Integrating Machine Learning in Geotechnical Engineering: A Novel Approach for Railway Track Layer Design Based on Cone Penetration Test Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1.Modifications need to be made to summarize abstracts critically from the perspective of academic techniques or achievements.
2. In the introduction section, existing academic and theoretical research techniques related to this study need to be reinforced and added to the relevant references.
3. Discussions and conclusions also need to be revised by critically organizing academic and scientific contributions.

Author Response

Comments1: Modifications need to be made to summarize abstracts critically from the perspective of academic techniques or achievements

 

Response: a New abstract was written

 

 

Comments2: the introduction is reinforced

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Author, Thank you for your research. To improve the manuscript, I have several recommendations-questions for you:

1. The title of the manuscript reads "A Novel Approach". I suggest specifying this phrase in the manuscript.

2. "Hence, this article presents an automated recommendation system for sub..." are you suggesting "A Novel Approach" or "recommendations"?

3. What software complex was used to create Fig.1?

4. The quality of Fig. 1, 3 should be improved.

5. How well does artificial intelligence help in such research?

6. Fig.4 - what information can the reader learn?

7. "2.7. Classification model" what is the idea of this paragraph?

8. Where is the Author's model?

9. "3. Results" for such a study, this section should be enlarged, more information should be added.

10. Therefore, increase "4. Discussion»

 The Author must fulfill points 8-10!

 Kind regards,

Author Response

comments1:  The title of the manuscript reads "A Novel Approach". I suggest specifying this phrase in the manuscript.

response1:  I added

This paper introduces a pioneering application of machine learning in the field of railway engineering, focusing on predicting the necessary soil reinforcement of sub-layers directly from in situ CPT samples.

By determining the depth of soil replacement in an automated manner, this novel approach facilitates the design strategies for renewing railway tracks.

The applicability of the methodology considered in this paper is of course not limited to the Belgian network as it is actually a generic approach which can be implemented by each operator while adhering to their specific regulations.

 

Comment 2 : "Hence, this article presents an automated recommendation system for sub..." are you suggesting "A Novel Approach" or "recommendations"?

Response 2 : The abstract is modified to clarify the message

 

Comment 3: What software complex was used to create Fig.1?

Response 3 : Gimp

 

Comment 4 : The quality of Fig. 1, 3 should be improved

Response 4:  Quality of figure 3 is improved. Figure 1 is an existing image, so it is not possible for me to improve the quality.

 

Comment 5: How well does artificial intelligence help in such research?

Response: In the research of designing an optimal trackbed, AI helps to: Enhanced decision-making support through analysis based on extensive data sets. Minimization of human errors in the interpretation of geotechnical data by automating the analysis process. Increased operational efficiency: faster analysis of survey data.

 

Comments 6 : what information can the reader learn?

Response: PCA is used for dimensionality reduction

resulting in a lower-dimensional projection of the data that preserves the maximal data variance.

In this case it allows to easily visualize 2500 soundings distributed on 560 sites and there new trackbed design.

 

Comment 7: "2.7. Classification model" what is the idea of this paragraph?

Response 7: I present the machine learning model I use to automate the design of a trackbed with in input CPT data.

 

Comment 8: Where is the Author's model?

Response: The applied model is a pipeline made of succession of transformation of the data (2.2) ,

the pipeline include: Label encoding, Pre-processing, Hyperparameter optimization using cross-validation ...  

 

Comments: 9. "3. Results" for such a study, this section should be enlarged, more information should be added.

Response: In the results section, I discuss the model's performance on the training data and finally on a test dataset. I provide the confusion matrix. What additional information would you like to have?

For comparison, in the article "Assessing the Performance of Machine Learning Algorithms for Soil Classification Using Cone Penetration Test Data," no additional information is provided.

 

Comment 10: Therefore, increase "4. Discussion»

Response: Without concrete suggestions for improvement, it is difficult for me to enhance the discussion. Could you indicate what you believe should be added that is currently missing?

Reviewer 3 Report

Comments and Suggestions for Authors

Assessment of the Condition of Ballast and Sub-Ballast Layers of Railway Track is an Intriguing Scientific and Practically Necessary Task. The soil beneath the railway track has the ability to accumulate residual deformations and can affect the condition of the track as a whole. Diagnosing its condition requires the use of specialized equipment, and the results require specialized methods for their correct interpretation. This manuscript discusses the application of cone penetration testing and describes a method for further analysis of the results based on machine learning. This work has the potential to be useful for the further development of railway track condition analysis.

The reviewer has several significant remarks.

1. The author proposes the method of cone penetration testing. Only drilling is mentioned as an alternative method. Today, there is a group of methods based on measuring the velocity of wave propagation in the soil, including specialized measuring vehicles and small portable devices. For this work, it is necessary to include such methods in the literature review and compare them with the method described in the paper.

2. The manuscript text (and even the title) uses the term "design" of the railway track. However, the described method pertains not to design but to the assessment of the railway track condition, this is fundamentally different.

3. The text mentions complex sections of the railway track such as level crossings, turnouts, and bridge approaches. However, the author does not explain how the proposed method takes into account such sections with variable stiffness and whether its application is possible under such conditions.

4. "Furthermore, to ensure minimal interference from lateral friction during the test, a preliminary excavation is conducted in the ballast using a crowbar. This preparatory step accounts for the observed zero soil resistance in the initial 50 cm of depth." (Line 132-135).

4.1 It is necessary to explain in detail the technology of installing the device on the ballast (or in a certain depression in the ballast), as this can significantly impact the result.

4.2 It is necessary to justify the depth for starting the recording of results. Will it always be exactly 50 cm? It will likely depend on the thickness of the ballast layer (crushed stone), which can vary.

5. Figure 2a shows the application of the device to the side of the railway track (in the background, a train can be seen and the distance to the rails can be estimated). Is it possible to install the device directly on the railway track, particularly considering question 4.1? If the measurements, which this study is based on, were not performed directly under the track superstructure, such results are not suitable (or have a very limited scope of application).

Author Response

Comment 1 : I added this 

Today, surface wave seismic methods, more specifically multichannel surface wave analysis (MASW), are establishing themselves as non-destructive techniques for fast, efficient track diagnostics.
These methods can provide information over longer linear distances and with greater density than standard geotechnical methods.
Although these methods are of great interest as they enable a rapid and efficient assessment of the condition of railway infrastructures, they are not discussed further in this article.

Comment 2 :
"The manuscript text (and even the title) uses the term "design" of the railway track.
However, the described method pertains not to design but to the assessment of the railway track condition, this is fundamentally different."
Response: The AI model presented in the paper directly proposes the thickness of sublayers for design based on the assessment of the track using the PANDA test

Comment 3:
"The text mentions complex sections of the railway track such as level crossings, turnouts, and bridge approaches.
However, the author does not explain how the proposed method takes into account such sections with variable stiffness and whether its application is possible under such conditions"

Response: The same design methodology is applied to level crossings and turnouts on the belgian railway.
However, an asset manager is free to create its own model taking into account its preferences or normes/requirements.

Comment 4 and 5: I added

The PANDA test is conducted along the axis of the track to study the substructure of the railway.
The outcomes of the test are methodically recorded and presented in the form of penetrograms, which are graphs depicting the evolution of cone resistance with depth.
The testing commences at a depth of 0, corresponding to the uppermost surface of the railway sleeper.
Notably, the test uses a 4 $cm^2$ cone avoiding lateral friction measurements.
Furthermore, to ensure minimal interference from lateral friction during the test, a preliminary excavation is conducted in the ballast using a crowbar.
This preparatory step accounts for the observed zero soil resistance in the initial 50 cm of depth, which corresponds to the combined thickness of the sleepers (20 cm) and the ballast layer (30 cm).

Also:
50 cm is a theoretical reference, but if the preliminary excavation is only 40 cm, it is not a problem because the PANDA test is designed to neglect lateral friction.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Personally, I do NOT need anything.

I wrote questions-recommendations with respect for the Author and his research

Author Response

Thank you for your feedback. I have reviewed your questions and recommendations and have made every effort to incorporate your suggestions into my work. 

Reviewer 3 Report

Comments and Suggestions for Authors

I did not receive a clear answer to point 5 of my review on the previous version of the article. I consider this question very important, so I will formulate it again. Is it possible to conduct measurements directly on the railway track (of course, between the sleepers) - I marked this area as "A" in Fig. 1 (see file), or only on the shoulder - area "B", as shown in the photograph in the manuscript (Fig. 2a)? In which of these zones were the measurements, mentioned in the manuscript text, conducted?

Comments for author File: Comments.pdf

Author Response

Thank you for your feedback. To clarify the method, I added the following sentence:

'The PANDA test is performed directly along the track axis between two sleepers to analyze the railway substructure.'

I also changed the picture, which was not clear enough about the location for performing the test.

Round 3

Reviewer 3 Report

Comments and Suggestions for Authors

The authors provided thorough responses to all of my questions and implemented my recommendations. The article has been revised, so I recommend it for publication in its current form.