System uncertainties in offshore structures in general have been found to be small but some measurements and some more or less sophisticated structural analyses indicate that they can be significant both with respect to extreme value failure and with respect to fatigue in some cases. A rigorous analysis by stochastic Finite element codes in conjunction with modern reliability methods such as FORM/SORM, however, is not yet feasible for realistic structures. This is primarily due to the fact that rigorous FORM/SORM methods require numerical or analytical gradients involving considerable computational effort. Even if this problem can be overcome the existence of multiple pVpoints especially in the dynamic case and/or the necessity to investigate many "hot spots" makes such an analysis difficult. As an alternative simulation methods have been proposed among which so-called adaptive conditional expectation methods appear most suitable as they can overcome some of the more important disadvantages of the more rigorous methods and retain the advantages of FORM/SORM. In the paper, a two dynamic structures exposed to random wave loading with/without current is investigated with emphasis on the methodological aspects. Apart from the loading environment the coefficients in the Morison equation, the stiffnesses of the structural members, the stress concentration factors and the deck load are assumed uncertain. In agreement with earlier studies it is found that adaptive and non-adaptive conditional expectation procedures are in fact appropriate means to investigate the influence of system uncertainties. However, adaptive schemes can be suboptimal whenever several "hot spots" need to be investigated simultaneously. Their convergence behaviour also depends strongly on the starting conditions for the simulations. In any case the coefficient of variation of the probability estimates must not be used as a convergence criterion, Furthermore the system uncertainties should be small relative to other uncertainties and not too high dimensional. The numerical effort required for real large structures may still be excessive and schemes should be designed which use more information from sample points than incorporated in the implemented procedure.
展开▼
