PSO-BP-Based Morphology Prediction Method for DED Remanufactured Deposited Layers
Round 1
Reviewer 1 Report
See attached, please.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Kindly find attached my comments.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The main Comment: We need more than 80 samples to analyze the dataset with AI. But in evaluating your dataset, there were 30 samples in tables 1 and 2. The authors should increase the dataset.
1- The authors use BP neural network; why did they not use Feed forward or Multiple layer perceptron?
2- The author used the PSO algorithm; why did they not use other algorithms similar to genetic, firefly, and whale optimization algorithms?
After the introduction, the authors should add the following parts: (Because the instruction is not good)
- Background study
- Methodology
- Dataset
- Results and discussions
- Conclusion
3- How many topologies are used for analyzing BP models? the authors should evaluate different topology
4- There is no validation with other papers or algorithms; the authors need to consider this accuracy with different models.
5- I suggest evaluating models; the authors use statistical indexes, including AAE, MSE, VAF, etc.
Author Response
Dear reviewer,
Thank you very much for your valuable comments.
These comments can significantly improve the quality of our manuscript and provide the important guiding significance to future research. We have studied these comments carefully and revised the manuscript according to suggestions.
The following is a point-to-point response to the comments.
The main Comment: We need more than 80 samples to analyze the dataset with AI. But in evaluating your dataset, there were 30 samples in tables 1 and 2. The authors should increase the dataset.
Respond: Thank you for your valuable suggestion. The design of 25 experiments in this article was carefully considered, for the following specific reasons:
The DED experiment is difficult to conduct and it is challenging to obtain a large amount of sample data. As the experimental results exhibit strong regularity, the use of 25 sets of experimental data for model training has a certain degree of persuasiveness.
The purpose of proposing the PSO-BP model in this article is to demonstrate its good prediction accuracy. In the future, with further research, the training dataset will be further increased and the model prediction accuracy will be further improved. The results of this article have already demonstrated the feasibility of the proposed method for predicting sedimentary layer topography. Data set issues will be further improved in future studies.
1- The authors use BP neural network; why did they not use Feed forward or Multiple layer perceptron?
Respond: Thank you for your positive comment. In the manuscript, the BP neural network is a type of feedforward neural network that uses backpropagation algorithm to update weights and biases, and does not require the use of feedforward perceptron or multilayer perceptron. Based on our experience with neural network learning and the specific application scenario of additive manufacturing, we believe that the BP neural network is more suitable.
2- The author used the PSO algorithm; why did they not use other algorithms similar to genetic, firefly, and whale optimization algorithms?
Respond: Thank you for your valuable comment. Compared to genetic algorithms (GA), firefly algorithms, and whale optimization algorithms (WOA), we think that PSO has several benefits as below:
- Ease of Implementation: PSO is easy to implement compared to other optimization algorithms. The algorithm only requires a few parameters to be set, such as the number of particles and the maximum number of iterations.
- Fast Convergence: PSO has the ability to converge quickly to the optimal solution. This is because the particles in PSO move towards the best solution found by the swarm and tend to explore the search space efficiently.
- Robustness: PSO is a robust algorithm that can handle noisy and complex optimization problems. It is less likely to get trapped in local optima than other optimization algorithms.
- Less Parameter Tuning: PSO has fewer parameters to tune compared to other optimization algorithms. This makes it easy to apply and reduces the time required for optimization.
- Wide Applicability: PSO can be used to optimize a wide range of problems, including linear and nonlinear optimization, discrete and continuous optimization, and single-objective and multi-objective optimization.
After the introduction, the authors should add the following parts: (Because the instruction is not good)
- Background study
- Methodology
- Dataset
- Results and discussions
- Conclusion
Respond: Thank you for your positive suggestion. This manuscript was transferred to the Sustainability journal through the ICoR conference, so it was written in accordance with the conference's required template (Introduction-results-discussion-conclusion-methods), resulting in some structural confusion. Currently, the format of the entire manuscript has been revised according to reviewers` requirements
3- How many topologies are used for analyzing BP models? the authors should evaluate different topology
Respond: Thank you for your good suggestion. We use three topologies for analyzing the BP models. Because we believe that the three-layer topology of the BP model is relatively mature and can meet the prediction needs. We have conducted experiments with a four-layer topology and found that increasing the number of layers in the network does not improve the prediction accuracy very well. On the contrary, it will reduce the training speed of the neural network. After comprehensive consideration, we used a three-layer topology structure in the research.
4- There is no validation with other papers or algorithms; the authors need to consider this accuracy with different models.
Respond: Thank you for your valuable suggestion. We have added a paragraph in the Discussion section that compared the predictive accuracy of our model in the article with the predictive accuracy of models presented in references 25 and 27. The conclusion was that the proposed PSO-BP model outperformed the models presented in those references when it comes to predicting the morphology of deposited layers.
5- I suggest evaluating models; the authors use statistical indexes, including AAE, MSE, VAF, etc.
Respond: Thank you for your positive comment. The models have been evaluated now. the AAE, MSE and VAF of the models can be found in Table 6.
Round 2
Reviewer 2 Report
Good job!
Author Response
Thank you. We will work harder to improve the manuscript.
Reviewer 3 Report
1-About the Main comments, I need to be more convinced.
2-About different neural networks, You should put the results for other ANN models and compare them in your paper.
3-About different optimization algorithms, You should put the results for other Algorithms and compare them in your paper.
Author Response
Dear reviewer,
Thank you very much for your valuable comments.
These comments can significantly improve the quality of our manuscript and provide the important guiding significance to future research. We have studied these comments carefully and revised the manuscript according to suggestions.
The following is a point-to-point response to the comments.
1-About the Main comments, I need to be more convinced.
Respond: Thank you for your valuable suggestion. we have collected more experimental data from within the research group. The number of test samples has been updated to 120.
2-About different neural networks, You should put the results for other ANN models and compare them in your paper.
Respond: Thank you for your positive comment. We conducted a horizontal comparison of the prediction results and statistical indicators of PSO-BP, CNN, and RNN, which can be found in Tables 3 and 4 of the manuscript.
3-About different optimization algorithms, You should put the results for other Algorithms and compare them in your paper.
Respond: Thank you for your positive suggestion. We conducted a horizontal comparison of the prediction results and statistical indicators of PSO-BP, BP, GA-BP, and GWO-BP, which can be found in Tables 3 and 4 of the manuscript.
Author Response File: Author Response.pdf
Round 3
Reviewer 3 Report
Dear Editor
As a response to my comment, the author provided a suitable response. Lastly, the authors may share their dataset in the appendix for future research. As a result of this, it will be possible to compare new research to the dataset in order to make comparisons.
Author Response
Dear reviewer,
thank you for your valuable suggestions. We have uploaded the dataset of 120 groups of DED experiments in the attachment for your review. Please check it.
