Pursuing Perfection

• Preliminary Exploration Methods:   used to provide an initial sampling of the design space, include algorithms derived from Design Of Experiments (DOE) theory.
• Multi-Objective Optimization Methods:   used to find designs that lie anywhere in the design space but near the Pareto Frontier, include genetic algorithms and evolution strategies.
• Local Refinement Methods:  used to investigate the space around a given design, these single objective methods look for a local maxima or minima employing so-called Hill Climbing algorithms.
• Special-Purpose Plug-ins:  third-party optimization algorithms that communicate with modeFRONTIER through its proprietary Plug-In Interface and provide solutions to specific problems.

The approach used by modeFRONTIER to find optimal designs can be controlled by the user and can vary from the use of a single technique, to an hybridization of several techniques (e.g., one or more DOE methods can be used to generate an initial population of designs, which is then evolved using a Genetic Algorithm, and whose best candidates are then refined through an Hill Climbing algorithm).

It is important to note that, as the numerical models used to simulate a design are often computationally heavy, it is often unfeasable to carry out a large number of design evaluations. To overcome this problem it is possible to complement these expensive computations with an approximation of the design objectives computed using the available data sets. The values obtained using the approximation surface (called a Response Surface) are then used to suggest new candidate designs, which are computed and whose values are used to refine the response surface.

The combination of this tool with the others provided by modeFRONTIER leads to the most effective and robust optimization technique available.