Forecasting the Dynamic Response of Rotating Machinery under Sudden Load Changes
Round 1
Reviewer 1 Report
The paper
‘Forecasting the dynamic response of rotating machinery under sudden load changes
By
Juan Carlos Jauregui-Correa
Investigates non-stationary signals recorded from a wind turbine under gusty winds, i.e. with sudden amplitude changes.
Specifically, the ARIMA method is applied to the vibrational data.
Overall, The topic is of good interest to researchers and practitioners in the field of vibration-based condition monitoring. However, some issues are listed here below and should be addressed to achieve full acceptance.
1. The main issue from this Reviewer regards the novelty and originality of the paper’s content. As a complication of the well-known ARMA procedure, the Autoregressive integrated moving average (ARIMA) method is known and used since the 80s. Here, the paper seems to only revolve around an application of this well-known method to a specific dataset.
2. Before considering applications to real wind turbines, the method is preliminarily verified on a numerically simulated damped oscillation (eq. 5). This example is quite trivial and does not add very much to the rest of the paper. A more compelling case study, more similar to the true response of a wind turbine to sudden loads, would be much better.
3. Related to the previous comment, also adding different cases of artificially-added white Gaussian noise would help in understanding the robustness of ARIMA to measurement noise.
4. The experimental dataset must be described in much, much more detail, in a dedicated Section.
5. The Discussion section is too synthetic and should be expanded to include a more detailed discussion about the main results and, especially, the novelty of the obtained outcomes.
6. Conclusions. Reporting that “In this case, the ¨p¨ parameter that provided the best results was ¼ of the database; the “D” parameter 373 must be one, and the “q” was set at fife” is of limited interest because it is a very case specific result. The comments should be generalised to fit the application also for datasets different from the specific one used here.
7. Related to the above comment, a sensibility analysis would be interesting. The Author considered 5 cases: Case 1 (p=1, D=2, and q=1), Case 2 (p=10, d=1, q=1), Case 3 (p=10, d=1, q=2), Case 4 (p=10, D=2, q=5), Case 5 (p=20, D=1, q=5). However, this should be done more systematically, perhaps showing in a 3D graph the accuracy of each combination, varying the three independent variables (p,d,q) one by one.
8. Finally, for being a paper specifically focused on Wind Turbine monitoring, its state-of-the-art review is a bit too generic and not very updated. Recent advancements, such as https://doi.org/10.3390/app12031059 and other works on the vibration-based or Non-Destructive Techniques (NDTs)-based Condition Monitoring of wind turbines should be added.
The English is overall fine and the writing of the paper is good and easy readable.
Author Response
- Artificial noise was added to the simulated function, and the predictions were analyzed for different noise levels.
- The experiment was better described, figures of the wind turbine and the sensor's location were included, and the data preprocessing was described.
- The procedure for determining p and q parameters was included, based on analyzing the variance between predictions and original data. A contour plot is included.
- The recommended reference was included.
Reviewer 2 Report
The well written manuscript provides original technical content on the application of the Auto-Regressive Integrated Moving Average (ARIMA) algorithm to forecast future events with an application related to analyzing the amplitude of vibration for the bearing in a gear box of a wind turbine subjected to sudden (intermittent) wind gusts.The review of the prior art is sound and provides criticism of the methods to predict future events (failures perhaps).
The discussion explaining the ARIMA method is clear and concise but gives no details on a logical method to determine the parameters p,D,q, and whose "reasonable" selection is fundamental for realistic predictions.
The example application shows the procedure to select sound parameters (p,D,q) remains rudimentary, cursory at best ("different p.D,q values were tested").
There is no logic on the selections and the end results may not be applicable to other problems (similar data or not).
The reviewer would prefer (highly recommends) the authors use ARIMA to develop the model with the (say) first 40 events or observations and then produce predictions for the next 40 instances (41-80) for which the authors obviously know the results. Case 5 (p=20, D=1, q=5) is the best to forecast the future behavior, alas there is no data to actually ascertain the predictions are accurate. Please note that the major aspect of failure or damage of a component is only referred on the passing since the study does not state limits for the amplitude of vibration of the said component.
In the conclusion section, please correct "fife" to "five" and check if p=1/4 data is correct (20 out of 80 observations).
Lastly, the reviewer recommends the authors to introduce the acronym ARIMA in the abstract and define it the first instance of its appearance in the main text.
Author Response
Thank you for your recommendation to compare predictions with actual data. A new section is included discussing the predictions based on 50% of the data and comparing them with the remaining 50% of the measurements.
The misspelling was corrected, and the abstract included the ARIMA acronym.
Round 2
Reviewer 1 Report
All the comments have been addressed positively. Thus, this Reviewer suggests the acceptance of the submitted paper and its publication, after careful grammar checking and proofreading.
The English of the paper is acceptable
