On propeller blade shape optimizations: Global vs. local optimum

Specifically for propeller shape optimizations, version V4.14 sometimes runs into local optima resulting in not-so-feasible (yet interesting) results, such as this one:

The main reason is, that propeller shape optimization problems have a very high modality (that is, there are many local optima) and the global optimum is more challenging to find. If you restart your optimization case multiple times and each time get a different result, then its a good indication that these are local optima and the optimizer struggles to find the global optimum.

To improve the optimization robustness, the algorithms in Aeolus ASP are getting tweaked at the moment. One example of improved robustness is shown below on a an eggholder test function. The left picture uses the current optimizer from Version 4.14 which exhibits limited design space exploration, whereas the right picture shows the new one to come in the next version. and which finds the global optimum

Here is a teaser for some more robust blade shape optimization in the upcoming version. Also, you can see that surface smoothing is now available 🙂