Section: Software and Platforms

FAMOSA

Participant : Régis Duvigneau [correspondant] .

Opale team is developing the software platform FAMOSA (C++), that is devoted to multidisciplinary design optimization in engineering. It integrates the following components:

• an optimization library composed of various algorithms : several descent methods from steepest-descent method to quasi-Newton BFGS method (deterministic, smooth), the Multi-directional Search Algorithm (deterministic, noisy), the Covariance Matrix Adaption Evolution Strategy (semi-stochastic, multi-modal) and the Efficient Global Optimization method (deterministic, multi-modal). It also contains the Pareto Archived Evolution Strategy to solve multi-objective optimization problems ;

• an evaluation library managing the performance estimation process (communication with external simulation tools) ;

• a metamodel library that contains tools to build a database and kriging models that are used to approximate the objective function for different purposes;

• a scenario library that allows to use the previous components to achieve various tasks:

  • Construct a design of experiments ;

  • Construct a metamodel ;

  • Find the design that minimizes a cost functional ;

  • Find the Pareto front for two cost functionals

  • Play a Nash game to find the equilibrium between two criteria ;

  • Apply a multiple gradient descent strategy to improve simultaneously two criteria.

The FAMOSA platform is employed by Opale project-team to test its methodological developments. The platform is also used by the Fluid Mechanics Laboratory at Ecole Centrale de Nantes for hydrodynamic design applications and ONERA for multidisciplinary design optimization (MDO). Moreover, it is presently tested by Peugeot Automotive industry for external aerodynamic design purpose.