This paper focuses on techniques for estimation of extreme response of flexible risers in a short-term environmental condition. It is shown that the degree of non-linearity is strongly system and excitation dependent and that conventional estimation techniques may introduce severe systematic errors. A new tail fitting technique is found to give more precise extreme value estimates at the expense of an increased statistical uncertainty. A simple method for prediction of the simulation length required to obtain a specified level of confidence in the estimated extreme values is also presented.
Design analyses of flexible risers are today often carried out by stochastic time, domain simulations for a given environmental condition. Several computer programs using finite element or finite difference method have been developed for this purpose. The resulting time record of the actual response type is then used to estimate the design value of the response (lifetime maximum), in most cases identified as the expected largest response for a certain duration of the selected short term environmental condition. Response of flexible risers subjected to ocean waves is known to be non-Gaussian due to nonlinearities introduced by the hydrodynamic loading and the stiffness properties of the pipe. Methods for prediction of extremes for non-Gaussian responses will therefore be fundamental for a reliable design of flexible risers. This paper presents methods for extreme value estimation based on fitting of a proposed distribution function to a sample found from simulation. Emphasis has been put on the involved systematic errors and statistical uncertainties. Systematic errors will be introduced if the selected distribution function is unable to describe the true tail behaviour of the sample distribution. In order to overcome this problem, a tail-fitting technique based on the Gumbel distribution is developed.
