Unsteady Incidence and the Correction of

Pneumatic Probe Measurements

John Coull (  john.coull@eng.ox.ac.uk )
University of Oxford
Henry Ng
RMIT University
Tony Dickens
University of Cambridge, United Kingdom (Now at Cambridge Consultants Ltd
José Serna
Universidad Politécnica de Cartagena

Research Article

Keywords:

Posted Date: October 6th, 2022

DOI: https://doi.org/10.21203/rs.3.rs-2104484/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
 Unsteady Incidence and the Correction of
Pneumatic Probe Measurements
John Coull1, Henry Ng2, Tony Dickens3, and José Serna4

1
University of Oxford, United Kingdom, ORCID 0000-0003-2392-2131
2
RMIT University, Australia, ORCID 0000-0003-4926-7622
3
University of Cambridge, United Kingdom (now at Cambridge Consultants Ltd., United Kingdom)
4
Universidad Politécnica de Cartagena, Spain, ORCID 0000-0001-5593-6330

ABSTRACT NOMENCLATURE
Pneumatic probes such as five-hole probes Symbols
(5HP) have many advantages and can measure flow 𝐴 sinusoidal perturbation amplitude
angles, total and static pressure, velocity and Mach 𝐶 calibration coefficient
number. In most applications, pressure transducers 𝐶𝐷 drag coefficient
are connected to the probe head by lengths of tubes,
𝐶𝑝 , 𝐶𝑝𝑠 pressure coefficients, eq (32), (38)
allowing a simple, miniaturised, low-cost probe
construction. Such “steady” probes therefore 𝑑 probe diameter
measure a “pneumatically-averaged” flow field, 𝐷 dynamic pressure coefficient = 𝑓(𝑡)
which is of sufficient accuracy for many purposes. 𝐸 error
However, this paper demonstrates that when the 𝑓 frequency
probe incidence angle fluctuates, pneumatic 𝑓𝑟 reduced frequency
averaging causes errors in the indicated total and 𝐹𝐷 drag force
static pressure. The aim of this paper is to understand
ℎ span
these pneumatic averaging errors and to demonstrate
practical means to post-correct data. 𝐻 channel height
A quasi-steady model provides an analytical 𝑃 pressure
framework to explain the effect of unsteady flow on 𝑢, 𝑣, 𝑤 fluctuating velocity components
5HPs. Total and static pressure coefficients have a 𝑉 time-mean streamwise velocity
largely symmetric response to positive and negative 𝑆𝑡 Strouhal number
incidence, causing a bias error in the pneumatic 𝑡 time
average. The steady calibration cannot account for
𝑊 width
these effects, leading to erroneous measurements.
𝑥 axial direction
These errors are demonstrated by comparing
5HP and Kiel-shrouded pitot traverses in the 𝑦 transverse direction
shedding wake of a D-shaped body. The 5HP 𝛼 flow angle
overestimates total pressure loss and drag 𝜌 density
coefficients by up to 44% and 60% respectively. 𝜙 phase angle
Similar pneumatic-averaging errors will have Sub-scripts
affected a substantial body of data in the literature.
0 total
A post-correction method is demonstrated that
1 inlet
can be applied to historical data. Estimates of
unsteady flow angles are obtained from a low-cost 2 traverse plane
unsteady computation. The quasi-steady model 𝑎𝑣𝑒 average
accurately corrects the 5HP data, reducing errors by 𝑐𝑜𝑚𝑝 compound
an order of magnitude. Similar corrections are 𝑐𝑜𝑟 corrected
obtained using simple correlations based on the root- 𝑑𝑦𝑛 dynamic pressure
mean-squared angle fluctuation.

1
 𝑓𝑠 freestream consider 5HPs, but the analysis is equally applicable
𝑝𝑖𝑡 pitch to other multi-hole pneumatic probes.
𝑅𝑀𝑆 root-mean-squared 1.1. Previous studies of Pneumatic-Averaging
𝑠 static Most previous studies of pneumatic averaging
Over-bars considered small-scale turbulence. For a pitot tube,
̅ time-averaged Goldstein (1936) argued that the turbulent kinetic
̂ pneumatically-averaged energy will be reversibly recovered if the eddies are
small compared to the probe. Assuming
1. INTRODUCTION incompressible flow, a pitot pointing into the flow
will therefore indicate a total pressure 𝑃̂0 :
Many aerodynamic flows are highly three-
𝜌 2 𝜌 2
dimensional by nature. These three dimensions of 𝑃̂0 ≈ 𝑃
̅+ 𝑉 ̅ + (𝑢 ̅̅̅ + 𝑣̅2 + ̅𝑤̅̅̅2 ) = 𝑃
̅0 (1)
velocity may be obtained from laser techniques (e.g. 2 2
stereoscopic Particle Imaging Velocimetry) or hot- where 𝑃̅ is the time-averaged pressure, 𝑉̅ is the mean
wire anemometry. However, such methods do not velocity and 𝑢, 𝑣, 𝑤 are the velocity fluctuation
measure total pressure, which is often required to components. By these assumptions the probe will
characterise performance, especially for internal indicate the true time-mean total pressure, 𝑃̅0 .
flows. In practice the Goldstein approximation is not
Measurements of flow velocity and pressure are always applicable. Bailey et al. (2013) compared
most simply achieved using a calibrated multi-hole hot-wire and pitot measurements in boundary layers,
probe, such as the five-hole-probe (5HP) shown in and found that only the streamwise fluctuations
Figure 1(a). This paper considers typical “steady” contributed to the indicated total pressure:
5HPs, where the probe head is connected to 𝜌 2 𝜌 2 𝜌
𝑃̂0 ≈ 𝑃̅+ 𝑈 ̅ + ̅̅̅𝑢 =𝑃 ̅0 − (𝑣̅2 + ̅𝑤̅̅̅2 ) (2)
transducers via tubing, and only averaged pressures 2 2 2
can be measured. Steady 5HPs have many
Thus the pitot indicated a total pressure below the
advantages: they are low-cost, simple to construct
true time-mean. A similar, but greater, effect is
and easy to operate. They can be made at small scale
observed in our current analysis of 5HPs.
(~1mm diameter) to minimise spatial errors,
Grimshaw and Taylor (2016). Small scale also Few authors have examined these effects for
enables measurement access, e.g. for intra-row multi-hole probes. Building on the work of
traverses in multi-stage turbomachines. Bradshaw & Goodman (1968), Christiansen &
Furthermore, 5HPs can be made sufficiently robust Bradshaw (1981) examined yaw meters at high and
to operate in harsh environments. With these low turbulence and found high static pressure errors
advantages, steady muti-hole probes are widely used when instantaneous yaw angles were large. This
across a variety of applications. work is again consistent with the current findings.
However, many of the flows where pneumatic
probes have been used are inherently unsteady.
Many turbomachinery experiments measure
efficiency using stationary-frame traverses
downstream of rotating blade rows, for example
Bauinger et al. (2017), Evans and Longley (2017),
Prahst et al. (2015). Other set-ups with unsteadiness
include measurements in shedding wakes, e.g.
Melzer and Pullan (2019) used an un-Kieled pitot to
assess turbine trailing edge loss.
In such unsteady flows, it has been generally
recognised that the “pneumatic-averaged” flow
indicated by a probe may differ from the true time-
average. However, these effects have been poorly
understood and are usually ignored.
This work is motivated by the need to address
these deficiencies. In particular, we develop a
simple, rigorous approach to understand and
quantify pneumatic-averaging errors. The focus on
low-frequency disturbances, which can cause
Figure 1: Five-Hole-Probe nomenclature and
greater errors than small-scale turbulence. We
response in steady and fluctuating flow.

2
1.2. Quasi-Steady Pneumatic-Averaging respectively; §6 presents the results and the
The primary difference between true flow correction. Finally, §7 considers the general
unsteadiness and turbulence is scale. Large-scale implications of the results and discusses strategies to
fluctuations are of most interest as they will affect avoid pneumatic-averaging errors in experiments.
the pressure field around the entire probe head. The
fluctuations can be characterised using the reduced 2. 5HP PNEUMATIC AVERAGING MODEL
frequency 𝑓𝑟 : 2.1. Steady Flow Calibration Coefficients
𝑓𝑑 Non-dimensional pressure coefficients obtained
𝑓𝑟 = (3)
𝑉 from the steady calibration are indicated by the
where 𝑓 is the frequency of the fluctuation, 𝑑 is the symbol 𝐶. For individual holes these are given by
probe diameter and 𝑉 is the flow velocity. A reduced (Dominy and Hodson 1993):
frequency below ~0.3 typically indicates quasi- 𝑃𝑋 − 𝑃0
steady behaviour; the case study in this paper is in 𝐶𝑋 = (4)
𝑃0 − 𝑃𝑠
this regime (see §5.3). At higher reduced frequencies
the flow becomes fundamentally unsteady. The where 𝑃0 is the flow total pressure, 𝑃𝑠 is the static
probe response will therefore depart from quasi- pressure and 𝑃𝑋 is the pressure of the hole in
steady behaviour, but the general trends are likely to question. The subscript 𝑋 indicates the index of the
be similar. hole: centre 𝐶, left 𝐿, right 𝑅, top 𝑇, bottom 𝐵,
Figure 1(a). The average side-hole pressure
Figure 1(b) and (c) respectively illustrate the
coefficient 𝐶𝑎𝑣𝑒 is expressed as:
response of a 5HP to steady flow and with
fluctuating yaw angle. The mean incidence is zero. 𝑃𝑎𝑣𝑒 − 𝑃0 (𝐶𝐿 + 𝐶𝑅 + 𝐶𝑇 + 𝐶𝐵 )
In steady flow (Figure 1(b)), the central hole 𝐶𝑎𝑣𝑒 = = (5)
𝑃0 − 𝑃𝑠 4
operates as a pitot and measures a pressure close to
Using these coefficients, the usual choice of
the flow total pressure, 𝑃𝐶 ≈ 𝑃0 . For fluctuating yaw
yaw and pitch angle coefficients are:
(Figure 1(c)), the flow is at times aligned with the
probe and again 𝑃𝐶 ≈ 𝑃0 . At other instances there is 𝑃𝐿 − 𝑃𝑅 𝐶𝐿 − 𝐶𝑅
𝐶𝑦𝑎𝑤 = = (6)
incidence onto the probe and the central hole will 𝑃𝐶 − 𝑃𝑎𝑣𝑒 𝐶𝐶 − 𝐶𝑎𝑣𝑒
measure a lower pressure, 𝑃𝐶 < 𝑃0 . Thus in the time-
average, the central hole measures a total pressure 𝑃𝑇 − 𝑃𝐵 𝐶𝑇 − 𝐶𝐵
𝐶𝑝𝑖𝑡 = = (7)
below the true time-mean, i.e. 𝑃̅𝐶 < 𝑃̅0 . This effect 𝑃𝐶 − 𝑃𝑎𝑣𝑒 𝐶𝐶 − 𝐶𝑎𝑣𝑒
can also be inferred from Bailey’s result for The total and dynamic pressure coefficients are
turbulent fluctuations, equation (2). typically taken as:
The reduction in central-hole pressure (𝑃̅𝐶 <
𝑃0 − 𝑃𝐶 −𝐶𝐶
̅
𝑃0 ) is deeply problematic for steady 5HPs because 𝐶𝑃0 = = (8)
they cannot distinguish between steady and 𝑃𝐶 − 𝑃𝑎𝑣𝑒 𝐶𝐶 − 𝐶𝑎𝑣𝑒
fluctuating flow. The fluctuating flow condition is 𝑃0 − 𝑃𝑆 1
therefore interpreted simply as having lower total 𝐶𝑑𝑦𝑛 = = (9)
𝑃𝐶 − 𝑃𝑎𝑣𝑒 𝐶𝐶 − 𝐶𝑎𝑣𝑒
pressure. This effect is evident in the current
experiments and in the data of Bauinger et al. Together these coefficients can also define a static
(2017), where a 5HP downstream of a turbine rotor pressure coefficient, here defined as:
indicated lower total pressure than Kiel-shrouded 𝑃𝐶 − 𝑃𝑆 𝐶𝐶 + 1
pitots (which are largely insensitive to angle). 𝐶𝑃𝑠 = = 𝐶𝑑𝑦𝑛 − 𝐶𝑃0 = (10)
𝑃𝐶 − 𝑃𝑎𝑣𝑒 𝐶𝐶 − 𝐶𝑎𝑣𝑒
1.3. Paper Aims and Outline Though rarely used, it is useful for the current
The key aims of the paper are (1) to explain how analysis to separate static pressure in this manner.
multi-hole probes produce erroneous readings in
2.2. Probes and Calibration Ranges
fluctuating flow, and (2) to demonstrate how to
correct for the errors. For simplicity the calibration is assumed to be
independent of Reynolds and Mach number. The
First, a theoretical framework is laid out: §2
methods can be readily extended to include these
details the quasi-steady probe response model, and
effects, which will be small for well-designed probes
§3 discusses the response of two different designs of
operating below transonic Mach numbers (<0.7).
5HP to sinusoidal flow fluctuations.
This paper uses calibration maps from two
A bluff body experiment demonstrates unsteady
different 5HP pyramid probes, designated as:
flow errors and how to post-correct the data using an
inexpensive computation. The experimental and 1) Ng 45° Probe: This probe has faces at 45 from
computational set-ups are detailed in §4 and §5 the probe axis and 2.2mm diameter. The
calibration covers ±26 in yaw and pitch. This

3
 probe was used in the bluff body experiment
and by Ng and Coull (2017).
2) Grimshaw 60° Probe: This probe has side faces
set at 60 from the probe axis and a diameter of
1.5mm. The calibration covers ±60 in yaw and
±20 in pitch (Grimshaw, S.D., 2020, private
communication).
As will be seen, the two probes exhibit similar total
pressure errors, but different static pressure errors.
2.3. Sinusoidal Fluctuations
For the analysis in §3, fluctuating flow is
specified by defining sinusoids of arbitrary
frequency. Flow angles are given by:
𝛼𝑦𝑎𝑤 (𝑡) = 𝛼̅𝑦𝑎𝑤 + 𝐴𝑦𝑎𝑤 𝑠𝑖𝑛(𝑡) (11)

𝛼𝑝𝑖𝑡 (𝑡) = 𝛼̅𝑝𝑖𝑡 + 𝐴𝑝𝑖𝑡 𝑠𝑖𝑛(𝑡 + 𝜙𝑝𝑖𝑡 ) (12)

where 𝜙𝑝𝑖𝑡 controls the relative phase. Fluctuating
pressures can be specified relative to the true time-
mean total (𝑃̅0 ) and static pressure (𝑃̅𝑠 ) of the flow:
𝑃0 (𝑡) − 𝑃̅0
𝐷𝑃0 (𝑡) = = 𝐴𝑃0 𝑠𝑖𝑛(𝑡 + 𝜙𝑃0 ) (13)
𝑃̅0 − 𝑃̅𝑠 Figure 2: Quasi-Steady model.
𝑃𝑠 (𝑡) − 𝑃̅𝑠 These probe-indicated values are denoted using
𝐷𝑃𝑠 (𝑡) = = 𝐴𝑃𝑠 𝑠𝑖𝑛(𝑡 + 𝜙𝑃𝑠 ) (14) over-hats, and may differ from the true time-average
𝑃̅0 − 𝑃̅𝑠
̅
of the instantaneous coefficients, 𝐶𝑦𝑎𝑤 ̅ .
and 𝐶𝑝𝑖𝑡
The symbol 𝐷 denotes a dynamic pressure
Indicated angles are calculated by interpolating
normalized by the time-mean flow.
from the calibration map in typical fashion:
2.4. Quasi-Steady Probe Response
𝛼̂𝑦𝑎𝑤 , 𝛼̂𝑝𝑖𝑡 = 𝑓(𝐶̂𝑦𝑎𝑤 , 𝐶̂𝑝𝑖𝑡 ) (19)
As shown in Figure 2, the instantaneous
response of each hole is estimated by interpolating The errors from the true time mean flow angles are:
from the calibration data. From equation (4): 𝐸𝑦𝑎𝑤 = 𝛼̂𝑦𝑎𝑤 − 𝛼̅𝑦𝑎𝑤 (20)
𝑃𝑋 (𝑡) = 𝑃0 (𝑡) + (𝑃0 (𝑡) − 𝑃𝑠 (𝑡))𝐶𝑋 (𝑡) (15)
𝐸𝑝𝑖𝑡 = 𝛼̂𝑝𝑖𝑡 − 𝛼̅𝑝𝑖𝑡 (21)
where 𝐶𝑋 (𝑡) = 𝑓(𝛼𝑦𝑎𝑤 (𝑡), 𝛼𝑝𝑖𝑡 (𝑡)) interpolates the
hole calibration coefficient for the instantaneous The indicated total, static and dynamic pressure
angles. Normalized by the time-mean flow, the coefficients are interpolated from the calibration
dynamic pressure coefficients 𝐷𝑋 (𝑡) are thus: using the indicated angles:

𝑃𝑋 − 𝑃̅0 𝐶̂𝑃0 , 𝐶̂𝑃𝑠 , 𝐶̂𝑑𝑦𝑛 = 𝑓(𝛼̂𝑦𝑎𝑤 , 𝛼̂𝑝𝑖𝑡 ) (22)

𝐷𝑋 = = 𝐷𝑃0 + 𝐶𝑋 (𝐷𝑃0 − 𝐷𝑃𝑠 + 1) (16)
𝑃̅0 − 𝑃̅𝑠 Again, these values will generally differ from the
true time averages. Non-dimensional errors in the
2.5. Pneumatically-Averaged Response indicated total, dynamic and static pressure, are then
A precise model of pneumatic averaging would defined relative to the true time-average flow:
need to consider the dynamics of the pressure tubes, 𝑃̂0 − 𝑃̅0
which may affect the average readings. However 𝐸𝑃0 = ̅𝐶 + 𝐶̂𝑃0 (𝐷
=𝐷 ̅𝐶 − 𝐷
̅𝑎𝑣𝑒 ) (23)
𝑃̅0 − 𝑃̅𝑠
such effects are likely to be small, and highly case-
dependent. Thus for simplicity, pneumatic averaging 𝑃̂𝑠 − 𝑃̅𝑠
is modelled by time-averaging each hole pressure, 𝐸𝑃𝑠 = ̅𝐶 − 𝐶̂𝑃𝑠 (𝐷
= 1+𝐷 ̅𝐶 − 𝐷
̅𝑎𝑣𝑒 ) (24)
𝑃̅0 − 𝑃̅𝑠
giving the following pitch and yaw coefficients:
̅𝐿 − 𝐷
𝐷 ̅𝑅 𝑃̂0 − 𝑃̂𝑠
𝐶̂𝑦𝑎𝑤 = (17) 𝐸𝑑𝑦𝑛 = − 1 = 𝐶̂𝑑𝑦𝑛 (𝐷
̅𝐶 − 𝐷
̅𝑎𝑣𝑒 ) − 1 (25)
̅𝐶 − 𝐷
𝐷 ̅𝑎𝑣𝑒 𝑃̅0 − 𝑃̅𝑠
Each error is positive when the quantity is over-
̅𝑇 − 𝐷
𝐷 ̅𝐵
𝐶̂𝑝𝑖𝑡 = (18) estimated by the probe and vice-versa. For steady
̅ ̅
𝐷𝐶 − 𝐷𝑎𝑣𝑒 flow all errors are zero by definition.

4
3. SINUSOIDAL FLOW FLUCTUATIONS In the absence of knowledge of fluctuations, the
5HP operator must use the indicated flow angles
We now examine the response of the quasi-
(𝛼̂𝑦𝑎𝑤 ≈ 0, 𝛼̂𝑝𝑖𝑡 ≈ 0) to interpolate for 𝐶𝑃0 from the
steady model to sinusoidal fluctuations, the most
steady calibration data. This gives an indicated total
important of which are angle fluctuations.
pressure coefficient 𝐶̂𝑃0 = 0.01, which suggests that
3.1. Angle Fluctuations in One Component the flow total pressure should be very close to the
For the Grimshaw probe, Figure 3 shows the central hole pressure. Without the correct factor
response to a sinusoidal fluctuation of ±25 in yaw ̅ ≈ 0.2), the probe thus indicates a lower total
(𝐶𝑃0
angle, for zero mean incidence. The instantaneous pressure than the true value:
points simply follow the calibration line for 𝛼𝑝𝑖𝑡 = 𝑃̂0 = 𝑃̅𝑐 + (𝑃̅𝑐 − 𝑃̅𝑎𝑣𝑒 )𝐶̂𝑃0
0. Because the calibration is approximately anti-
symmetric, positive and negative incidence effects ≈ 𝑃̅0 − 0.19(𝑃̅𝑐 − 𝑃̅𝑎𝑣𝑒 ) (27)
tend to cancel each other. Thus the pneumatic The calibration map for static pressure also has
average Yaw Coefficient 𝐶̂𝑦𝑎𝑤 is close to zero, and a generally symmetric response to angle, but it is
the probe indicates a yaw angle 𝛼̂𝑦𝑎𝑤 ≈ 0. more sensitive to the probe design and manufacture.
Figure 4 shows the total pressure response for Figure 5 shows the static pressure response for the
the same fluctuation. In contrast to yaw angle, 𝐶𝑃0 Grimshaw probe. Because of its high face angle
vs. 𝛼𝑦𝑎𝑤 is symmetric. Both positive and negative (60), the static pressure coefficient is relatively
yaw cause the central hole pressure to drop (higher insensitive to angles over this range. Thus the
𝐶𝑃0 ); this is the process illustrated schematically in underestimation of 𝑃𝑠 in Figure 5 is only around
Figure 1(c). Due to this effect, the true time-average 0.04(𝑃𝐶 − 𝑃𝑎𝑣𝑒 ). Figure 6 shows the response of the
̅ ≈ 0.2.
total pressure coefficient (equation (8)) is 𝐶𝑃0 Ng probe, which has face angles of 45 and is more
This value is the correction that should be applied to sensitive to angle. As shown, this probe
the central hole pressure to give the correct flow underestimates static pressure by 0.37(𝑃𝐶 − 𝑃𝑎𝑣𝑒 ).
total pressure, i.e.
3.2. Angle Fluctuations in Two Components
̅
𝑃̅0 = 𝑃̅𝑐 + (𝑃̅𝑐 − 𝑃̅𝑎𝑣𝑒 )𝐶𝑃0 (26) We now consider simultaneous fluctuations in
both angles components. Figure 7 shows total

Figure 3: Yaw Angle response to ±25 Figure 5: Static Pressure response to ±25
fluctuating Yaw, 60° Grimshaw Probe. fluctuating Yaw, 60° Grimshaw Probe.

Figure 4: Total Pressure response to ±25 Figure 6: Static Pressure response to ±25
fluctuating Yaw, 60° Grimshaw Probe. fluctuating Yaw, 45° Ng Probe.

5
 greater than for total pressure, reflecting the poorer
conditioning of the static pressure measurement
(Dominy and Hodson (1993)). The static pressure
coefficient is also less directionally symmetric
(Figure 5, Figure 6) than the total pressure
coefficient.
The static errors for 45° Ng probe (Figure 8(b))
are approximately equal to the total errors:
2.3
𝐸𝑃𝑠 ≈ 𝐸𝑃0 ≈ −1.6 × 10−4 (𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 ) (30)
As a result of these similar errors, the Ng probe has
generally low dynamic pressure error.
With its sharper face angles, the 60° Grimshaw
probe (Figure 8(a)) has lower static errors which can
Figure 7: Total Pressure Errors for be approximated by:
fluctuating yaw and pitch. 2.5
𝐸𝑃𝑠 ≈ −0.2 × 10−4 (𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 ) (31)
As indicated in Figure 8, this probe therefore suffers
from relatively large dynamic pressure errors.
The universality of these static error
correlations has not been tested, but one may expect
them to provide useful initial estimates for similar
probes. For greater accuracy, the quasi-steady model
can be readily applied to any given calibration map.
3.3. Other Factors Affecting Errors
The flow angle effects shown above are the
dominant source of pneumatic-averaging errors.
Nonetheless it is worth noting other factors that can
influence the errors:
• Reducing the time-averaged incidence, as in
Figure 8: Static Pressure Errors for “nulled” 5HP operation, tends to slightly reduce
fluctuating yaw and pitch, legend in Figure 7. errors but the benefits are insignificant.
pressure errors for a range of 𝐴𝑦𝑎𝑤 , 𝐴𝑝𝑖𝑡 and phase • The addition of pressure fluctuations to angle
𝜙𝑝𝑖𝑡 for the two probes, plotted against the root- fluctuations can increase or decrease error
mean-square (RMS) of the compound angle: slightly, depending on the phase. These effects
are driven by the covariance of angle and
2 2 pressure, which adds further biasing to the
𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 = √(𝛼𝑦𝑎𝑤−𝑅𝑀𝑆 ) + (𝛼𝑝𝑖𝑡−𝑅𝑀𝑆 ) (28) pneumatic averaging. While unlikely in most
The effect of probe design and phase is minimal, and cases, the quasi-steady model can be used to
the errors collapse to the following correlation: analyse these covariance effects.
2.3 3.4. Summary
𝐸𝑃0 ≈ −1.6 × 10−4 (𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 ) (29)
The calibration maps for total and static
The collapse for both probes reflects the pitot-like pressure coefficient are approximately symmetric.
configuration of the 5HP central hole and the The probe therefore effectively responds to the
similarity in pitch and yaw response. Figure 7(a) magnitude of a fluctuation in flow angle, |𝛼′|. When
includes the Bailey et al. (2013) pitot correction probe incidence fluctuates, pneumatic averaging
from equation (2), applied to the central hole. The therefore causes a bias which cannot be accounted
Bailey equation is accurate for small incidence for by the steady calibration. Having demonstrated
(𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 < 15°) but underestimates errors for how these pneumatic-averaging errors arise, we will
larger fluctuations. examine these effects in a canonical bluff body flow.
The static pressure errors are presented in
Figure 8; note the smaller scale compared to Figure 4. CASE STUDY: EXPERIMENTAL SET-UP
7. As observed above the two probes respond
The case study is a D-shaped bluff body in an
differently, but again the errors approximately
enclosed channel, as shown in Figure 13. Air is
collapse to 𝐸𝑃𝑠 ≈ 𝑓(𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 ). The scatter is

6
driven through the test section by an open-loop wind traverses were obtained along the midspan plane at
tunnel. The body has a semi-circular leading edge, a a streamwise distance of 323mm (≈ 4𝑊) behind the
width 𝑊 of 80mm, chord 187.5mm and span of trailing edge, covering a vertical height of 0.34m
375mm. The body is mounted at zero incidence in (= 4.25𝑊) which fully captured the wake.
the middle of a long channel of height (𝐻) of The 5HP measurements were complemented by
650mm, giving a blockage factor of 12%. The traversing a Kiel probe (Kiel-shrouded-pitot) of
Reynolds number was 98,500 based on the body 4mm outside diameter at the same location. This
width 𝑊 and upstream velocity 𝑉1 . The flow was Kiel probe is insensitive to flow angle over a range
incompressible (Mach < 0.15), and the inlet of at least ±30° and therefore does not suffer from
turbulence level was 0.5%. significant biasing errors as flow angles fluctuate.
Pressure measurements were obtained using a Figure 16 shows the differences between these
Scanivalve DSA 3217 with a 10” H20 range. The measurements, which are discussed in §6.1.
inlet total pressure 𝑃01 was measured using wall An array of 11 static tappings were situated on
mounted pitots in the upstream wind tunnel, prior to the side wall at the traverse plane, clustered towards
a ~4:1 flow contraction into the constant-area test the centreline and covering the full traverse height.
section. Inlet static pressure 𝑃1 was measured on the Area traverses with the 5HP and Kiel probe
wall 600mm (7.5𝑊) upstream of the leading edge. confirmed that the flow is approximately two-
4.1. Wake Traverses dimensional over the central 60-70% of span.
Linear wake traverses were obtained with the 4.2. Surface Static Pressure and Form Drag
Ng 5HP probe described in section 2.2. This probe Forty-six surface static tappings were
has a conical head with 45° face angles and a distributed around the D-shaped body as shown in
diameter of 2.5mm (0.035𝑊). The probe was Figure 10. The measured static pressure coefficients
calibrated in a separate facility for ±26° yaw and are shown in Figure 11:
pitch, at a similar velocity as in the experiments.
𝑃01 − 𝑃
Using an automated stepper-motor controller, 𝐶𝑝 = (32)
𝑃01 − 𝑃1
As can be seen, the front face pressures largely
follow the potential flow theory distribution for a
cylinder, while the trailing face has near-constant
pressure. The form drag coefficient can be
Figure 9: D-Shaped bluff body experiment. calculated by integration:
𝐶𝑝 𝑑𝑦
𝐶𝐷−𝑓𝑜𝑟𝑚 = ∮ (33)
𝑊
For this flow the form drag will dominate over
viscous shear and 𝐶𝐷 ≈ 𝐶𝐷−𝑓𝑜𝑟𝑚 . Numerical
integration using cubic up-sampling between
Figure 10: Surface Static Tapping Locations. measurement points gives a drag coefficient of 0.74.
4.3. Drag Coefficient from Wake Traverses
The drag coefficient is also calculated from the
wake traverse after the method of Jones (1936). A
virtual downstream plane is considered which has
the same static pressure as the inlet, 𝑃1 . The drag
coefficient is then given by:
𝐹𝐷 𝜃′
𝐶𝐷 = =2 (34)
1 2 𝑊
𝜌𝑉 𝑊
2 1
The wake momentum thickness 𝜃′ at the virtual
plane is calculated by assuming two-dimensional
flow from the traverse plane, applying continuity
and assuming isentropic flow:

𝑃02 − 𝑃2 𝑃02 − 𝑃1
𝜃′ = ∫ √ (1 − √ ) 𝑑𝑦 (35)
Figure 11: Surface Static Pressure 𝑃01 − 𝑃1 𝑃01 − 𝑃1
Coefficient.

7
The value of 𝐶𝐷 is sensitive to the assumptions coefficient in Figure 11 shows close agreement with
made, but alternative calculation methods show the measurements and gives a similar integrated drag
same trends observed in the current analysis. As coefficient, 𝐶𝐷−𝑓𝑜𝑟𝑚 = 0.77.
discussed in §6.3, the 5HP indicates a drag
5.3. Unsteady Flow Angles
coefficient of 1.18, 60% higher than 𝐶𝐷−𝑓𝑜𝑟𝑚 .
The reduced frequency of the 5HP probe
5. UNSTEADY COMPUTATIONS (equation (3)) can be related to the flow Strouhal
number (𝑆𝑡 = 𝑓𝑊/𝑉1 ):
5.1. Purpose
𝑓𝑑 𝑉1 𝑑
A key aim of this paper is to demonstrate how 𝑓𝑟 = = 𝑆𝑡 (36)
𝑉 𝑉𝑊
to post-correct 5HP data. To achieve this, we need
an estimate of the unsteady incidence angles onto where 𝑉 is the velocity at the probe head. The
the probe in question. velocity ratio 𝑉1 /𝑉 varies between about 0.9 and 1.4
through the wake traverse, and the probe diameter is
For some set-ups, unsteady flow angles may be
small relative to the trailing edge, 𝑑/𝑊 = 0.035.
measured directly, e.g. using cross-wires or laser
For the primary shedding frequency (𝑆𝑡 ≈ 0.2), the
techniques. However, these measurements will not
reduced frequency is therefore 𝑓𝑟 ≈ 0.01. Thus the
always be possible, and cannot be performed for
probe will respond quasi-steadily.
historical set-ups. Computational Fluid Dynamics
(CFD) simulations will therefore be the only viable Figure 14 shows time traces of angles at the
means to estimate flow angle fluctuations in many centreline of the traverse plane for around 12
cases. Rather than rely on high-order methods such shedding cycles. Yaw oscillations (within the plane
as Large Eddy Simulation, we here use a low-cost of Figure 13) of over ±35 are observed, with
Unsteady Reynolds-Averaged Navier-Stokes smaller variation in pitch. Figure 15(a) shows the
(URANS) calculation. Such calculations are used distribution of angle RMS through the wake. The
routinely across academia and industry, thus RMS of the compound angle is 27° at the wake
minimising the effort required to post-correct data. centreline, equivalent to a sinusoidal fluctuation of
amplitude ±40. Figure 15(b) shows the expected
The quasi-steady model can also be used to
total pressure error 𝐸𝑃0 from equation (29), which
perform “virtual” probe traverses in CFD. By
peaks at 𝐸𝑃0 = −0.32, i.e. 32% of the local dynamic
mimicking pneumatic-averaging effects, this
head at the wake centreline.
approach allows for a more rigorous like-for-like
comparison of simulations and experiments.
5.2. Computational Set-Up
The URANS calculation was performed with
Fluent v18.2 using a pressure-based incompressible
solver. Figure 12 shows the multi-block structured
mesh, which has 5.8 million cells and uses wall
functions to resolve the boundary layers. The full
span was simulated over an axial distance of 6.5
chords (~15.2𝑊). Uniform values of inlet
stagnation and turbulence quantities were specified,
and an average static pressure at the exit. The side-
walls where the body was mounted are treated as Figure 12: URANS Computational grid.
viscous walls and the top and bottom walls were set
to be periodic, which had minimal impact on the
results. The inlet reference pressure 𝑃1 was
calculated as the area- and time-average at the inlet
of the midspan plane. Around 100 timesteps were
used for each fundamental shedding cycle, and data
were collected for ~25 cycles after transients.
The Scale-Adaptive-Simulation turbulence
model of Menter and Egorov (2010) was used,
which has been developed specifically to help
capture the multiple scales of vortex interaction and
shedding flows. The presence of multiple vortex
Figure 13: Instantaneous Turbulent Kinetic
scales is evident in Figure 13, which shows a
Energy (m2/s2) in the midspan plane, URANS.
snapshot of turbulent kinetic energy from the
Dashed line indicates traverse plane.
URANS calculation. The surface static pressure

8
 𝑃01 − 𝑃
𝐶𝑝𝑠 = (38)
𝑃01 − 𝑃𝑓𝑠
Compared to the sidewall statics, the 5HP indicates
that the pressure at midspan is lower by
~0.30(𝑃01 − 𝑃𝑓𝑠 ).
Figure 16 also includes the time-average of the
URANS calculations. Total pressure (Figure 16(a))
is close to the Kiel measurements but the wake
appears slightly under-mixed, with higher loss at the
centreline. Sidewall static pressure in Figure 16(b) is
close to the measurements, especially in the
Figure 14: Instantaneous angles at the centreline of the wake. URANS predicts that the
traverse centreline, URANS. midspan pressure is only ~0.10(𝑃01 − 𝑃𝑓𝑠 ) below
the sidewall value, i.e. significantly higher pressure
than indicated by the 5HP.

6.2. Quasi-Steady Modelling of 5HP Response

To mimic the 5HP response, the time-series of
flow angles and pressure from the URANS
calculation can be directly inputted into the quasi-
steady model outlined in §2. Mimicking the probe
response in this way gives a like-for-like comparison
between the experiments and CFD.
The instantaneous angles exceed the calibration
range (±26) and the map was therefore extended by
linearly extrapolating individual hole coefficients
(equation (4)). This extrapolation introduces only
small uncertainty since similar results were obtained
by clipping the flow angles to ±26. The modelled
response of the 5HP is included in Figure 16 (“5HP
model”), and closely mimics the 5HP-indicated loss
and static pressure across the wake.
Figure 15: URANS angle fluctuations and Figure 16 also includes predictions of probe
predicted total pressure error (equation (29)). response using only the angle fluctuations i.e.
ignoring total and static pressure fluctuations. These
6. ERRORS AND CORRECTION
We now present the measurements and
demonstrate how to correct for the errors.
6.1. Measured Flow Properties in the Wake
Figure 16(a) shows traverse measurements of
total pressure loss from the 5HP and the Kiel probe:
𝑃01 − 𝑃0
𝑌𝑃 = (37)
𝑃01 − 𝑃𝑓𝑠
where 𝑃𝑓𝑠 is the static pressure in the freestream
outside the wake (|𝑦/𝑊| > 2). The Kiel is largely
insensitive to angle so has minimal fluctuating angle
error. In contrast, the 5HP indicates losses which are
0.22 (+44%) higher at the wake centreline. This
error is close to the predictions shown in Figure 15.
The local dynamic pressure at the wake centreline is
around 0.62(𝑃01 − 𝑃𝑓𝑠 ), so 𝐸𝑃0 = −0.32
corresponds to a overestimation of 0.20 in 𝑌𝑝 .
Figure 16(b) presents static pressure coefficient, Figure 16: Measured and Predicted Total and
𝐶𝑝𝑠 , defined in a similar manner: Static Pressure in the wake.

9
 Figure 17: (a) total and (b) static pressure
correlation with yaw angle, wake centreline.
lines are almost indistinguishable from the full
model, demonstrating that it is the angle fluctuations
that drive the error. Figure 17 shows that pressure
fluctuations are not strongly correlated with the yaw
angle fluctuations, which is why they have little
influence. Pressure fluctuations are therefore Figure 18: Correction of 5HP data: URANS
ignored in the following corrections. flow angles with quasi-steady model (“direct
calc”) and correlations (eqs. (29), (30)).
6.3. Error Correction using URANS flow angles
As Figure 18 shows, this method gives a similar
The URANS flow angles can be used to correct
correction to the direct calculation but is slightly less
the 5HP measurements either by (1) direct
accurate. The correlation approach has the
calculation, inputting the time-series data into the
advantage of being very simple to apply and does
quasi-steady model with a known calibration map,
not require the probe calibration map.
or (2) applying the correlations given in §3.2.
Table 1 shows calculated drag coefficients
Using the direct calculation method, the model
based on upstream conditions. As noted in Figure 11,
outputs the difference between the true time-
̅ , and those indicated integration of surface statics gives a form drag of
averaged coefficients, e.g. 𝐶𝑃0
𝐶𝐷−𝑓𝑜𝑟𝑚 = 0.74. For reference, URANS gives
by the probe, e.g. 𝐶̂𝑃0 . Applying these differences to
similar coefficients for both surface integration
the indicated coefficients allows a direct correction
(𝐶𝐷−𝑓𝑜𝑟𝑚 = 0.77) and from the wake traverse
of the data, e.g. for total pressure:
(𝐶𝐷 = 0.74). Combining the Kiel and sidewall
𝐶𝑃0−𝑐𝑜𝑟 = 𝐶̂𝑃0 + [𝐶𝑃0
̅ − 𝐶̂𝑃0 ] (39) statics data underestimates drag slightly (𝐶𝐷 =
𝑚𝑜𝑑𝑒𝑙
0.68) because the midspan static pressure is lower
𝑃̅0−𝑐𝑜𝑟 = 𝑃̅𝑐 + (𝑃̅𝑐 − 𝑃̅𝑎𝑣𝑒 )𝐶𝑃0−𝑐𝑜𝑟 (40) than the sidewall (Figure 16(b)). The uncorrected
Figure 18 shows that this correction (“5HP direct 5HP indicates a drag coefficient of 1.18, around
calc.”) brings the 5HP total pressure in line with the 60% higher than the form drag from static
Kiel (dashed blue line), and the static pressure is integration. The direct calculation correction brings
now in good agreement with URANS. Overall, the the value within 3% of the calculated form drag, and
errors have been reduced by about an order of the correlation correction is within 7%.
magnitude.
7. GENERAL IMPLICATIONS
The second approach uses the correlations
based on the RMS of the compound angle given in This section considers the implications of the
§3.2. For this 45° angle 5HP, the total and static results. Pneumatic-averaging errors are likely to
errors are of similar magnitude: affect a wide range of data collected over many
2.3
decades. We therefore first consider some common
𝐸𝑃𝑠 ≈ 𝐸𝑃0 ≈ −1.61 × 10−4 (𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 ) (41) set-ups where angle fluctuations may be significant.
From equations (23) and (24), the corrected total and We then consider practical steps to identify and post-
static pressure are then given by: correct potential errors in data, before discussing
how to minimize or design-out angle fluctuation
𝑃̂0 (1 − 𝐸𝑃𝑠 ) + 𝑃̂𝑠 𝐸𝑃0 errors in the first place.
𝑃̅0 = (42)
(1 − 𝐸𝑃𝑆 + 𝐸𝑃0 )
7.1. Examples of fluctuating flow angles
𝑃̅𝑠 = 𝑃̂𝑠 (1 + 𝐸𝑃0 ) − 𝑃̂0 𝐸𝑃𝑠 (43) We have demonstrated that fluctuating angles
can cause large 5HP errors in shedding wakes.

10
 5HP angle fluctuations. Several observations can be

Kiel + Sidewall
made:

Surface Statics
• Kiel probes should be used to cross-check 5HP

URANS

Correlation
Correction
Indicated

𝐸𝑃0 , 𝐸𝑃𝑠
data when fluctuations may be present.

Direct
• Turbomachinery traverses downstream of
rotors should be performed in the rotating frame
where possible to minimise fluctuations.
𝐶𝐷−𝑓𝑜𝑟𝑚 0.77 0.74 - - - - Alternatively, stationary-frame traverses should
𝐶𝐷 0.74 - 0.68 1.18 0.72 0.69 be performed as far downstream as possible.
Table 1 : Drag Coefficients calculated with • The errors can be avoided by use of fast-
equations (33) and (34). response multi-hole probes. More work is
Similar errors would be expected for any multi-hole needed to further miniaturise such probes and
probe or unshrouded pitot. improve their accuracy.
Turbomachinery measurements are particularly • Errors could be minimised with new multi-hole
prone to such errors because of blade-passing probes with low angle sensitivity for total and
fluctuations. Typically, efficiency is calculated from static pressure. For example, a hybrid 5HP with
shaft torque and stationary-frame 5HP area- a Kiel-pitot central hole would be much less
traverses, e.g. Evans and Longley (2017), Bauinger prone to errors. Challenges remain to develop,
et al. (2017). These 5HP measurements will be characterise and miniaturise such probes.
corrupted by periodic unsteadiness from upstream
rotating blade rows. There is direct evidence of such 8. CONCLUSIONS
errors in the comparison of Kiel and 5HP data in Quasi-steady arguments demonstrate how
Bauinger et al. (2017). Coull et al. (2020) estimate fluctuating incidence leads to pneumatic-averaging
that the angle fluctuations downstream of a turbine errors. Five-hole-probe calibration coefficients for
can lead to under-prediction of efficiency by up to total and static pressure are approximately
1.5%; such errors are sufficiently large to warrant symmetric with angle. When flow angles fluctuate,
the post-correction of a wide body of data. there is therefore a similar response to both positive
Significant errors may also be expected in flows and negative incidence. The resultant bias in the
where unsteady shear layers generate flow angle time-average cannot be accounted for by the steady
fluctuations. Examples include stalled aerofoils or calibration, leading to significant errors. Similar
diffusers, shedding trailing edges, and three- errors will occur for other multi-hole pneumatic
dimensional unsteady flows such as tip leakage, probes and un-Kieled pitots.
corner separation and stall in turbomachinery flows. The errors caused by fluctuating flow angles are
7.2. Identifying and Assessing Error Magnitude demonstrated in a shedding bluff-body wake.
Compared to a Kiel probe, a five-hole-probe over-
Yasa and Paniagua (2012) proposed an error- estimates total pressure loss by up to 44% and under-
detection method based on the departure of data estimates static pressure, leading to a ~60% over-
points from the individual hole calibration maps. estimation of drag coefficient.
However, this approach is not reliable for angle
A post-correction method is demonstrated using
fluctuations because some erroneous measurements
an inexpensive URANS calculation to estimate the
still lie close to the steady calibration data surfaces.
flow angle fluctuations. A simple correction is
An independent estimate of angle fluctuations is achieved using correlations with the root-mean-
therefore required, either from measurements (e.g. squared angle. A more accurate correction is
multi-element hot-wires), analysis or computations. obtained using a quasi-steady model with the probe
As discussed in §6.3, a simple correction can be calibration map. Both methods reduce errors by
made using the correlations based on compound around an order-of-magnitude.
RMS angle, 𝛼𝑐𝑜𝑚𝑝−𝑅𝑀𝑆 . More accurate results for a
A wide body of data is likely to have been
known probe calibration can be obtained by using
corrupted by fluctuating angle errors, in particular
the quasi-steady model directly. The precision of
turbomachinery performance measurements. The
these corrections will depend on the accuracy of the
approaches in the paper can be used to post-correct
estimated angle fluctuations and having sufficient
this data. Experiments should also be planned to
angle range on the calibration.
mitigate these errors in the first place, for example
7.3. Designing-Out Error by corroborating multi-hole probe data with Kiel
Finally, it is useful to consider how we can probes.
design experiments to minimise the errors from

11
ACKNOWLEDGEMENTS Memoranda 3527. Aeronautical Research
Council.
The authors would like to thank Rolls-Royce for
funding and permission to publish, with particular Christiansen, T. & Bradshaw, P. 1981 Effect of
thanks to Dr Raul Vazquez and Dr Duncan Simpson. turbulence on pressure probes. Journal of Physics
Prof Luca di Mare, University of Oxford, provided E: Scientific Instruments 14 (8), 992–997
advice on the URANS calculation set-up. Dr Sam Coull, J., Dickens, T., Ng, H. and Serna, J., 2020.
Grimshaw of the Whittle Laboratory, University of Five hole probe errors caused by fluctuating
Cambridge, provided a sample calibration map. incidence. XXV Symposium on Measuring
Techniques in Turbomachinery, Santorini, Greece,
DECLARATIONS Sept 2020
Dominy, R.G. and Hodson, H.P., 1993. An
Ethical Approval
investigation of factors influencing the calibration
Ethical approval is not applicable to this study. of five-hole probes for three-dimensional flow
Competing Interests measurements. Journal of Turbomachinery,
115(3), pp.513-519.
The authors have no conflict of interest.
Evans, K.R. and Longley, J.P., 2017. Clocking in
Author Contributions low-pressure turbines. Journal of
All authors contributed to the conception of this Turbomachinery, 139(10), p.101003.
study. The quasi-steady modelling was developed by Goldstein, S., 1936. A note on the measurement of
Dr Coull in collaboration with Dr Dickens. The total head and static pressure in a turbulent stream.
experiment was designed and commissioned by Dr Proceedings of the Royal Society of London.
Ng; and experiments were conducted by Dr Serna Series A-Mathematical and Physical Sciences,
and Dr Coull. Dr Coull also performed the CFD 155(886), pp.570-575.
simulations. The material in the paper was written Grimshaw, S.D. and Taylor, J.V., 2016, June. Fast
by Dr Coull with input from the other authors. settling millimetre-scale five-hole probes. ASME
Funding Turbo Expo: Power for Land, Sea, and Air
Funding was received from Rolls-Royce plc. Jones, B.M., 1936. Measurement of profile drag by
the pitot-traverse method. ARC R&M no. 1688
Availability of Data and Materials
Melzer, A.P. and Pullan, G., 2019. The role of vortex
Data and material are available on request to the shedding in the trailing edge loss of transonic
corresponding author. turbine blades. Journal of Turbomachinery,
141(4), p.041001.
REFERENCES Menter, F.R. and Egorov, Y., 2010. The scale-
Bailey, S.C.C., Hultmark, M., Monty, J.P., adaptive simulation method for unsteady turbulent
Alfredsson, P.H., Chong, M.S., Duncan, R.D., flow predictions. Part 1: theory and model
Fransson, J.H.M., Hutchins, N., Marusic, I., description. Flow, Turbulence and Combustion,
McKeon, B.J., Nagib, H.M., Örlü, R., Segalini, A., 85(1), pp.113-138.
Smits, A.J., and Vinuesa, R., 2013, “Obtaining Ng, H.C.H. and Coull, J.D., 2017. Parasitic loss due
Accurate Mean Velocity Measurements in High to leading edge instrumentation on a low-pressure
Reynolds Number Turbulent Boundary Layers turbine blade. Journal of Turbomachinery, 139(4)
Using Pitot Tubes,” J. Fluid Mech., 715, pp. 642– Prahst, P.S., Kulkarni, S. and Sohn, K.H., 2015,
670 June. Experimental results of the first two stages
Bauinger, S., Marn, A., Göttlich, E. and Heitmeir, F., of an advanced transonic core compressor under
2017. Influence of pressure fluctuations on the isolated and multi-stage conditions. Turbo Expo:
mean value of different pneumatic probes. Power for Land, Sea, and Air 2015
International Journal of Turbomachinery, Yasa, T. and Paniagua, G., 2012. Robust procedure
Propulsion and Power, 2(3), p.13. for multi-hole probe data processing. Flow
Bradshaw, P. & Goodman, D.G. 1968 The effect of Measurement and Instrumentation, 26, pp.46-54.
turbulence on static-pressure tubes. Reports and

12