Multi point evolutionary optimization without problem specific knowledge and applications to the design of UAV systems

It is well known that Evolutionary Algorithms (EAs) can
provide solutions to design problems that cause difficulty to
conventional deterministic optimizers. These problems are generally
multimodal, involve approximations, are non differentiable and deal more
and more with multi objectives.

In this lecture, we present recent results of a continuing research into
improving the speed and robustness of Hierarchical Asynchronous Parallel
EAS ( HAPEA) as black box optimizers. We will give results showing the
design and optimization of an airfoil geometrty for an unmanned aerial
vehicle (UAV) from completely random starting conditions, including
multi point flaps-up ( cruise) and multi point flaps down (take off)
design. We compute a set of Pareto Optimal airfoils satisfying both
specified constraints in both cases, using panels or Euler flow solvers
coupled with turbulent boundary layers effects.

We conclude by showing that EAs can provide very efficient solution to
the design problem, as well as verifying traditional design
methodologies, whilst not requiring any knowledge of the physics or
internal workings of the solver used.In addition, although efficient
shapes are found during the flap-down case, it is shown how an
occasional modelling limitation of the solver can precipitate spurious
results.

Finally, we investigate numerically the possibility of extending this
methodology to the difficult approach of design with unstructured adapted
mesh Navier Stokes solver.