Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
5 Efficient Deterministic Approaches for Aerodynamic Shape <strong>Optimization</strong> 113<br />
in aerodynamic shape optimization by Jameson in his works on potential<br />
equations [19] <strong>and</strong> later also for the Euler [21] as well as Navier-Stokes [20]<br />
equations. Its main advantage over the finite differences is the significant increase<br />
in speed since the corresponding numerical effort is independent <strong>of</strong> the<br />
number <strong>of</strong> design variables. However, the implementation <strong>of</strong> the continuous<br />
adjoint approach may be time-consuming <strong>and</strong> error-prone.<br />
On the other h<strong>and</strong>, in the discrete case, one takes the discretized flow<br />
equations for the derivation <strong>of</strong> the discrete adjoint problem (see e.g., [3, 12]).<br />
This can be automated by so-called algorithmic, or automatic, differentiation<br />
(AD) tools.<br />
AD is a comparatively new field <strong>of</strong> mathematical sciences. This technique<br />
is based on the observation that various elemental operations (like +, −, ×)<br />
build up the cost function as their concatenation. Therefore, applying the<br />
chain rule to this concatenation results in an automated differentiation <strong>of</strong> the<br />
cost function. Depending on the starting point <strong>of</strong> the differentiation process –<br />
either at the beginning or at the end <strong>of</strong> the respective chain <strong>of</strong> concatenations<br />
– one distinguishes between the forward mode <strong>and</strong> the reverse mode <strong>of</strong> AD.<br />
Using the reverse mode <strong>of</strong> AD, gradients can be computed very accurately at<br />
a computational cost that is independent <strong>of</strong> the number <strong>of</strong> design variables.<br />
This is the reason why this method is also called a discrete adjoint method.<br />
At DLR, a few attempts have been made in AD computations. One attempt<br />
is the differentiation <strong>of</strong> the DLR TAUij code by ADOL-C [14], which<br />
is presented in this chapter as part <strong>of</strong> a differentiated optimization chain.<br />
In the next sections, the different adjoint approaches will be explained<br />
for single disciplinary aerodynamic shape optimization first <strong>and</strong> then their<br />
extension to multidisciplinary design optimization (MDO) problems will be<br />
discussed for aero-structure cases.<br />
Finally, we will discuss the so-called one-shot methods. Here one breaks<br />
open the simulation loop for optimization.<br />
To start out, we explain how one can parameterize aerodynamic shapes<br />
by the use <strong>of</strong> deformation techniques.<br />
5.2 Parameterization by Deformation<br />
In aerodynamic shape optimization, a geometry is either given by a parameterization<br />
or can be changed by parameterized deformation. This means that<br />
based on these parameters, a shape can be built up or deformed by a design<br />
vector. Furthermore, the obtained shape has some aerodynamic properties<br />
like the drag coefficient or pressure distribution. Therefore, the task <strong>of</strong> the<br />
aerodynamic shape optimization is to optimize this design vector <strong>and</strong> its<br />
dependent shape for some aerodynamic cost function.<br />
When optimizing, there must be some chain to calculate the cost function<br />
value at a given parameterization. This can be done by deforming a static