You are on page 1of 43

Universit di Cagliari

Corso di Laurea Magistrale in Ingegneria Civile


Corso di Idraulica 2

A (very) quick introduction to


Computational Fluid Dynamics
Riccardo Rossi
Universit di Bologna
Stanford University
Karalit srl

Anno Accademico 2014/15


Todays lecture

Motivation and objectives


To provide a quick overview of CFD from basic principles to an illustrative example of an
actual case study

Topics
Overview of CFD workflow
Turbulent flows and turbulence modeling
A case study
Overview of CFD workflow
Overview of CFD workflow

What is CFD?

Computational Fluid Dynamics (CFD) is is a branch of fluid mechanics that uses


numerical methods and algorithms to solve and analyze problems that involve fluid
flows.

CFD simulation of the flow around


a F1 car (courtesy of Ferrari S.p.A)
Overview of CFD workflow

What is CFD?

Computational Fluid Dynamics (CFD) is is a branch of fluid mechanics that uses


numerical methods and algorithms to solve and analyze problems that involve fluid
flows.

CFD simulation of the wind flow


around the China Expo Pavillion
(courtesy of FM-ingegneria S.p.A.)
Overview of CFD workflow

How does CFD work?

The CFD workflow is made up of three main steps that allows to obtain a numerical
solution to the real fluid mechanic problem.

Pre-processing: Solving: Post-processing:


Problem geometry and Numerical approximation of Visualization of numerical results
computational domain setup integral and differentiation via contour maps, vector fields
operators and streamlines
Grid generation
Discretization of flow governing Integration of numerical results
Physical and mathematical model equations
setup Validation of the CFD solution
Solution of discrete transport
Fluid properties and boundary equations via numerical algorithms
conditions setup
Overview of CFD workflow

How does CFD work?

The CFD workflow is made up of three main steps that allows to obtain a numerical
solution to the real fluid mechanic problem.

Pre-processing: Solving: Post-processing:


Problem geometry and Numerical approximation of Visualization of numerical results
computational domain setup integral and differentiation via contour maps, vector fields
operators and streamlines
Grid generation
Discretization of flow governing Integration of numerical results
Physical and mathematical model equations
setup Validation of the CFD solution
Solution of discrete transport
Fluid properties and boundary equations via numerical algorithms
conditions setup
Overview of CFD workflow

How does CFD work?

The CFD workflow is made up of three main steps that allows to obtain a numerical
solution to the real fluid mechanic problem.

Pre-processing: Solving: Post-processing:


Problem geometry and Numerical approximation of Visualization of numerical results
computational domain setup integral and differentiation via contour maps, vector fields
operators and streamlines
Grid generation
Discretization of flow governing Integration of numerical results
Physical and mathematical model equations
setup Validation of the CFD solution
Solution of discrete transport
Fluid properties and boundary equations via numerical algorithms
conditions setup
Overview of CFD workflow

How does CFD work?

The CFD workflow is made up of three main steps that allows to obtain a numerical
solution to the real fluid mechanic problem.

Pre-processing: Solving: Post-processing:

Example of CFD workflow for the simulation of the flow around a GT car
(courtesy of Pagani Automobili S.p.A.)
Overview of CFD workflow

What are the pros of CFD?

The advantages of CFD mainly follow from the inherent virtual nature of the
technology.

Main advantages
Time and cost cut down in new designs and
prototyping
Access to challenging, hazardous or even
impossible experimental conditions
Virtually unlimited level of accuracy and detail

Experimental testing of a fighter


jet in a wind tunnel
Overview of CFD workflow

What are the cons of CFD?

The limits of CFD follow from the approximations involved in the CFD workflow and
from the complexity of fluid mechanics

and limits
The physical and mathematical model might be
approximated
Discretization techniques and numerical
algorithms lead to numerical errors

Most of flows in nature are turbulent

Numerical simulation of a
turbulent jet
Turbulent flows and
turbulence modeling
Turbulent flows and turbulence modeling

What is turbulence?

Fluid flow in which the fluid undergoes irregular1 fluctuations, or mixing2. The
speed of the fluid at a point is continuously undergoing changes3 in magnitude and
direction, which results in swirling and eddying4 as the bulk of the fluid moves in a
specific direction*

Turbulence characteristics
1. Chaotic process
2. Enhanced mixing
3. Unsteady phenomena
4. Vortical structures

*from Encyclopedia Britannica Online Visualization of a turbulent jet


Turbulent flows and turbulence modeling

Why turbulence is so important?

Turbulent flows are of outstanding practical relevance, because from the


engineering standpoint turbulence is the rule, not the exception!

Laminar dispersion
Turbulence and transport
phenomena:
Mass transport High concentration
enhancement
Momentum exchange
Turbulent dispersion
enhancement
Heat transfer enhancement

Low concentration

Dispersion of a passive scalar


Turbulent flows and turbulence modeling

When is a flow turbulent? (1/3)

The Reynolds experiment shows that the transition from a laminar to a turbulent
flow takes place depending on the values of the following non-dimensional group:

(Reynolds number)

where
velocity scale
length-scale
kinematic viscosity

Reynolds experiment
Turbulent flows and turbulence modeling

When is a flow turbulent? (2/3)


Turbulent flows and turbulence modeling

When is a flow turbulent? (3/3)

Flow regimes at various Reynolds


number for a circular cylinder in
crossflow
Turbulent flows and turbulence modeling

Is turbulence good or bad? (1/2)

Turbulent

Laminar
Turbulent flows and turbulence modeling

Is turbulence good or bad? (2/2)

Laminar Turbulent
Turbulent flows and turbulence modeling

Turbulence and chaos (1/3)

The governing equations of turbulence are known and established, so why


turbulence is so difficult to predict?:

because the Navier-Stokes equations represent a chaotic system!


Turbulent flows and turbulence modeling

Turbulence and chaos (2/3)

A chaotic system is a physical process exhibiting a sensitive dependence from


initial and boundary conditions

Deterministic process Bullet


trajectory

Stochastic process Light bulb


lifespan
Turbulent flows and turbulence modeling

Turbulence and chaos (3/3)

The sensitive dependence to initial and boundary conditions is due to the effect of
non-linear terms in the deterministic equations

Chaotic system

Diverging behavior of a nonlinear system


Turbulent flows and turbulence modeling

Statistical description of turbulence

Since the Navier-Stokes equations are sensitive to initial and boundary conditions,
a statistical description is introduced:

Average

Variance

Correlation
Turbulent flows and turbulence modeling

The RANS equations (1/2)

The statistical description of turbulence can be extended to the flow governing


equations via the Reynolds decomposition

Reynolds decomposition
Turbulent flows and turbulence modeling

The RANS equations (2/2)

Once the Reynolds decomposition for the field variables in introduced, the RANS
equations are obtained by taking the average of the original governing equations,
giving:

Reynolds-stress tensor

Turbulent heat flux


Turbulent flows and turbulence modeling

Turbulence spectrum (1/3)

Flow
Turbulent flows and turbulence modeling

Turbulence spectrum (2/3)

Fourier series

Amplitude

Period

Phase
Turbulent flows and turbulence modeling

Turbulence spectrum (3/3)

Production range

Inertial range

Dissipative range
Turbulent flows and turbulence modeling

Turbulent scales

In homogeneous and isotropic turbulence, it is possible to assume that the


following hypothesis are valid:

The small scales of turbulence are statistically


isotropic

At high-Reynolds numbers, the characteristics


of the small scales are universal and depends
on viscosity and dissipation only

The characteristics of scales associated with


the inertial range are also universal and
dependent upon the dissipation only
Turbulent flows and turbulence modeling

Numerical simulation of turbulence

The computational cost associated with solving directly the Navier-Stokes


equations for a turbulent flow is dictated by the ratio between the large and the
small scales of turbulence
Turbulent flows and turbulence modeling

Turbulence modeling (1/2)

+ Physics
- Applications
DNS

LES

RANS
- Physics
+ Applications
Turbulent flows and turbulence modeling

Turbulence modeling (2/2)

Spatial and time filtering is used in LES and RANS respectively to reduce the
computational effort

LES model

RANS model
A case study
A case study

Overview (1/2)

We examine the release from a localized source (volumetric) placed within a


staggered array of cubical obstacles of random height using DNS and RANS.

Flow parameters
A case study

Overview (2/2)

Summary of grid resolution

Flow angles
1-MOD 4-MOD

Staggered Staggered
Aligned Aligned
A case study

Simulation tools

Scalar dispersion analysis is performed by means of CFD computations based on


unstructured finite-volume techniques.

Direct Numerical Simulations (Numerical experiments):


Assessment of the reliability of unstructured finite-volume techniques

Establishment of a DNS database for scalar flux modeling

Analysis of turbulent mixing/dispersion in complex flows

Reynolds-Averaged Navier-Stokes simulations (Applications):


Assessment of capabilities/limitations of standard eddy-diffusivity models

Analysis/validation of advanced scalar flux models for scalar dispersion

Analysis of realistic flow conditions/surface topography


A case study

Modeling framework

Flow and scalar computations are performed in a Eulerian framework assuming an


incompressible fluid carrying a passive tracer.

Direct Numerical Simulations (Numerical experiments):

Reynolds-Averaged Navier-Stokes simulations (Applications):


A case study

Computational effort

DNS RANS
8.000.000 cells 1.300.000 cells
240.000 time steps 1000 time steps
8 days@240 cores <1 hour@96 cores
A case study

Turbulence transport modeling

Turbulent transport is usually modeled via ad hoc analogy to the molecular


transport or standard-gradient dffusion hypothesis (SGDH):

where is the eddy-diffusivity and is the


turbulent Schmidt number.

Note that the gradient-transport model


must satisfy the following constraint:

where is the length-scale of turbulent


transport.
A case study

Molecular vs turbulent transport

The SGDH model is unable to represent scalar fluxes in complex flows because of
the failure in the ad hoc analogy between molecular and turbulent transport.

Gradient Molecular transport

Flux

Turbulent transport
Gradient

Flux
A case study

Algebraic flux modeling

Algebraic flux models (AFMs) represent one way to abandon the standard gradient-
diffusion model without introducing the complexity of second-order models:

where is the turbulent kinetic energy, a characteristic time-scale and a model


coefficient. The second-rank tensor represents a tensorial eddy-diffusivity.

A simple way to construct anisotropic scalar flux models is given by the


Generalized Gradient-Diffusion Hypothesis (GGDH):

where time-scale is assumed equal to the mechanical scale of the kinetic


energy dissipation rate.
A case study

Validation of RANS model (1/2)


DNS
RANS
A case study

Validation of RANS model (2/2)

You might also like