Global optimisation on assembly problems using gradient-based surrogate model and multiparametric strategy - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2014

Global optimisation on assembly problems using gradient-based surrogate model and multiparametric strategy

Luc Laurent

Résumé

Optimization strategies on assembly design are often relatively time expensive because of the large number of non-linear calculations (due to contact or friction problems) required to localize the optimum of an objective function. In order to achieve this kind of optimization problems with an acceptable computational time, we propose to use a two- level model optimization strategy. More precisely we propose a method to achieve global parametric optimisation on interfaces between parts of an assembly (gaps, friction coefficients, preload). The process is based on two main tools: the first is called the MultiParametric Strategy. It allows us to compute with a significant reduction of the computation time many similar assembly problems in the sense that only few parameters vary. Based on the features LaTIn method, MultiParametic Strategy consists in a smart reinitialisation of the LaTIn algorithm using boundary magnitudes on interfaces (a mixed domain de- composition is used). Moreover the MultiParametric Strategy enables us to compute the gradients of the objective function using finite difference method. The second tool is a gradient-based metamodel called Cokriging. This metamodel is built using responses and gradients computed on many sample points generated using Latin Hypercube Sampling. The metamodel provides a very inexpensive approximate response of the mechanical problem with a very good quality of approximation in comparison with kriging metamodel (without gradients). A global optimisation using a Efficient Global Optimizer can be achieved using this surrogate model and a smart infill criterion. Thus the first level of our strategy provides an approximate minimum which supply the second level. Finally this second level is based on a local optimizer (SQP) initialized with the approximate minimum. The optimizer enables us to confirm and/or fix the optimum and converge to the global optimisation. The whole optimisation process and its different steps are studied on few 2D and 3D assembly problems on which significative reduction of the computation time can be obtain (with gains around 10 compared to the non use of the MultiParametric Strategy). The performance is presented and compared with classical methods with non-gradient-based metamodel and without the MultiParametric Strategy.
Fichier non déposé

Dates et versions

hal-01431902 , version 1 (11-01-2017)

Identifiants

  • HAL Id : hal-01431902 , version 1

Citer

Luc Laurent. Global optimisation on assembly problems using gradient-based surrogate model and multiparametric strategy. 11th World Congress on Computational Mechanics, Jul 2014, Barcelona, Spain. ⟨hal-01431902⟩
109 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More