Gradient, non-gradient and hybrid algorithms for optimizing 3D forging sequences with uncertainties

Lionel Fourment (1)

1 CEMEF - Centre de Mise en Forme des Matériaux

In the frame of computationally expensive 3D metal forming simulations, optimization algorithms are studied in order to find satisfactory solutions within less than 50 simulations and to handle complex optimizations problems with several extrema. Two types of algorithms are selected, which both utilize a meta-model to approximate the objective function and so reduce computational cost. This model either supports standard Evolutionary Algorithms, such as Genetic Algorithms, or is sequentially improved until finding a satisfactory and well approximated solution. The Meshless Finite Difference Method is the utilized meta-model, without (standard algorithm) or with (hybrid algorithm) the gradient information. This meta-model approach allows taking into account uncertainties on optimization parameters in an inexpensive way. The optimization procedure is modified accordingly. The proposed algorithms are first evaluated and compared on standard analytic functions, and then applied to a 3D forging benchmark, the shape optimization of preform tool in order to minimize the potential of fold formation.

Mots clés: Design of Experiments; Evolutionary Algorithm; Forging; Optimization Algorithm; Response Surface; Sequential Approximation; Shape Optimization; Tool Design; Uncertainties

Domaines: Matériaux

Lionel Fourment. Gradient, non-gradient and hybrid algorithms for optimizing 3D forging sequences with uncertainties. Materials Processing and Design, Modeling, Simulation and Applications, NUMIFORM '07: 9th International Conference on Numerical Methods in Industrial Forming Processes, Jun 2007, Porto, Portugal. pp.Pages 475-480, ⟨10.1063/1.2740856⟩. ⟨hal-00510567⟩