Enhancing Moisture-Induced Defect Detection in Insulated Steel Pipes through Infrared Thermography and Hybrid Dataset

Round 1

Reviewer 1 Report

The paper presents an application of a novel methodology that enriches machine learning training through the integration of experimental data with synthetic datasets created by Finite Element Method (FEM) simulations in tackling the challenge of detecting moisture-induced defects in steel pipe insulation to prevent Corrosion Under Insulation (CUI). My detailed comments are as follows:

1. The UNet and FEM used in the paper works very well for the detecting moisture-induced defects in steel pipe insulation. On the other hand, the two method are a well-established method, and the present research is a direct application of this method without new contribution in methodological research.

2. For the above reason, the presentation should be focused on the results. Unfortunately, the presentation is far from acceptable for publication. The material was not properly organized and it is strongly suggested that the authors check carefully the English writing and use standard terminologies in the technical area.

3. The shortcomings of methods such as GANs are mentioned in the manuscript, but only in the textual presentation, and the authors should have included comparative experiments of different methods to highlight the advantages of the highlighted proposed method.

4. The conclusion should be concise and only summarize the most important contribution of the research.

5. In general, there is a lack of explanation of UNet used in the study. Furthermore, an explanation of why the authors did these various experiments should be provided.

It is noted that your manuscript needs careful editing by someone with expertise in technical English editing paying particular attention to English grammar, spelling, and sentence structure so that the goals and results of the study are clear to the reader.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The abstract is rather vague in terms of terminology e.g. "advanced machine learning". What does the author mean by "advanced"? Moreover, the abstract lack information about the actual finding perhaps some numerical values or more precise information.

The equations 1-3 were not explained fully and symbols used are not defined including "p" and "v".

Table 4 shows the comparison between "Experiment" and "FEM_Experiment" models. it’s interesting to note that despite the lower loss and higher Mean IoU, the improvement in F1 score is relatively small. This could be due to the F1 score being more sensitive to the balance between precision and recall. If the “FEM + Experiment” model has improved recall but at the expense of precision (or vice versa), the F1 score may not show as much improvement. One model might be making different types of errors than the other. For example, one model might have more false positives (predicted positive instances that are actually negative), while the other might have more false negatives (actual positive instances that are predicted as negative). These differences can affect the IoU more than the F1 score. So it might not be a clear cut for the FEM+Experiment in terms of being more "accurate".

The authors did not mention about hyperparameters used in their models, nor the architecture of the model used, which are very important for repeatability of the experiment.

More information about the results of the testing could have been included including Confusion Matrix etc.

again, in conclusion, the term "significantly advanced" is a bold statement and in the absence of other models' comparison, this adjective is even more vague. What metrics did the authors used to come up with this "significantly advanced"? is it statistically significant? compared to which model?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

After reading this revised manuscript, I believe that the author has made some changes based on my comments. However, I believe that this manuscript still has significant flaws and there are many issues that need to be addressed and explained:

1. Integration of FEM with UNet is certainly novel, but I do not believe that this integration is sufficiently innovative for a paper. In addition there are many network models other than UNet that can be used to achieve the purpose of detection, but the authors need to experimentally compare the performance of different network models or they will not be able to justify the use of UNet. Unless the authors demonstrate the innovation and justification of this manuscript, this manuscript is more suitable for conference presentation.

2. All the tests in the manuscript are based on simulation data, but the content of the study is oriented to engineering practice. It is well known that there is a difference between simulation and reality. So how to ensure the effectiveness of the proposed method in real engineering, the authors must give the corresponding experimental proof.

3. The authors introduce F1 indexes which seem almost meaningless in the experiment. Without the comparison of different methods, more indexes can only prove that the proposed method can be used, but cannot prove that it is a relatively better choice.

The English of this revised manuscript is an improvement over the previous one, requiring only minor linguistic changes.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors has implemented the feedback and acted accordingly with a satisfactory results.

Author Response

Thanks for your comments and guidelines, it helped a lot to improve the quality of this research.

Round 3

Reviewer 1 Report

The authors provided a satisfactory response to my response. Considering the constraints of the experimental environment, I believe the author has done their best. Furthermore, I sincerely hope that future authors can conduct further research in this field. I think the shortcoming of this manuscript now is that its English proficiency still needs to be improved.

Before this manuscript is accepted, the authors need to invite native English speakers to proofread it.