Search

This paper proposes a new approach to calculate aeroelastic stability of operating wind turbines using system identification, and explores how an efficient numerical experiment can be designed. The data-driven method is based on time series including induced motion of the blades, which can stem from any fidelity simulation. While this study utilises HAWC2, the findings enable the usage of e.g. computational fluid dynamics in a fluid-structure-interaction simulation. Different forcing sequences are suitable to excite the blade, but especially the doublet and chirp show favourable characteristics and results for an application of no more than five seconds. System identification, the Multivariable Output-Error State sPace (MOESP) algorithm in particular, is used to find an algebraic equivalent of the simulated turbine. Using the output-only version, about 20 seconds of simulated time are sufficient to identify the system such that the predicted response matches the input with an R2-score of > 99.8 %. A comparison with HAWCStab2 shows frequency differences of < 0.5% across the wind speed range, matching mode shapes but some bigger deviations in aeroelastic damping.

In this work we present multi-fidelity steady-state aeroelastic framework that leverages the state-of-the-art simulation tool HAWC2 for the structural model, and a variety of aerodynamic models, comprising of the low fidelity blade element momentum (BEM) method, the medium fidelity blade element vortex cylinder (BEVC) method and the coupled near wake and vortex cylinder method, and finally the high-fidelity CFD solver EllipSys3D. The aeroelastic framework is part of AESOpt, an aerostructural framework for design of wind turbine blades. The different aerodynamic models are applied to compute the aeroelastic steady state of the newly designed IEA 22 MW Reference Wind Turbine. The results show a very good agreement between the medium- and high-fidelity aerodynamic models with a maximum of 2.7% difference between the high-fidelity aeroelastic response and that of the lower fidelities.

This reports presents comparison of the steady state HAWC2 [1, 2, 3] simulation results and the HAWCStab2 computations of the DTU 10 MW reference turbine [4], as well as a comparison of structural blade-only frequencies and damping ratios. It serves as a simple verification study of the HAWCStab2 [5, 6, 7] computations. The steady state comparison is shown in Section 2. For a fair comparison, the following simplifications of the DTU 10 MW reference turbine are made: • no gravity; • shaft tilt angle is set to zero, since HAWCStab2 assumes the inflow is perpendicular to the rotor plane; • uniform aligned inflow conditions (no turbulence, shear, veer or yaw); • tower and shaft flexibility are not considered to assure the shaft remains perfectly aligned with the wind inflow vector (horizontal). Furthermore, each main-body in the HAWC2 model must contain as many bodies as elements, to be equivalent to the co-rotational formulation used by HAWCStab2. There are four test cases considered in the steady state comparison: • Case 1: no blade flexibility, and the aerodynamic modelling reduced to strip theory: no induction and no tip correction, labelled as “no induction” or “without induction”. • Case 2: no blade flexibility in conjunction with BEM induction model and Prandtl tip correction (labelled as “induction+tip”). • Case 3: flexible blades in conjunction with “induction+tip”. Includes power and thrust curve comparison with HAWCStab2 2.15. • Case 4: flexible blades in conjunction with “induction+tip”, 15 degrees coning. Includes power and thrust curve comparison with HAWCStab2 2.15. Both HAWC2 and HAWCStab2 have the ability to use different aerodynamic models. For the “induction+tip” model, the rotor induced velocities are calculated with Blade Element Momentum theory, and the presence of the tip vortex is accounted for by the Prandtl tip loss model. For not completely planar rotors, for example

Surrogate models are vital for offshore wind farm optimisation, or digital twin for rapid execution. Reliable surrogates are difficult to make, especially for large wind farms and for fatigue loads. One limitation is the high resolution of turbulence needed for fatigue calculations and the vast size of a wind farm. We present code optimisations for implementing the Mann model for farm-sized flows, which are infeasible otherwise. Additionally, we present a case study of a full load surrogate for the Lillgrund offshore wind farm using large turbulence box generation in aeroelastic wind farm simulations using HAWC2Farm. Various mappings between the wind field in front of a turbine and the corresponding turbine structural fatigue loads are presented. The best-performing surrogate model, using a proper orthogonal decomposition of the input space with an artificial neural network, is able to significantly reduce the error in fatigue load estimates by as much as a factor of 5 compared to conventional methods. The presented recommendations are crucial to generating reliable wind farm load surrogates for wind farm optimisation purposes.

This repository contains supporting data for Camargo et al.: 'Regionalizing Sea-level Budget with Machine Learning Techniques', Ocean Sciences (2022), https://egusphere.copernicus.org/preprints/2022/egusphere-2022-876/.

A parked wind turbine can be subjected to large yaw angles, which causes vortex shedding. Under some particular inflow conditions, the interaction between the shed vorticity and the blade structure could lead to Vortex-Induced Vibrations (VIV). VIV are an instability where the response grows until it reaches a limit cycle oscillation. In this paper, we will focus on this last part of the response. We will see that the shedding makes the aeroelastic system linear time-periodic, and study its stability with Floquet theory. We will use high-fidelity simulations of different inflow conditions, and quantify the stability of the system using a model-free approach.