Graph Neural Networks for Aerodynamic Flow Reconstruction from Sparse Sensing

Duthé, Gregory; Abdallah, Imad; Barber, Sarah; Chatzi, Eleni

doi:10.48550/arxiv.2301.03228

Cited by 3 publications

(3 citation statements)

References 15 publications

(20 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We use an Encode-Process-Decode GNN architecture [18,19], which ensures that the dimensionality during the message-passing process is much larger than the physical parameter space, thus increasing the expressivity of the GNN. We display in Figure 1 a graphical overview of this GNN architecture.…”

Section: Architecture Of the Gnn Solutionmentioning

confidence: 99%

Multivariate prediction on wake-affected wind turbines using graph neural networks

Santos,

Duthé,

Abdallah

et al. 2024

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

Modern wind turbines are large and slender dynamical structures with a fatigue loading profile of complex nature. The guarantee of their structural integrity is paramount for materializing cost efficient and more reliable wind energy. The measurement of the global dynamic response and loads of wind turbines is fundamental for achieving this goal. However, an industry-wide, cost-effective direct sensing framework is yet to arise. Moreover, deploying physical sensors and measurement systems on every structural component of interest of a wind turbine induces prohibitive costs in deployment, maintenance and data management. Considering that direct fluid-structure interaction simulations on a farm level are not computationally feasible, the preferred path for structural response estimation on wind farms has been surrogate modelling. Within this landscape, new model architectures have risen in recent years which are able to take into account graph structured data (i.e. non-euclidean data). Wind turbines positioned in a farm, where there is a layout- and topology-dependant interplay of aerodynamic wake affecting the loading profile and power production, lend themselves perfectly to this paradigm. Thus, in this contribution, we introduce the use of graph neural networks (GNN) for layout agnostic saptio-temporal joint modelling of fatigue loads effects, rotor-averaged wind speed and power production on individual turbines of wind farms. To this end, we generate stochastic dependent samples of inflow conditions for wind speed, wind direction, wind shear and nacelle yaw angles. Additionally, wind farm layouts are randomly generated based on different geometric shapes (rectangle, triangle, ellipse and sparse circles) with random parametrization (varying orientations, length/width ratio) for different numbers of turbines and minimal distance (based on the rotor diameter). Both the arbitrary layouts and the random inflow conditions are used as inputs for PyWake, a wind farm simulation tool capable of calculating wind farm flow fields, power and fatigue loads. In our analysis, we develop and compare the performance of the GENeralized Aggregation Networks (GEN), the Graph Attention Networks (GAT) and the Graph Isomorphism Network with Edges (GINE) in their accuracy and ability to generalize their joint predictions for unseen layouts, uncertain inflow conditions and fatigue load estimation on the blade root, tower top and tower base of any wind turbine in the farm. Our results indicate that the GEN layer yields the best performance, followed by GINE, while the GAT layer under-performs and is unable to differentiate between different wake conditions. We further observe that the GAT layer causes a latent space collapse, due to the coupled effect of the manner in which we initialise node features and the way in which its messages are computed.

show abstract

Section: Architecture Of the Gnn Solutionmentioning

confidence: 99%

Multivariate prediction on wake-affected wind turbines using graph neural networks

Santos,

Duthé,

Abdallah

et al. 2024

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

show abstract

“…Machine learning methods have emerged as data-driven general-purpose tools, that can be used for reconstructing fields from sparse sensor measurements. Numerous studies [20][21][22][23][24][25] have demonstrated the potential of neural networks to efficiently interpret and utilize sparse data in various contexts. Fully connected neural networks, as discussed in Erichson et al [20] on shallow architectures, have been successfully trained on benchmark datasets.…”

Section: Introductionmentioning

confidence: 99%

Journey over destination: dynamic sensor placement enhances generalization

Marcato,

Guiltinan,

Viswanathan

et al. 2024

Mach. Learn.: Sci. Technol.

View full text Add to dashboard Cite

Reconstructing complex, high-dimensional global fields from limited data points is a challenge across various scientific and industrial domains. This is particularly important for recovering spatio-temporal fields using sensor data from, for example, laboratory-based scientific experiments, weather forecasting, or drone surveys. Given the prohibitive costs of specialized sensors and the inaccessibility ofcertain regions of the domain, achieving full field coverage is typically not feasible. Therefore, the development of machine learning algorithms trained to reconstruct fields given a limited dataset is of critical importance. In this study, we introduce a generalapproach that employs moving sensors to enhance data exploitation during the training of an attention based neural network, thereby improving field reconstruction. The training of sensor locations is accomplished using an end-to-end workflow, ensuringdifferentiability in the interpolation of field values associated to the sensors, and is simple to implement using differentiable programming. Additionally, we have incorporated a correction mechanism to prevent sensors from entering invalid regions within the domain. We evaluated our method using two distinct datasets; the results show that our approach enhances learning, as evidenced by improved test scores.

show abstract

“…The layout-dependent interaction between wind turbines' aerodynamics, which lends itself to be modelled as a graph, on the one hand, and the ability of graph neural networks to effectively extrapolate [3], on the other, has led GNNs to not only become one of the hottest themes in machine learning, but also being increasingly implemented in wind energy. The vast majority of employments of GNNs in wind energy has been related with two topics: power [4,5,6] prediction tasks and wake loss [7,8,9]. Despite these numerous examples, there is yet to be presented a generalizeable and wind farm layoutagnostic GNN approach that combines predictions on wind deficit due to wake interactions, with the fatigue loads these farm-wide interactions cause.…”

Section: Introductionmentioning

confidence: 99%

Local flow and loads estimation on wake-affected wind turbines using graph neural networks and PyWake

Duthé

Santos

Abdallah

et al. 2023

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

The estimation of the flow conditions throughout a modern wind farm is a relevant albeit complex problem involving several different modelling scales. In this context, recent years have seen the rise of the demand - normally industry-driven - of quick, computationally inexpensive, wind farm flow field simulators upon which further calculations might be undertaken (e.g. power output, fatigue loads, etc.). Naturally, the trustworthiness of such models hinges upon their ability to accurately estimate wakes, under the evident trade-off between computational time and accuracy. One such simulator that has been recently gaining traction within the wind community is PyWake, a Python-based flow solver, with several possible add-ons and modularity options. Based on runs from PyWake, this contribution studies the applicability of graph neural networks (GNN) as a simultaneous wind turbine fleet, wind inflow and loads surrogate. As graphs are discrete mathematical sets of dependent objects, one can see how the layout-dependent interaction between wind turbines’ aerodynamics, from which wake arises, lends itself to be modelled as a graph. Thus, we introduce the use of GNNs for layout-agnostic joint modelling of rotor-averaged wind speed, turbulence intensity, power production and damage equivalent loads (DEL) on individual turbines of wind farms. To this end, probabilistic samples of inflow conditions from a Weibull distribution for wind speed, uniform wind direction and conditional normal distributions of wind shear and nacelle yaw angles are generated as main inflow properties in the numerical simulations. Additionally, arbitrary wind farm layouts are created based on varied geometric shapes with random parametrization (varying orientations, length/width ratio). Both the arbitrary layouts and the random inflow conditions are then used as inputs for PyWake. PyWake’s output, namely, rotor averaged wind speed, turbulence intensity, power production and DELs of each individual wind turbine in the wind farms, are used to train a general GNN model. We elect to implement an Encode-Process-Decode GNN as a graph learning model. We compare the performance of various approaches of the relational dependency representation (edge-forming techniques) amongst the nodes (wind turbines) on the graph, such as Delaunay triangulation, KNN, radius-based methods and fully connected scheme. In our analysis, we evaluate the accuracy of GNNs and its ability to generalize their joint predictions for unseen layouts and varying inflow conditions.

show abstract

Graph Neural Networks for Aerodynamic Flow Reconstruction from Sparse Sensing

Cited by 3 publications

References 15 publications

Multivariate prediction on wake-affected wind turbines using graph neural networks

Multivariate prediction on wake-affected wind turbines using graph neural networks

Journey over destination: dynamic sensor placement enhances generalization

Local flow and loads estimation on wake-affected wind turbines using graph neural networks and PyWake

Contact Info

Product

Resources

About