ME3621

Applied Fluid
Mechanics (+CFD)
James Tyacke
Edward Smith
Jun Xia
 This session will be recorded

Brunel University London ME3621 Applied Fluid Mechanics 2

 Feedback

• Use the QR code to go to feedback

• You can ask questions or make comments at any time, either linked to
your name (if you put it in) or anonymously

Brunel University London ME3621 Applied Fluid Mechanics 3

 Summary

Aim
• To consider flows including separation and transition demonstrated with real
cases

Objectives
• Understand flow over bluff bodies
• Introduce modelling considerations
• Understand modelling impact

Brunel University London ME3621 Applied Fluid Mechanics 4

 Effect of turbulence on separation and drag

The drag of a body is made up of two components:

1. Skin friction drag due to surface shear stresses
2. Pressure drag due to separation

• For a streamlined (slender) shape like an aerofoil, most of the drag comes
from skin friction. In this case, laminar flow will tend to reduce overall drag.

• For bluff/blunt (non-slender) shapes, such as a cylinder or sphere, most of

the drag is from pressure drag due to separation. In this case, turbulent
flow will tend to delay separation and hence reduce overall drag.

Streamlined shape – laminar flow Bluff body – turbulent flow

For low drag:

Brunel University London ME3621 Applied Fluid Mechanics 5

 Separation on cylinders

Low Re Wide turbulent wake

• The flow remains

attached or separates,
with vortices detaching
periodically from either
side of the body (“Von (a) Re<1 Laminar separation
Karman vortex street”) (d) 400<Re<3x105
Separation Narrower turbulent wake

Moderate to high Re
• For laminar flow, a
large separation region
of low pressure (wake) Symmetric recirculating wake Laminar separation Turbulent separation
occurs behind the (b) Re=40 Turbulent reattachment
cylinder
Alternate vortex formation in broad wake (e) 3x105<Re<3x106

• By comparison, if the
flow is turbulent, the
separated region/ wake
is smaller in width
(c) 100<Re<200 (f) Re>3x106 Turbulent separation

Brunel University London ME3621 Applied Fluid Mechanics 6

 Reynolds-Averaged Navier-Stokes (RANS) is a
time averaged modelling of the turbulence
NS = Navier-Stokes • The time-averaging
process results in
additional unknown
terms (Reynolds
stresses)
• These require modelling
• The problem is how to
model the infinite range
of turbulent flows and
phenomena
• We use great
simplifications such as
Eddy-
Viscosity/Boussinesq
approximation
• This results in an
effective viscosity, by
adding the modelled
eddy-viscosity to the
real dynamic viscosity
We solve for time-averaged variables, so the mesh resolves the
Brunelmean flow London
University and gradients, not real, time evolving
ME3621 vortices
Applied Fluid Mechanics 7
 The relationship between the apparent du
τ =µ − ρ u′v′
turbulent shear stress and the local velocity dy
is given as • By using an Eddy
viscosity (simply
Here u’ and v’ are the fluctuations in the x using an effective
and z directions respectively. viscosity) in the
solved equations,
The expression for turbulent shear stress consists of two terms we assume
Reynolds stresses
du are aligned with
Laminar shear stress term µ mean flow
dy gradients
turbulent shear stress (known • This is one
as the Reynolds Stress) − ρ u ′v′ assumption of
many which are
violated and result
The correct estimation of turbulent flows depends strongly on in inaccurate,
accurately quantifying the Reynolds stresses in the fluid and this inconsistent or
is what makes the modelling of turbulent flows so complex and fundamentally
difficult. incorrect results
Brunel University London ME3621 Applied Fluid Mechanics 8
 2 equation models include transport or
turbulence variables, here k-ε (k-ω is similar)
You don’t need to
remember these. Just
know there is an equation
for k and epsilon similar in
form to the momentum
equations. The eddy-
viscosity is then evaluated
from these at each
computational point

• Note, these
constants are
tuneable and can
be calibrated for
different flows.
• Underlying
assumptions means
no turbulence
model can perform
well for all flows

Brunel University London ME3621 Applied Fluid Mechanics 9

 Shortcomings related to common EV models

Watch out for these flow features and be cautious:

• Ducts of varying section - Non-local effects (k, epsilon help to some degree)
• Secondary flows - Square ducts, aerofoil junctions
• Jet flows – spreading rate (a round shear layer is different from a plane shear
layer)
• Streamline curvature – no terms to account for centrifugal force in strongly
curved streamlines
• Acceleration – no pressure gradient sensitivity
• Transition or reverse transition (re-laminarization) – flows assume
turbulent (attached BL)

• RANS is widely used because it is cheap relative to LES

• RANS is accurate for certain flows – attached BL with moderate pressure
gradients – not good for transition prediction
• RANS is not good for separated or mixing type flows

Brunel University London ME3621 Applied Fluid Mechanics 10

 Large-Eddy Simulation (LES) is a spatial filtering
of the N-S equations
• LES is always unsteady and 3D!
• Imagine a turbulent velocity field • This filtering essentially removes small
overlaid on a CFD mesh: scales (vortices) which cannot be
captured

Big fish are

captured

Small fish
The grid only picks up samples at escape
these points of a continuous flow field
• These small scales have an overall
dissipative influence on the larger scales
and are modelled. I.e. the sub-grid scales
are modelled, again using a turbulence
model resulting in an additional viscosity but
appears in the equations in the same way:
• μeff = μ + μSGS .
• Note μSGS << μRANS (SGS = sub-grid scales)
Figure credits to K. Hanjalic 
Brunel University London ME3621 Applied Fluid Mechanics 11
 Periodic ribbed duct (infinite span)

• Does this flow have

typical attached BL?
• Are there any high
pressure gradients?
• Is it 2D or 3D?

Pressure source Temperature source

Brunel University London ME3621 Applied Fluid Mechanics 12
 Time-averaged / instantaneous flow

Brunel University London ME3621 Applied Fluid Mechanics 13

 Meshes used

Quite a range in relative grid spacing:

What kind of change in the flow errors do we see with LES?

Brunel University London ME3621 Applied Fluid Mechanics 14

 Heat transfer

RANS LES

• RANS is highly sensitive to a flow it is not designed for

• LES did not change too much due to large-scale free-turbulence

Brunel University London ME3621 Applied Fluid Mechanics 15

 More realistic geometries

Brunel University London ME3621 Applied Fluid Mechanics 16

 Introduction of sidewalls creating a square duct

Brunel University London ME3621 Applied Fluid Mechanics 17

 <u> and <u’>

Brunel University London ME3621 Applied Fluid Mechanics 18

 How might we resolve the peak in <u’u’>?

• We have a shear layer – basically high velocity gradient

• Use 10-20 points across vorticity thickness:

U1
du/dymax

U2

U1 du/dymax
U1 could be Umax
U2 may be zero at rib
U2 ΔU=U1-U2

Brunel University London ME3621 Applied Fluid Mechanics 19

 Heat transfer enhancement

With added complexity in flow, LES still did well!

LES (various) RANS (various)

• LES performs well for a vast range of flows including separated flow, mixing
flows + heat transfer, BL with pressure gradients, due to low modelled
content
• LES is more costly, especially high Re (bigger mesh, unsteady)
Brunel University London ME3621 Applied Fluid Mechanics 20
 A DNS of a similar case (1/3)
S.V. Mahmoodi-Jezeh and B.-C., 2021, Wang Direct numerical simulation of turbulent heat transfer in a
square duct with transverse ribs mounted on one wall

Note
• η (Eta) is the Kolmogorov microscale (smallest viscous scale)
• (a) only 20% mesh lines are shown

Brunel University London ME3621 Applied Fluid Mechanics 21

 A DNS of a similar case (2/3)

Study of various rib heights…

Brunel University London ME3621 Applied Fluid Mechanics 22

 A DNS of a similar case (3/3)

Note:
(a) strong pressure gradients
(b) Change in flow regime affects Nu – flow impinges on next rib at high blockage

Brunel University London ME3621 Applied Fluid Mechanics 23

 Ribs+topology+rotation+passage shape

Is it realistic to expect RANS to get this correct?

Brunel University London ME3621 Applied Fluid Mechanics 24

 Different rib shapes…

Brunel University London ME3621 Applied Fluid Mechanics 25

 What is the optimum?

• Broadly the same flow

features (Separation,
reattachment,
recirculation)
• Heat/momentum
transport details change
Nu and dp.

Brunel University London ME3621 Applied Fluid Mechanics 26

 What is the geometry? – manufacturing
tolerances + variability, wear, fouling…

• Consider other
uncertainties in
the problems
you tackle!

Brunel University London ME3621 Applied Fluid Mechanics 27

 Summary

• Simple geometry generates complex flows

• sensitive to turbulence modelling
• For bluff-body flows, RANS can be inconsistent and unreliable.
LES is consistent and accurate but more costly.
• CAD does not always represent real in-service geometry
• Small changes can alter flow regime
• Vorticity thickness can be used to improve resolution of flow features
(shear layers)
• Heat transfer is often traded against pressure drop
• Improvement to both can be relative to laminar conditions
• Ratio between heat transfer and pressure drop gives an “efficiency”

Brunel University London ME3621 Applied Fluid Mechanics 28