SENSITIVITY ESTIMATION WITHIN 1ST-ORDER AND 2ND-ORDER RELIABILITY METHODS

An efficient method to compute the sensitivity of failure probability estimates to changes in distribution parameters is proposed. The method is applicable to failure probability estimates obtained by first and second order reliability methods (FORM and SORM). It is based on the fact that current FORM and SORM approaches require transforming the problem into standard normal space and then numerically fitting first/second order approximations. To compute the change in failure probability due to a change in parameter, one can transform the points that were used to numerically fit the approximation into a new standard normal space corresponding to the changed parameter, fit a new approximation and use the new approximation to compute the changed failure probability. The proposed method is computationally efficient because it does not require any new structural response evaluations. In contrast to conventional methods that are accurate for FORM estimates of component problems, the proposed method is accurate for FORM and SORM estimates of both component and system problems.