Toward Synthetic Data Generation to Enhance Skidding Detection in Winter Conditions

Round 1

Reviewer 1 Report

The paper "Toward Synthetic Data Generation to Enhance Skidding Detection in Winter Conditions" proposes the use of a neural network to identify lateral skidding events of road vehicles used during winter driving conditions.  The article needs improvements for publication, as suggested:   1) Introduction: authors must explain works related with Artificial Neural Networks. 2) The article lacks the creation of Background section for: (i) to approach theoretical aspects of Artificial Neural Networks and  (ii) present a literature review for the problem. 3) The section on "2.2. Neural network structures" lacks more important references in the Artificial Neural Networks field, for example, in "The network was trained using the Levenberg-Marquardt training algorithm".  4) Explain better how to obtain the network structure in: "The final network size was obtained by manual optimization using the list of linearly important variables". What was manual optimization? 5) Figure 5: improve quality.  6) Conclusion: better point out the possibilities for future work. 7) References: include more recent references in the field of Artificial Neural Networks (last three years). 

Author Response

Here is the list of modifications to the original document submitted, thank you for your feedback.

1) Introduction: authors must explain works related with Artificial Neural Networks.

• Addition of several paragraphs detailing the work published in recent articles covering the use of neural networks in vehicle dynamics estimation. This ties in with the elements needed to be improved mentioned in point 2.
• Most of the information was added to the introduction but some of the elements were added in the Neural Network section as this directly supported the articles methodology.

2) The article lacks the creation of Background section for: (i) to approach theoretical aspects of Artificial Neural Networks and  (ii) present a literature review for the problem.

• see point 1

3) The section on "2.2. Neural network structures" lacks more important references in the Artificial Neural Networks field, for example, in "The network was trained using the Levenberg-Marquardt training algorithm".  

• Additional research on the use of the Levenberg-Marquardt training algorithm was done to increase the theoretical validity of the algorithm in standard and time-delayed neural networks. The added information can be found in section 2.2 of the article.
• New and updated references were also added throughout section 2.2, increasing the theoretical background in general.

4) Explain better how to obtain the network structure in: "The final network size was obtained by manual optimization using the list of linearly important variables". What was manual optimization?

• Wording of this element has been modified and a suitable explanation has been added. The "manual" method is in fact an automated run of 13, first and second layer node modifications, network structures were the MSE of the estimate was compared to find the optimal (size to performance ratio).

5) Figure 5: improve quality.

• figures 5 and 9 were modified to improve their quality clarity.

6) Conclusion: better point out the possibilities for future work.

• Two new paragraphs were added that include future work on the subject.

7) References: include more recent references in the field of Artificial Neural Networks (last three years).

• Additional literature was found, analyzed and added the the paper. Including several recent (2019 and after) article and proceeding concerning the use of neural networks in vehicle dynamics estimation, network training and tire road interactions.

Reviewer 2 Report

even if the topic is not in my research field, the article is understandable and pleasant to read. From my point of view, it can be published after corrections of  minor methodological errors and text editing

Author Response

Here is the list of modifications to the original document submitted, thank you for your feedback.

1) Introduction

• Addition of several paragraphs detailing the work published in recent articles covering the use of neural networks in vehicle dynamics estimation. This ties in with the elements needed to be improved mentioned in point 2.
• Most of the information was added to the introduction but some of the elements were added in the Neural Network section as this directly supported the articles methodology.

2) Are the methods adequately described?

• Additional research on the use of the Levenberg-Marquardt training algorithm was done to increase the theoretical validity of the algorithm in standard and time-delayed neural networks. The added information can be found in section 2.2 of the article.
• New and updated references were also added throughout section 2.2, increasing the theoretical background in general.
• Wording considering the manual optimization of the neural network has been modified and a suitable explanation has been added. The "manual" method is in fact an automated run of 13, first and second layer node modifications, network structures were the MSE of the estimate was compared to find the optimal (size to performance ratio).

3) Figure 5: improve quality.

• figures 5 and 9 were modified to improve their quality clarity.

4) Conclusion

• Two new paragraphs were added that include future work on the subject.

5) References

• Additional literature was found, analyzed and added the the paper. Including several recent (2019 and after) articles concerning the use of neural networks in vehicle dynamics estimation, network training and tire road interactions.

6) Report proof-read

Reviewer 3 Report

The neural network could be more exactly explained. Plus, figs. 5 and 9 could be improved because the first part of figure 9 and figure 5 are not very clear.

Some minor corrections in spelling could be made.

Author Response

Here is the list of modifications to the original document submitted, thank you for your feedback.

1) Introduction

• Addition of several paragraphs detailing the work published in recent articles covering the use of neural networks in vehicle dynamics estimation. This ties in with the elements needed to be improved mentioned in point 2.
• Most of the information was added to the introduction but some of the elements were added in the Neural Network section as this directly supported the articles methodology.

2) Are the methods adequately described?

• Additional research on the use of the Levenberg-Marquardt training algorithm was done to increase the theoretical validity of the algorithm in standard and time-delayed neural networks. The added information can be found in section 2.2 of the article.
• New and updated references were also added throughout section 2.2, increasing the theoretical background in general.
• Wording considering the manual optimization of the neural network has been modified and a suitable explanation has been added. The "manual" method is in fact an automated run of 13, first and second layer node modifications, network structures were the MSE of the estimate was compared to find the optimal (size to performance ratio).
• Additional information concerning the network structure was added in a footnote. Please inform me if this is adequate or if more information is needed.

3) Figure 5: improve quality.

• figures 5 and 9 were modified to improve their quality clarity.

4) Conclusion

• Two new paragraphs were added that include future work on the subject.

5) References

• Additional literature was found, analyzed and added the the paper. Including several recent (2019 and after) articles concerning the use of neural networks in vehicle dynamics estimation, network training and tire road interactions.

6) Report proof-read

Round 2

Reviewer 1 Report

In my opinion, authors have addressed the comments proposed in this new paper version.