An aspect ratio of 33.8 is chosen, showing that an extremely slender wing is optimal.
The method seems very knteresting to me, and I watched F Trevisi’s talk at the last AWEC, and it was very impressive.
The above quote states my issues with the current state of this paper. The algorithm can only optimize the model that is programmed in the system. When the model is not complete enough or not accurate enough, the conclusions will also not be accurate. Id like to say «garbage in garbage out», but without implying anything about the paper’s quality is poor quality.
From my own investigations which are much less structured, but which takes into account a few aspects that are not included in the paper’s model, such a high AR is just not feasible. As an initial counter argument I would say that such a AWE rig would have extremely low rigidity, unless the kite was made to loop in an extremely tight turn, effectively keeping one wingtip at standstill or moving quite slowly.
I think so far from what I’ve seen, many different designs may have similar performance. Tweaking a design may give extreme improvements along a tangent, but in reality that tangent may just not be feasible. In the end, the precision at which you can describe the model along with the lack of «stiffness» in optimization, makes it very difficult to make an optimizer that actually hits the nail on the head.
Let me state and example of what the problem is: Say you want to optimize the aspect ratio Æ of the kite. The kite kan loop only as tight as a diameter 10x the wingspan. To accomodate a looping «cone» of, say, less than 30 degrees, this limits the overall tether length to be longer than a certain length (to less than l > \frac{10 b}{2 \arcsin{15}} for wingspan b and tether length l).
Now, if the optimizer reduces the Æ, the drag of the kite is increased. But at the same time, the tether may be shortened, leading to a decreased overall drag. The overall drag may be nearly unchanged for a large selection of input values. The optimizer ends up choosing either a very low or a very high Æ due to model inaccuracy.
It seems that for many of these compromises, the overall performance is very constant, and very detailed analysis and design is required to make informed decisions. Eg requiring very accurate CFD simulations of a nearly complete design, and knowing the tether drag coefficient to a high accuracy.
In effect, when you make statements such as «Flygen is better than ground gen», you are basically just relaying the voice of the people who implemented the model and any assumptions about what it must look like
So to summarize my objections; maybe you can achieve the same results of figuring out an optimal configuration by optimizing each part isolated and then having some heuiristics to lead the direction of the design process. This may require a less detailed model, allowing engineers to make some of the more difficult decicions based on experience and gut feeling.
Also a bit unfair to the author, I am not really able to provide any sensible feedback on the optimization method itself.