


Multilevel optimization
Our strategy of multilevel optimization ensures maximum reduction
in time required to solve optimization problem. This becomes possible
due to the simultaneous use of mathematical models of various fidelity and
active automatic switching between them in the optimization process. The proposed
procedure minimizes the number of times the highest fidelity model is
involved without reducing the accuracy of the solution. Our experience
shows that it is possible to solve the optimization problem when the number
of times the highest fidelity model is involved is less than the number
of design variables. This fact allows for considerable (several
orders of magnitude) reduction in CPU time required to solve an optimization
problem.
This feature is equivalent to the case when for the same problem conditions (accuracy of the solution, computer speed, etc.) your analysis program is performing many (dozens) times faster without any actual modifications in the analysis code. Employing our methods, you can actually use very complex and accurate mathematical models (for example, 3D CFD codes) to solve practical multidimensional single and multiobjective optimization problems. Our strategy of multilevel optimization provides an efficient way to solve the complex problems of robust design optimization and robust optimal control.


© Sigma Technology 2001,2009 