Report based on Ph.D.-thesis (supervisor: Ove Ditlevsen)
The model correction factor method is proposed as an alternative to traditional polynomial based response surface techniques in structural reliability considering a computationally time consuming limit state procedure as a 'black box'. The class of polynomial functions is replaced by a limit statebased on an idealized mechanical model to be adapted to the original limit state by the model correction factor. Reliable approximations are obtained by iterative use of gradient information on the original limit state function analogously to previous response surface approaches. However, the strength of the model correction factor method, is that in simpler form not using gradient information on the original limit state function or only using this information once, a drastic reduction of the number of limit state evaluation is obtained together with good approximations on the reliability. Methods for solving convergence in especially difficult cases are proposed.