Benchmark study on methods to determine collapse failure probabilities of redundant structures

Abstract Methods to determine system failure probability are compared with respect to their results, capabilities and efficiency. In the case of normally distributed basic variables nearly all methods yield identical results. With non-normal variables, only those methods are reliable which are able to evaluate non-normal safety margins, though with small coefficients of variation the results may differ considerably. Those methods which depend on a plastic limit load analysis to find only one mechanism are unreliable. Since the former methods need the prior evaluation of mode limit state functions and have to consider all modes, their application is restricted to small systems. To evaluate failure probabilities of large systems, an automatic procedure to find the stochastically most relevant modes is necessary.