Study of a topology optimization tool associating a hybrid algorithm with a Voronoï-based formalism for the design of electromagnetic devices

Denies, Jonathan [UCL] Ben Ahmed, Hamid [ENS Cachan Antenne de Bretagne] Dehez, Bruno [UCL] Glineur, François [UCL]

This paper presents and compares three topology optimization tools differing from the optimization algorithms they implement. The first two are respectively based on a local optimization algorithm, i.e. a Nelder-Mead method, and a global one, i.e. a genetic algorithm. The third uses a hybrid algorithm associating the local and the global algorithms used in the first two. Each of them implements the same material distribution formalism, i.e. a dynamic formalism based on Voronoi cells. The results obtained on a practical study case, the design of an electromagnet, show the better performance of the hybrid algorithm as well in terms of speed of the convergence as in terms of quality of the final solution and therefore the interest of coupling the exploration abilty of the global search with the intensification abilty of the local search.