the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A digital-twin solution for floating offshore wind turbines validated using a full-scale prototype
Emmanuel Branlard
Jason Jonkman
Cameron Brown
Jiatian Zang
Abstract. In this work, we implement, verify, and validate a physics-based digital twin solution applied to a floating offshore wind turbine. The digital twin is validated using measurement data from the full-scale TetraSpar prototype. We focus on the estimation of the aerodynamic loads, wind speed, and section loads along the tower, with the aim at estimating the fatigue life-time of the tower. Our digital twin solution integrates: 1) a Kalman filter to estimate the structural states based on a linear model of the structure and measurements from the turbine, 2) an aerodynamic estimator, and 3) a physics-based virtual sensing procedure to obtain the loads along the tower. The digital twin relies on a set of measurements that are expected to be available on any existing wind turbine (power, pitch, rotor speed, and tower acceleration), and motion sensors that are likely to be standard measurements for a floating platform (inclinometers and GPS sensors). We explore two different pathways to obtain physics-based models: a suite of dedicated Python tools implemented as part of this work, or the OpenFAST linearization feature. In our final version of the digital twin, we use components from both approaches. We perform different numerical experiments to verify the individual models of the digital twin. In this simulation realm, we obtain estimated damage equivalent loads with an accuracy of the order of 5 % to 10 %. When comparing the digital twin estimations with the measurements from the TetraSpar prototype, the errors increased to 10 %–15 % on average. Overall, the accuracy of the results appears promising and demonstrates the possibility to use digital twin solutions to estimate fatigue loads on floating offshore wind turbines. A natural continuation of this work would be to implement the monitoring and diagnostics aspect of the digital twin, to inform operation and maintenance decisions. The digital twin solution is provided with examples as part of an open-source repository.
- Preprint
(2589 KB) - Metadata XML
- BibTeX
- EndNote
Emmanuel Branlard et al.
Status: closed
-
RC1: 'Comment on wes-2023-50', Anonymous Referee #1, 29 May 2023
General comments
The paper addresses the “construction” of a digital twin for floating wind turbines that is verified against numerical simulations in a state-of-the-art offshore code and validated against real-world data. At present, the first commercial floating wind turbines are being installed and digital tools as the one presented in this article can be fundamental for their success. Therefore, the article is relevant to the international scientific community and falls within the scope of WES.
In general, the objectives and hypotheses of the article are clear. The methods are valid and can be reproduced by the reader, also thanks to the fact that the source code of the digital twin will be shared in a GitHub repository once the article will be published. The analyses presented by authors are valid and backed up by widely used simulations tools and previous research. Results are outlined in a clear way and the article is generally well structured in all its parts (text, figures, mathematics).
For all this reasons, the article deserves to be considered for publication in WES after being revised.
As a general comment, I see the tools presented in this article are derived from those developed in previous works about digital twins for bottom-fixed turbines. I suggest the authors to point out in the introduction the additional challenges of making a digital tool for a wind turbine with floating foundations. It may sound trivial, but I would explain why one cannot simply use tools derived for bottom-fixed turbines and where these are inaccurate.
Specific comments are reported below, and technical corrections are in the attached document.
Specific comments
- 102-103. The OpenFAST model is complemented with a closed-loop controller: is it based on a reference controller like ROSCO, are you allowed to give the reader some information about it? The controller parameters are tuned so that OpenFAST simulations match SCADA data for selected operating conditions. Can you add something about the controller tuning?
- Can you say something more about the assumption of rotational symmetry of the platform? How do you think it might affect your study and its results?
- 137-138. The capabilities of the Python code are not clear. If important for the rest of the study, I would explain better how the Python tools work and how it carries out the operations listed at lines 137-138
- 139-147. This paragraph does not convey adequately the difference between OpenFAST and WELIB. I understand that: 1) WELIB is somehow simpler (and less accurate?) than OpenFAST for nonlinear simulations 2) WELIB is easier to linearize and since it is based on an analytical approach it provides better understanding of the wind turbine physics. I suggest the authors to work a bit on this paragraph to help the reader understand the complementarity of the two simulation tools with respect to the objectives of this study.
- 146, “to obtain analytical linear models and gain physical intuitiveness on the model”. From this sentence it's not clear the difference between a linear model obtained in WELIB and in OpenFAST. In the previous sentence it is said that WELIB can perform nonlinear simulations, so the difference for the linearized model is in how the model is obtained. I think you should briefly mention this issue and introduce what is explained in the next sections.
- 169, “in two different ways”. It's not clear the added value of performing the linearization with Python and the HydroDyn module. I suggest the authors to focus on one of the two, or clarify why it is important to show in this study the results obtained with the two methodologies.
- 180-181. Not clear what is Q0. I think you should rember that the degress of freedom of Mh are the platform motions which are only a subset of q. You should also recall the dimensions of Q0 and its structure.
- 3.3 Verification of the linear models. I suggest introducing this paragraph explaining why you used free decay simulations for verification of the linear models and no other types of simulation. Moreover, which is the purpose of simulations without hydrostatic restoring?
- 383, “adequate filtering of the input measurements can be used to tune the energy content at high frequencies”. Which is the impact of the frequency content above 1 Hz on fatigue loads? How meaningful it is compared to low-frequency harmonics? If you think they could be interesting, I suggest you add results of DEL obtained based on low-pass filtered signals.
- 421-422, “this is a crude first-order approximation”. Can you add some words about what you are neglecting using the tower top load cell?
- 428, “… or that the state estimator is failing”. Do you think that sensor issues can explain the different offset?
- 450, “gain-scheduling of the linear model, using nonlinear models”. You should briefly explain how these techniques can improve the results you already got.
Technical corrections
See attached document.
-
RC2: 'Comment on wes-2023-50', Anonymous Referee #2, 06 Jun 2023
Digital twins for floating wind turbines is a topic of interest lately within the wind energy community. This paper addresses this subject in a clear and elegant way that is easy to follow. It starts by introducing the building blocks separately before combining everything together to obtain the digital twin tool. Afterwards, measurements from numerical simulations and full-scale data of the TetraSpar prototype are used for the validation of the tool. Not only that the tool is validated, but also, every component forming the digital twin is verified before validating the whole thing.
The paper reveals the potential gains the digital twins have to offer, which further promotes the wind energy scientific community. The methods and analyses performed are clear and valid, the scientific approach is clearly explained, which helps with the repeatability of the work. There is a good balance between the methods and the results with proper referencing.
The results are well presented and properly discussed till arriving at a well-drawn conclusion summarizing the work done. The abstract efficiently conveys the main message of the paper.
Accordingly, the paper meets the criteria of the journal and qualifies for publication after revision.
I see my fellow referee have many similar comments to mine. So, for brevity, I will add a few more:
- Can you explain why you used a model of 8 DOFs in particular? Why not more?
- You mentioned that controller tuning was performed to match the SCADA data for OpenFAST simulations. What about the control system in WELIB? Does it use a wrapper like pHydrodyn but for ServoDyn? Can you elaborate on that a bit more?
- Lines 196-197, you mentioned that there exists a drift in the sway and roll responses, but you did not explain the reason for its existence. Can it be due to neglecting the hydrodynamic viscous term? Can it be the transformation matrix?
- In Figure 4, there is an agreement in the peaks of the linear and nonlinear OpenFAST, while WELIB does not. Can you say something about that?
- Line 389, can you cite a reference for the equations?
- Line 451-452, "adding a model to account for wave excitation forces", can you elaborate a bit more what do you mean by this sentence? what would be the difference between this model and HydroDyn/ pHydrodyn?
A final careful read should be done to brush off any spelling or grammatical mistakes.
Citation: https://doi.org/10.5194/wes-2023-50-RC2 -
RC3: 'Comment on wes-2023-50', Anonymous Referee #3, 07 Jun 2023
This research focuses on the implementation of a physics-based digital twin solution for a floating offshore wind turbine, specifically the TetraSpar prototype. The digital twin aims to estimate aerodynamic loads, wind speed, and section loads along the tower to assess the fatigue lifetime of the structure. The solution incorporates a Kalman filter for structural state estimation, an aerodynamic estimator, and a physics-based virtual sensing procedure to obtain loads along the tower. The digital twin utilizes measurements commonly available on wind turbines and motion sensors typically used for floating platforms.
Two different approaches are explored to develop physics-based models: dedicated Python tools and the OpenFAST linearization feature, with a combination of both approaches in the final digital twin version. The OpenFAST linearization is for estimation of system states, and WELIB is used for sectional load calculations with the calculated states. Numerical experiments are conducted to verify the individual models, achieving estimated damage equivalent loads with an accuracy of 5% to 10% in simulation. When compared to measurements from the TetraSpar prototype, the errors increased to an average of 10%-15%. The results demonstrate promising accuracy and highlight the potential of digital twin solutions in estimating fatigue loads on floating offshore wind turbines.
Given the rise in popularity of digital twins for estimation of long term damage, I believe this article is fit for publication with minor adjustments and clarifications and will serve its purpose in continuing the progress of efficient digital twins for floating wind turbines. For some general comments, I think the importance of computationally efficient digital twins should be highlighted and the computation time for each model should be evaluated, or explained. Does OpenFAST compute fast enough for real time estimation? If it doesn’t, is it possible to assume a significant increase in computational power would allow it to be run in real time? There are also questions on the hydrodynamic modeling of the floating platform. There seems to be an exclusion of the radiation damping of the platform. I assume this is due to the diameters relation to encountered wave lengths, but perhaps it would be helpful to explain this. If the platform is being modeled with Morrison’s element, a brief discussion of this would be enlightening as well. As a final statement about the hydrodynamics, what would be the effect of including wave excitation into the linearized model inputs, and why was this not used to begin with?
Comments/Questions:
- 29-30 : You mention the use of data driven models for digital twins, what are the strengths and weaknesses of these data driven models when compared to physics based? If any of these weaknesses led to the development of the physics based digital twin, it would be interesting to know.
- 65-66: This is an important statement, as there are multiple definitions for digital twins so clarification is needed to the definition of digital twin in the scope of this article.
- 72: The goal is to use currently available measurements on the turbine (SCADA), but are there significant effects to adding more measurements? If the effect of adding more measurements to the physical turbine is positive enough, it may warrant a new addition of sensors to turbines in the field. The cost to add more sensors might be allowable if it can reduce the long term cost of the project with O&M decisions coming from digital twin estimations.
- 102: What values were tuned based on measurements? The method for calculating excitation forces in the OpenFAST simulation is not apparent also.
- 139: You mention WELIB has some shortcomings, but don’t mention what they are. These options could lead to discrepancies when comparing models. Are these limited options for structural dynamics, hydrodynamics, control?
- 169-175: It is not explicitly stated which method was used.
- 192-194: This seems like a powerful use of the state-estimator, that is, to neglect forces and allow the estimator to account for the absence of these forces. To what degree is this valid?
- 322: You state the linear model is valid while close to the operating point. What is this region for a full scale turbine?
- 356: The WELIB library is chosen for calculation of the sectional loads, but the full digital twin uses both WELIB and OpenFAST linearization. Is the job of OpenFAST linearized model to calculate the states of the system, while the WELIB calculates the section loads? If this is not the case, clarification is needed on the responsibility of each model.
Citation: https://doi.org/10.5194/wes-2023-50-RC3 -
AC1: 'Comment on wes-2023-50 - Response to reviewers', Emmanuel Branlard, 16 Jun 2023
Dear reviewers
Thank you so much for your time and effort in reviewing our paper. Please find our answers to your
comments and the corresponding updates to the manuscript in the document attached. We hope to have addressed most of
your comments which helped us improve the manuscript, and we will be happy to continue the discussion.
At the end of the PDF, you can find the differences made to the manuscript with highlighted colors.
Emmanuel and co-authors
Status: closed
-
RC1: 'Comment on wes-2023-50', Anonymous Referee #1, 29 May 2023
General comments
The paper addresses the “construction” of a digital twin for floating wind turbines that is verified against numerical simulations in a state-of-the-art offshore code and validated against real-world data. At present, the first commercial floating wind turbines are being installed and digital tools as the one presented in this article can be fundamental for their success. Therefore, the article is relevant to the international scientific community and falls within the scope of WES.
In general, the objectives and hypotheses of the article are clear. The methods are valid and can be reproduced by the reader, also thanks to the fact that the source code of the digital twin will be shared in a GitHub repository once the article will be published. The analyses presented by authors are valid and backed up by widely used simulations tools and previous research. Results are outlined in a clear way and the article is generally well structured in all its parts (text, figures, mathematics).
For all this reasons, the article deserves to be considered for publication in WES after being revised.
As a general comment, I see the tools presented in this article are derived from those developed in previous works about digital twins for bottom-fixed turbines. I suggest the authors to point out in the introduction the additional challenges of making a digital tool for a wind turbine with floating foundations. It may sound trivial, but I would explain why one cannot simply use tools derived for bottom-fixed turbines and where these are inaccurate.
Specific comments are reported below, and technical corrections are in the attached document.
Specific comments
- 102-103. The OpenFAST model is complemented with a closed-loop controller: is it based on a reference controller like ROSCO, are you allowed to give the reader some information about it? The controller parameters are tuned so that OpenFAST simulations match SCADA data for selected operating conditions. Can you add something about the controller tuning?
- Can you say something more about the assumption of rotational symmetry of the platform? How do you think it might affect your study and its results?
- 137-138. The capabilities of the Python code are not clear. If important for the rest of the study, I would explain better how the Python tools work and how it carries out the operations listed at lines 137-138
- 139-147. This paragraph does not convey adequately the difference between OpenFAST and WELIB. I understand that: 1) WELIB is somehow simpler (and less accurate?) than OpenFAST for nonlinear simulations 2) WELIB is easier to linearize and since it is based on an analytical approach it provides better understanding of the wind turbine physics. I suggest the authors to work a bit on this paragraph to help the reader understand the complementarity of the two simulation tools with respect to the objectives of this study.
- 146, “to obtain analytical linear models and gain physical intuitiveness on the model”. From this sentence it's not clear the difference between a linear model obtained in WELIB and in OpenFAST. In the previous sentence it is said that WELIB can perform nonlinear simulations, so the difference for the linearized model is in how the model is obtained. I think you should briefly mention this issue and introduce what is explained in the next sections.
- 169, “in two different ways”. It's not clear the added value of performing the linearization with Python and the HydroDyn module. I suggest the authors to focus on one of the two, or clarify why it is important to show in this study the results obtained with the two methodologies.
- 180-181. Not clear what is Q0. I think you should rember that the degress of freedom of Mh are the platform motions which are only a subset of q. You should also recall the dimensions of Q0 and its structure.
- 3.3 Verification of the linear models. I suggest introducing this paragraph explaining why you used free decay simulations for verification of the linear models and no other types of simulation. Moreover, which is the purpose of simulations without hydrostatic restoring?
- 383, “adequate filtering of the input measurements can be used to tune the energy content at high frequencies”. Which is the impact of the frequency content above 1 Hz on fatigue loads? How meaningful it is compared to low-frequency harmonics? If you think they could be interesting, I suggest you add results of DEL obtained based on low-pass filtered signals.
- 421-422, “this is a crude first-order approximation”. Can you add some words about what you are neglecting using the tower top load cell?
- 428, “… or that the state estimator is failing”. Do you think that sensor issues can explain the different offset?
- 450, “gain-scheduling of the linear model, using nonlinear models”. You should briefly explain how these techniques can improve the results you already got.
Technical corrections
See attached document.
-
RC2: 'Comment on wes-2023-50', Anonymous Referee #2, 06 Jun 2023
Digital twins for floating wind turbines is a topic of interest lately within the wind energy community. This paper addresses this subject in a clear and elegant way that is easy to follow. It starts by introducing the building blocks separately before combining everything together to obtain the digital twin tool. Afterwards, measurements from numerical simulations and full-scale data of the TetraSpar prototype are used for the validation of the tool. Not only that the tool is validated, but also, every component forming the digital twin is verified before validating the whole thing.
The paper reveals the potential gains the digital twins have to offer, which further promotes the wind energy scientific community. The methods and analyses performed are clear and valid, the scientific approach is clearly explained, which helps with the repeatability of the work. There is a good balance between the methods and the results with proper referencing.
The results are well presented and properly discussed till arriving at a well-drawn conclusion summarizing the work done. The abstract efficiently conveys the main message of the paper.
Accordingly, the paper meets the criteria of the journal and qualifies for publication after revision.
I see my fellow referee have many similar comments to mine. So, for brevity, I will add a few more:
- Can you explain why you used a model of 8 DOFs in particular? Why not more?
- You mentioned that controller tuning was performed to match the SCADA data for OpenFAST simulations. What about the control system in WELIB? Does it use a wrapper like pHydrodyn but for ServoDyn? Can you elaborate on that a bit more?
- Lines 196-197, you mentioned that there exists a drift in the sway and roll responses, but you did not explain the reason for its existence. Can it be due to neglecting the hydrodynamic viscous term? Can it be the transformation matrix?
- In Figure 4, there is an agreement in the peaks of the linear and nonlinear OpenFAST, while WELIB does not. Can you say something about that?
- Line 389, can you cite a reference for the equations?
- Line 451-452, "adding a model to account for wave excitation forces", can you elaborate a bit more what do you mean by this sentence? what would be the difference between this model and HydroDyn/ pHydrodyn?
A final careful read should be done to brush off any spelling or grammatical mistakes.
Citation: https://doi.org/10.5194/wes-2023-50-RC2 -
RC3: 'Comment on wes-2023-50', Anonymous Referee #3, 07 Jun 2023
This research focuses on the implementation of a physics-based digital twin solution for a floating offshore wind turbine, specifically the TetraSpar prototype. The digital twin aims to estimate aerodynamic loads, wind speed, and section loads along the tower to assess the fatigue lifetime of the structure. The solution incorporates a Kalman filter for structural state estimation, an aerodynamic estimator, and a physics-based virtual sensing procedure to obtain loads along the tower. The digital twin utilizes measurements commonly available on wind turbines and motion sensors typically used for floating platforms.
Two different approaches are explored to develop physics-based models: dedicated Python tools and the OpenFAST linearization feature, with a combination of both approaches in the final digital twin version. The OpenFAST linearization is for estimation of system states, and WELIB is used for sectional load calculations with the calculated states. Numerical experiments are conducted to verify the individual models, achieving estimated damage equivalent loads with an accuracy of 5% to 10% in simulation. When compared to measurements from the TetraSpar prototype, the errors increased to an average of 10%-15%. The results demonstrate promising accuracy and highlight the potential of digital twin solutions in estimating fatigue loads on floating offshore wind turbines.
Given the rise in popularity of digital twins for estimation of long term damage, I believe this article is fit for publication with minor adjustments and clarifications and will serve its purpose in continuing the progress of efficient digital twins for floating wind turbines. For some general comments, I think the importance of computationally efficient digital twins should be highlighted and the computation time for each model should be evaluated, or explained. Does OpenFAST compute fast enough for real time estimation? If it doesn’t, is it possible to assume a significant increase in computational power would allow it to be run in real time? There are also questions on the hydrodynamic modeling of the floating platform. There seems to be an exclusion of the radiation damping of the platform. I assume this is due to the diameters relation to encountered wave lengths, but perhaps it would be helpful to explain this. If the platform is being modeled with Morrison’s element, a brief discussion of this would be enlightening as well. As a final statement about the hydrodynamics, what would be the effect of including wave excitation into the linearized model inputs, and why was this not used to begin with?
Comments/Questions:
- 29-30 : You mention the use of data driven models for digital twins, what are the strengths and weaknesses of these data driven models when compared to physics based? If any of these weaknesses led to the development of the physics based digital twin, it would be interesting to know.
- 65-66: This is an important statement, as there are multiple definitions for digital twins so clarification is needed to the definition of digital twin in the scope of this article.
- 72: The goal is to use currently available measurements on the turbine (SCADA), but are there significant effects to adding more measurements? If the effect of adding more measurements to the physical turbine is positive enough, it may warrant a new addition of sensors to turbines in the field. The cost to add more sensors might be allowable if it can reduce the long term cost of the project with O&M decisions coming from digital twin estimations.
- 102: What values were tuned based on measurements? The method for calculating excitation forces in the OpenFAST simulation is not apparent also.
- 139: You mention WELIB has some shortcomings, but don’t mention what they are. These options could lead to discrepancies when comparing models. Are these limited options for structural dynamics, hydrodynamics, control?
- 169-175: It is not explicitly stated which method was used.
- 192-194: This seems like a powerful use of the state-estimator, that is, to neglect forces and allow the estimator to account for the absence of these forces. To what degree is this valid?
- 322: You state the linear model is valid while close to the operating point. What is this region for a full scale turbine?
- 356: The WELIB library is chosen for calculation of the sectional loads, but the full digital twin uses both WELIB and OpenFAST linearization. Is the job of OpenFAST linearized model to calculate the states of the system, while the WELIB calculates the section loads? If this is not the case, clarification is needed on the responsibility of each model.
Citation: https://doi.org/10.5194/wes-2023-50-RC3 -
AC1: 'Comment on wes-2023-50 - Response to reviewers', Emmanuel Branlard, 16 Jun 2023
Dear reviewers
Thank you so much for your time and effort in reviewing our paper. Please find our answers to your
comments and the corresponding updates to the manuscript in the document attached. We hope to have addressed most of
your comments which helped us improve the manuscript, and we will be happy to continue the discussion.
At the end of the PDF, you can find the differences made to the manuscript with highlighted colors.
Emmanuel and co-authors
Emmanuel Branlard et al.
Emmanuel Branlard et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
944 | 344 | 26 | 1,314 | 17 | 17 |
- HTML: 944
- PDF: 344
- XML: 26
- Total: 1,314
- BibTeX: 17
- EndNote: 17
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1