François SCHOTT "Contribution to robust and fiabilist optimization Application to the design of acoustic radio-frequency filters"

François Schott PhD Works : "Contribution to robust and fiabilist optimization Application to the design of acoustic radio-frequency filters"

Abstract :This thesis aims to develop robust and fiabilist optimiztion means in order to face the future requirements of the radio-frequency (RF) filter market. The goals of this thesis are: to reduce the optimization process timespan, to be able to find a solution that fully satisfies a tough bill of specifications and to reduce failure rate due to manufacturing uncertainties. Several research works has been done to achieve these goals.
During the formulation phase of an engineering design optimization (EDO) process, ambiguities, leading to unsatisfying solutions, could happen. In this case, some phases of the EDO process has to be iterated increasing its timpespan. Therefore, a Framework to properly formulate an optimization problem has been developed. During a run of optimization, for the algorithm to solve the problem according to designer preferences and thus avoid un-satisfying solution, two challenges, among others, have to be faced. The variable challenge is about handling mixed variables with different order of magnitudes while the satisfaction challenge is about properly computing satisfaction. The Normalized Evaluations Approach has been developed to face these challenges. The resolution method efficiency strongly relies on the choice of its core element: the algorithm. Hence, the high number of optimization algorithms is a challenge for an optimizer willing to choose the correct algorithm. To face this challenge, a Benchmark, being a tool to assess the algorithm performance and to be able to select the correct algorithm for a given problem, has been developed. Algorithm efficiency depends on the values given to its parameters, its setting. A common practice is to tune parameters manually which does not guarantee the best performance. A solution to this problem is to perform meta-optimization (MO) which consists in optimizing an algorithm efficiency by tuning its parameters. A MO approach using a benchmark to evaluate settings has been tested. A fiabilist optimization method, taking the uncertainties into account, has to be developed. However, this method has to do so without degrading resolution time, which is usually the case with fiabilist method. Therefore, a Sparing Fiabilist Optimization method taking uncertainties into account without increasing too much the numerical resolution timespan has been developed.
These methods have been applied to optimize a RF filter, with a tough bill of specifications, for which no fully satisfying solution where found before the thesis. By using methods developed during this thesis, a determinist solution, not taking uncertainties into account, which fully satisfies the bill of specifications, has been found. Moreover, a fiabilist solution having a 71% success rate has been found. As a conclusion, it appears that optimization methods developed during these thesis where sufficient to face the future requirements of the radio-frequency filter market.

Jury composition :

Thomas BARON, Ingénieur de Recherche, Université Bourgogne - Franche-Comté, PhD Director

Yann MEYER, Maître de Conférences, Sorbonne Universités, Université de Technologie de Compiègne, CNRS, UMR 7337 Roberval, centre de recherche Royallieu, CS 60 319, PhD CoDirector

Dominique CHAMORET, Maître de Conférences, ICB UMR 6303, CNRS Univ. Bourgogne Franche-Comté, UTBM, PhD Co-Director

Omar ELMAZRIA, Professeur des Universités, Université de Lorraine - Institut Jean Lamour - CNRS Reviewer

Jean BIGEON, Directeur de Recherche,  CNRS, G-SCOP - Equipe SIREP  Reporter

Lhassane IDOUMGHAR, Professeur des Universités, Université de Haute-Alsace - Institut IRIMAS, Laboratoire LMIA, Equipe MAGE, Reporter

Localization : Amphithéâtre Gagnepain, 15B, Rue des Montboucons, 25000 Besançon