Computational Fluid Dynamics
Past, Present and Future
Prof. Antony Jameson
Department of Aeronautics & Astronautics
– Aerospace Computing Laboratory –
Stanford University
[email protected] SciTech 2015
January 7, 2015
1
Outline
I. The History of CFD V.Overview of Numerical Methods
• Van Leer’s View • Typical Requirements of CFD
• Emergence of CFD • A Review of the Literature
• Multi-disciplinary Nature of CFD • DG and related Discontinuous Finite
• Hierarchy of Governing Equations Element methods
• 50 Years of CFD VI. Research on the FR Methodology
• Advances in Computer Power • ESFR
II. Author’s Experience • OFR
• CFD Code Development • Shock detection
• FLO and SYN Codes • Non-linear stabilization via Filtering
• Wing Optimization Using SYN107 VII. Applications
III.Usage of CFD • Transitional Flow over SD7003 Airfoil
• Boeing’s Experience • Study of Flapping Wing Sections
• Airbus’s Experience • Flapping Wing Aerodynamics
IV. Current & Future Trends • Flow Over Spheres
• The Current Status of CFD VIII. LES Computations
• The Future of CFD (?) • Taylor-Green Vortex
• Large-Eddy Simulation • Flow past a Square Cylinder
IX.Summary and Conclusions
2 A. Jameson SciTech, January 7, 2015
Context
“When I die and go to Heaven there are two matters
on which I hope enlightenment. One is quantum
electrodynamics and the other is turbulence.
About the former, I am really rather optimistic.”
Sources: Wikipedia.org; NASA.gov; Hubblesite.org; et al.; H. Lamb (1932)
3 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VI. Research on the FR Methodology
• Van Leer’s View VII. Applications
• Emergence of CFD VIII. LES Computations
• Multi-disciplinary Nature of CFD IX. Summary and Conclusions
• Hierarchy of Governing Equations
• 50 Years of CFD
• Advances in Computer Power
II. Author’s Experience
III. Usage of CFD
IV. Current & Future Trends
V. Overview of Numerical Methods
4 A. Jameson SciTech, January 7, 2015
The History of CFD
History of CFD in Van Leer’s View
5 A. Jameson SciTech, January 7, 2015
The History of CFD
Emergence of CFD
• In 1960 the underlying principles of fluid dynamics and the formulation of the
governing equations (potential flow, Euler, RANS) were well established
• The new element was the emergence of powerful enough computers to make
numerical solution possible – to carry this out required new algorithms
• The emergence of CFD in the 1965–2005 period depended on a combination of
advances in computer power and algorithms.
Some significant developments in the ‘60s:
• birth of commercial jet transport – B707 & DC-8
• intense interest in transonic drag rise phenomena
• lack of analytical treatment of transonic aerodynamics
• birth of supercomputers – CDC6600
Sonic line
Shock wave
M<1 M>1
Boundary layer
DC-8 Transonic Flow CDC6600
6 A. Jameson SciTech, January 7, 2015
The History of CFD
Multi-Disciplinary Nature of CFD
7 A. Jameson SciTech, January 7, 2015
The History of CFD
Hierarchy of Governing Equations
8 A. Jameson SciTech, January 7, 2015
The History of CFD
50 Years of CFD
• 1960–1970: Early Developments
Riemann-based schemes for gas dynamics (Godunov), 2nd-order dissipative schemes for
hyperbolic equations (Lax-Wendroff), efficient explicit methods for Navier-Stokes
(MacCormack), panel method (Hess-Smith)
• 1970–1980: Potential Flow Equations
type-dependent differencing (Murman-Cole), complex characteristics (Garabedian),
rotated difference (Jameson), multigrids (Brandt), complete airplane solution
(Glowinsky)
• 1980–1990: Euler and Navier-Stokes Equations
oscillation control via limiters (Boris-Book), high-order Godunov scheme (van Leer),
flux splitting (Steger-Warming), shock capturing via controlled diffusion (Jameson-
Schmit-Turkel), approximate Riemann solver (Roe), total variation diminishing
(Harten), multigrids (Jameson, Ni), solution of complete airplane (Jameson-Baker-
Weatherill)
• 1990–2000: Aerodynamic Shape Optimization
adjoint based control theory
• 2000–2010: Discontinuous Finite Element Methods
Discontinuous Galerkin, Spectral Difference, Flux Reconstruction, etc.
9 A. Jameson SciTech, January 7, 2015
The History of CFD
Advances in Computer Power
1970 CDC6600 1 Megaflops 106
Cray 1
1980 100 Megaflops 108
Vector Computer
IBM SP2
1994 10 Gigaflops 1010
Parallel Computer
2007 Linux Clusters 100 Teraflops 1014
(affordable) Box Cluster in my house
2007 Four 3 GHz dual core CPUs (24 Gigaflops peak) 2.5 Gigaflops 2.5×109
$10,000
HP Pavilion Quadcore Notebook
2009 1 Gigaflops 109
$1,099
MacBook Pro Quadcore Laptop
2011 2.5 Gigaflops 2.5×109
$2,099
10 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VIII. LES Computations
II. Author’s Experience IX. Summary and Conclusions
• CFD Code Development
• FLO and SYN Codes
• Wing Optimization Using SYN107
III. Usage of CFD
IV. Current & Future Trends
V. Overview of Numerical Methods
VI. Research on the FR Methodology
VII. Applications
11 A. Jameson SciTech, January 7, 2015
Author’s Experience
CFD Code Development
• 1970–1980: Potential Flows
solution of inverse problem by conformal mapping (SYN1), solution of 2D potential
flow by conformal mapping (FLO1), 2D transonic potential flow using rotated difference
scheme (FLO6), first transonic potential flow solution for a swept wing (FLO22), 3D
potential flow in general grid with trilinear isoparametric elements (FLO27), multigrid
solution of 2D transonic potential flow (FLO36)
• 1980–1990: Euler & Navier-Stokes Equations
solution of 3D Euler (FLO57), multigrid solution of 3D Euler (FLO67), multigrid
solution of 2D Euler (FLO82), first solution of Euler equations for a complete aircraf
with tetrahedral meshes (FLOPLANE), cell-vertex and cell-centered schemes for 3D
Navier-Stokes (FLO107)
• 1990–2000: Aerodynamic Shape Optimization
airfoil design via control theory using 2D Euler (SYN83), wing design using 3D Euler
(SYN88), airfoil design using 2D Navier-Stokes (SYN103), wing design using 3D
Navier-Stokes (SYN107), aerodynamic design of complete aircraft with tetrahedral
mesh (SYNPLANE), viscous flow solution on arbitrary polyhedral meshes (FLO3XX)
• 2000–2010: High-order Methods for Navier-Stokes Equations
high-order discontinuous finite element methods for unsteady compressible Navier-
Stokes equations on unstructured meshes (Spectral Difference Method, Energy Stable
Flux Reconstruction Method)
12 A. Jameson SciTech, January 7, 2015
Author’s Experience
Wing Optimization Using SYN107
State of the Art Wing Design
Process in 2 Stages, starting
from Garabedian-Korn Airfoil and
NASA Common Research Model
13 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD IX. Summary and Conclusions
II. Author’s Experience
III. Usage of CFD
• Boeing’s Experience
• Airbus’s Experience
IV. Current & Future Trends
V. Overview of Numerical Methods
VI. Research on the FR Methodology
VII. Applications
VIII. LES Computations
14 A. Jameson SciTech, January 7, 2015
Usage of CFD – Boeing's Experience
Impact of CFD on Configuration Lines & Wind Tunnel Testing
15 A. Jameson SciTech, January 7, 2015
Usage of CFD – Boeing's Experience
Impact of CFD on B737-300 Program
16 A. Jameson SciTech, January 7, 2015
Usage of CFD – Boeing's Experience
Computational Methods at Boeing
TRANAIR:
• Full Potential with directly coupled Boundary Layer
• Cartesian solution adaptive grid
• Drela lag-dissipation turbulence model
• Multi-point design/optimization
Navier-Stokes Codes:
• CFL3D – Structured Multiblock Grid
• TLNS3D – Structured Multiblock Grid, Thin Layer
• OVERFLOW – Overset Grid
N-S Turbulence Models:
• S-A Spalart-Allmaras
• Menter’s k-ω SST
17 A. Jameson SciTech, January 7, 2015
Usage of CFD – Boeing's Experience
CFD Contributions to B787
18 A. Jameson SciTech, January 7, 2015
Usage of CFD – Airbus' Experience
CFD Development for Aircraft Design
19 A. Jameson SciTech, January 7, 2015
Usage of CFD – Airbus' Experience
Block-Structured RANS Capability: FLOWer
20 A. Jameson SciTech, January 7, 2015
Usage of CFD – Airbus' Experience
Unstructured RANS Capability: TAU
21 A. Jameson SciTech, January 7, 2015
Usage of CFD – Airbus' Experience
Numerical Flow Simulation
22 A. Jameson SciTech, January 7, 2015
Usage of CFD – Airbus' Experience
CFD Contribution to A380
23 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VIII. LES Computations
II. Author’s Experience IX. Summary and Conclusions
III. Usage of CFD
IV. Current & Future Trends
• The Current Status of CFD
• The Future of CFD (?)
• Large-Eddy Simulation
V. Overview of Numerical Methods
VI. Research on the FR Methodology
VII. Applications
24 A. Jameson SciTech, January 7, 2015
Current Status & Future Trends
The Current Status of CFD
• Worldwide commercial and government codes are based on
algorithms developed in the ‘80s and ‘90s
• These codes can handle complex geometry but are generally
limited to 2nd order accuracy
• They cannot handle turbulence without modeling
• Unsteady simulations are very expensive, and questions over
accuracy remain
25 A. Jameson SciTech, January 7, 2015
Current Status & Future Trends
The Future of CFD (?)
CFD has been on a plateau for the past 15 years
• Representations of current state of the art:
‣ Formula 1 cars
‣ Complete aircrafts
• The majority of current CFD methods are not adequate for vortex
dominated and transitional flows:
‣ Rotorcraft
‣ High-lift systems
‣ Formation flying
26 A. Jameson SciTech, January 7, 2015
Current Status & Future Trends
Large-Eddy Simulation
The number of DoF for an LES of turbulent flow over an airfoil scales as
Rec1.8 (resp. Rec0.4) if the inner layer is resolved (resp. modeled)
Rapid advances in computer hardware should make LES feasible
within the foreseeable future for industrial problems at high
Reynolds numbers. To realize this goal requires
• high-order algorithms for unstructured meshes (complex geometries)
• Sub-Grid Scale models applicable to wall bounded flows
• massively parallel implementation
Chapman (1979), AIAA J. 17(12)
27 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VII. Applications
II. Author’s Experience VIII. LES Computations
III. Usage of CFD IX. Summary and Conclusions
IV. Current & Future Trends
V. Overview of Numerical Methods
• Typical Requirements of CFD
• A Review of the Literature
• Discontinuous Galerkin and related
discontinuous Finite Element methods
VI. Research on the FR Methodology
28 A. Jameson SciTech, January 7, 2015
Overview of Numerical Methods
Typical Requirements of CFD
Traditional numerical schemes for engineering problems
are too dissipative and do not provide sufficient
accuracy for LES and DNS
• Accuracy: solution must be right
• Small numerical dissipation: unsteady flow features
• Unstructured grids: complex geometries
• Numerical flux: wave propagation problems
• High resolution capabilities: transitional and turbulent flows
• Efficiency: code parallelism
• ...
29 A. Jameson SciTech, January 7, 2015
Overview of Numerical Methods
A Review of the Literature
Past Research on DG Schemes:
• Modern development of DG schemes for hyperbolic conservation laws stems
from the work of Cockburn & Shu [1989a,1989b,1990,1998,2001]
Recent Research:
Attempts to reduce complexity and avoid quadrature:
• Spectral Difference (SD) scheme by Kopriva & Kolias [1996], Liu, Vinokur &
Wang [2006]
• Nodal Discontinuous Galerkin (NDG) scheme by Atkins & Shu [1998],
Hesthaven & Warburton [2007]
• Flux Reconstruction (FR) scheme by Huynh [2007,2009]
Cockburn, et al. (1989). J. Comput. Phys., 84(1); Cockburn, Shu (1989). Math. Comput., 52; Cockburn, et al. (1990). Math.
Comput., 54(190); Cockburn, Shu (1998). J. Comput. Phys., 141; Cockburn, Shu (2001). J. Sci. Comput., 16; Kopriva, Kolias
(1996). J. Comput. Phys., 125(1); Liu, et al. (2006). J. Comput. Phys., 216(2); Atkins, Shu (1998). AIAA J., 36(5); Hesthaven,
Warburton, (Springer Verlag, 2007); Huynh, (2007). AIAA P., 2007-4079; Huynh, (2009) AIAA P., 2009-403
30 A. Jameson SciTech, January 7, 2015
Overview of Numerical Methods
DG, NDG, SD, FR
Main Similarities Between Schemes:
• Local discretization with element-wise polynomials
• Inter-element communication achieved through Riemann solver across interfaces
Main Differences Between Schemes:
• Discontinuous Galerkin: weak formulation requires the use of high-order quadrature
rules to find flux discretization
• Nodal DG: discretizes flux in the same way solution is discretized: element-wise
polynomials. Flux and solution points are coincident and located at the Gauss-Lobatto
points
• Spectral Difference: uses differential formulation. Flux is discretized with element-wise
polynomials one order higher than those used to discretize the solution. Solution and
flux points are collocated.
• Flux Reconstruction: uses differential formulation. Flux and solution discretized with
element-wise polynomials of the same order. Correction functions correct the flux and
are polynomials of one order higher. Flux and solution points are coincident. Recovers
previous schemes, hence facilitating analysis and comparison.
31 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VII. Applications
II. Author’s Experience VIII. LES Computations
III. Usage of CFD IX. Summary and Conclusions
IV. Current & Future Trends
V. Overview of Numerical Methods
VI. Research on the FR Methodology
• Energy stability
• Optimality of dispersion and dissipation
properties
• Shock detection
• Filtering for non-linear stabilization
32 A. Jameson SciTech, January 7, 2015
The FR Methodology
Linear Energy Stability
• Energy stability analysis versus Fourier stability analysis
‣ Energy method is more general and rigorous
‣ Energy method enables stability proofs for all orders of accuracy
‣ Energy method applies to non-uniform meshes
‣ Fourier analysis provides more detailed information about the distribution
of dispersive and diffusive errors
‣ Fourier analysis identifies super accuracy for linear problems
• There exists a family of Flux Reconstruction schemes that are
guaranteed to be linearly stable
‣ Parameterized with a constant c
‣ Recover NDG, SD, plus other previously-found energy-stable FR schemes
‣ c changes the scheme, hence dispersion and dissipation properties too
33 A. Jameson SciTech, January 7, 2015
The FR Methodology
Optimal Flux Reconstruction (OFR)
Effective wavenumber
OESFR - optimized by varying c
OFR - optimized by modifying
zeros of correction function
Imaginary part of numerical
wavespeed
Asthana et al. (2014). J. Comput. Phys.
34 A. Jameson SciTech, January 7, 2015
The FR Methodology
Optimal Flux Reconstruction (OFR)
Advection of a Gaussian
bump, P = 5
- DG, OESFR: 61 elements
- OFR: 45 elements
- c+: 76 elements
OESFR - optimized by varying c
OFR - optimized by modifying
zeros of correction function
Asthana et al. (2014). J. Comput. Phys.
35 A. Jameson SciTech, January 7, 2015
The FR Methodology
Shock Detection with the Concentration Method
Based on concentration
method by Gelb, Gates,
and Tadmore. Adapted
to polynomials by
Sheshadri.
Steps:
1. Find polynomial
modal decomposition
(coefficients of Jacobi
or Chebyshev bases)
2. “Enhance”
decomposition via
convolution with Kernel
3. If magnitude of
coefficients is above a
selected threshold, a
discontinuity is present
Sheshadri, (2014) AIAA P., 2014-2688
36 A. Jameson SciTech, January 7, 2015
The FR Methodology
Filtering for non-linear stabilization
Dissipation can
stabilize FR
schemes.
Filtering can be
posed as
dissipation.
Local Fourier-
Spectral (LFS)
filters developed Burgers Equation, N = 3, P = 119 Shu-Osher shock-turbulence interaction, N = 56, P = 8
by Asthana et
al. perform
exact
convolution
locally and take
neighboring
information into
account.
Double Mach reflection, N = 56x224, P = 8 Asthana et al., (2014) submitted to JCP
37 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VIII. LES Computations
II. Author’s Experience IX. Summary and Conclusions
III. Usage of CFD
IV. Current & Future Trends
V. Overview of Numerical Methods
VI. Research on the FR Methodology
VII. Applications
• Transitional Flow over SD7003 Airfoil
• Study of Flapping Wing Sections
• Flapping Wing Aerodynamics
• Flow Over Spheres
38 A. Jameson SciTech, January 7, 2015
Applications
Transitional Flow over SD7003 Airfoil
SD scheme, N=4
Freestream Separation Transition Reattach.
Turbulence xsep/c xtr/c xr/c
Radespiel
et al.
0.08% 0.30 0.53 0.64
Ol et al. 0.10% 0.18 0.47 0.58
Galbraith
Visbal
0% 0.23 0.55 0.65
Uranga
et al.
0% 0.23 0.51 0.60
Present
ILES*
0% 0.23 0.53 0.64
Experiments in green
Iso-Q colored by Ma
Re=6×104, AoA=4°, 2.2×107 DoF
*1.7×106 DoF
Castonguay, et al. (2010). AIAA P., 2010-4626; Radespiel, et al. (2007). AIAA J., 45(6); Ol, et al. (2005). AIAA P., 2005-5149;
Galbraith, Visbal (2008). AIAA P., 2008-225; Uranga, et al. (2009). AIAA P., 2009-4131;
39 A. Jameson SciTech, January 7, 2015
Applications
Study of Flapping Wing Sections
Experiment (Jones, et al.)
NACA0012, Re=1850, Ma=0.2,
SD, 2D, N=5 on deforming grid St=1.5, ω=2.46, h=0.12c
Jones, et al. (1998). AIAA J., 36(7)
40 A. Jameson SciTech, January 7, 2015
Applications
Flapping Wing Aerodynamics
Iso-Entropy colored by Ma Iso-Entropy colored by Ma
Flapping NACA0012, Re=2000, Wing-Body, Re=5000,
SD N=5, 4.7×106 DoF SD N=4, 2.1×107 DoF
Ou, et al. (2011). AIAA P., 2011-1316; Ou, Jameson (2011). AIAA P., 2011-3068
41 A. Jameson SciTech, January 7, 2015
Applications
Flow Over Spheres
Mach contours + streamlines Iso-Vorticity colored by Mach
Flow over a spinning sphere, Flow over a sphere,
Re=300, Ma=0.2 Re=10000, Ma=0.2
Ou, et al. (2011). AIAA P., 2011-3668
42 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VIII. LES Computations
II. Author’s Experience • Taylor-Green Vortex
III. Usage of CFD • Flow past a Square Cylinder
IX. Summary and Conclusions
IV. Current & Future Trends
V. Overview of Numerical Methods
VI. Research on the FR Methodology
VII. Applications
43 A. Jameson SciTech, January 7, 2015
LES Computations
Taylor Green Vortex: ReD = 21400
Energy spectra for different FR-
derived schemes, compared to a
spectral method on a 5123 grid,
at t = 9 sec
Dissipation rate over time of FR-
derived SD scheme compared to a
13-point Dispersion-Relation-
Preserving (DRP) Finite
Difference scheme on a 5123 grid.
The number of degrees of
freedom remain constant in the
different schemes. 8th order
scheme becomes unstable because
of aliasing.
Iso-Q colored by velocity magnitude
44 A. Jameson SciTech, January 7, 2015
LES Computations
Flow past a Square Cylinder: ReD = 21400
3
iso-Q colored by velocity
x/D = 1.0
x/D = 1.5
x/D = 2.0
x/D = 2.5
x/D = 3.0
2
y/D
-1
-2
-3
hui/Ub
0 1 2 3 4 5
3
x/D = 1.0
x/D = 1.5
x/D = 2.0
x/D = 2.5
x/D = 3.0
2
y/D
-1
-2
-3
hu0 u0 i/Ub2
0 0.4 0.8 1.2 1.6
3
x/D = 1.0
x/D = 1.5
x/D = 2.0
x/D = 2.5
x/D = 3.0
2
y/D
-1
-2
-3
hu0 v 0 i/Ub2
0 0.2 0.4 0.6 0.8
45 A. Jameson SciTech, January 7, 2015
Outline
I. The History of CFD VIII. LES Computations
II. Author’s Experience IX. Summary and Conclusions
III. Usage of CFD
IV. Current & Future Trends
V. Overview of Numerical Methods
VI. Research on the FR Methodology
VII. Applications
46 A. Jameson SciTech, January 7, 2015
Summary and Conclusions
Predicting the future is generally ill advised.
However, the following are the author’s opinions:
• The early development of CFD in the Aerospace Industry was primarily driven by the
need to calculate steady transonic flows: this problem is quite well solved
• CFD has been on a plateau for the last 15 years with 2nd-order accurate FV methods for
the RANS equations almost universally used in both commercial and government codes
which can treat complex configurations. These methods cannot reliably predict complex
separated, unsteady and vortex dominated flows
• Ongoing advances in both numerical algorithms and computer hardware and software
should enable an advance to LES for industrial applications within the foreseeable future
• Research should focus on high-order methods with minimal numerical dissipation for
unstructured meshes to enable the treatment of complex configurations
47 A. Jameson SciTech, January 7, 2015
Summary and Conclusions
Current obstacles to the wider adoption of high-order methods which call for
further research include:
• slow convergence for steady state problems - this might be alleviated by a better design
of a multi-hp convergence acceleration scheme
• the need for a more efficient implicit time stepping scheme for unsteady problems
• more robust high-order schemes for nonlinear problems such as are encountered in high
speed gas dynamics
• more efficient and user friendly mesh generation techniques
48 A. Jameson SciTech, January 7, 2015
Summary and Conclusions
Current issues in LES include:
• the need for wall models to enable simulations of wall bounded flows at affordable
computational costs
• the need for further research on subgrid filtering techniques on unstructured meshes
• the need for continuing research on subgrid models, including approximate deconvolution
and exact SGS models, and a careful evaluation of implicit LES methods
Automatic shape design methods based on control theory or other optimization
methods will be increasingly used in aerospace design
Design problems in unsteady flow, such as turbomachinery, rotorcraft, or unsteady
separated flows are particularly challenging
49 A. Jameson SciTech, January 7, 2015
Summary and Conclusions
Eventually DNS may become feasible for high Reynolds number flows
hopefully with a smaller power requirement than a wind tunnel
50 A. Jameson SciTech, January 7, 2015
The Aerospace Computing Laboratory
Acknowledgement
The current research is a combined effort by
• Postdocs: Jonathan Bull, Guido Lodato
• Ph.D. students: Manuel López, Kartikey Asthana, Abhishek Sheshadri,
Jacob Cabrill, Joshua Romero, Jerry Watkins, David Manosalvas
It is made possible by the support of
• the Air Force Office of Scientific Research under grants FA9550-10-1-0418
and FA9550-14-1-0186 monitored by Dr. Fariba Fahroo
• the National Science Foundation under grants 0708071 and 0915006
monitored by Dr. Leland Jameson
• Stanford Graduate Fellowship
51 A. Jameson SciTech, January 7, 2015
Questions & Anwers
Thank you for listening
52 A. Jameson SciTech, January 7, 2015