New example: satellite brake

It's been a while since we last posted a new movie on our YouTube channel. Due to our work on the latest release of the Toolbox and several projects we simply lacked time.

Today we uploaded a movie of a Neural Network model of a new satellite brake system. The response corresponds to the displacement angle of the brake, and varies over time. The evolution of the angle is influenced by several parameters, three of them were included in this model. Such system is experimental and is used by the Finnish student satellite projects.

A simulator for this system was linked to our toolbox. To optimize the Neural Networks a genetic algorithm was used. The model fitness was multi-objective fashion: sample-error vs LRM. This setup can be used to avoid expensive crossvalidation. The adaptive sampling strategy used is a follow-up on LOLA-Voronoi, the algorithm generates similar designs but does it much faster (especially when the dimensionality of the problem increases), more updates on this can be expected later.