Optimisation for CFD

ANSYS R14 Fluids Update Seminar

David Mann, ANSYS UK Ltd.

Milton Park, February 16th, 2012

Sheffield, February 29th, 2012
Aberdeen, March 8th, 2012
1

2011 ANSYS, Inc.

March 2, 2012

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

2011 ANSYS, Inc.

March 2, 2012

Introduction
I have run my analysis, but ..
What happens if I increase/reduce the flow rate?
What do I need to adjust to unify the flow distribution?
How should the geometry change to maximise heat rejection?
What can be to done to mix out the species earlier?
Which parameters have greatest effect on the pressure drop?
What actions can I take to prevent the coolant from boiling?
How do I optimise my design .
3

2011 ANSYS, Inc.

March 2, 2012

Introduction
Optimisation refers to seeking the best possible design point
within the design space
Optimisation is a three-fold problem

Formulation of appropriate parameterisation

Parametric geometry using CAD tool or Design Modeler
Mesh morpher to define parametric mesh deformations

Selection of objective function

What are we seeking to maximise or minimise

Selection of robust optimiser

Gradient based optimisation method (Adjoint Solver)
Direct search based optimisation method (e.g. Simplex Method)
Statistical Optimisation method (ANSYS DesignXplorer)
4

2011 ANSYS, Inc.

March 2, 2012

Introduction
ANSYS provides a
comprehensive set of tools
for optimisation

Manual part mesh replacement

Design of Experiments

Response Surfaces

Goal Seek

Mesh Morphing

Adjoint Solutions

Optimisation can be based around

parametric geometry, arbitrary
freeform mesh deformation, or precise
geometrical mesh deformation
5

2011 ANSYS, Inc.

March 2, 2012

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

2011 ANSYS, Inc.

March 2, 2012

Manual Optimisation

Replace sub zones effectively as parameter for design points.

New in R14.

Allows automatic first order interpolation of face and cell data

Allows automatic grid manipulations (face slitting, node merging,

zone fusing, recreation of mesh interfaces etc)

Can use apriori grid preparation/decomposition but also works

well with cavity re-meshing tool in TGrid for conformal interfaces

Only requires re-meshing of sub zones (replaced part) and larger

mesh can be re-used across design points

Able to cope with large design changes and topology changes

(with consistent zone names)

2011 ANSYS, Inc.

March 2, 2012

Manual Optimisation
Part Swapping : Six
Each
Cavity
different
wing
remesh
has
wing
slightly
zone
configurations
fordifferent
conformal
rotation,
tosub-grid
be compared
position
transplant
and AoA

2011 ANSYS, Inc.

March 2, 2012

Manual Optimisation

Transplant of meshes using script

Design 2

Design 1

Design 3

Design 4

2011 ANSYS, Inc.

March 2, 2012

Design 5

Design 6

Manual Optimisation
Example Script for the case

Replace sub grid command

Path to new sub grid

Repeat for
N designs

Interpolate
New fluid zone name and
zone to be replaced (need
not be the same as here)
10

2011 ANSYS, Inc.

March 2, 2012

face/cell data?

Manual Optimisation

Design1 Design 2 Design 3 Design 4 Design 5 Design 6

11

2011 ANSYS, Inc.

March 2, 2012

Manual Optimisation

Design 1
Drag=97N

Design 2
Drag=137N

Design 3
Drag=131N

12

2011 ANSYS, Inc.

March 2, 2012

Design 4
Drag=141N

Design 5
Drag=139N

Design 6
Drag=173N

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

13

2011 ANSYS, Inc.

March 2, 2012

Gain Deeper Product Insights

ANSYS simulation software can give you more clarity into your products and
development processes
Move from a single point solution to understanding the design space so simulation
can guide design.

?
?
?

Single Point
14

2011 ANSYS, Inc.

March 2, 2012

What If?

Response
Surface

Design Xplorer (DX)

15

Workbench based optimisation tool

Input and output parameters from all types of analysis can be

shared with DX via WorkBench

Works with all ANSYS products from ANSYS structural to EMAG to

CFD, making it beneficial for multiphysics analysis

2011 ANSYS, Inc.

March 2, 2012

Example of DX usage Simple Diffuser

To illustrate how Design Xplorer can be used to optimise a simple

geometry consider the simple diffuser below:-

The diffuser
geometry is
characterised by four
input parameters

This allows the

diffusers shape to be
controlled by
WorkBench and
Design Xplorer

16

2011 ANSYS, Inc.

March 2, 2012

Initial setup and run Current Design Point

The initial run is setup and run as normal

with CFD post then being used to provide
output parameters

This forms the current design point

which can then be rerun with a different
geometry simply by changing the input
parameters, and the new results viewed
via the output parameters

17

2011 ANSYS, Inc.

March 2, 2012

Design of Experiments (DoE)

A Design of Experiments can then be used to

automatically generate a set of runs which cover the
design space specified by the parameters

18

2011 ANSYS, Inc.

March 2, 2012

Response Surface Results

The data from the Design of Experiments runs is then

used to generate a response surface from which the
performance of other designs can be predicted

A clear maximum pressure rise is visible in the results

beyond which additional diffusion causes separation

19

2011 ANSYS, Inc.

March 2, 2012

Goal Driven Optimisation

The response surface is used to predict the

parameters that give the optimum design

The objective function can be multi-variate, but in

this case is simply chosen to maximise the pressure
rise through the diffuser.

Once determined the optimum design can be run to

confirm the predicted results or to improve the
resolution of the response surface

20

2011 ANSYS, Inc.

March 2, 2012

Confirmation Run

21

2011 ANSYS, Inc.

Pressure Rise = 43.465 Pa (43.397 Pa predicted)

March 2, 2012

Design Xplorer (DX)

DesignXplorer features the following studies:

Design point analysis (default) - examines how the input parameters

affect the output parameters by creating designs in a spreadsheet
like view.

Response Surface
- Goal-Driven Optimization or GDO automatically change design
parameters to find optimal design.
- Six Sigma incorporates uncertainties of input parameters.

22

Min-Max Search examines the entire output parameter space from

a Response Surface to approximate the minimum and maximum of
each output parameter. You can perform this search at any time.

Parameter Correlation gives correlation data that has been used to

derive sensitivities and decide if individual sensitivity values are
significant or not.

2011 ANSYS, Inc.

March 2, 2012

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

23

2011 ANSYS, Inc.

March 2, 2012

Introduction
The mesh morphing technology introduced to ANSYS FLUENT at R13
allows a single mesh to be deformed in a freeform way to achieve new
designs without the need to create new geometries or meshes.

This gives us a platform on which we can carry out design optimisation

studies without the need to build a parametric model on a
topologically identical mesh.

24

2011 ANSYS, Inc.

March 2, 2012

Mesh Morpher
Introduction
The FLUENT Mesh Morpher utilizes Bernstein Polynomials to allow smooth
mesh deformations based upon movement of predefined control points.

The Morpher is hooked up to some basic optimization algorithms allowing

shape optimization to be carried out within FLUENT with the following
benefits:
Shape modification carried out quickly in parallel solver
Zero file I/O requirement
Quick convergence - data from previous design point can be used for
subsequent design points so initial data field is close to final solution
Scriptable by Text User Interface journals
Works on all mesh types, i.e. hex, tet, cutcell, hybrid, poly etc.

25

2011 ANSYS, Inc.

March 2, 2012

Mesh Morpher
Case Study Manual Morphing
Generic F1 car (Hexcore) nose extension before

Two control points moved in -x

26

2011 ANSYS, Inc.

March 2, 2012

Mesh Morpher
Case Study Manual Morphing
Generic F1 car (Hexcore) nose extension after

Two control points moved in -x

27

2011 ANSYS, Inc.

March 2, 2012

The morpher deformation modes

In FLUENT the user specifies a deformation region inside which the mesh is
morphed and an array of control points to define the deformation

Modes of deformation are specified by describing how all the control points move
together, for example the sine wave deformation shown

Multiple deformation modes can be specified and the relative weighting of each
controlled by parameters (three deformation modes = 3 parameters)

28

A sufficient number of deformation modes will allow arbitrary shapes to be formed

2011 ANSYS, Inc.

March 2, 2012

Mesh Morpher
Interface Deformation Setup

In R14 we now have the

ability to constrain
boundaries within
deformation regions!

29

2011 ANSYS, Inc.

March 2, 2012

Optimisation methods
Design Xplorer Design of Experiments

Internal FLUENT Simplex Optimiser

Two different optimisation techniques are available when using the FLUENT
morpher:1. Optimisation using Design of Experiments (DoE) in Design Xplorer (DX) requires
script
2. Optimisation using one of the internal FLUENT Mesh Morpher Optimiser (MMO)
methods such as the Simplex optimiser
30

2011 ANSYS, Inc.

March 2, 2012

Mesh Morpher
Optimiser

31

2011 ANSYS, Inc.

March 2, 2012

Mesh Morpher
Objective Functions and Definition

The Objective Function is a single scalar value that the chosen

optimizer method will drive towards a minimum.

Typical Objective Functions

Lift & drag

Mass flow-rate for inlets and outlets
Surface average pressures for walls/inlets/outlets
Min-max absolute pressure/temperature etc.

Objective Function can be defined by:

User defined functions
Scheme Function
NEW IN R14 GUI Driven Objective Function Definition that can call
FLUENT exposed parameters
32

2011 ANSYS, Inc.

March 2, 2012

Case Study 1 - Generic Sedan Drag

Optimization
Study effect of various vehicle shape parameters on drag force
Shape parameters are defined using mesh morphing technology in ANSYS
Fluent
ANSYS WB is used to drive the shape parameters, create DOE & perform
goal driven optimization
ANSYS WB makes the process automatic

Three Shape Parameters

1. Backlight angle (1)
2. Tumble home angle (2)
3. Windshield angle (3)

Generic Sedan Model

33

2011 ANSYS, Inc.

March 2, 2012

Generic Sedan Baseline Design

34

2011 ANSYS, Inc.

March 2, 2012

Generic Sedan Worst Design

35

2011 ANSYS, Inc.

March 2, 2012

Generic Sedan Optimal Design

36

2011 ANSYS, Inc.

March 2, 2012

Case Study 2 Optimisation of NACA0012

37

The initial geometry to be optimised is the NACA0012 symmetrical 2D aerofoil

section.

A grid of 5 rows of 18 control points is superimposed over the mesh to facilitate the
mesh morphing

Moving these control points causes the mesh to morph

A course tet mesh and a finer quad pave mesh were used

Seek to maximise Lift to Drag ratio

2011 ANSYS, Inc.

March 2, 2012

Sine deformation modes

applied to NACA 0012
section

1.5
Mode 1
Mode 2
1

Mode 3
Mode 4
Mode 5

0.5

Mode 6

0
0

10

12

-0.5

Mode n = sin(n.pi.x) where x is normalised

between 0 and 1 from leading edge to trailing edge

-1

-1.5

Mode 1 180o

Mode 2 360o

Mode 3 540o

Mode 4 720o

Mode 5 900o

Mode 6 1080o

38

2011 ANSYS, Inc.

March 2, 2012

14

16

18

Workbench Morpher Optimisation

Project Workflow

39

Workbench and Design Xplorer are used to drive the six morpher
parameters to optimise the design for lift/drag

The workflow is shown below

2011 ANSYS, Inc.

March 2, 2012

Output parameters from CFD Post

40

CFD Post is used to return the lift and drag from the aerofoil as
Workbench output parameters

A new workbench output parameter of Lift/Drag can then be derived in

WB parameter manager for use in the DoE optimisation

2011 ANSYS, Inc.

March 2, 2012

Optimisation using Design Xplorer

41

Data points using six morpher parameters are very noisy in their fit to
the response surface, as there are always four other parameters not
involved in the surface fit

The response surface fit can be poor if the search area is wide, so the
limits on parameters were tightened around the then best point and
the DoE rerun several times to get a better fit

2011 ANSYS, Inc.

March 2, 2012

Optimisation using sine modes and inbuilt

Simplex optimiser

Convergence history using Simplex optimiser

42

2011 ANSYS, Inc.

March 2, 2012

Comparison of optimised aerofoil shapes

with different optimisers
Sine modes using DoE in
Design Xplorer

Parameters {0.018027, 0.0042698, -0.001, 0.00083241, -0.002297,

0.0025603}
CL/CD = 26.746, CL = 7.3694, CD = 0.27553

Sine modes using internal

MMO simplex optimiser

Parameters {0.03000086, 0.00429696, 0.003607369, 0.001349214,

0.0003653693, 0.001880633}
CL/CD = 26.349, CL = 7.5189, CD = 0.28536

Interestingly two completely different sets of parameters were obtained giving very similar but
laterally displaced aerofoil shapes with the same morpher modes
The number of design points required were 3x 45 design points with DX and 148 with the MMO
running the Simplex optimiser. DX required some manual refinement points. For the same CL/CD DX
would have needed less. DX can spawn runs to multiple machines/cores via RSM
43

2011 ANSYS, Inc.

March 2, 2012

Comparison of DX and MMO results

The two optimisation methods ended up with very similar aerofoil sections,
but in each case the geometry ended up in a different y location

The DX route is less prone to get stuck in a local minimum, but requires more
manual fine tuning of limits

0.01

0.035

0.03
0.005
0.025

0
-0.06

-0.04

-0.02

0.02

0.04

0.02

0.06

Sine Modes DX

Sine Modes DX

Sine Modes MMO

Sine Modes MMO

0.015

-0.005
0.01

-0.01
0.005

0
-0.06

-0.015

Optimised designs translated and overlaid

44

2011 ANSYS, Inc.

March 2, 2012

-0.04

-0.02

0.02

0.04

Optimised designs untranslated

0.06

Case Study 3 - Optimisation of a UAV wing

A generic (representative, but not accurate) model of the Global Hawk UAV was downloaded
from http://www.nasa.gov/multimedia/3d_resources/assets/global_hawk.html
The non realistic wing was replaced with one with a realistic section and the correct
planform view derived from open literature relating to this aircraft
The Mesh Morpher Optimiser (MMO) tool in ANSYS Fluent was used to deform the wing
section to optimise the Lift-Drag ratio

45

2011 ANSYS, Inc.

March 2, 2012

Deformation Modes
Mode 1 Decrease/Increase
Aft Thickness

Mode 2 Decrease/Increase
Stagger

Mode 3 Redistribute Camber

Mode 4 Increase/Decrease
Camber
46

2011 ANSYS, Inc.

March 2, 2012

The Optimised Wing Section

1.4
6
5

1.2
4
1

CL

The simplex optimiser was used to find the

combination of the four deformation modes
that gave the maximum lift/drag ratio
The resultant wing section has a CL/CD ratio
24.8% higher than the original at the datum
3 degree Angle of Attack (AoA), and a wider
range of AoA capability
The optimised section is more aft loaded

0
0.4

-1

1
0.6

0
Optimised Section

-1
-2

-2

0.2
0.01

0.8

0.015

0.02

0.025

Original Section
0.03

0.035

CD

0.14

-2

Original PS

0.12

Original SS

-1.5

Optimised

0.1

Baseline
Optimised

0.08

-1

CP

0.06
0.04

-0.5

-0.2

0.2

0.4

0.6

0.8

0.02

Chord

0.5

0.2

0.4

0.6

0.8

-0.02
1

-0.04
-0.06

47

2011 ANSYS, Inc.

1.5

March 2, 2012

Baseline

Optimum

0.04

Mesh Morpher - Conclusions

The inbuilt freeform mesh morpher in Fluent provides a powerful tool for
arbitrary changes in the geometry without being limited by a constrained
parameterised geometry

If sufficient well designed deformation modes are used with the mesh
morpher, any arbitrary shape change can be achieved and true optima can
be approached

This mesh morpher is less suitable for cases where a high degree of control
is needed, for example part of the geometry is a given shape.

48

2011 ANSYS, Inc.

March 2, 2012

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

49

2011 ANSYS, Inc.

March 2, 2012

RBF-Morph
RBF Morph tool
3rd Party Add-on

50

2011 ANSYS, Inc.

March 2, 2012

RBF-Morph
Tool Objective
The aim of RBF Morph is to perform fast mesh morphing using a meshindependent approach based on state-of-the-art RBF (Radial Basis Functions)
techniques.
The use of RBF Morph allows the CFD user to perform shape modifications,
compatible with the mesh topology, directly in the solving stage, just adding a
single command line in the input file:

(rbf-morph (("sol-1" amp-1) ("sol-2" amp-2)...("sol-n" amp-n)))

The final goal is to perform parametric studies of component shapes and positions
typical of the fluid-dynamic design like:
Design Developments
Multi-configuration studies
Sensitivity Studies
DOE (Design Of Experiment)
Optimization

51

2011 ANSYS, Inc.

March 2, 2012

RBF-Morph
Features
Fully integrated within Fluent (GUI, TUI &
solving stage) and Workbench
Mesh-independent RBF fit used for surface
mesh morphing and volume mesh smoothing
Parallel support allows morphing of large
grids in a short time
Support for all mesh types (tetrahedral,
hexahedral, hexcore, polyhedral, etc.)
Ability to generate modified CAD file from
morphed surface mesh
Multi fit makes the Fluent case truly
parametric (only 1 mesh is stored)
High precision morphing : exact nodal
movement and exact feature preservation.

52

2011 ANSYS, Inc.

March 2, 2012

RBF-Morph
Background and Theory
A system of radial functions is used to fit a
solution for the mesh movement/morphing,
from a list of source points and their
displacements. This approach is valid for
both surface shape changes and volume
mesh smoothing. The RBF problem definition
does not depend on the mesh
Radial Basis Function interpolation
is used to derive the displacement
in any location in the space, so it is
also available in every grid node.

Radial Basis Function

(r )

Spline type (Rn)

r , n odd

Thin plate spline (TPSn)

r log r , n even

n
n

Multiquadric(MQ)

1+ r 2
1

Inverse multiquadric (IMQ)

1+ r 2
1
1+ r 2
2
e r

Inverse quadratic (IQ)

An interpolation function composed by a

radial basis and a polynomial is defined.
N

s (x ) = i ( x x i ) + h(x )
i =1

53

2011 ANSYS, Inc.

March 2, 2012

h(x ) = + 1 x + 3 y +

Gaussian (GS)

(
(
(

)
)
)

v x = s x (x) = ix x x ki + 1x + 2x x + 3x y + 4x z

4z
i =1
N

(
)
=
=
v
s
x
iy x x ki + 1y + 2y x + 3y y + 4y z
y

y
i =1

N
v = s (x) = z x x + z + z x + z y + z z

1
2
3
4
z
i
ki
z
i =1

RBF-Morph
How it Works : Problem Setup
The problem must describe correctly the desired changes and
must preserve exactly the fixed part of the mesh.
The prescription of the source points and their displacements fully
defines the RBF Morph problem.
The problem is mesh-independent, and could be defined using grid
nodes as well as arbitrary point locations.
Each problem and its fit define a mesh modifier or a shape
parameter.

54

2011 ANSYS, Inc.

March 2, 2012

RBF-Morph
Solid body motion and exact deformation

55

The main differentiator between RBF-Morph and the inbuilt morpher is that
RBF-Morph allows the mesh to be deformed to give precise geometry
changes, such as solid body motion. The inbuilt morpher is designed for
arbitrary shape change with little constraint.

RBF-Morph is a useful alternative to parameterising the geometry as the

mesh does not need to be recreated.

2011 ANSYS, Inc.

March 2, 2012

RBF-Morph
Industrial Applications

56

2011 ANSYS, Inc.

March 2, 2012

Motorbike driver height and position

The original motorbike model
is parameterized to
investigate the effect of
driver height and position:

57

1.

Changing of driver height

[-5 cm, 0 cm, 5 cm];

2.

Changing of driver
position acting on the
hunching angle
[0,7.5,15];

2011 ANSYS, Inc.

March 2, 2012

Set up of RBF Morph

The morphed action is limited in the
box region domain 1.
The motion of the surfaces inside the
encapsulation domain is imposed
to the points on the windshield
(fixed), the fairing (fixed) and the
helmet (moving).
Driver height is changed moving the
helmet
Driver position is changed rotating
the helmet around the ankle

58

2011 ANSYS, Inc.

March 2, 2012

59
2011 ANSYS, Inc.
March 2, 2012

(Bricomoto, MRA)

Motorbike Windshield

Formula 1 Front Wing

60

2011 ANSYS, Inc.

March 2, 2012

61
2011 ANSYS, Inc.
March 2, 2012

Sails Trim (Ignazio Maria Viola,

University of Newcastle)

Generic Formula 1 Front End

62

2011 ANSYS, Inc.

March 2, 2012

Generic Formula 1 Front End

63

2011 ANSYS, Inc.

March 2, 2012

Fluid Structure Interaction

64

2011 ANSYS, Inc.

March 2, 2012

Turbine Blade
65

2011 ANSYS, Inc.

March 2, 2012

MIRA Reference car

(MIRA ltd)
66

2011 ANSYS, Inc.

March 2, 2012

Conclusions
A shape parametric CFD model can be defined using ANSYS
Fluent and RBF Morph.
Such parametric CFD model can be easily coupled with
preferred optimization tools to steer the solution to an optimal
design that can be imported in the preferred CAD platform
(using STEP)
Proposed approach dramatically reduces the man time
required for set-up widening the CFD calculation capability
M.E. Biancolini, Mesh morphing and smoothing by means of
Radial Basis Functions (RBF): a practical example using Fluent
and RBF Morph in Handbook of Research on Computational
Science and Engineering: Theory and Practice (http://www.csebook.com/).

67

2011 ANSYS, Inc.

March 2, 2012

Using CFX moving mesh for optimisation

68

The preceding mesh

morphing sections have
been on Fluent, but the
moving mesh model in
CFX can also be used as
the basis of an
optimisation

Solid body motion and

deflections can be applied
and the mesh will morph
to accommodate the
movement

Multiple scenarios can

then be run with one
mesh

2011 ANSYS, Inc.

March 2, 2012

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

69

2011 ANSYS, Inc.

March 2, 2012

The Adjoint Solver

The Adjoint solver is an additional solver within Fluent that is run after
the conventional solution is converged

The Adjoint solver is used to assess the sensitivity of output parameters

such as lift, drag or pressure drop to input parameters such as the
geometrical shape without the need for additional runs

The output from the Adjoint

solver is typically a surface
vector field that illustrates how
the geometry would need to
change to increase or decrease
the output parameter of interest

70

As such it forms a very useful

tool in an optimisation study and
can be linked to the mesh
morpher
2011 ANSYS, Inc.

March 2, 2012

Drag sensitivity for NACA0012

Adjoint Solver
Key Ideas - Fundamentals

High-level system view of a conventional flow solver

Inputs

Outputs

Field data
Contour plots
Vector plots
xy-plots
Scalar values
Lift
Drag
Total pressure drop

Boundary mesh
Interior mesh
Material properties
Boundary condition 1
Flow angle
Inlet velocity
@
@

71

2011 ANSYS, Inc.

March 2, 2012

FLOW SOLVER

Adjoint Solver
Fundamentals
HOW ARE CHANGES TO KEY OUTPUTS DEPENDENT ON CHANGES TO
THE INPUTS?
Inputs

Outputs

Field data
Contour plots
Vector plots
xy-plots
Scalar values
Lift
Drag
Total pressure drop

Boundary mesh
Interior mesh
Material properties
Boundary condition 1
Flow angle
Inlet velocity
@
@

72

2011 ANSYS, Inc.

March 2, 2012

?
ADJOINT SOLVER

Adjoint Solver
Key Ideas

73

The Adjoint solver can be used to compute the derivative of a chosen

observation of engineering interest with respect to all the input data for
the system.

Solving an adjoint problem is not trivial about as much effort as a flow

solution.

The adjoint solution provides guidance on the optimal adjustment that

will improve a systems performance.

An adjoint solution can be used to estimate the effect of a change prior

to actually making the change.

Shape sensitivity data can be combined with mesh morphing to guide

smooth mesh deformations.

An adjoint solution can be used as part of a gradient-based optimization

algorithm.

2011 ANSYS, Inc.

March 2, 2012

Adjoint Solver
Mesh Morphing
Completing the design cycle
Mesh Morphing
Sensitivity of lift to surface shape
Use Bernstein polynomial-based
morphing scheme
Adjoint to deformation operation
Surface shape sensitivity becomes
control point sensitivity
Benefit of this approach is two-fold

Smooths the surface sensitivity
field

Provides a smooth interior mesh
deformation
Select portions of the geometry to be
modified

74

2011 ANSYS, Inc.

March 2, 2012

Flow

Adjoint Solver
Mesh Morphing
Constrained motion
Some walls within the control volume may be constrained not to
move.
A minimal adjustment is made to the control-point sensitivity field
so that deformation of the wall is eliminated. Cast as a least-squares
problem.
Actual change 3.1
P = -213.8
Total improvement
of 8%

75

2011 ANSYS, Inc.

March 2, 2012

Adjoint Solver
Current Functionality
ANSYS-Fluent flow solver has very broad scope
Adjoint is configured to compute solutions based on some assumptions

Steady, incompressible, laminar flow.

Steady, incompressible, turbulent flow with standard wall functions.
First-order discretization in space.
Frozen turbulence.

The primary flow solution does NOT need to be run with these restrictions

Strong evidence that these assumptions do not undermine the utility of the adjoint
solution data for engineering purposes.

Fully parallelized.
Gradient algorithm for shape modification

Mesh morphing using control points.

Adjoint-based solution adaption
76

2011 ANSYS, Inc.

March 2, 2012

Adjoint Solver
Example Test Cases S809 HAWT Blade
Objective Maximise Lift/Drag Ratio

77

2011 ANSYS, Inc.

March 2, 2012

Adjoint Solver
Example Test Cases S809 HAWT Blade
The best lift/drag ratio is observed when setting observable for lift, and using a scale factor of 2.5. The new
shape provides about 30% more lift than the original geometry

Highest lift/drag
ratio achieved

78

2011 ANSYS, Inc.

March 2, 2012

Adjoint Solver
Example Test Cases Internal ducting U bend

100
90

Base design

80

ptot [Pa]

70
60
50
40
30
20
10

End design
79

2011 ANSYS, Inc.

March 2, 2012

0
0

10

20

Run [-]

30

Adjoint Solver
Example Test Cases External Aero (Small car)

Surface map of the drag sensitivity to shape changes

Surface map of the drag sensitivity to shape changes

80

2011 ANSYS, Inc.

March 2, 2012

Surface map of the drag sensitivity to shape changes

Adjoint Solver
Example Test Cases External Aero (Full Generic Race Car)
Increase the downforce on the vehicle
Adjoint result shows regions of highest sensitivity of downforce to shape

81

2011 ANSYS, Inc.

March 2, 2012

Adjoint Solver
Example Test Cases External Aero (Full Generic Race Car)
Front Wing changes

Rear Wing Changes

Downforce (N)

82

Downforce (N)

Geometry

Predicted

Result

Geometry

Predicted

Result

Original

---

425.7

Original

---

425.7

Modified

447.4

451.1

Modified

481.3

492.5

2011 ANSYS, Inc.

March 2, 2012

Adjoint Solver Summary

The Adjoint solution is carried out as an addition to the primary flow solution

The Adjoint solver solves the Navier Stokes equations recast as derivatives of
output flow variables of interest such as drag, lift or pressure drop

As the equations are rewritten only a finite number of predefined flow

variables of interest are available

The output from the Adjoint is a field of the sensitivity of inputs such as the
geometry and boundary conditions to the output variables of interest.

This can form the basis for optimising these inputs.

83

2011 ANSYS, Inc.

March 2, 2012

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

84

2011 ANSYS, Inc.

March 2, 2012

The Remote Solve Manager (RSM)

RSM has a three tiered architecture

85

Client

Solver Manager

Compute Server

2011 ANSYS, Inc.

March 2, 2012

The Remote Solve Manager (RSM)

A queue can be set up on

the solver manager for
multiple compute cores
which may be available on
the network

All design points are then

submitted to this queue

Design points are run as

compute cores become
available

WB session can be
detached and reattached

Results picked up by WB
when complete

86

2011 ANSYS, Inc.

March 2, 2012

Workbench RSM Terminology

There are RSM controls at both the solution level, e.g. FLUENT or CFX, to
control how jobs are run, and also at the parameter set (global) level giving
good control of resource utilisation

Solution Component properties

Parameter Set properties

87

2011 ANSYS, Inc.

March 2, 2012

RSM Summary

88

Optimisation can create a large number of design points that

would be slow to run sequentially in serial.

In R14 the Remote Solve Manager provides a way of controlling

compute resource allocation for CFD design point/ optimisation
studies to allow multiple design points to be submitted to a
queue of available compute nodes.

2011 ANSYS, Inc.

March 2, 2012

Agenda
Optimisation Tools for CFD
Introduction
Manual Optimisation and Scripting
Design Xplorer (DX)
Mesh Morpher or Shape Optimiser
RBF-Morph
Adjoint Solver
Remote Solve Manager (RSM)
Summary

89

2011 ANSYS, Inc.

March 2, 2012

Summary
Where should I use these tools?
Manual Optimisation and Scripting

I want high level control on the meshes to be generated and I am

prepared to script the process
Design Xplorer

My optimisation may include a wider process or workflow in

Workbench with input and output parameters coming from multiple
simulations, and/or I am working with parametric geometry. My
design space may have local minima that I wish to avoid. I have a good
idea of nature and limits of the design space and want to enforce strict
control on any changes.

MMO

90

I am interested arbitrary shape optimisation where geometry

parameterisation may limit finding a true optimum. Constraints and
prescribed deformations are secondary concerns. No additional cost.
Can interface with Workbench but scripts needed.

2011 ANSYS, Inc.

March 2, 2012

Summary
Where should I use these tools?
RBF-Morph

I want high level control on the morphed geometry shape, for example
can do solid body transformations and modify geometry in a manner
consistent with using parameterised geometry. Highly beneficial
where remeshing a parametric geometry is costly. Morphed
modifications can be fed back to CAD. Additional cost.

Adjoint Solver

91

I am interested in finding where I could potentially modify my

geometry to make small improvements. Which areas are most
contributing to lift, drag, pressure drop Limitations on models and
objective functions. No additional cost. No interaction with
Workbench, solver only. Geometry deformation is free form as with
the MMO.

2011 ANSYS, Inc.

March 2, 2012

Conclusions

92

ANSYS can provide tools to drive your design to optimise for

chosen parameters

This can be done with geometric parameterisation, mesh

replacement or morphing technologies

Some tools can provide predicted off design performance

without running multiple design points

The Remote Solve Manager allows the efficient handling of

large numbers of design points

We are continually developing these new technologies to be

more efficient

We want to listen to your feedback to improve the usefulness

of the tools

Increase your ROI using the ANSYS optimisation tools

2011 ANSYS, Inc.

March 2, 2012