Aircraft Wing Shape Optimization
Aircraft Wing Shape Optimization
Aircraft Wing Shape Optimization
Patch 2.23
Baseline Optimized
Baseline Optimized
ANNAPOLIS, MD
2
Contents
List of Figures: ..................................................................................................................................................................... 4
1 Introduction .................................................................................................................................. 7
1.1 Complexity of Flow ................................................................................................................................................ 7
1.2 Computational Cost ............................................................................................................................................... 7
1.3 Aerodynamic Optimization ................................................................................................................................ 9
List of Figures:
Figure 1.1 Hierarchy of Models for Industrial Applications .............................................................................. 7
Figure 1.2 Cp Contours on High Lift Configuration with 22 M Cells Model ................................................ 8
Figure 2.1 The optimizer started with the initial shape of a circle (1st frame) and converged to a
super-critical airfoil (3rd frame). In this process, the optimizer had to go through infeasible
intermediate shapes, such as the one shown here (2nd frame) .......................................................................... 11
Figure 2.2 Multi-Point Design Process as Envisioned by Jameson – (Courtesy of Jameson et al.) .. 13
Figure 2.3 Concept of using Parallel Evaluation Strategy of Feasible and Infeasible Solutions to
Guide Optimization Direction in a GA ........................................................................................................................... 14
Figure 2.4 Wing Configurations at Different Flight Phases (Courtesy of Chiguluri) ............................. 15
Figure 3.1 Volume of Solid (VOS) design variables as grey-scale and RSVS profile in red; 1
corresponds to a completely full cell and 0 an empty cell – Courtesy of [Allen et al.] .............................. 21
Figure 3.2 Four Levels of Subdivision of a Four Point Control Polygon - Courtesy of [Allen et al.]
....................................................................................................................................................................................................... 22
Figure 3.3 Generic Non-Symmetric Airfoil Modes - a Mode 1. b Mode 2. c Mode 3. d Mode 4. e Mode
5. ................................................................................................................................................................................................... 23
Figure 3.4 Surface-Based Control Points and Example Deformation. a Control Points. b Example
Deformation - Courtesy of [Allen et al.]....................................................................................................................... 26
Figure 3.5 Surface Mesh and Control Points in 3D – Courtesy of [Allen et al.] ........................................ 27
Figure 3.6 Surface and Control Point Modal Deformations. a Mode 1 global. b Mode 3 global. c Mode
5 .................................................................................................................................................................................................... 28
Figure 3.7 Surface Mesh and off-Surface Control Points – Courtesy of [Allen et al.]............................. 29
Figure 3.8 Domain and Block Boundaries and far-field Mesh – Courtesy of [Allen et al.] .................. 32
5
Figure 3.9 Upper Surface Pressure Coefficient. an Initial Geometry. b Domain Element. c 10 Global
Modes ,d 10 Local Modes – Courtesy of [Allen et al.].............................................................................................. 34
Figure 3.10 Convergence Histories – Courtesy of [Allen et al.] ...................................................................... 35
Figure 3.11 Shape Design Variables are the z-Displacements of 720 FFD Control Points - (Courtesy
of Martins and Hwang)........................................................................................................................................................ 37
Figure 3.12 Optimized Wing with Shock-Free with 8.5% Lower Drag – (Courtesy of Lyu and
Martins) ..................................................................................................................................................................................... 40
Figure 3.13 Insensitivity of Number of Optimization Iterations to Number of Design Parameters
....................................................................................................................................................................................................... 41
Figure 3.14 Multipoint Optimization Flight Conditions .................................................................................... 43
Figure 3.15 Multi-Point Optimized - (Courtesy of Lyu and Martins) ........................................................... 43
Figure 3.16 Comparison of Baseline, Single, and Multipoint Optimization............................................... 44
Figure 3.17 Flow Field Around the Wing Section of a 3-Element Wing – Courtesy of [Brezillon et al.]
....................................................................................................................................................................................................... 47
Figure 3.18 Multi-Block Structured Mesh Around the DLR-F11 Model in Full Span Flap and Slat
Configuration – Courtesy of [Brezillon et al.]............................................................................................................. 52
Figure 3.19 Close View of the Wake Discretization on the Top of the DLR-F11 Wing – Courtesy of
[Brezillon et al.] ...................................................................................................................................................................... 53
Figure 3.20 Objective function according to the wall-clock time - Aerodynamic flows computed with
the structured FLOWer code running sequentially on a NEC-SX8 – Courtesy of [Brezillon et al.] ...... 54
Figure 3.21 Objective and Lift Coefficient According to the ............................................................................ 54
Figure 3.22 Drag Distribution Along the Spanwise Direction on the Baseline and Optimized
configurations – Courtesy of [Brezillon et al.] ........................................................................................................... 55
Figure 3.23 Optimizing Configuration for Slat and Flap – Courtesy of [Brezillon et al.] ..................... 56
Figure 3.24 Pressure Distribution for Transonic RAE 2822 Airfoil ............................................................. 66
Figure 3.25 Mesh for 2-element airfoil used in assessment of 2D design sensitivities ........................ 66
Figure 3.26 Location of design variables for 2-element airfoil ...................................................................... 66
Figure 3.27 Location of Design Variables for RAE 2822 Airfoil ..................................................................... 67
Figure 3.28 Location of design variables for ONERA M6 wing & Grid Assessment of 3D Design
Sensitivities .............................................................................................................................................................................. 68
Figure 3.29 Pressure distributions for 4-element airfoil.................................................................................. 70
Figure 3.30 Four-element airfoil in original and perturbed positions ........................................................ 71
Figure 3.31 Initial and final pressure distributions for drag reduction on RAE 2822 airfoil. ........... 71
Figure 3.32 Initial and final Mach contours for transonic airfoil optimization exercise ..................... 72
Figure 3.33 Velocity contours and vectors in flap region for multielement airfoil................................ 73
Figure 3.34 Initial and final density contours for inviscid wing design ..................................................... 74
Figure 3.35 Initial and Final Density Contours for Inviscid Wing Design.................................................. 74
Figure 3.36 Parametric CAD model of the XRF-1 wings.................................................................................... 78
Figure 3.37 Multi-Blocks Structured Mesh of the XRF-1 Model..................................................................... 79
Figure 3.38 Direct and Adjoint Calculations Convergence History on the XRF-1 baseline at the 6 80
Figure 3.39 XRF-1 single point optimization convergence history .............................................................. 82
Figure 3.40 Pressure coefficient at single point optimum ............................................................................... 83
Figure 3.41 An attempt of multipoint optimization with 6 operating conditions and explicit lift
constraints handled by the SLSQP optimizer [69] ................................................................................................... 84
Figure 3.42 6 points multi-Mach multi-Lift XRF-1 optimization history ................................................... 87
Figure 3.43 Pressure Coefficient at Multiple Operating Conditions Optimum ........................................ 88
Figure 3.44 Multipoint aeroelastic twist optimization of the XRF-1 model, starting from the
multipoint rigid optimum .................................................................................................................................................. 89
Figure 3.45 Steps of the XRF-1 multipoint aeroelastic optimization ........................................................... 90
6
7
1 Introduction
1.1 Complexity of Flow
The complexity of fluid flow is well illustrated in Van Dyke’s Album of Fluid Motion. Many critical
phenomena of fluid flow, such as shock waves and turbulence, are essentially nonlinear and the
disparity of scales can be extreme. The flows of interest for industrial applications are almost
invariantly turbulent. The length scale of the smallest persisting eddies in a turbulent flow can be
estimated as of order of 1/Re3/4 in comparison with the macroscopic length scale. In order to resolve
such scales in all three spatial dimensions, a computational grid with the order of Re 9/4 cells would
be required. Considering that Reynolds numbers of interest for airplanes are in the range of 10 to
100 million, while for submarines they are in the range of , the number of cells can easily overwhelm
any foreseeable supercomputer. [Moin and Kim] reported that for an airplane with 50-meter-long
fuselage and wings with a chord length of 5 meters, cruising at 250 m/s at an altitude of 10,000
meters, about 10 quadrillions (1016) grid points are required to simulate the turbulence near the
surface with reasonable details. They
estimate that even with a sustained
performance of 1 Teraflops, it would
take several thousand years to
simulate each second of flight time. RANS (1990s)
Spalart has estimated that if computer
performance continues to increase at
the present rate, the Direct Numerical Euler (1980s)
Simulation (DNS) for an aircraft will be
feasible in 2075.
Consequently mathematical models Non-linear
with varying degrees of simplification Potential (1970s)
have to be introduced in order to make
computational simulation of flow Linear Potential
feasible and produce viable and cost- (1960s)
effective methods. Figure 1.1.1
indicates a hierarchy of models at
different levels of simplification which
have proved useful in practice. Inviscid Figure 1.1.1 Hierarchy of Models for Industrial Applications
calculations with boundary layer
corrections can provide quite accurate
predictions of lift and drag when the flow remains attached. The current main CFD tool of the Boeing
Commercial Airplane Company is TRANAIR, which uses the transonic potential flow equation to
model the flow. Procedures for solving the full viscous equations are needed for the simulation of
complex separated flows, which may occur at high angles of attack or with bluff bodies. In current
industrial practice these are modeled by the Reynolds Average Navier Stokes (RANS) equations
with various turbulence models1.
1A. Jameson and M. Fatica, “Using Computational Fluid Dynamics for Aerodynamics”, Stanford University,
USA.
8
demonstrated that inviscid transonic potential flow or Euler solutions for an airfoil can be accurately
calculated on a mesh with 160 cells around the section, and 32 cells normal to the section. Using a
new non-linear Symmetric Gauss-Siedel (SGS) algorithm, which has demonstrated “text book”
multigrid convergence (in 5 cycles), two-dimensional calculations of this kind can be completed in
0.5 seconds on a laptop computer (with a 2Ghz processor). A three dimensional simulation of the
transonic flow over a swept wing on a 192 x 32 x 32 mesh (196,608 cells) takes 18 seconds on the
same laptop. Moreover it is possible to carry out an automatic redesign of an airfoil to minimize its
shock drag in 6.25 seconds, and to redesign the wing of a Boeing 747 in 330 seconds2.
Viscous simulations at high Reynolds numbers require vastly greater resources. On the order of 32
mesh intervals are needed to resolve a turbulent boundary layer, in addition to 32 intervals between
the boundary layer and the far field, leading to a total of 64 intervals. In order to prevent
degradations in accuracy and convergence due to excessively large aspect ratios (in excess of 1,000)
in the surface mesh cells, the chord wise resolution must also be increased to 512 intervals.
Translated to three dimensions, this implies the need for meshes with 5-10 million cells (for
example, 512 x 64 x 256 = 8,388,608 cells) for an adequate simulation of the flow past an isolated
wing. When simulations are performed on less fine meshes with, say, 0.5 M to 1 M cells, it is very
hard to avoid mesh dependency in the solutions as well as sensitivity to the turbulence model.
Currently Boeing uses meshes with 15-60 million cells for viscous simulations of commercial
aircraft with their high lift systems deployed3. Figure 1.2.1 show the Cp contours on High Lift
Configuration (wing) with 22 M Cells (Courtesy of Boeing). Using a multigrid algorithm, 2000 or
more cycles are required to reach a steady state, and it takes 1-3 days to turn around the calculations
on a 200 processor Beowulf cluster.
2 Antony Jameson & Massimiliano Fatica, “Using Computational Fluid Dynamics for Aerodynamics”,
Stanford University.
3 Boing using 22 M Cells on High lift Configuration.
9
4 Glauert, H. (1926), “The Elements of Aero foil and Airscrew Theory”, Cambridge University Press.
5 Prandtl, L. and Tietjens, O.G. (1934), “Applied Hydro and Aerodynamics”, Dover Publications.
6 Sears, W.D. (1947), ”On projectiles of minimum drag”, Q. Appl. Math., 4, 361–366.
7 Jones, R.T. (1981), ”The minimum drag of thin wings in frictionless flow”. J. Aerosol Sci., 18, 75–81.
8 Lighthill, M.J. (1945), ”A new method of two dimensional aerodynamic design”. Rep. Memor. Aero. Res. Coun.
9 Samareh, J.A., “Survey of shape parameterization techniques for high-fidelity multidisciplinary shape
optimization”. AIAA Journal, 2001.
10 See the previous.
11 Sripawadkul, V., M. Padulo and M.Guenov, “A Comparison of Airfoil Shape Parameterization Techniques for
AIAA-2007-59. 2007.
14 Mousavi, A., Castonguay P., and S. K. Nadarajah, “Survey of Shape Parameterization Techniques and Its Effect
cannot accurately predict the aerodynamic performance of such designs due to the massive
separation, the gradients provided the correct trends, and the optimization eventually converged to
an optimal supercritical airfoil for which RANS is valid.
Figure 2.1.1 The optimizer started with the initial shape of a circle (1st frame) and converged to a
super-critical airfoil (3rd frame). In this process, the optimizer had to go through infeasible intermediate
shapes, such as the one shown here (2nd frame)
Multiple Objectives:
Minimize F(x) = [F1(x) , F2(x) , . . . , Fk (x)]T
subject to gj (F(x)) ≤ 0 , j = 1, 2, . . . , m , and hL (F(x)) = 0, L = 1, 2, . . . , e ,
F(x) ∈ Ek are also called objectives, criteria, payoff functions, cost functions, or
value functions, where k is the number of objective functions, m is the number of
inequality constraints, and e is the number of equality constraints. x ∈ En is a vector
of design variables (also called decision variables), where n is the number
independent variables.
configuration, and is updated until a minimum of f(x) is identified or the optimization process runs
out of allocated time/iterations. The standard form of optimization problem statement is:
1 The level of information fidelity required from the flow solver, depending on problem ;
2 Parametrization of Design Space;
3 Types of Design Variables, e.g. Discrete and/or Continuous;
12
Multi-Level Optimization:
Minimize (x , y) , F(x , y)
subject to: y ∊ S(x); G(x, y) ≤ 0
where S(x) denotes the set of solutions of the lower level problem as:
Minimize (y) : f(x , y) subject to g(x, y) ≤ 0
16 Jane J. Ye, “Necessary optimality conditions for multi-objective bi-level programs”, 2010.
17 A. Fathi · A. Shadaram, “Multi-Level Multi-Objective Multi-Point Optimization System for Axial Flow Compressor
β1 + β2 + ⋯ + β𝑛 = 1 , g = β1 g1 + β2 g 2 + ⋯ + β𝑛 g 𝑛
Eq. 2.4.1
The gradient to determine an assigned a weight overall loss or a gain for the design is created by
summing all the gradients times each respective weight (see Eq. 2.4.1). What this allows for is if a
change drastically improves takeoff performance but results in a slight hit on cruising performance,
the cruising hit can override the takeoff gain due to weighting. Setting the simulation up in this
manner can significantly improve the designs produced by the software. This version of the modeler,
however, adds yet another complexity to the initial conditions, and a slight error on the designer’s
behalf can have a significantly larger effect on the resulting design. The calculation efficiency
improvement takes advantage of the multiple variables. (See Figure 2.4.1). The problem observed
is that changes that boosted one point of interest directly conflicted with the other, and the resulting
compromise severely hampers the improvement gained. Current research involves a better way to
resolve the differences and achieve an improvement similar to the single-point optimizations.
Figure 2.4.1 Multi-Point Design Process as Envisioned by Jameson – (Courtesy of Jameson et al.)
18Jameson, A., Leoviriyakit, K., and Shankaran, S., "Multi-point Aero-Structural Optimization of Wings Including
Planform Variations", 45th Aerospace Sciences Meeting and Exhibit, AIAA-2007-764, Reno, NV, 8–11 Jan 2007.
14
Optimisation and to Jetty Scheduling Problems”. Trends in Engineering Computational Technology, 2008.
21 T. Kato, K. Shimoyama, and S. Obayashi. “Evolutionary Algorithm with Parallel Evaluation Strategy of Feasible
and Infeasible Solutions Considering Total Constraint Violation”. IEEE, 1(978):986-993, 2015.
15
feasible and infeasible solutions could be evaluated in parallel to guide the optimization search
direction towards feasible design spaces. This is intuitively true for single discipline aerodynamic
optimization problems where often small modifications to design variables can largely impact the
performance rendering designs infeasible. Algorithm understanding of infeasible solutions can help
in the betterment of feasible solutions though algorithm learning/training and constraint based
reasoning. [Robinson et al.]22, comparing the performance of alternative trust-region constraint
handling methods, showed that reapplying knowledge of constraint information to a variable
complexity wing design optimization problem reduced high-fidelity function calls by 58% and
additionally compare the performance to alternative constraint managed techniques.
Elsewhere, [Gemma and Mastroddi]23 demonstrated that for the multi-disciplinary, multi-objective
aircraft optimizations the objective space of feasible and infeasible design candidates are likely to
share no such definitive boundary. With the adoption of utter constraints, structural constraints, and
mission constraints solutions defined as infeasible under certain conditions would otherwise be
accepted, hence forming complex Pareto fronts. Interdisciplinary considerations such as this help to
develop and balance conflicting constraints. For example, structural properties which may be
considered feasible, but are perhaps heavier than necessary will inflict aero-elastic instabilities at
lower frequencies. In the aerospace industry alone there are several devoted open-source
aerodynamic optimization algorithms with built-in constraint handling capability. Some studies
have also adopted MATLAB's optimization tool-box for successful optimization constraint
management.
1.1.1 Case Study - Multi-Point Optimization of Airfoil
Aerodynamically, an optimal airfoil shape produces high lift and low drag within the design
constraints often imposed by the structural
requirements [Chiguluri]24. An inverse
design technique was applied to NACA 0012
airfoil which resulted in an airfoil with drag
bucket at the normal flight operation
conditions. The most general form of an
airfoil (used on most commercial airplanes)
consists of three individual units: slat, main
element, and the flap. Each part has its
importance in obtaining the required
performance from the airfoil. Slat and flap
are often deployed or retrieved based on
the phase of the flight. Slat is used to delay
stall such that an increment in the angle of
attack doesn’t cause adverse effect on the
lift. The flap is used to increase the camber
of the airfoil so that additional lift is
obtained. Figure 2.5.2 summarizes the Figure 2.5.2 Wing Configurations at Different Flight
typical configuration of the wing at different Phases (Courtesy of Chiguluri)
22 T.D. Robinson, K.E. Willcox, M.S. Eldred, and R. Haimes. “Multi-fidelity Optimization for Variable-Complexity
Design”. 11th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, pages 1-18, Portsmouth,
VA, 2006. AIAA 2006-7114.
23 S. Gemma and F. Mastroddi. “Multi-Disciplinary and Multi-Objective Optimization of an Unconventional Aircraft
Concept”. 16th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, AIAA 2015-2327, pages
1-20, Dallas, TX, 2015.
24 B., Chiguluri, “Multi-Point optimization of Airfoils”, Undergraduate Research Thesis, Georgia Institute of
Technology, 2011.
16
phases of flight level flight, take-off, and landing. In the cruise phase i.e. when the slat and flap are
retracted, the multi-element airfoil can be simplified (by ignoring the small gaps between the
surfaces) to a single element airfoil. The simplified single element airfoil’s aerodynamic properties
are often used to design an optimum wing cross section. Often, landing could be ignored because it is
easily achieved through aileron deployment. Therefore, the two phases of flight that govern the airfoil
design are the steady flight and take-off conditions. In most studies, the optimization process is
applied to the cruise level condition while ignoring the take-off conditions. The results often result in
inefficient take-off conditions which result in excess fuel procurement.
17
25 Christian B. Allen, Daniel J. Poole, Thomas C. S. Rendall, “Wing aerodynamic optimization using efficient
mathematically-extracted modal design variables”, Optimum Engineering, 2018.
26 Hicks RM, Henne PA, “Wing design by numerical optimization”. J Aircr 15(7):407–412, 1978.
27 Qin N, Vavalle A, Le Moigne A, Laban M, Hackett K, Weinerfelt P, “Aerodynamic considerations of blended wing
47(2):638–646, 2010.
29 Lyu Z, Kenway GKW, Martins JRR, “Aerodynamic shape optimization investigations of the common research
element parameterization and mesh deformation”. Inter J Numerical Meth Fluids 58(8):827–860, 2008.
32 Morris AM, Allen CB, Rendall TCS, “Domain-element method for aerodynamic shape optimization applied to a
33 Allen CB, Rendall TCS, “Computational-fluid-dynamics-based optimization of hovering rotors using radial basis
functions for shape parameterization and mesh deformation”. Optimum Eng. 14:97–118.
34 Poole DJ, Allen CB, Rendall TCS, “Metric-based mathematical derivation of efficient airfoil design variables”.
design variables with a constrained global optimizer”. Computer Fluids 143:1–15, 2017.
37 L. J. Kedward , A. D. J. Payot y, T. C. S. Rendall, C. B. Allen, “Efficient Multi-Resolution Approaches for Exploration
optimization process, which is required to allow deformation of a body-fitted CFD mesh. An effective
parameterization method is
➢ flexible and robust enough to cover the design space, and
➢ efficient enough to represent a given shape with as few design variables as possible.
Methods are classified as either constructive, reformative or unified. In-depth reviews have been
presented by [Samareh]38, [Castonguay and Nadarajah]39, [Mousavi et al. ]40 and [Masters et al. ]41.
Constructive methods consider the definition of the surface and the deformation of the surface
separately. Examples of these methods are CST [Kulfan]42, PARSEC [Sobieczky]43, PDEs [Bloor and
Wilson]44 and splines [Braibant and Fleury]45. Other approaches that combine various
parameterizations in a hybrid approach can also be found. Because of the constructive nature of these
approaches, perturbation of the base geometry through the optimization process requires that the
new surface be reconstructed, which subsequently requires automatic mesh generation tools for
production of a new surface and volume mesh. This extra difficulty can make it advantageous to
consider approaches that manipulate an existing mesh.
An alternative to constructive methods are deformities methods which unify the geometry creation
and perturbation. This tends to make them simpler to integrate with mesh deformation tools and
allows the use of previously generated meshes; a considerably cheaper alternative to regeneration,
although the mesh deformation scheme is a separate algorithm. Analytic [Hicks and Henne]46 and
discrete [Jameson]47 methods are examples of deformities approaches. A further refinement of
unifying geometry creation and perturbation is the integration with a mesh deformation algorithm.
Methods of this type typically have some interpolation that describes a link between the surface and
volume, often via a set of control points that are independent of both, such that deformation of the
control points results in deformation of the surface and CFD mesh. These approaches are commonly
used in ASO, and the methods included in this unified category are free-form deformation
[Samareh]48, domain elements (Morris et al.]49 and direct manipulation [Yamazaki et al.]50.
Surface parameterizations developed around the FFD and domain element approach are very
popular in wing optimization as this type of approach allows the design space to be reduced from
38 Samareh JA, “Survey of shape parameterization techniques for high fidelity multidisciplinary shape
optimization”, AIAA J 39(5):877–884, 2001.
39 Castonguay P, Nadarajah SK, “Effect of shape parameterization on aerodynamic shape optimization”, 45th AIAA
Aerospace Sciences Meeting and Exhibit, Reno, Nevada, AIAA Paper 2007–59.
40 Mousavi A, Castonguay P, Nadarajah SK, “Survey of shape parameterization techniques and its effect on three
dimensional aerodynamic shape optimization”.18th AIAA CFD dynamics conference, Florida, 2007.
41 Masters DA, Taylor NJ, Rendall TCS, Allen CB, Poole DJ, “Geometric comparison of aerofoil shape
NW, Keane AJ, Holden CME (2010) Geometric filtration using proper orthogonal decomposition for
aerodynamic design optimization. AIAA J 48(5):916–928, 1998.
44 Bloor MIG, Wilson MJ, “Generating parameterizations of wing geometries using partial differential equations”.
44(3):247–267, 1984.
46 Hicks RM, Henne PA, “Wing design by numerical optimization”. J Aircraft 15(7):407–412, 1978.
47 Jameson A (1988) ,“Aerodynamic design via control theory”. J Scientific Computation 3(3):233–260.
48 Samareh JA, “Novel multidisciplinary shape parameterization approach”. J Aircraft- 38(6):1015–1024, 2001.
49 Morris AM, Allen CB, Rendall TCS, “CFD-based optimization of airfoils using radial basis functions for domain
element parameterization and mesh deformation”, Int J Numer Meth Fluids 58(8):827–860, 2008.
50 Yamazaki W, Mouton S, Carrier G, “Geometry parameterization and computational mesh deformation by
thousands of design parameters to hundreds of design parameters. Such techniques have been
developed by [Zingg] and colleagues [Hicken and Zingg]51 ; [Leung and Zingg]52, and have shown that
these types of methods can be flexible enough to allow the molding of a sphere into an aircraft like
shape under certain optimization conditions [Gagnon and Zingg]53. Further work has been performed
by Martins and others [Mader and Martins]54; [Lyu and Martins]55 who showed results for blended-
wing-body optimizations, and [Yamazaki et al.]56 who further reduced the number of design variables
by considering the direct manipulation method for wing optimization. A novel method, recently
developed by the authors, is to extract airfoil design variables using a mathematical approach. The
approach utilizes singular value decomposition in a manner that analyses an initial library of airfoils
and decomposes that library into a reduced set of optimum variables that are geometrically
orthogonal to each another. The method is based on perturbations, so is independent of the initial
geometry and can fit into any of the three categories outlined above; the deformities formulation is
used in this work, to allow a unified application of the design variables to control both the surface
and volume mesh within the ASO framework. Previous work has considered the method’s ability to
represent a wide-range of airfoil shapes [Poole et al.]; [Masters et al.], and their effectiveness in airfoil
optimization [Poole et al.]; [Masters et al.], wherein the efficiency of this modal approach was clearly
demonstrated. Hence, the aim of the work presented here is to develop an effective method to apply
these novel mathematically-extracted design variables, which have been extracted as two-
dimensional quantities, in three dimensions, and determine their effectiveness when applied to
aerodynamic optimization, in particular drag minimization of wings in transonic flow.
1.1.3.1 Other Parameterization Techniques
This section develops the R-Snake Volume of Solid (RSVS) parameterization method which blends
the topological flexibility of volume of solid design variables with the efficiency of established
aerodynamic parameterization methods. Achieving this level of efficiency requires the RSVS to
generate smooth surfaces fulfilling volumes specified on a predefined grid. To ensure the method is
flexible enough to support anisotropic design variable refinement and to facilitate the extension to
3D, the RSVS must be generic enough to work on arbitrary polygonal grids. One of the difficulties in
designing a parameterization with topological flexibility is to maintain smooth control close to
topology changes, as these are geometrically discontinuous regions of the design space. To define a
set of VOS variables a grid is superimposed on the design space, where the design variables become
the fraction of each cell within a geometry built from this information. This process is shown for a
simple grid in Figure 3.1.1. This parameterization procedure provides intuitive handling of
topology change without maintaining explicit control of it. It is important that topology is not
controlled explicitly as this would lead to a severely discontinuous design space which would not be
usable with many of the traditional local and global optimizers used for aerodynamic optimization.
51 Hicken JE, Zingg DW, “Aerodynamic optimization algorithm with integrated geometry parameterization and
mesh movement”. AIAA J 48(2):400–413, 2010.
52 Leung TM, Zingg DW, “Aerodynamic shape optimization of wings using a parallel newton-krylov approach”.
AIAA aviation technology, integration and operations (ATIO) conference and 14th AIAA/ISSMO
multidisciplinary analysis optimization conference, Indianapolis, Indiana, 2012.
54 Mader CA, Martins JRRA, “Stability-constrained aerodynamic shape optimization of flying wings”. J Aircraft
50(5):1431–1449, 2013.
55 Lyu Z, Kenway GKW, Martins JRRA , “Aerodynamic shape optimization investigations of the common research
Figure 3.1.1 Volume of Solid (VOS) design variables as grey-scale and RSVS profile in red; 1
corresponds to a completely full cell and 0 an empty cell – Courtesy of [Allen et al.]
Figure 3.1.2 Four Levels of Subdivision of a Four Point Control Polygon - Courtesy of [Allen et al.]
‖𝐌 − 𝐌 k ‖F ≤ ‖𝐌 − 𝐦‖F
Eq. 3.1.2
where m is any matrix of rank k and ‖ . ‖F is the Frobenius norm. Hence, the error in the low rank
approximation (found from SVD) will always be at least as good as the error between any other rank
k matrix and the full rank matrix. The SVD thus produces an optimal low order projection of the
higher dimensional space into the lower dimensional one, which is significant for optimization
parameters. The SVD method first requires a training library of N a airfoils to be collated from which
the airfoil deformation modes are extracted. Each airfoil surface is parameterized by N surface points,
where the i-th surface point has a position in the space (xi , zi). To ensure consistency of the surface
description of the training data all airfoils are parameterized with the same parametric distribution.
The x distribution is often defined as the controlling parameterization, with zi = f (xi), but this is not
the most flexible approach; instead all airfoil surfaces are parameterized in terms of peripheral
distance s ∈ [0 , 1] and then exactly the same si distribution is defined for all airfoils. Following the
surface point distribution, each airfoil has a rigid body translation, scaling and then rotation applied
23
to it to map the geometry into a consistent form where each section has unit chord and z(0) = z(1) =
0: A matrix is built from which SVD is performed, by evaluating the vector difference of the i-th
surface point between all airfoils, producing Ndef = Na(Na -1)/2 airfoil deformations. The x and z
deformations are stacked into a single vector of length 2N, for each airfoil deformation, so a matrix is
built of the airfoil deformations which has 2N rows and Ndef columns:
∆x1,1 ⋯ ∆x1,Ndef
⋮ ⋱ ⋮
∆xN,1 ⋯ ∆xN,Ndef
𝐌=
∆z1,1 … ∆z1,Ndef
⋮ ⋱ ⋮
[ ∆zn,1 … ∆zn,Ndef ]
Eq. 3.1.3
Performing a SVD decomposes the matrix into three constituent matrices:
𝐌 = 𝐔𝚺𝐕 𝐓
Eq. 3.1.4
where U is a matrix of vectors, each of length 2N. The structure is analogous to the decomposed
matrix, so the columns of this matrix are the airfoil mode shapes. Σ is a diagonal matrix of the singular
values, arranged in descending order. These can be considered the ‘relative energy’ of the modes, and
represent the ‘importance’ of the mode shapes in the original library. The total number of possible
mode shapes is governed by the number of singular values, which is the minimum of the number of
columns and rows of the decomposed matrix. A truncation of the U matrix, based on a certain total
energy required, then gives the number of design variables used in the optimization. The training
library is based on deformations, and this is an important choice such that design variables that result
from the decomposition are also deformations, ensuring they are independent of the topology of the
airfoils that are used. This allows direct insertion into an aerodynamic shape optimization
framework where deformation of the surface and mesh is important. If the constructive formulation
is used, however, then the columns of the training matrix, M, are absolute positions of the airfoil
surface points as opposed to deformations between surface points.
In this work, a generic, non-symmetric training library is considered based on the optimization being
performed. The library contains 100 different airfoils, extracted from a larger library by quantifying
Figure 3.1.3 Generic Non-Symmetric Airfoil Modes - a Mode 1. b Mode 2. c Mode 3. d Mode 4. e Mode 5.
f Mode 6 – Courtesy of [Allen et al.]
24
their performances in the transonic regime using the Korn technology factor [Poole et al.]59. The first
six modes of the library are shown in Figure 3.1.3; all modes are scaled up for illustration purposes,
and have been added to a NACA0012 section. Also the relative ‘energy’ of the first 20 modes, i.e. the
singular values of each mode normalized by the sum of all singular values. Once the design variables
have been extracted and the total number of modes has been truncated, a new airfoil can be formed
by a weighted combination of m modal parameters, as shown in Eq. 3.1.5. The weighting vector, β,
represents the magnitude of the modal deformations which are then the design variable values that
the optimizer works with. The truncation of the total number of modes, which is often very large,
down to a number which is useful for the optimization can either be user-specified or based on the
requirement for a total amount of energy to be preserved, e.g., if 99.0% of the energy of the original
library is required to be preserved then the first, say, six modes may cumulatively have 99.1% of the
energy so six modes would be used. In this work, a number of modes is specified and those modes
with the highest amount of energy are taken.
𝐗 new = 𝐗 old + ∑ βn 𝐔 n
n=1
Eq. 3.1.5
The modes extracted here are two-dimensional deformations, based on a large database of
aerodynamic surfaces, and so are effective in two dimensions, and this has been proven previously
[Poole et al.]. Hence, it would make sense to consider a similar approach in three dimensions.
However, this would require a database of wings, something that would not be easy to create, and
with variable parameters such as taper and sweep, and surface discontinuities such as crank
locations, would also require a complex parametric transformation to a normalized space.
Furthermore, this would still not contain global variables, and so it is more sensible and flexible to
consider a more local sectional approach. This is the approach considered here.
1.1.3.4 RBF Coupling of Point Sets for Airfoil Deformation
The airfoil design variables must be coupled to a control point-based approach to allow flexible
deformation of the CFD mesh. The control point method links deformations of the CFD mesh to
deformations of a small set of control points on or near the surface. At the center of this technique is
a multivariate interpolation using radial basis functions (RBFs), which provides a direct mapping
between the control points, the surface geometry and the locations of grid points in the CFD volume
mesh. The approach is meshless, so requires no connectivity and is applicable to any mesh type;
control points and volume mesh points are simply treated as independent point clouds. The system
is only the size of the number of control points, and so is not related to the mesh size.
The general theory of RBFs is presented by [Buhmann (2005)] and [Wendland (2005)], and the basis
of the method used here is described in detail by [Rendall and Allen (2008)]. If φ is the chosen basis
function and ‖ . ‖ is used to denote the Euclidean norm, then a general volume interpolation models
has the form
n
59Poole DJ, Allen CB, Rendall TCS, “Metric-based mathematical derivation of efficient airfoil design variables”.
AIAA J 53(5):1349–1361, 2015.
25
points) are used here to decouple the shape parameters from the surface mesh, and provide a flexible
framework through which to control the shape of a base geometry. Setting up a global RBF volume
interpolation for nc control points then requires a solution to a linear system [see Morris et al.
(2008)] to ensure exact recovery of the control point data, in this case deformations:
∆x1c α1x
∆𝐗 𝐜 = [ ⋮ ] , 𝛂n = [ ⋮ ]
c
∆xnc αxnc
Eq. 3.1.8
(analogous definitions hold for y and z coordinates) and the control point dependence matrix, C, takes
the form
φ11 … φ1nc
𝐂=[ ⋮ ⋱ ⋮ ], where φij = φ(‖𝐱 ic − 𝐱 jc ‖)
φnc 1 … φncnc
Eq. 3.1.9
For surface and volume mesh deformation, it is sensible to use decaying basis functions, to give the
interpolation a local character and ensure deformation is contained in a region near the moving body,
and [Wendland’s] C2 function is used here. It is also sensible to omit polynomial terms, since these
will transfer deformation throughout the entire mesh. Hence, in the case considered here the global
influence on any point in the aerodynamic mesh (denoted by superscript a) from the control points
is determined and applied as
nc
the control points to define the modal deformations are located on the surface of the airfoil section.
This ensures that there is direct coupling between the control point deformations and the surface
deformations that derived them. The deformation modes derived here by SVD are extracted from a
training library of airfoils. A complete library of airfoils is quantified in terms of aerodynamic
performance, using the Korn technology factor, and a form of library filtering applied to down-select
the library; see [Poole et al.]. A set of control points is used to control the aerodynamic surface; these
points are independent of a base geometry, and the surface deformations are defined in terms of
perturbations so the modes can be applied to any geometry. A ‘shrink-wrapping’ method is used to
map them onto the geometry being considered. Here, 24 control points are used; more than this are
not necessary unless small wavelength changes are required. The modal deformations can be defined
using a larger number of points than the N value in Eq. 3.1.3 and projected onto these 24 points, but
here N = 24 is used in the SVD extraction process. Figure 3.1.4 shows the surface control points and
an example deformation of the fourth mode for a NACA0012 mesh.
Figure 3.1.4 Surface-Based Control Points and Example Deformation. a Control Points. b Example
Deformation - Courtesy of [Allen et al.]
parameters, and Δxni is the modal deformation of point i for mode n. Once the deformation vector has
been evaluated for every control, the ΔX and ΔZ vectors are known (ΔY = 0 in 2D), can be solved, and
the deformation of all mesh nodes, including those on the surface, is evaluated.
1.1.6 Computation of Deformation Field in 3D
In three dimensions, a set of ns sectional slices of control points are applied to the surface at regular
intervals. However, when these are deformed, the variation of the deformation field between the
sections can either be defined explicitly or left to the global interpolation field. The latter is normally
used, but this means that interpolation properties, for example the basis function chosen and the
support radius set, will influence the deformed surface. That effect is undesirable, so it is eliminated
here, as it can result in a more global influence of an effectively local deformation. Intermediate
sections are thus defined between each deformed slice, and the deformation of these is controlled
analytically. The span wise region between each section is split into Nint intermediate regions, and so
the total number of sections becomes 1+(ns -1) nint .
The geometry considered here is the MDO wing [Allwright] and [Haase et al.]. The surface is
preprocessed to compute the local chord length at each section, i.e. cj, and the initial rotation angle of
each section, α0j , where j is the section location. The control point sections are then applied to the
surface by scaling by local chord, rotating by local incidence, and shrink-wrapping to the exact
geometry using a local geometric intersection algorithm. Figure 3.1.5 shows the control points
Figure 3.1.5 Surface Mesh and Control Points in 3D – Courtesy of [Allen et al.]
resulting from using ns = 10 and nint = 4, for the surface mesh used later. This means there are 37
control point sections but only 10 are deformed by the design parameters. The deformed points are
shown in green, and the controlled points in black. Consider first the deformation field for global
application of the modal parameters. In this case the modal deformations are applied using a single
global weighting, i.e. one design variable for each mode. As with the two-dimensional approach, all
deformations are computed at each sectional set of 24 control points defined at zero incidence, and
so there is deformation due to rigid rotation and that due to modal parameters. Global pitch and twist
variables are used later. In three dimensions the modal deformations must be scaled by the local
chord as well as being rotated by local section incidence, and so to compute the deformation field at
the 24 control points, i, at section j:
28
𝐧𝒔 𝐦
Figure 3.1.6 Surface and Control Point Modal Deformations. a Mode 1 global. b Mode 3 global. c Mode 5
global. d Mode 1 local. e Mode 3 local. f Mode 5 local – Courtesy of [Allen et al.]
sectional deformation decays to zero at the neighboring sections each side, a global modal
deformation can be recovered exactly. In this case βn would have a single value for all sections, and
so it can be shown that if
29
ns
∑ φ (j, s) = 1
s=1
Eq. 3.1.14
at any span wise point, Eq. 3.1.11 and Eq. 3.1.12 are equivalent. Hence, a trigonometric function of
(j, s) is used. Figure 3.1.6 shows the control locations and resulting surface mesh for deformations
using the first, third, and fifth modes; the upper row shows a global modal deformation, and the lower
row shows local modal deformations of the fifth control point section. The modal deformation
magnitude is exaggerated to 10% local chord for illustration purposes. This improved localization
process is also adopted to improve the application of off-surface domain element perturbations, used
previously by the authors. Figure 3.1.7 shows two views of the surface and domain element points,
again using ns = 10 and nint = 4. An exaggerated movement of all the points on the fifth section is
shown; magnitude 20% local chord. (see [Allen et al.]60).
Figure 3.1.7 Surface Mesh and off-Surface Control Points – Courtesy of [Allen et al.]
60Christian B. Allen, Daniel J. Poole, Thomas C. S. Rendall, “Wing aerodynamic optimization using efficient
mathematically-extracted modal design variables”, Optimum Engineering, 2018.
30
is highly dependent on the extent of the surface representation and the fidelity of the flow analysis
tool. The issue of degree of multi-modality in aerodynamic optimization problems is an unanswered
question, with work presented showing that multimodality exists in a number of cases, but unimodal
cases also exist (Namgoong et al]61, [Khurana et al.]62; [Buckley et al.]63; [Chernukhin and Zingg].
[Chernukhin & Zingg] have considered this issue by testing a number of different optimization
problems and have shown that for a b-spline parameterization of the surface, viscous, compressible
drag minimization of the RAE2822 airfoil has one global optimum. They also showed multiple local
optima for other three-dimensional problems. For maximum flexibility and efficiency, a gradient-
based method is used here, with a second-order finite-difference approach for gradient evaluation.
This approach allows a ‘wrap-around’ approach, i.e. any flow-solver can be implemented within the
framework.
1.1.7.1 Feasible Sequential Quadratic Programming (FSQP)
The feasible sequential quadratic programming (FSQP) algorithm is used here as implemented in
version 3.7 [Zhou et al.]64. FSQP is based on an SQP method, which is an approach for constrained
gradient-based optimization. It is constructed with a number of modifications to the conventional
SQP method to avoid the so-called ‘Maratos’ effect (restriction of a step size due to the requirement
of feasibility) [Maratos]65. The modifications include a number of strategies, the primary one being
combining a search along an arc [Mayne and Polack] with a non-monotone procedure for that search
[Grippo et al.]. The FSQP algorithm is briefly outlined below, and fully described and analyzed in
[Panier and Tits] and [Bonnans et al.]. At every major iteration, t, the design vector, b, at the next
iteration is given by:
𝛃(t + 1) = 𝛃 + a∆𝛃 + a2 ̅̅̅̅
∆𝛃
Eq. 3.1.15
where a is the step length, Δβ is the line step direction and Δβ→ is a correction direction used to create
a search arc. To find the line step direction, FSQP solves a quadratic programming (QP) sub problem.
Considering inequality constraints only, this QP sub problem at every major iteration is:
1 T
Minimize: ∆𝛃sqp 𝐇∆𝛃sqp + ∇(𝛃)T ∆𝛃 sqp
2
Subject to: ∇g i (𝛃)T ∆𝛃sqp + g i (𝛃) ≤ 0 i = 1, , , , , G
Eq. 3.1.16
where J is the objective function and gi is the ith inequality constraint of a total of G inequality
constraints. FSQP augments the SQP descent direction by a feasible step direction, Δβf , that is a
fraction of either ⊽J(β) or ⊽gi(β), depending on the constraint value. The overall step direction is
then a blend of the SQP and feasible step directions
61 Namgoong H, Crossley W, Lyrintzis AS, “Global optimization issues for transonic airfoil design”, 9th
AIAA/ISSMO symposium on multidisciplinary analysis and optimization, Atlanta, AIAA Paper 2002–5641.
62 Khurana MS, Winarto H, Sinha AK, “Airfoil optimization by swarm algorithm with mutation and artificial
neural networks.” ,47th AIAA aerospace sciences meeting including the new horizons forum and aerospace
Exposition, Orlando, Florida, AIAA Paper 2009–1278.
63 Buckley HP, Zhou BY, Zingg DW, “Airfoil optimization using practical aerodynamic design requirements”. J
where ρ ∈ [0,1] = O (‖ Δβsqp ‖2) such that as a solution is approached, q tends very quickly to zero to
enable the fast convergence of the pure SQP step direction to be inherited [Bonnans et al.]. The
correction direction is found such that both descent and feasibility are ensured by solving a further
quadratic programmed while avoiding the need for further constraint and function evaluations. The
exact implementation of the rules that govern the computation of the step size are given in Zhou et
al. (1997). The Hessian, or an approximation to the Hessian, at every major iteration is required,
which in turn requires sensitivity of the objective function and constraints with respect to the design
variables. The Hessian is approximated by the Broyden–Fletcher–Goldfarb–Shanno (BFGS) update
scheme where the Hessian approximation is initialized as the identity matrix. The gradients are
obtained by a second Aerodynamic optimization using efficient modal variables order central-
difference scheme, so the number of objective function evaluations is proportional to the number of
design variables. Once the search arc has been determined, the no monotone line search proceeds
(Zhou and Tits 1993). For this, the conventional backtracking line search requirement of requiring a
suitable reduction in the objective function from iteration t to t + 1 is relaxed, such that a reduction
in the objective function to iteration t + 1 is required against the maximum objective function from
the last four iterations. This has been shown to be highly effective for unconstrained optimization
when compared to a conventional backtracking search (Grippo et al. 1986), and when implemented
for FSQP shows similar results for constrained optimization (Zhou and Tits 1993).
The algorithm iterates until the Kuhn–Tucker conditions are satisfied, which then represent a
converged solution using a constrained gradient-based optimizer. For computational efficiency, the
sensitivity evaluation has been parallelized based on the number of design variables such that the
evaluation of the sensitivity of the objective function and constraints with respect to the design
variables is split between the number of CPUs available (Morris et al. 2008, 2009). This is necessary
as within the ASO environment, an objective function evaluation represents a CFD solution, so this
formulation allows parallel evaluation of the required sensitivities; second-order finite-differences
are used for the sensitivities. It is well known that the accuracy of the gradient evaluation is a critical
issue, and the authors have performed several studies on perturbation size for finite-differences; see
for example Morris et al. (2008). A relative perturbation of 10-4 is adopted here, i.e. a deformation
magnitude of 0.01% of local chord. Constraint and step-size evaluations and optimizer updates occur
in the master process, and each CPU controls the geometry (and CFD volume mesh) perturbations
corresponding to the different design variables, and calls the flow solver. Flow-solver results are then
returned to the master for optimizer updates.
1.1.8 Flow Solver
The flow-solver used is a structured multi-block finite-volume code, with upwind spatial
discretization, using the flux vector splitting of [van Leer], and multistage Runge–Kutta time-
stepping. Convergence acceleration is achieved through multigrid [Allen]66.
1.1.9 Application of Modal Design Variables in 3D
Optimization is applied here to the MDO wing [a large modern transport aircraft wing, the result of a
previous Brite–Euram project [Allwright]67 and [Haase et al.] in the economical transonic cruise
condition.
1.1.9.1 Problem Definition
The economical cruise flight Mach number for the MDO wing defined by [Allwright] and [Haase et
al.] is 0.85, with the wing trimmed to obtain a lift coefficient of 0.452. This design case is well-suited
to inviscid flow analysis, since induced and wave drag are dominant here. Compressible transonic
66 Allen CB, “Multigrid convergence of inviscid fixed- and rotary-wing flows”. International J Numerical Meth
Fluids 39(2):121–140, 2002.
67 Allwright S, “Multi-discipline optimization in preliminary design of commercial transport aircraft”.
wing optimization for drag minimization subject to strict constraints is investigated, and so the
problem definition is:
Figure 3.1.8 Domain and Block Boundaries and far-field Mesh – Courtesy of [Allen et al.]
68 Allen CB, “Towards automatic structured multi-block mesh generation using improved transfinite
interpolation”. International J Numerical Meth Eng. 74(5):697–733, 2008.
33
69Christian B. Allen, Daniel J. Poole, Thomas C. S. Rendall, “Wing aerodynamic optimization using efficient
mathematically-extracted modal design variables”, Optimum Engineering, 2018.
34
Figure 3.1.9 Upper Surface Pressure Coefficient. an Initial Geometry. b Domain Element. c 10 Global
Modes ,d 10 Local Modes – Courtesy of [Allen et al.]
shown in the table are optimization run times and the number of cores used for each. Note that the
costs presented do not include the cost of the twist-only case run first. The optimizer adopts a second-
order central finite-difference gradient evaluation, and so each iteration requires two flow solutions
per variable, and a further one or two solutions for the step size evaluation, and the step size
evaluation is always performed in serial on the master node. All cases were run with one core per
design variable and one for the master process. Hence, the total cost scales with the number of design
variables, but the parallel framework means the optimization run time scaling can be reduced to the
number of iterations. It is clear that the global modes are particularly efficient, requiring significantly
fewer evaluations than the off-surface domain element, for similar drag reduction. However, neither
of these approaches has eliminated the wave drag entirely, whereas the local modes have achieved
this for significantly lower cost than the domain element approach. an improved distribution.
35
1.1.10 Conclusions
Aerodynamic shape optimization has been considered, using mathematically derived design
variables. Orthogonal design variables have been extracted by a singular value decomposition
approach where a training library of airfoils is analyzed and decomposed to obtain an efficient and
reduced set of design variables. They are geometric ‘modes’ of the original library, representing
typical airfoil design parameters. In the aerodynamic shape optimization framework a surface and
mesh deformation algorithm is required, and a control point approach has been adopted. This adopts
a small number of control points which are linked to the numerical mesh points by a global volume
interpolation using radial basis functions to allow large, smooth deformations of the mesh. The
performance of the mathematical design variables has been demonstrated in three dimensions, with
results of optimization of the MDO wing in transonic flow. The modal deformations have been applied
as both local and global variables. An important aspect of effective geometric application of these
two-dimensional variables is localization of the deformation field. A basis function deformation
control approach has been developed and presented, allowing improved local control of
deformations, and ensuring exact recovery of global modes from local modes. It has been
demonstrated that the modal approach gives better results than the domain element approach, for
significantly fewer design variables and, furthermore, using global modes, an impressive result is
achieved with only O(10) variables.
36
70 Martins, J. R. R. A. and Hwang, J. T., “Review and Unification of Methods for Computing Derivatives of
Multidisciplinary Computational Models," AIAA Journal, Vol. 51, No. 11, pp. 2582-2599, 2013.
71 Dimitri Mavriplis, Department of Mechanical Engineering, University of Wyoming and the Vision CFD2030
Multidisciplinary Computational Models," AIAA Journal, Vol. 51, No. 11, pp. 2582-2599, 2013.
77 Kenway, G. K. W., Kennedy, G. J., and Martins, J. R. R. A., “Scalable Parallel Approach for High-Fidelity Steady-
State Aero-elastic Analysis and Adjoint Derivative Computations," AIAA Journal, Vol. 52, No. 5, pp. 935-951, 2014.
78 Kenway, G. K. W. and Martins, J. R. R. A., “Multi-point High-fidelity Aero-structural Optimization of a Transport
Aircraft Configuration," Journal of Aircraft, Vol. 51, No. 1, pp. 144-160, 2014.
37
here we use only the components which are relevant for aerodynamic shape optimization, namely,
the geometric parametrization, mesh perturbation, CFD solver, and optimization algorithm.
3.2.1.1 Geometric Parametrization
Many different geometric parameterization techniques have been successfully used in the past for
aerodynamic shape optimization. These include mesh coordinates (with smoothing), B-spline
surfaces, Hicks–Henne bump functions, camber-line-thickness parameterization79, and Free-Form
Deformation (FFD)80. In this work is done using a Free Form Design (FFD) volume approach. The
FFD approach can be visualized as embedding the spatial coordinates defining a geometry inside a
flexible volume. The parametric locations (u; v; w) corresponding to the initial geometry are found
using a Newton search algorithm. Once the initial geometry is embedded, perturbations made to the
FFD volume propagate within the embedded geometry by evaluating the nodes at their parametric
locations. Using B-spline volumes for the FFD implementation, and displacement of the control point
Figure 3.2.1 Shape Design Variables are the z-Displacements of 720 FFD Control Points - (Courtesy of
Martins and Hwang)
locations as design variables. The sensitivity of the geometric location of the geometry with respect
to the control points is computed efficiently using analytic derivatives of the B-spline shape
functions81.
The FFD volume parametrizes the geometry changes rather than the geometry itself, resulting in a
more efficient and compact set of geometry design variables, thus making it easier to manipulate
complex geometries. Any geometry may be embedded inside the volume by performing a Newton
search to map the parameter space to the physical space. All the geometric changes are performed
on the outer boundary of the FFD volume. Any modification of this outer boundary indirectly
modifies the embedded objects. The key assumption of the FFD approach is that the geometry has
constant topology throughout the optimization process, which is usually the case in wing design. In
addition, since FFD volumes are B-spline volumes, the derivatives of any point inside the volume can
be easily computed. Figure 3.2.1 demonstrations the FFD volume and the geometric control points
(red dots) used in the aerodynamic shape optimization. The shape design variables are the
displacement of all FFD control points in the vertical (z) direction.
79 Carrier, G., Destarac, D., Dumont, A., Meheut, M., Din, I. S. E., Peter, J., Khelil, S. B., Brezillon, J., and Pestana, M.,
“Gradient-Based Aerodynamic Optimization with the elsA Software,” 52nd Aerospace Sciences Meeting, 2014.
80 Kenway, G. K., Kennedy, G. J., and Martins, J. R. R. A., “A CAD-free Approach to High-Fidelity Aero-structural
Optimization,” Proceedings of the 13th AIAA/ISSMO Multidisciplinary Analysis Optimization Conference, Fort
Worth, TX, 2010.
81 De Boor, C., A Practical Guide to Splines, Springer, New York, 2001.
38
82 Kenway, G. K., Kennedy, G. J., and Martins, J. R. R. A., “A CAD-free Approach to High-Fidelity Aero-structural
Optimization," Proceedings of the 13th AIAA/ISSMO Multi-disciplinary Analysis Optimization Conference”, 2010.
83 Saad, Y. and Schultz, M. H., “GMRES: A Generalized Minimal Residual Algorithm for Solving Non-symmetric
Linear Systems," SIAM Journal on Scientific and Statistical Computing, Vol. 7, No. 3, 1986, pp. 856-869.
84 Gill, P. E., Murray, W., and Saunders, M. A., “SNOPT: An SQP Algorithm for Large-Scale Constrained
Optimization," SIAM Journal on Optimization, Vol. 12, No. 4, 2002, pp. 979-1006.
39
Table 3.2.1 Aerodynamic Shape Optimization Problem - (Courtesy of Martins and Hwang)
Table 3.2.2 Mesh Convergence Study for the Baseline CRM Wing - (Courtesy of Martins and Hwang)
and symmetry plane meshes for the L0, L1, and L2 grid levels are developed where O-grids of varying
sizes were generated using a hyperbolic mesh generator.
1.1.10.1 Optimization Problem Formulation
The aerodynamic shape optimization seeks to minimize the drag coefficient by varying the shape
design variables subject to a lift constraint (CL = 0.5) and a pitching moment constraint (CMy > -0.17).
The shape design variables are the z-coordinate movements of 720 control points on the FFD volume
and the angle-of-attack. The control points at the trailing edge are constrained to avoid any
movement of the trailing edge. Therefore, the twist about the trailing edge can be implicitly altered
by the optimizer using the remaining degrees of freedom. The leading edge control points at the wing
root are also constrained to maintain a constant incidence for the root section. There are 750
thickness constraints imposed in a 25 chord wise and 30 span wise grid covering the full span and
from 1% to 99% local chord. The thickness is set to be greater than 25% of the baseline thickness at
each location. Finally, the internal volume is constrained to be greater than or equal to the baseline
volume. The complete optimization problem is described in Table 3.2.1.
1.1.11 Single-Point Aerodynamic Shape Optimization
Here, we present our aerodynamic design optimization results for the CRM wing benchmark problem
(Figure 3.2.2) under the nominal flight condition (M = 0.85, Re = 5 x 106). We use the L0 grid (28.8
M cells) for the optimization, thanks to a multilevel optimization acceleration technique that
Figure 3.2.2 Optimized Wing with Shock-Free with 8.5% Lower Drag – (Courtesy of Lyu and Martins)
meaningfully reduces the overall computational cost of the optimization. Our optimization
procedure reduced the drag from 199.7 counts to 182.8 counts, i.e., an 8.5% reduction. The
corresponding Richardson-extrapolated zero-grid spacing drag decreased from 199.0 counts to
181.9 counts. Given that the CRM configuration was designed by experienced aerodynamicists, this
is a significant improvement (although they designed the wing in the presence of the fuselage, which
41
we are ignoring in this problem). At the optimum, the lift coefficient target is met, and the pitching
moment is reduced to the lowest allowed value. The lift distribution of the optimized wing is much
closer to the elliptical distribution than that of the baseline, indicating an induced drag that is close
to the theoretical minimum for a planar wake. This is achieved by fine-tuning the twist distribution
and airfoil shapes. The baseline wing has a near-linear twist distribution. The optimized design has
more twist at the root and tip, and less twist near mid-wing. The overall twist angle changed only
slightly: from 8.06 degree to 7.43 degree. The optimized thickness distribution is significantly
different from that of the baseline, since the thicknesses are allowed to decrease to 25% of the
original thickness, and there is a strong incentive to reduce the airfoil thicknesses in order to reduce
wave drag. The volume is constrained to be greater than or equal to the baseline volume, so the
optimizer drastically decreases the thickness of the gained value drag trade off more promising. To
ensure that the result of our single-point optimization has sufficient accuracy, we conducted a grid
convergence study of the optimized design.
iterations are needed on the L0 grid to converge the optimization. However, the L0 grid requires the
largest computational effort, due to the high cost of the flow and adjoin solutions on this fine grid.
Given that the cost per optimization iteration in the L0 grid is 770 process-hr (compared to 2.9
process-hr for the L2 grid) it is not feasible to perform an optimization using only the L0 grid.
Assuming that the same number of iterations used for the L2 grid (638) would be needed for the L0
grid, the computational cost would be 23 times higher than that of the multilevel approach, which
would correspond to 16 days using 1248 processors.
3.2.5 Multi-Point Aerodynamic Shape Optimization
Transport aircraft operate at multiple cruise conditions
because of variability in the flight missions and air
traffic control restrictions. Single-point optimization
under the nominal cruise condition could overstate the
benefit of the optimization, since the optimization
improves the on-design performance to the detriment
of the off-design performance. In previous sections, the
single-point optimized wing exhibited an
unrealistically sharp leading edge in the outboard of the
wing. This was caused by a combination of the low
value for the thickness constraints (25% of the
baseline) and the single-point formulation. A sharp Figure 3.2.4 Multipoint Optimization
leading edge is undesirable because it is prone to ow Flight Conditions
separation under off-design conditions. We address
this issue by performing a multi-point optimization. The optimization is performed on the L2 grid.
We choose five equally weighted flight conditions with different combinations of lift coefficient and
the Mach number. The flight conditions are the nominal cruise, 10% of cruise CL, and 0.01 of cruise
M, as shown in Figure 3.2.4. More sophisticated ways of choosing multipoint flight conditions and
their associated weights can be used. The objective function is the average drag coefficient for the
have flight conditions, and the moment constraint is enforced only for the nominal flight condition.
A comparison of the single-point and multi-point optimized designs is shown in Figure 3.2.5. Unlike
the shock-free design obtained with single-point optimization, the multipoint optimization settled on
an optimal compromise between the flight conditions, resulting in a weak shock at all conditions. The
leading edge is less sharp than that of the single-point optimized wing. Additional fight conditions,
such as a low-speed flight condition, would be needed to further improve the leading edge. The
overall pressure distribution of the multipoint design is similar to that of the single-point design. The
twist and lift distributions are nearly identical. Most of the differences are in the chord wise C p
distributions in the outer wing section. The drag coefficient under the nominal condition is
approximately two counts higher. However, the performance under the off design conditions is
considerably improved.
1.1.12 Strength of Multi-Point Optimization
To demonstrate the robustness of the multipoint design, we plot ML/D contours of the baseline,
single-point, and multipoint designs
with respect to CL and cruise Mach in
Figure 3.2.6 where ML/D provides a
metric for quantifying aircraft range
based on the Breguet range equation
with constant thrust specific fuel
consumption. While the thrust-specific
fuel consumption is actually not
constant, assuming it to be constant is
acceptable when comparing
performance in a limited Mach number
range. We add 100 drag counts to the
computed drag to account for the drag
due to the fuselage, tail, and nacelles,
and we get more realistic ML/D values.
The baseline maximum ML/D is at a
lower Mach number and a higher CL
than that of the nominal flight
condition. The single-point
optimization increases the maximum Figure 3.2.6 Comparison of Baseline, Single, and Multipoint
Optimization
ML/D by 4% and moves this maximum
toward the nominal cruise condition. If
we examine the variation of ML/D along the CL = 0.5 line, we see that the maximum occurs at the
nominal Mach of 0.85, which corresponds to a dip in a drag divergence plot. If we examine the
variation of ML/D along the CL = 0.5 line, we see that the maximum occurs at the nominal Mach of
0.85, which corresponds to a dip in a drag divergence plot. For the multipoint optimization, the
optimized flight conditions are distributed in the Mach-CL space, resulting in an attended ML/D
variation near the maximum, which means that we have more uniform performance for a range of
flight conditions.
If we examine the variation of ML/D along the CL = 0.5 line, we see that the maximum occurs at the
nominal Mach of 0.85, which corresponds to a dip in a drag divergence plot. For the multipoint
optimization, the optimized flight conditions are distributed in the Mach-CL space, resulting in an
attended ML/D variation near the maximum, which means that we have more uniform performance
for a range of flight conditions. In aircraft design, the 99% value of the maximum ML/D contour is
45
often used to examine the robustness of the design. The point with the highest Mach number on that
contour line corresponds to the Long Range Cruise (LRC) point, which is the point at which the
aircraft can fly at a higher speed by incurring a 1% increase in fuel burn. In this case, we see that the
99% value of the maximum ML/D contour of the multipoint design is larger than that of the single-
point optimum, indicating a more robust design.
46
German Aerospace Center, Institute of Aerodynamics and Flow Technology, Braunschweig, Germany, 26 th
International Congress Of The Aeronautical Sciences, ICASE 2008.
47
Figure 3.3.1 Flow Field Around the Wing Section of a 3-Element Wing – Courtesy of [Brezillon et al.]
In order to capture all of these critical flow features a grid of high quality is required. This implies the
use of thin cells not only normal to the wall for the accurate discretization of the boundary layers but
also behind the trailing edge of the upstream element and up to the downstream element in order to
accurately resolve the wakes until its confluence with the boundary layer of the following elements.
A thick layer of thin cells has then to be generated orthogonal to the wall in order to capture the
mixing shear layers and the ability of the boundary layer to sustain separation. Furthermore, the
high-lift configuration has complicated areas like slots, slat and flap coves, which have to be meshed
as well. The selection of the meshing procedure for high-lift design is then a compromise between
the flexibility to handle complex configurations, the number of cells for fast computations and the
capability to control the space discretization in particular areas.
A comparison between block-structured and hybrid unstructured approaches for high-lift
48
method. 40th AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper 2002-844, Reno NV, January 2002.
95 Storn R, Price K. Differential Evolution - a simple and efficient adaptive scheme for global optimization over
continuous spaces. Technical Report, International Computer Science Institute, TR-95-012, Berkley, 1995.
96 Rowan T. Functional stability analysis of numerical algorithms, Thesis, Department of Computer Sciences,
Soni, and N. Weatherill, CRC Press, Boca Raton, FL, 1999, pp. 25.1-25.18.
101 Wild J, Niederdrenk P, Gerhold, T. Marching generation of smooth structured and hybrid meshes based on
metric identity. 14th International Meshing Roundtable, edited by B. Hanks, Springer, Berlin, pp. 109-127, 2005.
50
NETGEN of [Schöberl]102 as black box for the generation of the unstructured mesh part. The
structured hexahedrons are connected to the unstructured part by collapsing the outer hexahedral
layer into prisms and tetrahedrons in order to achieve a smooth transition of the control volumes for
the reduction of numerical errors and instabilities. It has already been shown that applying this mesh
type can greatly speed up RANS flow computations without losing solution accuracy, mainly due the
reduced number of points103. This second approach has been selected for the setting optimization of
the FNG configuration that is a 3D high-lift configuration featuring a slat and a flap that does not
extend the whole wing span.
1.1.13 CFD Flow Solvers
For the resolution of the flow around the high-lift configuration the block-structured FLOWer-Code
104-105 and the unstructured TAU-Code 106-107-108 are available at DLR. Both codes are well established
tools for aerodynamic applications in DLR, aerospace industry and universities109-110. Both codes
solve the compressible, three dimensional, unsteady Reynolds-Averaged Navier-Stokes (RANS)
equations for rigid bodies in arbitrary motion. For spatial approximation a finite-volume method
with second order upwind or central discretization with scalar or matrix artificial dissipation is
available. In FLOWer cell centered and cell vertex formulations are provided, whereas TAU uses a
vertex centered dual mesh formulation.
The discrete equations are integrated explicitly by multistage Runge-Kutta schemes, using local time
stepping and multigrid acceleration. In FLOWer the explicit scheme is used in combination with
implicit residual smoothing, whereas in TAU the implicit LU-SGS scheme is additionally available.
Preconditioning is used for low speed flow simulations. Various turbulence models are available,
ranging from eddy viscosity to full differential Reynolds stress models including options for DES
(Detached Eddy Simulation). For aerodynamic optimization, the Spalart-Allmaras model with
Edwards modification is preferred for its accuracy and robustness. For a long time only the low
Reynolds number formulation has been available in both codes for accuracy reasons, but recently
model specific “universal” wall-function have been introduced in the TAU code to achieve a higher
efficiency of the solver, especially for use in optimization.
102 Schöberl J. NETGEN - An advancing front 2D/3D mesh generator based on abstract rules. Computing and
Visualization in Science, Vol. 1, pp. 41-52, 1997.
103 Wild J. Acceleration of aerodynamic optimization based on RANS-Equations by using semi-structured grids.
ERCOFTAC Design Optimization: Methods & Applications, Paper ERCODO2004_221, Athens (Greece), 2004.
104 Raddatz J, Fassbender JK. Block structured Navier-Stokes solver FLOWer. MEGAFLOW – Numerical Flow
Simulation for Aircraft Design. Notes on Numerical Fluid Mechanics and Multidisciplinary Design (NNFM), Vol.
89, pp. 27-44, ISBN 3-540-24383-6, 2005.
105 Kroll N, Radepiel R, Rossow CC. Accurate and efficient flow solvers for 3D applications on structured meshes.
Design. Notes on Numerical Fluid Mechanics and Multidisciplinary Design (NNFM), Vol. 89, pp. 81-92, ISBN 3-
540-24383-6, 2005.
108 Schwamborn D, Gerhold T, Heinrich R. The DLR TAU-Code: Recent Applications in Research and Industry.
ECCOMAS CDF 2006, In proceedings of “European Conference on Computational Fluid Dynamics“, Delft The
Netherland, 2006.
109 Kroll N, Rossow CC, Schwamborn D. The MEGAFLOW-project - numerical flow simulation for aircraft. A.
DiBucchianico, R.M.M. Mattheij, M.A. Peletier (Edts.), Progress in Industrial Mathematics at ECMI 2004,
Springer, New York, S. 3 – 33, 2005.
110 Kroll N, Fassbender JK. MEGAFLOW – Numerical flow simulation for aircraft design. MEGAFLOW - Numerical
Flow Simulation for Aircraft Design. Notes on Numerical Fluid Mechanics and Multidisciplinary Design (NNFM),
Vol. 89, ISBN 3-540-24383-6, 2005.
51
This approach is able to deliver results nearly as good as the low Reynolds number approach for
pressure and skin friction distributions over a wide range of y+ values for the first cell height at the
wall, while saving computational time and memory. Both codes can be run in target-lift mode in order
to compute the aerodynamic state at a given lift coefficient by automatically adjusting the angle of
attack. Since the efficiency of the flow solver is an important key to reduce the turnaround time
dedicated strategies have been developed. On multi-block structured meshes, the solver has to
handle large vectors and the FLOWer has been accordingly optimized to efficiently run on vector
supercomputers like the NEC-SX series.
In the same way, a solver based on unstructured mesh has to resolve a single large computational
domain that can easily be partitioned in an arbitrary number of domains. Based on domain
decomposition and the message passing concept using MPI TAU reaches a high level of efficiency on
parallel computers thanks to its optimization for cache processors through specific edge coloring
procedures and load balancing adjusted to the solver needs and the number of requested processors.
Another step forward has been achieved thank to the discrete adjoint solver that was developed
within TAU enabling cost-efficient gradient-based aerodynamic optimization. The advantage of the
adjoint method is its ability to evaluate the gradient of a single cost function with respect to a large
number of design variables with an effort that scales weakly with the number of design variables.
The method consists of the explicit construction of the exact Jacobian of the spatial discretization
with respect to the unknown variables. A wide range of spatial discretization available in TAU has
been differentiated, including the Spalart-Allmaras-Edwards one-equation turbulence model. The
effect of various approximations of the Jacobian was first investigated on 2D cases and the efficiency
of the adjoint approach has been demonstrated on several 2D and 3D optimization problems111-112.
In the present study, the robustness of the linear solver has been improved for high-lift application
by applying the Generalized Minimum Residual (GMRES) method in its restarted form that requires
storage of a given number of Krylov basis function113. The metric terms required for the gradient
computation are reliably evaluated by finite differences, which imply the necessity of keeping
constant the number of grid points while varying the design variables. This process can be done
during the CFD run and once the aerodynamic and adjoint states are available, the gradient is
extremely fast to compute. See [Brezillon et al]114.
3.3.4.1.1 Optimization of the DLR-F11 Configuration
The first configuration optimized is the so called DLR-F11 model with full span flap and slat in take-
off configuration. This model is a wide-body Airbus-type research configuration with a half span of
1.4 m that can feature different degrees of complexity115. In the present study, six design variables
are selected to modify the deflections, the horizontal and the vertical positions of the flap and the
slat. The geometric changes are propagated homogenously along the span. The goal is to maximize at
111 Brezillon J, Dwight R. Discrete adjoint of the Navier-Stokes equations for aerodynamic shape optimization.
Evolutionary and Deterministic Methods for Design, Optimization, and Control with Applications to Industrial and
Societal Problems EUROGEN 2005, edited by R. Schilling, W. Haase, J. Pé riaux, H. Baier and G. Bugeda, FLM, ISBN:
3-00-017534-2, Munich, 2005.
112 Brezillon J, Brodersen O., Dwight R, Ronzheimer A, Wild J. Development and application of a flexible and
efficient environment for aerodynamic shape optimization. ONERA-DLR Aerospace Symposium (ODAS),
Toulouse, 2006.
113 Saad Y, Schultz MH. A generalized minimum residual algorithm for solving non-symmetric linear systems. SIAM
German Aerospace Center, Institute of Aerodynamics and Flow Technology, Braunschweig, Germany, 26 th
International Congress Of The Aeronautical Sciences, ICASE 2008.
115 Rudnik R, Thiede P. European research on high lift commercial aircraft configurations in the EUROLIFT
a single take-off condition (M = 0.3, Re =20 x 106, AoA = 8°) a derived expression of the lift to drag
ratio: Obj = CL3 /CD2 .
This performance indicator, based on the climb index, has already been successfully employed for
flap design based on 2D computations and turned to be better suited than the lift to drag ratio.
Additionally, the lift is not allowed to decrease and the angle of attack is kept fix. In order to make a
more realistic optimizations the weight of the high-lift system kinematics, which depends on the
horizontal deployment capability, is taken into account by penalizing the objective function to avoid
too heavy a mechanism. The relation between the horizontal displacement and the penalty is set
according to industrial specifications.
Figure 3.3.2 Multi-Block Structured Mesh Around the DLR-F11 Model in Full Span Flap and Slat
Configuration – Courtesy of [Brezillon et al.]
The upper part of the main wing is discretized by 73 points to capture the boundary layer and the
wake of the slat, see Figure 3.3.3, while 49 points are set for the slot between the main element and
the flap. The complete process, from reading the design variables to writing the mesh in either
structured or unstructured formats takes about 1 minute on a single AMD Opteron 2.5 GHz processor.
The numerical simulations, either with the FLOWer or the TAU codes, are based on the RANS
equations and the Spalart-Allmaras-Edwards turbulence model. For fast convergence, the low Mach
number preconditioning approach is adopted and the steady state is reached by a Runge-Kutta
scheme using multigrid W-cycles on 3 levels. A fully converged solution with almost 5 orders of
density residual decrease is obtained after 1,500 FLOWer or 5,000 TAU cycles. Since the integration
scheme to get convergence is differently implemented with more operations for the structured code
and since the mesh for TAU computations can be partitioned in an arbitrary number of domains, the
total time needed to converge the flow can be almost identical for both codes.
Figure 3.3.3 Close View of the Wake Discretization on the Top of the DLR-F11 Wing – Courtesy of
[Brezillon et al.]
individuals, the evaluation of a single population on a cluster of 5 NEC-SX8 CPUs is reduced to 1.5
hours, wall-clock time.
In the same way, in order to speed up the optimization process with the NLPQLP strategy, the
components of the gradients are computed simultaneously. Since the convergence of the
optimization process here depends of the accuracy of the gradients, the aerodynamic computations
are fully converged and the gradients are computed with central finite differences: based on a NEC-
SX8 cluster with 6 processors, the turnaround time for the computation of the gradients is equivalent
to only 2 flow computations, i.e.
4 hours, wall clock time,
instead of 24 hours as in a
sequential approach.
The SubPlex approach, which
involves sequential evaluations
only, does not benefit from
possible simultaneous
computations. A speed up
could be obtained by running
the FLOWer code in parallel on
the NEC, but this was not tested
during the study.
Figure 3.3.4 presents the
evolutions of the optimizations
processes during 14 days, wall
clock time. For the clarity of Figure 3.3.4 Objective function according to the wall-clock time -
the figure, only the solutions Aerodynamic flows computed with the structured FLOWer code
that fulfill the constraint on the running sequentially on a NEC-SX8 – Courtesy of [Brezillon et al.]
lift level within 0.2% are
presented. In order to measure the convergence of the DE algorithm, the best objective and the
average objective within one generation are presented. The DE ran up to 162 generations,
representing more than 1,600 evaluations, and stopped since the convergence reaches a plateau and
the average solution is close to the best optimum. In total the optimizations process required 10 days
to converge. The SubPlex algorithm
that involves here only one single
CPU, presents a convergence rate
close to the average solutions from
DE. After 14 days and 168 CFD
computations, the SubPlex was
almost converged and the process
was voluntary stopped. On the
contrary, the gradient-based
strategy was completely converged
after 3 days of simulations and 123
evaluations including 108 for the
gradients only.
3.3.4.1.4 Optimizations Using the
TAU Code
In a second attempt, the Figure 3.3.5 Objective and Lift Coefficient According to the
aerodynamic flow is computed with wall-clock time - Aerodynamic and adjoint flows computed
the TAU code running in parallel and with the unstructured TAU code – Courtesy of [Brezillon et al.]
55
the solution is fully converged after 3 hours, wall-clock time, on a cluster of 32 AMD Opteron 2.4 GHz
processors. In order to plenty use the capability of the TAU code, the adjoint mode is here used to
efficiently compute the gradients of the lift and drag coefficients. The adjoint states, computed
simultaneously on 2 clusters of 16 processors each, the gradients of the objective function and the
constraint - derived from the drag and lift gradients are obtained within 3 hours. Figure 3.3.5
presents the evolution of the optimizations process obtained with the NLPQLP strategy coupled to
the adjoint approach for the computations of the gradients. After 13 evaluation and 78 hours of
simulations the optimizations converged to a maximum with a limited deviation on the lift coefficient.
Thanks to the adjoint approach, the process is now almost independent to the number of design
variables and a more complex optimization problem involving more design parameters should
require almost the same turnaround time.
3.3.4.2 Synthesis
All optimizations presented above were obtained within 14 days but with different solvers or levels
of convergence. For the sake of comparison, the optimums previously obtained are recomputed with
the same flow solver and the same level of convergence and the deviations to the baseline
configuration are
summarized in Table
3.3.1. One can sees that all
optimization strategies
give almost the same
improvement with a slight
advantage of the NLPQLP
strategy, independently to
the way of computing the
gradients. Figure 3.3.7 Table 3.3.1 Variations of the Objective Function, Drag (dc=drag count)
(a-b) present a and Lift Coefficients – Courtesy of [Brezillon et al.]
comparison of the slat and
flap positions at a given spanwise position and confirm that all optimums are located at almost the
same position and that both optimums given by the NLPQLP strategy are identical. It can be guessed
that the design space is rather shallow close to the optimum and the optimum can be reached only
with a fully converged optimizations
process, as for the NLPQLP cases. The
performance improvement is made
evident by plotting the drag
distribution in spanwise direction for
each element, see Figure 3.3.6, the
optimizations has almost no
influence on the drag of the body and
the flap but permits to made further
negative the drag on the slat. This
improvement has to be paid by a
lower drag increase on the main wing
and the optimized configuration has
in total 17.8 counts less drag than on
the baseline configuration.
(a) Slat Setting for the Baseline (b) Flap Setting for the Baseline
and Optimised Configurations and Optimised Configurations
Figure 3.3.7 Optimizing Configuration for Slat and Flap – Courtesy of [Brezillon et al.]
is made for a take-off configuration at flight conditions (M = 0.18, Re = 18 x 106) and a constant lift
coefficient of CL =1.25. In order to achieve the latter the calculations are all made in target lift mode,
where the angle of attack is adapted during the flow calculation. The objective of the optimization is
the minimization of drag coefficient. In order to eliminate even smallest deviations of the lift
coefficient and its influence on the induced drag, the value used was the actual drag coefficient minus
the ideal elliptic induced drag, commonly known as profile drag
(CL )2
CD 𝑝 = CD −
πΛ
Eq. 3.3.1
3.3.5.1 Parameterization and Meshing Procedure
The number of design parameters is increased compared to the previous case by allowing different
deflections of the inboard and outboard slat. This reflects that usually for transport aircraft with the
engines mounted under the wing the slat is not continuous at the pylon. So there is neither a need
nor a restriction for matching slat deflections inboard and outboard the engine. This extends the
geometry parameterization to 9 degrees of freedom for the rigid body positioning of the 3
independent devices. As for the DLR-F11 the geometric changes in gap and overlap are propagated
at this stage homogenously along the device span. Since the geometric restrictions at the device ends
116 Dargel, G., Hansen, H., Wild, J., Streit, T., Rosemann, H. Aerodynamische flügelauslegung mit
multifunktionalen steuerflächen (aerodynamic design of wings with multi-functional control devices). DGLR
Jahrbuch 2002, DGLR-2002-096, Vol I, DGLR, (2002).
57
and their interference with the remaining clean wing tip do not allow for an easy application of block
structured meshes, here the mixed mesh type was used.
3.3.5.2 Results
For this design only the SubPlex method was applied. The gradient method based on the adjoint could
not be applied here due to the variation of the number of grid points in the unstructured mesh part.
The subspace simplex method was run for 5 cycles resulting in overall 150 flow calculations. For the
flow calculations 96 processors have been used, resulting in an overall turn-around time of roughly
375 hours or 15 days. The optimum solution has already been found after the third loop, the rest of
the optimization secures that there is no better optimum within the range of the parametric changes.
Neglecting these iterations would decrease the turn-around time to approximately 11 days. The
objective was reduced by 0.83%, which corresponds to approximately 4 dc. The drag reduction
corresponds to a slight gap increase over the whole wing span. The lift coefficient was kept constant
within a margin of 0.03%. The relatively small amount of improvement may be explained by the fact
that the wing section of the FNG high-lift wing has already been designed applying numerical
optimization in 2D and the spanwise variation of the geometry was not included in full extend into
the presented exercise.
3.3.5.3 Conclusion
The methods and tools developed and successfully applied to perform the settings optimizations of
3D high-lift configurations within a limited time frame were presented. The major key technology to
optimize such complex configurations is the meshing procedure and two ways followed at DLR were
discussed in detail. The development of efficient and robust numerical methods and the increase of
the computer performance permit fast computations of complex flows and makes feasible numerical
optimizations within limited turn-around time. It was observed that the evolutionary algorithm is
very attractive thanks to his inherent possibility to compute a population in parallel. For strategies
involving only sequence of simulations, the reduction of the turnaround time can be achieved by
conducting CFD in parallel. The gradient-based strategy tested here has permitted to fully converge
the optimizations problem within a very short time frame and provides here the best solution. The
inclusion of the adjoint approach gives new opportunities to treat more complex optimizations
problems in limited turnaround time and will be further investigated at DLR.
58
117Eric J. Nielsen, W. Kyle Anderson, “Aerodynamic Design Optimization on Unstructured Meshes Using the
Navier-Stokes Equations”, AIAA-98-4809.
59
equations so that the resulting sensitivity derivatives are consistent with those obtained using finite
differences.
In addition to a compressible solver, an incompressible formulation is also differentiated in order to
accommodate a wide range of applications. The accuracy of the resulting derivatives is established
and sample calculations are shown for 2 and 3D cases using both implementations. Conclusions and
suggestions for future research are also given.
3.4.2 Nomenclature
Be advised that throughout this article the notation M is reserved as million. The Mach number is
self-explanatory. Others notations are listed as:
60
∂𝐐
V + ∮(𝐅⃗. 𝐧
̂) dΩ = 0
∂t
Ω
Eq. 3.4.1
118Eric J. Nielsen, W. Kyle Anderson, “Aerodynamic Design Optimization on Unstructured Meshes Using the
Navier-Stokes Equations”, AIAA-98-4809.
61
ρ ρu
2
ρu ρu + p − τxx
𝐐 = ρv , 𝐅⃗ = ρuv − τxy +
ρw ρuw + p − τxz
[E] [(E + p)u − uτxx − vτxy − wτxz + qx ]
ρv
ρuv − τxy
ρv 2 + p − τyy +
ρvw − τyz
[(E + p)v − uτxy − vτyy − wτyz + qy ]
ρw
ρuw − τxz
ρvw − τyz
ρw 2 + p − τzz
[(E + p)w − uτxz − vτyz − wτzz + qz ]
2M∞ ∂u ∂v ∂w
τxx = (μ + μt ) (2 − − )
3Re ∂x ∂y ∂z
2M∞ ∂v ∂u ∂w
τyy = (μ + μt ) (2 − − )
3Re ∂y ∂x ∂z
2M∞ ∂w ∂u ∂v
τzz = (μ + μ𝑡 ) (2 − − )
3Re ∂z ∂x ∂y
2M∞ ∂u ∂v
τxy = τyx = (μ + μt ) ( + )
3Re ∂y ∂x
2M∞ ∂w ∂u
τxz = τzx = (μ + μt ) ( + )
3Re ∂x ∂z
2M∞ ∂v ∂w
τyz = τzy = (μ + μt ) ( + )
3Re ∂z ∂y
M∞ μ μt ∂a2
qx = − ( + ) ,
Re(γ − 1) Pr Prt ∂x
M∞ μ μt ∂a2
qy = − ( + ) ,
Re(γ − 1) Pr Prt ∂y
M∞ μ μt ∂a2
qz = − ( + )
Re(γ − 1) Pr Prt ∂z
Eq. 3.4.2
The equations are closed with the equation of state for a perfect gas
62
u2 + v 2 + w 2
and p(γ − 1) [E − ρ ( )]
2
Eq. 3.4.3
and the laminar viscosity is determined through Sutherland's Law.
3.4.3.1 Adjoint and Design Equations
In the adjoint approach for design optimization, a cost function is defined and augmented with the
flow equations as constraints to form a Lagrangian given by
dL ∂f ∂𝐗 T ∂f ∂𝐐 T ∂f ∂𝐑 T ∂𝐑 T ∂𝐗 T ∂𝐑 T
={ +[ ] } + [ ] { + [ ] Λ} + {[ ] + [ ] [ ] } Λ
d𝐃 ∂𝐃 ∂𝐃 ∂𝐗 ∂𝐃 ⏟∂𝐐 ∂𝐐 ∂𝐃 ∂𝐃 ∂𝐗
0
Eq. 3.4.5
Because Λ is arbitrary, the terms multiplied by [∂Q/∂D]T may be eliminated using the following
equation
∂f ∂𝐑
− = [ ] ΛT
∂𝐐 ∂𝐐
Eq. 3.4.6
Eq. 3.4.6 is a linear system which represents the discrete adjoint equation for the optimization
problem. After the flow equations have been solved for Q , the adjoint equation can be solved for the
unknown vector of Lagrange multipliers Λ . The remaining terms in Eq. 3.4.5 can be used to evaluate
the sensitivity derivatives as follows:
dL ∂f ∂𝐗 T ∂f ∂𝐑 T ∂𝐗 T ∂𝐑 T
={ +[ ] } + {[ ] + [ ] [ ] } Λ
d𝐃 ∂𝐃 ∂𝐃 ∂𝐗 ∂𝐃 ∂𝐃 ∂𝐗
Eq. 3.4.7
After the solution for the costate variables is obtained using Eq. 3.4.6, the vector containing all of the
desired sensitivities can be evaluated as a single matrix-vector product, given by Eq. 3.4.7.
3.4.4 Numerical Implementation
3.4.4.1 Flow Equations
The flow solvers used in the current work are described at length in [1, 3, and 5]. The codes use an
implicit, upwind, finite-volume discretization, in which the dependent variables are stored at the
mesh vertices. Inviscid fluxes at cell interfaces are computed using the upwind schemes of Roe [21]
van Leer [29] or Osher [6]. Viscous fluxes are formed using an approach equivalent to a central-
difference Galerkin procedure. Temporal discretization is performed using a backward-Euler time-
stepping scheme, and multigrid acceleration can be used for the two-dimensional codes [5]. An
approximate solution of the linear system of equations formed at each time step is obtained using
63
several iterations of a point-iterative scheme in which the nodes are updated in an even odd fashion,
resulting in a Gauss-Seidel-type method.
The turbulence model is solved separately from the flow equations at each time step, using a
backward-Euler time-stepping scheme. The resulting linear system is solved using the same point
iterative scheme employed for the flow equations. The turbulence model is integrated all the way to
the wall without the use of wall functions. The incompressible solvers are based on an artificial
compressibility formulation, and are described in [3]. Time integration and solution of the linear
system at each time step are performed in the same manner as described above.
1.1.13.1 Adjoint and Design Equations
The adjoint equation given in Eq. 3.4.6 represents a linear set of equations for the costate variables
A . Although this system can be solved directly using GMRES [22], a time-like derivative is added and
the solution is obtained by marching in time, much like the flow solver:
V ∂𝐑 T n ∂f ∂𝐑
{ 𝐈 + [ ] }∆ Λ = − − [ ] Λn where Λn+1 = Λn + ∆n Λ
∆t ∂𝐐 ∂𝐐 ∂𝐐
Eq. 3.4.8
The time term can be used to increase the diagonal dominance for cases in which GMRES alone would
tend to stall. This ultimately results in a more robust adjoint solver. Due to the large amount of code
resulting from the linearization of the viscous terms and the turbulence model, these contributions
are stored in the present implementation. Because the stencil for the inviscid contributions is larger,
the linearization of these terms is recomputed at each step to avoid the need for extra storage and
data structure.
To precondition the linear system, an incomplete LU decomposition of the matrix obtained from a
first-order accurate discretization is used. The preconditioning is applied on the left and no fill-in is
allowed [3]. Numerical experiments using this preconditioner have shown that some turbulent cases
are slow to converge. An alternate means of preconditioning that has often been found useful is to
employ a point-iterative scheme similar to that used for the flow equations. This technique allows for
continual Improvement in preconditioning the first-order system but is most effective when the time
step is small and the matrix is diagonally dominant.
In the present work, the differentiation of both the flow equations and the turbulence model is
accomplished by "hand differentiating" the code. For obtaining the solution of the adjoint equations,
the turbulence model is tightly coupled during the solution process, whereas it is solved separately
during the flow analysis. During development, various treatments of the turbulence model have been
studied and it has been found that the close coupling of the turbulence model is required in order to
obtain sensitivity derivatives consistent with those obtained using finite differences. This will be
illustrated in a subsequent section.
1.1.13.2 Cost Functions
For both two and three dimensions, the cost function is composed of a linear combination of the lift
and drag coefficients:
f = ω1 (Cl − Cl∗ )2 + ω2 (Cd − Cd∗ )2
Eq. 3.4.9
For two-dimensional calculations, a target pressure distribution can also be specified. The drag can
be minimized while maintaining a specified lift by adjusting the weights associated with each term
in Eq. 3.4.9. The weights must be chosen such that neither term dominates the other. The current
method for choosing the initial weights is to simply set the ratio of ω2 to ω1 to be equal to the ratio of
the lift to the drag:
64
ω2 C l
=
ω1 Cd
Eq. 3.4.10
During the design process, these weights may require adjustment. However, this avoids the need to
solve separate adjoint equations for lift and drag.
3.4.4.2 Design Variables
For both incompressible and compressible flows, the angle of attack can be utilized as a design
variable in addition to the shape. For compressible flows, the free-stream Mach number can also be
specified as a design parameter. When the shape is evolving, the surfaces are parameterized as
described in the following sections.
3.4.4.2.1 Two-Dimensional Parameterization
For two-dimensional cases, the geometry is described using B-splines and the coordinates of the
control points are used as design variables as described in [2 and 4]. Translation and rotation of
individual bodies can also be used as design variables. Although not discussed further here, a
graphical interface has been developed which aids the user in selecting and placing limits on design
variables, as well as modifying target pressure distributions [2]. This interface has proven to be very
useful and helps reduce errors in setting up design cases.
3.4.4.2.2 Three-Dimensional Parameterization
The three-dimensional code has been coupled with a geometric parameterization scheme recently
introduced by [Samareh] method utilizes a free-form deformation technique similar to that used in
the motion picture industry for animating digital images. Here, a Bezier net describing the changes
in the geometry is placed around the baseline mesh. The control points in the net may be used directly
as the design variables, or they may be further grouped into design variables such as camber,
thickness, and twist. This parameterization technique has been chosen for its ability to handle
arbitrary geometries and because the mesh generation process does not depend on a prior
parameterization of the geometry. This allows meshes which have been previously generated solely
for analysis to be utilized for design purposes.
3.4.4.3 Geometric Constraints
During the design process, the feasibility of the geometry is maintained by limiting the movement of
the design variables. For the two-dimensional code, area and curvature constraints may also be
placed on the geometry although these are not employed in the present work. The curvature
constraints can be used to enforce a specified leading edge radius or to guarantee curvatures of a
specified sign.
3.4.5 Grid Generation and Mesh Movement Strategy
For all of the two-dimensional computations, the meshes have been generated using the method
described In Ref. 17. In three dimensions, the method of Ref. 20 is utilized. Both techniques employ
an advancing front methodology and generate good quality grids for both inviscid and viscous
calculations. When the design process requires modifications to the surface geometry, the
computational mesh must be deformed to reflect the changes. For inviscid flows, the mesh movement
strategy is based on the spring analogy described in Ref. 30. The edges of the mesh are treated as
tension-springs, and the following equation is solved using a Jacobi iteration process:
∑ K ij (∆xi − ∆xj ) = 0
j∈Nt
Eq. 3.4.11
65
Here Δxi, and Δxj, represent the change in the coordinates of nodes i and j from the initial mesh to the
desired mesh. The spring constants Kij are assumed to be lij-2 , where l is the length of the edge
connecting node i to node j . Since this technique may result in crossed grid lines, the required shift
of the surface coordinates is decomposed into a series of smaller movements (usually around 10),
and Eq. 3.4.11 is relaxed for each change in the surface. This strategy has been found to work well
for Euler-based designs.
For viscous meshes, the method described above is not adequate and can easily lead to crossing mesh
lines and negative volumes. For these cases, the nodes near viscous surfaces are shifted by
interpolating the changes in the coordinates at the boundaries of the nearest surface triangle or edge.
This technique is blended with a smoothing procedure so that away from the highly stretched cells
near the surface, the mesh movement reverts to that of the procedure described above for inviscid
meshes. Further details can be found in Ref. 4.
For two-dimensional viscous applications, the procedure described above is very robust and is
capable of successfully deforming the mesh in response to large changes in the surface geometry. For
multielement airfoils, there is a tendency to open "gaps" in the mesh between elements when the
elements are allowed to translate away from one another. A similar problem occurs when the
elements move closer together in which case there is a "jamming" together of mesh points. These
difficulties are simply due to the fact that no additional mesh points are inserted or removed during
the process so that as elements shift in relation to one another, voids can be created. This has not had
a detrimental effect on the flow solver and can be remedied by periodically regenerating the mesh.
For three-dimensional flows, the mesh movement procedure is inadequate when large changes in the
geometry are required. In these cases, negative cell volumes can occur around the edges of the
planform. Therefore, changes in the thickness and camber have been limited to only a few percent of
the chord. Further research is required to develop a more reliable methodology for large geometric
changes in three dimensions.
As the surface is deformed during the design process, there is a corresponding change in the interior
mesh points as well. The effect of the changing grid is reflected through the mesh sensitivity terms
given by ∂X/∂D in Eq. 3.4.7. The computation of these terms is achieved by differentiating the mesh
movement process described above.
3.4.6 Optimization Technique
The optimization technique used in all of the results presented below is the quasi-Newton method of
Davidon-Fletcher-Powell. [8,11]. The current implementation of this technique, referred to as
KSOPT, allows for multipoint optimization as well as both equality and inequality constraint. For the
present work, the multipoint capability is not utilized although this is an obvious future requirement.
3.4.7 Results and Discussion
3.4.7.1 Consistency of Linearization
During code development, great care has been taken at each step to ensure that the derivatives are
consistent with those obtained using finite differences. In this section, the accuracy of the resulting
derivatives is verified for compressible flow in both two and three dimensions. Similar results have
been obtained for incompressible flows and are included in Appendix B of [Nielsen and Anderson]119.
Comparisons are made between derivatives computed using finite differences with those obtained
using the adjoint method. When computing derivatives using finite differences, central-difference
formulas are used with a step size of l x 10 -5 and, the flow solver is converged to machine accuracy.
The two-dimensional results shown are calculated using the Osher flux function although similar
accuracy is obtained using either flux-difference splitting or flux-vector splitting. For the three
dimensional linearization, all results are obtained using the flux difference splitting scheme of Roe.
119Eric J. Nielsen, W. Kyle Anderson, “Aerodynamic Design Optimization on Unstructured Meshes Using the
Navier-Stokes Equations”, AIAA-98-4809.
66
In order to further demonstrate the accuracy of the differentiation, a case containing transonic flow
is examined. An RAE 2822 airfoil is used at an angle of attack of 2.81∘, a Mach number of 0.75, and a
Reynolds number of 6.2 M. The mesh contains 14,127 nodes and the spacing at the wall is 1 x 10 -5.
The computed pressure distribution is shown in Figure 3.4.1 along with the corresponding
experimental data. For
this case, a strong shock
is present on the upper
surface which separates
the flow immediately
downstream. The
locations of the three
design variables are
Figure 3.4.4 Location of Design Variables for RAE 2822 Airfoil
shown by the filled
circles in Figure 3.4.4.
The corresponding sensitivity derivatives
for the lift coefficient with respect to a
vertical movement of the control points
are listed in Error! Reference source not f
ound. along with those for Mach number
and angle of attack. The agreement with
finite differencing is very good, although
the derivatives for the design variables
located in the separate region of the flow
just downstream of the shock appears to
be slightly inconsistent. However,
numerical experiments using different
time steps have shown the finite
differences for this control point are
somewhat sensitive to the perturbation
level. For step size of 10-6 and 5 x 10-5 Table 3.4.1 Accuracy of 2D Derivatives for Drag
result in finite difference derivatives of Coefficient
1.9187 and 1.9150 respectively.
3.4.7.1.2 3D Accuracy
To verify the accuracy of the derivatives in three dimensions, a similar experiment is conducted. For
this case, an ONERA M6 wing [24] has been parameterized using 46 design variables describing the
planform, twist, shear, thickness, and camber. The design variables are depicted in Figure 3.4.5
where twist and wing shear have been parameterized at five spanwise locations. The thickness and
camber have also been parameterized using the six locations shown in the figure. The design
variables describing the planform are not shown in the figure nor are thickness and camber design
variables along the leading and trailing edges. The mesh used for these tests contains 16,391 nodes
and 90,892 tetrahedra and is shown in Figure 3.4.5 - right.
The flow conditions are an angle of attack of 2∘, a Reynolds number of 5 M based on the mean chord,
and a Mach number of 0.3. In this test, the cost function is a combination of the lift and drag
coefficients so that only one adjoint solution is required. The derivatives of the cost with respect to
the angle of attack and the Mach number as well as the derivatives with respect to four of the shape
parameterization variables. As can be seen, the consistency between the derivatives obtained with
the adjoint formulation and finite differences is excellent. Additional derivatives for the design
variables depicted in Figure 3.4.5 have also been verified with comparable accuracy.
68
Figure 3.4.5 Location of design variables for ONERA M6 wing & Grid Assessment of 3D Design
Sensitivities
3.4.7.2.2 "Frozen"
Turbulence Model
An accurate linearization of
the turbulence model can be
difficult to obtain. As seen from
the equations given in
Appendix A of [Nielsen and
Anderson]120, there are many
terms and additional functions
that must be properly
differentiated. These terms
exhibit complex dependency
on both the flow variables as
well as the distance to the wall.
By assuming that the
turbulence model is "frozen", a
significant reduction in the
required level of effort may be
obtained. This approach has
been previously used in [15
and 26] for structured grid
applications to airfoils and
wings. In these references,
successful optimizations have
been performed although the
accuracy of the derivatives has
not been explicitly
demonstrated. Table 3.4.2 Sensitivity derivatives for lift coefficient using various
Results obtained by making approximations
the assumption of a constant
eddy viscosity are listed. While the computed sensitivities show a large amount of error when
compared to finite differences, they all exhibit the correct sign. However, for derivatives associated
with horizontal changes in these same design variables, several are of incorrect sign. For example,
the finite-difference derivative obtained by perturbing point A in the horizontal direction is -0.18060
whereas the current approximations to the linearization yield 0.30328. For this same design variable,
the complete linearization yields -0.18052 which is less than 0.05 percent in error.
A similar technique that can be used to simplify the implementation is to neglect the contributions
from the turbulence model in Eq. 3.4.7. This is primarily motivated by the observation that the
costate variable associated with the turbulence model is very small and decays rapidly away from
the body. Although not shown, numerical experiments indicate that the resulting accuracy is poor
with many derivatives of incorrect sign.
3.4.7.2.3 Extent of Mesh Sensitivities
For each design variable, the evaluation of Eq. 3.4.7 requires a matrix-vector product of the costate
variables with the linearization of the residual. This also includes computation of the mesh
sensitivities for each design variable. For large numbers of design variables and mesh points, this can
potentially represent a significant expense due to the complicated linearization of the residual.
120Eric J. Nielsen, W. Kyle Anderson, “Aerodynamic Design Optimization on Unstructured Meshes Using the
Navier-Stokes Equations”, AIAA-98-4809.
70
Because nodes further away from the body are subjected to more moderate changes than those in
the immediate vicinity of the surface, it may be possible to neglect terms in Eq. 3.4.7 that are
sufficiently far from the body. This could help to reduce the cost of evaluating Eq. 3.4.7 by avoiding
the need to include terms from every mesh point in the field.
To investigate the validity of this assumption, a region around the surface of the airfoil is defined by
first "tagging" the nodes on the surface and then identifying nodes that lie within a set number of grid
layers adjacent to the surface. Outside of this region, the mesh sensitivities are set to zero to emulate
the effect of neglecting all the contributions outside of the tagged region. The sensitivity derivatives
for the lift coefficient with respect to vertical and horizontal translations of the flap are computed for
a varying number of grid layers.
In this figure, the curve labeled n/n,,, is the ratio of the number of nodes where mesh sensitivities are
employed to the total number of nodes in the mesh. It should be noted that examining a single
derivative may not be representative of the behavior of the rest of the derivatives and an accurate
computation of this derivative does not guarantee accuracy for the remaining derivatives. However,
inaccuracy of this derivative demonstrates that neglecting the full effects of the mesh sensitivities
may have an adverse effect on other derivatives as well.
As seen in the figure, the influence of the mesh sensitivities gradually decays away from the surface.
Accurate results are obtained when the number of mesh layers is greater than approximately 15]. At
this point, about half the total number of points in the mesh is included in the layers so that a factor
of 2 savings could be realized when evaluating Eq. 3.4.7. When many design variables are present,
neglecting some of the mesh sensitivities could lead to a substantial savings in computer time.
However, for the present study, the computer time required for evaluating Eq. 3.4.7 does not
dominate the overall optimization process so this strategy is not used.
3.4.7.3 Design Examples
3.4.7.3.1 Inviscid Multielement Airfoil
The objective of the first example computation is to position the elements on a multielement airfoil
to obtain a desired pressure distribution. For this case, the target pressure distribution is obtained
from analysis of a baseline configuration [28]. The individual elements are then perturbed by
translating in the x- and y-directions as well as by rotating by several degrees. The mesh used for this
test contains 3,820 nodes with 193 nodes on the surface of the main element and 129 on each flap.
The initial and target configurations are shown in Figure 3.4.7.
The initial pressure distributions over the elements are shown in Figure 3.4.6 along with the target
pressure distribution and the pressures
obtained after 15 design cycles. As seen in
the figures, the target pressure distributions
are obtained and each of the elements
returns to its desired position. This
experiment has been successfully
performed using both the compressible and
incompressible solvers, although only
results from the incompressible case are
shown.
3.4.7.3.2 Turbulent Airfoil
The objective for the next test case is to
reduce the drag on the RAE 2822 airfoil. The
initial flow conditions are a free-stream
Mach number of 0.75 , an angle of attack of
2.81∘ , and a Reynolds number of 6.2 M based
on the chord of the airfoil. These conditions
correspond to those presented earlier for Figure 3.4.8 Initial and final pressure distributions
verifying the shape sensitivity derivatives. for drag reduction on RAE 2822 airfoil.
72
Figure 3.4.9 Initial and final Mach contours for transonic airfoil optimization exercise
For this case, there are 47 active design variables. The lift coefficient is held fixed at 0.7336 and the
objective is to reduce the drag coefficient. After 10 design cycles, the drag has been reduced from
0.0263 to 0.0150 whereas after 20 design cycles, a modest improvement is further obtained, reducing
the drag coefficient to 0.0149. The initial and final pressure distributions are shown in Figure 3.4.8
and Mach number contours are shown in Figure 3.4.9. As seen in the figures, the shock wave on the
surface of the airfoil is eliminated although the curvature in the Mach contours for the final geometry
indicate that a shock may form at off-design conditions.
1.1.13.2.1 Multi-Element Airfoil
For this case, the objective is to increase the downforce for a multielement airfoil used for open-wheel
racing cars. This airfoil has been initially designed using the inviscid design techniques described in
[12]. The mesh has 15,446 nodes with 195 placed on the main element and 129 on the flap. The
spacing at the wall is 10-5 giving a y+ = 1 based on flat plate estimates. The Reynolds number is 2.4
million based on the chord of the airfoil and the angle of attack is held fixed at 12∘ which corresponds
to the operating point suggested in Ref. 12. After 25 cycles using 65 design variables, the downforce
coefficient has increased from -2.3068 to -2.4379. The initial and final pressure distributions
computed using the incompressible solver are shown in Figure 3.4.10 (a) along with the
corresponding shapes. It is seen from the figure that the redesigned main element carries more
downforce compared to the initial design, while the loading on the flap has decreased. Velocity
contours and vectors, shown in Figure 3.4.10 (b-c), indicate that the region of separated flow on
the rear of the flap has been reduced through the design process. Although not shown, comparable
cases have been run using the compressible code with similar results.
An example optimization is conducted for inviscid flow over the ONERA M6 wing [24]. The free-stream
Mach number for this case is 0.84 and the angle of attack is 3.06∘ . The mesh used for this computation
consists of 53,961 nodes and 287,962 tetrahedrons. The contours for the initial and final density
distribution on the surface of the wing are shown in Figure 3.4.11. The objective of the optimization
is to reduce the drag while maintaining a specified lift. For this design, the angle of attack is allowed
to change in addition to 10 shape design variables (4 twist, 4 camber, and 2 thickness). The twist
variables are located at the 4 outboard stations in Figure 3.4.5 and allowed to increase or decrease
by 1∘. The thickness and camber variables at positions 3 and 4 are also design variables as is the
73
camber at positions 5 and 6. Each of these is allowed to change by 2 percent of the span. After 10
design cycles, the drag has been reduced from 0.0182 to 0.0167 while the lift has been maintained.
Figure 3.4.10 Velocity contours and vectors in flap region for multielement airfoil
10 design cycles, the drag coefficient is reduced from 0.0200 to 0.0184. Density contours for the
initial and final design are shown in Figure 3.4.12. It is apparent from the increased spacing
A B
Figure 3.4.12 Initial and final density contours for inviscid wing design
between the contours that the strength of the shock at the mid chord location is somewhat weaker
for the final design which accounts for the lower drag.
A B
Figure 3.4.11 Initial and Final Density Contours for Inviscid Wing Design
the approximations yielded derivatives of acceptable accuracy and were often of incorrect sign.
Efficient surface parameterizations have been utilized in both two and three dimensions, and the
resulting codes have been integrated with an optimization package. Example optimizations have
been demonstrated in both two and three dimensions.
In order for large scale optimization to become routine, the benefits of parallel architectures should
be exploited. Although the three-dimensional flow solver has been parallelized using compiler
directives, the parallel efficiency is under 50 percent. Clearly, parallel versions of the codes will have
an immediate impact on the ability to design realistic configurations on fine meshes, and this effort
is currently underway (see e.g. Ref. 16). Another area that requires future work is the incorporation
of multipoint optimization capability for designing geometries that perform well at off-design
conditions. Further development of mesh movement strategies which enable large changes in the
geometry are also needed in three dimensions.
3.4.9 References
1 Anderson, W.K., and Bonhaus, D.L., "An Implicit Upwind Algorithm for Computing Turbulent Flows
on Unstructured Grids," Computers and Fluids, Vol. 23, No.1, 1994, pp. 1-21.
2 Anderson, W.K., and Bonhaus, D.L., "Aerodynamic Design on Unstructured Grids for Turbulent
Flows," NASA TM 112867, June 1997.
3 Anderson, W.K., Rausch, R.D., and Bonhaus, D.L., "Implicit/Multigrid Algorithms for Incompressible
Turbulent Flows on Unstructured Grids," J. Comp. Phys., Vol. 128, 1996, pp. 391-408.
4 Anderson, W.K., and Venkatakrishnan, V., "Aerodynamic Design Optimization on Unstructured Grids
with a Continuous Adjoint Formulation," AIAA Paper No. 97-0643, 1997.
5 Bonhaus, D.L., "An Upwind Multigrid Method for Solving Viscous Flows on Unstructured Triangular
Meshes," M.S. Thesis, George Washington University, 1993.
6 Chakravarthy, S.R., and Osher, S., "Numerical Experiments with the Osher Upwind Scheme for the
Euler Equations," AIM Journal, Vol. 21, No. 9, 1983, pp. 1241-1248.
7 Cook P.. M. cDonald, M.. and Firmin. M.. "Airfoil RAE 2822-Pressure Distributions and Boundary
Layer Wake Measurement.", AGARD AR-138, Paper A6, 1979.
8 Davidon, W.C., "Vanable Metnc Method for Minimization," AEC Research and Development Report,
ANL-5990, 1959.
9 Elliott, J., and Peraire, J., "Aerodynamic Design Using Unstructured Meshes," AIAA Paper, 1996.
10 OElliott, J., and Peraire, J., "Aerodynamic Optimization on Unstructured Meshes with Viscous
Effects," AIAA Paper No. 97-1849, 1997.
11 Fletcher, R., and Powell, M.J.D., "A Rapidly Convergent Descent Method for Minimization," The
Computer Journal, Vol. 6, July 1963, pp. 163-1 68.
12 Gopalarathnam, A,, Selig, M.S., and Hsu, F., "Design of High-Lift Airfoils for Low-Aspect Ratio Wings
with Endplates," AIAA Paper No. 97-2232, 1997.
13 Hackbusch, W., Iterative Solution of Large Sparse Systems of Equations, Springer-Verlag, NY, 1994.
14 Hou, G. J.-W., Taylor, A.C., and Korivi, V.M., "Discrete Shape Sensitivity Equations for Aerodynamic
Problems," Znt. J. for Numerical Methods in Engineering, Vol. 37, 1994, pp. 2251-2266.
15 Jameson, A,, Pierce, N.A., and Martinelli, L., "Optimum Aerodynamic Design Using the Navier-Stokes
Equations," AIAA Paper No. 97-0101, 1997.
16 Kaushik, D.K., Keyes, D.E., and Smith, B.F., "On the Interaction of Architecture and Algorithm in the
Domain-Based Parallelization of an Unstructured Grid Incompressible Flow Code," in Proceedings of
the 10th Intl. Conf. on Domain Decomposition Methods, J. Mandel et al., eds., AMs, pp. 311-319.
17 Marcum, D.L., "Generation of Unstructured Grids for Viscous Flow Applications," AIAA Paper No.
95-0212, 1995.
18 Mohammadi, B., "Optimal Shape Design, Reverse Mode of Automatic Differentiation and
Turbulence." AIAA Paper No. 97-0099, 1997.
76
19 Newman, J.C., and Taylor, A.C., "Three-Dimensional Aerodynamic Shape Sensitivity Analysis and
Design Optimization Using the Euler Equations on Unstructured Grids," AIAA Paper No. 96-2464, 1996.
20 Pirzadeh, S., "Viscous Unstructured Three-Dimensional Grids by the Advancing-Layers Method,"
AIAA 94-0417, 1994.
21 Roe, P.L., "Approximate Riemann Solvers, Parameter Vectors, and Difference Schemes," J. Comp.
Phys., Vol. 43, No. 2, 1981, pp. 357-372.
22 sad, Y., and Schultz, M.H., "GMRES: A Generalized Minimal Residual Algorithm for Solving
Nonsymmetric Linear Systems," SZM J. Sci. Stat. Comp., Vol. 7, 1986, pp. 856-869.
23 Samareh, J., "Geometry Modeling and Grid Generation for Design and Optimization,"
ICASE/LaRC/NSF/ARO Workshop on Computational Aero sciences in the 21st Century," 1998.
24 Schmitt, V., and Charpin, F., "Pressure Distributions on the ONERAM6 Wing at Transonic Mach
Numbers," Experimental Data Base for Computer Program Assessment, AGARD-AR-138, May 1979,
pp. B1-1-B1-44.
25 Sherman, L.L., Taylor, A.C., Green, L.L., Newman, P.A., Hou, G.J. W., and Korivi, V.M., "First- and
Second-Order Aerodynamic Sensitivity Derivatives via Automatic Differentiation with Incremental
Iterative Methods," AIAA Paper No. 94-4262, 1994.
26 Soemanvoto, B., "Multipoint Aerodynamic Design by Optimization," Ph.D. Thesis, Delft University
of Technology, 1996.
27 Spalart, P.R., and Allmaras, S.R., "A One-Equation Turbulence Model for Aerodynamic Flows," AIAA
Paper No. 92-0439, 1991.
28 Suddhoo, A,, and Hall, I.M., "Test Cases for the Plane Potential Flow Past Multi-Element Airfoils,"
Aeronautical Journal, December 1985.
29 Van Leer, B., "Flux Vector Splitting for the Euler Equations," Lecture Notes in Physics, Vol. 170,
1982, pp. 501-512.
30 Venkatalaishnan, V., and Mavriplis, D.J., "Implicit Method for the Computation of Unsteady Flows
on Unstructured Grids," J. Comp. Phys., Vol. 127, 1996, pp. 380-397.
31 Williams, B.R., "An Exact Test Case for the Plane Potential Flow About Two Adjacent Lifting
Aerofoils," R. & M. No. 3717, British Aeronautical Research Counci, 1973.
32 Wrenn, G.A., "An Indirect Method for Numerical Optimization Using the Kresselmeier-Steinhauser
Function," NASA CR 4220, March 1989.
77
3.5 Case Study 5 – Single & Multipoint Optimization for XRF-1 Wing Shape
3.5.1 Summary
In this chapter, the wing of the XRF-1 wing-body configuration, representative of modern civil
transport aircraft, is optimized in viscous and transonic conditions with the adjoint method. A single
point optimization and a multi-lift-multi-Mach optimization are performed. The polar of the
resulting designs are compared with the ones of the XRF-1 baseline, and the single point optimum
design. The test case aims at showing the applicability of the robust optimization method proposed
in the present thesis to real-life cases. The effect of the noise on the gradients of the objective
functions on both the multipoint optimization and the Gradient Span Analysis (GSA) algorithm are
explained and solutions are proposed. Finally, the XRF-1 is optimized taking into account for
aeroelasticity, using coupled aero-elastic CFD simulations. The robust rigid multipoint optimum is
taken as starting point for a twist derivative-free multipoint optimization.
the drag, but it would increase the mass of the internal structure. The compromise is
multidisciplinary. Because no structural sizing is involved in the pure aerodynamic optimization
process, the two structural beams are built in the wing CAD model, see Figure 3.5.1 in red, and their
thickness are geometrical constraints. The wing sweep is kept constant, also for its impact on the
structure that is not simulated in our case. Similarly, the planform of the wing is fixed, the wingspan,
sweep and reference surface are frozen. Their choice require the multidisciplinary compromise
between high-speed fuel burn, low-speed performance such as take-off field length, weight balance,
max take-off weight, operating cost etc.; which are not addressed in the present project. The whole
model, displayed in Figure 3.5.1, contains 91 design variables.
121 M = million
80
and 1.7% on the lift, while the 74 M cells mesh divides this
error by two for a more than three times higher CPU cost.
The 13.5 M cells mesh provides a relative precision of 4.3%
on the drag, which is the order of magnitude of the gains
expected by the optimizations. In addition, Table 3.5.2
shows that the spurious drag quantifying non-physical
dissipation, is of 14.3 D.C.122 on the 13.5M cells mesh, which
is also the order of magnitude of the expected gains by the
optimization. On the other hand, it is of 3.4 D.C., on the 22M
cells mesh. One could still perform a numerical
optimization using the 13.5 cells.
mesh, but it has to be kept in mind that the aim of Table 3.5.1 XRF-1 ; M = 0.83 ; AoA =
2.607o ; Re = 49.9 106. Far-field
optimization is to modify a shape in order to improve its in-
spurious drag on a mesh hierarchy
flight performance. Consequently, the observed from Nguyen-Dinh’s PhD thesis [98]
performance variations have to be related to physical
phenomena modifications and
not to a numerical behavior of
the model. If the measured
performance variations due to
optimization are of a similar
order of magnitude as the
precision of the numerical
model, then it can be difficult to
justify. So we consider that the
13.5M cells mesh is not the
best-suited to the Table 3.5.2 XRF-1 near field analysis; M = 0.83 ; AoA = 2.607o ; Re =
demonstration. 49.9 106. CL, CLp, CD, CDp and CDf on a mesh hierarchy from Nguyen-
In addition to mesh Dinh’s phd thesis [98]
convergence, the residuals
convergence is a critical point in optimization. The norms of the residuals quantifies how much the
Figure 3.5.3 Direct and Adjoint Calculations Convergence History on the XRF-1 baseline at the 6
operating conditions of the Multipoint Optimization
flow state equations are satisfied, so having them as close to zero as possible is required to link the
optimization gains to a real performance improvement, and not a violation of the state equations due
to modification of the numerical model during the optimization. It has been shown that the adjoint
to the flow supposes that the residuals of the state equation are null. If it is not the case, then the
obtained gradients are inexact, which is known to slow down the optimization process and degrade
the results [33]. Direct and adjoint calculations convergence are displayed in Figure 3.5.3, on the
XRF-1 baseline. The residuals L2-norm of the direct calculations are reduced to 10−6 for all the flight
conditions.
The residuals of the adjoint calculations are plotted against LU-SSOR iterations, each of them using
10 LU relaxation iterations. The optimal number of LU relaxations has been chosen to minimize the
CPU cost for a given convergence level. The levels of convergence of the adjoint calculations are
between 0.8 and 2.1 orders of magnitude. The number of iterations that provides such a level of
convergence has been chosen by comparison of the obtained gradients and the level of noise due to
the linearization hypotheses: frozen turbulence, thin layer approximation. Besides, we noted that the
L2-norm being the sum of all square residuals, the quantity can be greatly affected by a poor
convergence in some areas of the shape that have a low influence on optimization, far from
parametrized shape areas.
The situation for adjoint is quite different from a direct calculation, a high residual in the RANS
equation means that the flow is ill-resolved, so that the solution is non-physical. In the adjoint
approach, the objective is to obtain sufficiently precise gradients to achieve optimizations, and
gradient-based optimizers are robust to noise [44]. To this aim, the adjoint field has to be adequately
resolved close to the parametrized surfaces. Ideally, of course, the adjoint problem should be solved
up to machine precision. In practice however, the algorithms should be robust to noise, since
industrial applications with complex geometrical details often lead to non-ideal meshes and poor
convergence. We will see that this noise also has impacts on the multipoint optimization strategy,
detailed in the following sections.
3.5.4 Single Point Optimization Strategy
A single point optimization of the configuration is run at Mach = 0.83, Reynolds = 49.9 M and CL =
0.531. This enables to check the behavior of the optimization process, and to compare the results
with multipoint optimizations. The L-BFGS-B code from Zhu et al. [145] is used as optimizer. The lift
constraint CL − CL0 = 0 is handled in the flow solver by a target lift approach. Newton iterations with
finite differences on the function CL(AoA) are performed during the convergence of the flow and the
cost is about 20% more than a fixed AoA calculation. A Lagrangian is derived to compute the
derivative of the drag at constant lift in Eq. 3.5.1 and Eq. 3.5.2, the derivatives being calculated by
discrete adjoints of CL and CD:
dCD (χ , AoA)
dχ
λCL =
dCL (χ , AoA)
dχ
Eq. 3.5.3
and therefore the gradient of the Lagrangian is:
dCD (χ , AoA)
dL(CD , CL − CL0 )(χ) dCD (χ , AoA) dχ dCL (χ , AoA)
= +
dχ dχ dCL (χ , AoA) dχ
dχ
Eq. 3.5.4
This Lagrangian gradient is provided to the L-BFGS-B algorithm. Figure 3.5.4 shows that the target
lift approach succeeds in maintaining the lift during the optimization by automatic adjustment of the
angle of attack. The overall drag reduction of 11% is mainly due to the shock smoothing, which also
leads to a viscous pressure
drag reduction. Figure
3.5.5 also shows smooth
pressure coefficient iso-
lines and the global view of
the configuration.
The direct calculation
convergence levels are
satisfactory, 100% of them
being higher that 5 orders
of magnitude. 200 adjoint
calculations have a
convergence level higher
or equal to 1 order of
magnitude, and 60 of them
have a convergence level
higher than 2 orders of
magnitude. However 25
Figure 3.5.4 XRF-1 single point optimization convergence history
adjoint calculations did
not converge adequately,
which represents 8% of them and are mostly at the second iteration of the algorithm, which
represents 12 calculations. This second step was rejected in the line-search phase of the L-BFGS-B
algorithm, so the gradient was not taken into account in the BFGS approximation. The second
iteration of the algorithm is always a special case, since the Hessian approximation is the identity
matrix, meaning that the full steepest descent step is used, often leading to too important design
variables modification and then a drag increase, which is observed in Figure 3.5.4. A solution for
that is to modify the L-BFGS-B algorithm to decrease the first step length, modification that we
accomplished during the present project. For further information regarding the optimization details,
please refer to Gallard, F. (2014)123.
123 Gallard, François. Aircraft shape optimization for mission performance. PhD, Institute National Polytechnique
de Toulouse, 2014
83
dCD (χ , AoA)
dχ
∆λTCL =∆
dCL (χ , AoA)
dχ
Eq. 3.5.8
By derivation of Eq. 3.5.8 we obtain, at first order
85
dC (χ , AoA)
dCD (χ , AoA) ∆ D
dχ
∆rel =‖ ‖ and
dχ dCD (χ , AoA)
dχ
dC (χ , AoA)
dCL (χ , AoA) ∆ L
dχ
∆rel =‖ ‖
dχ dCL (χ , AoA)
dχ
Eq. 3.5.11
to finally obtain an upper bound for the Lagrange multiplier noise propagation in a target lift
formulation.
dCD (χ , AoA)
∆ dCD (χ , AoA) dCL (χ , AoA)
dχ
‖∆λTCL ‖ ≤ ‖ ‖ (∆rel + ∆rel )
dCL (χ , AoA) dχ dχ
dχ
Eq. 3.5.12
And the relative error ΔrelλTCL = ΔλTCL/λTCL
∇𝑟𝑒𝑙 λ𝑐𝑠𝑡 = ∇𝑟𝑒𝑙 ∇χi 〈CD (χi ) , ∇χ𝑖 CL (χi )〉 + 2 ∇𝑟𝑒𝑙 ‖∇χ𝑖 CL (χi )‖
Eq. 3.5.16
Eq. 3.5.13 shows that in the target lift approach, the relative error on the Lagrangian is the sum of
the relative errors of drag and lift derivatives with respect to the angle of attack only, while Eq.
3.5.16 shows a much more complex expression involving the noise of the derivatives with respect
to all the design variables in the case where explicit constraints are used. The errors on the gradients
are evaluated by second order finite differences in the aim of comparing the two approaches. Finite
differences are expensive, so the 2D RAE2822 case is used, at Mach = 0.74 and CL = 0.782 Re = 6.5M.
The average errors on the gradient are of 0.36% for the lift, and 1.9% for the drag. Numerical
evaluations of Eq. 3.5.13 and Eq. 3.5.16 gave ΔrelλTCL = 15.2 and ΔrelλCST = 43.9. It means that the
Lagrangian of target lift approach amplifies the noise on the gradients 2.9 times less than the
Lagrangian of an explicit constraints formulation. In this 2D case, there are only 16 design variables,
including the angle of attack, but on the XRF-1 case, the parametrizations have between 100 and 300
design variables. It is interesting to note that ΔrelλTCL does not depend on the design variables number,
while ΔrelλCST does. It is possible to extrapolate the gradient errors estimation from 2D to the 3D case
by copying the gradients to reach the targeted number of design variables, since the wing
parametrization uses copies of airfoil parametrization along wingspan. Of course, this is not accurate
to estimate the XRF-1 true gradient error since the adjoint convergence is not the same and the flows
are different, but we aim here at obtaining tendencies for the variations of ΔrelλCST with the number
of design variables. For 96 design variables, we obtain ΔrelλCST = 236.3, and ΔrelλCST = 736. for 304
design variables.
The SLSQP algorithm used to solve the drag minimization under explicit lift constraints does not
directly minimize the Lagrangian of the optimization problem. Quadratic Programming minimizes a
second order model of the function under feasibility constraints. However, the second order terms
are modeled by the BFGS approximation computed using the gradients of the Lagrangian. Similarly,
the L-BFGS-B algorithm, used to minimize the drag at constant lift ensured with the target lift
approach also uses the BFGS approximation of the Lagrangian of the drag at constant lift. Therefore,
the previous comparisons of the gradients noise propagation on the Lagrangian is using target lift
than explicit lift constraints.
124 Gallard, François. Aircraft shape optimization for mission performance. PhD, Institute National Polytechnique
de Toulouse, 2014
87
125 Gallard, François. Aircraft shape optimization for mission performance. PhD, Institute National Polytechnique
de Toulouse, 2014
88
number (M = 0.86, CL = 0.5). The only condition affected by a performance degradation is the M =
0.83, CL = 0.48 operating condition, with an increase of 0.4%. In average on the 6 operating
conditions, the drag gain is of
0.38% compared to the rigid
optimum.
Multiple conclusions can be
drawn from this experiment.
First, aeroelasticity
introduces a new degree of
freedom in the multipoint
optimization. This degree of
freedom can be exploited to
gain in performance, and the
optimal rigid design is not
optimal any more when
aeroelasticity is taken into
account, even if it enabled to
design the wing sections.
Second, the weights used for
the rigid optimization,
computed by an estimation of
the utopia point with a rigid
structure, in the aim of
gaining drag at all the
operating conditions, are not
fully adapted to the problem
any more, since the drag has
increased at one operating
condition. This means that
aeroelasticity is changing the
trade-offs of gains between
the operating conditions. The
weight calculations for
multipoint optimizations
must then take aeroelasticity
into account to well balance
the gains between operating
conditions. Figure 3.5.10
summarizes the process used
for the XRF-1 optimization,
leading to a wing optimized Figure 3.5.10 Steps of the XRF-1 multipoint aeroelastic optimization
for a multi-Mach and multi-
Lift operating conditions, and
accounting for aeroelastic effects. These steps are automated and integrated in the workflow
manager, which shows that the methodology can be scaled to an industrial level.
3.5.8 References
[1] Global Market Forecast, Future Journeys 2013. Technical report, Airbus, 2013. URL http:
//www.airbus.com/company/market/forecast/.
[2] John D Anderson. Aircraft performance & design. 1999.
91
[3] Emir Mahmut Bahsi. Dynamic Workflow Management for Large Scale Scientific Applications. PhD
thesis, Louisiana State University, 2008.
[4] S Balay, J Brown, K Buschelman, V Eijkhout, W Gropp, D Kaushik, M Knepley, L Curfman McInnes,
B Smith, and H Zhang. Petsc user’s manual revision 3.4. 2013.
[5] Arnaud Barthet. Amelioration de la prevision des coefficients aerodynamiques autour de
configurations portantes par methode adjointe. 2007.
[6] Richard Bellman. Adaptive control processes: a guided tour, volume 4. Princeton university press
Princeton, 1961.
[7] Hans-Georg Beyer and Bernhard Sendhoff. Robust optimization – A comprehensive survey.
Computer Methods in Applied Mechanics and Engineering, 196(33–34):3190–3218, 2007. ISSN 0045-
7825. doi: 10.1016/j.cma.2007.03.003.
URL http://www.sciencedirect.com/ science/article/pii/S0045782507001259.
[8] Leon Bottou. Online learning and stochastic approximations. On-line learning in neural networks,
17:9, 1998.
[9] Leon Bottou and Olivier Bousquet. The tradeoffs of large-scale learning. Optimization for Machine
Learning, page 351, 2011.
[10] Stephen P Boyd and Lieven Vandenberghe. Convex optimization. Cambridge university press,
2004.
[11] Thierry Braconnier, Marc Ferrier, Jean-Christophe Jouhaud, Marc Montagnac, and Pierre Sagaut.
Towards an adaptive POD/SVD surrogate model for aeronautic design. Computers & Fluids, 40:195–
209, January 2011. doi: 10.1016/j.compfluid.2010.09.002.
[12] Howard P. Buckley and David W. Zingg. Approach to aerodynamic design through numerical
optimization. AIAA Journal, 51(8):1972–1981, 2014/06/25 2013. doi: 10.2514/1.J052268. URL
http://dx.doi.org/10.2514/1.J052268.
[13] Howard P. Buckley, Beckett Y. Zhou, and David W. Zingg. Airfoil Optimization Using Practical
Aerodynamic Design Requirements. Journal of Aircraft, 47(5):1707–1719, September- October 2010.
doi: 10.2514/1.C000256.
[14] Howard P Buckley, Beckett Y Zhou, and David W Zingg. Airfoil optimization using practical
aerodynamic design requirements. Journal of Aircraft, 47(5):1707–1719, 2010.
[15] L. Cambier, M. Gazaix, S. Heib, S. Plot, M. Poinot, J.-P. Veuillot, J.-F. Boussuge, and M. Montagnac.
CFD Platforms and Coupling : An Overview of the Multi-Purpose elsA Flow Solver. Aerospace Lab, 2,
March 2011.
[16] L. Cambier, S. Heib, and S. Plot. The Onera ElsA CFD Software: Input from Research and Feedback
from Industry. Mechanics & Industry, 14(03):159–174, January 2013. doi: 10.1051/meca/2013056.
[17] Richard L. Campbell. Efficient Viscous Design of Realistic Aircraft Configurations. In AIAA Paper,
number 98-2539, 1998.
[18] G Carrier. Single and multi-point aerodynamic optimizations of a supersonic transport aircraft
wing using optimization strategies involving adjoint method and genetic algorithm. Proceedings of
ERCOFTAC Workshop" Design optimization: methods and applications", Las Palmas, 2006.
[19] Oleg Chernukhin and DavidWZingg. Multimodality and global optimization in aerodynamic
design. AIAA Journal, 51(6):1342–1354, 2013.
[20] AC Chiang and Kevin Wainwright. Fundamental methods of mathematical economics. McGraw-
Hill, New York, 2005.
[21] Haecheon Choi and Parviz Moin. Grid-point requirements for large eddy simulation: Chapman’s
estimates revisited. Physics of Fluids, 24:011702, 2012.
[22] Hyoung-Seog Chung and Juan J Alonso. Using gradients to construct cokriging approximation
models for high-dimensional design optimization problems. AIAA paper, 317:14–17, 2002.
[23] A.R. Collar. The expanding domain of aeroelasticity. J. Royal Aero Society, 1:613–634, 1946.
[24] I. Das and J. E. Dennis. A closer look at drawbacks of minimizing weighted sums of objectives for
Pareto set generation in multicriteria optimization problems. Structural and Multidisciplinary
92
URL http://doi.acm.org/10.1145/962437.962439.
[43] Serge Gratton, Philippe L Toint, and Anke Troltzsch. An active-set trust-region method for
derivative-free nonlinear bound-constrained optimization. Optimization Methods and Software,
26(4-5):873–894, 2011.
[44] Serge Gratton, Philippe L Toint, and Anke Troltzsch. How much gradient noise does a gradient-
based linesearch method tolerate? Technical report, CERFACS, 2011.
[45] Justin Gray, Kenneth T Moore, and Bret A Naylor. Openmdao: an open source framework for
multidisciplinary analysis and optimization. In Proceedings of the 13th AIAA/ISSMO Multidisciplinary
Analysis Optimization Conference, Fort Worth, TX, USA. MIT Press, 2010.
[46] S Gurol, AT Weaver, AM Moore, A Piacentini, HG Arango, and S Gratton. B-preconditioned
minimization algorithms for variational data assimilation with the dual formulation. Quarterly
Journal of the Royal Meteorological Society, 2013.
[47] M. Harbeck and A. Jameson. Exploring the limits of shock-free transonic airfoil design. In AIAA
43rd aerospace sciences meeting and exhibition, 2005.
[48] Raymond M. Hicks and Garret N. Vanderplaats. Application of Numerical Optimization to the
Design of Supercritical Airfoils without Drag Creep. In Business Aircraft Meeting, number Paper
770440, Wichita, Kansas, March 29 - April 1 1977. Society of Automotive Engineers. URL
http://papers.sae.org/770440.
[49] Chen-Hung Huang, Jessica Galuski, and Christina L Bloebaum. Multi-objective pareto concurrent
subspace optimization for multidisciplinary design. AIAA Journal, 45(8):1894–1906, 2007.
[50] Luc Huyse, Sharon L Padula, R Michael Lewis, and Wu Li. Probabilistic approach to freeform
airfoil shape optimization under uncertainty. AIAA Journal, 40(9):1764–1772, 2002.
[51] John T Hwang, Dae Young Lee, James W. Cutler, and Joaquim R. R. A. Martins. Largescale
multidisciplinary optimization of a small satellite’s design and operation. Journal of Spacecraft and
Rockets, 2013. (Accepted Nov. 17, 2013).
[52] Laurenceau J. Surfaces de réponse par krigeage pour l’optimisation de formes aérodynamiques -
TH/CFD/08/62. PhD thesis, Institut National Polytechnique de Toulouse, France, 2008. URL
http://www.cerfacs.fr/~cfdbib/repository/TH_CFD_08_62.pdf. phd.
[53] A. Jameson. Steady State Solutions of the Euler Equations for Transonic Flow by a Multigrid
Method, chapter Advances in Scientific Computing, pages 37–70. Academic press, 1982.
[54] A. Jameson, L. Martinelli, and N. A. Pierce. Optimum Aerodynamic Design Using the Navier-
Stokes Equations. Theoretical and Computational Fluid Dynamics, 10(1-4):213–237, 1998. doi:
10.1007/s001620050060.
[55] Antony Jameson. Aerodynamic design via control theory. J. Sci. Comput., 3(3):233–260, 1988.
[56] Antony Jameson, Kasidit Leoviriyakit, and Sriram Shankaran. Multi-point Aero-Structural
Optimization of Wings Including Planform Variations. In 45th Aerospace Sciences Meeting and Exhibit,
number AIAA Paper 2007-764, Reno, Nevada, January 8-11 2007.
[57] Shinkyu Jeong, Mitsuhiro Murayama, and Kazuomi Yamamoto. Efficient optimization design
method using kriging model. Journal of Aircraft, 42(2):413–420, 2005.
[58] Forrester T. Johnson, Edward N Tinoco, and N Jong Yu. Thirty years of development and
application of CFD at Boeing commercial airplanes, Seattle. Computers & Fluids, 34(10): 1115–1151,
2005.
[59] S Johnson. The nlopt nonlinear-optimization package (2011). URL http://ab-initio. mit.edu/nlopt.
[60] Eric Jones, Travis Oliphant, and Pearu Peterson. Scipy: Open source scientific tools for python.
http://www. scipy. org/, 2001.
[61] Eric Jones, Travis Oliphant, Pearu Peterson, et al. SciPy: Open source scientific tools for Python,
2001–. URL http://www.scipy.org/.
[62] William Karush. Minima of functions of several variables with inequalities as side constraints. PhD
thesis, Masterś thesis, Dept. of Mathematics, Univ. of Chicago, 1939.
[63] Gaetan K. W. Kenway and Joaquim R. R. A. Martins. Multi-point high-fidelity aero structural
94
[85] R Timothy Marler and Jasbir S Arora. The weighted sum method for multi-objective
optimization: new insights. Structural and multidisciplinary optimization, 41(6):853–862, 2010.
[86] Joaquim R. R. A. Martins and Andrew B. Lambe. Multidisciplinary design optimization: A survey
of architectures. AIAA Journal, 51:2049–2075, 2013. doi: 10.2514/1.J051895.
[87] Michael D McKay, Richard J Beckman, and William J Conover. Comparison of three methods for
selecting values of input variables in the analysis of output from a computer code. Technometrics,
21(2):239–245, 1979.
[88] A. Messac, A. Ismail-Yahaya, and C.A. Mattson. The normalized normal constraint method for
generating the Pareto frontier. Structural and Multidisciplinary Optimization, 25:86–98, 2003. ISSN
1615-147X. doi: 10.1007/s00158-002-0276-1.
URL http://dx.doi.org/10. 1007/s00158-002-0276-1.
[89] B. Mohammadi and O. Pironneau. Applied Shape Optimization for Fluids. Oxford Science
Publications, 2001.
[90] A. M. Morris, C. B. Allen, and T. C. S. Rendall. CFD-based optimization of aerofoils using radial
basis functions for domain element parameterization and mesh deformation. International Journal
for Numerical Methods in Fluids, 58(8):827–860, 2008. ISSN 1097-0363. doi: 10.1002/fld.1769. URL
http://dx.doi.org/10.1002/fld.1769.
[91] Daniel Mueller-Gritschneder, Helmut E. Graeb, and Ulf Schlichtmann. A Successive Approach to
Compute the Bounded Pareto Front of Practical Multiobjective Optimization Problems. SIAM Journal
on Optimization, 20(2):915–934, 2009.
[92] John M Mulvey, Robert J Vanderbei, and Stavros A Zenios. Robust optimization of largescale
systems. Operations research, 43(2):264–281, 1995.
[93] D. My-Ha, K.M. Lim, B.C. Khoo, and K. Willcox. Real-time optimization using proper orthogonal
decomposition: Free surface shape prediction due to underwater bubble dynamics. Computers &
Fluids, 36(3):499–512, 2007. ISSN 0045-7930. doi: 10.1016/j.compfluid.2006. 01.016.
[94] Uwe Naumann. The Art of Differentiating Computer Programs, volume 24. SIAM, 2012.
[95] M. Nemec, M. J Aftosmis, and T. H. Pulliam. Cad-based aerodynamic design of complex
configurations using a cartesian method. AIAA Paper, 113, 2004.
[96] Marian Nemec and David W Zingg. Newton-krylov algorithm for aerodynamic design using the
navier-stokes equations. AIAA Journal, 40(6):1146–1154, 2002.
[97] Marian Nemec, David W Zingg, and Thomas H Pulliam. Multipoint and multi-objective
aerodynamic shape optimization. AIAA Journal, 42(6):1057–1065, 2004.
[98] M Nguyen-Dinh. Qualification des simulations numériques par adaptation anisotropique de
maillages. PhD thesis, Universite de Nice-Sophia Antipolis, 2004.
[99] E. J. Nielsen and W. K. Anderson. Aerodynamic Design Optimization on Unstructured Meshes
Using the Navier-Stokes Equations. AIAA Journal, 37(11):1411–1419, 1999. doi: 10.2514/2.640.
[100] Emma Nygren, Kjell Aleklett, and Mikael Hook. Aviation fuel and future oil production
scenarios. Energy Policy, 37(10):4003–4010, 2009.
[101] Travis E. Oliphant. Python for scientific computing. Computing in Science & Engineering,
9(3):10–20, 2007. doi: http://dx.doi.org/10.1109/MCSE.2007.58. URL http://scitation.
aip.org/content/aip/journal/cise/9/3/10.1109/MCSE.2007.58.
[102] Valerie Pascual and Laurent Hascoet. Extension of tapenade toward fortran 95. In Automatic
Differentiation: Applications, Theory, and Implementations, pages 171–179. Springer, 2006.
[103] Sergey Peigin and Boris Epstein. Robust optimization of 2d airfoils driven by full navier-stokes
computations. Computers and Fluids, 33:1175–1200, 2004.
[104] Ruben E Perez, Peter W Jansen, and Joaquim RRA Martins. pyopt: a python-based object
oriented framework for nonlinear constrained optimization. Structural and Multidisciplinary
Optimization, 45(1):101–118, 2012.
[105] Jacques Peter, Maxime Nguyen-Dinh, and Pierre Trontin. Goal oriented mesh adaptation using
total derivative of aerodynamic functions with respect to mesh coordinates aĂŞ with applications to
96
euler flows. Computers & Fluids, 66(0):194 – 214, 2012. ISSN 0045-7930. doi:
http://dx.doi.org/10.1016/j.compfluid.2012.06.016.
URL http://www.sciencedirect.com/ science/article/pii/S0045793012002393.
[106] Le. A Piegl and W. Tiller. The NURBS book. Springer, 1997.
[107] Xavier Pinel and Marc Montagnac. Block krylov methods to solve adjoint problems in
aerodynamic design optimization. AIAA Journal, 51(9):2183–2191, 2013.
[108] Michael JD Powell. A direct search optimization method that models the objective and
constraint functions by linear interpolation. In Advances in optimization and numerical analysis, pages
51–67. Springer, 1994.
[109] L. Prandtl. Theory of lifting surfaces. Technical report, NACA, 1920. URL http://naca.
larc.nasa.gov/reports/1920/naca-tn-9/.
[110] Ludwig Prandtl. Tragflugeltheorie. i. mitteilung. Nachrichten von der Gesellschaft der
Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse, 1918:451–477, 1918. Cited in p. 25
[111] G. Puigt, M. Gazaix, M. Montagnac, M.-C. Le Pape, M. de la Llave Plata, M. Marmignon, J.-F.
Boussuge, and V. Couaillier. Development of a new hybrid compressible solver inside the CFD elsA
software. In 20th AIAA Computational Fluid Dynamics Conference, number AIAA–2011–3379,
Honolulu (HI), USA, June 27-30 2011.
[112] Hicks R. M., Murman E. M., and Vanderplaats G. N. An assessment of airfoil design by numerical
optimization. Technical Report TMX-3092, NASA Ames Research Center, Moffett Field, California,
1974.
[113] J. Reuther, J.J. Alonso, Joaquim R. R. A. Martins, and S. C. Smith. A coupled aero-structural
optimization method for complete aircraft configurations. In Proceedings of the 37th AIAA Aerospace
Sciences Meeting and Exhibit, Reno, NV, January 1999. AIAA 99-0187.
[114] James J. Reuther, Antony Jameson, Juan J. Alonso, Mark J. Rimlinger, and David Saunders.
Constrained Multipoint Aerodynamic Shape Optimization Using an Adjoint Formulation and Parallel
Computers, Part 1. Journal of Aircraft, 36(1):51–60, January-February 1999. doi: 10.2514/2.2413.
[115] James J. Reuther, Antony Jameson, Juan J. Alonso, Mark J. Rimlinger, and David Saunders.
Constrained Multipoint Aerodynamic Shape Optimization Using an Adjoint Formulation and Parallel
Computers, Part 2. Journal of Aircraft, 36(1):61–74, January-February 1999. doi: 10.2514/2.2414.
[116] P. L. Roe. Approximate Riemann Solvers, Parameter Vectors and Difference Schemes. Journal
of Computational Physics, 43(2):357–372, October 1981. doi: 10.1016/0021-9991(81)90128-5.
[117] Youcef Saad and Martin H Schultz. Gmres: A generalized minimal residual algorithm for solving
nonsymmetric linear systems. SIAM Journal on scientific and statistical computing, 7(3):856–869,
1986.
[118] Jerome Sacks, William J Welch, Toby J Mitchell, and Henry P Wynn. Design and analysis of
computer experiments. Statistical science, 4(4):409–423, 1989.
[119] Jamshid A Samareh. Survey of shape parameterization techniques for high-fidelity
multidisciplinary shape optimization. AIAA Journal, 39(5):877–884, 2001. Cited in page 84
[120] Jamshid A Samareh. Aerodynamic shape optimization based on free-form deformation. AIAA
paper, 4630:2004, 2004.
[121] S Shahpar. Challenges to overcome for routine usage of automatic optimization in the
propulsion industry. Aeronautical Journal, 115(1172):615–625, 2011. Cited in pages 8, 35, and 82
[122] Donald Shepard. A two-dimensional interpolation function for irregularly-spaced data. In
Proceedings of the 1968 23rd ACM national conference, pages 517–524. ACM, 1968. Cited in page 88
[123] Jaroslaw Sobieszczanski-Sobieski and Raphael T Haftka. Multidisciplinary aerospace design
optimization: survey of recent developments. Structural optimization, 14(1):1–23, 1997.
[124] P. R. Spalart and S. R. Allmaras. A One-Equation Turbulence Transport Model for aerodynamic
flows. In 30th AIAA Aerospace Sciences Meeting and Exhibit, Reno, Nevada, January 1992. AIAA Paper
92-0439.
[125] E. Stanewsky. Adaptive wing and flow control technology. Progress in Aerospace Sciences,
97
37(7):583–667, 2001.
[126] Ralph E. Steuer. Multiple criteria optimization: theory, computation, and application. Krieger
Malabar, 1989.
[127] Krasimira Stoilova and Todor Stoilov. Comparison of workflow software products. In
International Conference on Computer Systems and Technologies-CompSysTech, pages 15–16, 2006.
[128] Per Stromholm. Fermat’s methods of maxima and minima and of tangents. a reconstruction.
Archive for History of Exact Sciences, 5(1):47–69, 1968. ISSN 0003-9519. doi: 10.1007/ BF00328112.
URL http://dx.doi.org/10.1007/BF00328112.
[129] Marius Swoboda, Andre Huppertz, Akin Keskin, Dierk Otto, and Dieter Bestle. Multidisciplinary
compressor blading design process using automation and multi-objective optimization. In 25th
Congress of International Council of the Aeronautical Sciences. Paper number ICAS2006-5.6 S, 2006.
[130] Knut Sydsaeter and Peter J Hammond. Essential mathematics for economic analysis. Pearson
Education, 2008.
[131] Genichi Taguchi. Introduction to quality engineering: designing quality into products and
processes. 1986.
[132] Bui T. Thanh, M. Damodaran, and K. Willcox. Proper Orthogonal Decomposition Extensions for
Parametric Applications in Transonic Aerodynamics (AIAA Paper 2003-4213). In Proceedings of the
15th AIAA Computational Fluid Dynamics Conference, 2003.
[133] G. D. van Albada, B. van Leer, and W. W. Roberts. A Comparative Study of Computational
Methods in Cosmic Gas Dynamics. Astronomy and Astrophysics, 108:76–84, 1982.
[134] B. van Leer. Towards the ultimate conservative difference scheme. V. A second-order sequel
to Godunov’s method. Journal of Computational Physics, 32(1):101–136, 1979. doi: 10.1016/0021-
9991(79)90145-1.
[135] GN Vanderplaats. A robust feasible directions algorithm for design synthesis. In Proc., 24th
AIAA/ASME/ASCE/AHS Structures, Structural Dynamics and Materials Conf, 1983.
[136] R.V. Gamkrelidze. L.S. Pontryagin V.G. Boltyanskii. Towards a theory of optimal processes
(Russian), volume 110(1). Reports Acad. Sci. USSR, 1956.
[137] Alessandro Vicini and Domenico Quagliarella. Inverse and direct airfoil design using a multi
objective genetic algorithm. AIAA Journal, 35(9):1499–1505, 1997.
[138] Li W. and Padula S. Performance trade study for robust airfoil shape optimization. In 21th AIAA
Applied Aerodynamics Conference, 2003.
[139] David C Wilcox. Turbulence modeling for CFD, volume 2. DCW industries La Canada, 1998.
[140] Jochen Wild, Joel Brezillon, Olivier Amoignon, Jurgen Quest, Frederic Moens, and Domenico
Quagliarella. Advanced high-lift design by numerical methods and wind tunnel verification within the
european project eurolift ii. In 25th AIAA applied aerodynamics conference, 2007.
[141] David H Wolpert and William G Macready. No free lunch theorems for optimization.
Evolutionary Computation, IEEE Transactions on, 1(1):67–82, 1997.
[142] S. Yoon and A. Jameson. Lower-Upper Symmetric-Gauss-Seidel Method for the Euler and
Navier-Stokes Equations. AIAA Journal, 26(9):1025–1026, 1988. doi: 10.2514/3.10007.
[143] D. P. Young, W. P. Huffman, R. G. Melvin, C. L. Hilmes, and F. T. Johnson. Nonlinear elimination
in aerodynamic analysis and design optimization. In Lorenz T. Biegler, Matthias Heinkenschloss,
Omar Ghattas, Bart Bloemen Waanders, Timothy J. Barth, Michael Griebel David E. Keyes, Risto M.
Nieminen, Dirk Roose, and Tamar Schlick, editors, Large-Scale PDE-Constrained Optimization, volume
30 of Lecture Notes in Computational Science and Engineering, pages 17–43. Springer Berlin
Heidelberg, 2003. ISBN 978-3-642-55508-4. URL http://dx.doi.org/10.1007/978-3-642-55508-4_2.
[144] L Zadeh. Optimality and non-scalar-valued performance criteria. Automatic Control, IEEE
Transactions on, 8(1):59–60, 1963.
[145] Ciyou Zhu, Richard H. Byrd, Peihuang Lu, and Jorge Nocedal. L-bfgs-b - fortran subroutines for
large-scale bound constrained optimization. Technical report, ACM Trans. Math. Software, 1994.
[146] David Zingg and Samy Elias. Aerodynamic Optimization Under a Range of Operating Conditions.
98