Validation of steady RANS modelling of isothermal plane

turbulent impinging jets at moderate Reynolds numbers

Citation for published version (APA):
Khayrullina, A., van Hooff, T., Blocken, B., & van Heijst, G. J. (2019). Validation of steady RANS modelling of
isothermal plane turbulent impinging jets at moderate Reynolds numbers. European Journal of Mechanics. B,
Fluids, 75, 228-243. https://doi.org/10.1016/j.euromechflu.2018.10.003

Document license:
TAVERNE

DOI:
10.1016/j.euromechflu.2018.10.003

Document status and date:

Published: 01/05/2019

Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can be
important differences between the submitted version and the official published version of record. People
interested in the research are advised to contact the author for the final version of the publication, or visit the
DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page
numbers.
Link to publication

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please
follow below link for the End User Agreement:
www.tue.nl/taverne

Take down policy

If you believe that this document breaches copyright please contact us at:
[email protected]
providing details and we will investigate your claim.

Download date: 25. Jul. 2025

 European Journal of Mechanics / B Fluids 75 (2019) 228–243

Contents lists available at ScienceDirect

European Journal of Mechanics / B Fluids

journal homepage: www.elsevier.com/locate/ejmflu

Validation of steady RANS modelling of isothermal plane turbulent

impinging jets at moderate Reynolds numbers
∗
Adelya Khayrullina a , , Twan van Hooff a,b , Bert Blocken a,b , GertJan van Heijst c
a
Department of the Built Environment, Eindhoven University of Technology, Eindhoven, The Netherlands
b
Department of Civil Engineering, KU Leuven, Belgium
c
Department of Applied Physics, Eindhoven University of Technology, Eindhoven, The Netherlands

highlights

• Detailed validation study of RANS turbulence models for impinging jets is performed.
• Reduced-scale particle image velocimetry measurements are used for comparison.
• Best overall performance for velocities is provided by realisable k − ε model.
• Best predictions of potential core length are shown by realisable and RNG k − ε model.
• The standard k − ε model provides the most accurate estimation of jet decay rate.

article info a b s t r a c t

Article history: The numerical modelling of impinging jet flows is not straightforward as it should not only solve the
Received 29 March 2018 shear layer development in the free jet region, but also the near-wall behaviour (streamline curvature)
Received in revised form 9 August 2018 and the resulting wall jets after impingement. This study presents a validation study of steady Reynolds-
Accepted 1 October 2018
averaged Navier–Stokes turbulence models for predicting isothermal plane turbulent impinging jets at
Available online 15 October 2018
two different slot Reynolds numbers, i.e. Re = 8,000 (case I) and Re = 13,000 (case II), based on 2D particle
Keywords: image velocimetry measurements. In addition, an in-depth analysis of the results provided by the five
Plane turbulent jet different turbulence models: standard k − ε (SKE), realisable k − ε (RKE), RNG k − ε , SST k − ω, and a
Turbulence model validation Reynolds stress model (RSM), is performed. The results show that: (1) for both Reynolds numbers the
RANS best agreement with measured velocities and turbulent kinetic energy in the region near the jet nozzle
CFD
is achieved with SST; (2) the best predictions of potential core length are provided by RNG (case I) and
Computational fluid dynamics
RKE (case II); (3) centreline distributions of velocities and turbulent kinetic energy are most accurately
predicted by RNG and RKE for case I, while for case II the best agreement with experimental data is
obtained by SKE and RNG; (4) the best overall performance for both cases in predicting velocities is
provided by RKE, and by RKE and RNG when considering turbulent kinetic energy; (5) all models more
accurately predict the jet spreading rate in the intermediate region than in the potential core region; (6)
for both Reynolds numbers SKE provides the most accurate estimation of jet decay rate.
© 2018 Elsevier Masson SAS. All rights reserved.

1. Introduction of the following three distinctive flow regions (e.g. [8–10]): (1)
the potential core region with a centreline velocity more than or
Air curtains are used in many practical applications to separate equal to 98% of the inlet jet velocity (length of the potential core
two environments in terms of heat and mass transfer, for example, region (hp ) is generally 3wjet ≤ hp ≤ 8wjet , with wjet the width
at entrances of buildings to reduce heat losses (e.g. [1–3]), in clean- of the nozzle from which the jet is issued), (2) the intermediate
rooms and hospitals to prevent pollutant spreading (e.g. [4,5]), region with decaying centreline velocity, including the transition
inside buildings (e.g. [6]) and tunnels (e.g. [7]) to prevent smoke region from the potential core to the developed region, with self-
similar profiles for mean velocity and turbulence intensity, (3) the
propagation between two or more zones.
impingement region, where the flow is influenced by the pressure
An air curtain can be represented by a plane turbulent imping-
gradient created by the opposing impingement plate (height of the
ing jet (PTIJ) as schematically shown in Fig. 1. The jet consists
impingement region hi ≈ (0.10–0.13)hjet , with hjet the distance
from the nozzle to the floor). The length of the potential core
∗ Corresponding author. (hp ) is influenced by the turbulence intensity at the jet nozzle,
E-mail address: [email protected] (A. Khayrullina). jet Reynolds number (Re), the shape of the jet nozzle and its

https://doi.org/10.1016/j.euromechflu.2018.10.003
0997-7546/© 2018 Elsevier Masson SAS. All rights reserved.
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 229

lacking accuracy in predicting another. For example, Aziz et al.

[31] found a better performance of SKE than RNG in predicting
the jet spreading rate, jet decay rate and velocity profiles of a free
plane turbulent jet, while the turbulent kinetic energy was only
accurately predicted by RNG. Angioletti et al. [18], Jaramillo et al.
[21] and Dutta et al. [25] indicated the absence of generality in
results for different studied cases of PTIJ due to different boundary
conditions and computational settings.
In spite of the rather large number of numerical studies on
PTIJs performed in the past, there is still a scarcity of validation
studies for PTIJ at moderate Reynolds numbers (i.e. 5000 < Re <
15,000) for a jet height larger than 10 jet nozzle widths, including
an evaluation of both mean velocity and turbulence intensity (see
Table 1). Moreover, most of the numerical studies on PTIJs only
focused on the jet impingement region. Furthermore, there is an
evident inconsistency in conclusions of validation studies on PTIJs.
Therefore, the aim of the present study is to validate steady RANS
CFD simulations of PTIJ flows with a jet height to width ratio of
hjet /wjet = 22.5 at two moderate Reynolds numbers, i.e. Re = 8000
and Re = 13,000, based on 2D particle image velocimetry (PIV) data
Fig. 1. Structure of turbulent impinging jet. for the whole vertical centreplane of the PTIJ. Section 2 provides
a short description of the experimental set-up. Section 3 contains
an overview of the computational model. The results of the CFD
simulations are presented in Section 4. Section 5 (discussion) and
aspect ratio (AR, which is the ratio of nozzle depth to width wjet )
Section 6 (conclusions) conclude this paper.
(e.g. [9,10]). The height of the impingement region hi is found to be
slightly dependent on jet Reynolds number (e.g., [10]). However, 2. Reduced-scale PIV measurements
Maurel and Solliec [9] stated that the jet behaviour did not vary
with Reynolds number or jet height. A number of studies provided For validation purposes, isothermal reduced-scale experiments
correlations for jet velocity distributions within the intermediate of PTIJ flow are performed in a water channel, where a plane
region (e.g. [11,12]) and near the impingement plate (e.g. [9]). jet is issued from a smoothly-shaped nozzle (Fig. 2). Hydrostatic
The scientific literature available on air curtains and plane im- pressure is created by a water column and upstream of the noz-
pinging jets can be divided into two categories: (1) application- zle, water flows through a conditioning section consisting of one
oriented studies on air curtains and (2) basic studies on plane honeycomb and two screens in order to reduce lateral and longi-
impinging air jets. Applied and basic studies can both be per- tudinal turbulence intensity and the mean velocity gradients (more
formed experimentally (at full or reduced scale, on-site or in a uniform flow). The Reynolds number is defined based on the mean
laboratory) and numerically, i.e. using computational fluid dynam- streamwise velocity at the jet exit V0 and the jet nozzle width wjet ;
ics (CFD). Application-oriented studies are generally focused on Re = (V0 wjet )/ν , with ν the kinematic viscosity. In the experiment,
the dimensioning of air curtains, determination of the air-curtain V0 is equal to 0.50 m/s and 0.81 m/s, corresponding to Re = 8000
efficiency by development of empirical formulae, and defining and Re = 13,000, respectively. The jet is issued vertically on a
the most influential parameters with respect to the air curtain horizontal plate at a distance of hjet = 360 mm from the jet inlet.
separation efficiency. Basic studies generally analyse the structure The value of wjet is equal to 16 mm resulting in hjet /wjet = 22.5 and
of the jet, turbulence levels and vortical structures by means of the nozzle depth (dimension in spanwise direction; z direction) is
different measurement, visualisation and numerical techniques. 300 mm, resulting in a nozzle aspect ratio AR of 18.75.
This paper focuses on the capability of steady Reynolds-averaged For the PIV measurements, a 2D PIV system is used consisting of
Navier–Stokes (RANS) CFD simulations to predict plane turbulent a solid-state frequency-doubled Nd:Yag laser (wavelength 532 nm
impinging jet flows. Table 1 provides a non-exhaustive overview and repetition rate 15 Hz) as a light source, and a charge coupled
of previous numerical validation studies with the steady RANS device (CCD) camera (1600 x 1200 pixels resolution, up to 30
approach on PTIJs. This overview is used to define the novel con- frames/s) for image recordings. The light sheet is delivered at the
tribution of the present paper. bottom of the water channel by a set of mirrors and cylindrical
Modelling PTIJ involves solving the free shear layer within the and spherical lenses. Seeding is provided by polyamide particles
jet potential core and intermediate region, and the jet near-wall (D = 50 µm, density ρs = 1030 kg/m3 ) added to the water.
behaviour in the impingement region. Seyedein et al. [13], Hey- Three sets of measurements are performed, which focus on (Fig. 2):
erichs and Pollard [14], Shi et al. [16], Habli et al. [29], Achari and (1) ROI1 — the entire measurement plane of (W x H = 192 x
Das [26] and Moureh and Yataghene [28] and others questioned 360 mm2 ) (2) ROI2 — the region close to the jet nozzle (W x
the performance of the standard k − ε model (SKE) for PTIJ, while a H = 128 x 192 mm2 ), and (3) ROI3 — the impingement region (W x
number of validation studies for free plane turbulent jets (i.e. with- H = 128 x 176 mm2 ). The PIV measurements provide information
out impingement) showed a sufficient performance of steady RANS on the mean velocities and turbulence intensities in the vertical
in combination with SKE (e.g. [20,30,31]). According to Le Song centreplane. More information on the experimental set-up and
and Prud’homme [19], Rhea et al. [22] and Kubacki et al. [32], the measurement results can be found in Khayrullina et al. [34].
Boussinesq relationship that uses the turbulent viscosity to relate
turbulent shear stresses and mean strain rate is responsible for 3. CFD simulations
the inability of RANS models to accurately predict the turbulence
quantities in the free jet region and specifically close to the wall. 3.1. Computational geometry and grid
However, the performance of turbulence models can differ per
characteristic zone of a PTIJ (e.g. [33]). In addition, one turbulence The computational geometry (L x H x D = 1.0 × 0.36 x 0.15
model can provide accurate predictions for one flow variable while m3 ) replicates one-fourth of the experimental setup described in
230 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243

Table 1
Overview of CFD studies with RANS on PTIJ.
Authors hjet /wjet Re Turb. model Inlet boundary condition Validation Provided results
parameters
Jet region Impingement region
Seyedein et al. [13] 2.5–7.5 5000–10,000 LRKELS , Vel, k, ε P, Nu Vel P, Nu
LRKEC ,
LRKELB
Heyerichs and Pollard [14] 2.6 10,000 SKE, LRKE, Vel, k, ε Nu Nu, C f
SKO
Chen et al. [15] 2–8 250–20,000 SKO Vel, k, ω Nu, Vel Vel Nu, Sh
Shi et al. [16] 2, 6, 12 1500–71,300 SKE, RSML Vel, I, l Nu Vel Nu
9800
Park et al. [17] 0.5–4 16,400 SKO Vel, k, ω Nu, Vel Vel Nu, C f
25,100
Angioletti et al. [18] 4.5 1000–4000 RNG, RSML , Vel, I Vel, Nu Vel Nu
SST
Le Song and Prud’homme [19] 10 6000 URANS LRKE Vel, k, ε Vel, RMS Vel, k Vel, k
Isman et al. [20] 4–10 4000–12,000 SKE, RNG Vel, k, ε Nu k, ε , Nu
Jaramillo et al. [21] 4, 9.2 20,000 SKO, SKE Vel, k, ε Vel, RMS, Nu Vel, RMS Nu
Rhea et al. [22] 8 10,000 RSMD Vel, l Vel, RMS, Re Vel, RMS Vel, RMS
stresses
Koseoglu and Baskaya [23] 2, 6, 12 10,000 LRKE Vel, k, ε Nu Nu
13,500 Vel, RMS, Re
Kubacki and Dick [24] 4, 9.2, 10 WKO Vel, k, ω Vel, I, Re stresses Cf , Nu, Vel, k
20,000 stresses, C f , Nu
Dutta et al. [25] 4, 9.2 20,000 LRKE, RNG, Vel, k, ε, ω, RMS Vel, RMS, Nu Vel, I Nu, Vel, I
RKE, SKE,
SKO, SST, v2 f
Achari and Das [26] 4, 9.2 20,000 LRKE, SKE, Vel, k, ε, ω Vel, RMS, I, P, Vel, RMS P , Cf , Nu
SKO C f , Nu, T
Benmouhoub and Mataoui [27] 8 10,000–25,000 RSML Vel, k, ε Vel, Nu Vel, T Nu
Moureh and Yataghene [28] 10 8000 SKE, RSMLR Vel, I, Re stresses Vel, RMS Vel, I, T

Table legend: WKO — k − ω by Wilcox, LRKEC — Chien Low-Re k − ε , LRKELS — Launder and Sharma Low-Re k − ε , LRKELB — Lam–Bremhorst Low-Re k − ε , RSML — Launder
Reynolds stress model (RSM), RSMLR — Wilcox RSM, RSMD — Dianat RSM, v 2 f — normal velocity relaxation model of Durbin, Cf — skin friction coefficient, I — turbulence
intensity, l — turbulence length scale, k — turbulent kinetic energy, Nu — Nusselt number, P — surface pressure, RMS — RMS velocity distributions, Sh — Sherwood number,
T — temperature distributions, Vel — velocity distributions, ε — turbulence dissipation rate, ω — specific turbulence dissipation rate.

results with those obtained from a simulation with the full do-
main (dimensions equal to experimental set-up). In addition, a 3D
computational geometry is chosen for this validation study, since
small but noticeable differences are observed between the results
from 3D and 2D simulations. The computational grid is created
by the surface-grid extrusion technique (e.g. [35]) resulting in a
structured hexahedral mesh for the water channel section and an
unstructured hexahedral mesh for the smoothly shaped nozzle. A
high spatial resolution is applied for the boundary layers (near the
nozzle surfaces and the impingement plate), in regions with high
velocity gradients at the jet nozzle (y = 0), and at the outlet of
the domain. The grid resolution is obtained from a grid-sensitivity
analysis using three different grids√(see Section 3.4 for results)
with a grid-refinement factor of 2 in each direction; coarse
grid (1,003,696 cells), basic grid (2,809,800 cells) and fine grid
(7,915,275 cells). Across the computational inlet (half of the nozzle
width) 14, 20 and 28 cells are used for coarse, basic and fine grids,
respectively (Fig. 4). The dimensionless wall distances (y∗ , average
values) for the coarse, basic and fine grids in the jet nozzle are equal
to 4, 2.5 and 1, respectively, while at the impingement plate they
are equal to 5, 1.25 and 0.5, respectively. The value of y∗ is defined
1/4 1/2
as y∗ = ρ Cµ kP yP /µ with ρ the density of the fluid (water),
µ the dynamic viscosity, kP the turbulent kinetic energy in the
centre point P of the wall-adjacent cell, yP the distance from point
Fig. 2. Experimental set-up with indication of measurement regions of interest P to the wall, Cµ the model constant generally equal to 0.09 (exact
(ROI) [34]. value depends on the used turbulence model). The low values of
y∗ (below 5) enable the use of low-Reynolds number modelling,
which implies resolving the flow in the immediate vicinity of the
Khayrullina et al. [34] (Fig. 3). A symmetry boundary condition wall including the thin viscous sublayer.
with zero normal velocities and zero normal gradients of all vari-
3.2. Boundary conditions
ables is used to limit the computational costs, as the geometry
and time-averaged flow pattern are regarded to be symmetri- At the inlet of the computational domain a uniform jet exit
cal. This assumption is successfully verified by comparison of the velocity in vertical direction of V0 = 0.085 m/s and V0 = 0.138 m/s
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 231

3/4
the turbulence length scale l (= 0.07Lgap /Cµ ) is calculated based
on the opening size (Lgap = 1 mm) in the finest screen in the
conditioning section with Cµ equal to 0.09 [36]. At the outlet, zero
static gauge pressure is applied, whereas zero normal velocities
and zero normal gradients of all the variables are imposed at the
two symmetry planes. The remaining surfaces are modelled as no-
slip walls.

3.3. Solver settings

The 3D steady RANS equations are solved with ANSYS Fluent 16

[36] using five different turbulence models to provide closure: (1)
standard k − ε model (SKE) [37]; (2) realisable k − ε model (RKE)
[38]; (3) renormalisation group k − ε model (RNG) [39]; (4) shear
stress transport k −ω model (SST) [40] with low-Re corrections and
a limiter applied to the production of turbulent kinetic energy [41];
(5) linear pressure–strain Reynolds stress model (RSM) [42]. These
models are chosen due to their good performance in previous
PTIJ studies (see Section 1) and their common use in engineering
applications due to their good balance between computational
demand and accuracy (e.g. [43]). Moreover, these turbulence mod-
els represent models from two different groups: (1) first-order
closure models based on the Boussinesq eddy-viscosity hypothesis
(k − ε and k − ω models); and (2) second-order closure models;
i.e. Reynolds stress models (RSM).
All models utilise low-Re number modelling for near-wall re-
gions, which is possible due to the sufficiently low dimensionless
Fig. 3. (a) Computational geometry with indication of boundary conditions. (b–d)
Computational grid near the: (b) smoothly-shaped nozzle; (c) impingement plate; wall distances that are present (y∗ < 5, see Section 3.1).
(d) outlet. Second-order upwind discretisation schemes are used for both
the convection and viscous terms of the governing equations,
pressure–velocity coupling is performed by the SIMPLEC algo-
rithm, and pressure interpolation is second order. Oscillatory con-
vergence of the scaled residuals is observed for RKE, RNG, SST and
RSM. Due to oscillations of the velocity magnitude observed at a
reference point on the jet centreline at y/hjet = 0.9 (see Fig. 5a),
the results of RKE, RNG, SST presented in the remainder of the
paper are averaged over a sufficient number of iterations (70,000–
100,000 iterations). The values of the cumulative moving average
(CMA) of velocity magnitude in Fig. 5b show a steady behaviour for
all models meaning that the results are averaged over a sufficient
number of iterations.

3.4. Grid-sensitivity analysis

The grid-sensitivity analysis is conducted with RKE and SST for

case I (Re = 8000) with coarse (1,003,696 cells), basic (2,809,800
cells) and fine (7,915,275 cells) grids. The results of the grid-
sensitivity analysis are provided in Fig. 6; i.e. the centreline profiles
(x = 0, z = 0) of normalised jet velocity in the vertical direction
(V /V0 ). The grid-convergence index (GCI) by Roache [44,45] is
calculated for the basic grid, which is shown by shaded bounds in
Fig. 6, using:
⏐ ( basic fine
)⏐
⏐ rp V − VV ⏐⏐
⏐ V0 0
V
GCIbasic = FS ⏐⏐ ⏐ (3.1)
⏐ 1 − rp ⏐
⏐

Fig. 4. Configurations of the computational grid with indication of the boundary with FS the safety factor equal to the recommended value of 1.25
conditions and the number of cells over the half of the contraction width: coarse when at least three
√ grids are analysed, r the linear grid refinement
(14 cells per inlet), middle (20 cells) and fine (28 cells). factor equal to 2, p the formal order of accuracy, which is assigned
the value of 2 as second-order discretisation schemes are used for
the simulations [45]. The requirement regarding the minimum grid
is imposed, which lead to Re = 8000 (case I) and Re = 13,000 (case refinement factor is not straightforward. Guidelines by Casey and
Wintergerste [46] stated the grid should be doubled twice in each
II) at the nozzle exit, respectively. The turbulence intensity is set to
direction for grid-sensitivity analyses, resulting in r = 2. How-
15%, which resulted in a turbulence intensity on the jet centreline ever, Roache [47] and Celik et al. [48] stated that for engineering
near the jet nozzle that corresponded to the measured values, and applications, an appropriate grid refinement factor should be at
232 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243

Fig. 5. (a) Velocity magnitude |V | monitored over number of iterations on the jet centreline at y/hjet = 0.9 for different turbulence models. (b) Cumulative moving average
of velocity magnitude at the same point. Note that the zeroth iteration corresponds to the iteration where the scaled residuals levelled off and started to oscillate.

least 1.3 and 1.1, respectively. In order to limit the increase in respectively. The square brackets indicate averaging over all data
computational demand per grid refinement, and to be in line with points. This study provides an analysis of 150 data points along the
the different aforementioned
√ guidelines, the grid refinement factor jet centreline and 50 cross-jet values per line (three lines), resulting
in this study is r = 2. in a total number of 300 data points used in the analysis, which is
Fig. 6 shows that the results of the grid-sensitivity study vary considered to be sufficient for this particular case. The number of
per turbulence model. The prediction of velocities by RKE is sen- required data points depends on the flow pattern, flow complexity,
sitive to the grid resolution within the intermediate (0.2 < y/hjet and aim of the study.
< 0.87) and the impingement (y/hjet ≥ 0.87) regions. However, The values of FAC1.1 and FAC1.5 show the fraction of considered
the results obtained with the basic and the fine grids are very data points, where the simulation results fall within a factor of
close to each other. The solution provided by SST shows a higher 1.1 and 1.5 of the experimentally obtained values, respectively.
sensitivity to the grid resolution for basic and fine grids in the FAC1.1 and FAC1.5 provide a clear picture of randomly occurring
intermediate (0.35 < y/hjet < 0.87) jet region. The average values high or low differences between measured and predicted values of
of GCI V basic along the jet centreline are 1.0% and 4.3% for RKE and characteristic flow quantities [49]. The values of R reflect the lin-
SST, respectively. The maximum values of GCI V basic are 2.2% at ear relationship between experimental and numerical results and
y/hjet = 0.55 for RKE and 9.7% at y/hjet = 0.89 for SST. Based provide an overall assessment of model performance. However, it
on this analysis it is concluded that the basic grid provides nearly can be influenced by a few either low or high outliers To the best
grid-independent results and it is therefore used in the remainder knowledge of the authors, the quantitative requirements regarding
of this study. model validation are available only for prediction of pollution
dispersion cases, e.g. Chang and Hanna [50] and Schatzmann et al.
4. Results [49], who used a threshold of FAC2 > 0.5 in the model assessment
for pollution dispersion predictions. In the present study, FAC 1.1
4.1. Validation metrics and FAC 1.5 are used in order to clearly illustrate the performance
differences between different RANS models.
In order to provide a quantitative assessment of the perfor-
mance of different turbulence models the following validation 4.2. Mean velocity and turbulent kinetic energy distributions near the
metrics are applied: the factor of 1.1 and 1.5 of the observations jet nozzle
(FAC1.1 and FAC1.5, respectively) and the correlation factor (R).
These metrics are calculated as follows:
Fig. 7 shows normalised time-averaged velocity magnitude
n
{ Pi
} (|V |/V0 ) and turbulent kinetic energy (TKE; k/V02 ) profiles near the
1∑ 1 for 0.91 ≤ ≤ 1.1
FAC 1.1 = Ni with Ni = Oi jet nozzle from the PIV measurements [34] and CFD simulations
n 0 else of case I (Fig. 7a, b; Re = 8000) and case II (Fig. 7c, d; Re =
i=1
13,000). The distributions are taken at a distance of y = 2wjet
(4.1)
from the jet inlet. Note that the simulations are performed with
symmetry boundary conditions, therefore the results are reflected
in the jet centreline (x/wjet = 0). In the remainder of the paper, all
Pi
n
{ }
1∑ 1 for 0.67 ≤ ≤ 1.5 results will consist of the time-averaged measurement data from
FAC 1.5 = Ni with Ni = Oi PIV and the mean values from the steady RANS CFD simulations.
n 0 else
i=1 A quantitative comparison is made in Table 2 using the validation
(4.2) metrics from Section 4.1.
Based on the velocity distributions provided in Fig. 7a, c and the
∑n validation metrics provided in Table 2 the following observations
1
n i=1 (Oi − [O]) (Pi − [P]) are made:
R = (4.3)
σP σO – For both cases, the distributions of velocities predicted by
with Pi and Oi the time-averaged values obtained from CFD sim- the numerical simulations generally show a less clear top-hat
ulations and PIV experiments, respectively, n the number of mea- shape profile compared to the measured profiles, except for
surement data points, σP and σO the standard deviation of Pi and Oi , the results from SST that most clearly represent the top-hat
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 233

Fig. 6. Results of grid-sensitivity analysis: V /V0 along jet centreline. (a) RKE; (b) SST. Grey bounds indicate GCI index calculated for the basic grid.

Table 2
Validation metrics defined for cross-jet velocity distributions near the jet inlet at y = 2wjet .
Model Velocity magnitude at y = 2wjet Turbulent kinetic energy at y = 2wjet
Case I Case II Case I Case II
R FAC1.1 R FAC1.1 R FAC1.5 R FAC1.5
SKE 0.909 0.32 0.923 0.40 0.634 0.08 0.710 0.16
RKE 0.918 0.40 0.929 0.44 0.659 0.08 0.724 0.32
RNG 0.919 0.40 0.930 0.44 0.636 0.08 0.701 0.28
SST 0.947 0.36 0.951 0.36 0.741 0.40 0.800 0.44
RSM 0.919 0.32 0.929 0.32 0.662 0.08 0.720 0.20
Perfect performance 1 1 1 1 1 1 1 1

Fig. 7. Comparison of PIV results with results of steady RANS CFD simulations. Cross-jet profiles of: (a,c) normalised jet velocity magnitude (|V |/V0 ); (b,d) normalised
turbulent kinetic energy (k/V02 ). (a,b) Case I (c,d) Case II. x/wjet = 0 corresponds to the jet centreline.

profile. The results of the validation studies by Berg et al. [30] with the measured velocities in this initial region based on R
and Rhea et al. [22], who investigated SKE and RSM models, is achieved with SST (R = 0.947 and 0.951 for cases I and II,
respectively, are in line with this finding. The best agreement respectively) and the worst with SKE (R = 0.909 and 0.923
234 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243

for cases I and II, respectively). When considering FAC1.1, the Table 3
Spanwise component wRMS 2 for TKE.
best agreement is obtained with RKE and RNG; FAC1.1 = 0.40
Profile name Spanwise component wRMS 2
and 0.44 for case I and case II, respectively, while the worst
agreement is for case I is obtained with SKE and RSM, and for y/hjet > 0.85: wRMS 2 = a(y)uRMS 2 a(y) — variable defined from
PIV-1
y/hjet ≤ 0.85 : wRMS 2 = uRMS 2 Gutmark et al. [52]
case II with RSM.
– For both cases the velocities in the outer jet region (0.5 < PIV-2 wRMS 2 = 0
|x|/wjet < 1) are overestimated by all the turbulence models. PIV-3 wRMS 2 = 0.5(uRMS 2 + vRMS 2 )
– The ambient fluid starts to entrain the jet potential core at a
distance |x|/wjet from the jet centreline, where |V |/V0 = 0.98
[51]. In case I, |x|/wjet = 0.15 for SKE, |x|/wjet = 0.20 for RKE
4.3. Centreline mean velocity and turbulent kinetic energy distribu-
and RNG, |x|/wjet = 0.30 for SST and |x|/wjet = 0.18 for RSM. tions
Note that in Fig. 7a the line for SKE overlaps with that of RKE
and RSM for |x|/wjet < 0.5. In case II, |x|/wjet = 0.23 for RKE Fig. 8 shows V /V0 and k/V02 along the jet centreline obtained
and RNG, |x|/wjet = 0.20 for SKE, |x|/wjet = 0.30 for SST and with different turbulence models.
|x|/wjet = 0.23 for RSM. The value obtained from PIV is equal For case I (Fig. 8a, b) the following observations are made:
to |x|/wjet = 0.33 and 0.32 for cases I and II, respectively.
– RNG and RKE accurately predict V /V0 along the jet centreline.
Fig. 7b, d shows three profiles of k/V0 2 (PIV-1, PIV-2, PIV-3) calcu- SKE underestimates V /V0 in the potential core and the be-
lated using: ginning of the intermediate jet region (0 < y/hjet < 0.5), and
accurately predicts V /V0 for y/hjet > 0.5. RSM is consistent
k/V02 = 0.5(uRMS 2 + vRMS 2 + wRMS 2 )/V0 2 (4.4)
with the experimental data within 0 < y/hjet < 0.4, while it
with uRMS , vRMS , wRMS the lateral, streamwise and spanwise RMS overestimates V /V0 further downstream (y/hjet > 0.4). SST
velocities, respectively. Due to the absence of the spanwise compo- overestimates V /V0 for y/hjet > 0.2.
– Values of k/V02 within the jet potential core (0 < y/hjet <
nent (wRMS 2 ) in the 2D PIV experiments by Khayrullina et al. [34],
0.2) are overestimated by SKE, RNG and RKE compared to the
its value needs to be estimated, which is done using the assump-
experimental values.
tions described in Table 3. The profile of PIV-1 uses the relation
– At the beginning of the intermediate jet region (0.2 < y/hjet
between the centreline profiles of uRMS 2 and wRMS 2 provided by
< 0.4), profiles of k/V02 by RKE and RSM correspond well to
Gutmark et al. [52]; uRMS 2 = wRMS 2 on the jet centreline for y/hjet ≤
the experimental data.
0.85. For y/hjet > 0.85 the values of wRMS 2 become slightly higher
– Within 0.60 < y/hjet < 1 all turbulence models overestimate
than uRMS 2 (factor a is larger than 1), which can be explained by
values of k/V02 . In contrast, Rhea et al. [22] revealed some un-
vortex stretching along the impingement plate, which enhances derpredictions, particularly by RSM, of turbulence quantities
spanwise (wRMS ) and streamwise (vRMS ) velocity fluctuations. For in the free and wall jet regions.
PIV-2, the spanwise component is equal to zero; for PIV-3 wRMS 2 is – The maximum experimentally obtained value of k/V02 at the
assumed to be equal to the average of the lateral and streamwise centreline within the intermediate jet region occurs at y/hjet
RMS velocities. Fig. 7b, d shows that: ≈ 0.53 and reaches k/V02 ≈ 0.023. The location of this peak is
accurately predicted by RKE and RSM (y/hjet ≈ 0.52 and 0.53,
– The CFD profiles of k/V02 near the jet inlet correspond best
respectively), and their predicted values (k/V02 = 0.019 and
to the profile with the assumption that wRMS 2 is zero (PIV-
0.018, respectively) provide the closest agreement with the
2). Therefore, only the results of PIV-2 are provided in the
experimental data.
remainder of the paper.
– The maximum experimentally obtained TKE in the jet im-
– SST provides the closest agreement (R = 0.741 and 0.8, and
pingement region occurs at y/hjet ≈ 0.98 and is equal to
FAC1.5 = 0.40 and 0.44 for cases I and II, respectively), es- k/V02 ≈ 0.020. SST estimates the maximum TKE near the
pecially within the jet potential core |x|/wjet < 0.3. SKE, impingement region to be k/V02 ≈ 0.024 at y/hjet ≈ 0.98,
RKE, RNG and RSM underestimate k/V02 in the shear layer which is close to the experimental data. The other models
(0.3 < |x|/wjet < 0.75) and overestimate k/V02 close to the overestimate the maximum value of TKE in the impingement
jet centreline (x/wjet = 0). region.
– The values of k/V02 near the jet nozzle in the shear layer
are underestimated by CFD simulation compared to that of A slightly different performance of the turbulence models is ob-
experiments for both cases. For example, for case I (Fig. 7b) served for case II (Fig. 8c, d):
SST provides a maximum value of k/V02 = 0.028, while ex- – RNG and SKE predict mean velocities that are consistent with
perimental data (PIV-2) shows a maximum of k/V02 = 0.068. the experimental data, while RKE, SST and RSM overestimate
Similarly, Rhea et al. [22] and Achari and Das [26] observed V /V0 in the intermediate jet region (0.3 < y/hjet < 0.9).
underestimations of turbulent kinetic energy by k − ε models – The maximum measured value of k/V02 at the centreline in the
in the shear layer, especially close to the inlet. Achari and Das intermediate region reaches k/V02 ≈ 0.021 at y/hjet ≈ 0.55.
[26] demonstrated that the standard k − ω model provided The location and magnitude of this peak is quite accurately
better predictions of turbulent kinetic energy near the inlet; predicted by RNG, showing k/V02 ≈ 0.021 at y/hjet ≈ 0.53.
however, this turbulence model is not tested in the present – The maximum measured value of k/V02 in the jet impinge-
study. ment region occurs at y/hjet ≈ 0.99 and is equal to k/V02 ≈
– In both cases SKE shows the largest overestimation of k/V02 0.022. The maximum value of k/V02 near the impingement
within the jet potential core. For example, for case I (Fig. 7b), region by SST is k/V02 ≈ 0.024 at y/hjet ≈ 0.98, which
SKE provides a maximum value of k/V02 = 0.012, while the corresponds well to the value from PIV. The other models
experimental data show that k/V02 approaches zero (k/V02 ≈ again overpredict k/V02 in the impingement region. Heyerichs
0.0005) near the jet centreline (x/wjet = 0). and Pollard [14] and Chen et al. [15] found that the standard
k − ω model could adequately predict turbulence levels in
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 235

Table 4 Table 6
Validation metrics for centreline distributions of velocity and turbulent kinetic Validation metrics for cross-jet distributions of velocity and TKE for case I.
energy. Model V /V0 k/V02
Model Centreline velocity Turbulent kinetic energy
R FAC1.1 R FAC1.5
Case I Case II Case I Case II
SKE 0.97 0.56 0.92 0.52
R FAC1.1 R FAC1.1 R FAC1.5 R FAC1.5 RKE 0.98 0.62 0.95 0.75
SKE 0.98 0.93 0.98 0.96 0.70 0.63 0.79 0.66 RNG 0.97 0.56 0.94 0.66
RKE 0.98 0.96 0.97 0.72 0.91 0.89 0.88 0.72 SST 0.98 0.62 0.94 0.63
RNG 0.98 0.91 0.98 0.94 0.94 0.80 0.92 0.83 RSM 0.98 0.51 0.94 0.62
SST 0.96 0.37 0.96 0.36 0.68 0.53 0.81 0.51 Perfect performance 1 1 1 1
RSM 0.97 0.72 0.97 0.55 0.78 0.91 0.79 0.64
Perfect performance 1 1 1 1 1 1 1 1
Table 7
Validation metrics for cross-jet distributions of velocity and TKE for case II.
Table 5 Model V /V0 k/V02
Normalised length of the jet potential core hp /wjet . R FAC1.1 R FAC1.5
Case I Case II SKE 0.97 0.42 0.92 0.64
hp /wjet % difference hp /wjet % difference RKE 0.97 0.43 0.94 0.68
RNG 0.97 0.42 0.91 0.71
PIV 4.85 4.85
SST 0.97 0.45 0.93 0.67
SKE 2.76 −43 3.67 −24
RSM 0.96 0.40 0.92 0.66
RKE 3.80 −21 4.48 −8
RNG 4.48 −8 5.33 +10 Perfect performance 1 1 1 1
SST 8.74 +80 8.41 +73
RSM 4.34 −10 5.91 +22

less spreading of the jet than the other models (Fig. 9a). This is due
to the lower predicted levels of TKE by SST near the jet centreline
the stagnation zone. Moreover, Angioletti et al. [18] and Dutta and in the shear layer, and therefore less intensive entrainment of
et al. [25] stated that SST provided more reliable predictions ambient fluid compared to the other models. The velocity profiles
of turbulent kinetic energy in the stagnation region in their of RKE, RNG and RSM are nearly identical. The jet potential core
particular case. predicted by SST runs until |x|/wjet = 0.2. RKE, RNG and RSM
Table 4 shows the validation metrics for velocities and TKE provide a fairly accurate prediction of TKE within |x|/wjet < 0.3
along the jet centreline, from which it can be concluded that RKE (Fig. 9b), with a total average deviation of 38%, 22% and 37%, respec-
and RNG show the best performance for predicting centreline tively. Fig. 9b also shows that SKE predicts the highest turbulence
velocities for case I, while SKE and RNG show the best performance levels among all the models within |x|/wjet < 1. The maximum
for case II. RNG and RKE show the best agreement of centreline values of TKE in the shear layer reach k/V02 ≈ 0.040 for SKE, which
distributions of k/V02 for case I and case II. corresponds well to the experimental value (k/V02 ≈ 0.044). Fig. 9c,
The measured and predicted normalised length of the jet po- d indicates that at y/wjet = 11.3 the cross-jet velocity profiles of
tential core (hp /wjet ) for both cases are provided in Table 5. For RKE nearly overlap with the measurement data for |x|/wjet < 1.
case I the best agreement with experiments is provided by RNG and Fig. 9d shows that SKE provides the best correspondence to the
RSM, resulting in an underestimation by 8% and 10%, respectively. measured TKE profile for |x|/wjet < 1, while RKE provides the best
For case II, RKE underestimates the jet potential core length by 8%. agreement for |x|/wjet > 1. The highest values in jet shear layer
For both cases, SKE fails to predict the length of the potential core (k/V02 ≈ 0.032) are predicted by SST. Finally, Fig. 9e, f shows the
due to excessive levels of TKE along the jet centreline and in the profiles at y/wjet = 19.3, which is at the end of the intermediate jet
shear layer near the jet nozzle, as shown in Figs. 7b, d and 8b, d. The region and at the beginning of the impingement region. The cross-
opposite behaviour is shown by SST, which underestimates values jet velocity profiles show a good overall correspondence to the PIV
of TKE on the jet centreline resulting in a substantially longer measurements. The best agreement appears to be present for SKE
potential core (+ 80% and + 73% for case I and case II, respectively). and RNG, followed by RKE (Fig. 9e). All models overestimate the
This finding is in line with the results from the study of Achari values of TKE, with a maximum value of k/V02 ≈ 0.021 (SST), while
and Das [26]. In contrast, Achari and Das [26] and Moureh and the measured maximum value is k/V02 ≈ 0.016 (Fig. 9f).
Yataghene [28] revealed an overestimation of jet potential core by
SKE model, while the present study indicates an underestimation 4.4.2. Case II
of the potential core length (−43% and −24% for case I and case II, The cross-jet distributions of V /V0 and k/V02 for case II are
respectively). compared in this section. The results in Table 7 show that SST
provides the best agreement with experimental data for velocity
4.4. Cross-jet mean velocity and turbulent kinetic energy distributions distributions (R = 0.97 and FAC1.1 = 0.45), while RKE shows the
best agreement with experimental data for TKE (R = 0.94 and
4.4.1. Case I FAC1.5 = 0.68).
The cross-jet distributions of V /V0 and k/V02 for case I are com- Fig. 10a, b shows the cross-jet profiles at y/wjet = 4. Again, SST
pared at three different distances from the jet nozzle: y/wjet = 4, provides the best agreement with PIV experiments for the cross-
y/wjet = 11.3 and y/wjet = 19.3. The validation metrics averaged jet velocity distribution close to the inlet. All models overestimate
over these three distances are provided in Table 6 that indicates the values of k/V02 , with maximum differences of 21% (SST) to 28%
that the results of RKE show overall the closest agreement with (SKE). The maximum values of TKE in the shear layer reach k/V02 ≈
the experimental data (R = 0.98 and FAC1.1 = 0.62 for velocities; 0.036 for SKE, which predicts the highest turbulence level among
R = 0.95 and FAC1.5 = 0.75 for TKE). all the models and deviates most from the experimental data
Fig. 9a, b shows the cross-jet velocity and TKE profiles at y/wjet (maximum k/V02 ≈ 0.028). Fig. 10c, d depicts the cross-jet profiles
= 4. The velocity profile of SST, which provides a close correspon- of V /V02 and k/V02 aty/wjet = 11.3. The numerically obtained
dence to experimental data (R = 0.98 and FAC1.1 = 0.62), shows cross-jet profiles of velocity do not show very large differences. As
236 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243

Fig. 8. Comparison of PIV results with results of steady RANS CFD simulations obtained along the jet centreline (x/wjet = 0): (a,c) normalised centreline velocity (V /V0 );
(b,d) normalised turbulent kinetic energy (k/V02 ). (a,b) Case I. (c,d) Case II. y/wjet = 0 corresponds to the jet nozzle.

for TKE, SST again provides the highest values in the shear layer Table 8
(maximum k/V02 ≈ 0.030) and overestimates the experimentally Overall validation metrics for distributions of velocity and TKE (cases I and II
combined).
obtained values (maximum k/V02 ≈ 0.025) by 20% in this region.
Model Velocity Turbulent kinetic energy
Finally, cross-jet profiles at y/wjet = 19.3 are shown in Fig. 10e,
R FAC1.1 R FAC1.5
f. SKE and RNG appear to provide the best agreement with the
measured velocity profile (Fig. 10e). SST and RNG provide the SKE 0.961 0.553 0.833 0.502
RKE 0.964 0.567 0.883 0.554
highest values for TKE (maximum k/V02 ≈ 0.02) and overestimate
RNG 0.963 0.564 0.873 0.612
the experimentally obtained values (maximum k/V02 ≈ 0.018) by SST 0.966 0.468 0.863 0.545
11%, while SKE, RKE and RSM show a very good agreement (within RSM 0.961 0.464 0.851 0.486
3%) (Fig. 10f). Perfect performance 1 1 1 1

4.5. Overall performance

As shown in previous sections, no single model shows the and FAC1.1 = 0.567). RKE and RNG provide the best agreement
best performance in predicting both mean velocities and TKE at with the measured TKE (R = 0.883 and FAC1.1 = 0.554 for RKE,
different distances from the jet inlet. The overall performance R = 0.873 and FAC1.1 = 0.612 for RNG). Note that the differences
(along jet centreline and three cross-jet profiles; in total 300 data in performance between the models are small.
points) is provided in Table 8. The values provided are calculated
as mean values of R, FAC1.1 and FAC1.5 as provided in Tables 2, 4, 4.6. Spatial distribution of mean velocity and vector plots
6 and 7. The values for case I and II are combined in these mean
values, however, separate values for mean velocities and TKE are Figs. 11 and 12 show spatial distribution of |V |/V0 and the mean
calculated, resulting in four values per model (R and FAC1.1 for velocity vectors in the vertical centreplane for case I and case II,
velocity and R and FAC1.5 for TKE). respectively. The white areas at the top and bottom parts of Fig. 11a
The results show that SST provides the best agreement with and Fig. 12a indicate zones where erroneous measurement data
experimental data for mean velocity distributions when consider- due to obstructions and reflections are omitted. The development
ing the R value (R = 0.966), although differences in R values with of the jet half-width is indicated by dashed white lines. The slope
the other models are small. However, SST predicts only 46.6% of of these lines is used to characterise the jet spreading rate Ky =
the considered mean velocities within a factor of 1.1 (FAC1.1 = dδ0.5 /dy [53]. The values of Ky are defined separately for the poten-
0.468). Hence, based on both R and FAC1.1, RKE provides the best tial core and the intermediate jet regions and are listed in Table 9.
agreement with experimental data for mean velocities (R = 0.964 Fig. 10e clearly shows less intensive jet spreading for SST compared
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 237

Fig. 9. Cross-jet profiles for case I of normalised streamwise velocity V /V0 (a,c,e) and turbulent kinetic energy k/V02 (b,d,f) at: (a,b) y/wjet = 4; (c,d) y/wjet = 11.3; (e,f) y/wjet
= 19.3.

to the other models. However, the jet spreading rate within the in- core region as calculated by the steady RANS CFD simulations
termediate region for SST differs only by 2% and 7% from the value are lower than those from the PIV measurements (case I: 0.05,
from the PIV experiments for cases I and II, respectively (Table 9). case II: 0.04), and thus also lower than the values reported in
However, note that the spreading rate in the intermediate region other experimental studies. Achari and Das [26] attributed the
can be similar while the local width of the jet at the beginning of the underestimation of jet spreading rate as predicted by k − ε models
intermediate region is different due to different jet spreading rates
to a lower entrainment from surrounding stagnant fluid towards
in the potential core region. The substantial difference between
the jet centreline compared to what was found in the experiments.
the values of jet spreading rate for the potential core region and
Table 10 shows the values of the jet decay rate Ku , defined as
the intermediate jet region for SST (Table 9) is caused by a steep
growth of TKE at y/hjet > 0.35, which is depicted in Fig. 8b, d, due to Ku = (V0 /V )2 /((y − y0 )/wjet ), with y0 the vertical distance between
the entrainment of ambient fluid into the jet. The experimentally jet nozzle and the kinematic virtual jet origin. The experimentally
obtained values for the potential core region are Ky = 0.05 and obtained values are Ku = 0.15 and Ku = 0.14, for case I and case
Ky = 0.04, for case I and case II, respectively. SKE provides the II, respectively. For case I, SKE and RNG accurately predict the jet
closest agreement for jet spreading rate within the jet potential decay rate within the intermediate jet region (Ku = 0.15), and for
core for both cases (Ky = 0.04 and 0.03, corresponding to a 22% case II SKE provides an accurate prediction as well (Ku = 0.14).
and 20% difference, for case I and II, respectively). Similarly, Aziz This observation is in line with Aziz et al. [31], who found that SKE
et al. [31] showed that SKE outperformed RNG in predicting jet predicted jet decay rate more accurately than RNG. For both cases
spreading rate. The spreading rate within the jet potential core for the values of jet decay rate are significantly overestimated by SST
case II predicted by the other models differs with 33% to 58% from
(Ku = 0.18), and underestimated by RSM (Ku = 0.12 and Ku = 0.10
the experimental data. All the models provide a good agreement
for case I and case II, respectively). The overestimation of Ku by SST
with the experiments within the intermediate jet region for both
can also be observed in Fig. 8, in which large velocity gradients in
cases(Ky = 0.10). Largest differences are present for RSM and RNG
for case I, with predictions that deviate by 16% and 21% (Table 9). the streamwise direction are present for y/hjet > 0.4. In previous
Experimental studies for free plane turbulent jets with side- experimental studies of free plane turbulent jets with sidewalls,
walls in the range of Re = 6000–16,000 (e.g. [51,54–59]) reported with Re = 6000–16,000 (e.g. [51,54–57,59,60]), a range of values
values of Ky in the range from 0.05 to 0.11 for the potential core and of Ku from 0.11 to 0.22 was reported, which encompasses both the
intermediate jet regions. The jet spreading rates in the potential experimental and numerical values obtained in the present study.
238 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243

Fig. 10. Cross-jet profiles for case II of normalised streamwise velocity V /V0 (a,c,e) and turbulent kinetic energy k/V02 (b,d,f) at: (a,b) y/wjet = 4; (c,d) y/wjet = 11.3; (e,f)
y/wjet = 19.3.

Table 9
Spreading rates Ky for the potential core region and intermediate region.
Case I Case II
Potential core region % difference Intermediate region % difference Potential core region % difference Intermediate region % difference
PIV 0.05 0.10 0.04 0.10
SKE 0.04 −22 0.11 +9 0.03 −20 0.10 +4
RKE 0.03 −46 0.09 −8 0.02 −40 0.10 −3
RNG 0.03 −48 0.12 +16 0.02 −43 0.11 +8
SST 0.01 −72 0.10 −2 0.02 −58 0.09 −7
RSM 0.03 −40 0.08 −21 0.03 −33 0.10 +4

Table 10 4.7. Production of turbulent kinetic energy

The jet decay rate Ku for intermediate jet region.
Case I Case II In order to support the discussion on differences between re-
Intermediate % difference Intermediate % difference sults provided by the different turbulence models, cross-jet pro-
PIV 0.15 0.14 files of the production of turbulent kinetic energy per unit volume
SKE 0.15 0 0.14 −2 (Gk = ρ Pk , with ρ the density and Pk the production of turbulent
RKE 0.18 +19 0.13 −9 kinetic energy per unit mass) are depicted at two distances from
RNG 0.15 +1 0.17 +20 the jet nozzle; i.e. at y/wjet = 0.1 and y/wjet = 2, for cases I and
SST 0.18 +23 0.18 +30
II in Fig. 13. At y/wjet = 0.1, there is a clear difference in predicted
RSM 0.12 −21 0.10 −31
values of Gk in the shear layer of the jet; the highest levels are
predicted by SKE, RKE and RNG and the lowest levels are present
for SST. The maximum values of Gk for case I are located in the
shear layer and range from 250 kg/ms3 for SST to 2510 kg/ms3 for
SKE (Fig. 13a) at y/wjet = 0.1, while they are reduced to about
210 kg/ms3 for SST and around 190 kg/ms3 for the other models at
y/wjet = 2 (Fig. 13b). For case II, the maximum values in the shear
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 239

Fig. 11. Spatial distributions of dimensionless velocity magnitude (|V |/V0 ) for case I (Re = 8000): (a) PIV measurements; (b–f) steady RANS CFD simulations: (b) SKE; (c)
RKE; (d) RNG; (e) SST; (f) RSM.

layer range from 1160 kg/ms3 for SST to 10,340 kg/ms3 for SKE 5. Discussion and future work
(Fig. 13c). At y/wjet = 2, the values of Gk for all models decreased
to around 800 kg/ms3 in the shear layer with the highest values for This paper presented the results of a validation study of a PTIJ at
3 Re = 8000 and Re = 13,000. The results obtained from steady RANS
SST (980 kg/ms ) (Fig. 13d).
CFD simulations with the SKE, RKE, RNG, SST and RSM turbulence
The excessive production of TKE in the initial shear layer by models were compared with experimental data of Khayrullina
SKE explains the highest entrainment and largest jet spreading rate et al. [34].
within the jet potential core, resulting in a shorter potential core
5.1. Differences in turbulence model performance
region (see Table 4). On the contrary, near the jet nozzle, the lowest
levels of Gk are present for SST, resulting in a lower spreading The substantial differences in turbulence model performance
rate within the jet potential core region and an extended potential as described in Section 4 are – among others – caused by dif-
core length (see Table 4). The production term can be regarded as ferent formulations for the eddy viscosity µt . The k − ε models
utilise the equation relating µt to turbulent kinetic energy k and
an important factor in the prediction of the mean velocities and
its dissipation rate ε , and applying a model constant or variable
turbulence levels in impinging jets by steady RANS, as studied in Cµ representing an empirical ratio of Reynolds shear stresses to
this paper. turbulent kinetic energy (e.g. [61]). While SKE and RNG use a model
240 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243

Fig. 12. Spatial distributions of dimensionless velocity magnitude (|V |/V0 ) for case II (Re = 13,000): (a) PIV measurements; (b–f) steady RANS CFD simulations: (b) SKE; (c)
RKE; (d) RNG; (e) SST; (f) RSM.

constant Cµ , RKE uses a model variable Cµ , which depends on the viscosity is an isotropic scalar quantity, which is not always the
deformation of mean flow (strain) and turbulent kinetic energy case. The Boussinesq relationship is known for overestimation of k
and its dissipation in the flow (terms k and ε ). The SST model
in the impingement jet region (e.g., [61–64]). In the current study,
employed uses a modified eddy-viscosity formulation to account
for transport effects of the turbulent shear stress near walls by too high levels of turbulence are exhibited in the impingement
applying a limiter to the eddy-viscosity formulation. In addition, region by the k − ε models, as expected based on literature.
a turbulence damping term is introduced in the ω equation. Please The lower values of turbulent kinetic energy in the impingement
note that SST also utilises a limiter applied to the production of region obtained with SST, compared to the results of other eddy-
turbulent kinetic energy, as mentioned in Section 3.3.
Furthermore, the production term Gk is treated differently in the viscosity models, can be explained by the limiting term applied in
transport equations of different turbulence models. The values of the equation for the production of turbulent kinetic energy. As for
Gk near the jet nozzle are presented in Section 4.7. SKE, RKE and RSM, the linear pressure–strain model used in this study is known
RNG utilise a simplified production term based on the Boussinesq for the overestimation of k in the impingement region due to the
relationship, which relates the Reynolds (turbulent shear) stresses
erroneous redistribution of normal stresses in the impingement
to the mean strain rate using the turbulent viscosity, resulting
in the inability to accurately predict an anisotropic distribution regions by the wall-reflection term applied to the pressure–strain
of Reynolds stresses due to the assumption that the turbulence equation (e.g., [62,63]).
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 241

Fig. 13. Production of turbulent kinetic energy Gk at: (a,c) y/wjet = 0.1; (b,d) y/wjet = 2. (a,b) Case I (c,d) Case II.

5.2. Applicability of considered RANS models in application studies In addition, the observed oscillatory convergence of the resid-
uals and resolved parameters near the impingement plate could
As mentioned before, RANS models are commonly used in en- suggest the application of URANS, LES, and hybrid URANS–LES
gineering applications due to their good balance between compu- models. Furthermore, the present study can be extended by
tational demand and accuracy. The current paper showed that RKE analysing the development of the wall jet near the impingement
and RNG models accurately predicted the general jet flow patterns plate. Finally, the application of non-linear eddy-viscosity models
and characteristics. Therefore, it is expected that moderate errors and, for example, the v 2 − f model, which utilises a depen-
from these predictions will not have a significant impact on appli- dency of turbulent viscosity on the velocity scale v 2 and over-
cation case results. However, if the focus of engineering studies lies comes the problem of inaccurate prediction of near-wall turbu-
in the impingement region (e.g., heat or mass transfer particularly lence anisotropy by the linear eddy-viscosity models (i.e. k − ε
in this region) some discrepancies near the impingement plate are and k − ω models), can be worthwhile to improve the prediction
expected. Furthermore, realistic cases involve a large number of accuracy in the impingement region [67,68], without having to
external processes influencing the jet flow, e.g. pressure gradient resort to LES or hybrid RANS–LES methods.
across the jet due to, for example, wind and temperature differ-
ences between two sides of the jet. Here, turbulence models should 6. Conclusions and recommendations
be able to accurately predict these physical processes as well as the
jet flow. The aim of the present study is the validation of steady RANS
CFD simulations of PTIJ flows with a jet height to width ratio of
5.3. Future work hjet /wjet = 22.5 at Re = 8000 and 13,000 based on 2D PIV experi-
mental data for the whole vertical centreplane of the PTIJ. Previous
The current study showed that RANS was not able to accurately numerical studies on PTIJ showed inconsistencies in conclusions
regarding the performance of turbulence models. Moreover, there
predict turbulence levels in the shear layer, especially near the inlet
is still a scarcity of basic studies on PTIJ at moderate Reynolds
region and in the impingement region. The latter is known to be a
numbers for a jet height larger than 10 jet nozzle widths, including
complex flow region, where the effect of anisotropy of turbulent
an evaluation of mean velocity and turbulence intensity over the
fluctuations and streamline curvature are pronounced (e.g. [28]).
entire height of the impinging jet.
The accuracy of numerical predictions in these regions will benefit
Numerically predicted distributions of mean velocity and tur-
from LES (e.g. [32]). Moreover, temporal flow features, including jet
bulent kinetic energy were compared with experimental data
flapping, development of vortical structures in the shear layers, etc.
along both the jet centreline and several cross-jet lines in lateral
can be solved by LES (e.g. [19,65]). However, LES will require high-
direction. Furthermore, the production of turbulent kinetic energy
resolution grids throughout the flow domain, which would in-
in the jet potential core region near the jet nozzle was analysed.
crease the computational demand substantially; a hybrid unsteady
The most important findings of this study regarding the applica-
RANS–LES (URANS–LES) approach might therefore be a good com-
bility of steady RANS models to predict isothermal plane turbulent
promise between computational demand and increased prediction
jets at Re = 8000 (case I) and 13,000 (case II) are:
accuracy. Hybrid URANS–LES can couple a URANS model in the
near-wall region with LES for the outer flow, reducing the need for – For both cases the best agreement with measured mean ve-
a very fine near-wall grid to resolve the small-scale eddies near locity and turbulent kinetic energy in the region near the jet
the wall when a full LES would be performed (e.g. [66]). Therefore, nozzle is achieved with SST.
future work will include an elaboration of hybrid URANS–LES – For case I, the best correspondence in potential core length
models. is provided by RNG, which underestimates the experimental
242 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243

value by 8%. For case II, RKE underestimates the experimental [5] Z. Zhai, AL. Osborne, Simulation-based feasibility study of improved aircondi-
jet potential core length by 8%, providing the closest agree- tioning systems for hospital operating room, Front. Archit. Res. 2 (2013) 468–
475.
ment with experimental data.
[6] N. Luo, R. Gao, W. Zhang, Z. Tian, An experiment and simulation of smoke
– Validation metrics show that centreline distributions of mean confinement utilizing an air curtain, Safety Sci. 59 (2013) 10–18.
velocity and turbulent kinetic energy are most accurately [7] JP. Rydock, T. Hestad, H. Haugen, JE. Skaret, An isothermal air curtain for
predicted by RNG and RKE for case I, while for case II the best isolation of smoking areas in restaurants, in: HB Awbi (Ed.), ROOMVENT
agreement with experimental data is obtained by SKE and 2000, Proc of the 7th International Conference on Ventilation for Health and
Sustainable Environment, 2000.
RNG. For both cases, the peaks of TKE in the jet impingement [8] L. Guyonnaud, C. Solliec, M. Dufresne de Virel, C. Rey, Design of air curtains
region are most accurately predicted by SST. used for area confinement in tunnels, Exp. Fluids 28 (4) (2000) 377–384.
– Based on the overall validation metrics the best performance [9] S. Maurel, C. Solliec, A turbulent plane jet impinging nearby and far from a flat
in predicting mean velocity for both cases is provided by RKE plate, Exp. Fluids 31 (2001) 687–696.
and in predicting turbulent kinetic energy by RKE and RNG. [10] A. Koched, M. Pavageau, F. Aloui, Experimental investigations of transfer
phenomena in a confined plane turbulent impinging water jet, J. Fluid Eng.-
– All the models, except RSM, provide a good agreement with Transfer. ASME 133 (6) (2011) 061204(1–13)..
the experimental values of jet spreading rate within the in- [11] N. Rajaratnam, Turbulent Jets, Elsevier Scientific, New York, 1976.
termediate jet region for both cases. [12] HB. Awbi, Ventilation of Buildings, England: Chapman & Hall, London, 1991.
– For case I, SKE and RNG accurately predict the jet decay rate [13] SH. Seyedein, M. Hasan, AS. Mujumdar, Modelling of a single confined turbu-
lent slot jet impingement using various k-ε turbulence models, Appl. Math.
within the intermediate region and for case II SKE provides
Model (18) (1994) 526–537.
the most accurate estimation as well. [14] K. Heyerichs, A. Pollard, Heat transfer in separated and impinging turbulent
– For several of the lines of analysis considered, the differ- flows, Int. J Heat Mass. Tran. 39 (12) (1996) 2385–2400.
ences in validation metrics between the considered turbu- [15] K. Chen, Y. Lawrence Yao, V. Modi, Numerical simulation of oxidation effects
lence models are negligibly small. Therefore, based on these in the laser cutting process, Int. J. Adv. Manuf. Technol. 15 (1999) 835–842.
[16] Y. Shi, MB. Ray, AS. Mujumdar, Effect of large temperature differences on
metrics, it is difficult to draw general straightforward conclu- local nusselt number under turbulent slot impingement jet, Dry Technol. 9
sions regarding the best performing RANS turbulence model (9) (2002) 1803–1825.
for predicting PTIJs. [17] TH. Park, HG. Choi, JY. Yoo, SJ. Kim, Streamline upwind numerical simulation
of two-dimensional confined impinging slot jets, Int. J. Heat Mass. Transfer.
Based on the results of this study the following recommenda- 46 (2003) 251–262.
tions can be made with respect to the performance of the different [18] M. Angioletti, E. Nino, G. Ruocco, CFD turbulent modeling of jet impingement
RANS turbulence models: and its validation by particle image velocimetry and mass transfer measure-
ments, Int. J. Therm. Sci. 44 (2005) 349–356.
– RKE and RNG are suitable for the prediction of jet flow in [19] G. Le Song, M. Prud’homme, Prediction of coherent vortices in an impinging
jet with unsteady averaging and a simple turbulence model, Int. J. Heat Fluid
general. SKE should not be used due to erroneous profiles of
Flow 28 (2007) 1125–1135.
turbulent kinetic energy on the jet centreline, especially near [20] MK. Isman, E. Pulat, AB. AEtemoglu, M. Can, Numerical investigation of tur-
the jet nozzle. bulent impinging jet cooling of a constant heat flux surface, Numer. Heat
– SKE is able to quantify the jet decay rate. Transfer. A 53 (2008) 1109–1132.
– SST is able to quantify jet spreading rate in the jet intermedi- [21] JE. Jaramillo, CD. Pérez-Segarra, I. Rodriguez, A. Oliva, Numerical study of
plane and round impinging jets using RANS models, Numer. Heat Transfer.
ate region. B 54 (3) (2008) 213–237.
– RNG and RKE are able to quantify the potential core lengths. [22] S. Rhea, M. Bini, M. Fairweather, WP. Jones, RANS modelling and LES of a
SST should not be used due to overestimation of this length. single-phase, impinging plane jet, Comput. Chem. Eng. 33 (8) (2009) 1344–
– Only SST is able to fairly accurately predict turbulent kinetic 1353.
[23] MF. Koseoglu, S. Baskaya, The role of jet inlet geometry in impinging jet heat
energy in the impingement zone.
transfer, modeling and experiments, Int. J. Therm. Sci. 49 (2010) 1417–1426.
Overall, RANS models are suitable to assess general flow charac- [24] S. Kubacki, E. Dick, Simulation of plane impinging jets with k–ω based hybrid
RANS/LES models, Int. J. Heat Fluid. F 31 (2010) 862–878.
teristics such as jet spreading rate, jet decay rate and estimating of
[25] R. Dutta, B. Srinivasan, A. Dewan, LES of a turbulent slot impinging jet to
potential core region. These quantities are relevant, for example, in predict fluid flow and heat transfer, Numer. Heat Transfer. A 64 (10) (2013)
prediction of heat and mass transfer through air curtains. However, 759–776.
steady RANS models are not suitable for basic research focusing [26] AM. Achari, MK. Das, Application of various RANS based models towards
predicting turbulent slot jet impingement, Int. J. Therm. Sci. 98 (2015) 332–
on the effect of transient flow features (jet flapping, development
351.
and advection of vertical structures, etc.), and are in general less [27] D. Benmouhoub, A. Mataoui, Heat transfer control of an impinging inclined
suitable for a detailed analysis of the impingement region. slot jets on a moving wall, Heat Transfer. Asian Res. 44 (2015) 568–584.
[28] J. Moureh, M. Yataghene, Numerical and experimental investigations on jet
Acknowledgements characteristics and airflow patterns related to an air curtain subjected to
external lateral flow, Int. J. Refrig 67 (2016) 355–372.
[29] S. Habli, NM. Saïd, G. Le Palec, H. Bournot, Numerical study of a turbulent plane
Twan van Hooff is currently a postdoctoral fellow of the Re- jet in a coflow environment, Comput. Fluids 89 (2014) 20–28.
search Foundation - Flanders (FWO) and acknowledges its financial [30] JR. Berg, SJ. Ormiston, HM. Soliman, Prediction of the flow structure in a
support (project FWO 12R9718N). The authors gratefully acknowl- turbulent rectangular free jet, Int. Commun. Heat Mass. 33 (2006) 552–563.
edge the partnership with ANSYS CFD. [31] TN. Aziz, JP. Raiford, AA. Khan, Numerical simulation of turbulent jets, Eng.
Appl. Comp. Fluid 2 (2) (2008) 234–243.
[32] S. Kubacki, J. Rokicki, E. Dick, Hybrid RANS/LES computations of plane imping-
References ing jets with DES and PANS models, Int. J. Heat Fluid F 44 (2013) 596–609.
[33] N. Zuckerman, N. Lior, Jet impingement heat transfer: physics, correlations,
[1] K. Siren, Technical dimensioning of a vertically upwards blowing air curtain— and numerical modeling, Adv. Heat. Transfer. 39 (2006) 565–631.
part I, Energ Buil. 35 (2003) 681–695. [34] A. Khayrullina, T. van Hooff, B. Blocken, GJF. van Heijst, PIV measurements
[2] AM. Foster, MJ. Swain, R. Barrett, P. D’Agaro, SJ. James, Effectiveness and of isothermal plane turbulent impinging jets at moderate Reynolds numbers,
optimum jet velocity for a plane jet air curtain used to restrict cold room Exp. Fluids. 58 (2017) 31.
infiltration, Int. J. Refrig. 29 (2006) 692–699. [35] T. van Hooff, B. Blocken, Coupled urban wind flow and indoor natural venti-
[3] T. Gil-Lopez, J. Castejon-Navas, MA. Galvez-Huerta, PG. O’Donohoe, Energetic, lation modelling on a high-resolution grid: A case study for the Amsterdam
environmental and economic analysis of climatic separation by means of air ArenA stadium, Environ. Modell. Softw. 25 (1) (2010) 51–65.
curtains in cold storage rooms, Energ Build 74 (2014) 8–16. [36] ANSYS Inc. ANSYS Fluent 15.0. User × Guide, ANSYS Inc, 2013.
[4] YC. Shih, AS. Yang, CW. Lu, Using air curtain to control pollutant spreading for [37] WP. Jones, BE. Launder, The prediction of laminarization with a two-equation
emergency management in a cleanroom, Build Environ. 46 (2011) 1104–1114. model of turbulence, Int. J. Heat Mass Transfer 15 (2) (1972) 301–314.
 A. Khayrullina, T.v. Hooff, B. Blocken et al. / European Journal of Mechanics / B Fluids 75 (2019) 228–243 243

[38] TH. Shih, J. Zhu, JL. Lumley, A new Reynolds stress algebraic equation model, [55] LWB. Browne, RA. Antonia, S. Rajagopalan, AJ. Chambers, Interaction region
Comput. Method. Appl. M 125 (1–4) (1995) 287–302. of a two-dimensional turbulent plane jet in still air, in: Structure of Complex
[39] V. Yakhot, SA. Orszag, S. Thangam, TB. Gatski, CG. Speziale, Development of Turbulent Shear Flow, Vol. 1, Dumas, R. and Fulachier, L., 1983, pp. 411–419.
turbulence models for shear flows by a double expansion technique, Phys. [56] FO. Thomas, HC. Chu, An experimental investigation of the transition of a
Fluids A 4 (7) (1992) 1510–1520. planar jet: subharmonic suppression and upstream feedback, Phys. Fluids A-
[40] FR. Menter, Two-equation eddy-viscosity turbulence models for engineering Fluid 1 (9) (1989) 1566–1587.
applications, AIAA J. 32 (8) (1994) 1598–1605. [57] Y. Sakai, N. Tanaka, T. Kushida, On the development of coherent structures in
[41] FR. Menter, Zonal two-equation k-ω turbulence models for aerodynamic a plane jet, JSME Int. J. B 49 (1) (2006) 115–124.
flows, AIAA Paper 9 (1993) 3–2906. [58] RC. Deo, GJ. Nathan, J. Mi, Comparison of turbulent jets issuing from rectan-
[42] BE. Launder, Second-moment closure and its use in modelling turbulent gular nozzles with and without sidewalls, Exp. Therm. Fluid Sci. 32 (2) (2007)
industrial flows, Internat. J. Numer. Methods Fluids 9 (1989) 963–985. 596–606.
[43] B. Blocken, LES over RANS in building simulation for outdoor and indoor [59] PR. Suresh, T. Sundararajan, SK. Das, Experimental investigation of the influ-
applications: a foregone conclusion? Build Simul (2018) http://dx.doi.org/10. ence of momentum thickness on the development of a slightly heated plane
1007/s12273-018-0459-3. jet, Int. Commun. Heat Mass. 35 (3) (2008) 282–288.
[44] PJ. Roache, Perspective: A method for uniform reporting of grid refinement [60] M. Alnahhal, T. Panidis, The effect of sidewalls on rectangular jets, Exp. Therm.
studies, ASME J. Fluids Eng. 116 (3) (1994) 405–413. Fluid Sci. 33 (5) (2009) 838–851.
[45] PJ. Roache, Quantification of uncertainty in computational fluid dynamics, [61] PA. Durbin, BA. Pettersson Reif, Statistical theory and modeling for turbulent
Ann. Rev. Fluid Mech. 29 (1) (1997) 123–160. flows, Wiley, Chichester, England, 2001.
[46] M. Casey, T. Wintergerste, ERCOFTAC best practice guidelines. European Re- [62] TJ. Craft, LJW. Graham, BE. Launder, Impinging jet studies for turbulence
search Community on Flow, Turbulence and Combustion, London, 2000. model assessment – II An examination of the performance of four turbulence
[47] PJ. Roache, Verification and validation in computational science and engineer- models, Int. J. Heat Mass. Tran. 36 (1993) 2685–2697.
ing. Science and Engineering, Hermosa, 1998. [63] S. Murakami, A. Mochida, Y. Hayashi, S. Sakamoto, Numerical study on
[48] IB. Celik, U. Ghia, PJ. Roache, CH. Freitas, Procedure for estimation and report- velocity-pressure field and wind forces for bluff bodies by k-ε , ASM and LES,
ing of uncertainty due to discretization in CFD applications, J. Fluids Eng. 130 J. Wind Eng. Ind. Aerodyn. 44 (1–3) (1992) 2841–2852.
(7) (2008) 078001–078004. [64] NG. Wright, GJ. Easom, Non-linear k–ε turbulence model results for flow over
[49] M. Schatzmann, H. Olesen, J. Franke, COST 732 model evaluation case studies: a building at full-scale, Appl. Math. Model. 27 (12) (2003) 1013–1033.
approach and results. COST Office, Brussels, 2010. [65] T. van Hooff, B. Blocken, Y. Tominaga, On the accuracy of CFD simulations of
[50] JC. Chang, SR. Hanna, Air quality model performance evaluation, Meteorol cross-ventilation flows for a generic isolated building: comparison of RANS,
Atmos. Phys. 87 (2004) 1–3. LES and experiments, Build. Environ. 114 (2017) 148–165.
[51] RC. Deo, J. Mi, GJ. Nathan, The influence of nozzle aspect ratio on plane jets, [66] J. Fröhlich, D. von Terzi, Hybrid LES/RANS methods for the simulation of
Exp. Therm. Fluid Sci. 31 (8) (2007) 825–838. turbulent fows, Prog. Aerosp. Sci. 44 (5) (2008) 349–377.
[52] E. Gutmark, M. Wolfshtein, I. Wygnanski, The plane turbulent impinging jet, [67] PA. Durbin, On the k-ε stagnation point anomaly, Int. J. Heat Fluid Flow. 17
J. Fluid Mech. 88 (1978) 737–756. (1996) 89–90.
[53] SB. Pope, Turbulent flows, Cambridge University Press, Cambridge, 2000. [68] M. Behnia, S. Parneix, PA. Durbin, Prediction of heat transfer in an axisymmet-
[54] PE. Jenkins, VW. Goldschmidt, Mean temperature and velocity measurements ric turbulent jet impinging on a flat plate, Int. J. Heat Mass. Tran. 41 (1998)
in a plane turbulent jet, J. Fluid Eng.-Transfer. ASME 95 (1973) 81–584. 1845–1855.