TURBULENCE GENERATED

BY FRACTAL GRIDS
D. Hurst, R.E. Seoud & J.C. Vassilicos

Department of Aeronautics and

Institute for Mathematical Sciences
Imperial College London, U.K.

. – p.1/45
 MOTIVATIONS
1. Many applications in environmental and geophysical
flows as well as in industry
of fractal-forced or fractal-generated turbulence,
e.g. polydisperse droplets/particles in turbulent carrier fluid
that are large enough to force the turbulence over a wide
range of scales corresponding to a wide range of particle
wake sizes (combustion applications, ocean wind-wave
sprays); turbulent flows through trees, over plant canopies,
over multi-sized breaking ocean waves, etc; various novel
mixing devices for the process, oil and other industries as
wel as novel ventilation systems (recent patents by Imperial
College London) which can impact on the environment by
requiring less power to mix...
AT THE VERY LEAST, A REFERENCE LABORATORY
EXPERIMENT IS REQUIRED . – p.2/45
 MOTIVATIONS
2. How to create ideal turbulence experiments with
(i) a very wide range of outer-to-inner scales
(ii) fully controlled conditions in the laboratory
(iii) the possibility to accurately measure down to the
smallest scales
3. Better: how to tamper with the turbulence in the
laboratory?
Various theories exist where the exponents p, q in
E(k) ∼ k −p , L/u03 ∼ Req are determined by
to one or many fractal dimensions of a fractal/multifractal,
spiral/multispiral field:
is it possible to modify E(k) ∼ k −p and/or L/u03 ∼ Req away
from p = 5/3 and q = 0 by tampering with the fractal/spiral
field and changing these dimensions?
. – p.3/45
 MOTIVATIONS
4. Effects on drag properties?

5. How does a turbulence decay when it is generated by

creating many eddies of many different sizes at once?

6. How does a turbulent flow scale when it is generated by

a fractal which has its own intrinsic scaling?

7. Multiscale flow control? in the present case, passive.

. – p.4/45
 Wind tunnels
0.912 m2 width; test section 4.8m; max speed 45m/s;
background turbulence ≈ 0.25%.

0.462 m2 width; test section ≈ 4.0m; max speed 33m/s;

background turbulence ≈ 0.4%.

. – p.5/45
FRACTAL CROSS GRIDS

. – p.6/45
FRACTAL I GRIDS

. – p.7/45
FRACTAL SQUARE GRIDS

. – p.8/45
 Three families of fractal grids
Three fractal-generating patterns

The fractal grids are totally characterised by

(i) the number of fractal iterations N
(ii)the lengths Lj = RLj L0 and thicknesses tj = Rtj t0 ,
j = 0, ..., N − 1
(iii) the number B j of patterns at iteration j : always here,
B = 4 and RL ≤ 1/2, Rt ≤ 1

. – p.9/45
 Important parameters
logB
Fractal dimension of fractal perimeter: Df = log(1/R L)
.
1 ≤ Df ≤ 2.
WE FIND THAT BEST MEAN FLOW HOMOGENEITY IS
ACHIEVED FOR MAXIMUM Df i.e. Df = 2:
Thickness ratio tr ≡ t0 /tN −1 ≡ tmax /tmin . (Note tr = Rt1−N .)
WE FIND THAT THE TURBULENCE INTENSITY
INCREASES WITH BOTH PRESSURE DROP (WHEN
INCREASING BLOCKADGE RATIO) AND THICKNESS
RATIO tr (KEEPING BLOCKADGE RATIO CONSTANT).
4T 2
√
Effective mesh size Mef f = P 1 − σ where T = tunnel
width, P = fractal perimeter, σ = blockadge ratio.
WE FIND THAT THE TURBULENCE SCALES WITH Mef f
IN THE CASE OF CROSS AND I GRIDS. Statistical
homogeneity can be as good as for classical grids, but
further dowstream in multiples of Mef f . . – p.10/45
Minimal complete description of grids

Cross grids require 4 parameters: e.g. T, N, tmax , Rt .

(T = Lmax , RL = 1/2 hence Df = 2.)
I and Square grids require 5 parameters: e.g.
N
1−RL
T, N, Lmax , tmax , tmin . (T ≈ Lmax 1−RL .)
VARIOUS WIND TUNNEL TESTS WERE CARRIED OUT
WITH A NUMBER OF GRIDS FROM EACH FAMILY.
GROUPS OF GRIDS FROM GIVEN FAMILIES WERE
CHOSEN SO AS TO HAVE THE SAME VALUES OF
PARAMETERS BUT ONE, IN ORDER TO DETERMINE
THIS ONE PARAMETER’S EFFECT WHEN EVERYTHING
ELSE IS KEPT CONSTANT:
E.G. KEEPING BLOCKADGE RATIO, AND/OR NUMBER
OF ITERATIONS AND/OR Mef f AND/OR tmin CONSTANT,
ETC, ETC, ETC... . – p.11/45
I grids: N = 6 and Df = 1.98, 1.87, 1.79, 1.6
Equal σ = 25 %, tmin = 1mm, T = 0.91m tunnel.

. – p.12/45
 I grid: N = 5 and Df = 2.0
σ = 31%, tmin = 4mm, T = 0.91m tunnel.

. – p.13/45
Df = 2 fractal I grids; T = 0.46m tunnel
Equal N = 4, σ = 25%, Mef f between 36mm and 37mm.
tr = 2.5, 5.0, 8.5, 13.0, 17.0

. – p.14/45
 Results: turbulence decay
Many possible ways to collapse the I grid data have been
tried. It is found that
(u0 /U )2 = tr C∆P (T /Lmax )2 f ct(x/Mef f )
collapses the turbulence decay data generated by all fractal
I grids in both wind tunnels.
0.08
Df = 1.98
0.07 Df = 1.87
Df = 1.79
0.06 D = 1.68
f
(u’/U) normalised

Df = 2.00
0.05
tr = 2.5
0.04 t = 5.0
r
tr = 8.5
0.03 tr = 13.0
t = 17.0
r
0.02

0.01

0
0 50 100 150 200
x/M
eff . – p.15/45
Df = 2, σ = 25% fractal square grids
and equal Mef f ≈ 2.6cm, Lmax ≈ 24cm, Lmin ≈ 3cm, N = 4,
T = 0.46m.
BUT tr = 2.5, 5.0, 8.5, 13.0, 17.0

. – p.16/45
Profiles at x = 3.25m in T = 0.46m tunnel
6
1.3
tr = 2.5
1.25 tr = 5.0 5
t = 8.5
r
1.2 t = 13.0
r 4
t = 17.0
1.15 r

u’/U (%)
3
U/U∞

1.1

1.05 2

1
1
0.95

0.9 0
−0.5 0 0.5 −0.5 0 0.5
y/T y/T
6

4
v’/U (%)

0 . – p.17/45
−0.5 0 0.5
y/T
 Results: turbulence intensity
10 10
tr = 2.5
t = 5.0
r
8 tr = 8.5 8
t = 13.0
r
t = 17.0
r
6 6
u’/U (%)

v’/U (%)
4 4

2 2

0 0
0 1 2 3 4 0 1 2 3 4
x (m) x (m)

10 10
tr = 2.5 tr = 2.5
tr = 5.0 tr = 5.0
8 t = 8.5 8 t = 8.5
r r
tr = 13.0 tr = 13.0
t = 17.0 t = 17.0
r r
6 6
u’/U (%)

v’/U (%)
4 4

2 2

0 0
0 50 100 150 200 0 50 100 150 200
(x L ) / (t T) (x L ) / (t T)
min min min min

L2max
xpeak = 75 tLmin
min
T
(Hurst & V PoF 2007) but xpeak = 1.2 tmax
(Mazellier, Bruera & V (to appear)) . – p.18/45
 Isotropy collapse using xpeak
xpeak helps collapse u0 /v 0 as fct of x

1.6 1.6

1.5 1.5

1.4 1.4

1.3 1.3

u’/v’
u’/v’

1.2 1.2
tr = 2.5 1.1 tr = 2.5
1.1
tr = 5.0 tr = 5.0
1 tr = 8.5 1 tr = 8.5
tr = 13.0 tr = 13.0
0.9 0.9
tr = 17.0 tr = 17.0
0.8 0.8
0 1 2 3 4 0 0.5 1 1.5 2 2.5 3
x (m) x/xpeak

T = 0.46m tunnel with U∞ = 10m/s.

. – p.19/45
Results: power-law turbulence decay?
How does the Taylor microscale evolve?
8 8

7 7

6 6

5 5

λ (mm)
λ (mm)

4 4

3 3 t = 2.5
t = 2.5 r
r
tr = 5.0 tr = 5.0
2 2
tr = 8.5 tr = 8.5
t = 13.0 1 t = 13.0
1 r r
tr = 17.0 tr = 17.0
0 0
0 1 2 3 4 0 0.5 1 1.5 2 2.5 3
x (m) x/x
peak

T = 0.46m tunnel and U∞ = 10m/s

. – p.20/45
 Results: integral length-scales

0.06 0.06
tr = 2.5
t = 5.0
r
0.05 0.05 t = 8.5
r
tr = 13.0
0.04 0.04 tr = 17.0
Lu (m)

L (m)
0.03 0.03

v
0.02 tr = 2.5 0.02
tr = 5.0
tr = 8.5
0.01 t = 13.0 0.01
r
t = 17.0
r
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x/xpeak x/xpeak

At U∞ = 10m/s, for all x  xpeak and for all grids, λ ≈ 6mm,

Lu ≈ 48mm, Lv ≈ 22mm (about Lu /2 as required by
isotropy) all  T = 0.46m.

. – p.21/45
Exponential turbulence decay at x  xpeak
u02 = u02
peak exp[−(x − xpeak )/lturb ]
where
xpeak = 75 tLmin
min
T
and l turb = 0.1λ 0
U λ0
ν
11
tr = 2.5 tr 17
10 tr = 5.0 exponential theory
1.2
tr = 8.5
9 tr = 13.0
1
tr = 17.0

u’2 (m/s)2
ln(U/u’)2

8 0.8

7 0.6

6 0.4

5 0.2

4 0
0 1 2 3 4 0 1 2 3 4
x (m) x (m)

. – p.22/45
Comparison with classical grid turbulence
(measurements taken by N. Mazellier)
−9 400

−9.5
300
ln( (λ / T)2 )

−10

Rλ
−10.5 200

−11
100
−11.5

−12 0
3 3.5 4 4.5 5 0 50 100
ln( (x − x0) / Meff) (x − x0) / Meff

−6

−6.5
ln ( <u > / U )

ln ( <u2> / U2 )
2

−5
−7

−7.5
2

−8 −5.5

−8.5

−9 −6
3 3.5 4 4.5 5 0 50 100
ln( (x − x0) / Meff) (x − x0) / Meff . – p.23/45
 02
Dissipation during exponential u decay

u02 = u02
peak exp[−(x − xpeak )/lturb ]

and

Lu , Lv independent of x

ARE INCOMPATIBLE WITH

− 32 U dx
d 02
u =  = C u03 /Lu
with C a universal constant.
All the results presented in what follows have been obtained
in the decay region (x > xpeak ) of the turbulence generated
by fractal square grids in the T = 0.46cm tunnel.
. – p.24/45
In fact, Lu, Lv , λ, Lu/λ and Lv /λ
are independent of x, tr and U∞ during decay.
25
tr 17 7.0 m/s
tr 17 10.5 m/s
tr 17 16.2 m/s
tr 17 19 m/s
tr 13 7 m/s
tr 13 16.3 m/s
tr 13 19 m/s
20
L / λ v Re tr 8.5 7.3 m/s
11 λ tr 8.5 16 m/s
7≤ U / m / s ≤ 19
inf
Grids = tr17 , tr 13 and tr 8.5
L11 / λ

15

10

5
100 200 300 400 500 600 700 800 900 1000
Re
λ

u02 −1 u03 −1 u03

Hence,  = = ∼ 15ν λ2 15Reλ λ Reλ Lu
VERY DIFFERENT FROM TAYLOR-KOLMOGOROV
u03
SCALING WHERE  ∼ Lu AND Lu /λ ∼ Reλ . – p.25/45
 03 −1
Lu/u ∼ Reλ
2
tr 17 7.0 m/s
tr 17 10.5 m/s
1.8 tr 17 16.2 m/s
tr 17 19 m/s
3 tr 13 7 m/s
Cε ( ε.L11/urms) v Reλ tr 13 16.3 m/s
1.6
tr 13 19 m/s
7 ≤ Uinf / m/s ≤ 19 tr 8.5 7.3 m/s
tr 8.5 16 m/s
1.4 Grids = tr17 , tr 13 and tr 8.5

1.2
Cε

1 Cε = 140 /Reλ

0.8

0.6

0.4

0.2

100 200 300 400 500 600 700 800 900 1000
Re
λ

Measurements taken on the centreline at x/xpeak between

about 1 and 3, i.e. x/Mef f between about 50 and 110 (end
of test section). NOTE HIGH Reλ VALUES IN SMALL
TUNNEL. . – p.26/45
Statistical homogeneity at x > xpeak

370
320
1.6 300
280
210
180
1.4 U /U v y / cm
local inf
U = 10.5 m/s, 180 ≤ x/cm ≤ 370
inf
1.2 Grid =tr17
Ulocal / Uinf

0.8

0.6

0.4

0 2 4 6 8 10 12 14 16 18

y / cm

. – p.27/45
Statistical homogeneity at x > xpeak
0.2
180 0.2
210 180
280 210
0.18
300 0.18 280
u /U v y / cm 320 300
rms local 320
0.16 U = 16.2 m/s, 180 ≤ x / cm ≤ 370 370
370
inf 0.16
Grid = tr 17 v /U v y / cm
rms local
0.14
0.14
U
inf
= 16.2 m/s , 180 ≤ x / cm ≤ 370
Grid = tr 17
urms / Ulocal

0.12

vrms / Ulocal
0.12

0.1
0.1

0.08
0.08

0.06
0.06

0.04
0.04

0.02
0.02

0 2 4 6 8 10 12 14 16 18
0 2 4 6 8 10 12 14 16 18
y / cm y / cm

. – p.28/45
 Statistical isotropy at x > xpeak
1
370
320
0.9 300
280
210
0.8 180
2
C ( |E (k)| / [E (k).E (k)] ) v kη
xy 12 11 22
0.7 U = 16 . 2 m/s , CL data, Grid = tr 17
Cxy, Cxy (45) inf

0.6

0.5

Cxy(45)
0.4

0.3

0.2

0.1

0
−4 −3 −2 −1 0
10 10 10 10 10
kη

Coherence spectrum at various x positions on the

centreline y = 0. Coherence spectra are very much the
same off centreline at y = 3cm and y = 6cm.
. – p.29/45
Statistical local isotropy at x > xpeak
1.3
0 cm 10.5 m/s
3 cm 10.5 m/s
1.25 6 cm 10.5 m/s
K1 v Re 0 cm 16.2 m/s
λ 3 cm 16.2 m/s
Uinf = 10.5 m/s and 16.2 m/s 6 cm 16.2 m/s
1.2
Grid = tr17, 280 ≤ x / cm ≤ 370
1.15

1.1
K1

1.05

0.95

0.9

0.85

0.8
150 200 250 300 350 400 450
Re
λ

Derivative ratio K1 ≡ 2 < ( ∂u

∂x ) 2 > / < ( ∂v )2 > as function of
∂x
Reλ at locations (x, y) downstream from the tr = 17 fractal
grid where x is larger than 2xpeak and y = 0, 3, 6cm.
Local isotropy implies K1 = 1. . – p.30/45
 03 −1
Lu/u ∼ Reλ at x > xpeak and |y| < Lmax/
2
tr 17 7.0 m/s CL + 3cm
tr 17 10.5 m/s CL + 3 cm
1.8 tr 17 16.2 m/s CL + 3 cm
tr 17 7.0 m/s CL + 6 cm
tr 17 10.5 m/s CL + 6 cm
tr 17 16.2 m/s CL + 6 cm
1.6
tr 13 16.3 m/s CL + 3 cm
tr 13 16.3 m/s CL + 6cm
tr 8.5 16 m/s CL + 3 cm
1.4 tr 8.5 16 m/s CL + 6 cm

1.2
Cε

1
C ( ε.L /u3 ) v Re
ε 11 rms λ
7 ≤ U / m/s ≤ 16, CL + 3 cm / + 6 cm
inf
0.8 Grids = tr17 , tr 13 and tr 8.5

0.6
Cε = 143 / Reλ
0.4

0.2

100 200 300 400 500 600 700 800 900 1000
Reλ

Measurements taken at x/xpeak between about 1 and 3, i.e.

x/Mef f between about 50 and 110 (end of test section), at y
positions between -12cm and +12cm (Lmax = 24cm,
Lu ≈ 5cm) . – p.31/45
Does Kolmogorov scaling hold here?
−5/3
We still get a power-law range where E11 (k1 ) ∼ at k1
high enough Reλ even though  is Reλ -dependent!
But we can collapse spectra E11 (k1 ) at different x with only
one length-scale: e.g. E11 (k1 ) = u02 Lu f (k1 Lu ).
0
10
370
320
−1
10 300
280
210
−2
10 180

(kL11)−1.67
−3
10
E11(k) / u2rms L11

−4
10

−5
10

−6
10

E (k) / u2 L v kL
−7 11 rms 11 11
10
Uinf = 16.2 m/s, CL data
−8
10
Grid = tr 17
−9
10

−10
10
−1 0 1 2 3
10 10 10 10 10 . – p.32/45
kL
11
 02
or E11(k1) = u λf (k1λ)
−3
10
370
320
−4
300
10 280
210
180
−5
10

(kλ)−1.67
E11(k) / urms λ

−6
10
2

−7
10

−8
10
2
E11(k) / urmsλ v kλ
−9 < Uinf > = 16.2 m/s, CL data
10
Grid = tr 17

−10
10

−11
10
−2 −1 0 1 2 3
10 10 10 10 10 10
kλ

. – p.33/45
 Non-Kolmogorov -5/3
The energy spectrum of decaying turbulence generated by
space-filling fractal square grids scales with only one
length-scale l(x), i.e. E11 (k1 ) = u02 lf (k11 l).
This implies that Lu ∼ l and λ ∼ l, hence L ∼ λ as
observed.
This also implies that  ∼ Re−1
λ u 03 /L as also observed.
u

And it also implies that in the power-law range, if a -5/3

u03 2/3 −5/3
spectrum exists, then E11 (k1 ) ∼ ( Lu ) k1 instead of
−5/3
E11 (k1 ) ∼ 2/3 k1 .
There exist fractal, i.e. multiscale, generators of turbulence
which lock the turbulence into a single length-scale! Yet, the
-5/3 is present even though the dissipation anomaly is not.
. – p.34/45
 Vortex Stretching?
The nonlinear rate of change of the enstrophy results from
vortex stretching and equals < ω · sω >.

In isotropic homogeneous turbulence,

< ( ∂u
∂x ) 3 >= − 2 < ω · sω >
35
and
< ( ∂u
∂x ) 2 >= 1 < ω 2 >.
15

Hence, the derivative skewness

S ≡< ( ∂u
∂x ) 3 > / < ( ∂u )2 >3/2
∂x
is a normalised dimensionless measure of the average
vortex stretching rate and can be obtained from a single hot
wire if use is made of Taylor’s frozen flow hypothesis.

. – p.35/45
Vortex stretching in single-scale turbulenc
The scale-by-scale energy balance
∂ 2 E(k, t)
E(k, t) = T (k, t) − 2νk
R ∞∂t 2
implies that 0 k T (k)dk is the rate of change of the
R∞ 2
average enstrophy 0 k E(k)dk as a result of nonlinear
interactions.

Hence, in isotropic homogeneous turbulence,

R∞ 2
R∞ 2
2 15 3/2 R 0 k T (k)dk 1/2 k T (k)dk
S= − 35 ( 2 ) ( ∞ k2 E(k)dk)3/2 = −(135/98) ( ∞ k2 E(k)dk)3/2
R 0

0 0

which can be evaluated from the scale-by-scale energy

budget and the single-scale spectrum property
E(k, t) = u02 λf (kλ) to give:

B d
S = ARe−1λ + u0 dt λ
in terms of two dimensionless constants A and B . . – p.36/45
Mean vortex stretching drops as Reλ grows

. – p.37/45
 −0.15 −1
but as S ∼ Reλ rather than S ∼ Reλ
This apparent -0.15 scaling is caused by the small
time-dependence of λ. Indeed
B d
S = ARe−1 λ + u0 dt λ
which can be recast as
SReλ = A + B Ulocal d 2
ν dx λ
if use is made of Taylor’s frozen flow hypothesis
Ulocal dt = dx.

This slow increase of λ with x, if fitted by λ ∼ (x − x0 )s with

0 < s < 1/2, implies a stretched exponential decay of u02
instead of the exponential form mentioned earlier. We leave
this correction for future detailed measurements and
studies.

. – p.38/45
 Ulocal d 2
SReλ = A + B ν dx λ for grid tr = 17.0

. – p.39/45
 Ulocal d 2
SReλ = A + B ν dx λ for grid tr = 13.0

. – p.40/45
 Ulocal d 2
SReλ = A + B ν dx λ for grid tr = 8.5

. – p.41/45
Can intermittency grow with L/λ constant
F ≡< ( ∂u
∂x ) 4 > / < ( ∂u )2 >2
∂x

. – p.42/45
 CONCLUSIONS
Df , tr and Mef f are important fractal grid parameters. Best
homogeneity is obtained for Df = 2. For space-filling fractal
I and square grids, homogeneity can be further improved by
increasing tr . In all cases of fractal grids, turbulence
intensity and Reynolds number can also be increased by
increasing tr .
Turbulence decay, fractal I grids:
(u0 /U )2 = tr C∆P (T /Lmax )2 f ct(x/Mef f )
Turbulence decay, fractal square grids, at x  xpeak :
u02 = u02
peak exp[−(x − xpeak )/lturb ]
where
xpeak = 75 tLmin
min
T
and l turb = 0.1λ 0
U λ0
ν
u02
peak increases linearly with tr . The Taylor microscale
λ = λ0 and the integral length scales are independent of tr
and U∞ and remain approx constant during decay. . – p.43/45
 CONCLUSIONS
In the decay region of space-filling fractal square grids the
turbulence is approximately homogeneous and locally
isotropic and such that (see W.K. George, PoF 1992):
E11 (k1 ) = u02 Lu f (k11 Lu ) = u02 λf (k11 λ)

L/λ = Const independent of x, tr and U∞

 ∼ Re−1
λ u 03 /L
u

u03 2/3 −5/3

A -5/3 power-law range exists where E11 (k1 ) ∼ ( L u ) k1
2/3 −5/3
instead of E11 (k1 ) ∼  k1 .

. – p.44/45
 CONCLUSIONS
Furthermore, in this decay region of space-filling fractal
square grids where turbulence is approximately
homogeneous and locally isotropic, the turbulence is also
such that
(i) vortex stretching decreases in the mean as the Reynolds
number is increased
(ii) and “intermittency” does not grow but remains constant
with increasing Reynolds number.

It is possible to tamper with the deepest properties of

homogeneous isotropic turbulence: the dissipation
anomaly, vortex stretching and intermittency. This points at
new possibilities for turbulence control. Also, if you can
tamper with something, you can start understanding it.

. – p.45/45