Beruflich Dokumente
Kultur Dokumente
Amitay Isaacs, D Ghate, A G Marathe, Nikhil Nigam, Vijay Mali, K Sudhakar, P M Mujumdar
Centre for Aerospace Systems Design and Engineering Department of Aerospace Engineering, IIT Bombay http://www.casde.iitb.ac.in
SAROD 2003 1
About CASDE
5 years old Masters program in Systems Design & Engineering MDO MAV Modeling & Simulation Workshops/CEPs/Conferences
SAROD 2003 2
Propulsion system
Engine sizing & cycle design Intake duct design using CFD
SAROD 2003
Problem Formulation
Objective/Constraints Pressure Recovery Distortion Swirl
CFD takes huge amounts of time for real life problems Design requires repetitive runs of disciplinary analyses Parametric geometry modeling Grid generation CFD solution Objective/Constraint function evaluation Optimization Finite difference step size (??), (NDV + 1) analyses required Exact formulations Automatic differentiation (ADIFOR), Adjoint method, Complex step method All require source code
6
Gradient Information
SAROD 2003
Flow Solver
Distortion & Swirl calculation requires NS solution In-house NS Solver
Analytical gradients possible Easy to integrate Gradients using finite difference only Difficult to integrate
FLUENT Inc. S-shaped non-diffusing duct Results validated with a NASA test case (Devaki Ravi Kumar & Sujata Bandyopadhyay)
SAROD 2003
Strategies
Reducing Time
Parameterization Variable fidelity to shrink the search space Surrogate modeling Meshing Parallel computing Continuation Wrapping executables and user interfaces Offline analysis (Surrogate models) semiautomatic
8
SAROD 2003
Our Strategy
SAROD 2003
Our Methodology
CFD analysis at DOE points
Constraints
Optimization
SAROD 2003
10
Parametrization
Y Z
Duct Centerline
X
11
Parametrization
Y Z
Duct Centerline
X
12
Typical 3D-Ducts
SAROD 2003
13
X2-MAX
Wall angle < 6 Diffusion angle < 3 6 * Equivalent Radius < ROC of Centerline
X2-MIN X1-MIN
X1-MAX
SAROD 2003
14
SAROD 2003
15
DC60
SAROD 2003
Optimization
Continuation Solution
DC60
SAROD 2003
SAROD 2003
18
Optimization Post-processing
Distortion Analysis
where, PA0 - average total pressure at the section, P60min - minimum total pressure in a 600 sector, q - dynamic pressure at the cross section. User Defined Functions (UDF) and scheme files were used to generate this information from the FLUENT case and data file. Iterations may be stopped when the distortion values stabilize at the exit section with reasonable convergence levels.
SAROD 2003
20
Continuation Method
Methodology
Store the solution in case & data files Open the new case (new grid) with the old data file Setup the problem Solution of (0.61 0.31 1 1) duct slapped on (0.1 0.31 0.1 0.1)
3-decade-fall 6-decade-fall Journal file Duct Parameters Generate new case file
4996 1493
9462 6588
70%
30%
FLUENT Solution
Simulation Time
Strategies
Serial Parallel
Slapping
20
40
60
80
100
Time (hrs)
SAROD 2003
22
SAROD 2003
23
SAROD 2003
24
26
Design Tools
Lift Calculation: CL from VLM Drag Calculation: CD0 from a/c data CDi from VLM DOE: Design Expert
D-optimality Criterion
quadratic/cubic
Optimizers : FFSQP
SAROD 2003 27
SAROD 2003
28
SAROD 2003
29
SAROD 2003
30
Observations
Quadratic model found better than cubic model in subspaces. Global model inadequate.
SAROD 2003
31
SAROD 2003
32
Euler
FORTRAN source code that can evaluate gradients
SAROD 2003
35
=0.2
% Error =0.02 Value % Error =0.002
Value
% Error Value
5.45
0.61 5.56 1.54
-0.41
7.08 -0.67 77.25
-1.18
1.56 -1.02 15.09
36
=0.0002
% Error
SAROD 2003
Optimization - ADIFOR vs FD
Single design variable unconstrained optimization problem
opt
2.810 2.810
L/Dopt
11.99 11.99
Calls
15 17
Time
(min.)
424 111
SAROD 2003
37
Thank You Please visit www.casde.iitb.ac.in for details and other information
SAROD 2003
38
Thank You
http://www.casde.iitb.ac.in/mdo/3d-duct/
Problem Statement
SAROD 2003
40
Duct Parameterization
Geometry of the duct is derived from the Mean Flow Line (MFL)
MFL is the line joining centroids of crosssections along the duct Any cross-section along length of the duct is normal to MFL
Cross-section area is varied parametrically Cross-section shape in merging area is same as the exit section
SAROD 2003 41
Centry 0
y1, z1
Lm/2 Lm x
y1(Lm) = y2 (Lm), y1 (Lm) = y2 (Lm), y1 (Lm) = y2 (Lm) z1(Lm) = z2 (Lm), z1 (Lm) = z2 (Lm), z1 (Lm) = z2 (Lm) y1 (Centry) = y2 (Cmerger) = z1 (Centry) = z2 (Cmerger) = 0
The shape of the MFL is controlled by 2 parameters which control the y and z coordinate of centroid at Lm/2 y(Lm/2) = y(0) + (y(L) y(0)) y 0 < y < 1 z(Lm/2) = z(0) + (z(L) z(0)) z 0 < z < 1
SAROD 2003 43
SAROD 2003
44
SAROD 2003
45
Turbulence Modeling
Relevance: Time per Solution Following aspects of the flow were of interest:
Boundary layer development Flow Separation (if any) Turbulence Development S-shaped duct Circular cross-section Doyle Knight, Smith, Harloff, Loeffer
Baldwin-Lomax model (Algebraic model)
Literature Survey
Computationally inexpensive than more sophisticated models Known to give non-accurate results for boundary layer separation etc.
SAROD 2003
46
Turbulence Modeling
Standard k- model
(contd.)
Realizable k- model
Shih et. al. (1994) C is not assumed to be constant A formulation suggested for calculating values of C1 & C Computationally little more expensive than the standard k- model
model)
SAROD 2003
47
Results
Mass imbalance: 0.17% Energy imbalance: 0.06% Total pressure drop: 1.42% Various turbulence related quantities of interest at entry and exit sections: Entry Exit
Turbulent Kinetic Energy (m2/s2) Turbulent Viscosity Ratio 124.24 5201.54 45.65 3288.45
y+ at the cell center of the cells adjacent to boundary throughout the domain is around 18.
SAROD 2003
48
Flow Separation
SAROD 2003
49
Flow Separation
SAROD 2003
50