Ductile–Brittle Mode Classification for Micro-End Milling of Nano-FTO Thin Film Using AE Monitoring and CNN

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Review Summary:

The authors have conducted commendable research to formulate the application of CNN (AI) for efficiently classifying ductile-brittle modes using a deep learning algorithm by interpreting Acoustic Emission (AE) data from end-milling operation. Fluorine-doped tin-oxide thin films were end-milled at varying milling parameters, causing differential high-frequency waves that generated acoustic emissions.  While the research is of practical significance, some concerns need to be addressed to enhance the quality of the publication. Some suggestions are made here to benefit the research community and enhance the article's content.

1. For the Introductory statement, “Metallic thin films have limited use because of their low permeability and poor durability. In contrast, oxide semiconductors have high mechanical properties, chemical resistance, and low resistivity.”, lines 38-40 look misleading. Why is there a reference to low permeability- was the author referring to implied transparency? Another aspect of Low resistance in Oxide semiconductors, as opposed to the previous statement about metallic thin films, is whether it is electrical resistance or roughness? If low resistance property is not of significance for this research, can these statements be rephrased/omitted?
2. On Line 49, is comparing the Photolithography process with end milling suitable?
3. Line 115-116, reference to “Materials can be quantified for their brittle/ductile properties through indentation testing [8]”, what is the use of this reference in this article? Is indentation testing used anywhere in the article? Is it needed?
4. What is the basis of using r_e = 1 μm on line 200
5. Line 213 - The ductile/brittle mode distinction of the machined surface was made using an optical microscope. – Is OM used for observation only, or is it suggested here as a criterion for mode distinction?
6. Lines 274-275 clearly state that the End-Mill tool Machining analysis was not useful in identifying ductile/brittle modes; Thus, what is the relevance of all the end-milling machining theory discussed? It's good readers get the understanding, but then the focus of this article which is to demonstrate that the application of CNN is diluted; thus, it is recommended to specify the equations and parametric conditions derived, and move the explanation to the “Supplemental Information” section.
7. What is the correlation of RHS in Equation 9 and Equation 10? Was V(t): AE signal calculated, or is the AE sensor output? If this is AE sensor output, it is suggested to move all end mill machining theory, i.e., Sections 2 and 3, to “Supplemental Information”, keeping only an account of the formulae and parameters selected to run the End Mill experiment.
8. Can a flow chart of the AE Signal file format be submitted to CNN, and data formats as it is processed at each step, as per the algorithm, be shared?
9. How can 97.37% ductile/brittle mode classification efficiency of CNN be translated to the process control of the End Milling operation leading to improved quality of FTO thin film? Is not there any necessity of this explanation?

Author Response

Responses to reviewers

On behalf of my co-authors, I extend our sincere gratitude to the reviewers for their comprehensive and insightful feedback. We have revised the manuscript thoroughly based on their valuable suggestions and comments. We hope these revisions have improved the paper to their satisfaction. Our responses to their specific comments, suggestions, and queries are detailed below.

Reviewer #1: The authors have conducted commendable research to formulate the application of CNN (AI) for efficiently classifying ductile-brittle modes using a deep learning algorithm by interpreting Acoustic Emission (AE) data from end-milling operation. Fluorine-doped tin-oxide thin films were end-milled at varying milling parameters, causing differential high-frequency waves that generated acoustic emissions.  While the research is of practical significance, some concerns need to be addressed to enhance the quality of the publication. Some suggestions are made here to benefit the research community and enhance the article's content.

1. For the Introductory statement, “Metallic thin films have limited use because of their low permeability and poor durability. In contrast, oxide semiconductors have high mechanical properties, chemical resistance, and low resistivity.”, lines 38-40 look misleading. Why is there a reference to low permeability- was the author referring to implied transparency? Another aspect of Low resistance in Oxide semiconductors, as opposed to the previous statement about metallic thin films, is whether it is electrical resistance or roughness? If low resistance property is not of significance for this research, can these statements be rephrased/omitted?
   
   

: We agree that the original expression may have caused confusion. Accordingly, the phrase "low permeability" has been revised to "low optical transmittance" to more accurately reflect the intended meaning. In addition, the reference to the "low resistance" of oxide semiconductors has been removed, as this electrical property is not directly relevant to the scope of the present study. Instead, we have rephrased the sentence to emphasize the key characteristics of oxide semiconductors—namely, their high optical transmittance, mechanical robustness, and chemical durability—which are more pertinent to the context of this work. (line 40~44)

2. On Line 49, is comparing the Photolithography process with end milling suitable?
   
   

: We fully acknowledge that photolithography and micro end-milling are fundamentally different in terms of their processing mechanisms. However, in the context of process controllability and applicability to complex geometries, micro end-milling offers distinct advantages over conventional photolithography, especially for three-dimensional fabrication of brittle thin films if actual depth of cut is less than ductile brittle transition value. To better clarify this point and avoid potential misunderstanding, the sentence in line 56 has been revised to highlight micro end-milling as a controllable process technology, emphasizing its strength in precise machining control.

3. Line 115-116, reference to “Materials can be quantified for their brittle/ductile properties through indentation testing [8]”, what is the use of this reference in this article? Is indentation testing used anywhere in the article? Is it needed?
   
   

: Although indentation testing was not employed in this study, the sentence was originally intended to explain that mechanical properties obtained from indentation tests can be used to estimate brittle/ductile behavior.

However, as the statement may disrupt the overall flow of the manuscript, it has been removed as per the reviewer’s suggestion.

4. What is the basis of using re = 1 μm on line 200
   
   

: Since the tool manufacturer (JJtools) did not provide the specification for the cutting edge nose radius, we measured the edge radius of end mills with diameters of 0.5, 0.7, and 1.0 mm using an optical microscope. The average value obtained from these measurements was approximately 1~1.5 μm, which was adopted as the representative edge radius in the analysis. Details of this procedure have been described in lines 155~156, 177~181 of the revised manuscript.

5. Line 213 - The ductile/brittle mode distinction of the machined surface was made using an optical microscope. Is OM used for observation only, or is it suggested here as a criterion for mode distinction?
   
   

: The machining modes were identified based on the surface images obtained using an optical microscope (OM), which served not only for observation but also as a criterion for mode classification. The specific visual criteria used to distinguish between ductile, partial brittle, and brittle modes have been clearly described in lines 194~197 of the revised manuscript. In addition, representative optical microscope images have been included on page 7 to support the classification and provide visual reference for each mode.

6. Lines 274-275 clearly state that the End-Mill tool Machining analysis was not useful in identifying ductile/brittle modes; Thus, what is the relevance of all the end-milling machining theory discussed? It's good readers get the understanding, but then the focus of this article which is to demonstrate that the application of CNN is diluted; thus, it is recommended to specify the equations and parametric conditions derived, and move the explanation to the “Supplemental Information” section.
   
   

: We would like to clarify that the statement in lines 274~275 of the original manuscript refers to the limited effectiveness of conventional AE signal analysis methods (such as AE RMS and FFT analysis) in distinguishing between ductile and brittle machining modes in our specific experimental context. As discussed in the manuscript, the theoretical basis presented emphasizes that the critical feed per tooth—determined by the machining conditions—is a key factor influencing the machining mode. The application of the CNN model was therefore proposed to classify the machining modes defined by this criterion. To avoid potential confusion, the discussion of AE RMS and FFT analysis has been moved to Appendix A, as it may interrupt the logical flow of the main text. We hope this revision resolves any misunderstanding, and we would be happy to provide further clarification if needed.

7. What is the correlation of RHS in Equation 9 and Equation 10? Was V(t): AE signal calculated, or is the AE sensor output? If this is AE sensor output, it is suggested to move all end mill machining theory, i.e., Sections 2 and 3, to “Supplemental Information”, keeping only an account of the formulae and parameters selected to run the End Mill experiment.
   
   

: The purpose of Equations (9) and (10) was to illustrate the correlation between AE energy and MRR. However, we acknowledge that the omission of intermediate derivation steps may have made the relationship unclear. To improve clarity, we have added supplementary equations (Equation (5), Equation (7)) to explain the connection more explicitly and described how AE energy can be used to monitor the material removal process. V(t) represents the output signal from the AE sensor. Following the reviewer’s suggestion, we have relocated the theoretical background on end milling from Sections 2 and 3 to the Appendix to keep the main text focused on the experimental model and its application.

8. Can a flow chart of the AE Signal file format be submitted to CNN, and data formats as it is processed at each step, as per the algorithm, be shared?
   
   

: To address this point, we have revised the original Figure 13 and newly added Figure 14, which illustrates the data processing flow of the AE energy graph image files used as inputs to the CNN. This new figure provides a clearer overview of how the AE signal is transformed and handled at each stage of the classification process within the CNN framework.

9. How can 97.37% ductile/brittle mode classification efficiency of CNN be translated to the process control of the End Milling operation leading to improved quality of FTO thin film? Is not there any necessity of this explanation?
   
   

: We have revised the conclusion section to clarify the practical implications of the CNN-based process model developed in this study. Specifically, we have added additional explanations in lines 376~390 regarding how the classification model can contribute to the control of end-milling operations and its potential applicability in relevant fields such as selective patterning of FTO thin films.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

(1) The authors used Bifano’s model to calculate the critical depth of the brittle-to-ductile transition, but this model has been modified by Huang et al. There is a literature as below on the critical depth of the brittle-to-ductile transition involved in precision machining of brittle solids. The authors should cite it in the manuscript to modify the critical depth of the brittle-to-ductile transition: [A] https://doi.org/10.1016/j.ijmachtools.2024.104197 . Eq. (9) in this paper gave the modified model of brittle-to-ductile transition depth.

(2) In Fig. 5, the authors marked that the surface roughness values of Figs. (a) and (b) were 18 nm and 62 nm, respectively. What equipment did the author use to measure the roughness values? Please provide the images of the measurement results. Achieving nanometer-level surface roughness through milling processes is highly challenging.

(3) More optical images of the machined surfaces under different milling parameters should be given in the manuscript. Only two images in Fig. 5 are not enough.

(4) What is the value of the parameter a in Eq. (9)?

(5) The discussion about brittle-to-ductile transition is not enough. More literatures about brittle-to-ductile transition involved in machining of brittle solids should be added in the Introduction section.

Author Response

On behalf of my co-authors, I extend our sincere gratitude to the reviewers for their comprehensive and insightful feedback. We have revised the manuscript thoroughly based on their valuable suggestions and comments. We hope these revisions have improved the paper to their satisfaction. Our responses to their specific comments, suggestions, and queries are detailed below.

Reviewer #2: This study proposed a polyvinylidene fluoride (PVDF)-type acoustic emission (AE) sensor to verify the replacement of AE sensors. Hsu-Nielsen test was employed to collect the signal and determine if the PVDF-type AE sensor can classify direct signal and reflect signal. By training the CNN with AE image after WPT, the algorithm learned from the data can classify the direct and reflect waves of PVDF-type AE sensor. Additionally, transfer learning was employed to increase the validation accuracy and reduce training time. The authors revised a lot and detailly responded to the reviewers. The details have been complemented in the revise. However, there are still some key points that should be noticed.

1. The authors used Bifano’s model to calculate the critical depth of the brittle-to-ductile transition, but this model has been modified by Huang et al. There is a literature as below on the critical depth of the brittle-to-ductile transition involved in precision machining of brittle solids. The authors should cite it in the manuscript to modify the critical depth of the brittle-to-ductile transition: [A] https://doi.org/10.1016/j.ijmachtools.2024.104197. Eq. (9) in this paper gave the modified model of brittle-to-ductile transition depth.
   
   

: We have revised Equation (3) in the original manuscript by incorporating the modified brittle-to-ductile transition model proposed by Huang et al. [A], as suggested. Accordingly, the calculation of the critical feed per tooth has been updated in lines 448~455 based on the revised model.

2. In Fig. 5, the authors marked that the surface roughness values of Figs. (a) and (b) were 18 nm and 62 nm, respectively. What equipment did the author use to measure the roughness values? Please provide the images of the measurement results. Achieving nanometer-level surface roughness through milling processes is highly challenging.
   
   

: The surface roughness was measured using a surface roughness tester (SJ-310 by Mitutoyo) by acquiring surface profiles over the machined regions. To enhance clarity regarding the measured surface conditions, we have added Figure 7, which presents the surface profile graphs of the machined areas. Furthermore, the quantitative roughness values have been specified in line 203 of the revised manuscript.

3. More optical images of the machined surfaces under different milling parameters should be given in the manuscript. Only two images in Fig. 5 are not enough.
   
   

: To improve the clarity of the manuscript and provide a more comprehensive understanding of the machined surface characteristics under various milling conditions, we have added additional optical microscope images in Figures 4 to 6. The corresponding machining parameters and classification criteria for each machining mode have also been described in detail in lines 193~200 of the revised manuscript.

4. What is the value of the parameter a in Eq. (9)?

: In Equation (9) of the original manuscript, the parameter a is an empirical constant introduced to express the proportional relationship between AE power and material removal rate (MRR). It is affected by various factors such as tool condition, geometry, and material properties, and is not treated as a fixed or measured value in this study. Therefore, it does not hold a specific numerical value in our analysis but rather serves to represent the theoretical correlation between AE power and MRR.

5. The discussion about brittle-to-ductile transition is not enough. More literatures about brittle-to-ductile transition involved in machining of brittle solids should be added in the Introduction section.
   
   

: To address this point, we have incorporated a discussion of recent research trends related to the brittle-to-ductile transition (BDT) in the machining of brittle solids into the Introduction section (lines 64~80). These additional references provide a broader and more updated perspective on BDT phenomena, thereby strengthening the theoretical foundation of the present study.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The revised manuscript can be accepted

Author Response

Thank you for your thoughtful comments and constructive feedback.
We have responded thoroughly and revised the manuscript, particularly in the abstract and introduction sections, to improve the clarity and precision of English expressions.
We sincerely appreciate the positive evaluation and the recommendation to accept the revised manuscript. Thank you for your time and thoughtful consideration.