Optimisation of Aluminium Alloy Variable Diameter Tubes Hydroforming Process Based on Machine Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

You can find comments in the attached document. 
English language needs to be refined (you can use MDPI servces for it, or use lector).

FEM part of the paper is severely lacking, you need to update this part. The neural network part is OK, and you need to "widen" comparison section - both with the FEM and neural network results - make more measuring points.

Tables in the appendix are not large as I have expected and there is no need for them to be in the appendix. Place them in the normal textual part. Check with the Editor for best effect.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

Some textual parts are abruptly ended, with no meaning and explanation. Parts of the text are missing for full understanding of the meaning. Best use professional lectors for advice.

Author Response

Please see the attachment！#Detailed Response to Reviewer-1

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript presents a fast prediction model based on a GA-PSO-BP neural network to predict the hydroforming outcomes of aluminum alloy variable diameter tubes. The work integrates a genetic algorithm with particle swarm optimization and a backpropagation neural network to enhance prediction accuracy. This model is trained and validated against results from finite element simulations and experimental tests.

While the idea is relevant and timely—particularly with the growing importance of machine learning in process optimization—the manuscript has significant issues that should be addressed before it can be considered for publication. These issues fall under the categories of language clarity, technical completeness, result interpretation, and overall presentation quality.

1. The manuscript requires major revision in terms of language. The English is often unclear and grammatically incorrect, which affects the readability and interpretation of the work. For example:

• Line 1 of abstract:
  "In order to study the forming effect of aluminum variable diameter tubes through hydroforming, this paper proposes the introduction of genetic algorithm..."
  This sentence is vague and awkward. A better formulation could be:
  "To predict the forming behavior of aluminum variable-diameter tubes during hydroforming, a genetic algorithm-enhanced particle swarm optimization (GA-PSO) is used to optimize a backpropagation neural network (BP-NN).”
• Unclear fragment:
  "In this paper, using the PSO-BP neural network based on genetic improvement."
  This sentence is incomplete and grammatically incorrect.

A thorough professional language edit is strongly recommended.

2. Section 2.1 lacks sufficient detail.

• There is no information about the experimental setup or testing conditions used to validate the FE simulations or the prediction model.
• The tube thickness is not clearly reported.
• There is no clear description of how the material properties (e.g., plasticity, hardening behavior) are derived from the engineering stress–strain curve and input into Abaqus or the finite element software.

3. The finite element software used is not identified
4. Friction plays a crucial role in tube hydroforming (see https://doi.org/10.1177/09544054231184914), but the friction model and coefficient used in simulations are not clearly described.
5. Failure prediction is not discussed: how is rupture or thinning limit modeled in your simulations?
6. Figure 3:

• The labels are unclear. Please specify which curve corresponds to axial feed and which to pressure.
• The figure is titled as a simulation result, but the word “experiment” is used in the caption or text—please clarify this.
• The conclusions drawn from Figure 3 are not explained—what insight does it offer, and how is it used in your model development?

7. Figures 7, 8, and 12: The image quality is poor, and axis labels and legends are difficult to read. All figures should be high-resolution and professionally formatted.
8. It is unclear how the GA-PSO-BP model was validated. Was cross-validation used? Was a test set separated from the training data?
9. The experimental validation mentions wall thickness and expansion height, but there is no comparison table or figure showing the simulation vs experimental results.
10. How do you justify the optimal process parameters (40 MPa, 4 mm axial feed, μ = 0.15)? Are they compared with other feasible conditions? How were they experimentally verified?
11. Controlling friction in hydroforming is extremely difficult in practice. How did you ensure or control the assumed coefficient (μ = 0.15) during experiments?
12. The conclusion does not sufficiently highlight the novelty of the method or its limitations.
13. Please consider adding discussion on the applicability of the model to other tube geometries, materials, or forming processes.
14. Mention any limitations or future work (e.g., extension to rupture prediction, integration with real-time control, etc.).

Author Response

Please see the attachment！#Detailed Response to Reviewer-2

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

In this article, the authors used selected artificial intelligence tools to analyze the tube hydroforming process. The GA-PSO-BP back-propagation neural network model is constructed, which provides prediction of wall thickness as well as forming height of formed parts based on the results of numerical modeling.

Overall assessment: Reject.

Justification:

The material model was not used properly. In the FEM model, the authors did not take into account the elastic properties of the sheet metal. Without these parameters, mechanical analysis FEA is not possible. The results presented in the paper are therefore not reliable. 

The applied plastic material model was not provided. The authors presented only the engineering stress-engineering strain curve, which does not contain important information in relation to FEM. In FEM, true stress-strain relation is used. Therefore, I again consider that FE-based results are not reliable.

The reliability of the FEM results is also related to the lack of mesh sensitivity analysis. The authors only provided the number of elements in Fig. 2b, but this parameter does not carry any significant information. It is not known whether the FEM mesh was created adaptively or by a structured approach. How was the optimal mesh size selected?

There are also no parameters of the friction model used and a justification for these parameters. It is not known how the boundary conditions were modeled. The pressure conditions and the justification for these conditions, etc., were not presented. Even the thickness of the sheet metal used was not presented.

Another curious approach of the authors is in the sentence: ‘The relevant settings of the finite element model are referred to [9,10].’ They refer here to the works of other authors. Methods should be described with sufficient detail to allow others to replicate and build on published results. The current version of the article does not allow this.

section 2.2. It is not known under what conditions the tensile tests were performed. It is not known whether and how statistical repeatability of the results was ensured.

type line: "fully laminated state". What laminated state did the authors have in mind? I recommend that the authors review professional literature thematically related to the analyzed topic before writing an article.

In figure 5 it is extremely difficult to figure out what the values in the legend between 1100 and 7460 mean.

type lines 209-210: 'Based on the 70 data sets obtained from the previous Latin superlative method, 60 data sets were randomly selected as the training set and 10 data sets as the test set.' It is not known why a separate subset was not planned for the validation set to independently check the learning algorithm. At the same time, Figure 8a shows the convergence for validation. Which is a contradiction in this paper.

If the test set contained 10 data sets (type lines 209-210), why do Figures 7a-c contain a prediction for 11 test set points?

Figure 8. How was performance determined for validation if such a data set was not used in the analyses?

Figure 10. What parameter is defined on the ordinate axis?

Measurement points are completely invisible in Figure 12.

Subsection 4.5 presents experimental results. However, experimental methods have not been presented before.

section 4.1. How was the coefficient of friction 0.15 (type line 351) achieved in the experimental conditions? Avoid general answers. A discussion of the strategy for the shaping process in such a way that the coefficient of friction was equal to 0.15 is required.

Article title: Test material (5A02) is an aluminum alloy, not aluminum.

The resolution and graphic quality of some figures is very poor.

Comments on the Quality of English Language

Looks like it may have been written by a non-native English speaker, if this is the case, it would be important to ask one to proofread the paper. There are a few awkward phrasing.

Author Response

Please see the attachment！#Detailed Response to Reviewer-3

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Everything is corrected in the paper according to cover letter, and by doing so the presentation of results became much clearer and easier to understand to the wider audience.

Reviewer 2 Report

Comments and Suggestions for Authors

After careful evaluation of the responses and the revised manuscript, I would like to suggest the acceptance of the manuscript in its current form.

Reviewer 3 Report

Comments and Suggestions for Authors

After reviewing the revised manuscript, I note that the authors have made an improvement on the old version. The authors have addressed the reviewer's comment properly. I believe that this manuscript is acceptable in the current form.