Prediction of Curing Time/Shear Strength of Non-Conductive Adhesives Using a Neural Network Model

Round 1

Reviewer 1 Report

Manuscript: Machine Learning based Prediction of Curing Time/Shear Strength of Non-Conductive

Adhesives and its Formulation Optimization by Min et al.

The manuscript is well researched and well written. Here, I suggest some modifications to be

considered with the aim to improve the overall quality/readability of the manuscript.

1- Improve the English language to eliminate any grammatical errors. For instance, in line 166,

do not start a sentence with a number. Check for other errors.

2- Title: modify the title to “Prediction of Curing Time/Shear Strength of Non-Conductive

Adhesives using a neural network model”. This title better reflects the work conducted in

this research. You used a simple 3 layer neural network model, and this should be

mentioned in the title. The words “machine learning” and “optimization” are rather

confusing in the title.

3- In the abstract, you also used the “neural network” to describe your work, which is correct.

This should also be clearly mentioned in other sections of the manuscript. Do not use

“machine learning” when describing your work.

4- Which platform did you use to implement the neural network model?

5- You only mentioned that you employed a 3-layer neural network. Did you used a FFNN

model? Mention this clearly in the manuscript.

6- Provide a better error analysis to evaluate the performance of the ANN model. There are

lots of statistical approaches available.

7- I suggest to remove Tables 2 1nd 3 from the main manuscript, and present them in an

appendix or supplementary material. This is too much information, which can be deleted

from the main text.

8- Please provide some discussion on the results provided. For instance, mention the

importance of the work for the current literature, the future research direction, etc.

Author Response

1. Improve the English language to eliminate any grammatical errors. For instance, in line 166, do not start a sentence with a number. Check for other errors.
• We have a grammar check in the university writing center, and the sentence in the line 166 was revised as “The input parameters are 4 types of resins, 2 types of hardeners, 8 types of catalysts, and a coupling agent, and 65 datapoints were used for the prediction of the curing time with regression models.“
2. Title: modify the title to “Prediction of Curing Time/Shear Strength of Non-Conductive Adhesives using a neural network model”. This title better reflects the work conducted in this research. You used a simple 3 layer neural network model, and this should be mentioned in the title. The words “machine learning” and “optimization” are rather confusing in the title.
• Title has been revised as advised.
3. In the abstract, you also used the “neural network” to describe your work, which is correct. This should also be clearly mentioned in other sections of the manuscript. Do not use “machine learning” when describing your work.
• Section 2.3 has been revised: “2.3. Neural network model”
4. Which platform did you use to implement the neural network model?
• We added the following sentence: “Modeling was implemented by Python and Keras application programming interfaces.”
5. You only mentioned that you employed a 3-layer neural network. Did you used a FFNN model? Mention this clearly in the manuscript.
• The paper has been revised in lines 173-174.: “The ANN model, as a multilayer perceptron (MLP) network, with one hidden later with 3 nodes was used.”
6. Provide a better error analysis to evaluate the performance of the ANN model. There are lots of statistical approaches available.
• MAPE is added in Table 2.
• The paper has been revised in lines 193-194: “The calculated MAEs and MAPEs for training and validation datasets for curing time and shear strength models were shown in Table 2.”
• RMSEs may be added in Table 2, but it can be redundant with MAEs.
• R2 data are shown in Figs 4 and 5.
7. I suggest to remove Tables 2 and 3 from the main manuscript, and present them in an appendix or supplementary material. This is too much information, which can be deleted from the main text.
• Tables 2 and 3 were move to Appendix A.
8. Please provide some discussion on the results provided. For instance, mention the importance of the work for the current literature, the future research direction, etc.
• The used materials in this study are epoxy types of resin, anhydride types of hardener.
Imidazole and amine types of catalyst, and silane coupling agent. In this paper, a
framework of prediction of curing time/shear strength of NCA and optimization of NCA
formulation using ANN model.
• Currently, data mining is progressing to improve accuracy of ANN model.
• In the future research, polyurethane and polyamide resins will be considered, and
hardeners, catalysts, and coupling agent will be selected based on the selected resin. In
addition, various materials having different chemical chain length and reaction under same
types of materials used in this study will be considered.
• The paper has been revised in lines 312-318: “This study proposed a neural network
framework to predict the curing time/the shear strength and optimize NCA formulation
based on epoxy resins, anhydride hardeners, imidazole and amine catalysts, and silane
coupling agent. In the future research, polyurethane and polyamide resins will be
considered, and data mining will be implementing using matching hardeners, catalysts,
and coupling agents. In addition, in order to generalize the developed neural network
model, it is necessary to develop characteristic indices including chemical chain length
and reaction that can characterize polymer components.”

Reviewer 2 Report

The research is quite interesting but there are a few areas that need to be improved/addressed.
It is suggested to validate the research model with some scientific / data acquired from an authentic source.
The novelty of the research work should be addressed properly.
Comparative analysis of material selection, catalyst and their hardness should be added as there are many tables. 
During machine learning, it is suggested to show a black box/ white box model.
Show machine learning Algorithms, and optimization techniques.
How regression models are selected and what is the % error among each model?

In Tables 2 and 3, there are many variables that show null results i,e, 0, this may cause overfitting results. Kindly justify

On referencing section. there are a few latest papers cited on most recent research. It is strongly recommended to add the latest trends from recent publications, furthermore, ref 27 is not right.

It is strongly recommended to add more results and comparative analysis on selection of materials/ catalyst 

Author Response

Thank you very much for your careful review.

We revised the manuscript according to your advice.

The details are following:

 

The research is quite interesting but there are a few areas that need to be improved/addressed.
It is suggested to validate the research model with some scientific / data acquired from an authentic source.

• Because there are only a few data points in this field, we conducted experiments to gather data points.
• The trained data used in this study were verified using third-party data from another researcher.

The novelty of the research work should be addressed properly.

• The comment for novelty was added in Conclusions.

 

Comparative analysis of material selection, catalyst and their hardness should be added as there are many tables.

• Details of catalysts and all materials were listed in Table 1. The details were from the supplier.

During machine learning, it is suggested to show a black box/ white box model.

• In this study, an artificial neural network (ANN) model was used, and neural network is a black box model. The comment was added in section 2.3.
• The paper has been revised in lines 162-163: “In this study, an artificial neural network (ANN), as a black-box model, was selected to consider complex nonlinear relationships among numerous variables.”

Show machine learning Algorithms, and optimization techniques.

• In this study, the ANN model with one hidden layer with 3 nodes was used. A rectified linear unit (ReLU) activation function was employed. An MSE loss function and adaptive moment estimation (ADAM) optimizer were used. The details were explained in the Sec. 2.3.

How regression models are selected and what is the % error among each model?

• In this study, the ANN model with ReLU activation nodes was used..
• The Mean absolute errors, such as errors of each model, were shown in Table 4. The errors were indicated using the units of curing time (s) and shear strength (MPa).

 

• The calculated MAPEs for the training and validation datasets for curing time and shear strength were shown in Table 4. MAPEs of training and validation datasets for curing time were 24.7 % and 25.4 %, respectively. MAPEs of training and validation datasets for shear strength were 18.8 % and 22.5 %, respectively.

 

• The paper has been added and revised in Table 4.

 

 

• The paper has been revised in lines in 194-195: “The calculated MAEs and MAPEs for training and validation datasets for curing time and shear strength models were shown in Table 4.”
• The paler has been revised in lines in 197-199: “The MAPEs for the training and validation datasets for the curing time were 24.7 % and 25.4 %, respectively. The MAPEs for the training and validation datasets for the shear strength were 18.8 % and 22.5 %, respectively.”

In Tables 2 and 3, there are many variables that show null results i,e, 0, this may cause overfitting results. Kindly justify

• In NCA formation, the number of components in the formulation is limited. This is the reason why the input was very sparse as seen in Tables 2 and 3.
• In the design of metals, the chemical composition is often employed to avoid this kind of sparse issue; however, in polymer formulation, the chemical composition-based approaches are difficult to apply because the properties show severe nonlinearity.
• Fortunately, as shown in Fig. 3, overfitting issues were not happened.

On referencing section, there are a few latest papers cited on most recent research. It is strongly recommended to add the latest trends from recent publications, furthermore, ref 27 is not right.

• References were added from recent publications and revised.

It is strongly recommended to add more results and comparative analysis on selection of materials/ catalyst 

• Comparative analyses for predictions of curing time and shear strength was shown in the case studies.
• In this study, a framework for prediction of the curing time/shear strength and optimization of NCA formulation were demonstrated. However, the number of datapoints in this study were limited, and currently, data mining is progressing.
• After completing the current experiments, a new paper will be submitted.

Round 2

Reviewer 1 Report

The authors have properly addressed my comments, and thus I suggest the manuscript to be considered for publication in its current form. 

Author Response

Thank you very much for your comments.