A Low-Code Visual Framework for Deep Learning-Based Remaining Useful Life Prediction
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors- The paper is well-written, but some sentences are long and could be simplified for better readability. Consider breaking down complex sentences and removing redundant phrases.
Ensure consistent terminology. For example, the paper mentions "Flash framework" (page 5, line 224) but later "Flask framework" (page 9, line 332). This should be corrected to "Flask" throughout.
A more concise version could be considered, for example: "A Low-Code Visual Framework for Deep Learning-Based Remaining Useful Life Prediction".
The abstract is good. To make it even stronger, you could briefly mention the validation on the benchmark NASA dataset.
Section Numbering: There is a typo in the section number on page 8: "3.1.. Modular Design Principles" should be "3.1.3. Modular Design Principles". Similarly, "3.3.5. Key Classes, Interfaces, and Functions" on page 11 should likely be renumbered to 3.3.
The title "3.2. Application Function Module Contact" (page 9) is unclear. Consider renaming it to something more descriptive like "3.2. Application Module Details" or "3.2. Functional Module Implementation".
The text claims the Transformer model shows a good fit visually, but the table shows it has the highest MSE and RMSE by a large margin (0.1485 vs. 0.0136 for LSTM). This is a significant contradiction that must be addressed. Explain why this might be the case.
Author Response
Dear Reviewer,
We sincerely appreciate your meticulous review and the valuable feedback provided on our manuscript. Your insights are of great significance and have offered us a clear - sighted perspective on the areas that require improvement. We have carefully considered each of your comments and suggestions, and are committed to making the necessary revisions to enhance the quality and rigor of our work.
Question1:
The paper is well-written, but some sentences are long and could be simplified for better readability. Consider breaking down complex sentences and removing redundant phrases.
Answer1:
Thank you for your feedback! We have reviewed the text and worked on breaking down any long, complex sentences into shorter, clearer ones. Additionally, we have identified and removed any redundant phrases to enhance the overall clarity. We have used the English editing service recommended by MDPI, which assisted us in revising and polishing the English in the text.
Your suggestions have been valuable in helping us to improve the quality of the manuscript. If you have specific examples or sections in mind, please feel free to share them, and we will address those areas directly. Thank you again for your constructive criticism!
Question2:
Ensure consistent terminology. For example, the paper mentions "Flash framework" (page 5, line 224) but later "Flask framework" (page 9, line 332). This should be corrected to "Flask" throughout.
Answer2:
Thank you for pointing that out! We have corrected the inconsistency and now use "Flask" consistently throughout the paper. Maintaining uniform terminology is important, and we appreciate your attention to detail in helping ensure accuracy. We have reviewed the paper and made the necessary changes to maintain consistency. If you notice any other inconsistencies, feel free to let us know!
Question3:
A more concise version could be considered, for example: "A Low-Code Visual Framework for Deep Learning-Based Remaining Useful Life Prediction".
Answer3:
Thank you for your suggestion regarding a more concise title! We have considered your input, and the proposed title, "A Low-Code Visual Framework for Deep Learning-Based Remaining Useful Life Prediction," now effectively captures the essence of the paper in a straightforward manner. We have evaluated its relevance and clarity in relation to the content of the paper, and it aligns well. We appreciate your help in refining this aspect of the work!
Question4:
The abstract is good. To make it even stronger, you could briefly mention the validation on the benchmark NASA dataset.
Answer4:
Thank you for the positive feedback on the abstract! We agree that mentioning the validation on the benchmark NASA dataset would strengthen the Abstract and provide a clearer picture of the paper's contribution. We have revised the abstract to briefly include this information, highlighting the validation process and its relevance.
Question5:
Section Numbering: There is a typo in the section number on page 8: "3.1.. Modular Design Principles" should be "3.1.3. Modular Design Principles". Similarly, "3.3.5. Key Classes, Interfaces, and Functions" on page 11 should likely be renumbered to 3.3. The title "3.2. Application Function Module Contact" (page 9) is unclear. Consider renaming it to something more descriptive like "3.2. Application Module Details" or "3.2. Functional Module Implementation".
Answer5:
Thank you for your detailed observations regarding the section numbering and titles in the paper! We appreciate your attention to these details, as they are crucial for clarity and organization.
1) The issue with "3.1. Modular Design Principles" has been corrected to “4.1.3. Modular Design Principles”, and we have also adjusted and reviewed the numbering of other sections. Thank you for pointing this out.
2) Regarding the renumbering of "3.3.5. Key Classes, Interfaces, and Functions," as suggested by another reviewer, we have moved the content to the appendix to ensure clarity and logical consistency in the section numbering.
3) For the ambiguous title "3.2. Application Function Module Contact," we have changed it to "4.2. Application Module Details" to more accurately reflect the content of this section.
Question6:
The text claims the Transformer model shows a good fit visually, but the table shows it has the highest MSE and RMSE by a large margin (0.1485 vs. 0.0136 for LSTM). This is a significant contradiction that must be addressed. Explain why this might be the case.
Answer6:
Thank you for bringing this important contradiction to our attention. It is crucial to ensure that the reported findings in the text align with the data presented in the tables.
To address this inconsistency, we have revised the relevant sections of the paper to clarify that the LSTM model is, in fact, the best-performing model according to the metrics of Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). Additionally, we have modified the conclusion to emphasize that the LSTM model demonstrates superior performance.
Thank you for taking the time to review my manuscript and provide valuable comments and suggestions. Your feedback has played a crucial role in improving the quality of my paper, and I will continue to strive to enhance my research. Once again, I deeply appreciate your support and assistance with this study.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis paper tackles an important practical problem in industrial maintenance by developing a low-code platform for remaining useful life (RUL) prediction using deep learning. The authors recognize that while deep learning shows great promise for equipment health monitoring, the high technical barriers prevent many practitioners from adopting these methods. Their solution is a user-friendly Streamlit-based tool that packages multiple neural network models (LSTM, CNN, GRU, Transformer) into an intuitive interface, allowing users to perform RUL predictions without extensive programming knowledge. They validate their approach using NASA's turbofan engine dataset and demonstrate that users can achieve reasonable prediction accuracy through simple point-and-click operations. While the engineering contribution is solid and addresses a real need in the industry, the paper would benefit from stronger technical novelty and more comprehensive evaluation.
1. Beef up the technical contribution - Right now this feels more like a software demo than a research paper. Consider adding novel algorithmic improvements, automated model selection features, or innovative ensemble methods that go beyond just wrapping existing models in a GUI.
2. Relying solely on the NASA turbofan data is pretty limiting. Add bearing datasets, battery degradation data, manufacturing equipment, or other industrial cases to show this actually works across different domains and isn't just tuned for one specific problem.
3. Please compare against real competition. The paper doesn't really position itself against existing tools. What about commercial PHM software, other open-source solutions, or even general AutoML platforms? Show us why someone should choose this over alternatives.
4. Please get actual users involved - The "low-code" claim needs validation from real users. Run usability studies with engineers who would actually use this tool. How much time does it really save? What's the learning curve? Get some testimonials or case studies from industry folks.
5. Please expand the evaluation metrics - MSE and RMSE are fine but pretty basic for RUL prediction. Add prognostic horizon, early/late prediction penalties, uncertainty quantification, and other metrics that actually matter for maintenance decision-making.
6. Please fix the writing and presentation - There are quite a few grammatical issues and the flow could be smoother. The figures need better quality and more informative captions. Also, separate out a proper related work section instead of cramming everything into the intro.
7. Please deployment in industrial environments with security requirements? Discuss computational limits, memory usage, and real-world deployment challenges.
This paper can be accepted after all the above aspects are resolved.
Author Response
Dear Reviewer,
Thank you very much for your meticulous review and the valuable feedback on our manuscript. We truly appreciate the time and effort you have dedicated to providing such in - depth and constructive comments, which have offered us clear directions for improvement.
Question1:
Beef up the technical contribution - Right now this feels more like a software demo than
a research paper. Consider adding novel algorithmic improvements, automated model
selection features, or innovative ensemble methods that go beyond just wrapping existing
models in a GUI.
Answer1:
Thank you for your valuable feedback. We understand that the current paper focuses more on showcasing the software's functionality, rather than adequately emphasizing its technical contributions. To address this, in the revised version, we have added an overview of the algorithmic principles and techniques in Section 3.1 to enhance the technical depth of the paper. Additionally, to demonstrate the applicability of our approach, we have included test validations based on datasets in Section 5.2. In the future, we will further strengthen the research aspect of the paper by exploring and incorporating novel algorithmic improvements and automated model selection features to enhance the technical value of the work.
Question2:
Relying solely on the NASA turbofan data is pretty limiting. Add bearing datasets,
battery degradation data, manufacturing equipment, or other industrial cases to show this
actually works across different domains and isn't just tuned for one specific problem.
Answer2:
Thank you for your suggestion. We also recognize that relying solely on the NASA turbofan engine dataset has certain limitations. To validate the applicability of our method across different domains, we have included another battery lifespan prediction dataset, constructed from experimental data provided by the Hawaii Natural Energy Institute (HNEI), in Section 5.1 of the paper. Through the experimental results obtained on this dataset, we aim to demonstrate that our method is not just targeted at a specific problem but is widely applicable to various industrial applications. In the future, we will further explore and test additional datasets, such as bearing fault data, battery degradation data, and manufacturing equipment data.
Question3:
Please compare against real competition. The paper doesn't really position itself against
existing tools. What about commercial PHM software, other open-source solutions, or even
general AutoML platforms? Show us why someone should choose this over alternatives.
Answer3:
Thank you for your feedback. We have strengthened our research on existing tools, particularly through a comparative analysis with other open-source solutions and general AutoML platforms. In the comparison, we evaluate them across multiple dimensions, such as performance, ease of use, and deployment costs, while highlighting the unique advantages of our tool. Our platform is primarily aimed at beginners, designed to help them to quickly get started and understand the application of deep learning in Remaining Useful Life (RUL) prediction. To better illustrate this, we have further emphasized the educational guidance aspect of the platform in the paper and provided a detailed description of how customizing the network layers can achieve this goal.
Question4:
Please get ctual users involved - The "low-code" claim needs validation from real users.
Run usability studies with engineers who would actually use this tool. How much time does
it really save? What's the learning curve? Get some testimonials or case studies from
industry folks.
Answer4:
Thank you for your suggestion. We agree that the effectiveness of low-code platforms needs to be validated through actual user testing. In the future, we plan to conduct usability studies by inviting engineers who have used the tool to participate in testing, evaluating the time saved and the learning curve. We will also seek feedback from industry experts, including case studies and user testimonials, in order to demonstrate the tool's value in real industrial settings. If you are interested in our research, you can continue to follow our progress.
Question5:
Please expand the evaluation metrics - MSE and RMSE are fine but pretty basic for RUL
prediction. Add prognostic horizon, early/late prediction penalties, uncertainty
quantification, and other metrics that actually matter for maintenance decision-making.
Answer5:
Thank you for your suggested improvements. We have added more evaluation metrics that are meaningful for RUL prediction. In addition to MSE and RMSE, we have incorporated metrics like R² and WMAPE, which are crucial for maintenance decision-making. These new evaluation methods are presented in detail in Section 5.2 of the revised paper, along with an explanation of how they enhance the practical application value of the model. In future research, we will consider validating more evaluation metrics.
Question6:
Please fix the writing and presentation - There are quite a few grammatical issues and the
flow could be smoother. The figures need better quality and more informative captions.
Also, separate out a proper related work section instead of cramming everything into the
intro.
Answer5:
Thank you for your feedback. We have conducted a comprehensive grammar revision of the paper and optimized its structure to make it more fluent. We have used the English editing service recommended by MDPI, which assisted us in revising and polishing the English text.
Additionally, we have improved the quality of the figures and tables, ensuring their clarity and informativeness, and enhanced the captions. We have also created a separate "2. Related Work" section, instead of placing all the content in the Introduction, in order to improve the paper's readability and structure.
Question7:
Please deployment in industrial environments with security requirements? Discuss
computational limits, memory usage, and real-world deployment challenges.
Answer7:
Thank you for your feedback. We have provided a detailed description of the software's positioning and applicable scenarios in the paper, particularly focusing on the characteristics of beginners. Additionally, we have included more experimental data on training time and efficiency, so that readers can better understand the performance of different models in practical use. We have also described the computational constraints and memory usage in Section 5.3 of the paper. If you have any specific suggestions or additional details you would like to be included, please feel free to let us know.
Thank you for taking the time to review my manuscript and provide valuable comments and suggestions. Your feedback has played a crucial role in improving the quality of my paper, and I will continue to strive to enhance my research. Once again, I deeply appreciate your support and assistance with this study.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe manuscript investigates deep learning techniques—specifically LSTM, CNN, and GRU —for remaining useful life (RUL) prediction, and presents a visual, low-code application developed using PyTorch and Streamlit. However, the current version of the manuscript places disproportionate emphasis on the software implementation aspects, while underdeveloping the scientific and engineering rationale behind the work.
Major Comments:
- Lack of Research challenge identification:
- The introduction does not adequately establish the need for the study. What specific gaps in the current literature or industry practice does this work address? Apart from the aspect of low-code applications.
- The relevance of RUL prediction across different equipment types (e.g., turbines, pumps, bearings) is not sufficiently discussed. Does the proposed framework generalize across domains? If not, what are the limitations?
- There is limited discussion of the operational context in which RUL prediction is valuable
- Insufficient Literature Review:
- The literature review is too brief and omits important recent work on RUL prediction using deep learning. Key papers on LSTM/GRU for prognostics, hybrid models, or domain adaptation should be cited and critically discussed.
- Similarly, existing low-code/no-code ML platforms are not mentioned or compared. This leaves the reader with little understanding of the novelty or utility of the proposed Streamlit-based application.
- Overemphasis on Implementation Details and paper structure:
- A large portion of the manuscript focuses on Streamlit/PyTorch code structures and UI features. While the practical implementation is useful, it overshadows the theoretical, methodological, and experimental contributions.
- Consider moving technical implementation details to an appendix or supplementary material, and shift focus in the main text to modeling rationale, architecture design choices, and performance analysis.
- Evaluation:
- The manuscript lacks a thorough description of the datasets.
- There is little discussion of model validation.
- Can this software be improved and how?
Minor Comments:
- Line 56, what IR represent?
- The RUL abbreviation has been introduced, keep using RUL, for example, line 89.
- Reduce the length of section 3
Author Response
Dear Reviewer,
Thank you very much for your meticulous review and valuable feedback on our manuscript. We highly appreciate the time and effort you have dedicated to providing such in - depth and constructive comments, which will undoubtedly help us improve the quality of our work significantly.
Major Comments:
Question1:
The introduction does not adequately establish the need for the study. What
specific gaps in the current literature or industry practice does this work
address? Apart from the aspect of low-code applications.
Answer1:
Thank you for your feedback on the Introduction. In the revised version, we have more clearly articulated the necessity of this study in Section 1. The platform we have developed for remaining useful life (RUL) prediction based on deep learning is primarily targeted at beginners, helping them learn how to use deep learning for predicting the RUL of equipment. To achieve this goal, we allow users to customize the number of network layers in the model. Additionally, as the software is implemented in Python, we have provided an introduction to the key classes, interfaces, and functions in the appendix to help users to understand how to build their own neural network models.
Question2:
The relevance of RUL prediction across different equipment types (e.g., turbines, pumps, bearings) is not sufficiently discussed. Does the proposed framework generalize across domains? If not, what are the limitations? There is limited discussion of the operational context in which RUL prediction is valuable.
Answer2:
Thank you for the valuable comments from the reviewers. These questions directly address the key points of our study. In addition to the low-code application aspect, we also address other important issues related to RUL prediction. In Section 5.2, we have added dataset validation, emphasized the significance of RUL prediction across different types of equipment, and verified whether the proposed framework is generalizable across domains. Furthermore, we have expanded the review of the technical applications of RUL prediction in Section 2.1.
Question3:
The literature review is too brief and omits important recent work on RUL prediction using deep learning. Key papers on LSTM/GRU for prognostics, hybrid models, or domain adaptation should be cited and critically discussed. Similarly, existing low-code/no-code ML platforms are not mentioned or compared. This leaves the reader with little understanding of the novelty or utility of the proposed Streamlit-based application.
Answer3:
Thank you for your valuable suggestions! We highly value the literature review section, as it is crucial for understanding the research background and existing work. In the revised version, we have added Chapter 2, providing a supplementary literature review on relevant models and a survey of the existing literature on low-code platforms. If you have any specific suggestions or additional references that should be included, please feel free to let us know, and we will further improve this section.
Question4:
A large portion of the manuscript focuses on Streamlit/PyTorch code structures and UI features. While the practical implementation is useful, it overshadows the theoretical, methodological, and experimental contributions. Consider moving technical implementation details to an appendix or supplementary material, and shift focus in the main text to modeling rationale, architecture design choices, and performance analysis.
Answer4:
Thank you for your valuable suggestions! We recognize that the manuscript somewhat overemphasized the implementation details and technical features of Streamlit/PyTorch. In the revised version, we have reduced the focus on these aspects and shifted the emphasis to discussing the theoretical framework, methodology, and experimental results. In Section 3.1, we have added an overview of the relevant technical principles and, in Section 5.2, we have included dataset validation. Additionally, we have moved some of the technical implementation details to the Appendix or Supplementary Materials, ensuring that the main text is more focused on discussing modeling principles, architectural design choices, and performance evaluation.
Question5:
The manuscript lacks a thorough description of the datasets. There is little discussion of model validation.
Answer5:
Thank you for your valuable suggestions!We have provided detailed information about the publicly available dataset from the NASA Ames Prognostics Center of Excellence used in the paper in Section 5.1, including its source, characteristics, and how it was used for model training and validation.In Section 5.2, we have further refined the discussion on the performance of different models on this dataset, ensuring that readers can clearly understand our experimental results. If you have any specific suggestions or additional details that should be included, please feel free to let us know.
Question6:
Can this software be improved and how?
Answer6:
Thank you for your feedback. While using popular large language models for predicting remaining useful life (RUL) could improve prediction accuracy, as mentioned at the beginning, the goal of this software is to help beginners learn about RUL prediction based on deep learning. Therefore, this feature has not been included. However, future versions of the software can incorporate additional evaluation metrics, offer more model options, and implement other improvements.
Minor Comments:
Question1:
Line 56, what IR represent?
Answer1:
Thank you for your feedback."IR" stands for industrial robots, and we have clarified the meaning of this term to avoid any confusion.
Question2:
The RUL abbreviation has been introduced, keep using RUL, for example, line 89.
Answer2:
Thank you for your feedback. We have ensured that the abbreviation "RUL" is consistently used throughout the manuscript.
Question3:
Reduce the length of section 3
Answer3:
Thank you for your feedback.We have adjusted Section 3 to improve its readability by removing unnecessary details and focusing on the most relevant content.
Thank you for taking the time to review my manuscript and provide valuable comments and suggestions. Your feedback has played a crucial role in improving the quality of my paper, and I will continue to strive to enhance my research. Once again, I deeply appreciate your support and assistance with this study.
Author Response File: Author Response.pdf