Optimization and robustness analysis have become important tools for the virtual development of industrial products. In parametric optimization, the optimization variables are systematically modified by mathematical algorithms in order to get an improvement of an existing design or to find a global optimum. The design variables are defined by their lower and upper bounds or by several possible discrete values. In real world industrial optimization problems, the number of design variables can often be very large. Unfortunately, the efficiency of mathematical optimization algorithms decreases with increasing number of design variables. For this reason, several methods are limited to a moderate number of variables, such as gradient based and Adaptive Response Surface Methods. With the help of sensitivity analysis the designer identifies the variables which contribute most to a possible improvement of the optimization goal. Based on this identification, the number of design variables may be dramatically reduced and an efficient optimization can be performed. Additional to the information regarding important variables, sensitivity analysis may help to decide, if the optimization problem is formulated appropriately and if the numerical CAE solver behaves as expected.

A modern approach to search for better designs or to compute the ”best” design has to introduce all available engineering know how and has to automate a multidisciplinary optimization process. By specifying the design criteria as objectives and constraints and specifying the space of all possible designs with optimization parameters, a framework for numerical optimization can be defined. Part of the challenge of defining a multidisciplinary optimization problem will be the communication of different design groups about conflicting objectives, fixed criteria or weighted com-promise objective functions. Consequently, the degree of non-linearity has to be taken into account in the optimization process. Because of that the resulting optimization problem may become very noisy, very sensitive to design changes or ill conditioned for mathematical function analysis (non-differentiable, non-convex, non-smooth).

That defines the requirements to a modern optimization tool. Arbitrary solvers have to be connected, optimization strategies for smooth as well as for non-smooth or even ill-posed problems have to be available.

Besides the problem to find an optimal design, the evaluation of the robustness, for example, the sensitivity of unavoidable scatter of design variables due to the structural response, becomes more and more important. Very often, optimized designs tend to be very sensitive to small (sometimes random) fluctuations of parameters. Such phenomena may occur due to system instabilities like bifurcation problems in the structure. Design robustness can be checked by applying a systematic perturbation analysis based on a randomly generated design sample set. Statistics on the sample set allows the evaluation of robustness of the optimized design.