Fatigue Design of Wind Turbines - Wind Turbine Technology: Fundamental Concepts of Wind Turbine Engineering - page 678

Fatigue Load Spectra from HAWT Tests
Deterministic and Probabilistic Models
Mathematical models of the structural-dynamic behavior of wind turbines are now devel-
oped to the point where median ( i.e. , 50th percentile) loads can be predicted with sufficient
accuracy for design purposes. Median loads result from normal wind conditions and normal
aerodynamic responses that are predictable enough to be referred to as deterministic , and
these have been incorporated successfully into computer models of wind turbine structures.
Unfortunately, it is not these median loads that cause fatigue damage but the infrequent, high-
er-amplitude loads with probabilities of exceedance in the range of 10 percent or less. These
fatigue-driving loads, often called probabilistic or stochastic loads, are much more difficult
to predict. Their frequency of occurrence depends on random combinations of configuration,
site, wind, and control characteristics that are not known a priori , if ever.
One solution to the difficult problem of predicting probabilistic loads is to modify a
deterministic model on the basis of operating load data like that illustrated in Table 12-3. For
example, the variability of the wind input in time and space can be made larger and larger
until the output loads from the model equal higher and higher percentile loads measured at
selected points in the structure. This produces a probabilistic wind model that can then be
used as an input to new structural-dynamic models, and the resulting loads will have known
probabilities of exceedance. Analytical tools of this type can be used to
-- modify deterministic mathematical models for the estimation of probabilistic
loads;
-- guide the preliminary fatigue design of critical wind turbine components;
-- establish baseline loads for future field tests.
The semi-empirical approach just described has been found to be both practical and ac-
curate when field test data on a prototype turbine are available, or when an operating turbine
is being modified. In most other cases, however, there is usually a lack of operational test
data on the specific configuration being designed. Instead, engineers must look for guidance
from the probabilistic loads measured on similar wind turbines. In this section we will ex-
amine a set of fatigue load data from tests on a variety of medium- and large-scale HAWTs
as well as empirical equations which correlate these data with basic wind turbine and site
properties.
Representative Field Test Data
The fatigue loads data presented here were measured during various HAWT research
and development projects at the NASA Lewis Research Center (Spera et al. 1985). The
configurations of the rotors tested are described in Table 12-4, and tower and site data are
given in Table 12-5. The parameters listed in these two tables are those that were found to
correlate with cyclic load levels, and they appear in empirical equations to be discussed later.
The 13 data cases in the set include 11 two-bladed rotor designs with diameters from 28 m
to 91.4 m, power ratings from 50 kW to 4,000 kW, truss and shell towers, and eight coastal
and inland sites in the U.S.
Defining the scope of the data set is important when using an empirical or semi-empiri-
cal method to predict fatigue loads, since the objective is to make the process one of interpo-
lation rather than extrapolation. For example, only isolated HAWTs with two-bladed rotors
are represented in this data set, so additional tests would be required to model the effect of