Abstract. We present a digital shadow Kalman filtering framework based on the direct linearization of a trusted multibody aeroservoelastic wind turbine model. In contrast to shadowing based on ad hoc modeling approaches, reusing validated industrial or research-grade models reduces the development effort, leverages resources invested in tuning and validation, and, eventually, increases confidence in the results. Building on earlier work, the filter-internal model is extended to improve applicability under non-symmetric, waked, and yaw-misaligned inflow conditions. In addition to tower fore–aft and rotor-speed dynamics, the model incorporates tower side–side motion as well as blade flapwise and edgewise degrees of freedom. Real-time inflow observers estimate rotor-equivalent wind speed, vertical and horizontal shear, and yaw misalignment, enabling operating-point-dependent scheduling of the linearized model. To further enhance predictive accuracy, the white-box model is augmented with data-driven corrections, considering both a bias-correction approach that acts on states and outputs, and a neural-network-based output correction. The proposed method is validated in simulation under freestream, waked, and wake-steering scenarios and subsequently on field data from an instrumented wind turbine. Additional field cases with extreme shear and waked operation are used to assess robustness. Even without data-driven correction, damage-equivalent loads estimated from field data exhibit accuracy comparable to simulation-based results. When correction strategies are applied, accuracy improves substantially, with damage-equivalent load errors reduced to only a few percent.
Hoghooghi et al. (Fri,) studied this question.