Estimation of blade loads from nacelle acceleration to detect stall-induced vibrations in standstill wind turbines

Stall Induced Vibrations (SIV) are an important consideration for wind turbine blade design, especially for large, modern wind turbines which require flexible blades. SIV are characterised by growing vibrations and can be detected by monitoring the edgewise bending moments at the blade root. However, the sensors to measure the edgewise bending moments are normally not available in many turbines, and the closest possible measurements available are the nacelle accelerations, from which it is not always possible to detect the presence of SIV by direct monitoring. In this paper, a data-driven method is proposed to reconstruct the blade root bending moment vibrations from nacelle acceleration measurements through system identification techniques. We first start with an analysis of data collected in a measurement campaign and propose that nacelle accelerations and blade load vibrations can be modelled as the outputs of a Linear Time-Invariant (LTI) system around an operating point defined by the mean wind speed, inflow yaw angle and rotor azimuth angle, and with the instantaneous wind speed as the input. An existing subspace method is enhanced to identify the LTI model of the wind turbine from the offline training data when all the inputs and outputs of the system are available. Then later during real-time, when only one of the outputs namely the nacelle accelerations is available, a structure of a Kalman filter estimator is proposed to estimate the other output. The proposed method is evaluated on test data and it estimates the blade edgewise vibrations with good accuracy in the presence of SIV.