the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
One-to-one aeroservoelastic validation of operational loads and performance of a 2.8 MW wind turbine model in OpenFAST
Abstract. This paper presents a validation study of the popular aeroservoelastic code suite OpenFAST leveraging weeks of measurements obtained during normal operation of a 2.8 MW land-based wind turbine. Measured wind conditions were used to generate one-to-one turbulent flow fields (i.e., comparing simulation to measurement in 10-minute increments, or bins) through unconstrained and constrained assimilation methods using the kinematic turbulence generators TurbSim and PyConTurb. A total of 253 10-minute bins of normal turbine operation were selected for analysis, and a statistical comparison in terms of performance and loads is presented. We show that successful validation of the model is not strongly dependent on the type of inflow assimilation method used for mean quantities of interest, which have modeling errors generally within 5–10 % of the measurement. The type of inflow assimilation method does have a larger effect on the fatigue predictions for blade-root flapwise and tower-base fore-aft quantities, which surprisingly see larger errors from the assumed higher-fidelity assimilation methods. Further work including improvements to the induction modeling in OpenFAST during high shear, as well as other possible improvements to the aerodynamic, blade, and controller modeling, may offer insight on the origin of the ∼5–40 % overprediction of fatigue for these quantities.
- Preprint
(13510 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on wes-2023-166', Anonymous Referee #1, 06 Feb 2024
General comments:An interesting paper comparing aero-elastic simulations to field data. The novelty is concentrated around different inflow models and their impact on the agreement with measurements. Not being able to disclose information from the manufacturer complicates things and makes some comparisons not very meaningful. For many of the observed differences potential explanations are given without substantiating, it would be good to make the paper less speculative. The figures on the results, from Fig 6 to the end, can be revised to make them easier to read (larger, different line styles, etc.)Specific comments-p3 section 2 p77/78 A description of the test set-up of the experiment including turbine, instrumentation and visuals would help the reader. What is the rated wind wind speed? Part of section 3.1 could be moved here.-p4 section 2.1 AerodynamicsIn relation to the later observed differences and their hypothesized origin it would be good to add some more details-Details of the airfoil data weighting process (% clean/rough?)-Some info on BEM implementation, e.g. considering local / annulus avg momentum-p4 line 105Clarify what is meant by 'tuning' environmental conditions and its relevance for the aerodynamic modeling section.-p4/5 section 2.2.1 and 2.2.2This comparison should seriously be reconsidered in my opinion. What is the value of comparing against another dataset if no detail is given about it?-Numerical values from a GE tool without knowing its background-A lab test without any info on the set-upFurthermore it is not clear what kind of load is under consideration here?-p4 line 115Can a twist difference of 0.5deg be considered small?-p5 Fig.1If this plot still makes sense (see comment above), without values on the axis, also a plot highlighting the % difference would help to aid interpretation.-p5 line 134What was the originally specified damping and how was this modified?-p7 line 169What is the sample frequency and what frequency would be needed?-p8 line 175Not sure which linearized unsteady aero model is referred to from the description in section 2.1?-p9 line 202Clarify what is meant by 'more sophisticated system monitoring and control mode switching'-p11 line 260Quantify what is meant by some vs persistent IPC activity-p13 line 294Not sure what is meant here, if a wind field turbulent box is created, the appropriate tower incident wind velocity could be interpolated from that I suppose?-p17 line 342 and throughout manuscriptNot sure what is meant by Region II and III?-p19 Fig. 13A plot of power coefficient Cp would allow to zoom in better on differences irrespective of wind speed-p19 line 359As mentioned, power differences can arise from a large nr of reasons. Is there a particular reason to suspect the ground effect here, e.g. looking at past comparisons to field data? Acknowledging the corresponding trend differences in rpm (fig 11, below rated, ) and pitch (fig. 11 above rated), would that help to substantiate further?-p20 1st paragraph, overprediction of flapwise fatigue loadsIt is hypothesized that differences arise from the BEM model in OpenFAST, in agreement with previous literature. Would it be worthwhile to compare against results from a different model like vortex wake for a particular case under investigation to substantiate this further?-p20/21 line 407/408Consider to reformulate 'changes to the relative magnitude of the fluctuating component of the velocity (i.e. stretching of turbulence structures) due to the induction field' for clarity. To my understanding the publication of Branlard (https://doi.org/10.1016/j.jweia.2016.01.002.) seems in contradiction to this observation, suggest clarification.-p21 line 412Would the proximity of blade frequencies to excitation frequencies be important to consider as well?-p21 line 414Hypothesizing about the cause for differences is commonplace in a scientific article. Suggested to be careful with pinpointing these to a numeric value without argumentation.-p22 Fig. 15 Acknowledging the atmospheric relation between TI and shear, can we exclude bias here? Can we plot the trend against shear for a constant TI and vice-versa for a constant shear against TI.-p23 line 446-449Should the recommendation for surface pressure measurements be part of a section dedicated to inflow measurements?-p24 section 5.3Control modeling should not be part of a section about Suggestions for a future experiment.Citation: https://doi.org/
10.5194/wes-2023-166-RC1 -
RC2: 'Comment on wes-2023-166', Anonymous Referee #2, 10 Feb 2024
The studies are aimed at investigating the characteristics of wind turbine in turbulent inflow conditions by simulating the experimentally measured flow field. While the studies are relevant for the community, I do feel the paper is not straightforward and tends to bring in some other focus to the readers, especially at the beginning of the discussion. I believe the main objective and focus shall be the effects of different inflow conditions, and the authors may remove most parts of the paper which do not support the discussion. In fact, extend the discussion of the turbine loads and response. Constraining the timeseries of the wind is an interesting approach which may be used to better understand the turbine behavior. I think a similar story was/is under investigations within the IEA Wind Task 47 consortium. Probably providing some more link toward similar studies/report will be of benefit for the audience. Last but not least, since the wind is constrained, I would expect some time domain comparison of the loads, not just the overall statistics (which can be achieved anyway without constraining in usual manner).
Citation: https://doi.org/10.5194/wes-2023-166-RC2 - RC3: 'Comment on wes-2023-166', Anonymous Referee #3, 08 Mar 2024
-
RC4: 'Comment on wes-2023-166', Anonymous Referee #4, 20 Mar 2024
The paper assesses prediction capabilities of OpenFAST aeroelastic code through comparisons against detailed site measurements of the loads and performance characteristics of a GE wind turbine. The innovative element of the paper is that the authors apply data assimilation techniques with the aim to reconstruct as closely as possible the measured wind field which then serves as input in the aeroelastic analyses.
The work is novel and relevant to the wind community. The paper is well structured and well written. Furthermore, the work constitutes an enormous measurements and simulations effort by the authors. It attempts to provide answers to the open question of how closely the site specific environmental conditions can be replicated in aeroelastic analyses.
In the accompanying pdf the authors can find some specific comments. Below, the key points necessitating further attention or rebuttal are outlined:
- As mentioned in page 14, the advection time from the met tower to the turbine is completely omitted. It is claimed that this time ranges between 20-100s. However, 100s is not a negligible time interval. By doing so, all the benefits from assimilating the actual wind field are compromised. If there is a 100s shift between the simulated wind and the actual wind experienced by the turbine, then the analyzed loads and the resulting DELs correspond to totally different load time series which are only statistically equivalent as in the case of TurbSim simple. Could it be that the better agreement in this case by the assimilated inflow is because of the better representation of the spatial variation of the inflow? In the reviewer’s opinion at least the frozen turbulence approach could have been considered in order to mitigate this effect. This is one key point that deserves further discussion. Perhaps by comparing not only the simulated wind speed time series but also some representative load series.
- The authors claim that the overprediction in the flapwise moment DELs is due the induction modelling in openFAST which fails to correctly predict azimuthal variation of the induction at high shear exponent values. However, even at low shear exponent values the overprediction is substantial (~20%). Another point is that overprediction is equally high both at low and high wind speeds where induction is overall much lower (normally its variations will be lower too). An easy explanation could be that the overprediction is due to the omission of IPC. This viewpoint is refuted thought by the fact that deviation is independent of turbulence intensity. Finally, in the reviewer’s opinion this point remains not properly answered while some of the possible reasons claimed by the authors are not supported by results (e.g. not matching flapwise frequency). I would recommend some further elaboration on that matter or maybe some more concrete explanation (perhaps leveraging TurbSim simple results which seems to do better).
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
385 | 97 | 15 | 497 | 14 | 7 |
- HTML: 385
- PDF: 97
- XML: 15
- Total: 497
- BibTeX: 14
- EndNote: 7
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1