Abstract
One of the more difficult optimal design tasks occurs when the data describing the system to be optimized is either highly nonlinear or noisy or both. This situation arises when trying to design restraint systems for automotive crashworthiness using the traditional lumped parameter analysis methods. The nonlinearities in the response can come from either abrupt changes in the occupants interaction with the interior or from relatively minor fluctuation in the response due to the interactions of two restraint systems such as belts and airbags. In addition the calculated response measures are usually highly nonlinear functions of the accelerations. Two approaches using an approximate problem formulation strategy are proposed. One approach uses a first-order approximation based on finite difference derivatives with a nonlocal step size. The second and more effective approach uses a second-order curve fitting strategy. Successful example problems of up to 16 design variables are demonstrated. A conservative design strategy using a derivative-based constraint padding is also discussed. The approach proves effective because analytical expressions are available for the second-order terms.