flow separated by a (usually short) time interval.
Figure 4.1 shows typical interrogation areas from two successive records - selected small cells
within base and cross images, in this case of 32 x 32 pixels. A number of approaches may be
taken to finding the distance moved by the tracers between frames. For example, if the number
density of particle images is low, one approach might be to carry out some image processing to
establish the positions of all images and in some way try to pair the images. With a higher
density of images a method based on finding the cross-correlation of the images within the
interrogation areas is a very efficient and robust approach.
The cross-correlation function C(Δx, Δy) is arrived at by comparing the base and cross images
given by intensity distributions Ibase(x, y) and Icross(x, y). In physical terms, the function is moving
the cross image relative to the base image, seeking the best match between the intensity patterns.
The clearly visible peak in the function indicates the offset in x and y between the base and cross
images at which the best correlation between the images was found. Once the software has
calculated the cross-correlation function, it locates this peak and records the Δx, Δy values.
Knowledge of the optical magnification and the time interval between images allows Δx, Δy
subsequently to be scaled to give a velocity vector relating to the location of the interrogation
area.
Illumination for must usually also define a 2-d plane or “sheet” through the flow. The source of
the light is typically a pulsed laser. Lasers are used because the low divergence of their beams
allows thin sheets to be formed, and because the energy density in the beam is very high, eg we
may need enough energy to image a 50 µm tracer particle with a camera 1 m away within a few
microseconds.
PIV illumination is usually achieved either by an expanded beam method (Figure 4.3). The
pulsing of the illumination may be achieved by employing a pulsed laser or (more historically)
combining a high speed shutter with a “continuous wave” (CW) laser.
4.6.1 Introduction
The process of acquiring good PIV flow records involves two major stages. The first is selecting
appropriate hardware for the application eg, illumination, seeding and camera, and the second is
using this hardware to best effect. This Section addresses both of these stages.
4.6.2 Hardware
The hardware considerations may be divided into three broad (and somewhat interdependent)
areas: illumination, flow seeding and camera. The illumination system preferred for the
application of PIV to hydrodynamics has been discussed (Section 4.5).
Seeding
The selection of a suitable flow seeding is very important. The seeding used in hydrodynamics
experiments at Edinburgh is conifer pollen. This meets the most important criteria for a suitable
seeding: once soaked with water, it is almost exactly neutrally buoyant, quite reflective at the
wavelength of the laser, and small enough to follow the flows studied (typical particle diameters
are ∼70µm). Importantly, it is also quite inexpensive - the cost to seed a flume containing six
tonnes of water is ∼ £2 - £5.
For air flows of high and low speeds, hollow glass spheres have proved a very successful
seeding. Diameters are ∼10 - 20 µm. Such small particles have a tiny terminal velocity in air, and
so do not rapidly fall to Earth under gravity.
Generally now, a wide range (sizes and densities) of synthetic seed particles are available from
PIV equipment suppliers.
The choice of the camera and lens is also important. The lens should be a flat-field lens in order
that distortions of the image plane are minimised. Choosing a lens of longer focal length reduces
the apparent effect of any out-of-plane motions of particles in the field, but increases the
difficulty of achieving a really sharp focus and makes the process more sensitive to vibrations.
In the early years of Edinburgh PIV, a Hasselblad 500 EL/M camera was the mainstay of PIV
measurements in hydrodynamics. In the early 90s, it was joined by a Kodak MegaPlus “still
image video camera” (what would now be called simply a “digital camera”). It was a
monochrome camera with 4 Mpixels (2048 x 2048 pixels resolution). It cost over £20k - compare
this with the cost of a colour 4 Mpixel digital camera today!
Very high speed digital cameras have also been used successfully for PIV. Frame rates of up to
30kHz have been used, though at some cost in spatial resolution.
After the selection of the hardware, the important issues remaining are the choice of camera
position, achieving the correct seeding density, optimising the focus and optimising the exposure
parameters to give good, high contrast flow images
Seeding
The optimum seeding density is determined by considering the subsequent analysis of the
photograph. In typical applications in the wave flumes at Edinburgh, the local flow velocity at a
“point” on the flow record is averaged over a 32 x 32 pixel interrogation area (IA). Thus the
seeding density should be high enough that there will always be several (∼5 to 15) tracer particles
in each IA on the flow record. Experience is perhaps the best guide to getting an optimum level
of seeding; once good results have been obtained, the successful level of seeding can be repeated.
The size of the images of the seeding should also fall within reasonable, intuitive bounds. Seeing
images which fill a substantial part of the IA would be too large to offer a good correlation
between base and cross images. Similarly, if the image is so small as to be less than one pixel in
size, then the analysis will only ever be able to “see” particle image movements of whole units of
pixels (ie 1 pixel; 2 pixels; 3 pixels etc) whereas the shift (base to cross) of an image spread over
a few pixels can be measured to “sub-pixel” accuracy.
The seeding should also be of a size that forms images of the desired size. The size of a particle
image on the recording medium, di is determined by three factors: the particle diameter, dp , the
magnification (recorded:actual size) M, and the lens aperture used. defined by its “f-number3”, f#
:
di = M 2 d p 2 + d spot
2
where dspot is the diffraction-limited image size - the smallest image that can be recorded by a
lens working at the given f-number, f# ,
d spot = 2.44(1 + M ) f # !
Thus for larger magnifications, larger particles and larger apertures (smaller f-numbers), the size
of the diffraction-limited image becomes less significant, and the size of the recorded image
approaches that which would be expected on the basis of the magnification alone.
Illumination Interval
Like the optimisation of seeding density, the choice of illumination interval is dictated by
consideration of the subsequent analysis. The interval should be set such that the largest velocity
3
The working diameter or “aperture” of the lens is usually given in terms of the lens focal length, f, eg an
aperture of f/8 implies that the aperture is 1/8th of the focal length. In this case the “f-number” f# = 8.
Focus
Achieving a really sharp focus is extremely important if good PIV photographs are to be
obtained, and can be quite difficult, especially when a large lens aperture or long focal length lens
are being used. Tests in which the focus is varied can provide a route to optimisation.
Photographic Magnification
The magnification from the measurement zone to the image depends upon the focal length of the
lens and the distance from the camera to the measurement zone. Its selection is in general
another compromise. It is often desirable to measure as large a region as possible, but it is again
important to consider the analysis phase. The velocities are calculated from averaging particle
image displacements over a small (eg 32 x 32 pixel) interrogation region on the flow record. The
implicit assumption is that particle image displacements over this small region are uniform: if
there is a strong displacement gradient present, errors are introduced and the resulting data point
may be at best inaccurate and at worst, spurious. Therefore the size of the area imaged onto the
flow record is typically limited to that which will result in displacement gradients of less than
∼5% over any interrogation area, limiting systematic error from this source to less than ∼1%.
Exposure
In general, tests trying various settings of camera aperture and laser power is required for a new
application. The quality of the resulting flow records may then be assessed under analysis and
the best settings finalised.
If there is scope for choosing an aperture setting, it should be remembered that the largest
apertures (smallest f numbers) give the poorest depth of field, so focusing is more difficult.
However, if the particle image size is diffraction limited, the smaller apertures (larger
f numbers) will result in larger particle images. Using f/4 or f/5.6} is usually a good compromise.
The geometric relation between the measurement plane in the flow and the PIV image plane in
the camera may be generally specified by six coordinates - three positional and three angular
(Figure 4.7). The selection of these coordinates corresponds to the positioning and alignment of
the camera.
The position of the camera in x1 and x2 (Figure 4.4) is determined by the characteristics of the
measurement zone. In the simplest cases, the camera field of view will be centred on the centre
of the measurement zone. However, if there is some feature in the measurement zone which
partially obstructs the view such as a test object or a free surface, then the selection of the camera
position may demand some care if as much as possible of the region of interest is to be imaged.
The x3 position, the distance from camera to measurement plane defines the magnification. The
magnification is chosen as a compromise between the size of the measurement area and the
resolution of the velocity map.
Aligning the camera fixes the rotational degrees of freedom, ϕ1 , ϕ2 and ϕ3 . The pitch ϕ1 and
yaw ϕ2 axes must be set so that the camera image plane is parallel to the measurement plane. If
this is not achieved precisely, not only will a systematic distortion of the image plane result, but
also it may prove impossible to maintain a sharp focus over the whole of the plane. The roll axis
ϕ3 needs to be set if the object and image planes are not to be rotated relative to one another.
Vibration
The sharpness of the particle images recorded on the film is an important factor in the
signal-to-noise ratio of the resulting data. Therefore it is important to consider possible sources of
mechanical vibration in the recording system and how they might be minimised. Two possible
sources can, in general, be identified:
1. the mounting of the camera
2. the internal workings of the camera
If the camera is mounted on a good quality tripod, itself standing on a solid lab floor, there should
be little or no problem from this quarter. However, if the camera is mounted otherwise, eg upon a
moving stage in an image shifting system, or if the tripod rests on a floor which vibrates for any
reason, related to the experiment or not, then steps may have to be taken to minimise or at least
quantify the effect of vibration.
Registration
Calibration
Finally, provision should be made to measure the photographic magnification from the laboratory
frame to the image frame. Again, it may be possible to measure this directly from the image of
an object of known size in the measurement zone. However, in most cases, it may be preferable
to take a dedicated calibration photograph before the start of a run of experiments, eg of an object
or grid of accurately known dimensions.
A summary of the errors inherent in the acquisition phase of autocorrelation PIV is presented in
Skyner [1992], from which the table below is reproduced.
4.7 Applications
A random selection of applications with which the Edinburgh group has been associated...
Combustion
• Flow in power station coal burners (live flame tests)
• Flow of pulverised coal into power station combustion chamber
Coastal Engineering
• Beach erosion /accretion
• Wave impacts on breakwaters
• Flow around seabed pipelines
Offshore Engineering
• Wave enhancement due to structure blockage
• Oceanic internal waves
• Deep water breaking waves
Wind Energy
• Measurement of wake behind a wind turbine
Heat Transfer
• Recirculation in horizontal kettle reboilers
Fire Safety
• Flow of fumes and smoke out of burning buildings