Minkowski logarithmic error: A physics-informed neural network approach for wind turbine lifetime assessment

Santos, Francisco de Nolasco; D’Antuono, Pietro; Noppe, Nymfa; Weijtjens, Wout; Devriendt, Christof

doi:10.14428/esann/2022.es2022-16

Cited by 4 publications

(5 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this case, the hyperparameters

x \in scriptX

were

h \in scriptH = false{1, \dots, 5 false}

hidden layers,

n \in scriptN = false{32,64, 96, \dots, 512 false}

neurons,

a \in scriptA = false{ReLU, GELU, SELU false}

activation function types, 63

d \in scriptD = false{0,0 . 1,0 . 2,0 . 3 false}

dropout rate and

o \in scriptO = false{1 \cdot e^{- 2}, 1 \cdot e^{- 3}, 1 \cdot e^{- 4} false}

learning rate of the optimizer (Adam 64 ). The monitored loss function was the Minkowski logarithmic error (MLE) introduced in de N Santos et al 65 This function attempts to guide the neural network learning by including physical knowledge specific to the problem at hand, in a so‐called physics‐guided ML approach (

bold-italicΦ - ML

). It is described by Equation (), with

m = 5

, and

normalY, normalŶ

, the measured and predicted vectors.…”

Section: Methodsmentioning

confidence: 99%

“…By using probabilistic models and acquisition functions, Bayesian modeling captures the relationship between hyperparameters and model performance, with fewer computational resources and less manual effort. Thus, for a differentiable loss function LðxÞ, we have x * ¼ argmin 65 This function attempts to guide the neural network learning by including physical knowledge specific to the problem at hand, in a so-called physics-guided ML approach (Φ À ML). It is described by Equation ( 3), with m ¼ 5, and Y, Ŷ, the measured and predicted vectors.…”

Section: Model Trainingmentioning

confidence: 99%

See 1 more Smart Citation

Farm‐wide interface fatigue loads estimation: A data‐driven approach based on accelerometers

de N Santos,

Noppe,

Weijtjens

et al. 2024

Wind Energy

Self Cite

View full text Add to dashboard Cite

Fatigue has become a major consideration factor in modern offshore wind farms as optimized design codes, and a lack of lifetime reserve has made continuous fatigue life monitoring become an operational concern. In this contribution, we discuss a data‐driven methodology for farm‐wide tower‐transition piece fatigue load estimation. We specifically debate the employment of this methodology in a real‐world farm‐wide setting and the implications of continuous monitoring. With reliable nacelle‐installed accelerometer data at all locations, along with the customary 10‐min supervisory control and data acquisition (SCADA) statistics and three strain gauge‐instrumented 'fleet‐leaders', we discuss the value of two distinct approaches: use of either fleet‐leader or population‐based data for training a physics‐guided neural network model with a built‐in conservative bias, with the latter taking precedence. In the context of continuous monitoring, we touch on the importance of data imputation, working under the assumption that if data are missing, then its fatigue loads should be modeled as under idling. With this knowledge at hand, we analyzed the errors of the trained model over a period of 9 months, with monthly accumulated errors always kept below . A particular focus was given to performance under high loads, where higher errors were found. The cause for this error was identified as being inherent to the use of 10‐min statistics, but mitigation strategies have been identified. Finally, the farm‐wide results are presented on fatigue load estimation, which allowed to identify outliers, whose behavior we correlated with the operational conditions. Finally, the continuous data‐driven, population‐based approach here presented can serve as a springboard for further lifetime‐based decision‐making.

show abstract

“…In this case, the hyperparameters

x \in scriptX

were

h \in scriptH = false{1, \dots, 5 false}

hidden layers,

n \in scriptN = false{32,64, 96, \dots, 512 false}

neurons,

a \in scriptA = false{ReLU, GELU, SELU false}

activation function types, 63

d \in scriptD = false{0,0 . 1,0 . 2,0 . 3 false}

dropout rate and

o \in scriptO = false{1 \cdot e^{- 2}, 1 \cdot e^{- 3}, 1 \cdot e^{- 4} false}

bold-italicΦ - ML

). It is described by Equation (), with

m = 5

, and

normalY, normalŶ

, the measured and predicted vectors.…”

Section: Methodsmentioning

confidence: 99%

Section: Model Trainingmentioning

confidence: 99%

Farm‐wide interface fatigue loads estimation: A data‐driven approach based on accelerometers

de N Santos,

Noppe,

Weijtjens

et al. 2024

Wind Energy

Self Cite

View full text Add to dashboard Cite

show abstract

“…In this contribution we attempt to guide the neural network learning by including physical knowledge specific to the problem at hand, in a so-called physics-guided machine learning approach (Φ−ML). Specifically, we include the Minkowski logarithmic error (MLE) introduced in [23], described by Equation 3 for m = 5 and Y, Ŷ, the measured and predicted vectors. It is based on the L p norm and the logarithm function, and attempts to prioritise long-term DEL performance (DEL L T ), whilst maintaining ten-minute level prediction accuracy.…”

Section: Methodology and Datamentioning

confidence: 99%

Fleetwide Interface Fatigue Load Prediction of an Operational Offshore Wind Farm Using a Single Accelerometer and Scada Data

SANTOS,

NOPPE,

WEIJTJENS

et al. 2023

Proceedings of the 14th International Workshop on Structural Health Monitoring

View full text Add to dashboard Cite

The operational life of offshore wind turbines is in part driven by the fatigue life of key structural components, such as the substructure. In recent years, fatigue life management of operational assets has become evermore important, as older farms are closing on their design lifetime and newer farms are designed with tighter margins. To support decisions on fatigue lifetime, it is advantageous to monitor the fatigue progression in these structures through SHM. However, a full instrumentation of every asset in a farm to assess the fatigue life of the substructure is considered economically infeasible. The drive for SHM in wind has been accompanied by the increased availability of intelligent data-driven methodologies which have attempted to provide the same information without the need for additional hardware. One such cost-effective approach, as described in [1], uses the available supervisory control and data acquisition (SCADA) systems, coupled with acceleration measurements to predict the fatigue life of an offshore wind turbine. The use of acceleration measurement data has been proven critical for capturing the complex dynamics of offshore wind turbines. In this contribution, we present the results of said data-driven approach using SCADA and Internet of Things (IoT) accelerometer installed at nacelle-level to monitor the fatigue life for the entirety of a real-world offshore wind farm comprised of 23 turbines, with a specific focus on long-term damage equivalent fatigue loads (DEL) estimation [2]. The availability of acceleration measurements for all locations in particular, is fundamental, as to cover all possible differences in natural frequencies between turbines would be nearly impossible if solely relying on wave and tidal data. To achieve this goal, a neural network architecture is used, enhanced by physics-informed learning focused on long-term estimation, trained and validated on so-called fleet-leader (three turbines instrumented with strain gauges, which provide the ground truth). In this study, we pay special attention to the farm-wide validation, cross-validation and extrapolation of these models as well as the performance for different operational conditions. Finally, this study is undertook for a real-world instrumentation setup, unparalleled in its scale, and can thus be indicative of future trends for SHM in offshore wind: farm-wide instrumentation and monitoring of structural health based on acceleration measurements, enabling a greater trustworthiness on reliability and durability estimation over the serviceable lifetime.

show abstract

“…To generate the final graphs, this information is encoded for the random layouts of Section 3.1 using Delaunay's triangulation to establish the connectivity. The full data set can be found in [34].…”

Section: Pywake Simulationsmentioning

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

Minkowski logarithmic error: A physics-informed neural network approach for wind turbine lifetime assessment

Cited by 4 publications

References 3 publications

Farm‐wide interface fatigue loads estimation: A data‐driven approach based on accelerometers

Farm‐wide interface fatigue loads estimation: A data‐driven approach based on accelerometers

Fleetwide Interface Fatigue Load Prediction of an Operational Offshore Wind Farm Using a Single Accelerometer and Scada Data

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

Contact Info

Product

Resources

About