BBA Set 3 - Turbulence Modelling
BBA Set 3 - Turbulence Modelling
Other sources:
Dr. Taka Nishino, Cranfield University, Turbulence Modelling course
Stephen Pope, Turbulent flows, Cambridge University Press, 2000
Lecture overview
• Introduction to turbulence modelling
• Reynolds-averaged decomposition
• Closure problem
• Hybrid RANS-LES
1
What you will learn
Turbulence modelling is complicated and difficult to learn in two lectures
The goal here is to understand the basic concepts so that you can begin to use StarCCM+
2
Turbulence?
3
Examples
nasa.gov
All fluid flows of interest in BBA are turbulent
Very important to predict the role of turbulence
4
Evolution of square jet
5
Need to be careful
Visualization of von Karam type vortex shedding
behind a circular cylinder (Re = 100)
Owen et al., Physics of Fluids, 2000
Laminar
Transitional
6
Reynolds number: We need to know the Reynolds number
• Ratio of (nonlinear) inertial forces to the linear viscous force.
• The inertial forces are responsible for the flow instability while the viscous forces convert the
kinetic energy into thermal energy.
• The small scales are determined by the fluid viscosity and they depend on Reynolds number.
The smallest length scale is called the Kolmogorov length scale, 𝑙 = 𝜈 3 𝐿/𝑈 3 1/4
7
8
Computation of Turbulent Flow
• Turbulent flow is governed by the Navier-Stokes equations: Continuity, Momentum (3
directions) and Energy
• For incompressible fluid (𝜌 = constant) with constant viscosity and no body forces,
• Continuity equation:
𝜕𝑈𝑖
=0
𝜕𝑥𝑖
• Navier-Stokes equation:
𝜕𝑈𝑖 𝜕𝑈𝑖 1 𝜕𝑃 𝜕 2 𝑈𝑖
+ 𝑈𝑗 =− +𝜈
𝜕𝑡 𝜕𝑥𝑗 𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗 𝑥𝑗
9
Computation of Turbulent Flow
• The equations are highly (the 3 N’s)
• Non-linear,
• Non-integrable and
• Non-local
• However, the equations can be modelled through computational methods such as direct
numerical simulation (DNS).
10
Computational Methods
Direct Numerical Simulations (DNS)
Principle: Directly solves the complete Navier-Stokes equations.
Pros: Resolves all the spatial and temporal scales
Cons: Computationally expensive and not applicable for high Reynolds number flows
11
Computational Methods
Reynolds Averaged Navier-Stokes (RANS)-based models
Principle: Solves a Reynolds-averaged form of the Navier-Stokes equations with the use of
models developed to represent the Reynolds stresses.
Pros: Simple and lesser computational cost.
Cons: Doesn’t resolve the turbulence phenomena since it solves an averaged equations.
Hybrid RANS-LES
Principle: Resolves the large-scales eddies away from the wall using LES formulation and
models the small-scales near the wall using RANS
Pros: Lesser computational cost compared to full LES
Cons: Doesn’t fully resolve all the scales.
12
DNS
13
Engineers at Work
• The computational domain size must be much larger than the largest eddy size, while the grid size
must be small enough the resolve the smallest eddy.
• In a 1D flow, calculate the number of computational nodes required to resolve all the length scales
from the largest (L) to the smallest (𝑙 = 𝜈 3 𝐿/𝑈 3 1/4 )?
3/4
𝐿 𝐿 𝑈𝐿
= 1 = = 𝑅𝑒 3/4
𝑙 𝜈
𝜈3𝐿 4
𝑈3
𝑹𝒆𝟗/𝟒
14
Computation of Turbulent Flow
• Since the smallest scale in a flow domain is 𝑙 = 𝜈 3 𝐿/𝑈 3 1/4 , the number of computational
nodes required to cover the length (largest scale, L) while resolving the smallest eddies is of
the order
3/4
𝐿 𝐿 𝑈𝐿
= 1 = = 𝑅𝑒 3/4
𝑙 3 𝜈
𝜈 𝐿 4
𝑈3
• Since turbulent flow is 3D, we need these many points in each direction. Therefore, the total
number of nodes required increases to
𝑹𝒆𝟗/𝟒
• Thus, the grid requirement increase rapidly as Reynolds number increases
15
Engineers at Work
• What is the largest and smallest scales in an open channel flow of water depth, h = 200 mm
and freestream velocity, U = 0.5 m/s? (N.B. 𝑙 = 𝜈 3 𝐿/𝑈 3 1/4 )
16
Computation of Turbulent Flow
• Therefore, DNS offer some promise, but most turbulent flows of engineering importance
occur at large Reynolds numbers (i.e. very wide range of scales, 𝑹𝒆𝟗/𝟒 ) that is practically
impossible to simulate even with the most advanced computational resources.
17
Instantaneous vs. mean flow characteristics
20
Learning objectives
At the end of this section, you will be able to
• Express instantaneous turbulent fields into mean, fluctuations and higher-order moments
based on Reynolds decomposition
21
Background: Statistics
• When you have a set of discrete random data points (𝑥𝑖 ) of size N, you can easily
calculate the
• Mean (𝑋)
𝑁
1
𝑋 = 𝑥𝑖
𝑁
𝑖=1
• Fluctuations or deviations as
𝑥𝑖′ = 𝑥𝑖 − 𝑋
• Variance
𝑁
1
𝜎 2 = 𝑥𝑖 − 𝑋 2
𝑁
𝑖=1
22
Reynolds Decomposition: Mean and Turbulence Field
• In turbulent flows, instantaneous flow variables such as velocity, 𝑈 and pressure, P can be
decomposed into mean values and fluctuations:
𝑈=𝑈 ഥ + 𝑢′
𝑉 = 𝑉ത + 𝑣 ′
𝑊=𝑊 ഥ + 𝑤′
𝑃 = 𝑃ത + 𝑝′
• As a property, the mean of the fluctuation is zero
𝑢ഥ′ = 0
• Therefore, we represent the fluctuations by the root-mean-square
𝑢𝑟𝑚𝑠 = 𝑢′ 𝑢′
often referred to as the intensity of turbulence.
• The (relative) turbulence intensity is given as
ഥ
𝑇𝑢 = 𝑢𝑟𝑚𝑠 /𝑈
23
Second order moments (Reynolds stress)
𝑖 𝑗 𝑘
𝑖 𝑢′2 𝑢′ 𝑣 ′ 𝑢′ 𝑤 ′
𝑗 𝑢′ 𝑣 ′ 𝑣 ′2 𝑣 ′ 𝑤 ′
𝑘 𝑢′ 𝑤 ′ 𝑣 ′ 𝑤 ′ 𝑤 ′2
• Symmetric along the diagonal so although the tensor has 9 variables, it is reduced to 6.
• The diagonals are called the Reynolds normal stresses
• Streamwise (𝑢′2 ), transverse (𝑣 ′2 ) and spanwise (𝑤 ′2 ) according to the direction
• The Reynolds normal stresses add up to form the turbulent kinetic energy (per unit mass)
𝑘 = 0.5 𝑢′2 + 𝑣 ′2 + 𝑤 ′2 = 0.5𝑢𝑖′ 𝑢𝑖′
• The off-diagonal terms are called the Reynolds shear stresses
24
Third order moments
(Triple velocity correlations)
𝑢′3 , 𝑣 ′3 , 𝑤 ′3
𝑢′2 𝑣 ′ , 𝑢′2 𝑤 ′ , 𝑢′ 𝑣 ′ 𝑤 ′
𝑢′ 𝑣 ′2 , 𝑣 ′2 𝑤 ′ , 𝑢′ 𝑤 ′2 , 𝑣 ′ 𝑤 ′2
25 25
Engineers at Work
Rules of averaging
• If 𝐴 = 𝐴ҧ + 𝑎 and 𝐵 = 𝐵ത + 𝑏 are any independent flow variables and 𝑠 is any of the spatial
coordinates (x, y, z), what is
• 𝐴ҧ + 𝑎 = 𝐴ҧ
• 𝐴ҧ𝐵ത = 𝐴ҧ𝐵ത
ҧ =0
• 𝐴𝑏
• 𝐴𝐵 = 𝐴ҧ𝐵ത + 𝑎𝑏
𝜕𝐴 𝜕𝐴ҧ
• =
𝜕𝑠 𝜕𝑠
• 𝐴 = 𝑠𝑑 𝐴 ҧ 𝑑𝑠
Key: The average of a variable’s fluctuation is zero
26
RANS Equations
Continuity Equation:
𝜕𝑈𝑖 𝜕𝑈 𝜕𝑉 𝜕𝑊
= + + =0
𝜕𝑥𝑖 𝜕𝑥 𝜕𝑦 𝜕𝑧
Reynolds decomposition:
ഥ+𝑢
𝜕 𝑈 𝜕 𝑉ത + 𝑣 ഥ +𝑤
𝜕 𝑊
+ + =
𝜕𝑥 𝜕𝑦 𝜕𝑧
ഥ 𝜕 𝑉ത 𝜕𝑊
𝜕𝑈 ഥ 𝜕𝑢 𝜕𝑣 𝜕𝑤
+ + + + + =0
𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑥 𝜕𝑦 𝜕𝑧
Averaging:
ഥ 𝜕 𝑉ത 𝜕 𝑊
𝜕𝑈 ഥ
+ + =0
𝜕𝑥 𝜕𝑦 𝜕𝑧
27
RANS Equations
Momentum Equation:
𝜕𝑈𝑖 𝜕𝑈𝑖 𝜕𝑃 𝜕 2 𝑈𝑖
𝜌 + 𝑈𝑗 =− +𝜇
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝑥𝑗
ഥ
𝜕𝑈 ഥ
𝜕𝑈 ഥ
𝜕𝑈 ഥ 𝜕𝑢𝑢 𝜕𝑢𝑣 𝜕𝑢𝑤
𝜕𝑈 𝜕 𝑃ത
𝜌 ഥ
+𝑈 + 𝑉ത ഥ
+𝑊 + + + =−
𝜕𝑡 𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑥
Rearranging:
ഥ
𝜕𝑈 ഥ
𝜕𝑈 ഥ
𝜕𝑈 ഥ
𝜕𝑈 𝜕𝑃ത
𝜌 ഥ
+𝑈 ത
+𝑉 ഥ
+𝑊 =−
𝜕𝑡 𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑥
𝜕 ഥ
𝜕𝑈 𝜕 ഥ
𝜕𝑈 𝜕 ഥ
𝜕𝑈
+ 𝜇 − 𝜌𝑢𝑢 + 𝜇 − 𝜌𝑢𝑣 + 𝜇 − 𝜌𝑢𝑤
𝜕𝑥 𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑧 𝜕𝑧
28
RANS Equations
y-component
𝜕𝑉ത 𝜕𝑉ത 𝜕𝑉ത 𝜕𝑉ത 𝜕𝑃ത
𝜌 ഥ
+𝑈 ത
+𝑉 ഥ
+𝑊 =−
𝜕𝑡 𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑦
z-component
𝜕𝑊ഥ ഥ
𝜕𝑊 ഥ
𝜕𝑊 ഥ
𝜕𝑊 𝜕𝑃ത
𝜌 ഥ
+𝑈 + 𝑉ത ഥ
+𝑊 =−
𝜕𝑡 𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑧
𝜕 ഥ
𝜕𝑊 𝜕 ഥ
𝜕𝑊 𝜕 ഥ
𝜕𝑊
+ 𝜇 − 𝜌𝑢𝑤 + 𝜇 − 𝜌𝑣𝑤 + 𝜇 − 𝜌𝑤𝑤
𝜕𝑥 𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑧 𝜕𝑧
29
NS vs RANS Equations
• NS Equation:
𝜕𝑈𝑖 𝜕𝑈𝑖 𝜕𝑃 𝜕 𝜕𝑈𝑖
𝜌 + 𝑈𝑗 =− + 𝜇
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗
• RANS Equation:
ഥ𝑖
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝜕𝑃ത 𝜕 ഥ𝑖
𝜕𝑈
𝜌 ഥ𝑗
+𝑈 =− + 𝜇 − 𝜌𝑢𝑖 𝑢𝑗
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗
• The mean momentum equation is complicated by the introduction of the new term referred
to as the Reynolds stress tensor
• The Reynolds stress tensor introduces 9 new variables that can be defined only through
(unavailable) knowledge of the detailed turbulent structure
30
Closure Problem
• So, by averaging the Navier-Stokes equations to obtain the RANS equations, we introduced
ഥ 𝑉,
• Together with the mean variables, 𝑈, ത 𝑊,
ഥ and 𝑃,
ത we have a total of
• 10 variables to be determined from
• Thus, the RANS equations are open (i.e. not closed) – big problem!
• To effect “Closure”, we must develop models for the unclosed terms (𝑢𝑖 𝑢𝑗 )
31
RANS-based Turbulence Models (I)
32
Learning objectives
At the end of the section, you will be able to
33
RANS Equation
ഥ𝑖
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝜕𝑃ത 𝜕 ഥ𝑖
𝜕𝑈
𝜌 ഥ𝑗
+𝑈 =− + 𝜇 − 𝜌𝑢𝑖 𝑢𝑗
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗
• The transport equations of the Reynolds stresses (Second moment closure (SMC)
models)
34
RANS Equation
ഥ𝑖
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝜕𝑃ത 𝜕 ഥ𝑖
𝜕𝑈
𝜌 ഥ𝑗
+𝑈 =− + 𝜇 − 𝜌𝑢𝑖 𝑢𝑗
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗
ഥ𝑖
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝜕𝑃ത 𝜕
𝜌 ഥ
+ 𝑈𝑗 =− + 𝜏𝑖𝑗
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗
ഥ𝑗
ഥ𝑖 𝜕𝑈
𝜕𝑈
𝜏𝑖𝑗 = 𝜇 + − 𝜌𝑢𝑖 𝑢𝑗
𝜕𝑥𝑗 𝜕𝑥𝑖
Laminar Turbulent
(Viscous stresses) (Reynolds stresses)
35
Boussinesq Eddy Viscosity Assumption
• Assuming that the turbulent (Reynolds) stresses act like the viscous stresses, thus, they are
also proportional to the mean velocity gradients
ഥ𝑗
ഥ𝑖 𝜕𝑈
𝜕𝑈
−𝜌𝑢𝑖 𝑢𝑗 = 𝜇𝑡 + ҧ
= 𝜇𝑡 𝑆𝑖𝑗
𝜕𝑥𝑗 𝜕𝑥𝑖
where
ഥ𝑖
𝜕𝑈 ഥ𝑗
𝜕𝑈
ҧ =
𝑆𝑖𝑗 + is the deformation tensor
𝜕𝑥𝑗 𝜕𝑥𝑖
36
What is Eddy viscosity?
37
Boussinesq Eddy Viscosity Assumption
• Furthermore, the above relation is generally not correct because if we apply contraction and
divide by 𝜌, we obtain
ഥ𝑖
𝜕𝑈
−𝑢𝑖 𝑢𝑖 = 2𝜈𝑡
𝜕𝑥𝑖
• The left-hand side of the relation is twice the turbulent kinetic energy.
• For incompressible flow, the right hand side is zero (continuity equation), the left-hand side
is non-zero (in general) unless there is no turbulence at all.
• Instead, we write
1 2
ҧ ҧ
−𝑢𝑖 𝑢𝑗 = 𝜈𝑡 𝑆𝑖𝑗 − 𝑢𝑘 𝑢𝑘 𝛿𝑖𝑗 = 𝜈𝑡 𝑆𝑖𝑗 − 𝑘𝛿𝑖𝑗
3 3
where 𝛿𝑖𝑗 is the Kronecker delta.
38
Boussinesq Eddy Viscosity Assumption
2
ҧ
−𝑢𝑖 𝑢𝑗 = 𝜈𝑡 𝑆𝑖𝑗 − 𝑘𝛿𝑖𝑗
3
𝜕𝑈 ഥ𝑗
ഥ𝑖 𝜕𝑈 2
−𝑢𝑖 𝑢𝑗 = 𝜈𝑡 + − 𝑘𝛿𝑖𝑗
𝜕𝑥𝑗 𝜕𝑥𝑖 3
where
• 𝜈𝑡 is the kinetic turbulent/eddy viscosity
• 𝑘 = 0.5𝑢𝑖′ 𝑢𝑖′ is the turbulent kinetic energy
• The modelling effort is shifted from the Reynolds stresses to the eddy-viscosity
39
Eddy Viscosity Model
• A common formulation of RANS-based model:
ഥ𝑖
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝜕 𝑃ത 2 𝜕 ഥ𝑗
ഥ𝑖 𝜕 𝑈
𝜕𝑈
ഥ𝑗
+𝑈 =− + 𝑘 + [𝜈𝑡 + 𝜈] +
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜌 3 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑖
𝑃ത 2
• Notice that 𝜌
+ 3 𝑘 are combined because they are both normal stresses
𝜈𝑡 ∝ [𝑉][𝐿]
40
What is the muddiest point?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
41
Learning objectives
At the end of the lesson, you will be able to
42
Resources
• Pope, S. B. (2000). Turbulent flows. Cambridge University Press.
• White, F. M. (1991). Viscous fluid flow. 2nd edition (Chapter 6). New York: McGraw-Hill.
• George, W. K. (2013) Lecture notes:
https://www.uio.no/studier/emner/matnat/math/MEK4300/v13/undervisningsmateriale/t
b_16january2013.pdf
43
Turbulence Modelling
Part II
Previously…
• Challenges of turbulence modelling and the different approaches
• Reynolds decomposition: Mean velocity and higher-order moments
ഥ + 𝑢′
𝑈=𝑈 Triple velocity
Reynolds stresses
𝑈 correlations
ഥ
𝑈 𝑢𝑖′ 𝑢𝑗′
𝑢𝑖′ 𝑢𝑗′ 𝑢𝑘′
𝑡
• Development of RANS Equations from the NS Equations and the Closure problem
ഥ𝑖
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝜕 𝑃ത 𝜕 ഥ𝑖
𝜕𝑈
𝜌 ഥ
+ 𝑈𝑗 =− + 𝜇 − 𝜌𝑢𝑖 𝑢𝑗
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗
• Eddy viscosity models based on the Boussinesq Assumption:
𝜕𝑈 ഥ𝑗
ഥ𝑖 𝜕 𝑈 2
−𝑢𝑖 𝑢𝑗 = 𝜈𝑡 + − 𝑘𝛿𝑖𝑗
𝜕𝑥𝑗 𝜕𝑥𝑖 3
• General formulation of Eddy viscosity models:
ഥ𝑖
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝜕 𝑃ത 2 𝜕 ഥ𝑗
ഥ𝑖 𝜕𝑈
𝜕𝑈
ഥ
+ 𝑈𝑗 =− + 𝑘 + [𝜈𝑡 + 𝜈] +
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜌 3 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑖
where 𝜈𝑡 ∝ [𝑉][𝐿]
45
Eddy viscosity models
46
Energy Cascade
&
Zero Equation Models
Learning objectives
At the end of the lesson, you will be able to
48
Energy Cascade: Scales of Turbulent flows
Energy
Containing Inertial Viscous
Range Sub-range Range
E (𝜅) Production (P)
Dissipation (𝜀)
ℓ 𝜆𝑇
η
Wave number (𝜅)
ℓ > 𝜆𝑇 > 𝜂
49
Energy Cascade: Scales of Turbulent flows
• The integral length scale is a measure of the large-scale eddies in the production or
energy containing range: 𝑅𝑢𝑢
ℓ ∝ 𝑘 3/2 Τ𝜀
∆𝑥
The proportionality constant is of the order one
• The Taylor microscale is a measure of the size of the eddies in the inertial subrange:
1/2
𝜆 𝑇 ≈ 10𝜈𝑘/𝜀
2
′2
𝜕𝑢′
𝜆𝑇 = 𝑢 ൘
𝜕𝑥
• The Kolmogorov microscale is the size of the smallest eddies present in the flow:
𝜂 ≈ 𝜈 3 /𝜀 1/4
50
What are the issues?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
51
Eddy Viscosity Models
• They are also known as first order models and can be classified into
depending on the number of transport equations solved to model the eddy viscosity.
52
Zero Equation Models
• This is a simple model that does not solve any transport equation (i.e. zero equation)
• For thin shear layers, the total stress depends on only the mean shear and Reynolds shear stress:
ഥ 𝜕𝑉ത
𝜕𝑈 ഥ
𝜕𝑈
𝜏𝑥𝑦 = 𝜇 + − 𝜌𝑢𝑣 = 𝜇 − 𝜌𝑢𝑣
𝜕𝑦 𝜕𝑥 𝜕𝑦
53
Zero Equation Models
• Therefore, the Boussinesq Eddy Viscosity Assumption reduces to
𝜕𝑈ഥ 𝜕𝑉ത
−𝑢𝑣 = 𝜈𝑡 +
𝜕𝑦 𝜕𝑥
Applying the B.C.: 𝜕Τ𝜕𝑦 ≫ 𝜕Τ𝜕𝑥 or 𝜕Τ𝜕𝑥 = 0 (fully developed flow)
ഥ
𝜕𝑈
−𝑢𝑣 = 𝜈𝑡
𝜕𝑦
where 𝜈𝑡 ∝ [𝑉][𝐿]
• Since, we do not want to solve any transport equation, the “easiest” approach is
use dimensional analysis to reduce the number of unknowns:
ഥ
𝜕𝑈
𝑉 = 𝑐𝐿
𝜕𝑦
• Thus
ഥ
𝜕𝑈 ഥ
𝜕𝑈
𝜈𝑡 = 𝑐𝐿2 2
= 𝑙𝑚
𝜕𝑦 𝜕𝑦
where 𝑙𝑚 is the Prandtl’s mixing length
54
Zero Equation Models
• Therefore
ഥ
𝜕𝑈
−𝑢𝑣 = 𝜈𝑡
𝜕𝑦
2 ഥ
𝜕𝑈 ഥ
𝜕𝑈
−𝑢𝑣 = 𝑙𝑚 𝜕𝑦 𝜕𝑦
• So the modelling effort is now shifted from the eddy viscosity, 𝜈𝑡 to the mixing length, 𝑙𝑚
• It depends on the type of flow, geometric/boundary conditions (e.g. smooth or rough), etc.
55
What is the muddiest point?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
56
Mixing length Models
Examples of mixing lengths for 2D turbulent flows
57
Mixing length Models: Pros & Cons
• Advantages
• Easy to implement
• Economical (computational resources and time)
• Good predictions for simple flows where experimental correlations for the mixing length exist
• Disadvantages
• Cannot describe flows where the turbulent length scale varies (e.g. separated and reattached
flows)
• No history effects
• Only calculates mean flow properties and turbulent shear stress
• Application
• Sometimes used for simple external aero flows
• No longer popular
58
59
60
61
62
63
64
One & Two-Equation Models
Learning objectives
At the end of this part, you will be able to
• Develop the one-equation eddy viscosity model
• Discuss the pros and cons of the Spalart-Allmaras model
• Identify the two main types two-equation models
• Develop the equations for the eddy viscosity used in the two-equation models
• Identify the terms in the k-transport equation that are modelled
• Differentiate between the exact PDE and modelled transport equations
• Discuss the two main limitations of the two-equation models
66
One Equation Models
• One-equation model solves one transport equation (PDE) for either the velocity or length scales
• Most one-equation models solve a transport equation for the turbulent kinetic energy (𝑘) to get
the velocity scale:
𝑉= 𝑘
• Kolmogorov-Prandtl expression:
𝜈𝑡 = 𝐶𝜇 𝑘𝐿
• For the length scale, most of the models use the mixing length
67
Spalart-Allmaras Model: Pros & Cons
• Advantages
• Economical and accurate for:
• Attached wall-bounded flows
• Flows with mild separation and recirculation
• Disadvantages
• Not suitable for
• Massively separated flows
• Free shear flows
• Decaying turbulence
• Application
• Used in unstructured codes in the aerospace industry
• Less popular
68
What is the muddiest point?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
69
Two-Equation Models
• Two-equation models solve two transport equations for the velocity or
length scales
𝑉= 𝑘
• The major difference in the models is the approach used to get the
length scale
• Turbulent dissipation rate, 𝜀
𝐿= 𝑘 3 ൗ𝜀 Recall: ℓ ∝ 𝑘 3/2 Τ𝜀
𝐿 = 𝑘 Τ𝜔
70
Two-Equation Models
• 𝜀- based models
𝜈𝑡 = 𝐶𝜇 𝑘 𝑘 3 ൗ𝜀 = 𝐶𝜇 𝑘 2 Τ𝜀
• Examples
• Standard 𝑘 − 𝜀 model (Launder and Spalding, 1974)
• Realizable 𝑘 − 𝜀 model (Shih et al., 1995)
• Renormalization Group (RNG) 𝑘 − 𝜀 model (Yakhot et al. 1986)
• 𝜔- based models
𝜈𝑡 = 𝑘 𝑘Τ𝜔 = 𝑘Τ𝜔
• Hybrid (𝜔 − 𝜀) models
• Baseline (BSL) model (Menter 1994)
• Shear Stress Transport (SST) model (Menter 1994)
71
What is the muddiest point?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
72
𝑘 −Transport Equation (Exact PDE)
𝑘 = 0.5 𝑢′2 + 𝑣 ′2 + 𝑤 ′2 = 0.5𝑢𝑖′ 𝑢𝑖′
73
Transport Equations: 𝑘 − 𝜀 model
• Exact PDE
𝜕𝑘 𝜕𝑘 𝜕𝑈𝑖 𝜕 𝜕𝑘 𝑝
+ 𝑈𝑗 = −𝑢𝑖 𝑢𝑗 + 𝜈 − 𝑘𝑢𝑗 − 𝑢𝑗 − 𝜀
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜌
𝜕 𝜕 𝜇𝑡 𝜕𝜀 𝜀
𝜌𝑈𝑗 𝜀 = 𝜇+ + 𝐶𝜀1 − 𝜌𝐶𝜀2 𝜀
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜎𝜀 𝜕𝑥𝑗 𝑘
Constants:
𝐶𝜀1 = 1.44, 𝐶𝜀2 = 1.92, 𝜎𝑘 = 1.0, 𝜎𝜀 = 1.3
74
Transport Equations: 𝑘 − 𝜔 model
• Sstandard 𝑘 − 𝜔 model (Steady flow, 𝝏Τ𝝏𝒕 = 𝟎)
𝜕 𝜕 𝜇𝑡 𝜕𝑘
𝜌𝑈𝑗 𝑘 = 𝑃𝑘 + 𝜇+ − 𝛽∗ 𝜌𝑘𝜔
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜎𝑘 𝜕𝑥𝑗
𝜕 𝜕 𝜇𝑡 𝜕𝜔 𝜔
𝜌𝑈𝑗 𝜔 = 𝜇+ + 𝛼 𝑃𝑘 − 𝛽𝜌𝜔2
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜎𝜔 𝜕𝑥𝑗 𝑘
where
Constants:
𝛽 ∗ = 0.09,𝛼 = 5/9,𝛽 = 0.075,𝜎𝑘 = 2,𝜎𝜔 = 2
75
What is the muddiest point?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
76
Limitations of Eddy Viscosity Models
1) Negative eddy viscosity for thin shear layers
ഥ
𝜕𝑈
−𝑢𝑣 = 𝜈𝑡
𝜕𝑦
Thus
ഥ
𝜕𝑈
𝜈𝑡 = (−𝑢𝑣)ൗ
𝜕𝑦
ഥ
𝜕𝑈
• Flow which (−𝑢𝑣) and do not have the same sign will give negative 𝜈𝑡 - big problem!
𝜕𝑦
77
Limitations of Eddy Viscosity Models
• E.g. Wall jets
U −𝑢𝑣
ഥ
𝜕𝑈
• Since ym ≠ y−uv=0, the mismatch region (marked green) will yield negative 𝜈𝑡 since (−𝑢𝑣) and do
𝜕𝑦
not have the same sign.
78
Limitations of Eddy Viscosity Models
2) Anisotropy
• The eddy viscosity models (Boussinesq expression and transport equations) are based on isotropic
turbulence
• Isotropic turbulence assumes that all turbulence statistics are independent of rotation and
reflection of the coordinate system
• Simply, all the Reynolds normal stresses are constant/the same
𝑢′2 = 𝑣 ′2 = 𝑤 ′2
Thus
𝑘 = 1.5 𝑢′2 Recall: 𝑘 = 0.5 𝑢′2 + 𝑣 ′2 + 𝑤 ′2
79
What is the muddiest point?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
80
Learning objectives
At the end of the lesson, you will be able to
• Develop the one-equation eddy viscosity model
• Discuss the pros and cons of the Spalart-Allmaras model
• Identify the two main types two-equation models
• Develop the equations for the eddy viscosity used in the two-equation models
• Identify the terms in the k-transport equation that are modelled
• Differentiate between the exact PDE and modelled transport equations
• Discuss the two main limitations of the two-equation models
81
82
Second Moment Closure Models
Learning objectives
At the end of the lesson, you will be able to
• Discuss the two main approaches used in Second moment closure models
84
Second Moment Closure (SMC) Models
• These models address the limitations of the Eddy Viscosity models based on two main approaches
85
ASM vs RSM
• Algebraic Stress Models (ASM)
• Generally, solves two transport equations like the two-equation eddy viscosity models
• However, the Reynolds stress tensor is modelled using the Anisotropic tensor (𝑎𝑖𝑗 ) formulation:
2
𝑢𝑖 𝑢𝑗 = 𝑘 𝑎𝑖𝑗 + 𝛿𝑖𝑗
3
• Example: 𝑘 − 𝜀 EARSM solves the 𝑘 and 𝜀 transport equations and the constitutive expression for 𝑎𝑖𝑗
86
Challenges of RSM
• Exact PDE of 𝑢𝑖 𝑢𝑗
𝜕𝑢𝑖 𝑢𝑗 𝜕𝑢 𝑢
ഥ𝑘 𝑖 𝑗 =
+𝑈
𝜕𝑡 𝜕𝑥𝑘
ഥ𝑗
𝜕𝑈 ഥ𝑖
𝜕𝑈 𝑝 𝜕𝑢𝑖 𝜕𝑢𝑗 𝜕𝑢𝑖 𝜕𝑢𝑗
− 𝑢𝑖 𝑢𝑘 + 𝑢𝑗 𝑢𝑘 + + − 2𝜈
𝜕𝑥𝑘 𝜕𝑥𝑘 𝜌 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑘 𝜕𝑥𝑘
𝜕 𝜕𝑢𝑖 𝑢𝑗 𝑝
+ 𝜈 − 𝑢𝑖 𝑢𝑗 𝑢𝑘 − 𝑢𝑖 𝛿𝑗𝑘 + 𝑢𝑗 𝛿𝑖𝑘
𝜕𝑥𝑘 𝜕𝑥𝑘 𝜌
𝑣 𝑡 𝑝
𝐿𝑖𝑗 + 𝐶𝑖𝑗 = 𝑃𝑖𝑗 + 𝜙𝑖𝑗 − 𝜀𝑖𝑗 + 𝐷𝑖𝑗 + 𝐷𝑖𝑗 + 𝐷𝑖𝑗
𝑡 𝑝
• Terms that require modelling: 𝜙𝑖𝑗 , 𝜀𝑖𝑗 , 𝐷𝑖𝑗 and 𝐷𝑖𝑗
• Modelling the dissipation and pressure-strain terms is challenging, therefore the different approaches used
often differentiates the various RSM models
87
Example of RSM: LRR-IP Model
• 𝑢𝑖 𝑢𝑗 transport equation
𝜕 𝜕 2 𝑘2 𝜕 𝜕𝑈𝑖 𝜕𝑈𝑖 2
𝜌𝑈𝑘 𝜌𝑢𝑖 𝑢𝑗 = 𝜇 + 𝐶𝑠 𝜌 𝜌𝑢𝑖 𝑢𝑗 − 𝜌𝑢𝑘 𝑢𝑗 − 𝜌𝑢𝑘 𝑢𝑗 + 𝜙𝑖𝑗 − 𝛿𝑖𝑗 𝜌𝜀
𝜕𝑥𝑘 𝜕𝑥𝑘 3 𝜀 𝜕𝑥𝑘 𝜕𝑥𝑗 𝜕𝑥𝑘 3
• 𝜀 −transport equation
𝜕 𝜕 𝜇𝑡 𝜕𝜀 𝜀 𝜕𝑈𝑖
𝜌𝑈𝑗 𝜀 = 𝜇+ + 𝐶𝜀1 −𝜌𝑢𝑖 𝑢𝑗 − 𝜌𝐶𝜀2 𝜀
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜎𝜀 𝜕𝑥𝑗 𝑘 𝜕𝑥𝑗
where
𝑢𝑖 𝑢𝑗 2 1 𝜕𝑈𝑖 𝜕𝑈𝑗 1 𝜕𝑈𝑖 𝜕𝑈𝑗
𝑎𝑖𝑗 = − 𝛿𝑖𝑗 , 𝑆𝑖𝑗 = + , 𝛺𝑖𝑗 = −
𝑘 3 2 𝜕𝑥𝑗 𝜕𝑥𝑖 2 𝜕𝑥𝑗 𝜕𝑥𝑖
Constants
𝐶𝑠 = 0.22, 𝐶𝑠1 = 1.8, 𝐶𝑟2 = 0.8, 𝐶𝑟4 = 0.6, 𝐶𝑟5 = 0.6, 𝐶𝜀1 = 1.45, 𝐶𝜀2 = 1.90, 𝜎𝜀 = 1.10
88
SMC Models: Pros & Cons
• Advantages
• Potentially the most complete turbulence models
• Provides the mean velocities and all the Reynolds stresses
• Suitable for simulation of complex turbulent flows
• Disadvantages
• Computationally expensive compared to the Eddy viscosity models
• The ASMs are less expensive compared to RSMs.
• RSM: The modelling of the dissipation and pressure-strain terms is challenging and can cause unstable
simulation (difficulty in convergence to a solution)
• Increased number of coefficients and constants that may require calibration for different types of flows
89
What is the muddiest point?
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
90
Learning objectives
At the end of the lesson, you will be able to
• Discuss the two main approaches used in Second moment closure models
91
Introduction to LES
92
Learning objectives
At the end of the lesson, you will be able to
93
Large Eddy Simulations (LES)
• LES resolves the large-energy containing scales and model the small scales
• The velocity vector is decomposed into filtered velocity and subgrid scale or residual (analogous to
the Reynolds decomposition)
ഥ 𝒙, 𝑡 + 𝑢′ (𝒙, 𝑡)
𝑈 𝒙, 𝑡 = 𝑈
• The filtered velocity is obtained by using a filter function, say 𝐺(𝑟) (Leonard 1974)
∞
ഥ 𝒙, 𝑡 = න 𝐺 𝑟, 𝒙 𝑈 𝒙 − 𝑟, 𝑡 𝑑𝑟
𝑈
−∞
• The required grid spacing is proportional to the specified filter width (∆) in the filter function
Types of filters
• Box filter
• Gaussian filter
95
Filtered equations
• Continuity
ഥ𝑖
𝜕𝑈
=0
𝜕𝑥𝑖
• Momentum
𝜕𝑈 ഥ𝑖 𝑈
ഥ𝑖 𝜕𝑈 ഥ𝑗 1 𝜕𝑃ത ഥ𝑖
𝜕 2𝑈 𝜕𝜏𝑖𝑗
+ =− +𝜈 −
𝜕𝑡 𝜕𝑥𝑗 𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗
ഥ𝑖 𝑈
𝜏𝑖𝑗 = 𝑈𝑖 𝑈𝑗 − 𝑈 ഥ𝑗
• Scale-similarity model
• Dynamic model (Germano et al. 1991, Lilly 1992, Meneveau et al. 1996)
Smagorinsky model
• Analogous to the linear eddy viscosity model
∗
𝜏𝑖𝑗 = −2𝜈𝑠𝑔𝑠 𝑆𝑖𝑗
• The eddy viscosity of the residual stress (𝜈𝑠𝑔𝑠 ) is model based on the analogy of the mixing length hypothesis
𝜈𝑠𝑔𝑠 = 𝑙𝑠2 𝑆ҧ
where
𝑙𝑠 = 𝐶𝑠 ∆
∗
𝜏𝑖𝑗 = −2(𝐶𝑠 ∆)2 𝑆ҧ 𝑆𝑖𝑗
https://www.prodigygame.com/main-en/blog/active-learning-strategies-examples
99
Learning objectives
At the end of the lesson, you will be able to
100
Concluding Remarks
101
Lecture overview
• Reynolds-averaged decomposition
• Closure problem
• Hybrid RANS-LES
102
Engineers at Work
Assuming you have access to all the models (zero-equation to Hybrid LES-RANS), which model
will you use for simulating the following flows? (N.B. Time and computational resources are
factors to consider)
• A simple fully developed channel flow where only the U and −𝑢𝑣 are required.
• A flow where the mean velocity and all the Reynolds stresses are required due to high
anisotropy in the flow.
• A flow around an Ahmed body where you are interested in the unsteady wake characteristics.
103
Resources
• Pope, S. B. (2000). Turbulent flows. Cambridge University Press
• White, F. M. (1991). Viscous fluid flow. 2nd edition (Chapter 6). New York: McGraw-Hill
• Wilcox, D. C. (1994). Turbulence modeling for CFD. La Canada, CA: DCW industries
• Hoffmann, K. A., & Chiang, S. T. (2000). Computational fluid dynamics volume III. Fourth edition,
Engineering Education System
• ANSYS CFX-solver Theory Guide. (2018) ANSYS CFX Release (Vol. 19.2).
104