Sei sulla pagina 1di 11

Environmental Modelling & Software 82 (2016) 218e228

Contents lists available at ScienceDirect

Environmental Modelling & Software


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

Sensitivity of turbulent Schmidt number and turbulence model to


simulations of jets in crossflow
Daniel Valero a, b, *, Daniel B. Bung a
a
Hydraulic Engineering Section, FH Aachen University of Applied Sciences, Bayernallee 9, 52066 Aachen, Germany
b
Research Group of Hydraulics in Environmental and Civil Engineering (HECE), Dept. of ArGEnCo, University of Liege (ULg), Chemin des Chevreuils 1, Bat
B52/3 þ1, B-4000 Li
ege, Belgium

a r t i c l e i n f o a b s t r a c t

Article history: Environmental discharges have been traditionally designed by means of cost-intensive and time-
Received 13 September 2015 consuming experimental studies. Some extensively validated models based on an integral approach
Received in revised form have been often employed for water quality problems, as recommended by USEPA (i.e.: CORMIX). In this
11 April 2016
study, FLOW-3D is employed for a full 3D RANS modelling of two turbulent jet-to-crossflow cases,
Accepted 29 April 2016
including free surface jet impingement. Results are compared to both physical modelling and CORMIX to
better assess model performance. Turbulence measurements have been collected for a better under-
standing of turbulent diffusion's parameter sensitivity. Although both studied models are generally able
Keywords:
Momentum jet
to reproduce jet trajectory, jet separation downstream of the impingement has been reproduced only by
Turbulent Schmidt number RANS modelling. Additionally, concentrations are better reproduced by FLOW-3D when the proper
RANS turbulent Schmidt number is used. This study provides a recommendation on the selection of the tur-
Physical modelling bulence model and the turbulent Schmidt number for future outfall structures design studies.
Numerical modelling © 2016 Elsevier Ltd. All rights reserved.

1. Introduction can be also found in environmental flows.


In any case, proper design of these environmental flow dis-
Conventionally, design of submerged outfall structures has been charges is of paramount importance for the environmental sys-
carried out by use of cost-intensive and time-consuming experi- tems' biotic media. Traditionally, outfalls have been designed by
mental studies which allow a detailed investigation of all relevant means of physical modelling. However, due to potential scale ef-
flow parameters when model laws and possible scale effects are fects (particularly when dealing with different densities between
properly considered. Outfall structures are used to discharge ef- water body and jet) in small scale models and the mostly very large
fluents from municipal or industrial plants into rivers or the sea. space being needed to model a river reach (since vertical model
Generally, the design of these diffusors aims to enhance the mixing distortion should be avoided to properly reproduce buoyant jets;
between the effluent jet and the ambient water body in order to not Kobus, 1980), numerical methods arise as a feasible and reliable
exceed critical concentrations. Along the jet trajectory, the near tool. Hence, different simulation methods have been developed
field (with a free development of the jet) needs to be distinguished during last decades. For application to complex buoyant jets,
from the far field where boundaries may affect the jet mixing (e.g. modelling accuracy of simple non-buoyant jets must be first
due to attachment at the bottom, the free surface or the banks). In addressed.
the near field of submerged outfalls, shear flow takes place slowing Simplified mixing zone models are usually employed for the
down the jet and generating a turbulent flow. Besides neutrally- design of these environmental jets. Cornell Mixing Zone Expert
buoyant jets, negatively (e.g. due to high salinity of the jet) or System (CORMIX) is one of the most validated software packages
positively buoyant (e.g. due to high temperature of the effluent) jets used for this purpose (Jirka, 2004, 2007; Bleninger and Jirka, 2008).
CORMIX includes 1) a length scale approach to classify a flow sce-
nario and to account for boundary interaction and 2) an integral
* Corresponding author. Hydraulic Engineering Section, FH Aachen University of approach using hydrodynamic equations to fulfill conservation
Applied Sciences, Bayernallee 9, 52066 Aachen, Germany laws and determine jet trajectories and concentrations. A third
E-mail addresses: valero@fh-aachen.de (D. Valero), bung@fh-aachen.de method, i.e. passive diffusion, becomes relevant in the far field
(D.B. Bung).

http://dx.doi.org/10.1016/j.envsoft.2016.04.030
1364-8152/© 2016 Elsevier Ltd. All rights reserved.
D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228 219

which is less significant for the present study. This commercial models is still an issue (Palomar et al., 2012; Roberts and Tian,
software package has been widely used for water quality modelling, 2004) and CFD benchmarking can provide relevant information
e.g. its methodology is supported by the U.S. Environmental Pro- for future best practice guidelines (Blocken and Gualtieri, 2012).
tection Agency (USEPA, 1991a, b). However, some limitations are To the knowledge of the authors, this is the first conducted study
held in the basic assumption of the employed integral model (e.g.: analyzing sensitivity of turbulent Schmidt number using simulta-
Gaussian concentration profiles). Modelling is also limited to very neously these two turbulence models for a jet-to-crossflow case.
basic boundary conditions (e.g. rectangular channels, steady The conducted study aims to reduce uncertainty and shortcomings
ambient flows and jet discharges, 1st order description of of the employed techniques and analyze their current limitations
contaminant degradation). Bounded flows with arbitrary boundary and feasibilities for environmental discharges modelling. Despite
conditions cannot be reproduced as well as jet re-entrainment of the current study is dealing with water, the results may be of in-
contaminated recirculated water. terest for other different types of turbulent environmental flows
In this regard, other more complex techniques as Computational where transport of velocity fluctuations must be considered, e.g.
Fluid Dynamics (CFD) can fill this gap with a more general atmospheric flows, density currents or air-water flows. All de-
approach. When using CFD techniques, different approximations cisions in the selection of the model, parameters and case studies
are available depending on the turbulence scales solved (Hirsch, fulfill the recommendations of Blocken and Gualtieri (2012) for
2007). Direct Numerical Simulation (DNS) allows the full solution model development and evaluation; and consequently the ten
of Navier-Stokes equations but is restricted to cases with low iterative steps of Jakeman et al. (2006), thus ensuring a good
Reynolds numbers for smaller spatial domains given the actual disciplined model practice.
computational limitations. Large Eddy Simulations (LES) are more
affordable and compete in accuracy although it is still very 2. Physical model setup
expensive for large domain simulations as the ones involved in
typical environmental flows (i.e.: study of Ruiz et al., 2015) or when A horizontal flume of 58 cm width located at the hydraulic
complex multi-physics are involved (Gorle  and Iaccarino, 2013). laboratory of FH Aachen has been used for the physical tests (see
Also, as mentioned by Piomelli (2014), the advanced level of Figs. 1 and 2). Illumination was established using several halogen
competence required to run a LES is an obstacle to its widespread spot lights of 500 Watts each, properly located to achieve homo-
application. Hence, current engineering approach for a wide range geneous illumination. Inlet boundary condition was composed of
of complex flows is based on Reynolds-Averaged Navier-Stokes linear tubes and foam in order to ensure an undisturbed inflow
(RANS) equations as pointed out by Davidson (2015), Bradshaw condition. A conical diffusor, located at a distance of 1 m down-
et al. (1996) and Spalart (2000) and, consequently, RANS has stream the foam, was especially prepared allowing the injection
become a standard practice in CFD (Tucker and DeBonis, 2014). pipe to be set at a certain angle q0 relative to the stream flow di-
Despite existence of more complex models than the eddy viscosity rection (with an inner diameter D ¼ 3 mm, and u0 ¼ 2 m/s average
based ones (i.e.: Reynolds-Stress Transport models or RST), it seems injection velocity). The other end of the injection pipe was con-
that CFD community predicts continued use of one and two- nected to a tank - built using a closed PE pipe with 200 mm
equation turbulence models (Slotnik et al., 2014). Also, RST diameter and installed above the flume - which was pressurized by
models can lack of robustness and are occasionally less accurate means of an 8 bars air compressor in order to set a constant in-
than standard RANS models (Slotnik et al., 2014). A detailed jection velocity. The jet velocity (or jet discharge) was measured
description of RANS equations can be found in (Pope, 2000; before starting the tests by checking the volume discharged over
Ferziger and Peric, 2012). RANS equations are solved aiming to 60 s. The efflux vertical position was fixed at h0 ¼ 5 cm from the
obtain the mean velocity field. As a consequence of the Reynolds channel bed. Injection inclination q0, as defined in Fig. 3, was set to
averaging procedure, some additional assumption is needed to 45 and 90 , respectively. Photos of both model setups are shown in
account for transport of velocity fluctuations. For this turbulent Fig. 2. Efflux Reynolds number (based on efflux velocity u0, orifice
transport the so called gradient diffusion hypothesis can be dimensions D and fluid kinematic viscosity n) is of the order of
assumed being necessary the definition of the turbulent Schmidt 6000; thus being over 2000, which is commonly accepted as the
number (Gualtieri and Bombardelli, 2013; Tominaga and critical value for the jet to be independent of viscosity in a turbulent
Stathopoulos, 2007; He et al., 1999; Spalding, 1971). Further dis- ambient flow (Fischer et al., 1979; Kobus, 1980). Ambient water
cussion on this parameter can be found in section 3.2. flow rate was set to 13 l/s by use of a frequency regulator and an
Otherwise, other approaches based on Lagrangian description of electromagnetic flowmeter. Flow depth (Ha) was controlled with a
the flow can be used; however large number of particles can result downstream weir and set to 28 cm at the flow region of interest,
on excessive computational costs. Different Lagrangian approaches
for near field modelling and detailed discussion on their accuracy
and performance can be found in Israelsson et al. (2006), Suh
(2006) and Liu and Du (2003).
In this study, two neutrally-buoyant jet-to-crossflow cases
(Rajaratnam, 1976) with different discharge angle have been
analyzed by means of three different methods: 1) physical
modelling, 2) CORMIX package and 3) CFD RANS modelling using
FLOW-3D commercial code. One of the studied cases includes a jet
impingement at the free surface thereby resulting on a complex
three-dimensional flow pattern with forward, lateral and partially
reverse spreading (Bleninger and Jirka, 2008). Moreover, two
widely used eddy viscosity closures for RANS equations (namely
RNG k  є and k  u) have been employed. Results are compared to
both physical data and CORMIX results in order to assess the ac-
curacy and highlight the advantages of the employed techniques. Fig. 1. Physical model setup. Pressurized tank over the chute and inlet boundary
As pointed out by previous authors, detailed validation of near field condition at the left side of the image.
220 D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228

Fig. 2. Investigated cases during the experiments.

Fig. 3. Mesh setup and boundary conditions and ADV profiler measurement.

yielding an average ambient velocity ua ¼ 0.08 m/s. The role of measuring techniques as three-dimensional Laser Induced Fluo-
ambient flow is to gradually deflect the turbulent jet into the flow rescence (3D LIF) or Planar Concentration Analysis (PCA), reader
direction while inducing additional mixing. might be addressed to Roberts and Tian (2004) or Weitbrecht et al.
The effluent water has been dyed by adding potassium per- (2008) respectively.
manganate (dilution: 1 g per 9 l water) to allow visual inspection of
the jet trajectory, boundaries and mixing. A Canon EOS 7D camera
3. Numerical models
was used to capture the complete domain from side and top views.
Stickers were located in the bed and side of the channel every 5 cm
3.1. CORMIX
to help later location of the jet boundaries. A high-speed camera
(type: Phantom Miro M120, used sample rate: 200 Hz, duration: 5 s,
This FORTRAN program basically includes the module CorJet
resolution: 1920  1200 px) has been employed in order to gain
which implements an integral model for turbulent buoyant jets
insight in the complex hydrodynamic processes occurring in the
assuming a self-similar distribution (i.e. a Gaussian distribution) of
Zone of Flow Establishment (ZOFE) and the near field. Additionally,
different flow parameters (e.g. velocity, concentration, tempera-
a velocity and turbulence measurement has been carried out by
ture, etc.) over a given cross-section of the jet (Jirka, 2004, 2007).
means of a Nortek Vectrino ADV profiler in the location marked in
This model was originally developed for two-dimensional round
Fig. 3.
jets (Jirka and Fong, 1981). For the assumption of self-similarity, an
For image post-processing, Python open source libraries where
unbounded flow field is required and boundary interaction is
employed, i.e.: NumPy and SciPy for data handling and 2D inter-
consequently not contemplated. CorJet is thus applicable in the
polation (Van der Walt et al., 2011) and Matplotlib for results
near field of a jet or for deep water only; unstable effects such as
plotting (Hunter, 2007). Image distortion has been compensated by
recirculation are not supported. However, for free unbounded jets,
interpolating the coordinates of over 150 reference points for the
this assumption of a known parameter distribution allows an easy
top views and 25 for side views (marked with stickers, as shown in
integration yielding estimation of variables such as volume flux,
Figs. 1 and 2). Light refraction at the air-water interface has also
momentum flux or buoyancy flux. One-dimensional integral fluxes
been taken into account when reference points are not located at
can then be considered to simplify basic three-dimensional hy-
the same plane as the observed concentration field.
drodynamic differential equations. The closure of the resulting 1D
With the described imaging techniques, it is not possible to
differential equation system requires a turbulence model which is
obtain full detailed concentrations but some qualitative analysis
implemented as an empirical entrainment model describing the
can be performed. Computing transversal integrals of the concen-
rate of volume entrained from the ambient water into the jet. This
tration field allows some comparison to averaged intensity ob-
model has been previously validated by experimental studies (Jirka,
tained from the side views (as later shown in section 4.4). In this
2004). The entrainment produces an increase of transported vol-
way, it must be noted that most of the physical data from jets being
ume and therefore a decrease of concentrations. It may be noted
used for model validation has been traditionally reported as visual
that a similar module, i.e. CorSurf, implements the simulation of
information (Bleninger and Jirka, 2008). For more complex
surface discharges.
D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228 221

In order to take into account boundary interaction at the banks, Turbulence modelling allows calculating a turbulent viscosity
bottom and free surface, an additional length scale model with a as:
coupled empirical flow classification including more than 100 flow
regimes is implemented. Depending on geometrical boundary k2
conditions, e.g. vertical inclination q0 of the jet, and the comparison nt ¼ Cm
ε
of length scales of hydrodynamic effects from the jet on length
scales of the ambient flow (typically the water depth) a flow regime where Cm takes the value 0.085 for the RNG version of the k  є
is determined and additional modules are consequently run. For turbulence model (Yakhot et al., 1992). This viscosity is used as a
instance, one significant length scale for the classification of the variable fluid viscosity accounting for the effect of Reynolds stresses
present study is the jet-to-crossflow length scale on RANS equations. When using k  u, turbulent viscosity can be
defined as:
1=2.
Lm ¼ M0 ua
k
nt ¼
where M0 ¼ u0Q0, being Q0 the jet discharge. Then, based on the u
ratio Lm/Ha < 1, a flow regime of a deep layer or weak momentum is For advection flux approximation, a second order explicit
expected for both studied cases. scheme was employed including a slope limiter algorithm ensuring
However, as pointed out in section 1, CORMIX is limited to very monotonicity preserving of the advected quantities (Van Leer, 1977;
simple boundaries, i.e. rectangular channels; and steady flows. Hirsch, 2007). Resulting linear systems were solved using a Krylov
Detailed discussion on CORMIX accuracy can be found in Palomar Subspace method: Generalized Minimal Residual method (GMRES)
et al. (2012); Loya-Fern andez et al. (2012); Roberts and Tian as described by (Saad, 2003). Krylov subspace dimension values
(2004); Bleninger and Jirka (2008, 2004, 2007). The model is ranging between 8 and 12 were chosen based on the simulations'
known to perform well for simple discharges and has been similarly elapsed time performance. Iterative convergence at each time step
validated for a variety of buoyant flows. Usually, inconsistency is achieved by reducing residuals four orders of magnitude (104)
among different experimental studies is larger than the deviation of following recommendations of ASME (Celik et al., 2008).
CORMIX results (Bleninger and Jirka, 2008). As only cases with low Spatial discretization of equations is achieved by means of the
or no unstable recirculation upstream the discharge location are Finite Volume Method (Versteeg and Malalasekera, 2007). For this
investigated in the present study, CORMIX is expected to perform reason, a computational mesh must be defined. Simulations carried
optimally since assumptions of the employed integral method are out use hexahedral cells held in three different Cartesian meshes
fulfilled (Jirka, 2004). (as sketched in Fig. 3). The first mesh works as a “buffer” mesh of
As CORMIX provides jet centreline and Gaussian related physical 2 m length, in a similar way to Bombardelli et al. (2011) for the inlet
parameters, results have been analyzed using previously condition. This “buffer” mesh decreases the cell size smoothly
mentioned packages NumPy and SciPy and interpolated over a matching the domain of interest meshes with similar cell sizes.
regular grid. Then, contour surfaces have been obtained using Thus, uniform flow with ua velocity is imposed upstream and de-
Matplotlib and compared to FLOW-3D results and physical data in a velops along this 2 m coarse mesh (1 cm average cell size) repro-
comparable format. ducing a more realistic flow field (in terms of velocities and
turbulence quantities) immediately upstream the jet.
Use of such a “buffer” mesh allows the flow to self-develop,
3.2. FLOW-3D
hence reducing the dependence of the solution on the inlet
boundary conditions. Mean flow is fed with turbulence from the
RANS numerical models have been set up using FLOW-3D for jet
contours up to an equilibrium level, this level has been observed to
discharge with q0 ¼ 45 and 90 . For both cases, two different two
correspond to a turbulent kinetic energy k equivalent to a 3.2%
equations turbulence models performance has been evaluated. The
turbulent intensity computed using:
first one, i.e. RNG k  є as defined by Yakhot and Orszag (1986) and
Yakhot et al. (1992), has been chosen as it is expected to reproduce
3
flow separation better than the standard k  є (Speziale and k ¼ ðTu Ua Þ2
Thangam, 1992; Bradshaw, 1997; Kim and Baik, 2004), has shown 2
better performance also for impinging jets (Dutta et al., 2013) and Which is in agreement with Pope (2000), recommending a value
good overall performance for complex hydraulic problems (Bayon below 5% and with observations of Tachie et al. (2000) for open
et al., 2016). This model uses different values for equations' pa- channel flow (2.4e3.5%). This value is then used as a boundary
rameters and adds a new term to the є equation. The second, condition in the inlet for subsequent simulations. Similarly, є has
namely the k  u developed by Wilcox (1998, 2008) which sub- been set at the inlet using the self-developed and self-established
stitutes the equation for є by another for u ¼ є/k. This model is flow value.
known to perform superiorly for many types of flows when Mesh 1 is linked to the “buffer” mesh and uses finer cells. It is
compared to the standard k  є (Pope, 2000; Ferna ndez Oro, 2012) extended 60 cm downstream the jet injector location and 20 cm
although it has not been as widely tested as k  є model (Bradshaw upstream (mesh 1 in Fig. 3). As for the “buffer” mesh, all the width
et al., 1996). Furthermore, it has been previously noticed that k  є of the channel is also meshed. Mesh 2 is the finest mesh and is
significantly overpredicts the spreading of the round jet (Pope, located embedded into the second one but not covering all the
2000) although is known to be more insensitive to free stream width of the channel. Cell size of 1 mm is set at the injector location
values. In the question of the choice of a second variable in two- and increases smoothly up to 1.5 mm in the mesh borders, allowing
equations models, freestream sensitivity should be given high a correct definition of jet hydrodynamic features in the ZOFE.
priority (Spalart, 2000). In this study turbulence model parameters Maximum cell size in mesh 2 holds for 5 mm. However, 2 mm cells
have been used as default. For an in-depth discussion on turbulence are forced in the border with mesh 2. Transition between “buffer”
models performance, strengths and limitations, reader might be mesh and mesh 1 has shown little effect upon the jet flow and
addressed to Pope (2000), Spalart (2000), Wilcox (1998) and trajectories although cell size ratio does not hold for 1:2, as usually
Bradshaw et al. (1996). recommended to avoid partial variables interpolation. The defined
222 D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228

distances in the employed mesh are greater than the used in the (molecular and turbulent). But as far as turbulence arise, it is
previous similar study (He et al., 1999), and then little influence reasonable to assume that turbulence transport will prevail over
from mesh boundaries might be expected. molecular diffusion; hence D z Dt. Definition of this turbulent
Described meshing settings are the result of an iterative process diffusion requires an additional relation. The so-called gradient
where meshes have been structurally refined obtaining a Grid diffusion hypothesis is herein studied. Gradient diffusion hypoth-
Convergence Index (GCI) under 10% for the maximum velocity esis assumes that the advective transport of a scalar due to turbu-
decay in the near field region (and a median GCI below 1%, see lent fluctuations is related to the spatial gradient of the
Fig. 4); according to methodology recommended by ASME in Celik concentration by means of a turbulent diffusivity. When using the
et al. (2008). The GCI is a measure of the percentage the computed gradient diffusion hypothesis, definition of the turbulent Schmidt
value is away from the value of the asymptotic value (when cell size number (Sct) becomes necessary:
tends to zero). It is based on the discretization error estimation
named Richardson extrapolation, which is the most reliable nt
method to estimate numerical uncertainty (Celik et al., 2008). For Sct ¼
Dt
the q0 ¼ 90 case study it has been also studied GCI for the mini-
mum velocity in the recirculation region, which is a representative As noted by previous authors, this number plays a key role for
aspect of the flow structure and a difficult flow feature for RANS turbulent transport in RANS models (Valero et al., 2016; Gualtieri
modelling (Bayon et al., 2016). and Bombardelli, 2013; Tominaga and Stathopoulos, 2007; He
As the free surface was constant in the investigated cases its et al., 1999; Spalding, 1971). Performance assessment of more
numerical determination was thus not required. Hence, it has been complex scalar transport models can be found in (Tominaga and
considered as a symmetry boundary condition in order to speed up Stathopoulos, 2013; Rossi and Iaccarino, 2009).
the simulations. Downstream mesh 1, stagnant pressure gradient is No universal values have been accepted by the research com-
imposed (noted as pa in Fig. 3). Side and bed of the channel are munity. Values are usually distributed in the range of 0.2e1.3 with
modeled as walls. All simulations have been performed until the the specific value depending on the local flow characteristics
whole domain mass-averaged turbulent energy has shown a sta- (Tominaga and Stathopoulos, 2007). Koeltzsch (2000) deeply
tionary trend and additionally the jet boundary did not change analyzed classic experimental studies and found that values are
during some seconds. comprised between 0.5 and 0.9 for boundary layer flows although
For transport and diffusion of a contaminant C the following local value depends on height over the wall. Flesch et al. (2002)
equation can be employed: reported values ranging from 0.18 to 1.34 for different flow condi-
tions. Launder and Spalding (1974) studied values ranging from 0.2
 
vC vC v vC to 1.5. Gualtieri and Bombardelli (2013) recently analyzed suit-
VF þ vi Ai ¼ Ai D
vt vxi vxi vxi ability of values comprised between 0.8 and 1.3 and Valero et al.
(2016) showed that using values between 0.7 and 1.4 can reduce
where VF is the cell volume fraction and accordingly Ai is the cell one order of magnitude error in the dispersion width estimation
area in i direction. This formulation is coherent with FAVOR method compared to not considering any turbulent Schmidt number.
employed for the definition of obstacles in the meshed domain Gualtieri (2010) argued feasibility of Sct ¼ 1.0 based on the
(Hirt and Sicilian, 1985). Generally, D accounts for total diffusion assumption that momentum and mass turbulent diffusivities are

Fig. 4. Numerical uncertainty for case q0 ¼ 90 (top) and q0 ¼ 45 (bottom).
D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228 223

similar since both depend on the same eddies for transport. near field of the diffuser. Turbulent diffusivity is thus too high.
The early CFD study of Spalding (1971) recommended the value It is remarkable that when the turbulence intensity was
of 0.7 for a round turbulent free jet. Same value was employed in compared to data from the physical modelling obtained with a
the numerical studies of Li and Stathopoulos (1997) and Wang and Nortek Vectrino ADV profiler, the absolute maximum error (AME,
McNamara (2006). Yimer et al. (2002) used LDV to determine Sct of as described by Bennett et al., 2013) was above 2. However, this
a turbulent free round jet, reporting increasing values from 0.62 to error holds in a very small region of the flow as it is produced by the
1.1 depending on distance to the jet axis. Other interesting peak value. Otherwise, the turbulence intensity in the self-
conclusion of the study is that, whereas Sct ¼ 0.7 is the recom- developed approaching flow was in good agreement with early
mended value, Sct ¼ 0.9 (slightly higher) can compensate the experimental studies and the jet trajectories are properly repro-
inaccuracies in the calculated velocity field in the RANS simulation. duced. Similar issues with turbulence related phenomenon in RANS
Nonetheless, this kind of cancellation of errors may not be directly modelling have been extensively discussed by Slotnik et al. (2014)
generalized for all types of flows. for flow separation, bed shear stress differences in Baykal et al.
He et al. (1999) recommended Sct ¼ 0.2 for a round jet in a cross (2015), turbulent heat transfer in Dutta et al. (2013) or Reynolds
flow. Wu et al. (2001) employed similarly a value of Sct ¼ 0.25 for a stresses in Kumar and Dewan (2015). Results obtained with
thermal shock flow in a T-junction. Additionally, they based their Sct ¼ 1.4 show better agreement with experimental data. While the
decision on previous study of He et al. (1999). However, such small upper jet boundary is nearly unaffected by the change of turbulent
values could be indeed too low based on Tominaga and diffusivity, experimental data at the lower jet boundary next to the
Stathopoulos (2007, 2013) discussions. On the light of previous diffuser is better reproduced. However, given the numerical un-
authors' results, different values have been tested in this study. certainty Sct ¼ 0.9 could already be considered accurate enough,
Sct ¼ 0.7 has resulted in excessive diffusion while Sct ¼ 1.4 has not differing importantly from solution with Sct ¼ 1.4 (see Fig. 6)
shown better agreement with experimental data. Given the phys- and being closer to physically based Sct value. The turbulence
ical and numerical uncertainty (shown in Fig. 4), Sct values over 0.9 overprediction must be related to the effective value of Sct number.
might be already acceptable. Similarly, excessive turbulence peak In the far field, on the other hand, momentum becomes less
value has been obtained in the numerical model, i.e. in the ratio of predominant and passive diffusion becomes more relevant, the
1:2 compared to the conducted ADV measurement, what may turbulent Schmidt number has therefore minor effect and main
justify using higher values of Sct than the physically suggested. component of the contaminant transport is due to mean velocity
(advection).
4. Results When analyzing a jet section for different Sct numbers, effect of
this parameter is better observed. In this regard, Fig. 6 shows the
4.1. General remarks range of employed values for results location x ¼ 0.50 m. Dt ¼ 0.0 is
also represented in order to show importance of turbulent diffusion
The following sections include the results on jet mixing and on modelling. It can also be appreciated non-linearity of results to Sct
the development of concentration profiles along the jet trajectory. value, which reinforces the importance of considering a range of Sct
To evaluate the simulated jet mixing, the isosurface for a C ¼ 0.01 values in design studies to increase safety levels (i.e.: between 0.9
concentration from both, FLOW-3D and CORMIX simulations, are and 1.4).
compared to the boundaries from physical modelling. Extraction of
those jet boundaries is based on later discussion of section 4.4. 4.3. Turbulence model sensitivity
It is generally pointed out that CORMIX results accurately pre-
dict jet trajectories as far as flow remains unbounded. Conversely, it The results obtained with a constant turbulent Schmidt number
has been observed that CORMIX tends to underestimate the of Sct ¼ 1.4 but using two different turbulence models, namely RNG
physical data jet boundaries and thus the contaminant dispersion. k  є and k  u, are presented in Figs. 7e9. Fig. 7 presents the side
It must be pointed out that CORMIX originally reports the 1/ view on the jet boundaries for q0 ¼ 45 with k  u turbulence
e ¼ 37% half width of the Gaussian distribution as jet boundary; model which can be compared to the results shown in Fig. 5b (for
which is indeed a concentration changing arbitrary value. On the RNG k  є and Sct ¼ 1.4). In this case, the choice of turbulence model
other hand, visual observations may correspond to a constant is not significantly affecting the simulation results. However, for the
concentration value when homogeneous illumination is achieved. most complex modeled jet (q0 ¼ 90 ), some differences in the near
In order to ensure a better comparability of both numerical results, field arise. Turbulent diffusion is additionally higher in the k  u
the original assumption for the jet half width is disregarded and full simulation in the near field while differences become small in the
concentration field is generated using NumPy and SciPy in order to far field after the surface impingement. However, the main flow
obtain the 1% concentration isosurfaces from CORMIX results files; features of this case, i.e. the jet separation and downstream surface
which has been used as jet boundary in the following figures. spiral rollers, are better reproduced with the RNG k  є turbulence
model (Fig. 8). The onset of jet separation is well reproduced, while
4.2. Turbulent Schmidt number sensitivity the spreading sets in too far downstream with the k  u turbulence
model. Despite, for RNG k  є turbulence model the spreading is
Different turbulent Schmidt numbers have been tested, thus still higher than the physically observed. As a result, inner and outer
allowing sensitivity analysis. Fig. 5 shows the boundaries of the jet jet boundaries for both turbulence models tend to be shifted to the
for Sct ¼ 0.7, which is a typically recommended value in the sidewalls. It may be noticed that the described phenomenon is not
experimental literature. This value, based on experimental results, present in the CORMIX results, where jet separation does not take
fails to accurately reproduce turbulent diffusion. It is shown in Fig. 5 place.
that the upper boundary, directly bended by the crossflow, is well In conclusion to the above conducted sensitivity analysis, a
modeled. Indeed, the jet boundary from FLOW-3D is higher than RANS simulation with RNG k  є turbulence model and turbulent
those from the laboratory experiments and CORMIX, but differ- Schmidt number of Sct ¼ 1.4 gives reasonable results which
ences are rather small given the uncertainty involved of the compare fairly well to experimental data and CORMIX results. RNG
employed physical modelling techniques. However, the lower jet k  є turbulence model allows capturing the 3D structure of the jet
boundary is found significantly lower than the other results in the better, including the jet separation, and the Sct number can
224 D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228

Fig. 5. Side view: q0 ¼ 45 , RNG k  є turbulence model.

Fig. 6. Effect of Sct on jet spreading at a given section (x ¼ 0.50 m).

improvement in the determination of scalar concentrations might


be similar although domain of study and time of simulation are
restricted by computational costs.

4.4. Concentration profiles

In order to evaluate the concentration profiles along the jet


trajectory and to justify the comparison of physical data with
C ¼ 0.01 boundaries, a comparison of numerical results with images
from the high-speed camera as shown in Fig. 10 is carried out. For
this purpose, homogeneous illumination was achieved by placing
Fig. 7. Side view: q0 ¼ 45 , Sct ¼ 1.4, k  u turbulence model. different halogen spot lights. Then, average intensity profiles of the

Fig. 8. Side view: q0 ¼ 90 , Sct ¼ 1.4.

compensate the observed excess of turbulence in the numerical images in the ZOFE and the near field can be determined. For
model. However, this turbulence value might yield accurate mean completeness of the study, the standard deviation of these 1000
velocity profiles estimations not ensuring a satisfactory prediction images is also presented, which allows proper visualization of the
of scalar concentration as previously noticed by Rossi and Iaccarino, shear regions. In Fig. 10, it can be clearly observed the regions of
2009. Study of Ruiz et al. (2015) obtained better turbulence quan- higher intensity variation. Out of these regions, negligible con-
tities agreement using a LES approach; however accuracy centrations might be expected. As it can be observed in Fig. 10,
D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228 225

Fig. 9. Top view: q0 ¼ 90 , Sct ¼ 1.4.

Fig. 10. Zone of Flow Establishment (ZOFE) and near field of the physically modeled jets captured with a high-speed camera. Top: instantaneous intensity; middle: averaged
intensity; bottom: intensity standard deviation.

when background and jet are homogeneously illuminated, the intensity can provide some information on the jet concentrations.
pixel intensity of the background can be easily subtracted and any More specifically on the opacity of the jet, which is related to the
fluctuation on the pixel intensity at the jet is due to concentration total contaminant in the transversal direction (thus the contami-
variations. After extraction of background, remaining jet pixel nant integral, as shown in Fig. 11).
226 D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228

(2004), this initial Zone of Flow Establishment (ZOFE) is a transition


region that lacks self-similarity as the initial unsheared profiles
undergo changes in form of peripherally growing axisymmetric
mixing layers until the final jet profiles are reached. For a better
description of turbulence processes taking place at the ZOFE reader
may be addressed to Ruiz et al. (2015).
Average intensity values reported from acquired images are
linked to the opacity of the jet. In order to better compare FLOW-3D
and CORMIX results, transversal integrals of the concentration
fields have been assessed. Although it is not precise for the whole
section's concentration distribution, maximum and minimum
values location of scalar can be therefore identified in Fig. 11.
Although, exact scalar concentration values cannot be reported
since the concentration e intensity transfer function might not be
linear.
Therefore, this allowed determination of C ¼ 0.01 representa-
tiveness for results comparison with physical data observed
boundaries. However, this value might depend on light conditions
and camera properties; and consequently it should not be extrap-
olated to other studies.
Finally, scalar concentration field is shown in Fig. 12 for FLOW-
3D with Sct ¼ 1.4 and RNG k  є turbulence model. For clarity
reasons, transversal maximum values are plotted rather than the
top plane view. It can be observed the deviation from the Gaussian
profile hypothesis even for the considered cases in this study. For
comparison, CORMIX concentrations are also shown in Fig. 13.
Reader may note that when impingement takes place, CORMIX
defines a new jet yielding a superposition of both jets in a small
area.

Fig. 11. Normalized profiles along the jet centerline for q0 ¼ 90 .

Pixel intensities for q0 ¼ 90 at different elevations are shown in


Fig. 11a). As it can be observed in the average image as well, the
concentrations show some skewness in terms of a higher turbulent
diffusion downstream of the jet, and thus a non-Gaussian distri-
bution. Similar results are found by FLOW-3D in Fig. 11b) what does
not happened otherwise for CORMIX results as a consequence of
the basic hypothesis.
It must be also noted that CORMIX gives no results for the ZOFE
very close to the outfall (here up to z ¼ 0.07 m). According to Jirka
Fig. 12. FLOW-3D concentration distributions for q0 ¼ 90 , Sct ¼ 1.4.
D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228 227

Mech. 11 (3), 263e288.


Bradshaw, P., Launder, B.E., Lumley, J.L., 1996. Collaborative testing of turbulence
models. J. Fluids Eng. 118 (2), 243e247.
Bradshaw, P., 1997. Understanding and prediction of turbulent flowd1996. Int. J.
Heat Fluid Flow 18 (1), 45e54. ISSN: 0142-727X.
Celik, I.B., Ghia, U., Roache, P.J., 2008. Procedure for estimation and reporting of
uncertainty due to discretization in CFD applications. ASME J. Fluids Eng. 130
(7), 1e4.
Davidson, P., 2015. Turbulence: an Introduction for Scientists and Engineers. Oxford
University Press, USA.
Dutta, R., Dewan, A., Srinivasan, B., 2013. Comparison of various integration to wall
(ITW) RANS models for predicting turbulent slot jet impingement heat transfer.
Int. J. Heat Mass Transf. 65, 750e764.
Ferna ndez Oro, J.M., 2012. Te cnicas nume ricas en Ingeniería de Fluidos. Ed. Reverte
,
Barcelona, Espan ~ a.
Fig. 13. CORMIX mid plane concentrations for q0 ¼ 90 . Ferziger, J.H., Peric, M., 2012. Computational Methods for Fluid Dynamics. Springer
Science & Business Media.
Fischer, H.B., List, E.J., Koh, R.C.Y., Imberger, J., Brooks, N.H., 1979. Mixing in Inland
and Coastal Waters. Academic Press, New York, NY.
5. Conclusions
Flesch, T.K., Prueger, J.H., Hatfield, J.L., 2002. Turbulent Schmidt number from a
tracer experiment. Agric. For. Meteorol. 111 (4), 299e307.
In the present study, two jets to crossflow have been modeled Gualtieri, C., 2010. RANS-based simulation of transverse turbulent mixing in a 2D
geometry. Environ. Fluid Mech. 10 (1e2), 137e156.
both with FLOW-3D and CORMIX. In order to improve under-
Gualtieri, C., Bombardelli, F., 2013. In: Parameterization of the Turbulent Schmidt
standing of employed RANS techniques, sensitivity to turbulence Number in the Numerical Simulation of Open-channel Flows. XXXV IAHR
model and turbulent Schmidt number has been assessed. For this Congress.
Gorle , C., Iaccarino, G., 2013. A framework for epistemic uncertainty quantification
purpose, physical data has been collected using techniques such as
of turbulent scalar flux models for Reynolds-averaged Navier-Stokes simula-
high-speed imaging and ADV. tions. Phys. Fluids 25, 055105. http://dx.doi.org/10.1063/1.4807067.
Turbulence model has shown to influence on the determination He, G., Guo, Y., Hsu, A.T., 1999. The effect of Schmidt number on turbulent scalar
of 3D jet structure whereas turbulent Schmidt number affects, non- mixing in a jet-in-crossflow. Int. J. Heat Mass Transf. 42 (20), 3727e3738.
Hirt, C.W., Sicilian, J.M., 1985. A porosity technique for the definition of obstacles in
linearly, jet diffusion. Modelling of the transport of velocity fluc- rectangular cell meshes. In: Proc. Fourth International Conf. Ship Hydro,
tuations is of paramount importance for turbulent environmental pp. 1e19.
discharges being noticeably underpredicted jet boundaries when Hirsch, C., 2007. Numerical Computation of Internal and External Flows: the Fun-
damentals of Computational Fluid Dynamics: the Fundamentals of Computa-
only modelling advection of the mean flow. In most cases, when tional Fluid Dynamics. Butterworth-Heinemann.
using a two equation eddy viscosity model, turbulence quantities Hunter, J.D., 2007. Matplotlib: a 2D graphics environment. Comput. Sci. Eng. 9,
may allow providing accurate mean velocity distributions; however 90e95. http://dx.doi.org/10.1109/MCSE.2007.55.
Israelsson, P.H., Do Kim, Y., Adams, E.E., 2006. A comparison of three Lagrangian
turbulence quantities may result excessive when compared to the
approaches for extending near field mixing calculations. Environ. Model. Softw.
physical values. Under these circumstances turbulent Schmidt 21 (12), 1631e1649.
number must be an effective value rather than a physically based Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development
and evaluation of environmental models. Environ. Model. Softw. 21, 602e614.
one and use of values in the range of 0.9e1.4 might be encouraged;
Jirka, G.H., Fong, H.L.M., 1981. Dynamics and bifurcation of buoyant jets in crossflow.
being 0.9 closer to the physically based Sct value and 1.4 an effective J. Eng. Mech. Div. 107, 479e499.
parameter. As far field turbulence production is limited, turbulent Jirka, G.H., 2004. Integral model for turbulent buoyant jets in unbounded stratified
Schmidt number influence is reduced and hence can be calibrated flows. Part I: single round jet. Environ. Fluid Mech. 4 (1), 1e56.
Jirka, G.H., 2007. Buoyant surface discharges into water bodies. II: jet integral model.
by using only near field data. J. Hydraul. Eng. 133 (9), 1021e1036.
Furthermore, RANS approach has demonstrated to provide Kim, J.J., Baik, J.J., 2004. A numerical study of the effects of ambient wind direction
reasonable results where classical approach hypothesis failed. This on flow and dispersion in urban street canyons using the RNG keε turbulence
model. Atmos. Environ. 38 (19), 3039e3048.
more general approach can help reduce uncertainty in the design of Kobus, H., 1980. Hydraulic Modelling. Bulletin 7. German Association for Water
new outfall structures yielding better environmental policies Resources and Land Improvement, Parey, Hamburg.
fulfillment and safer discharges operation as well as allows Koeltzsch, K., 2000. The height dependence of the turbulent Schmidt number
within the boundary layer. Atmos. Environ. 34 (7), 1147e1151.
modelling of new complex environmental fluid mechanics Kumar, R., Dewan, A., 2015. Partially-averaged NaviereStokes method for turbulent
problems. thermal plume. Heat and Mass Transfer 1e13.
Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent flows.
Comput. Methods Appl. Mech. Eng. 3 (2), 269e289.
References Li, Y., Stathopoulos, T., 1997. Numerical evaluation of wind-induced dispersion of
pollutants around a building. J. Wind Eng. Ind. Aerodyn. 67, 757e766.
Baykal, C., Sumer, B.M., Fuhrman, D.R., Jacobsen, N.G., Fredsøe, J., 2015. Numerical Liu, L., Du, S., 2003. A computationally efficient particle-puff model for concentra-
investigation of flow and scour around a vertical circular cylinder. Philos. Trans. tion variance from steady releases. Environ. Model. Softw. 18 (1), 25e33.
Loya-Fern andez, A., Ferrero-Vicente, L.M., Marco-Me ndez, C., Martínez-García, E.,
R. Soc. Lond. A Math. Phys. Eng. Sci. 373 (2033), 20140104.
Bayon, Arnau, Valero, Daniel, García-Bartual, Rafael, Valle s-Moran, Francisco Jose, Zubcoff, J., Sanchez-Lizaso, J.L., 2012. Comparing four mixing zone models with
Amparo Lo pez-Jimenez, P., 2016. Performance assessment of Open FOAM and brine discharge measurements from a reverse osmosis desalination plant in
FLOW-3D in the numerical modeling of a low Reynolds number hydraulic jump. Spain. Desalination 286, 217e224.
Environ. Model. Softw. 80, 322e335. ISSN 1364-8152. http://dx.doi.org/10.1016/ Palomar, P., Lara, J.L., Losada, I.J., 2012. Near field brine discharge modeling part 2:
j.envsoft.2016.02.018. validation of commercial tools. Desalination 290, 28e42.
Bennett, N.D., Crok, B.F.W., Guariso, G., Guillaume, J.H.A., Hamilton, S.H., Piomelli, U., 2014. Large eddy simulations in 2030 and beyond. Philos. Trans. R. Soc.
Jakeman, A.J., Marsili-Libelli, S., Newhama, L.T.H., Norton, J.P., Perrin, C., A 372 (20130320). http://dx.doi.org/10.1098/rsta.2013.0320.
Pierce, S.A., Robson, B., Seppelt, R., Voinov, A.A., Fath, B.D., Andreassian, V., 2013. Pope, S.B., 2000. Turbulent Flows. Cambridge University Press.
Characterising performance of environmental models. Environ. Model. Softw. Rajaratnam, N., 1976. Turbulent Jets. Elsevier.
40, 1e20. Roberts, P.J., Tian, X., 2004. New experimental techniques for validation of marine
Bleninger, T., Jirka, G.H., 2008. Modelling and environmentally sound management discharge models. Environ. Model. Softw. 19 (7), 691e699.
of brine discharges from desalination plants. Desalination 221 (1), 585e597. Rossi, R., Iaccarino, G., 2009. Numerical simulation of scalar dispersion downstream
Blocken, B., Gualtieri, C., 2012. Ten iterative steps for model development and of a square obstacle using gradient-transport type models. Atmos. Environ. 43
evaluation applied to computational fluid dynamics for environmental fluid (16), 2518e2531.
mechanics. Environ. Model. Softw. 33, 1e22. http://dx.doi.org/10.1016/ Ruiz, A.M., Lacaze, G., Oefelein, J.C., 2015. Flow topologies and turbulence scales in a
j.envsoft.2012.02.001. jet-in-cross-flow. Phys. Fluids 27, 045101. http://dx.doi.org/10.1063/1.4915065.
Bombardelli, F.A., Meireles, I., Matos, J., 2011. Laboratory measurements and multi- Saad, Y., 2003. Iterative methods for sparse linear systems. Siam.
block numerical simulations of the mean flow and turbulence in the non- Slotnik, J., Khodadoust, A., Alonso, J., Darmofal, D., Gropp, W., Lurie, E., Mavriplis, D.,
aerated skimming flow region of steep stepped spillways. Environ. Fluid 2014. CFD vision 2030 study: a path to revolutionary computational
228 D. Valero, D.B. Bung / Environmental Modelling & Software 82 (2016) 218e228

aerosciences. NASA/CRe;2014e218178. Valero, D., Bung, D.B., Oertel, M., 2016. Turbulent dispersion in bounded horizontal
Spalart, P.R., 2000. Strategies for turbulence modelling and simulations. Int. J. Heat jets. RANS capabilities and physical modeling comparison. In: E-Proceedings of
Fluid Flow 21 (3), 252e263. the 4th IAHR Europe Congress, Liege Belgium 27e29 July 2016.
Spalding, D.B., 1971. Concentration fluctuations in a round turbulent free jet. Chem. Van Leer, B., 1977. Towards the ultimate conservative difference scheme. IV. A new
Eng. Sci. 26 (1), 95e107. approach to numerical convection. J. Comput. Phys. 23 (3), 276e299.
Speziale, C.G., Thangam, S., 1992. Analysis of an RNG based turbulence model for Versteeg, H.K., Malalasekera, W., 2007. An introduction to computational fluid dy-
separated flows. Int. J. Eng. Sci. 30 (10), 1379eIN4. namics: the finite volume method. Pearson Education.
Suh, S.W., 2006. A hybrid approach to particle tracking and EulerianeLagrangian Wang, X., McNamara, K.F., 2006. Evaluation of CFD simulation using RANS turbu-
models in the simulation of coastal dispersion. Environ. Model. Softw. 21 (2), lence models for building effects on pollutant dispersion. Environ. Fluid Mech. 6
234e242. (2), 181e202.
Tachie, M.F., Bergstrom, D.J., Balachandar, R., 2000. Rough wall turbulent boundary Weitbrecht, V., Socolofsky, S.A., Jirka, G.H., 2008. Experiments on mass exchange
layers in shallow open channel flow. J. Fluids Eng. 122 (3), 533e541. between groin fields and main stream in rivers. J. Hydraul. Eng. 134 (2),
Tominaga, Y., Stathopoulos, T., 2007. Turbulent Schmidt numbers for CFD analysis 173e183.
with various types of flow field. Atmos. Environ. 41 (37), 8091e8099. Wilcox, D.C., 1998. Turbulence Modeling for CFD. DCW industries, La Canada, CA.
Tominaga, Y., Stathopoulos, T., 2013. CFD simulation of near-field pollutant Wilcox, D.C., 2008. Formulation of the k-omega turbulence model revisited. AIAA J.
dispersion in the urban environment: a review of current modeling techniques. 46, 2823e2838.
Atmos. Environ. 79, 716e730. Wu, H., Chen, T., Luo, Y., Wang, H., 2001. Numerical simulation of transient 3-D
Tucker, P.G., DeBonis, J.R., 2014. Aerodynamics, computers and the environment. turbulent heated jet into crossflow in a thick-wall T-junction pipe. J. Therm. Sci.
Philos. Trans. R. Soc. A 372, 20130331. http://dx.doi.org/10.1098/rsta.2013.0331. 10 (1), 46e51.
United States Environmental Protection Agency (USEPA), 1991a. Technical Support Yakhot, V., Orszag, S.A., 1986. Renormalization group analysis of turbulence. I. Basic
Document for Water Quality-based Toxics Control. Office of Water, Washington, theory. J. Sci. Comput. 1 (1), 3e51.
DC.. EPA/505/2-90-001. Yakhot, V., Orszag, S.A., Thangam, S., Gatski, T.B., Speziale, C.G., 1992. Development
United States Environmental Protection Agency (USEPA), 1991b. Assessment and of turbulence models for shear flows by a double expansion technique. Phys.
Control of Bioconcentratable Contaminants in Surface Waters. Office of Water, Fluids A Fluid Dyn. 4 (7), 1510e1520 (1989e1993).
Washington, DC.. EPA-833-D94-001. Yimer, I., Campbell, I., Jiang, L.Y., 2002. Estimation of the turbulent Schmidt number
Van der Walt, S., Colbert, S.C., Varoquaux, G., 2011. The NumPy array: a structure for from experimental profiles of axial velocity and concentration for high-
efficient numerical computation. Comput. Sci. Eng. 13, 22e30. http://dx.doi.org/ Reynolds-number jet flows. Can. Aeronaut. Space J. 48 (3), 195e200.
10.1109/MCSE.2011.37.

Potrebbero piacerti anche