Surrogate-Based Optimization at ONERA: Some Recent Examples
published: July 20, 2009, recorded: July 2009, views: 4704
Report a problem or upload filesIf you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.
With the development of computational resources and the increasing complexity of industrial needs, stohastic optimization strategies, and especially those based on evolutionary concepts, have had a growing success among the community in the recent years. Indeed, they offer global-focusing minimization opportunities to potentially complex problems (e.g. with multiple minima over non-connected search spaces of (dis)continuous state variables) without relying on the computation of the objective function´s gradient, and as such remain almost completely independent on the physical nature of the problem to be treated (analyzer and optimizer are two distinct and autonomous processes to be interfaced with one another, hence the usual "black-box˝ denomination). Yet, the search of an optimum based on the entire range of possible solutions (where, formally, nothing guarantees the absolute global character of the result) is usually penalized by its important computational cost, which can increase exponentially with the complexity of the problem (Belmann´s curse of dimensionality) and become rapidly prohibitive (especially when one deals with precise aerodynamic evaluations on large configurations). This remark partly explains the popularity of surrogate-based optimization procedures, where the expensive analyzer is replaced by a low-fidelity but cheap model on which the optimizer browses. ONERA has been active on the field of surrogate optimization for some time now, with topics ranging from surrogate modeling itself (RBF/ANN, (Co)Kriging, high order RSM...) to efficient coupling and optimization (sampling methods, refinement criteria, on- or off-line implementation...), for a wide variety of applications (single or multi-objective performances of multidisciplinary problems). It is the aim of this presentation to review some of the results obtained throughout some of the most recent projects, such as the optimization of flow control parameters for novel high-lift design.
Link this pageWould you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !