Abstract. This article presents an application of the Kalman filtering technique to estimate loads on a wind turbine. The approach combines a mechanical model and a set of measurements to estimate signals that are not available in the measurements, such as wind speed, thrust, tower position, and tower loads. The model is severalfold faster than real time and is intended to be run online, for instance, to evaluate real-time fatigue life consumption of a field turbine using a digital twin, perform condition monitoring, or assess loads for dedicated control strategies. The mechanical model is built using a Rayleigh–Ritz approach and a set of joint coordinates. We present a general method and illustrate it using a 2-degrees-of-freedom (DOF) model of a wind turbine and using rotor speed, generator torque, pitch, and tower-top acceleration as measurement signals. The different components of the model are tested individually. The overall method is evaluated by computing the errors in estimated tower-bottom-equivalent moment from a set of simulations. From this preliminary study, it appears that the tower-bottom-equivalent moment is obtained with about 10 % accuracy. The limitation of the model and the required steps forward are discussed.

No abstract

Abstract. The paper presents an application of the Kalman filtering technique to estimate loads on a wind turbine. The approach combines a mechanical model and a set of measurements to estimate signals that are not available in the measurements, such as: wind speed, thrust, tower position, and tower loads. The model is several fold faster than real-time and is intended to be run online, for instance, to evaluate real-time fatigue life consumption of a field turbine using a digital twin. The mechanical model is built using a Rayleigh-Ritz approach and a set of joint coordinates. The paper presents a general method and illustrates it using a 2 degrees of freedom model of a wind turbine, and, using rotor speed, generator torque, pitch, and tower-top acceleration as measurement signals. The different components of the model are tested individually. The overall method is evaluated by computing the errors in estimated tower bottom equivalent moment from a set of simulations. From this preliminary study, it appears that the tower bottom equivalent moment is obtained with about 10 % accuracy. The limitation of the model and the required steps forwards are discussed.