Research on Normal Behavior Models for Status Monitoring and Fault Early Warning of Pitch Motors
Round 1
Reviewer 1 Report
The paper titled “Research on Normal Behavior Models for Status Monitoring and Fault Early Warning of Pitch Motors” presents monitoring process of the pitch motors. Comments are below:
1) The paper discusses the predictive model based on echo state networks. However, the contributions are not clear. The mathematical novelty is weak and the benefits of the proposed method are not explained in a convincing way. In my opinion, the novelty is the weakest part of the paper.
2) Comparison work should be performed to show the effectiveness of the proposed method. In particular, quantitative comparison work should be conducted. Thus, the benefits of the method can be highlighted.
3) The other prediction based method can be reviewed in Introduction section. Several research articles are provided below. Please discuss them in Introduction section, and compare them with your strategy.
[1] Predictive dual-induction machine control using nine-switch inverter for multi-drive systems
[2] Finite state model predictive control for 3× 3 matrix converter based on switching state elimination
[3] Enhanced Robust Deadbeat Predictive Current Control for PMSM Drives
[4] Finite Control Set Model Predictive Control of dual-output four-leg Indirect Matrix Converter under unbalanced load and supply conditions
[5] A predictive control scheme for a dual output indirect matrix converter
[6] Simplified Model Predictive Control for Dual Inverter-Fed Open-Winding Permanent Magnet Synchronous Motor
4) More simulation and experimental results should be provided to support the theoretical study.
5) Is the state assessment model novel? It seems that no theoretical novelty is provided.
6) The quality of Fig. 1 should be improved.
7) Section 2.3 Other Model is weak. Please provide more details about other models that have been reported before. If it is necessary, please provide mathematical background to discuss other techniques. It will be better for readers.
Author Response
Thank you very much for your valuable comments which has helped us to improve the manuscript further. The specific comments are laid out below in italicized font. Our responses are given in standard font. In the revised manuscript, the changes/additions to the manuscript are given in blue text.
Comment 1: The paper discusses the predictive model based on echo state networks. However, the contributions are not clear. The mathematical novelty is weak and the benefits of the proposed method are not explained in a convincing way. In my opinion, the novelty is the weakest part of the paper.
Response: Thank you for this reminder. When revising the paper, we have revised the abstract and the corresponding part in introduction.
Comment 2: Comparison work should be performed to show the effectiveness of the proposed method. In particular, quantitative comparison work should be conducted. Thus, the benefits of the method can be highlighted.
Response: We are grateful to the reviewer for this comment. In the experiments, the proposed method has been compared with the other methods ANFIS, NN and NN+dropout. The results listed in Table 2 show that the proposed ESN-based method performs better. Due to the limited revision time, we are sorry that more experiments weren’t conducted and provided currently.
Comment 3: The other prediction based method can be reviewed in Introduction section. Several research articles are provided below. Please discuss them in Introduction section, and compare them with your strategy.
Response: We really appropriate this comment. In the revised paper, some recommended articles have been cited in the near end of introduction. Meanwhile, they are also added into the list of references. On considering that the topics of these articles are similar, we only selected some of them to include into the paper.
Comment 4: More simulation and experimental results should be provided to support the theoretical study.
Response: Thank you very much for this useful comment. Of course, more simulations and experiments can more convincingly demonstrate the advantage of the proposed ESN-based method. Because of limited time, we cannot finish the related experiments and we are sorry for this.
Comment 5: Is the state assessment model novel? It seems that no theoretical novelty is provided.
Response: We admitted that the state assessment model described in section 3 is not novel. We just employed the commonly used EMVA method due to its simplicity and good performance.
Comment 6: The quality of Fig. 1 should be improved.
Response: Thank you for this comment. Originally, we just copy the figure from the original paper and annotated the reference. At present, we have replotted it by ourselves. We believe that the architecture of an ECN is much clearer.
Comment 7: Section 2.3 Other Model is weak. Please provide more details about other models that have been reported before. If it is necessary, please provide mathematical background to discuss other techniques. It will be better for readers.
Response: We are grateful to the reviewer for this comment. We totally agree with your opinion on section 2.3. On considering that much more contents need to be added if providing the mathematical background of the other methods. Due to page constraints of the paper, we only added some descriptions of these methods. And we provided the closely related literature and interested readers can refer to them for more details.
Reviewer 2 Report
The topic is timely and well presented. However, I have few comments to enhance its overall quality. The abstract has to be rewritten again to reflect the proposed work with some obtained values. The same for the conclusion, it is very short. Please, add a flow chart and more writing to the introduction section to show the complete proposed modeling process and show the novelty of your proposed methodology. Please, repeat the training and testing of your model for at least ten times and get the average. Add a complete section for future work or use the LSTM now to enhance your results. Change the construction of your model to obtain less error as it is not acceptable in some regions. The English writing has to be revised to avoid some typos and grammar mistakes. Thank you so much.Author Response
Thank you very much for your valuable comments which has helped us to improve the manuscript further. The specific comments are laid out below in italicized font. Our responses are given in standard font. In the revised manuscript, the changes/additions to the manuscript are given in blue text.
Comment 1: The topic is timely and well presented. However, I have few comments to enhance its overall quality. The abstract has to be rewritten again to reflect the proposed work with some obtained values. The same for the conclusion, it is very short. Please, add a flow chart and more writing to the introduction section to show the complete proposed modeling process and show the novelty of your proposed methodology. Please, repeat the training and testing of your model for at least ten times and get the average. Add a complete section for future work or use the LSTM now to enhance your results. Change the construction of your model to obtain less error as it is not acceptable in some regions. The English writing has to be revised to avoid some typos and grammar mistakes. Thank you so much.
Response: We are grateful to the reviewer for these helpful comments. The abstract as well as introduction has been revised to reflect the main work in the current work. The conclusion part has also been modified to discuss the strengths and weakness of the proposed method. In experiments, the data have been dealt with the sliding window to obtain the average RAME and MAPE values to compare the methods fairer. In conclusion section, we added more discussions about LSTM and other advanced methods. At the same time, we also corrected some grammatical and typographical errors in the original version of the paper.
Author Response File: 
Author Response.docx
Reviewer 3 Report
I expected more information about the experiment and the problems with the execution and data collection.
I am interested in what was the range of the variables, for example you specify torque and three values are displayed, likewise for other variables. Are the maximum and minimum values of the variables shown there, Table 1.
Table 2. I would like you to tabulate a portion of the predicted temperature as well, not just the error values.
I'm missing a picture of the experiment, where is the temperature measurement position? A little more information about the experiment.
You are talking about the moment of failure of wind turbines, you point out a date that is not recorded on the diagram. Why highlight the date if we don't have a time view of the tracking.
Author Response
Thank you very much for your valuable comments which has helped us to improve the manuscript further. The specific comments are laid out below in italicized font. Our responses are given in standard font. In the revised manuscript, the changes/additions to the manuscript have been highlighted with blue color.
Comment 1: I am interested in what was the range of the variables, for example you specify torque and three values are displayed, likewise for other variables. Are the maximum and minimum values of the variables shown there, Table 1.
Response: We are grateful to the reviewer for this comment. Table 1 only lists several observations of the used experimental data. In fact, the shown values are not the maximum and minimum values of them. We have checked all data and found that the range of these variables are very wide. To facilitate the understanding the paper, we also provided the ranges for the variables in the revised paper. Please refer to the paragraph before Table 1 for the detailed descriptions.
Comment 2: Table 2. I would like you to tabulate a portion of the predicted temperature as well, not just the error values.
Response: Thank you for this suggestion. When computing RMSE and MAPE, we first obtain the value of each observation predicted by one algorithm. And then compute the average values of these two metrics. In other words, the predicted temperatures need to be listed for each observation. On considering the large amount of test set, we just reported the average error metrics RMSE and MAPE for each method as shown in Table 2.
Comment 3: I'm missing a picture of the experiment, where is the temperature measurement position? A little more information about the experiment.
Response: Thank you for pointing out this issue. The experimental data are provided by one of our cooperated enterprises. For the data collecting system, they equip many sensors to the corresponding places of the wind machine. Due to confidential reasons, they didn’t provide the details. Thus, we cannot provide the details about the variables currently. In our experiments, we just selected some variables to study the influence of these variables to the temperature of the pitch motor.
Comment 4: You are talking about the moment of failure of wind turbines, you point out a date that is not recorded on the diagram. Why highlight the date if we don't have a time view of the tracking.
Response: We are sorry that the data shown on the x axis has not been made to correspond to the time points. Since there is not much places to annotate time on the x axis, we just used their indices in the test set. When revising the paper, we have indicated the corresponding place where there occurs a fault.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Thank you so much for addressing my comments.
