Prediction of extreme response of nonlinear structures by extended stochastic linearization

The paper describes stochastic linearization methods that are specifically designed for the purpose of predicting extreme responses. Standard stochastic linearization applied to a nonlinear dynamic system is based on a mean square deviation measure to derive the ‘equivalent’ linear system. Experience indicates that the obtained linear system may have approximately the same mean square response as the original nonlinear system. For predicting extreme response this procedure is not, however, equally suitable. In the case of, e.g. strong hardening type nonlinearities in the dynamic model, application of the standard method of linearization may lead to substantial overestimation of extreme responses. It is shown by example studies that the proposed methods may offer significant improvement in the ability to predict extreme responses.