1. report date 2. report type briefing chartsbriefing charts presented at 67th american physical...
TRANSCRIPT
REPORT DOCUMENTATION PAGE Form Approved
OMB No. 0704-0188 Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing this collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number. PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
1. REPORT DATE (DD-MM-YYYY) November 2014
2. REPORT TYPEBriefing Charts
3. DATES COVERED (From - To) November 2014- November 2014
4. TITLE AND SUBTITLE
5a. CONTRACT NUMBER N/A
Optimal Numerical Schemes for Time Accurate Compressible Large Eddy Simulations: Comparison of Artificial Dissipation and Filtering Schemes
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S)
5d. PROJECT NUMBER
Edoh, A., Karagozian, A., Merkle, C. and Sankaran, V. 5e. TASK NUMBER
5f. WORK UNIT NUMBER Q12J
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT NO.
Air Force Research Laboratory (AFMC) AFRL/RQR 5 Pollux Drive. Edwards AFB CA 93524-7048
9. SPONSORING / MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)Air Force Research Laboratory (AFMC) AFRL/RQR 5 Pollux Drive. 11. SPONSOR/MONITOR’S REPORT
Edwards AFB CA 93524-7048 NUMBER(S)
AFRL-RQ-ED-VG-2014-321
12. DISTRIBUTION / AVAILABILITY STATEMENT Distribution A: Approved for Public Release; Distribution Unlimited.
13. SUPPLEMENTARY NOTES Briefing Charts presented at 67th American Physical Society (APS) Meeting, San Francisco, CA, 23-25 Nov, 2014. PA#14565
14. ABSTRACT N/A
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF:
17. LIMITATION OF ABSTRACT
18. NUMBER OF PAGES
19a. NAME OF RESPONSIBLE PERSON
Venke Sankaran
a. REPORT Unclassified
b. ABSTRACT Unclassified
c. THIS PAGE Unclassified
SAR 23 19b. TELEPHONE NO
(include area code)
661-275-5634 Standard Form
298 (Rev. 8-98) Prescribed by ANSI Std. 239.18
November 23-25, 2014 Distribution A – Approved for public release; distribution is unlimited
Optimal Numerical Schemes for Time Accurate Compressible Large Eddy Simulations:
Comparison of Artificial Dissipation and Filtering Schemes
67th American Physical Society (APS) Meeting Division of Fluid Dynamics
San Francisco, CA
Ayaboe Edoh, Ann Karagozian (UCLA) Charles Merkle (Purdue), Venke Sankaran (AFRL)
ENERGY & PROPULSION RESEARCH
LABORATORY
Research Supported by AFOSR (Drs. Fahroo and Li, PMs)
Distribution A – Approved for public release; distribution is unlimited
Challenges: Reactive LES
CHARLES (Stanford)
LESLIE3D (Georgia Tech)
OpenFOAM (OpenCFD)
Fluent (Ansys)
Ref: 2013 - Cocks, Sankaran, Soteriou, “Is LES of reacting flow predictive? Part 1: Impact of Numerics”
Need to determine BEST discretization schemes for Reacting LES
Algorithm comparisons: - identical subgrid modeling - differences reside in numerics
2
Distribution A – Approved for public release; distribution is unlimited
Objectives GOAL:
damp high frequency errors while preserving low wave content (ie: low-pass response)
1) compare damping character of Artificial Dissipation and Filtering 2) formulate filter as an equivalent Artificial Dissipation scheme - consequence of filter damping for stiff problems 3) insight on achieving “ideal” low-pass response for general problems • von Neumann Analysis
• Crank-Nicolson w/ 6th order central differencing
3
von Neumann Analysis
Eigenvalues of the amplification matrix specify growth factor and phase errors.
Phase Error Growth Factor
Qn+1 = G θ( )Qn
∂Q∂t
+ A ∂Q∂x
= 0
gi 2φ
φexact=− tan−1 Im gi( ) / Re gi( ){ }
CFLi ×θ
Qi = Q̂ k( )eikxk∑
Qi+i = Q̂ k( )eik x+Δx( )
k∑ = Q̂ k( )eikΔxeikx
k∑
1D Euler System (quasi-linear form):
kΔx = θ with θ ∈ −π ,π[ ]
4 Distribution A – Approved for public release; distribution is unlimited
Artificial Dissipation ∂Q∂t
+ A ∂Q∂x
= −1( )m−1 Δx( )2m−1 ε2m λu+c∂2mQ∂x2mm
∑
ε2 = 1/ 2ε4 = 1/12ε6 = 1/ 60
Qi+1 −Qi−1
2Δx− 12
Qi+1 − 2Qi +Qi−1
Δx⎛⎝⎜
⎞⎠⎟ =
Qi −Qi−1
Δx−Qi+2 + 8Qi+1 − 8Qi−1 +Qi−2
12Δx+ 112
Qi+2 − 4Qi+1 + 6Qi − 4Qi−1 +Qi−2
Δx⎛⎝⎜
⎞⎠⎟ =
4Qi+1 + 6Qi −12Qi−1 + 2Qi−2
12Δx
m = 1: m = 2: etc…
governing equations augmented with dissipation terms
upwind biased stencils
Distribution A – Approved for public release; distribution is unlimited 5
Filtering (Explicit)
represent total amplification of filter scheme as: with
Qi = 1+ S2m Δx( )2m ∂2m
∂x2mm∑⎡
⎣⎢
⎤
⎦⎥Qi
*
G θ( ) = R θ( )G* θ( )
R θ( ) 2≤1
R θ = ±π( ) = 0
NOTE: filter is purely dissipative and does not alter original scheme’s phase behavior
smoothing of solution as post-process of integration step
Distribution A – Approved for public release; distribution is unlimited 6
Filtering (Implicit)
represent total amplification of filter scheme as: with
G θ( ) = R θ( )G* θ( )
R θ( ) 2≤1
R θ = ±π( ) = 0
NOTE: filter is purely dissipative and does not alter original scheme’s phase behavior
smoothing of solution as post-process of integration step
1+ S '2m Δx( )2m ∂2m
∂x2mm∑⎡
⎣⎢
⎤
⎦⎥Qi = 1+ S2m Δx( )2m ∂2m
∂x2mm∑⎡
⎣⎢
⎤
⎦⎥Qi
*
Distribution A – Approved for public release; distribution is unlimited 7
Artificial Dissipation: Growth Factor
2nd Order AD
4th Order AD
6th Order AD
no AD
2nd Order AD
4th Order AD 6th Order AD
no AD
Distribution A – Approved for public release; distribution is unlimited 8
Explicit Filter: Growth Factor
Qi = 1+ Sm Δx( )2m ∂2m
∂x2m
⎡
⎣⎢
⎤
⎦⎥Qi
* with Sm =−1( )m−1
22mShapiro Filter (1975)
500th Order
10th Order
6th Order
2nd Order
4th Order
500th Order
10th Order
6th Order
4th Order
2nd Order
Distribution A – Approved for public release; distribution is unlimited 8
2nd Order Implicit Filter: Growth Factor
1+ 1−δ( )S Δx( )2 ∂2
∂x2
⎡
⎣⎢
⎤
⎦⎥Qi = 1+ S Δx( )2 ∂2
∂x2
⎡
⎣⎢
⎤
⎦⎥Qi
* with δ ∈ 0,1]Long Filter (1971)
δ = 1e−1
δ = 1e− 2
δ = 1e− 3
δ = 1e− 4 δ = 1e− 5
Distribution A – Approved for public release; distribution is unlimited 10
2nd vs. 4th Order Implicit Filters: Growth Factor Error
1+ S Δx( )2 ∂2
∂x2
⎡
⎣⎢
⎤
⎦⎥Qi = 1+ S Δx( )2 ∂2
∂x2 +−δ
4 − 8δ⎛⎝⎜
⎞⎠⎟ S Δx( )4 ∂4
∂x4
⎡
⎣⎢
⎤
⎦⎥Qi
* with δ ∈ 0,14th order Lele Filter (1992)
* 500th Order expFil
decreasing delta
Distribution A – Approved for public release; distribution is unlimited 11
Filtering as an Artificial Dissipation Scheme
Qin+1 = 1+ 1
4Δx( )2 ∂2
∂x2
⎡
⎣⎢
⎤
⎦⎥Qi
*
= 1+ 14
Δx( )2 ∂2
∂x2
⎡
⎣⎢
⎤
⎦⎥ Qi
n − Δt 1−θ( ) ∂En
∂x+ Δtθ ∂E*
∂x⎡
⎣⎢
⎤
⎦⎥ with θ ∈ 0,1[ ]
Qin+1 −Qi
n
Δt+ 1−θ( ) ∂E
n
∂x+ θ( ) ∂E
*
∂x= 14
Δx( )2Δt
∂2Qn
∂x2− 1−θ( ) 1
4Δx( )2 ∂
3En
∂x3− θ( ) 1
4Δx( )2 ∂
3E*
∂x3
dispersive terms: • restore phase of original scheme
dissipation term scales as • increased damping w/ decreasing time-step
ε2 ~ 1 /CFLu+c
Distribution A – Approved for public release; distribution is unlimited 12
Effect of CFL
6th Order expFil
10th Order expFil
AD (CFL = 1e0)
AD (CFL = 1e-1)
* AD shown is 6th order
AD (CFL = 1e-2)
10th Order expFil
6th Order expFil
AD (CFL = 1e0)
AD (CFL = 1e-1)
AD (CFL = 1e-2)
Distribution A – Approved for public release; distribution is unlimited 13
Cumulative Low CFL Damping
CFLu+c = 10−4
Nsteps = 104
6th Order AD
10th Order expFil
6th Order expFil
2nd Order impFil δ = 1e− 7
Distribution A – Approved for public release; distribution is unlimited 14
Summary • Filtering is a form of artificial dissipation
– damping behavior more predictable and tunable • Explicit filters require very high order for low-pass response
– overly dissipative for small time-steps • Implicit filters can be efficiently designed for low-pass response
– superior to artificial dissipation or explicit filters
Acknowledgments guidance from advisors and collaborators: • Dr. Venke Sankaran (Edwards AFRL) • Professor Ann Karagozian (UCLA) • Professor Charles Merkle (Purdue) * support from Dr. Fariba Fahroo and Dr. Chiping Li (AFOSR)
Distribution A – Approved for public release; distribution is unlimited 15
Distribution A – Approved for public release; distribution is unlimited
THANK YOU
Distribution A – Approved for public release; distribution is unlimited
Back-Up Slides
Staggered Grid Von Neumann Analysis
Eigenvalues of the amplification matrix specify growth factor and phase errors.
Continuity/Energy Momentum
Staggered Grid Scheme/ Quasi-Linear Form
Phase Error Growth Factor
1D Euler Eqns
Distribution A – Approved for public release; distribution is unlimited 18
Runge Kutta 4: von Neumann Collocated Staggered
No damping of highest modes
No convection of highest modes
No damping of PARTICLE WAVE’s highest modes Highest ACOUSTIC
modes damped
No convection of PARTICLE WAVE’s highest modes
Slow convection of highest ACOUSTIC modes
Distribution A – Approved for public release; distribution is unlimited 19
Kinetic Energy Preservation (KEP)
∂∂t
12
ui2⎛
⎝⎜⎞⎠⎟+ ∂∂x j
12
ui2uj
⎛⎝⎜
⎞⎠⎟= −
∂ui P∂xi
+ ui
∂τ ij
∂x j
⎛
⎝⎜
⎞
⎠⎟
ui
∂ui
∂t+∂uiu j
∂x j
= − ∂P∂xi
+∂τ ij
∂x j
⎛
⎝⎜
⎞
⎠⎟
⎧⎨⎪
⎩⎪
⎫⎬⎪
⎭⎪
Incompressible Flow: 0j
j
ux
∂=
∂
Compressible Flow: 2
02i
j i i i j ijj j i j
u u u u u u Pt x t x x xρ ρ ρ ρ τ
⎧ ⎫ ⎧ ⎫− ∂ ∂ ∂ ∂ ∂ ∂⎪ ⎪ ⎪ ⎪+ + + + − =⎨ ⎬ ⎨ ⎬∂ ∂ ∂ ∂ ∂ ∂⎪ ⎪ ⎪ ⎪⎩ ⎭ ⎩ ⎭
∂∂t
12ρui
2⎛⎝⎜
⎞⎠⎟+ ∂∂x j
ρuj
ui2
2⎛
⎝⎜⎞
⎠⎟= −ui
∂P∂xi
+ ui
∂τ ij
∂x j
⎛
⎝⎜
⎞
⎠⎟
- K = ½ ui2 bounded and constant at inviscid limit
- KEP schemes satisfy secondary equation discretely - Richtmeyer & Morton (1967) - Arakawa (1966)
- Discrete analogue seeks: - Accurate transport of KE accurate physical transfer of energy: E = KE + Uint
- “in computations of turbulent flow fields, dissipative errors show up at the level of kinetic energy” (Mahesh 2004)
- Robust at inviscid limit (Re ∞)
Distribution A – Approved for public release; distribution is unlimited 20
KEP: Applied to 1D Euler
21
-‐‑ compare Crank-‐‑Nicolson (CN) with Fully KEP scheme (F-‐‑KEP)
-‐‑ discrete secondary equation satisfied to machine zero if KEP
(CN) (F-‐‑KEP)
with
(ρφk )in+1 − (ρφk )i
n
Δt+ 1
Vi
φ( )f∑
f
m(ρ u j ) f
n+1/2 ⋅Si +1Vi
∂pυk , j
∂x j
⎛
⎝⎜
⎞
⎠⎟
f∑
f
n+1/2
⋅Si = 0
11 ( )2
m n nφ φ φ+= +1
1
( ) ( )( ) ( )
n nm
n n
ρφ ρφφρ ρ
+
+
+=
+
2 1 2 21/2 1/2
, , , ,( ) ( ) 1 1( u ) ( )
2 2
mn nn m nk i k i kj f f i k i k j f i
f fi if
S p S RESIDUALt V V
ρφ ρφ φρ φ υ+
+ +⎛ ⎞− + ⋅ + ⋅ =⎜ ⎟Δ ⎝ ⎠∑ ∑
2 12 2 2 2
m m mk k k
i nbrf
φ φ φ⎛ ⎞ ⎛ ⎞ ⎛ ⎞=⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠⎝ ⎠
(Collocated Grid)
Subbareddy/Candler(2009) Merkle (2013)
1
k
ueY
φ
⎡ ⎤⎢ ⎥⎢ ⎥=⎢ ⎥⎢ ⎥⎣ ⎦
Distribution A – Approved for public release; distribution is unlimited
Evaluating KEP: u2
22
Mach = 0.859 disturb = 0.01%
disturb = 1%
- F-KEP always secondary conservative - CN is KEP for low compressibility effects
Distribution A – Approved for public release; distribution is unlimited