Seismic risk analysis of wind turbine support structures - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 718

ments using accurate nonlinear dynamic analyses of offshore wind turbine
support structures as explained next.
26.3.3 Virtual experiment data
A set of representative confi gurations is selected to generate virtual experi-
ments used to calibrate the proposed probabilistic demand models. The
representative confi gurations are selected by conducting an experimental
design to maximize their information content and minimize the computa-
tional costs associated with running 3D nonlinear fi nite element analyses.
Latin hypercube sampling
In the 'classical' design of physical experiments, a random variation is
accounted for by spreading the sample points out in the design space and
by taking multiple (replicated) data points (Simpson et al. , 2001). However,
when it comes to deterministic computer experiments, there is no random
error and no replication is required. According to Simpson et al. (2001), the
design space is defi ned as the region bounded by the upper and lower limits
of each design (input) variable being studied and the sample points should
be chosen to fi ll the design space for computer experiments such that they
spread as far from each other as possible. There are several 'space fi lling'
designs in the literature; here the Latin hypercube sampling technique
introduced by McKay et al. (1979) is used to select representative confi gura-
tions of the support structure. The Latin hypercube sampling technique
maximizes the minimum distance between sample points, while the range
of each variable, x i , is divided into N strata of equal marginal probability
1/ N ; therefore, this method ensures that the sampling has a good coverage
of the design space. A total of 100 confi gurations are generated. Variables
considered to characterize each wind turbine confi guration and their ranges
are presented in Table 26.2.
Analytical modeling
Finite element models are developed in ABAQUS (2007) to simulate the
dynamic response of the support structure of typical offshore wind turbines,
subject to wind, wave, current, and turbine operational loading as well as
earthquake. The fi nite element model of the support structure is constructed
such that it accounts for the nonlinearity of the soil behavior and soil-struc-
ture interaction.
Representative ground motions for the virtual experiments must be
properly selected in order to assess the seismic demand variables of interest
and their associated uncertainties. In particular, the selected ground motion