Influence of Meteorological Parameters on Explosive Charge and Stemming Length Predictions in Clay Soil during Blasting Using Artificial Neural Networks
Round 1
Reviewer 1 Report
The topic addressed is interesting, but it only presents the prediction results. Little consideration has been made about the findings derived from the results. For example, different models have different reproducibility and there would be many things that need consideration by going back to the various parameters, such as considering what the factors are. I believe that simulation research does need verification and calibration based on the analysis results. In conclusion, this paper lacks such a process and can be regarded as only a superficial analysis. For this reason, I cannot recommend this paper for publication.
Author Response
Rev 1
Authors:
First of all, we want to thank Rev 1 for helpful comments and suggestions. All suggestions by Rev 1 were accepted and implemented in the text of the manuscript.
Comments and Suggestions for Authors:
The topic addressed is interesting, but it only presents the prediction results. Little consideration has been made about the findings derived from the results. For example, different models have different reproducibility and there would be many things that need consideration by going back to the various parameters, such as considering what the factors are. I believe that simulation research does need verification and calibration based on the analysis results. In conclusion, this paper lacks such a process and can be regarded as only a superficial analysis. For this reason, I cannot recommend this paper for publication.
Answer:
We thank Rev 1 for constructive remarks and suggestions, and below the authors give their explanations:
The manuscript was sent for proofreading to an English language expert in Croatia for English language editing.
In this paper, the authors show that it is possible to compare and show the direct impact on the obtained data by field research (Volume of the resulting cavity, resulting expansion of the borehole and Deepening of the resulting expansion) with the amount of precipitation (rain) and air temperature, were 200 days before the blasting in clay soil. The behavior of clay in conditions of increased humidity (rain) is presented and described in detail in the doctoral dissertation Težak, D., 2018 and in the paper Težak, D. et al.: Impact of Seasonal Changes of Precipitation and Air Temperature on Clay Excavation. Sustain. 2019, 11.
As two types of explosive charge were used in the research, Pakaex and Permonex V19, ie different amounts of charge, and stem length of 0.5 - 1.0 m, and these parameters were taken into account when predicting the success of blasting, the formation of the volume of spherical expansion and expansion of borehole during blasting.
It has been unequivocally proven (Težak, D. Influence of the Blasting Features on the Expansion in Clay Soil, Faculty of mining, geology and petroleum engineering, 2018.; Težak, D.; Stanković, S.; Kovač, I. Dependence Models of Borehole Expansion on Explosive Charge in Spherical Cavity Blasting. Geosciences 2019, 9, 383, doi:10.3390/geosciences9090383.; Kovač, I.; Težak, D.; Mesec, J.; Markovinović, I. Comparative Analysis of Basic and Extended Power Models of Boreholes Expansion Dependence on Explosive Charge in Blasting in Clay Soil. Geosci. 2020, 10, 1–11, doi:10.3390/GEOSCIENCES10040151. and Težak, D.; Soldo, B.; Đurin, B.; Kranjčić, N. Impact of Seasonal Changes of Precipitation and Air Temperature on Clay Excavation. Sustain. 2019, 11.) that the amount of precipitation and air temperature have a direct impact on the success of blasting and the formation of spherical expansion in the clay soil after blasting.
Guided by the above facts, the authors developed a prediction model based on an artificial neural network that combines Long short-term memory (for processing time-dependent data - precipitation and air temperature) and Fully connected layers (processes data related to blasting and output from the LSTM part) with assistance by which civil mining, geotechnical engineers, and other engineers of related professions, when preparing their projects (improvement of clay soil by blasting, conducting blasting in clay soil, etc.) will receive direct guidelines for successful blasting. Under direct guidance, the authors think that engineers will be able to estimate the optimal amount of explosive charge and stem length based on our predictions about the impact of precipitation and air temperature, taking into account the amount of explosive charge (which they plan to use when conducting their project or research) and stem length, or use our model to be able to estimate the optimal amount of explosives and/or stem length when blasting in clay soil, given the previous 200 (or some other number they deem appropriate) weather conditions.
Also, considering your suggestions, results of the predicted vs. observed Stemming Lengths and Explosive Charge Mass were added (MSE, R2, RSE, MBE and leverage) which confirmed the success of the model for predicting the impact of precipitation and air temperature on the success of blasting.
In conclusion, the intention of the authors of this paper was to encourage new research that will allow additional verification and calibration based on the analysis results. In that sense, there is a significant interest of young researchers from our close academic community.
In order to make it easier for readers to follow the issues of the manuscript, the authors supplemented the "Introduction". Also, the authors added the above mentioned evaluation metrics that only further confirmed the initial premise of the manuscript.
Author Response File: 
Author Response.docx
Reviewer 2 Report
- The English language needs to improve.
- References need to check.
Author Response
Rev 2
Authors:
First of all, we want to thank Rev 2 for helpful comments and suggestions. All suggestions by Rev 2 were accepted and implemented in the text of the manuscript.
Comments and Suggestions for Authors:
The English language needs to improve.
Answer:
The manuscript was sent for proofreading to an English language expert in Croatia for English language editing.
Comments and Suggestions for Authors:
References need to check.
Answer:
References in the manuscript are checked and aligned with the template and requirements recommended in the “Instructions for authors” of the journal Applied Sciences.
Author Response File: Author Response.docx
Reviewer 3 Report
Loskovar et al. study the influence of of meteorological parameters on explosive charge and stemming length predictions in clay soil during blasting using ANNs. The work falls into the scope of applied sciences. I provide the following suggestions with the aim to improve the overall quality of the manuscript.
1- have the manuscript read by an English language professional to avoid any grammatical mistakes.
2- avoid phrases like “During the research”, “Following the above”, and “The subject research” in the manuscript. There are lots of phrases similar to these in the manuscript. Please look for the rest.
3- avoid bulk citations in the manuscript, and only cite references that you actually used. For instance line 31 and line 34. Please look for the rest in the manuscript.
4- the introduction is short and does not provide enough background to the problem. You should also mention the application of artificial intelligence in the field, bringing a few example without proper description is not sufficient.
5- please mention the novelty of the research in the introduction, and mention why this research is a good addition to the available literature.
6- again “Therefore, this research manuscript presents”. This does not sound really scientific, try to avoid such phrases.
7- figure 1, again bulk citation. Also this figure is not really informative.
8- figure 2, bulk citation.
9- better describe the blasting results presented in table 1.
10- lines 116-118, you should not quote the text here. You can easily write it with your own words.
11- delete figures 4 and 6. they are too simple.
12- better describe the implementation process of ANN model. The acquisition function, input parameters, etc should be better described. Which platform did you use for the implementation of ANN models? Can provide the code to readers? It would be better to publish the developed ANN codes.
13- how did you avoid over-fitting?
14- you need to provide statistical analysis on the results to prove the validity of your model. For instance, you can use the leverage statistical method for analyzing errors. You can also report the relative errors.
15- simply using MSE and R2 values is not sufficient for assessing the performance of ANN models. These criteria are only provide a very simple evaluation of the model. A more thorough evaluation of ANN models is required.
16- how did you evaluate the generalization ability of your models? This should be discussed in the manuscript with plots, etc.
17- the conclusion is very narrative and cannot be considered as proving the applicability of the models. You need to expand the discussion section, and also investigate the applicability of you models.
18- what is the take home message? How the readers can benefit from the findings?
19- how the ANN models can be used in the future? Is it possible to reproduce the results.
20 similar to the discussion section, the conclusion is very abstract and does not clearly mention the findings.
21- I am not sure if the results of the paper valid as the error analysis is basic and limited.
Author Response
Rev 3
Authors:
First of all, we would like to thank the Rev 3 for the thorough and benevolent analysis of our manuscript. Also, we thank Rev 3 for the suggested corrections since they will greatly contribute to the quality of the manuscript as a whole. We also provide explanations and/or corrections according to each of the individual comments.
Comments and Suggestions for Authors:
Leskovar et al. study the influence of meteorological parameters on explosive charge and stemming length predictions in clay soil during blasting using ANNs. The work falls into the scope of applied sciences. I provide the following suggestions with the aim to improve the overall quality of the manuscript.
Comments and Suggestions for Authors:
1- have the manuscript read by an English language professional to avoid any grammatical mistakes.
Answer:
Corrected.
The manuscript was sent for proofreading to an English language expert in Croatia for English language editing.
Comments and Suggestions for Authors:
2- avoid phrases like “During the research”, “Following the above”, and “The subject research” in the manuscript. There are lots of phrases similar to these in the manuscript. Please look for the rest.
Answer:
Corrected.
All the phrases have been replaced when editing the English language by the English language expert, especially the "Introduction" was corrected according to the suggestion.
Comments and Suggestions for Authors:
3- avoid bulk citations in the manuscript, and only cite references that you actually used. For instance, line 31 and line 34. Please look for the rest in the manuscript.
Answer:
Corrected. – line 31-37
“Introduction” was written according to suggestion, avoiding bulk citations, and with further explanations of each of the referenced studies.
Comments and Suggestions for Authors:
4- the introduction is short and does not provide enough background to the problem. You should also mention the application of artificial intelligence in the field, bringing a few example without proper description is not sufficient.
Answer:
Corrected.
The "Introduction" was rewritten, according to the suggestions. In the field of Geotechnics and blasting in clay soils, Artificial Intelligence (AI) was mainly investigated in terms of soil properties prediction models, more accurately in predicting water content (Zhou et al., 2016, W.-H. Zhou, A. Garg, A. Garg Study of the volumetric water content based on density, suction and initial water content Measurement, 94 (2016), pp. 531-537), temperature (Y. Feng, N. Cui, W. Hao, L. Gao, D. Gong - Estimation of soil temperature from meteorological data using different machine learning models - Geoderma, 338 (2019), pp. 67-77; Mohammad Sadik Khan, John Ivoke, Masoud Nobahar & Farshad Amini (2021) Artificial Neural Network (ANN) based Soil Temperature model of Highly Plastic Clay, Geomechanics and Geoengineering, DOI: 10.1080/17486025.2021.1928765), ground vibrations or peak particle velocity (Zeng, J.; Roussis, P.C.; Mohammed, A.S.; Maraveas, C.; Fatemi, S.A.; Armaghani, D.J.; Asteris, P.G. Prediction of Peak Particle Velocity Caused by Blasting through the Combinations of Boosted-CHAID and SVM Models with Various Kernels. Appl. Sci. 2021, 11, 3705. https://doi.org/10.3390/app11083705). However, none of the studies involved predicting blasting charge mass and/or stem length according to the desired borehole parameters and weather conditions which certainly impact the clay in general, and also the blasting in clay, therefore we came with this novel approach which should serve as starting point for further investigations involving predicting the best possible charge mass which should avoid unnecessary expenses and help reduce impacts on the environment.
Comments and Suggestions for Authors:
5- please mention the novelty of the research in the introduction, and mention why this research is a good addition to the available literature.
Answer:
Corrected.
Comments and Suggestions for Authors:
6- again “Therefore, this research manuscript presents”. This does not sound really scientific, try to avoid such phrases.
Answer:
Corrected.
Removed the phrase.
Comments and Suggestions for Authors:
7- figure 1, again bulk citation. Also this figure is not really informative.
Answer:
Fixed: added explanation of Figure 1 and removed bulk citation.
Comments and Suggestions for Authors:
8- figure 2, bulk citation.
Answer:
Fixed: removed bulk citation.
Comments and Suggestions for Authors:
9- better describe the blasting results presented in table 1.
Answer:
Corrected.
Plan and methods of field research, as well as data processing, are shown in Mesec J. et al., 2015 [1], Težak D., 2018 [2], Težak, D. et al., 2019 [3] and Kovač I. et al., 2020 [6]. Data used in data processing are shown in Table 1.
Line 228-272: Based on the data displayed in Table 1, the volume of the resulting cavity, Vrc [m3] and resulting expansion of the borehole, Lre [m] at the same explosive charge mass Q [kg] were higher after the activation of Permonex V19 than Pakaex because Permonex V19 has a higher density and VOD. Also, deepening of the resulting expansion, Dre [m] at the same explosive charge mass Q [kg], is larger with Pakaex than with Permonex V19.
Comments and Suggestions for Authors:
10- lines 116-118, you should not quote the text here. You can easily write it with your own words.
Answer:
Corrected.
Line 259-260: An artificial neural network is a data processing system built of simple processing units (neurons) with the tendency of storing experience for later use [27].
Comments and Suggestions for Authors:
11- delete figures 4 and 6. they are too simple.
Answer:
Corrected.
Figures 4 and 6 are deleted.
Comments and Suggestions for Authors:
12- better describe the implementation process of ANN model. The acquisition function, input parameters, etc should be better described. Which platform did you use for the implementation of ANN models? Can provide the code to readers? It would be better to publish the developed ANN codes.
Answer:
Corrected.
As mentioned in Section 2.6. the Python programming language together with the PyTorch framework was used to create the ANN-s used in this study. Also, other popular libraries like Numpy and Pandas have been used during data exploration, pre-processing and metrics calculation. For visualizations the package Matplotlib with addition os Seaborn was used. The code was written in a Jupyter Notebook, and the notebook (code) is in the attachment, for the reviewers to inspect. The code is also published on the GitHub profile of the first author, Karlo Leskovar, however we would like to keep the code as private until article publication. After the article is published, te code will be made public for the readers.
Comments and Suggestions for Authors:
13- how did you avoid over-fitting?
Answer:
Corrected.
Over-fitting was avoided with multiple trial and error runs of the training process. The goal was to find an optimal combination of epochs and learning rate. Care has been taken that the validation data loss does not surpass the training loss and that validation loss is decreasing. By doing so, it was ensured that the validation loss is at the final epoch lower than training loss, meaning the model generalizes well to unknown data samples, and does not overfit to a certain training dataset. Then, the optimal solution (combination of epochs and learning rate) was used for training and inference where the presented data/plots were calculated.
Comments and Suggestions for Authors:
14- you need to provide statistical analysis on the results to prove the validity of your model. For instance, you can use the leverage statistical method for analyzing errors. You can also report the relative errors.
Answer:
Corrected.
According to the suggestion, leverages and studentized on predicted stem lengths and charge masses were calculated and plotted. Additional statistics were also added, in form of Relative Squared Error (RSE) and Mean Bias Error (MBE) and leverage statics.
Comments and Suggestions for Authors:
15- simply using MSE and R2 values is not sufficient for assessing the performance of ANN models. These criteria are only provide a very simple evaluation of the model. A more thorough evaluation of ANN models is required.
Answer:
Corrected.
Added Relative Squared Error (RSE) and Mean Bias Error (MBE) and leverage statics.
Comments and Suggestions for Authors:
16- how did you evaluate the generalization ability of your models? This should be discussed in the manuscript with plots, etc.
Answer:
Corrected.
The generalization ability of the models is demonstrated by plotting the validation and testing (independent data) phases together with the training phase plots – Figure 6-13, where clearly none of the predicted plotted points (training, validation, and testing phase) displayed any outliers. Furthermore, the generalization ability has been reinforced by adding the leverage statistic to the plots.
Comments and Suggestions for Authors:
17- the conclusion is very narrative and cannot be considered as proving the applicability of the models. You need to expand the discussion section, and also investigate the applicability of you models.
Answer:
Corrected.
Applicability of the model has been proven with the validation and testing period (data), which are independent datasets, measured in different time than the training data (also with different meteorological conditions), meaning that if the model performs well on those datasets, it generalizes well to unknown data.
Comments and Suggestions for Authors:
18- what is the take home message? How the readers can benefit from the findings?
Answer:
Corrected.
The conclusion and discussion sections have been enhanced according to the comment. Also we provide a brief explanation of our idea here.
With our study we wanted to show the that:
- Models based on ANN can be used in predicting required stemming length and explosive charge mass required when blasting in clay soil
- Given the already proven significant influence of meateorological conditions on blasting in clay soil (previous research by the second author dr. Težak), this influence is mirrored when creating an ANN based predictor model, however due to the nature of ANN-s adding more features (training data) additional hyperparameter fine tuning is required in order to yield best possible results
Comments and Suggestions for Authors:
19- how the ANN models can be used in the future? Is it possible to reproduce the results.
Answer:
Corrected.
Special care has been taken to enable result reproducibility, by making sure that all random number generators like i.e. weigths and biases in the ANN have been seeded to a certain value, I our case seed = 14. Meaning, if this exact model structure is used by other investigators and other computer, Pytorch seed will ensure same training and validation results by setting identical starting weights and biases in the ANN. So, the answer is YES, the results can be reproduced, by using same input data (in terms of blasting and meteorological data) and same model hyperparameters and using the torch_seed parameter equal to 14.
Comments and Suggestions for Authors:
20 similar to the discussion section, the conclusion is very abstract and does not clearly mention the findings.
Answer:
Corrected.
Conclusion has been corrected according to the suggestion.
Comments and Suggestions for Authors:
21- I am not sure if the results of the paper valid as the error analysis is basic and limited.
Answer:
Corrected.
According to prior suggestions, the results, discussion and conclusion section have been corrected, the requested evaluation statistics have been added in order to improve the error analysis and overall manuscript quality.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The revised manuscript has much more academic value than the previous one. I think it is worthy of publication.
Reviewer 3 Report
The authors have addressed all my comments. I suggest publication of the manuscript.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
