Node-Loss Detection Methods for CZ Silicon Single Crystal Based on Multimodal Data Fusion

Round 1

Reviewer 1 Report

The paper deals with the Research on Node Loss Detection Method of CZ Silicon Single

Crystal Based on Multimodal Data Fusion. 

According to the reviewer, the paper is worth publishing at Sensors Journal, 

but corrections are needed and then the paper can be accepted for publication in the journal.

While the authors have made considerable research effort, 

the presentation of the paper must be improved. 

Additionally make the following corrections to the manuscript:

Comment 1

Line 10

In the CZ (Czochralski) method

The authors should replace

In the Czochralski (CZ) method

Line 68

the controlled object, Finally, the complementarity

Extended text editing

the controlled object. Finally, the complementarity

Extended text editing:

The authors must format the paper according to the journal's instructions

Line 45

Figure1 The whole normal ingot(left), node normal (middle) , node loss (right) 

The authors must replace

Figure 1. The whole normal ingot (left), node normal (middle), node loss (right) 

Line 213

Figure2 Diameter difference between normal and node loss

The authors must replace

Figure 2. Diameter difference between normal and node loss

The same comment for the rest of the Figures and the Tables.

Figures 11, 13 and 15

Input(V).......

The authors must replace

Input (V).......

Lines 99 - 100

work(MMFN),

The authors must replace (insert a space)

work (MMFN),

Comment 2

Lines 16, 18, 75, 83, 104, 200, 201, 240, 243, 249, 255, 256, 258, 296, 311, 316, 317, 318, 339, 339, 360, 373, 374, 417, 431, 444, 454, 456, 463, 466, 469, 475 and 477 

It is not so good to use the word "we".

The authors must rephrase.

Comment 3

The authors must comment the ref. [5] (is missing).

Comment 4

Line 117, 119, 121, 124, 128, 138 and 147

There is not a author at the Ref. [8], [9], [10], [11], [13], [14] and [16]

The authors must insert the "et al.", for examble

Pei [8]

The authors must replace

Pei et al. [8]

Comment 5

The authors must check:

Lines 166 - 167

such as Alexnet proposed in 2012 [21], (or 2017?)

Line 167

Resnet proposed in 2015[23].

The authors must check:

Resnet proposed in 2016 [23].

Comment 6

The authors must give more details how the Figure 2 occur, and they must insert the axis x and axis y, the units and explain the 104.

The authors must give more details how the Figure 3 occur (experiment with the using equipment).

Comment 7

Line 224

The authors must give more details for the limits of the diameter, pulling speed, temperature signal and the meniscus image data.

Line 228

The authors must give more details for the equipment (type, model)

Comment 8

The authors must give a typical result for the experiment.

Also, the authors must give the whole results for all the data in the Supplementary Materials. 

Major problem:

How the paper is presented, is like a closed box, without being able to check the reliability of the method proposed by the authors.

Comment 9

Line 329

The "Kaiming He et al. in 2015." is not at the References. The authors must insert the paper at the References.

Comment 10

Line 355

uses Adam as the overall

The authors must explain the word "Adam" with more details.

Comment 11

Line 357

the batch size of all datasets is set to 32, and the epoch is set to 20.

The authors must explain why they choose those values.

Comment 12

The authors must format the References according to the journal's instructions

References should be described as follows, depending on the type of work:

Journal Articles:

1. Author 1, A.B.; Author 2, C.D. Title of the article. Abbreviated Journal Name Year, Volume, page range.

The authors must delete the "[J]" and "C".

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The work titled ' Research on Node Loss Detection Method of CZ Silicon Single  Crystal Based on Multimodal Data Fusion ' is related to the detection of loss of monocrystalline silicon nodes in industrial objects. Based on the analysis of diameter, temperature, pulling speed, and two-dimensional images of the bent sample surface, the authors have developed a methodology for determining the presence or absence of monocrystalline silicon node loss. The methodology is largely based on machine learning algorithms. Satisfactory agreement has been found between predictive data and experimental data. In my opinion, the work is interesting and brings new scientific elements to the optimization of the monocrystalline manufacturing process. No substantive errors have been identified. However, before publication, the authors should address the following comments:

1.      On what basis was the number of training data determined? The authors should provide an answer to this question to explain the rationale for selecting the training data.

2.      Was the risk of overfitting or noise robustness examined in the developed procedures? The authors should discuss whether research was conducted in these areas and what results were obtained.

Furthermore, the axis of the graph presented in Figure 2 is not described. The variables 'a' and 'b' in Equations 1 and 2 are not explained.

The authors should take these comments into account to ensure clarity and precision in the presentation of their work.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

First of all, an important topic on using machine learning methods to identify CZ silicon single crystal node loss detection. However, I would suggest reconsidering the paper after inviting the authors to respond to the following comments.

1.     A lot of excessive information, which makes it difficult to read and follow. It is better to focus on the most important findings. The article would benefit from a more condensed style.

2.     P363-364: The calculation formulas for the recall rate, precision, F1 score, and accuracy are different, there should be numbered one by one, rather than as a whole.

3.     English must be improved through the article. I would suggest the authors ask a native for correct English, and then submit again thereafter.

4.     References must be formatted according to journal style. Use the standard abbreviations for journal names given in the International Standard ISO 4.

For examples:

(1)   Somewhere the authors list “four authors” e.g. the reference [18], somewhere “three authors with et al.” e.g. references [13],[25],[30]. Be consistent.

(2) The “and” should lie in the second author and the third, rather than between the first and second author in reference[24].

5.     Introduction needs better orientation while writing with three steps- state of art process, research gaps and highlighting the current innovative step adopted.

6.     Please indicate with arrows in Figure 1 what where what.

7.     Please give the meaning of symbols in the formulas, particularly in equations (1) to (2).

8.     Please give some specific and quantitated information in the conclusion.

  English must be improved through the article, particularly some miswriting, grammatical and  conciseness issues.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comment 1

Lines 131 - 132

Pei et 

al. [9] studied

The authors must replace

Zhijun et 

al. [9] studied

Comment 2

Line 138

Zhang et al. [12] used a finite

The authors should replace

Jing et al. [12] used a finite

Comment 3

Line 353

The ResNet network was proposed by Kaiming He et al. in 2015.

The authors should replace

The ResNet network was proposed by Kaiming He et al. in 2015 [24].

Comment 4

Line 378

Adam [32] was used

The authors should replace

Kingma et al. [32] was used

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

I think the authors have taken into consideration the comments and suggestions of the reviewers of the original manuscript and attempted to address them all. However, some new questions also have existed and the quality of this paper should be improved. The related suggestion is as follows:

1.     Please double-check the equation of SoftMax in Figure 7 as it is different from that in Figure 8.

2.     I also think too much detailed information in the manuscript makes it hard to follow at times. For example, the authors give much detailed information about Two deep learning-based networks, ICAM-CNN and MMFN in Section 3.2-3.3. please delete the excessive information and re-constructed them in a more condensed style.

3.     Please make a comparison between the two networks of ICAM-CNN and MMFN based on Training Results and give information about the feasibility of further application of detecting node loss while adapting to the crystal growth process.

No.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf