Page 316 - Wind Energy Handbook
P. 316

290                        DESIGN LOADS FOR HORIZONTAL-AXIS WIND TURBINES


          (2) A probability density function for the load cycle ranges can be derived
             empirically, based on the spectral properties of the power spectrum of the
             stochastic and periodic components of loading combined.

          The second approach is considered in the next Section.


          5.9.3 Fatigue prediction in the frequency domain


          The probability density function (p.d.f.) of peaks of a narrow band, Gaussian
          process are given by the well-known Rayleigh distribution. As each peak is
          associated with a trough of similar magnitude, the p.d.f. of cycle ranges is Rayleigh
          likewise.
            Wind turbine blade loading cannot be considered as narrow band, despite the
          concentration of energy at the rotational frequency by ‘gust slicing’ (Section 5.7.5),
          and neither can it be considered as Gaussian because of the presence of periodic
          components. Dirlik (1985) produced an empirical p.d.f. of cycle ranges applicable to
          both wide and narrow band Gaussian processes, in terms of basic spectral proper-
          ties determined from the power spectrum. This was done by considering 70 power
          spectra of various shapes, computing their rainflow cycle range distributions and
          fitting a general expression for the cycle range p.d.f. in terms of the first, second and
          fourth spectral moments. Dirlik’s expression for the cycle range p.d.f. is:

                                                   2
                                 D 1    Z=Q  D 2 Z   ( Z =2R )   ( Z =2)
                                                      2
                                                                  2
                                    e    þ     e        þ D 3 Ze
                                 Q          R 2
                           p(S) ¼                p ffiffiffiffiffiffiffi                    (5:117)
                                                2 m o
          where
                                                           2
                                     2
                   p ffiffiffiffiffiffiffi  2(x m   ª )    (1   ª   D 1 þ D )
             Z ¼ S= m o , D 1 ¼        , D 2 ¼             1  , D 3 ¼ 1   D 1   D 2
                                1 þ ª 2            1   R
                                                                 r ffiffiffiffiffiffiffi
                                                     2
                 1:25(ª   D 3   D 2 R)     ª   x m   D 1      m 1  m 2       m 2
             Q ¼                   , R ¼              2  , x m ¼      , ª ¼ p ffiffiffiffiffiffiffiffiffiffiffiffiffi ,
                                        (1   ª   D 1 þ D )
                         D 1                                  m 0  m 4      m 0 m 4
                                                      1
                 ð
                  1
                     i
            m i ¼   n S ó (n)dn  S ó (n) is the power spectrum of stress,
                  0
                and S is the cycle stress range:
            Although the Dirlik cycle range p.d.f. was not intended to apply to signals
          containing periodic components, several investigations (Hoskin et al. (1989), Mor-
          gan and Tindal (1990), Bishop et al. (1991)) have been carried out to determine its
          validity for wind turbine fatigue damage calculations, using monitored data for
          flapwise bending from the MS1 wind turbine on Orkney. Cycle range p.d.f.s were
          calculated from power spectra of monitored strains using the Dirlik formula and
          fatigue damage rates derived from these p.d.f.s compared with damage rates
   311   312   313   314   315   316   317   318   319   320   321