Prediction of Stochastic Blade Responses Using a Filtered Noise Turbulence Model in the FLAP Code

The wind turbine structural dynamics model, FLAP (Force and Loads Analysis Program), has been modified to include turbulent wind fluctuations based on a filtered-noise model, The importance of including the effects of turbulent wind fluctuations in the structural-loads predictive model has long been. recognized. These stochastic loads are the dominant fatigue loads for many structural components in horizontal-axis wind turbines. The turbulence field at the rotor plane is approx­ imated by interpolating functions, which allow the velocity field to vary quadratically for ·velocity com­ ponents normal to the rotor plane and linearly for in­ plane velocity components. The velocity field repre­ sented in this fashion is constructed to vary randomly with time and space and give the proper correlation between spatial locations and velocity components. For the normal velocity components, the spectral represen­ tations of these velocity fluctuations approximate those observed from a rotating turbine blade up to a frequency of two times per rotor revolution (2P). For the less-important in-plane components, the spectral representations approximate those observed from a rotating turbine blade up to lP. The time-domain model described in this paper uses a random number generator to construct a white-noise time series with uniform spectral density over the fre­ quency range of interest. This signal is then filtered to obtain the various wind fluctuations, which are used as input to the FLAP code. This requires as input only the mean wind speed, the turbulence intensity, and an estimate for the integral scale. The modeling is based on the assumptions that the velocity fluctuations are statistically stationary, homogeneous, and isotropic, and satisfy Taylor' s frozen field hypothesis. The von Karman model is used to characterize the correlation between velocities of points separated in space. To gain insight into the usefulness of this turbu­ lence simulation for predicting stochastic turbine loads, the FLAP code was used to model three data cases from the 1986 testing done on the Howden 330-kW tur­ bine. The FLAP code was run with the wind turbulence parameters set to model the actual test conditions to generate simulated time series of bending moments. The response spectra calculated from these time series were then compared with the experimental measurements obtained from the field test. Comparing the simulation results with actual test measurements generally shows good agreement. It takes about an hour to run a 450-revolution FLAP simulation using an 80386-based personal computer (PC) running at 20 MHz. The required knowledge of the actual turbulence characteristics is modest. The mean wind speed and the turbulence intensity are easily com­ puted from time-series wind data. As discussed here, the integral scale of the turbulence can be estimated from calculations of the longitudinal wind spectrum. This type of simulation should become part of the pro­ cess of designing wind turbines, It produces time­ series results that can be used to determine peak loads, and it can be rainflow-counted for estimating fatigue damage rates. In addition, the computational and data input requirements are within the means of even the smallest design team.