Tool Wear Monitoring System Using Seq2Seq

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have presented a study using the Seq2Seq model for monitoring tool wear. The tool wear monitoring is an important research area. However, the present work seems ambiguous and shallow.

There needs to be a clear indication of the motivation, and scientific and technological novelty in the introduction. RNNs are already popular and numerous studies on using RNNs to predict tool wear exist.

Describe what top metal and low metal data are.

In the abstract it is written, "The Seq2Seq model outperformed LSTM and GRU with an R2 of ap-14 proximately 0.03~0.037 in step drill data, 0.540.57 in top metal data, and 0.16~0.45 in low metal data" The sentence is very confusing. I think the authors meant R2 improvement. Even then both R2 values are to be given, since 0.03 is very less.

Seq2Seq is again an off-the-shelf model. The study compares 3 models using public/open repository datasets.

Within LSTM there are a lot of hyperparameter adjustments (number of hidden layers, training algorithms, activation functions, loss functions etc) which could have fetched a better result. No comprehensive evaluations on the other state-of-the-art are being done to conclude the better performance of seq2seq.

The study presents no practical machining/validations. In addition, the models and data are all off-the-shelf, without any significant scientific insights on the possible reasons for the better performance of seq2seq (R2 of ~0.8 is still not very good).

I regret to suggest rejection since I cannot see any significant scientific or technological contributions to the existing research in this area.

Author Response

Reviewer 1.

The authors have presented a study using the Seq2Seq model for monitoring tool wear. The tool wear monitoring is an important research area. However, the present work seems ambiguous and shallow.

There needs to be a clear indication of the motivation, and scientific and technological novelty in the introduction. RNNs are already popular and numerous studies on using RNNs to predict tool wear exist.

Describe what top metal and low metal data are.

In the abstract it is written, "The Seq2Seq model outperformed LSTM and GRU with an R2 of ap-14 proximately 0.03~0.037 in step drill data, 0.540.57 in top metal data, and 0.16~0.45 in low metal data" The sentence is very confusing. I think the authors meant R2 improvement. Even then both R2 values are to be given, since 0.03 is very less.

Seq2Seq is again an off-the-shelf model. The study compares 3 models using public/open repository datasets.

Within LSTM there are a lot of hyperparameter adjustments (number of hidden layers, training algorithms, activation functions, loss functions etc) which could have fetched a better result. No comprehensive evaluations on the other state-of-the-art are being done to conclude the better performance of seq2seq.

The study presents no practical machining/validations. In addition, the models and data are all off-the-shelf, without any significant scientific insights on the possible reasons for the better performance of seq2seq (R2 of ~0.8 is still not very good).

I regret to suggest rejection since I cannot see any significant scientific or technological contributions to the existing research in this area.

• Our methodology presents the results of a testbed process to predict wear using EOCR and sensors to acquire data from an old factory line in Changwon, South Korea. This methodology allows the actual wear rate to be accurately predicted through additional sensor attachment and verification. The paper has been revised to fit the overall flow. However, there were limited issues with the model's behavior with the current data.
• This is not an existing open data, but the result of testing with data directly acquired using EOCR and sensors on the production line and applying it to actual monitoring.
• Actual performance is low, but there is room for improvement. All the points you pointed out have been corrected.
• In future research, we plan to conduct various sensor data and field verification.

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript proposed a novel method to predict the cutting tool wear and tool replacement in real time, which is meaningful for the industry. There are some suggestions to be considered:

(1) The training data was collected by sensors as mentioned in Section 3.1, please describe more details about what’s the relation between the collected data and tool wear. How these data can reflect the tool wear features?

(2) Tool wear has different wear mechanisms such as abrasive wear, chemical wear, fatigue wear, and thermal wear. Does the proposed model can recognize different wear behaviors?

(3) What’s the cutting tool materials used in this manuscript? Is the model feasible for different materials and geometries of cutting tools and workpieces?

(4) Some expression need be explained more specifically, such as top metal and low metal.

Comments on the Quality of English Language

Moderate editing of English language required.

Author Response

This manuscript proposed a novel method to predict the cutting tool wear and tool replacement in real time, which is meaningful for the industry. There are some suggestions to be considered:

Thank you for reviewing the paper. Our methodology presents the results of a testbed process to predict wear using EOCR and sensors to acquire data from an old factory line in Changwon, South Korea. This methodology allows the actual wear rate to be accurately predicted through additional sensor attachment and verification. In future research, we plan to conduct various sensor data and field verification.

(1) The training data was collected by sensors as mentioned in Section 3.1, please describe more details about what’s the relation between the collected data and tool wear. How these data can reflect the tool wear features?
-> At present, we believe that current data affects wear and tear and show the results of our analysis. Additional analysis is required by attaching additional sensor data.

(2) Tool wear has different wear mechanisms such as abrasive wear, chemical wear, fatigue wear, and thermal wear. Does the proposed model can recognize different wear behaviors?

-> No, we are currently only trying to predict fatigue wear and thermal wear using current.

(3) What’s the cutting tool materials used in this manuscript? Is the model feasible for different materials and geometries of cutting tools and workpieces?

-> The parts that go into automobile parts are processed from metal.

(4) Some expression need be explained more specifically, such as top metal and low metal.

-> This is explained in Chapter 3, System Structure.

Reviewer 3 Report

Comments and Suggestions for Authors

1. Introduction:

   • Consider refining the connection between the evolving needs of customers, the role of smart factories, and the specific challenges in tool wear prediction. Ensure a seamless transition between these points.

2. Literature Review (Related Works):

   • Add a brief summary or critical analysis after presenting each referenced study. Discuss the strengths and limitations of these studies and highlight how your work contributes to or improves upon existing methodologies.
   • While your literature review is comprehensive, ensure that the studies mentioned are recent and directly related to your research goals.

3. Methodology:

   • Provide more details on the specific models used in the study (LSTM, GRU, Seq2Seq). A brief explanation of why these models were chosen and their relevance to the problem at hand would enhance the methodology section.
   •  

4. • Your experimental setup and results section is informative and provides a clear overview of the methodology used and the outcomes of your study. However, here are a few suggestions to enhance this section:

     1. Provide More Details on Hyperparameters:

        • While you've mentioned some key hyperparameters, consider providing additional details such as the batch size used during training, the number of layers in your LSTM and GRU models, and any other relevant hyperparameters that might impact the model's performance.

     2. Include Training and Validation Performance:

        • If applicable, include not only the performance metrics on the test dataset but also on a validation dataset. This can provide insights into how well your models generalize to unseen data.

     3. Discuss Computational Efficiency:

        • Given the hardware and deep learning framework used, you might want to briefly discuss the computational efficiency of your models. If there were any challenges in terms of training time or resource utilization, consider mentioning them.

     4. Highlight Significance of Results:

        • In the discussion of results, emphasize the practical significance of the improvements observed with the Seq2Seq model. How do these improvements translate into real-world applications? Providing context for the significance of your findings can strengthen your discussion.

     5. Address the Less Noticeable Differences:

        • You mention that for the low metal data, there are marginal differences in MAE and MSE despite a significant difference in R2. Explain why this might be the case and discuss the implications of these subtler differences.

     6. Consider Providing Confidence Intervals:

        • To quantify the uncertainty around your performance metrics, you might consider providing confidence intervals or statistical significance tests, especially if there are close comparisons between models.

     7. Connect Results to Research Questions:

        • Remind the reader of the specific research questions or hypotheses that your study aimed to address. Discuss how your results align with these original objectives.
     8.  

     9. Your conclusion provides a concise summary of the key findings and contributions of your study. Here are a few suggestions to further enhance the conclusion section:

        1. Emphasize Practical Implications:

           • Discuss how the superior performance of the Seq2Seq model in predicting tool wear translates into practical benefits for the manufacturing industry. Highlight the potential impact on minimizing downtime, preventing accidents, and optimizing tool replacement schedules.

        2. Address Limitations and Future Work:

           • Briefly mention any limitations of your study, such as constraints in data collection or assumptions made during the modeling process. Additionally, outline potential avenues for future research or improvements to your methodology. This can demonstrate a forward-looking perspective.

        3. Connect with the Introduction:

           • Revisit the problems and challenges outlined in the introduction and explicitly connect them with the solutions and advancements presented in your study. This reinforces the relevance of your work to the broader context.

        4. Discuss the Monitoring System:

           • Elaborate on the significance of the monitoring system you developed. How does it contribute to the overall tool wear prediction approach, and what benefits does it offer in a real-world manufacturing environment?

        5. Consider the Generalizability of Findings:

           • Comment on the potential generalizability of your findings to different manufacturing settings or types of machinery. Discuss whether the Seq2Seq model's success is likely to extend beyond the specific tools and conditions investigated in this study.

        6. Address Hardware Scalability:

           • Acknowledge that, with improved hardware performance, handling more data becomes feasible. However, discuss the scalability challenges and considerations that might arise when applying your methodology to larger datasets or different hardware configurations.

        7. Highlight the Innovation:

           • Reiterate the innovative aspects of your work. How does the use of a Seq2Seq model, particularly in the context of tool wear prediction, contribute to the existing body of knowledge? Emphasize any novel approaches or techniques introduced.

        8. Closing Statement:

           • End the conclusion with a strong and forward-looking statement that summarizes the overall significance of your study. Reinforce the potential impact on smart manufacturing and the advancement of predictive maintenance strategies.
        9.  
     10.  
5.  
6. •  

Comments on the Quality of English Language

No comments

Author Response

• Our methodology presents the results of a testbed process to predict wear using EOCR and sensors to acquire data from an old factory line in Changwon, South Korea. This methodology allows the actual wear rate to be accurately predicted through additional sensor attachment and verification.
• The paper has been revised to fit the overall flow. However, there were limited issues with the model's behavior with the current data.
• In future research, we plan to conduct various sensor data and field verification.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have now sufficiently addressed all concerns of the reviewer and can now be accepted for publication

Comments on the Quality of English Language

English is fine