NEWPAN 3D Optimisation
An exciting area of on-going work at Flow Solutions, in conjunction
with certain customers and development partners, is the coupling of
the NEWPAN panel method to various 3D optimisers.
Optimisation techniques afford the possibility of minimising or maximising
key aerodynamic objective functions such as lift and drag, whilst
constraining other functions controlling aspects of the geometry
(e.g. thickness) and other aerodynamic characteristics (e.g. flow
separation). Such techniques are showing promise for accelerating
the path through complex multi-disciplinary design cycles where
difficult trade offs often need to be made.
The NEWPAN method is in many respects an ideal code for coupling to
an optimiser, thanks to its rapid execution and robustness. An example
follows of a coupling to the QinetiQ CODAS
(Constrained Optimisation Design of Aerodynamic Shapes) method.
NEWPAN-CODAS CFD Optimisation
CODAS is a powerful CFD-based aerodynamic optimisation tool that improves designs and increases design productivity. The tool consists of the following modules:
- Constrained quadratic otimisation tool.
- Parametric surface manipulation tool.
- Interface to NEWPAN (and other CFD codes).
- Cost function and constraints definition tool. (Standard aerodynamic and
geometric constraint functions are supplied. New design requirements can be
defined through a flexible coding interface.)
The designer specifies the number and position of optimisation
variables controlling geometric surface design to the code. The
optimisation proceeds through iterative changes to the geometry and
flow conditions that, for example, optimise a defined cost function
(e.g. drag) while satisfying geometric and aerodynamic side
constraints (e.g. fuel volume, cruise lift coefficient). When used in
combination with NEWPAN, the complete iterative design process for
performance optimisation, geometry shaping and CFD analysis may be
Geometry can be optimised for single or multiple flow conditions,
including variable geometry. Target pressure distributions can be
specified for inverse design, or as a constraint during optimisation
of other performance drivers.
Recent applications of NEWPAN-CODAS have included "Aerodynamic Design Optimisation applied to a Formula One Car", presented at the 4th MIRA International Vehicle Aerodynamics Conference, 16-17 October 2002.
The aim of the design exercise was to maximise the downforce coefficient (-CL) from the complete rear wing geometry, subject to the geometric constraints imposed by the current FIA Formula One Technical Regulations. These rules limit the span of the wing assembly, and specify a rectangular box in cross-section within which a maximum of three upper wing elements must lie. The optimisation constraints and variables may be summarised as follows:
Objective Function: Minimise downforce coefficient (-CL)
Each of three upper wing elements constrained to
lie within a box of these dimensions in millimetres:
3150 < X < 3500
-500 < Y < 500
600 < Z < 800
The lower wing element was fixed in position and shape.
Specified in this study as maximum permissible boundary layer separations:
Vane element: up to 3% forward of the trailing edge
Main element: up to 3% forward of the trailing edge
Flap element: up to 10% forward of the trailing edge
The boundary layer properties were predicted using PANBL.
For each of the three upper wing elements:
2 camber variables
1 twist variable (applied at the centreline through the quarter chord)
2 rigid body translation variables (X and Z directions)
1 rigid body rotation variable (about Y-axis through quarter chord)
Hence a total of 18 design variables.
CODAS was run for a total of 20 gradient-search cycles, requiring a total of
386 NEWPAN/PANBL calculations. Such an optimisation may be performed
comfortably in an overnight run on a contemporary single CPU workstation.
Results demonstrated an increase in downforce whilst avoiding an increase
in viscous drag.
The study demonstrates the viability of performing complex design optimisations using NEWPAN.
|Wing elements before (black) and after (red) optimisation
||Downforce v/s iteration
||Percent separation v/s iteration