POD-Based Sparse Stochastic Estimation of Wind Turbine Blade Vibrations
Lorenzo Schena, Wim Munters, Jan Helsen, Miguel A. Mendez · 2025
Read POD-Based Sparse Stochastic Estimation of Wind Turbine Blade Vibrations by Lorenzo Schena, Wim Munters, Jan Helsen, Miguel A. Mendez (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-08505.txt.
Introduction
Schena, Munters, Helsen, and Mendez presented a method in twenty twenty-five for estimating wind turbine blade vibration from very few sensors. Knowing a blade's full bending state helps prevent damage and control loads. Fitting sensors everywhere is impractical, so the full picture must be inferred from sparse readings.
Their framework first compresses high-fidelity simulations into a small set of main patterns. A statistical estimator then rebuilds the blade's state in that compressed space. A Kalman filter blends in a simpler wind-driven rotation model fed by measured wind speed. Tests in a synthetic setting with turbulent wind and sensor noise were used. Readings at only four points sufficed to reconstruct three-dimensional bending and acceleration well, suggesting uses in live monitoring and sensor placement.
Catalog record
| CREATOR | Lorenzo Schena, Wim Munters, Jan Helsen, Miguel A. Mendez |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-08505
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
POD-Based Sparse Stochastic Estimation of Wind Turbine Blade Vibrations Lorenzo Schena, Wim Munters, Jan Helsen, Miguel A. Mendez Abstract This study presents a framework for estimating the full vibrational state of wind turbine blades from sparse deflection measurements. The identification is performed in a reduced-order space obtained from a Proper Orthogonal Decomposition (POD) of high-fidelity aeroelastic simulations based on Geometrically Exact Beam Theory (GEBT). In this space, a Reduced Order Model (ROM) is constructed using a linear stochastic estimator, and further enhanced through Kalman fusion with a quasi-steady model of azimuthal dynamics driven by measured wind speed. The performance of the proposed estimator is assessed in a synthetic environment replicating turbulent inflow and measurement noise over a wide range of operating conditions. Results demonstrate the method’s ability to accurately reconstruct three-dimensional deformations and accelerations using noisy displacement and acceleration measurements at only three spatial locations. These findings highlight the potential of the proposed framework for real-time blade monitoring, optimal sensor placement, and active load control in wind turbine systems. Journal: Journal of Sound and Vibration List of Acronyms LNM linear normal modes (eigenmodes) LOM low order models NNM nonlinear normal modes POD Proper Orthogonal Decomposition POM proper orthogonal modes ROM reduced order models 1 Introduction Wind turbine blades are exposed to various sources of unsteady loads arising from rotation, turbulence, control inputs (e.g., cyclic pitch), and other environmental factors [ Söker, 2013 , Schubel and Crossley, 2012 ] . These loads, coupled with the unprecedented flexibility of modern blades, lead to large deflections and motions. Blade deformations can become large enough to change the inertial properties of the blades, altering their modal structure and resonant response depending on the operating conditions [ Lopez-Lopez et al., 2020 , Skjoldan and Hansen, 2012 , Bottasso and Cacciola, 2015 , Riva et al., 2016 ] , or even azimuthal positions [ Acar and Feeny, 2018 ] . Moreover, large deflection can degrade aerodynamic performances [ Larsen et al., 2004 ] , induce harmful aeroelastic instabilities [ Rasmussen et al., 2003 , Kallesøe, 2011 ] , increase aerodynamic-driven fatigue loads [ Liu et al., 2017 ] and intensify mode couplings induced by blade geometry (e.g., pre-bending) and the anisotropic properties of modern composite materials [ Stäblein et al., 2017 ] . Monitoring and predicting blade behaviour is therefore critical for wind turbine operation, to guide active load mitigation strategies [ Kragh et al., 2014 , Cooperman and Martinez, 2015 ] , to provide advanced indicators for predictive maintenance [ Hameed et al., 2009 ] , and to validate widespread aeroelastic tools employed by the industry and academia alike [ Lehnhoff et al., 2020 ] . Yet, high-fidelity modelling of these fluid-structure interactions requires computationally expensive numerical solvers [ Wang et al., 2016 , Li et al., 2020 ] , unsuitable for applications in which rapid feedback is of the essence. This has driven extensive research into Low-Order Models (LOMs) and Reduced-Order Models (ROMs), which offer complementary pathways for fast predictions. On the one hand, LOMs are based on lumped descriptions that make simplifying yet reasonable assumptions [ Pao and Johnson, 2009 , Kallesøe, 2006 ] . These models prioritize interpretability and simplicity and are generally amenable to analytical treatment, offering the simplest approach to control, stability analysis, and optimization. However, the derivation of LOMs becomes impractical as the complexity of the structural response increases [ Volk et al., 2020 ] . In contrast, ROMs reduce the dimensionality of a high-fidelity simulation by projecting the high-dimensional state onto a reduced-dimensional space. Projection-based ROMs differ primarily in the choice of basis functions used for projection. Common choices are mode shapes obtained from linearized formulations [ Sønderby, 2013 , Jonkman et al., 2018 ] , potentially prescribed according to the operative condition [ Adegas et al., 2013 ] , or general-purpose bases such as Rayleigh-Ritz [ Branlard et al., b ] or finite element methods (FEM) [ Rezaei et al., 2015 ] . Alternatively, hybrid linear methods could be obtained by expanding traditional bases of eigenmodes (or Linear Normal Modes, LNM), which cannot handle strong nonlinearities [ Gözcü and Stolpe, 2020 ] , with modes such as FEM [ Tarpø et al., 2020 , Iliopoulos et al., 2016 ] or heuristic bases tailored by optimization [ Gözcü et al., 2022 ] . The nonlinear blade dynamics can also be directly addressed via the identification of the invariant manifolds of the system dynamics, described by Nonlinear Normal Modes (NNM) [ Shaw and Pierre, 1991 , Pesheck et al., 2002 , Touzé et al., 2021 , Martin et al., 2023 ] . However, inferring nonlinear manifolds from sparse measurements is more challenging, both because of the ill-posedness of the problem and the difficulty in inverting the nonlinear mapping from low to high dimension. In monitoring applications, these reduced-order representations enable the integration of model predictions with measurements, providing computationally efficient methods for identifying and tracking a physical system state in real-time. This integration can enable (1) sparse sensing if low-dimensional measurements are used to infer the full-dimensional state; (2) virtual sensing if sparse sensing is combined with a process model that maps observables to unobserved quantities of interest (QoIs); and (3) digital twinning if the continuous stream of information is used to tailor a model to a specific system or machine. Sparse sensing methods optimally locate sensors to retrieve the maximum amount of non-redundant information [ Manohar et al., 2018 ] , and are mostly used in the context of Operational Modal Analysis (OMA) [ Schulze et al., 2016 , Eichner et al., 2023 ] and structural health monitoring [ Ostachowicz et al., 2019 ] . Virtual sensing applications use these measurements to infer unmeasured quantities combining measurements and a process model. Typical inference approaches are Kalman Filters (KF, Welch and Bishop [1995] ) in their Augmented (AKF) formulation, to simultaneously estimate both observed (or primary) states and unmeasured (or secondary) states Lourens et al. [2012] . These have been used for strain estimation during so-called pull and release tests of wind turbine blades [ Vettori et al., 2021 ] , or the inference of tower-bottom moments [ Branlard et al., a ] , to give some examples. In these approaches, process models can be LOR, ROM, or purely data-driven. For example, Bilbao et al. [2022] used a Gaussian process and Kalman filtering to estimate tower loads, while [ Azzam et al., 2021 ] combined neural networks with multibody models for virtual sensing of gearboxes. An example of the application of ROM as a process model for statistical inference via Kalman filtering is the work by [ Mehrjoo et al., 2022 ] on the use of data-driven modal analysis for optimal sensor placement to estimate the compressions of the support jacket. The combination of a simplified model for fast prediction, enhanced or tuned by sensor data, is the essence of digital twinning [ Wright and Davidson, 2020 ] , which aims to derive adaptive models that can cope with the evolution of physical systems. Twinning methods vary in model complexity and tuning procedure, depending on whether LOR or ROMs are used. Examples of LOR-based twinning are provided by Schena et al. [2024] and Branlard et al. [2024] . Schena