Predicting Geotechnical Parameters from Seismic Wave Velocity Using Artificial Neural Networks
Round 1
Reviewer 1 Report
The authors have used ANN to predict parameters from seismic wave velocity. The work is traditional, and there are some issues to be addressed.
1.How many neurons are included in the 6 ANNs, and how many layers are they included?
2.How many data are included in this ANN?
3.Are the parameters the authors use as the imput in ANN is related to the output parameter? The authors should be clear if they have studied fully the parameters or if the parameter is appropriate.
4.What is 'LL' in the text?
5. The abstract should be improved.
Other comments:
1. There is a ')' after 0.73 in the following expression:
For sandy soil, R-value is 0.65, for 35
clayey soil R-value is 0.75, and the combined data R-value resulted in 0.73
2.Check the format carefully in the text, for example, spaces should be reserrved in the beginning of the following phrase:
Researchers have extensively studied...
3.Check if the ',' is appropriate in the expression:
'excitation frequencies,...'
4. Units should be added after the parameters:
'M is the constraint modulus, B is the bulk modulus, G is the shear modulus, and ρ is the mass density of the medium'
5. Spaces should be added between number and unit.
6. I do not understand the following expression:
'35 different soils samples were compacted in 4-inch diameter by 4.58 in'.
Author Response
Thank you for your questions and suggestions. Please find my answer below:
1.How many neurons are included in the 6 ANNs, and how many layers are they included?
Answer: ANN networks are organized in the following way
input_ (initial hidden nodes_final hidden nodes_iteration) _output)
For example, the network 4_ (4_4_3100) _1 there are 4 inputs, 4 initial nodes/neurons, 4 final nodes/neurons, 3100 iterations, and 1 output
2.How many data are included in this ANN?
Answer: For this study the total number of data used for the ANN model development are as follows
Field data = 34
Lab data = 35
Field plus lab data = 69 (Line: 138 & 150)
3.Are the parameters the authors use as the input in ANN is related to the output parameter? The authors should be clear if they have studied fully the parameters or if the parameter is appropriate.
Answer: A correlation matrix analysis was used before selecting the input and output parameters for the prediction model development. (Line 198-199)
4.What is 'LL' in the text?
Answer: LL is the abbreviation of Liquid limit. (Line 155)
- The abstract should be improved.
Answer: I have edited the abstract and added some numerical values to support the results.
Other comments:
- There is a ')' after 0.73 in the following expression:
For sandy soil, R-value is 0.65, for 35
clayey soil R-value is 0.75, and the combined data R-value resulted in 0.73
Answer: Thank you for your detailed review. I have made corrections. (Line: 44)
2.Check the format carefully in the text, for example, spaces should be reserved in the beginning of the following phrase:
Researchers have extensively studied...
Answer: I have made the corrections.
3.Check if the ',' is appropriate in the expression:
'excitation frequencies,'
Answer: I have corrected (Line: 107)
- Units should be added after the parameters:
'M is the constraint modulus, B is the bulk modulus, G is the shear modulus, and ρ is the mass density of the medium'
Answer: Units have been added in the text after the parameter definition. We use the MKS (meters, kilogram, seconds) system of units. Therefore, moduli are in units of Pa (Pascals). (Line 101-111)
- Spaces should be added between number and unit.
Answer: I edited the spaced between number and units.
- I do not understand the following expression:
'35 different soils samples were compacted in 4-inch diameter by 4.58 in'.
Answer: I have edited the sentence to read “35 samples were compacted using 4 inch diameter by 4.58 inch high molds. (Line 152-153)
Reviewer 2 Report
The paper applied ANN to predict some geotechnical parameters as well as wave velocity for a collected dataset. The topic is interesting but there are still some issues needed to be improved before the paper can be accepted.
1) The artificial intelligence technique has been widely applied in predicting key geotechnical parameters. A more comprehensive review should be added in the introduction.
2) Seismic velocity is familar to the geotechncal community, thus Section 2 is redundant and should be much more concise.
3) A correlation analysis should be conducted between the input and output parameters as an effective and efficient neural network should avoid incorporating unrelevant factors in the input.
4) How may trial and error cases have the authors conducted to determine the hyperparameters? Only those listed in the tables?
5) A detailed description on the regression method should be provided.
6) The biggest issue as mentioned by the authors is the side of data base. The accuracy of an ANN model largely relies on the accuracy and side of the database. The authors may consider to employ some special techniques that can tackle with the prediction based on small size database.
7) There are mistyping issues at some places, for example line 216. The authors should conduct a thorough check on these issues during revision.
Author Response
Thank you for your questions and suggestions. Please find my answer below:
1) The artificial intelligence technique has been widely applied in predicting key geotechnical parameters. A more comprehensive review should be added in the introduction.
Answer: I have added more literature review in the introduction section. (Page 3 & 4)
2) Seismic velocity is familiar to the geotechnical community, thus Section 2 is redundant and should be much more concise.
Answer: I agree with you that geophysical measurements are being used more frequently in engineering. The objective of placing some theoretical seismic models in the paper is to describe simplest models for seismic velocity dependence on soil properties. Essentially, although there is a density dependence, the seismic velocity is primarily controlled by the elasticity of the porous material. For granular material, this elasticity depends on the grain contacts which are very difficult to quantify in terms of effective stress, capillary pressure and cementation. Therefore an ANN approach is better suited for establishing relationships. Other reviewers requested more information regarding underlying dependence of seismic velocity to geotechnical parameters as such we have decided to retain the seismic descriptions.
3) A correlation analysis should be conducted between the input and output parameters as an effective and efficient neural network should avoid incorporating unrelevant factors in the input.
Answer: Before selecting the input and output parameters for ANN model development a correlation matrix was developed. I have added a line referring to this correlation matrix (Line: 198 & 199).
4) How may trial and error cases have the authors conducted to determine the hyperparameters? Only those listed in the tables?
Answer: For each ANN final model shown in the table, a total of 5 to 6 models were developed for different numbers of initial and final nodes/neurons. For each model, a total of 20000 iterations were used. The three best models were chosen for testing, validation and training using all the data. The final ANN model was chosen from these three best models depending on the statistics of training, testing and validation.
5) A detailed description on the regression method should be provided.
Answer: I have added a description for regression model development. (Page 16 & 17)
6) The biggest issue as mentioned by the authors is the size of data base. The accuracy of an ANN model largely relies on the accuracy and size of the database. The authors may consider employing some special techniques that can tackle the prediction based on small size database.
Answer: Since the number of data is limited, two different ANN approaches are used for predicting seismic wave velocity. The first approach is the typical approach, where the data is used for training, testing, and validation. Around 50% of data is used for training, 25% for testing, and 25% for validation. In the second approach, the validation stage is excluded so that 75% data is used for training and 25% for testing. The analysis showed that excluding the validation provides more accuracy and good correlation. Further analysis was conducted without the validation steps for ANN model development. (Line: 191-195).
7) There are mistyping issues at some places, for example line 216. The authors should conduct a thorough check on these issues during revision.
Answer: I have corrected the mistyping errors.
Reviewer 3 Report
1. The abstract should state briefly the purpose of the research, the principal results and major conclusions. An abstract is often presented separately from the article, so it must be able to standalone. The authors should endeavour to put down some numerical values to support their claims in the abstract and conclusion part of the manuscript.
2. Aside from the aim stated in the title, the research gap and the goals of the research are not specified which leads to the reader missing the significance of the research.
3. Please emphases the scientific value your paper added in your abstract and end introduction.
4. The major defect of this study is the argument is not clear stated in the introduction session. Hence, the contribution is weak in this manuscript. I would suggest the authors to enhance your theoretical discussion and arrives your debate or argument.
5. I would like to request the authors to emphasis on the contributions on practically and academically in implication session.
6. It is suggested to add more articles, recent ones and relevant to the research question to the literature review in enhance the number of references.
7. Please make sure your conclusions' section underscore the scientific value added of your paper, and/or the applicability of your findings/results, as indicated previously. Please revise your conclusion part into more details. Basically, the authors should enhance your contributions, limitations, underscore the scientific value added of your paper, and/or the applicability of your findings/results and future study in this session.
8. Methods section determines the results. Kindly focus on three basic elements of the methods section.
a. How the study was designed?
b. How the study was carried out?
c. How the data were analyzed?
9. Especially, the introduction section needs to re-organize. The major debate or Argument is not clear stated in the introduction session. Hence, the contribution debates are weak in this manuscript. I would suggest the authors to enhance your literature discussion and arrives your debate or argument.
Author Response
Thank you for your questions and suggestions. Please find my answer below:
- The abstract should state briefly the purpose of the research, the principal results and major conclusions. An abstract is often presented separately from the article, so it must be able to standalone.The authors should endeavour to put down some numerical values to support their claims in the abstract and conclusion part of the manuscript.
Answer: I have edited the abstract and inserted some numerical values to support the claims. (Page:1)
- Aside from the aim stated in the title, the research gap and the goals of the research are not specified which leads to the reader missing the significance of the research.
Please emphases the scientific value your paper added in your abstract and end introduction.
Answer: More literature reviews were added, and the scientific values were discussed. (Page 3 & 4)
- The major defect of this study is the argument is not clear stated in the introduction session. Hence, the contribution is weak in this manuscript. I would suggest the authors to enhance your theoretical discussion and arrives your debate or argument.
Answer: Suggestion was followed. (Page 3 & 4)
- I would like to request the authors to emphasis on the contributions on practically and academically in implication session.
Answer: Introduction and conclusion were modified according to the suggestion. (Page 3, 4, 18 &19)
- It is suggested to add more articles, recent ones and relevant to the research question to the literature review in enhance the number of references.
Answer: More literature review was conducted. (Page 3 and 4)
- Please make sure your conclusions' section underscore the scientific value added of your paper, and/or the applicability of your findings/results, as indicated previously. Please revise your conclusion part into more details. Basically, the authors should enhance your contributions, limitations, underscore the scientific value added of your paper, and/or the applicability of your findings/results and future study in this session.
Answer: Conclusion was modified. (Page 18&19)
- Methods section determines the results. Kindly focus on three basic elements of the methods section.
How the study was designed?
b. How the study was carried out?
c. How the data were analyzed?
Answer: Thank you for your suggestions. I rechecked the method section.
- Especially, the introduction section needs to re-organize. The major debate or Argument is not clear stated in the introduction session. Hence, the contribution debates are weak in this manuscript. I would suggest the authors to enhance your literature discussion and arrives your debate or argument.
Answer: Introduction has been modified (page 3 & 4).
Round 2
Reviewer 1 Report
The manuscript has been improved. However, there are some corrections to be made before the acceptance of this paper. The authors need to check carefully the manuscript to avoid unnecessary mistakes.
For example:
'Howver' in the abstract should be 'however'.
Another question is that the authors reply that there are only 69 data for the ANN, is it enough for a machine learning algorithm, how much is the uncertainty? The author may need to understand or clarify this in the paper.
Author Response
Thank you for your questions. I have added my answers in the attached file.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors have made some changes according to the previous comments. However, there are still two issues needed to be addressed before final acceptance.
Firstly, a critical review of current application of intelligent methods in geotechincal engineering should be provided as I suggested previously. Through such, the authors may provide basis for selecting ANN as the prediction methods rather than other methods.
Secondly, the authors' 'numerical tests' in different training mode is not what I meant to better exploit the small sized data. A larger training data size gave a higher accuracy, which would be expected intuitive. The authors may refer to some newly published work in exploiting small-sized data set to get acceptable accuracy.
There are still some minor language issues, especially in the newly added texts highlighted in yellow. A thorough check on typos and grammatical issues is compulsory.
Author Response
Thank you for your comments/ questions. Manuscript has been edited based on your suggestions.
Author Response File: Author Response.docx
