Complex systems are subject to uncertainties and related simulations are prone to numerical errors. Therefore, the naive use of a standard optimization algorithm may fail or converge to solutions of poor efficiency in real life conditions. To overcome these difficulties, we investigate the use of surrogate models, like Gaussian Processes (GP), that can account for uncertainty and errors to drive efficiently an optimization procedure.

Research axes

  • Surrogate models for optimization accounting for numerical errors. : We investigate the use of GP models to infer simulation errors and solve efficiently complex optimization problems.

  • Characterization of model uncertainty for turbulent flows : Surrogate models are used to quantify the impact of the model choice, or coefficients calibration, while optimizing flow configurations.

Some related papers

  • Efficient optimization procedure in non-linear fluid-structure interaction problem: Application to mainsail trimming in upwind conditions, M. Sacher, F. Hauville, R. Duvigneau, O. Le Maître, N. Aubin, M. Durand, Journal of Fluids and Structures, No 69, February 2017

  • Comparison of turbulence closures for optimized active control, R. Duvigneau, J. Labroquère, E. Guilmineau, Computers & Fluids, No 124, January 2016

Example pic Example pic

GP models representing the reciculation length for a backward facing step flow w.r.t. synthetic jet parameters, without (left) and with (right) noise filtering.