Downloaded from https://www.cambridge.org/core.

Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

J. Fluid Mech. (2020), vol. 892, A34.

c The Author(s), 2020. 892 A34-1
Published by Cambridge University Press
doi:10.1017/jfm.2020.207

Roughness effect in an initially laminar

channel flow

N. N. Anika1 , L. Djenidi1, † and S. Tardu2

1 School of Mechanical Engineering, The University of Newcastle, NSW 2308, Australia
2 LEGI, Université Grenoble Alpes Domaine Universitaire CS 40700, 38058, Grenoble CEDEX 9, France

(Received 3 July 2019; revised 5 February 2020; accepted 11 March 2020)

The possibility of generating and maintaining turbulence in an initially laminar

channel flow is investigated for two Reynolds numbers Reb (= ub h/ν) = 880 and
2100 (based on bulk velocity, ub and half-height of the channel, h). The study is
carried out through a direct numerical simulation based on the lattice Boltzmann
method (LBM). The channel consists of two parallel walls separated by a distance 2h
where the roughness elements are mounted on both walls. It was observed that when
the transverse square bars span half the width of the channel and are mounted in a
‘staggered’ formation, the flow becomes fully turbulent with strong similarities to fully
rough wall turbulent channel flows at much higher Reynolds number, as reported in
the literature. For example, the rough wall mean velocity profile exhibits a significant
downward shift when compared to the smooth wall one. Also, the turbulent kinetic
energy budget is similar to its counterpart in rough wall turbulent channel flows
at much higher Reynolds numbers than the present ones. It is further shown that
the present velocity spectra compared very well with that obtained in a rough wall
turbulent boundary layer. Finally, some elements of the possible physical mechanism
allowing the generation, growth and sustainability of turbulence are proposed.
Key words: turbulence simulation, turbulent transition

1. Introduction
While it is of interest to investigate the natural laminar–turbulence transition in a
flows (Henningson & Kim 1991; Tuckerman et al. 2014), it is nevertheless not evident
how to generate local turbulence in a laminar channel flow at low Reynolds numbers.
Further, even if local turbulence is generated, it is not simple, if not impossible, to
maintain it whilst it is entrained with the mean flow because of the strong dissipative
effect of the fluid viscosity. Indeed, once turbulence is generated in a low Reynolds
number channel flow, it eventually dissipates through the viscosity effects, unless
its dissipation is somewhat compensated by production of turbulent energy; this is
particularly so when the Reynolds number (Reb = ub h/ν, based on bulk velocity, ub
and half-height of the channel, h; where ν is the fluid kinematic viscosity) is below
the critical value of approximately 3000. Accordingly, the challenge to generating,

† Email address for correspondence: [email protected].

Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-2 N. N. Anika, L. Djenidi and S. Tardu

maintaining and/or enhancing turbulence without any external active input is difficult
in subcritical Reynolds number channel flows (Obot 2002; New, Lim & Luo 2006;
Lammers, Jovanović & Delgado 2011). In order to counteract the dissipative effect
of viscosity on the turbulence growth, one should provide a mechanism that not only
generates perturbations along the wall, where the effects of viscosity are the strongest,
but also allows these perturbations to grow as they are entrained with the main flow,
to eventually lead to a fully developed turbulent channel flow. Evidently, the lower
the Reynolds number, the more important is the dissipative effect of the viscosity.
This is why the task of devising strategies for achieving effective mixing in mini-
or micro-channels, such as in microfluidic applications, presents a challenge. Indeed,
in these applications, where the flow is laminar, mixing is mostly due to molecular
diffusion and thus requires a rather long time if no means is used to enhance mixing.
It is no wonder that miniaturization of channels, whose dimensions are reduced to
the extreme (of the order of micrometres), poses a challenge. In these situations,
mixing between two streams, for example, can be enhanced mainly by increasing
the contact areas (or interfaces) between the streams. This is achieved via chaotic
advection, which involves stretching and folding, e.g. Ottino (1990) and Djenidi &
Moghtaderi (2006). Ideally, if possible, one would prefer to use turbulence which,
due its three-dimensional (3-D) nature and to the presence of multiscale ‘eddies’, is
an effective means for leading to efficient and homogeneous mixing.
In this study, we investigate the possibility of triggering and maintaining turbulence
in an initially laminar low Reynolds number channel flow, using roughness elements
consisting of transverse bars. Such bars, generally spanning the entire width of the
channel, are very effective in generating and producing turbulence, at least when
the Reynolds number is at or above the critical Reynolds number (Leonardi et al.
2003; Krogstad et al. 2005). In particular, when the spacing between the bars is
approximately 8 times the diameter of the bars (Leonardi et al. 2003), the drag is
maximum, implying that mixing is also maximum. Leonardi et al. (2003) showed
clearly that when this spacing is between 7 and 16 the viscous drag is quite small
and negligible in comparison to the form drag due to the roughness elements; the
flow is in a fully rough regime. However, it is not clear whether the bars with that
particular spacing can also be effective in reducing the viscous drag in a laminar
channel flow. Also, it is not known whether they are effective at generating and
maintaining turbulence in an initially laminar channel flow at Reynolds numbers lower
than commonly reported in the literature, especially when there is no background
turbulence that can induce some instabilities in the flow to help trigger turbulence.
It is common practice in numerical simulations to add a turbulent background or
turbulence noise to the velocity field to bypass the laminar/turbulent transition regime
and reach a fully turbulent flow quickly.
Recently, Anika, Djenidi & Tardu (2018) investigated the use of a combination of
passive and active actuations in a laminar channel flow with the aim of triggering
turbulence. The passive actuation was in the form of two-dimensional (2-D) bars
spanning the entire channel width and the active control consisted of two local wall
pulsed jets, pulsed only once. The authors demonstrated that, despite the relatively
low initial Reynolds number, such a combination does actually trigger a 3-D localized
turbulence which grows into a maintained ‘pseudo’ fully rough regime, as illustrated
by the relatively negligible (averaged) viscous drag as compared to the form drag
generated by the roughness elements. The authors also showed that, for the same
Reynolds number, no turbulence could be initiated when the jets were not activated
with or without the roughnesses mounted on the channel walls, although the flow
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-3

in the case with roughnesses consisted mainly of steady secondary motions in the
canopies (spaces between bars) and a skimming two-dimensional laminar flow with
some oscillations above the canopies. It was further found that turbulence was initiated
but could not be sustained when the jets were activated without the roughnesses. The
authors argued that, while the two pulsed jets break up the spanwise symmetry of the
flow and introduce a spanwise velocity component w and a local spanwise gradient
of this velocity component, ∂w/∂z (this latter plays an important role in turbulence
generation (Tardu, Nacereddine & Doche 2008; Anika et al. 2018)), the roughnesses
helped reduced the dissipative effect of the viscosity in the near-wall region, allowing
the instability generated by the jets to grow.
The present study follows and builds on the work of Anika et al. (2018). It is
an exploratory work aimed at determining whether turbulence can be generated
and maintained in an initially laminar low Reynolds number channel flow, without
turbulence background, using 2-D transverse bars only. The spacing p between two
consecutive staggered bars (see figure 1) is taken equal to p = 8k (k is the roughness
height), with the assumption that, if turbulence is generated, then the viscous drag
would be negligible, allowing the growth and sustainability of turbulence at low
Reynolds numbers. Also, as shown by Anika et al. (2018) and since a critical feature
of transition is the break up of the spanwise symmetry with the generation of w and
its spanwise gradient, the bars are only spanning half the width of the channel and
mounted in a staggered manner (see figure 1) to induce a spanwise inhomogeneity.
The rationale for using only passive means is to avoid any energy cost invariably
required when active actuation is used.

2. Numerical procedure
2.1. Lattice Boltzmann method
A direct numerical simulation (DNS) is carried out using the lattice Boltzmann method
(LBM). Rather that solving directly the governing fluid equations (Navier–Stokes
equations), the LBM solves the Boltzmann equation on a lattice. The basic idea of
the LBM is to construct a simplified kinetic model that incorporates the essential
physics of microscopic average properties, which obey the desired (macroscopic)
Navier–Stokes equations (Frisch, Pomeau & Hasslacher 1986). With a sufficient
amount of symmetry of the lattice, the LBM implicitly solves these latter equations
with second-order accuracy. For the present calculations, each computational node
consists of a three-dimensional lattice composed of 18 moving particles and a rest
particle (lattice model D3Q19, for a developed account of LBM see for example
Chen & Doolen (1998) and Succi (2001)). The method was successfully used to
simulated both laminar (e.g. Bernsdorf et al. 1998) and turbulent flows (e.g. Djenidi
2006).
The standard lattice Boltzmann equation with the Bhatnagar–Gross–Krook (Djenidi
2006) approximation governing the time and space variations of the single-particle
distribution fi (x, t) at the lattice site x is
1
fi (x + ei 1t, t + 1t) − fi (x, t) = − ( fi (x, t) − fi (x, t)),
eq
(2.1)
τ
where τ is the relaxation time, 1t the time step, ei (= 1x/1t) is the particle velocity
eq
in the i-direction and fi is the equilibrium single-particle distribution

fi = ρωi (1 + 3(ei .u) + 29 (ei .u)2 − 32 u2 ),

eq
(2.2)
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-4 N. N. Anika, L. Djenidi and S. Tardu

1.0

0.5

0
y/h
-0.5 5

-1.0 3
-1.1
0.6 0 1 x/h
-0.6
z/h -1.1 0

F IGURE 1. Schematic of a segment of channel’s configuration with asymmetric rough

walls of staggered elements.

where ρ(= Σi fi ) is the fluid density, u(ρu = Σi fi ei ) is the local fluid velocity and ωi
are the corresponding weights (ωi = 13 for i = 0, 181 for i = 1 to 6 and 361 for i = 7 to 18;
i = 0 corresponds to the rest particle in the centre of the square lattice, i = 1, . . . , 6
correspond to the particles on the axis aligned with x, y and z and i = 7, . . . , 18 are
related to the particles on the diagonal directions).

2.2. Computational domain and boundary conditions

Figure 1 shows the schematic of the computational domain. The simulation is carried
out in a three-dimensional Cartesian grid. The grid mesh increments in the three
directions are 1x = 1y = 1z. The cross-section of the channel has 120 × 128
mesh points and the streamwise length of the channel extends up to 32 h. The
simulations were carried out at Reynolds numbers of 2096 and 880 with periodic
boundary conditions applied in the streamwise and spanwise directions. A no-slip
boundary condition is implemented at the walls of the channel using a mid-grid
bounce back scheme (Succi 2001) which is of a second-order accuracy. It was
first verified that the LBM simulation of a smooth wall laminar flow reproduced
the Poiseuille distribution (not shown here). The channel was then roughened by
mounting square bars transversally on the walls. The bars initially spanned the entire
width of the channel. However, when it was found that this configuration did not
generate turbulence, the bars were made to span only the half-width of the channel
and mounted in a staggered fashion (see figure 1). The square bars height, k, is
represented by 16 mesh points. The separation, p, between two consecutive aligned
roughness elements is p = 18k, while the distance between two consecutive staggered
elements is 8k. Table 1 summarizes the flow conditions.

3. Results
Since it was found that no turbulence was generated when the bars spanned the
entire width of the channel (see Anika et al. (2018) for a discussion of this case), we
do not present the results pertaining to this flow roughness configuration. The focus
is only on the configuration with the bars spanning half the width of the channel.
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-5

ub h uτ h 1yuτ kuτ
Reb = Resolution Reτ = 1y+ = h/k k+ =
ν ν ν ν
2093 1932 × 120 × 128 397.4 6.68 3.71 106
880 1932 × 120 × 128 122.7 2.06 3.71 35
TABLE 1. Channel flow conditions.

3.1. Flow visualizations

‘Numerical’ flow visualizations are performed using the λ2 -method (Jeong & Hussain
1995). The method is based on the eigenvalues of the symmetric 3 × 3 tensor

M ij = Σk (Sik Skj + Dik Dkj ), (3.1)

with k = 1, 2 and 3 (along the x, y and z directions, respectively) and

1 ∂ui ∂uj
 
Sij = + , (3.2)
2 ∂xj ∂xi
1 ∂ui ∂uj
 
Dij = − (3.3)
2 ∂xj ∂xi

are the symmetric and antisymmetric components, respectively, of the velocity gradient
tensor ∇u; note that in the rest of the text we use interchangeably the notation u1 , u2 ,
u3 and u, v, w, respectively. The tensor M ij has three real eigenvalues: λ1 , λ2 and λ3 .
Jeong & Hussain (1995), argued that M ij , can be used to determine the existence of
a local pressure minimum associated with vortical motion and showed that a vortex
core can be identified as any continuous region where two of the eigenvalues of M ij
are negative. If the eigenvalues are sorted such that λ3 6 λ2 6 λ1 , then any region
for which λ2 6 0 corresponds to a vortex core. This method allows following of the
development of any vortical structures in the flow.
Figure 2 shows a snapshot of λ2 structures for both Reynolds numbers. Recall that
no initial turbulent field or noise was used to trigger turbulence. The figure reveals
coherent elongated structures across the channel, which appear to evolve considerably
in a folding and stretching manner as they are transported with the mean flow. Note
the significant more populated field for the highest Reynolds number, where the
structures are much finer and shorter that in the case of the lower Reynolds number.
These structures exhibit similar features to those present in a fully turbulent channel
flow (either on a smooth or rough wall) indicating that this roughness configuration
not only generates but also maintains turbulence once generated.
Anika et al. (2018) showed that two pulsed jets shifted in a spanwise direction
activate the bypass the transition mechanism through ejection of spanwise vorticity
from the canopy (their figure 5) which is consistent with the ejection of vorticity
above arrays of cubical obstacles (Leonardi & Castro 2010). The passive bypass
transition mechanism investigated here differs from that in Anika et al. (2018) in
many respects. In order to investigate the initial turbulence generation mechanism we
carried out simulations in a non-periodic channel flow in the streamwise direction
where the width was doubled. This allows us to follow any disturbance and assess its
growth as it is entrained in a laminar regime before the flow becomes turbulent. This
is not possible with a periodic condition in x. Further, this approach is more realistic
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-6 N. N. Anika, L. Djenidi and S. Tardu

(a) Y
X
Z

(b) Y
X
Z

F IGURE 2. Flow visualization based on λ2 -contour (λ2 /λ2,max = −0.02) on staggered rough
wall at: (a) Reb = 2093, (b) Reb = 880. The section of the channel shown here is (8 6
x/h 6 20).

and more similar to experiments where the incoming flow at the inlet of the working
section is in the laminar regime. Finally, the ratio k/h was halved in comparison to
the one used in the periodic channel flow.
Figure 3 shows a plane view of the magnitude of the instantaneous vorticity, ω,
at several heights from the bottom wall: y/k = 0.5, 1, 1.5, 2, 3 and 4 and for the
distance 1 6 x/h 6 19. Starting at the plane y/k = 0.5, that is below the bar height, we
observe that, in the early stage of the flow (1 6 x/h 6 6), ω is practically zero, except
at the lateral ends of the roughness elements where a spanwise variation is seen.
This variation increases while spreading laterally as the flow progresses along the
channel. At the end of the last section (18 6 x/h 6 19), the vorticity has spread over
the whole span of the channel. This scenario is observed at all heights shown on the
figure. Note, however, that, as y/k increases, the flow remains longer in the pseudo
laminar regime. This is because the disturbances originated at the roughness level
are simply transported with the main flow. Figure 3(a,b) shows that each roughness
element induces a disturbance in the spanwise direction at its ends. Also, one can
discern patterns in figure 3(b–d) in the form of ‘pseudo-streaks’ (discussed in the
next paragraph). To better illustrate these patterns, we report in figures 4(a) and 4(b)
a single contour of the enstrophy (ω2 ) coloured by ωy , the vertical component of ω.
We can see the presence of pseudo-periodic longitudinal structures over the section of
the channel shown, although the structures become less coherent in the last section.
It seems that, at first, there are two flows in parallel, each delimited by the roughness
extremities. This is clearly visible in the first section. In this section, the flow is in
a pseudo-laminar regime, which explains the strong coherence of the longitudinal
structures. In the second (or middle) section, new structures are generated at the
roughness ends below the already established structures that are convected by the
flow. In the last section, new structures are also formed at the roughness extremities,
and they quickly interact with the existing ones, resulting in a multiplication of
structures, as seen in figure 5 where we reproduced for convenience the last section
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-7

(a)

(b)

(c)

(d)

(e)

(f)

F IGURE 3. Flow visualization at the entrance of the non-periodic channel flow (Reb =
2093). Contours of the vorticity magnitude at several heights from the bottom wall, y/k =
0.5 (a), 1 (b), 1.5 (c), 2, (d), 3 (e) and 4 ( f ). The three sections of the channel shown
here cover the range 1 6 x/h 6 19. On each image, the flow is from left to right. The
levels of the vorticity magnitude range from 0 (dark blue) to 0.03 (red).

of the three reported in figure 4. This figure shows well the interlacing and merging
of the newly created and old coherent structures.
Elongated positive and negative ∂u/∂z shear layers that are part of the wall-normal
component of vorticity ωy = ∂u/∂z − ∂w/∂x are established upon the channel entry
(figure 6). They are set up between the staggered roughness elements and slightly
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-8 N. N. Anika, L. Djenidi and S. Tardu

(a)

(b)

F IGURE 4. Flow visualization (non-periodic channel flow, Reb = 2093): enstrophy contour
coloured with the vertical component of the vorticity, ωy , for better visualization effect;
the level of the ωy -contours range from −0.002 to 0.002. (a) Perspective view; (b) top
view. Same range of x/h as in figure 3. Flow is from left to right.

F IGURE 5. Flow visualization (non-periodic channel flow, Reb = 2093): enstrophy contour
coloured with the vertical component of the vorticity, ωy ; same as in perspective view of
figure 4 but for 13 6 x/h 6 19.

oscillate in the streamwise direction before the formation of a fully developed

turbulent spot at in the zone C delimited by 14 6 x/h 6 19. In a fully developed
smooth wall turbulent flow, thin wall regions of ∂u/∂z mark the spanwise ends of the
high and low speed streaks induced by the quasi-streamwise vortices. This is not the
case in the early stages of the bypass transition process in zone A, i.e. 0 6 x/h 6 10,
wherein only a few quasi-streamwise vortices are triggered, as we will discuss later.
The ∂u/∂z shear layers in the zone A of figure 6 originate from the flow kinematics
as a result of the spanwise asymmetry between the staggered roughness elements. The
−∂w/∂x component of ωy , in turn, plays a key active role in the early stages of the
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-9

C
B

F IGURE 6. Contours of (∂u/∂z) ∗ h/ub shear layers at the entrance of the non-periodic
channel flow (Reb = 2093). The colour code is as follows: pink, −1.45; blue, 1.45. The
entry zone is divided into three parts; zone A corresponds to the early transition zone
and is delimited by 0 6 x/h 6 10; the zone B is the intermediate transition region wherein
the ensemble of transition ingredients take place (10 6 x/h 6 14) and the turbulent spot
appears in C, at x/h > 14. Same range of x/h as in figure 3.

transition process initiated in region A. The spanwise velocity fluctuations are rapidly
set up through the local pressure gradient ∂p/∂z and positive and negative ∂w/∂x
layers appear near the tips of the roughness elements at the channel entrance zone
(figure 7). The main production term of the streamwise vorticity transport equation
is the tilting of the wall-normal vorticity, which reduces to (−∂w/∂x)(∂(U + u)/∂y)
(Brooke & Hanratty 1993). There are intense ∂(U + u)/∂y local shear zones near
the tips and between the staggered roughness elements, as seen in figure 8. The
accumulation of the streamwise vorticity and the strong discontinuity, induced by the
tips of the roughness elements, constitute the necessary ingredients for the roll-up
of the ωx layers into quasi-streamwise vortices (Jiménez & Orlandi 1993). It is
clearly seen in figure 9 that the λ2 structures are sporadically triggered during the
very beginning of the bypass transition process wherein −∂w/∂x-shear layers are
concentrated. To better show the correspondence between the −∂w/∂x layers and
quasi-streamwise vortices, we zoom in on an arbitrary zone in the intermediate
region B in figures 7 and 9; we can observe such correspondences anywhere in the
whole transition zone.
The wall-normal vorticity layers are the key elements that sustain the wall
turbulence over a smooth wall, but also over the rough wall, and their suppression
brings the flow back to a completely laminar state. The linearized ωy transport
equation under these circumstances is

Dωy ∂U ∂v
=− + ν∇ 2 ωy (3.4)
Dt ∂y ∂z
(Landahl 1980, 1990). This equation is subject to algebraic growth (Schmid &
Henningson 2012). Supposing that the normal velocity remains constant over time,
the wave-type solutions of (3.4) lead indeed to a linear growth of ωy before the
limiting effect of viscosity comes into the play. Figure 10 shows that the shear
layers of −∂v/∂z are already seen at the channel entry, and are also concentrated
near the roughness tips. They are, however, less intense compared to −∂w/∂x up
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-10 N. N. Anika, L. Djenidi and S. Tardu

F IGURE 7. Contours of −(∂w/∂x) ∗ h/ub shear layers at the entrance of the non-periodic
channel flow (Reb = 2093). The colour code is as follows: pink, −1.45; blue, 1.45. Same
range of x/h as in figure 6.

F IGURE 8. Contours of local (∂(U + u)/∂y) ∗ h/ub shear layers at the entrance of the
non-periodic channel flow. (Reb = 2093) The colour code is as follows: pink, 6.78; blue,
−6.78. Same range of x/h as in figure 6.

to the start of the fully developed turbulent spot in the region B. The −∂v/∂z
patches grow and multiply in the latter and the algebraic streak growth mechanism
(Schoppa & Hussain 2002) comes also into play. The early key triggering element
of the quasi-streamwise coherent vortices generation is, however, the −∂w/∂x shear
layers. The roughness tips clearly play a role similar to that of vortex generators.
As discussed before, there are no classical streaks at the very beginning of this
process, which is different from the self-sustaining mechanism proposed by Waleffe
(1997) and Hamilton, Kim & Waleffe (1995). However, once the bypass mechanism
is achieved, and the quasi-streamwise vortices are generated, one has the set-up of
near-wall streaks of the classical self-sustaining process similar to that over a smooth
wall.
Finally, figure 11 shows λ2 /λ2,max = −0.02 at an early stage of the periodic channel
flow simulations with k/h = 0.27. As for the case k/h = 0.13 and the non-periodic
channel flow simulation, structures are generated at the edges of the roughness
elements, indicating that the turbulence generation mechanism is similar to that
discussed above. However, due to the periodic nature of the simulation, the transition
is faster.
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-11

F IGURE 9. Contours of λ2 /λ2,min = 0.0035 at the entrance of the non-periodic channel

flow (Reb = 2093). Same range of x/h as in figure 6.

F IGURE 10. Contours of local (∂v/∂z) ∗ h/ub shear layers at the entrance of the non-
periodic channel flow (Reb = 2093). The colour code is as follows: pink, −1.45; blue,
1.45. Same range of x/h as in figure 6.

Y
X
Z

F IGURE 11. Contours of λ2 /λ2,max = −0.02 at the entrance of the periodic rough wall
channel (Reb = 2093) for k/h = 0.27.

We complete this flow visualization section with images of the time averaged flow
field. Due to the relatively small number of instantaneous flow fields (approximately 50),
we followed the procedure used by Krogstad et al. (2005) to increase the number of
statistically independent samples and compute the time average. We used only the
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-12 N. N. Anika, L. Djenidi and S. Tardu

(a) (b) (c)

F IGURE 12. Time-averaged flow field in the non-periodic rough wall channel.
(a) Contours of U (ranging from 0 to U c ; c stands for centreline) in a vertical
plane and a horizontal plane just above the roughness base plane; also shown are
streamlines behind a roughness element. (b) Top view of the recirculatory motion.
(c) Top view of contours of W (ranging from −0.3U c to 0.3U c ) in an horizontal plane
at y/k = 0.5. The flow is from left to right. The last section of the non-periodic channel
(24 6 x/h 6 32) was used to carry out the averaging process.

last section of the non-periodic channel (24 6 x/h 6 32) where the flow was almost
fully developed, and exploited the streamwise periodicity of the average flow field:
statistical equivalence between two points (x, y, z) and (x + nq, y, z), where q is the
pitch (between two consecutive aligned bars) and n an integer, and with respect to
its vertical anti-symmetry about the channel middle plane.
Figure 12 shows that a recirculatory motion takes place behind the roughness
elements which extends over a distance of approximately k. As expected, this
recirculation shows variations along the roughness element, although it is symmetric
about the roughness mid-cross-section. The converging behaviour of the streamlines
indicates the presence of a non-negligible mean spanwise velocity component W at
the edges of the roughness elements, as seen in the figure. The velocity U exhibits
a spanwise variation, which is better seen in figure 13 where contours of U at
various horizontal planes are shown. The spanwise variation gradually dissipates
as the distance to the wall increases. However, even at the last plane, which is at
the half-height of the channel, U exhibits traces of the spanwise variation induced
by the roughness. This lasting effect is certainly due to the value of the ratio k/h
which is approximately 0.13. One can expect, however, that the spanwise variation
will eventually vanish as the computation domain is extended along the axis of the
channel, so completely fully developed conditions are established.
In order to demonstrate that the spanwise variation is indeed due to an insufficiently
long channel, a similar averaging procedure was applied to the periodic channel with
the larger roughness elements k/h = 0.27. The results are shown in figures 14
and 15. While the present roughness exhibits an obstacle-like behaviour, as reflected
in the large recirculation behind the roughness elements, the spanwise variation
is practically negligible for y/k > 2.5. Not surprisingly, the recirculation observed
behind these staggered roughness elements differs from that behind transverse bars
spanning the width of the channel. In this latter case the recirculation consists of a
vortical motion with its axis aligned in the spanwise direction (as can be seen in,
for example, figure 2 of Leonardi et al. (2003)), namely a spanwise vorticity Ωz .
The present recirculation is more complicated, as revealed by figures 12 and 14,
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-13

F IGURE 13. Time-averaged flow field in the non-periodic rough wall channel. Contours
of U (ranging from 0 to U c ) in horizontal planes at y/k = 0.125, 1, 2, 4 and 8, from far
left to far right; flow on each image is from left to right. The variation of U along a
white line (seen in the left image for convenience) is 100 %, 91 %, 34 %, 20 % and 14 %
for y/k = 0.125, 1, 2, 4 and 8, respectively. The last section of the non-periodic channel
(24 6 x/h 6 32) was used to carry out the averaging process.

(a) (b) (c)

F IGURE 14. Time-averaged flow field in the periodic rough wall channel and k/h = 0.27.
(a) Contours of U in a vertical plane and a horizontal plane just above the roughness
base plane; also shown are streamlines behind a roughness element. (b) Top view of the
recirculatory motion. (c) Top view of contours of W in an horizontal plane at y/k = 0.5.
The flow is from left to right. The ranges of the U- and W-contours are the same as in
figure 12.

and is primarily made of Ωz and Ωy , the vertical vorticity components. The former
dominates the centre of the bar, while the latter dominates toward the edges of the
bar. Further, the aspect ratio lb /k (lb is the bar length in the z direction) is likely
to play a role in the size and features of the recirculation region; lb /k = 15.6 in
figure 12, while lb /k = 4 in figure 14. Note the larger (and certainly with a stronger
intensity in Ωy ) lateral motion at the sides of the bar when lb /k = 4.
In the next section we attempt to quantify the turbulent flow in order to ascertain
whether it is similar to that of a fully developed turbulent channel flow, as reported in
the literature. It the following we used a time-double space (along x and z) averaging
procedure on the periodic channel flow simulation to ensure proper convergence of the
statistics. Although this averaging hides any spanwise variations, particularly near the
roughness elements, it is appropriate when the focus of the analysis is on the global
effect of the roughness on the flow, as it is in this study.
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-14 N. N. Anika, L. Djenidi and S. Tardu

F IGURE 15. Time-averaged flow field in the periodic rough wall channel and k/h = 0.27.
Contours of U (ranging from 0 to U c ) in horizontal planes at y/k = 0.125, 1, 2, 3 and
3.71 from far left to far right; flow on each image is from left to right. The ranges of
the U-contours are the same as in figure 12.

3.2. Mean velocity

The distributions of the wall-unit-normalized mean velocity (U + = U/uτ ) for both
cases of Reb with the staggered roughness elements are reported in a semi-log
representation in figure 16 as a function of y+ = yuτ /ν; the friction velocity, uτ , is
calculated via the pressure gradient as follows:
s
h(dp/dx)
uτ = , (3.5)
ρ

where ρ is the density of fluid, and p the pressure; the overbar represents time
and space averaging along the x and z directions. We also show in the figure the
distributions for our smooth wall and rough walls with the bars spanning the entire
width of the channel. Reported too, for comparison, are the fully developed smooth
wall turbulent channel flow of Moser, Kim & Mansour (1999), and the rough wall
turbulent channel flows of Krogstad et al. (2005) and Leonardi et al. (2003); the
roughness geometry used in the two latter studies are 2-D transverse bars spanning
the entire channel flow (only one wall is rough in the study of Leonardi et al.
(2003)). Only the distributions over half the heights of the channels are shown
and, for simplicity, the origin of the profiles is taken at the base of the roughness
elements. It should be recalled that the profile of Krogstad et al. (2005) represents
a time-averaged mean velocity profile measured above a roughness element, while
the profiles of Leonardi et al. (2003) represent, as for the present profiles, space and
time-averaged mean velocity distributions.
When the channel walls are smooth for the present simulations, the flow remains, as
expected, two-dimensional and laminar at both Reynolds numbers. The flow remains
also two-dimensional and laminar when the roughness elements spanning the entire
width of the channel are mounted on the walls, although steady state recirculatory
motions sit in the spaces between the roughness elements. In this latter case, the
friction velocity is increased, as reflected by the downward shift of the velocity
profile when compared to the smooth wall case. This indicates a drag increase which
stems from an increase of the wall velocity gradient at the top the roughness elements,
even though the viscous drag is slightly negative between the roughness elements due
to the recirculatory motion. The situation is significantly different when the bars span
only half the width of the channel. While the profiles exhibit an expected downward
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-15

22

20 6 z2
z1
4
18
U+
2
16
0 100 200 300 400 500
14 y+
12
U+
10

0
100 101 102 103
y+

F IGURE 16. Mean velocity profiles normalized by uτ . Smooth wall, Reb = 999: – – – –,
blue; transverse rough wall, Reb = 1000: –u–; staggered rough wall, Reb = 880: ——,
blue and Reb = 2093: ——, black. DNS smooth wall, Reh = 10 700: · · · · · · (Moser et al.
1999). Rough wall, Reb = 4200, h/k = 0.035, p/k = 8: – – – –, black (Krogstad et al.
2005); Reb = 4200, k/h = 0.2, p/k = 19: –u– – – and p/k = 8: –A– (Leonardi et al. 2003).
The straight thin solid lines represent the log law. Inset: comparison between time- and
space-averaged distribution and the ‘pseudo’ time-averaged ones for Reb = 2093; z1 and z2
mark the spanwise locations of the ‘pseudo’ time-averaged profiles.

shift, associated with an increase of the friction velocity, the shift is relatively large.
Indeed, although the Reynolds numbers of the present simulations are smaller than
those of Krogstad et al. (2005) and Leonardi et al. (2003), the downward shift is of
similar magnitude to that exhibited by the profiles of these latter studies. It should be
mentioned that Reb , which was approximately 1000 for both the smooth wall and the
wall with non-staggered roughness, increased to approximately 2000 after mounting
the staggered roughness on the walls while keeping all other parameters fixed. To
achieve the lower Reynolds number in the staggered roughened channel flow, the
velocity had to be reduced, implying that Reb for the corresponding smooth wall
channel flow would be even smaller than 1000, well within the laminar regime.
A remarkable feature exhibited by the staggered rough wall velocity profiles is that
they are similar in shape to those observed in a fully developed turbulent channel
flow. In particular, they reveal an apparent logarithmic region, albeit over a rather
shorter range of y+ than their DNS counterparts, certainly due to the lower value of
Reb achieved in the present case. The present distributions show a somewhat different
behaviour than those of Leonardi et al. (2003) as the wall is approached. Notice the
inflection point at y+ ' 15 and 40 in the present distribution for Reb = 880 and 2093,
respectively; both locations correspond to y = 0.375k, well below the roughness height.
This difference is not surprising considering the actual difference in the roughness
arrangement between the two simulations. Certainly, the spanwise inhomogeneity of
the time-averaged velocity profile near the present roughness as seen in figure 13
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-16 N. N. Anika, L. Djenidi and S. Tardu

accounts for the difference with Leonardi et al. (2003). The spanwise inhomogeneity
is well captured in the ‘pseudo’ time-average distributions computed at the same x
position, but two different spanwise locations, one above and middle of the bar, the
other shifted off the bar. These distributions are reported in the inset of figure 16
where a linear–linear scale is used; also reported for comparison only in the inset
is the time- and spaced-averaged distribution of the main figure. As anticipated from
the flow visualization in figure 15, the spanwise variation reduces as y increases; the
variation is 13 %, 3 % and 0 at y/k = 2, y/k = 3 and y/k = 3.71 (the channel centreline),
respectively.
A final interesting observation concerning the velocity profiles over the staggered
roughness elements relates to the fact that the downward shift magnitude quantified
by 1U + = Ur+ − Us+ (where Ur and Us are the rough wall and smooth wall velocities
measured within or beyond the log region of the velocity profiles) is larger for the
staggered rough wall case than that of Krogstad et al. (2005), even though Reb of the
former cases is four and two times, respectively, smaller than that of the latter study.
The relatively strong downward shift in our data in comparison to that of Krogstad
et al. (2005) may suggest that 3-D wall geometries lead to disturbances larger than
2-D wall geometries. However, the 2-D wall geometry of Leonardi et al. (2003)
exhibits a larger downward shift for the same ratio p/k = 8. It is likely that the effect
of the ratio h/k affects the magnitude of the downward shift. The present ratio is
approximately 0.26, while that of Krogstad et al. (2005) is approximately 0.03. This
effect of h/k is also reflected in the data of Leonardi et al. (2003) reported here
and for which h/k = 0.2. The large value of the present downward shift reflects the
large value of the drag, likely to be associated with the relatively large value of k/h.
Indeed, the mean velocity profiles show that ∂U + /∂y+ at the wall is practically zero
(this is very evident when a linear–linear plot is used) indicating that the viscous drag
is practically negligible. The dominance of the form drag indicates that the density,
the shape and layout of the 2-D and 3-D roughnesses certainly play roles in the
effect on the roughness on the flow. For example, the effect of the density was well
demonstrated by Leonardi & Castro (2010). Also, Bakken et al. (2005) reported that
the 2-D rough wall produced a larger downward shift in the mean velocity profile
than a wire-meshed rough wall.

3.3. Reynolds stresses

Figure 17 shows the distributions u+2 i (i = 1, 2 and 3) for Reb = 880 and Reb = 2093.
Due to the symmetry of the distributions across the channel height, only the part of
the distributions across the lower half-height is reported; the origin y/h = 0 is taken
at the base of the roughness elements. Also reported for comparison are the DNS
distributions for the smooth wall channel flow of Moser et al. (1999). Near the wall,
the streamwise Reynolds stress u+2 1 is the largest component, while the wall-normal
+2
Reynolds stress u2 is the smallest. Notice that the maximum for each distribution
is reached at different heights. These maxima are reached at y/h ' 0.07, 0.19 and
0.33 for u+2 +2
3 , u1 and u+22 , respectively. Of interest, u3
+2
exhibits a clear hump at
+2
approximately the same height where u2 is maximum. We also observe that while the
magnitude of the Reynolds stresses decreases as Reb decreases, the maximum of u+2 3
appears to present the stronger reduction, which is indicative of a reduced spanwise
activity within the roughness vicinity; u+2 1 shows the lowest reduction. The behaviour
of u+2
3 below the location y/k = 1 departs in a major manner from that of u+2 +2
1 and u2 .
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-17

5
ui„2+

0
10-2 10-1 100
y/k = 1
y/h

F IGURE 17. Present distributions of u0+2 i (i = 1, 2 and 3) for Reb = 880 (lines with solid
circles), Reb = 2093 (solid lines). Solid lines: present simulations. Dashed, dotted and dash-
dotted lines: DNS smooth wall (Moser et al. 1999). Blue, u0+2 0+2 0+2
1 ; red, u2 ; black, u3 . The
origin y/h = 0 for the rough wall data is the base wall and the location y/k represents
the location of the roughness crest plane.

Indeed, both u+2 1 and u+22 decrease monotonically from their maximum to reach zero
at the wall while, u+2 3 decreases slightly after a local maximum at y/h = 0.33 then
increases, reaches a maximum and finally decreases. The near-wall behaviour of u+2 3
is likely to reflect a strong spanwise activity caused by the present roughness elements,
as anticipated from the flow visualizations. No such behaviour is observed in the case
of 2-D transverse bars spanning the entire width of the channel (Ikeda & Durbin 2007;
Burattini et al. 2008).
The comparison between the smooth wall data and the present ones shows
differences in behaviour and magnitude. For example, the maximum of u+2 1 is
significantly reduced on the rough wall. This is also observed in the data of Ikeda
& Durbin (2007), Burattini et al. (2008) and Krogstad et al. (2005). Worthy to note
are the levels of u+2 3 in the region y/k 6 1 and that of u+2
2 above the plane y/k = 1
in comparison to the smooth wall data. These levels indicate that the increase in u+2 3
is related to the activity taking place at the sides of roughness while the increase
of u+2
2 is associated with an upward deflection of the flow caused by the roughness.
However, one can expect that the magnitude of these increases to be controlled by
the ratios k/h and p/k.
Similar observations can be made when we compare the present Reynolds stress
distributions with those of Krogstad et al. (2005). For example, we report in figure 18
the distributions for Reb = 2093 and those of Krogstad et al. (2005). Recalling that
the latter distributions are only time averaged and measured above a roughness
element, the values of u0+2 , v 0+2 and w0+2 for these distributions must be zero at
y/h = 0.034, the location of the roughness crest plane; caution must be taken in
analysing the differences between the two simulations. While the present distributions
show similarities with those of Krogstad et al. (2005), there are noticeable differences.
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-18 N. N. Anika, L. Djenidi and S. Tardu

ui„2+ 4

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
y/k = 1 y/h

F IGURE 18. Distributions of u0+2 , v 0+2 and w0+2 for Reb = 2093 (solid lines). DNS rough
wall data of Krogstad et al. (2005): dashed lines. Blue, u0+2 0+2 0+2
1 ; red, u2 ; black, u3 . The
origin y/h = 0 for the rough wall data is the base wall and the location y/k = 1 represents
the location of the roughness crest plane for the present simulation.

For example, the present distribution of u0+2 exhibits a larger maximum than that of
Krogstad et al. (2005). Also, in the present case v 0+2 is larger than that of Krogstad
et al. (2005) for y/k > 1. Considering that Reb for Krogstad et al. (2005) is more
than twice the present one, these differences in magnitudes of u0+2 and v 0+2 are
certainly related to the difference in the ratio k/h between the two simulations. The
ratio k/h for the present simulation is approximately 0.25 while it is approximately
0.034 for Krogstad et al. (2005). Note, however, (as inferred from figure 17) that
the magnitudes of the Reynolds stresses are smaller for the present simulation when
Reb = 880 than those of Krogstad et al. (2005), showing that at this Reynolds
number the roughnesses are less effective than at the larger one for producing strong
turbulence level. This is expected as the lower the Reynolds number, the stronger the
damping effect of the viscosity.
A comment is warranted regarding the relatively large value of k for the present
simulation. One may consider that the roughness is akin to an obstacle rather than
a roughness. However, as seen above, the distributions of the mean velocity and the
Reynolds stresses are similar to those observed in a rough wall channel flow. Perhaps
the combination of using bars spanning only half of the width of the channel and
the present spacing between the bars helps reduce the blockage effect and allows the
flow to resemble that of a channel flow over a rough wall; hence, one can refer to
these bars as roughness elements rather than obstacles. This resemblance to a fully
developed channel flow is further illustrated by the Reynolds shear stress distributions
−u+ v + shown in figure 19. Such distributions should present a linear variation across
the channel height for a fully developed channel flow, as illustrated by the distribution
of Moser et al. (1999) and Krogstad et al. (2005) also reported in the figure. Such
behaviour is well observed in the present distributions for both values of Reb .
The data of figures 16, 18 and 19 suggest a fully developed turbulent channel
flow. This is further supported by figure 20, showing the correlation coefficient
ρuv = −u+ v + /(u0 v 0 ). Indeed, in the region 0.2 6 y/h 6 0.6, ρuv falls in the range
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-19

0.9

0.8

0.7

0.6

0.5
-u+√+

0.4

0.3

0.2

0.1

0 0.2 0.4 0.6 0.8 1.0

y/h

F IGURE 19. Distributions of −(u+ v + ). Symbols: present simulation, Reb = 880 (blue) and
Reb = 2093 (red). Solid line: smooth wall (Moser et al. 1999). Dashed line: rough wall
(Krogstad et al. 2005). The origin y/h = 0 is taken at the roughness crest plane.

0.35–0.41, which is the range reported in the literature for fully developed turbulent
channel flows and turbulent boundary layers. Also, the present ρuv exhibits the
expected Reynolds number dependence: ρuv decreases as Reτ increases. Reported
for comparison in the figure is the ρuv -distribution of Krogstad et al. (2005), whose
values of ρuv in the region 0.2 6 y/h 6 0.4 exceed 0.4.

3.4. Energy budget

The kinetic energy budget can be written as
Dk
= P + T + D + Π − , (3.6)
Dt
where P = −uv(dU/dy) is the turbulence production, T = −∂(vk)/∂y is the turbulent
diffusion, D = ν(∂ 2 k/∂y2 ) is the viscous diffusion, Π = −∂(vp/ρ)/∂y is the ‘pressure
diffusion’ and  = 2ν(∂ui /∂xk )(∂uj /∂xk ) is the dissipation. In this equation, the terms
have been averaged in time and in both the streamwise and spanwise directions.
Also, the dispersive effects, which arise due to the non-uniform distribution of
mean quantities in the streamwise and spanwise directions, are not accounted for in
(3.6). These effects are mostly important in the vicinity of the roughness elements
(Leonardi et al. 2006) where the streamwise non-uniformity is important. Note
that, for the present rough wall, there is a further non-uniformity in the spanwise
direction in the vicinity of the roughness. Thus, in the following discussion, it
should be kept in mind that the present energy budgets do not account for any local
streamwise and spanwise variations in the mean turbulent kinetic energy. To account
for these variations requires time averaging only, which would necessitate running
the simulations over an impractical extremely long time to ensure convergence of
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-20 N. N. Anika, L. Djenidi and S. Tardu

0.5

0.4

0.3
-u√/u„√„

0.2

0.1

0 0.2 0.4 0.6

y/h

F IGURE 20. Correlation coefficient ρuv = uv/(u0 v 0 ). Symbols same as figure 19. Line:
rough wall (Krogstad et al. 2005). The origin y/h = 0 is taken at the roughness crest
plane.

the various terms in (3.6). Nevertheless, the use of the time–space-average procedure
allows comparisons with the time–space-averaged energy budget of Ikeda & Durbin
(2007) and Leonardi et al. (2006).
Figure 21 shows the distributions of each term in (3.6), except the advection term
(left side of (3.6)) which is nil, for Reb = 880 and 2093. The energy budgets for
the two Reynolds numbers, present similar features. Both P and  are maximum
near the roughness crest plane and decrease as the distance to the wall increases.
In the central region of the channel P becomes negligible and  is balanced by the
turbulent diffusion, T, which too exhibits a maximum near the roughness elements.
Note that,  = T at y/h = 1 (the channel centreline) as in a smooth channel flow
(Hoyas & Jiménez 2006) and as it should be. Also near the roughness, Π increases
significantly while it is practically zero in the central region of the channel. The
viscous diffusion, D, is negligible across the channel, except near the roughness. In
the region below the roughness crest, the balance changes. At the base (y = 0),  is,
as expected, practically balanced by D. Notice that, just below the roughness crest,
Π has changed sign and reached a minimum. As the bottom wall is approached it
reverses sign again, reaches a maximum then drops; T first drops practically to zero,
below the roughness crest, before reaching a maximum near the base and drops again
to zero at the base. Figure 21 shows clearly that the flow around the roughness is
dynamically controlled by the roughness. Figure 21(a) shows also the balance of the
terms in (3.6). This balance represents the contribution from the dispersive effects.
These are large in the vicinity of the roughness and dissipate quickly as the distance
to the wall increases above the roughness elements. Similar results are observed for
Reb = 2093.
The present budgets are remarkably similar to that of Ikeda & Durbin (2007)
obtained in a turbulent rough wall channel flow with transverse bar k/h = 0.11 and
p = 9 spanning the entire width of the channel (figure 22). The similarity between the
present budgets and that of Ikeda & Durbin (2007) strongly suggests that the present
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-21

(a)
10

-5

-10

-15
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Roughness top plane

(b) 12
P
10
T
8
D
6 Ô
4 ´

2
0
-2
-4
-6
-8
-10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Roughness top plane

F IGURE 21. Turbulent kinetic energy budget, normalized by u3τ /h, as a function of y/h;
Reb = 880 (a) and 2093 (b). Solid line in the top figure: balance of the budget terms
in (3.6).

turbulent flow shares the same characteristics as those of fully developed turbulent
channel flow over a 2-D rough wall. Notice thought that the largest difference between
budgets in the Π-term. The present one is much larger than that of Ikeda & Durbin
(2007).
We report in figure 23 the same Reb -distributions as in figure 21, but only in the
region near the roughness crest and compare them with the DNS results of Leonardi
et al. (2006). These authors carried out a direct numerical simulation of a rough wall
turbulent channel flow at Reb = 7000. The roughness consisted of square transverse
bars mounted on one wall of the channel only and spanning the entire width of
the channel, with k/h = 0.1; several different spacings w between the bars were
investigated. Despite the differences in the Reynolds number, the roughness geometry
and, more importantly, the values of k/h between the simulation configurations, there
is an acceptable correspondence between the results in the region 50 < y+ < 100, where
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-22 N. N. Anika, L. Djenidi and S. Tardu

Production
0.06 Dissipation
Pressure
0.04 Turbulent
diffusion

Gain
Viscous
0.02 diffusion
Uneven
0

-0.02
Loss

-0.04

-0.06

0 1 2 3 4
y/h

F IGURE 22. Reproduction of the turbulent kinetic energy budget near the rough wall of
Ikeda & Durbin (2007). The budget terms are normalized by u4τ /ν; see Ikeda & Durbin
(2007) for details on the various symbols.

0.1 0.10
T+
0.08
0.06 ´+
0.08 0.04 D+
0.02
Ô+
0.06 -0.02
0
P+
-0.04 ´+
0.04 -0.06
-0.08
P+
T+
-0.10
0.02 0 0.2 0.4 y/k 0.6 0.8 Ô+
D+
0

-0.02

-0.04

-0.06

0 10 20 30 40 50 60 70 80 90 100
+
y

F IGURE 23. Turbulent kinetic energy budget for Reb = 2093, normalized by u4τ /ν, as
function of y+ = yuτ /ν in the vicinity of the lower wall; y = 0 represents the top plane
of the roughness. Lines with symbols: present simulation (same as in figure 21); symbols
only: simulation of Leonardi et al. (2006), k/h = 0.1, p/k = 7 and Reb = 7000. Inset: same
as main figure but plotted as function of y/k.

there is an approximate equilibrium between the production and the dissipation; this
is well illustrated in the inset of figure 23, which shows the same data as the main
figure but as a function of y/k. It should be recalled that the equilibrium zone P = 
extends from y+ = 30 to y/h = 0.6 in smooth turbulent channel flows and it is exactly
reached only at infinite Reynolds numbers. Further, this balance between P =  leads
to the logarithmic velocity distribution (Tardu 2013). This appears to be the case here
too, as suggested by the mean velocity profile (figure 16) where a nascent logarithmic
region is visible. Large differences are, however, observed near the top plane of the
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-23

104

102

Eu*(k*)
100 Reb = 2093, ´iso
Reb = 2093, ´actual
Reb = 880, ´iso
10-2 Reb = 880, ´actual
y/∂ = 0.4, R¬ = 133; TBL

10-4 -3
10 10-2 10-1 100
*
k

F IGURE 24. Kolmogorov-normalized velocity spectra at the channel centreline for Reλ ≈
63 (Reb = 2093) and Reλ ≈ 32 (Reb = 880). The symbols correspond to the velocity
spectrum measured in a rough wall turbulent boundary layer. The asterisk represents
Kolmogorov normalization, using either the isotropic expression, iso , or the actual one,
actual , for calculating .

roughness in the present configuration compared to Leonardi et al. (2006). The

production P+ at y+ = 0 (y = 0 is taken the roughness crest) is for instance six times
smaller here. There are several reasons for these differences. Perhaps the major one
relates to the dispersive effects. As mentioned earlier, the non-homogeneity in the
spanwise direction is likely to result in a more important dispersive effect than that
of Leonardi et al. (2006). Consequently, the pressure term peaks at around y+ = 18
and relaxes slowly after. Note that the Π -term contribution of Leonardi et al. (2006)
is larger than that of Ikeda & Durbin (2007). Further, the roughness height ratio k/h
is three times larger in this investigation than in Leonardi et al. (2006). One can thus
expect that the combination of the spanwise inhomogeneity and larger ratio k/h leads
to some differences in the energy budget around the roughness region. The maximum
production is P+ = 0.25 over a smooth wall. It decreases to 0.08 in Leonardi et al.
(2006) and to 0.05 in the case k+ = 100 for Ikeda & Durbin (2007), which is close
to the value we have here but where k/h = 0.12 is three times smaller. The roughness
parameters that play a critical role in the weakening of production near the top plane
are not clearly identified in the literature, but large k/h values lead to a sluggish P+ .

3.5. Velocity spectra

We have seen above that the addition of bars in a staggered manner in a laminar
channel flow led to a fully developed turbulent channel flow even though Reb is
well below the critical Reynolds number. It is then of interest to ascertain how
the energy is distributed among the scales of motion or, in other words, ascertain
what is the velocity spectrum exhibited by this turbulence. Figure 24 shows the
longitudinal velocity spectrum at the channel centreline for both Reb . The spectra
are Kolmogorov normalized (the asterisk represents Kolmogorov normalization). To
obtain the Kolmogorov length scale, η = (ν 3 /)1/4 , and velocity scale, η = (ν)1/4 , we
calculate , the mean turbulent kinetic energy dissipation, using its isotropic surrogate,
i.e.  iso = 15ν(du/dx)2 and its actual expression. Note that a time series of u was used
to compute the spectra and ; the Taylor hypothesis to convert the time derivative
into a spatial derivative was also used. Also reported in the figure for comparison
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-24 N. N. Anika, L. Djenidi and S. Tardu

is the Kolmogorov-normalized velocity spectrum measured in a rough wall turbulent
boundary layer at y/δ = 0.4 where the Taylor microscale Reynolds number Reλ is
equal to approximately 133;  iso was used to normalize the spectrum (details of the
measurements can be found in Kamruzzaman et al. (2015)).
Firstly, we observe that the present spectra normalized using  iso , collapse well
with the boundary layer spectrum in the range 0.03 6 k∗ 6 0.9. While the use of
 iso for normalization purposes may not be strictly correct for wall shear flows, it
is nevertheless appropriate to compare the spectra using the same normalization
parameter,  iso here. The collapse of the spectra in the high range of k∗ is consistent
with the results of Antonia, Djenidi & Danaila (2014), who showed that the
Kolmogorov-normalized spectra obtained in various turbulent flows collapse very
well in the dissipative region for Reλ as low as 30, thus validating Kolmogorov 1st
similarity hypothesis (Kolmogorov 1941), although Djenidi et al. (2014) showed that
this similarity breaks down for Reλ 6 20.
Secondly, for the present simulation, we see that spectra normalized using  iso
collapse relatively well with the ones normalized using  actual , the actual expression
of , albeit the collapse is less remarkable for Reb = 880 than Reb = 2083. This result
suggests that the local isotropy is better satisfied for the higher Reynolds number.
It is likely that the inhomogeneity affects more significantly the small scales at the
lower Reynolds number than the higher one.
Finally, the results of figure 24, which shows a continuous distribution of energy
among all scales of motion, confirms that the turbulence characteristics generated
in this relatively low Reynolds number channel flow are similar to those of a fully
developed turbulence albeit at a low Reλ .

4. Concluding remarks
A direct numerical simulation based on the lattice Boltzmann method is carried
out in a rough wall channel flow initially in a laminar regime. Two (bulk velocity)
Reynolds numbers are used, Reb = 880 and 2093. Although not shown here, it is
found that, when the roughness elements consisted of transverse bars of square
cross-section spanning the entire width of the channel with p = 8k and h/k = 3.71,
the flow remains laminar, with a steady low intensity vortical motion taking place
between the roughness elements. Only when the roughness elements spanned half the
channel width (in a staggered fashion) is turbulence generated and maintained for
both cases of Reb . The turbulent flow in the channel exhibits similar characteristics to
a fully developed turbulent channel flow on a rough wall. For example, the roughness
function, measured as a downward shift of the mean velocity profile in comparison
to that of a smooth wall, is comparable to that observed in fully rough turbulent
channel flows at much higher Reynolds numbers. Even the turbulent kinetic energy
budget is similar to those observed in the direct numerical simulations of a fully
rough turbulent channel flow reported in the literature. Finally, the velocity spectra
for both Reb confirmed that the turbulence produced in the channel is relatively well
developed. Indeed, even though the Taylor microscale Reynolds number is low (32
and 63, respectively), the present spectra collapse well with a velocity spectrum
measured in a rough wall turbulent boundary layer.
The present results answer positively the initial question raised in the introduction:
can turbulence be generated and maintained in an initially laminar low Reynolds
number channel flow, without a turbulent background, using 2-D transverse bars? It
appears that the critical factor that helps produce turbulence at low Reynolds number
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

Roughness effect in an initially laminar channel flow 892 A34-25

is the non-homogeneity induced in the flow by the present roughness. The staggered
configuration of the bars induces local spanwise symmetry break ups in the flow
which, in turn, lead to the generation of a spanwise velocity component w and local
zones of streamwise gradient of this component. The latter is one of the components
of the wall-normal vorticity and enters directly into the transport equation of the
streamwise vorticity (Tardu et al. 2008). Resulting thin shear layers are subsequently
tilted by the mean shear to give birth to new coherent structures. This scenario is
reminiscent of that induced by the use of two pulsed jets in the 2-D rough wall
channel flow study of Anika et al. (2018). Once the turbulence is generated it is
sustained by at least two mechanisms. One is related to the fact that the dissipative
effect of the viscosity in the near-wall region is dramatically weakened by the choice
of the roughness element spacing. The second is associated with the extra energy
production associated with the shedding of the shear layers taking place at roughness
trailing edge. The balance of these two mechanisms is positive, resulting in a net
energy production, allowing the turbulence to be sustained in a low Reynolds number
channel flow.
Of course, one issue worth investigating but which is out of the scope of this paper
is whether similar results can be obtained for smaller ratios k/h and various roughness
spacing. Regarding the ratio k/h and keeping the same spacing p = 8 between two
consecutive staggered roughness elements, it was found that even when the ratio is
reduced by a factor two turbulence could be generated.

Declaration of interests
The authors report no conflict of interest.

REFERENCES

A NIKA , N. N., D JENIDI , L. & TARDU , S. 2018 Bypass transition mechanism in a rough wall channel
flow. Phys. Rev. Fluids 3 (8), 084604.
A NTONIA , R. A., D JENIDI , L. & DANAILA , L. 2014 Collapse of the turbulent dissipation range on
Kolmogorov scales. Phys. Fluids 26, 045105.
BAKKEN , O. M., K ROGSTAD , P.-Å., A SHRAFIAN , A. & A NDERSSON , H. I. 2005 Reynolds number
effects in the outer layer of the turbulent flow in a channel with rough walls. Phys. Fluids
17 (6), 065101.
B ERNSDORF , J., Z EISER , T., B RENNER , G. & D URST, F. 1998 Simulation of a 2D channel flow
around a square obstacle with lattice-Boltzmann (BGK) automata. Intl J. Mod. Phys. C 9
(08), 1129–1141.
B ROOKE , J. W. & H ANRATTY, T. J. 1993 Origin of turbulence-producing eddies in a channel flow.
Phys. Fluids A 5 (4), 1011–1022.
B URATTINI , P., L EONARDI , S., O RLANDI , P. & A NTONIA , R. A. 2008 Comparison between
experiments and direct numerical simulations in a channel flow with roughness on one wall.
J. Fluid Mech. 600, 403–426.
C HEN , S. & D OOLEN , G. D. 1998 Lattice Boltzmann method for fluid flows. Annu. Rev. Fluid
Mech. 30 (1), 329–364.
D JENIDI , L. 2006 Lattice-Boltzmann simulation of grid-generated turbulence. J. Fluid Mech. 552,
13–35.
D JENIDI , L. & M OGHTADERI , B. 2006 Numerical investigation of laminar mixing in a coaxial
microreactor. J. Fluid Mech. 568, 223–242.
D JENIDI , L., TARDU , S. F., A NTONIA , R. A. & DANAILA , L. 2014 Breakdown of Kolmogorov’s
first similarity hypothesis in grid turbulence. J. Turbul. 15, 596–610.
Downloaded from https://www.cambridge.org/core. Tsinghua University, on 14 Jul 2021 at 09:59:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/jfm.2020.207

892 A34-26 N. N. Anika, L. Djenidi and S. Tardu

F RISCH , U., P OMEAU , Y. & H ASSLACHER , B. 1986 Lattice-gas automata for the Navier–Stokes
equation. Phys. Rev. Lett. 56 (14), 1505–1508.
H AMILTON , J. M., K IM , J. & WALEFFE , F. 1995 Regeneration mechanisms of near-wall turbulence
structures. J. Fluid Mech. 287, 317–348.
H ENNINGSON , D. S. & K IM , J. 1991 On turbulent spots in plane Poiseuille flow. J. Fluid Mech.
228, 183–205.
H OYAS , S. & J IMÉNEZ , J. 2006 Scaling of the velocity fluctuations in turbulent channels up to
Reτ = 2003. Phys. Fluids 18 (1), 011702.
I KEDA , T. & D URBIN , P. A. 2007 Direct simulations of a rough-wall channel flow. J. Fluid Mech.
571, 235–263.
J EONG , J. & H USSAIN , F. 1995 On the identification of a vortex. J. Fluid Mech. 285, 69–94.
J IMÉNEZ , J. & O RLANDI , P. 1993 The rollup of a vortex layer near a wall. J. Fluid Mech. 248,
297–313.
K AMRUZZAMAN , M., D JENIDI , L., A NTONIA , R. A. & TALLURU , K. M. 2015 Scale-by-scale energy
budget in a turbulent boundary layer over a rough wall. Intl J. Heat Fluid Flow 55, 2–8.
K OLMOGOROV, A. N. 1941 The local structure of turbulence in incompressible viscous fluid for
very large Reynolds number. Dokl. Akad. Nauk SSSR 30, 301–305; (see also Proc. R. Soc.
Lond. A (1991), 434, 9–13).
K ROGSTAD , P.-Å., A NDERSSON , H. I., BAKKEN , O. M. & A SHRAFIAN , A. 2005 An experimental
and numerical study of channel flow with rough walls. J. Fluid Mech. 530, 327–352.
L AMMERS , P., J OVANOVI Ć , J. & D ELGADO , A. 2011 Persistence of turbulent flow in microchannels
at very low Reynolds numbers. Microfluid. Nanofluid. 11 (2), 129–136.
L ANDAHL , M. T. 1980 A note on an algebraic instability of inviscid parallel shear flows. J. Fluid
Mech. 98 (2), 243–251.
L ANDAHL , M. T. 1990 On sublayer streaks. J. Fluid Mech. 212, 593–614.
L EONARDI , S. & C ASTRO , I. P. 2010 Channel flow over large cube roughness: a direct numerical
simulation study. J. Fluid Mech. 651, 519–539.
L EONARDI , S., O RLANDI , P., D JENIDI , L. & A NTONIA , R. A. 2006 Guidelines for modeling a 2D
rough wall channel flow. Flow Turbul. Combust. 77 (1–4), 41–57.
L EONARDI , S., O RLANDI , P., S MALLEY, R. J., D JENIDI , L. & A NTONIA , R. A. 2003 Direct
numerical simulations of turbulent channel flow with transverse square bars on one wall.
J. Fluid Mech. 491, 229–238.
M OSER , R. D., K IM , J. & M ANSOUR , N. N. 1999 Direct numerical simulation of turbulent channel
flow up to Reτ = 590. Phys. Fluids 11 (4), 943–945.
N EW, T. H., L IM , T. T. & L UO , S. C. 2006 Effects of jet velocity profiles on a round jet in
cross-flow. Exp. Fluids 40 (6), 859–875.
O BOT, N. T. 2002 Toward a better understanding of friction and heat/mass transfer in microchannels
– a literature review. Microscale Therm. Engng 6 (3), 155–173.
OTTINO , J. M. 1990 Mixing, chaotic advection, and turbulence. Annu. Rev. Fluid Mech. 22 (1),
207–254.
S CHMID , P. J. & H ENNINGSON , D. S. 2012 Stability and Transition in Shear Flows, Applied
Mathematics Sciences, vol. 142. Springer.
S CHOPPA , W. & H USSAIN , F. 2002 Coherent structure generation in near-wall turbulence. J. Fluid
Mech. 453, 57–108.
S UCCI , S. 2001 The Lattice-Boltzmann Equation. Oxford University Press.
TARDU , S. 2013 Statistical Approach to Wall Turbulence. John Wiley and Sons.
TARDU , S., NACEREDDINE , R. & D OCHE , O. 2008 An interactive bypass transition mechanism in
wall-bounded flows. J. Fluid Mech. 615, 345–369.
T UCKERMAN , L. S., K REILOS , T., S CHROBSDORFF , H., S CHNEIDER , T. M. & G IBSON , J. F. 2014
Turbulent-laminar patterns in plane Poiseuille flow. Phys. Fluids 26 (11), 114103.
WALEFFE , F. 1997 On a self-sustaining process in shear flows. Phys. Fluids 9 (4), 883–900.