Combustion Regime Identification in Turbulent Non-Premixed Flames with Principal Component Analysis, Clustering and Back-Propagation Neural Network

Round 1

Reviewer 1 Report

The presented manuscript is not prepared according to the journal requirements. The authors should use the MDPI template and check the instructions for organizing the text and results presentation. The article has potential; it is well written, and the presented results have practical significance for identifying combustion regimes for industrial applications. However, it should be sent to the authors for reorganization.

The authors should pay attention to the following:

Write the chemical formulas correctly.

What program was used for PCA, Cluster, and ANN analysis? Add Statistical analysis section.

Provide more information about ANN. What kind of model was obtained? What were the activation functions?

Provide more discussion and compare obtained results with the existing literature.

Add more novel literature and organize references according to the guidelines for authors.

Author Response

We would like to thank you for valuable comments on our manuscript. Following the comments, we have revised it, and revisions are written in blue color in the updated manuscript. The point by-point responses are:

Point 1: The presented manuscript is not prepared according to the journal requirements. The authors should use the MDPI template and check the instructions for organizing the text and results presentation. The article has potential; it is well written, and the presented results have practical significance for identifying combustion regimes for industrial applications. However, it should be sent to the authors for reorganization.

Response 1: We are grateful to the reviewer for pointing out this issue. We have modified the article using MDPI template.

Point 2:Write the chemical formulas correctly.

Response 2: Thank you for pointing out this issue. Chemical formulas have being corrected in new manuscript. The corresponding modifications can be reflected in each chemical formula.

Point 3:What program was used for PCA, Cluster, and ANN analysis? Add Statistical analysis section.

Response 3: We have used MATLAB to perform PCA and clustering analyses. As already stated in the manuscript, the training of the BPNN is performed using the TensorFlow Python library. Specific revisions (highlighted with red text) can be found in Section 3.

Point 4:Provide more information about ANN. What kind of model was obtained? What were the activation functions?

Response 4: Thank your for the suggestions. The network structure is a fully-connected neural network, which is consisted of an input layer, an output layer, and a hidden layer. Activation function is the Softmax function to convert the output values of the multi-classification into relative probabilities. Loss function is the cross-entropy loss function, Modifications (highlighted with blue text ) can be found in Section 3.3 of the newly submitted manuscript.

Point 5: Provide more discussion and compare obtained results with the existing literature.

Response 5: We conducted more discussions in the results and referenced more literature. Details can be found in Section 6 of the newly submitted manuscript.

Point 6: Add more novel literature and organize references according to the guidelines for authors.

Response 6: Thanks for the suggestion, which would improve the manuscript. We have included more current research in recent years in the introduction.

Author Response File: Author Response.pdf

Reviewer 2 Report

The following work analyzes the possibility of identifying combustion regimes to understand combustion phenomena and the structure of flames.

This study proposes a method of identification of the combustion regime (CRI) based on the analysis of the rotated principal components (PCA), on the analysis of the clustering and on the method of the neural network of back-propagation (BPNN). Several LES simulations were used as data. In my opinion the manuscript is an application of a known technique, which does not bring great advantage from the point of view of research and / or application. This work appears to be presented in a superficial way and without an innovative character. I do not recommend the acceptance of such work.

Author Response

We would like to thank you for valuable comments on our manuscript. Following the comments, we have revised it, and revisions are written in blue color in the updated manuscript.

Point 1: This study proposes a method of identification of the combustion regime (CRI) based on the analysis of the rotated principal components (PCA), on the analysis of the clustering and on the method of the neural network of back-propagation (BPNN). Several LES simulations were used as data. In my opinion the manuscript is an application of a known technique, which does not bring great advantage from the point of view of research and / or application. This work appears to be presented in a superficial way and without an innovative character. I do not recommend the acceptance of such work.

Response 1: We thank the reviewers for the comments. In this study, we have proposed a new CRI method based on the integration of rotated PCA, clustering analysis and BPNN. We agree with the reviewer, that these are known techniques, and some of them have also been used in relevant combustion studies. For instance, Himanshu et al.[1] have adopted clustering algorithm to characterize MILD combustion. Jigjid et al. [10] have developed a predictive tool on the basis of neural network to identify combustion modes.

We proposes a combustion regime identification (CRI) method based on rotated principal component analysis (PCA), clustering analysis and back-propagation neural network (BPNN) method. While these several machine learning tools are used by many for data analysis, we propose a complete new flame identification process. To our knowledge, there is no previous study using cluster analysis in combination with neural networks for CRI In addition, other flame classification methods such as the GFRI method proposed by Hartl [2]et al. and the PCA and NN neural network methods proposed by Jigjid [3]et al. are based on the flame index method[4It evaluates the alignment of fuel and oxidizer gradients and thus gives an in-dication of the nature of local combustion regime ranging between premixed flame and diffusion flame. Our method does not require a scalar gradient. In contrast to the flame index method, our method is designed to distinguish the state of the combustion field, such as whether it is on fire or not.

In addition, we used the unsupervised learning clustering analysis results as input labels for the ANN compared to the CNN approach based on GFRI results used by Wen et al. This method avoids the tedious GFRI process and has a higher recognition accuracy than the method of Wen et al.

With reference to your suggestion, we refer to the studies of others and cite more references in the manuscript. In addition, we describe the significance and innovation of the study in more detail in the last section of the introduction.

[1]:Himanshu Dave, N. Swaminathan, Alessandro Parente, Interpretation and characterization of MILD combustion data using unsupervised clustering informed by physics-based, domain expertise, Combustion and Flame, Volume 240, 2022, 111954, ISSN 0010-2180, https://doi.org/10.1016/j. combustflame. 2021.111954.

=

[2]: S. Hartl, D. Geyer, A. Dreizler, G. Magnotti, R. S. Barlow, and C. Hasse, “Regime identification from Raman/Rayleigh line measurements in partially premixed flames,” Combustion and Flame, vol. 189, pp. 126–141, 2018.

[3]: K. Jigjid, C. Tamaoki, Y. Minamoto, R. Nakazawa, N. Inoue, M. Tanahashi, Data driven analysis and prediction of MILD combustion mode, Combust. Flame 223 (2021) 474–485.

[4]: H. Yamashita, M. Shimada, and T. Takeno, “A numerical study on flame stability at the transition point of jet diffusion flames,” Symposium (International) on Combustion, vol. 26, no. 1, pp. 27–34, 1996]

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper has serious technical and content issues. However, it seems interesting and based on a credible study. Therefore, a major revision is recommended at this stage, to give the authors the opportunity to render a proper manuscirpt.

At the end of the abstract, please add a summary of the obtained results and their significance. Please make the summary meaningful avoiding generic statements that would need no research. Characterising the obtained accuracy is fine, but what follows from that? It would be good if the readers get an idea of this, to improve the paper impact.

All references in the text are rendered with an error message from Word update of fields. This prevents appropriate mapping of the references to the text. Before this is corrected, the "Introduction" section cannot be comprehended or evaluated.

A similar problem exists also with the other cross-references in the text - to tables and to figures. This makes the manuscirpt uncomprehensible.

Before proceeding to describe your chosen model and actions, please describe your scientific hypothesis, concepts and the relevant reasoning for choosing the particular modelling approach. This should be accompanied by an overall description of the followed procedure. A block diagram of the procedure would be also very useful.

In the method presentation, I can reasonably guess that lines 161-179 describe Figure 2. However, the text provides only some fairly obvious comments about fuel concentration and temperature, while omitting the proper context and the way how this type of representation is used. 

Figure 3 - what is "Z" and its measurement units? How are these diagrams used? It is unclear what is presented on those plots, despite the provided captions. Similar issues exist for the plots in Figures 4 to 8, 9-10, 12.

Author Response

We would like to thank you for valuable comments on our manuscript. Following the comments, we have revised it, and revisions are written in blue color in the updated manuscript. The pointby-point responses are:

Point 1: At the end of the abstract, please add a summary of the obtained results and their significance. Please make the summary meaningful avoiding generic statements that would need no research. Characterising the obtained accuracy is fine, but what follows from that? It would be good if the readers get an idea of this, to improve the paper impact.

Response 1: Thanks for the suggestion, which would improve the manuscript. We have taken into account the comments of the reviewers and added a summary of the obtained results and their significance in the abstract.

Point 2: All references in the text are rendered with an error message from Word update of fields. This prevents appropriate mapping of the references to the text. Before this is corrected, the "Introduction" section cannot be comprehended or evaluated.

A similar problem exists also with the other cross-references in the text - to tables and to figures. This makes the manuscript uncomprehensible.

Response 2: We are grateful to the reviewer for pointing out these issues. These issues have been corrected in the updated manuscript.

Point 3: Before proceeding to describe your chosen model and actions, please describe your scientific hypothesis, concepts and the relevant reasoning for choosing the particular modelling approach. This should be accompanied by an overall description of the followed procedure. A block diagram of the procedure would be also very useful.

Response 3: Thank your for the suggestions. Our choice of the particular modelling approach is based on our previous research and extensive numerical experiments. We have described scientific hypothesis, concepts and the relevant reasoning for choosing the particular modelling approach at the section 1 in the updated manuscript.

We have added a block diagram to better describe our machine learning tools.

Please see the attachment to see new block diagram.

Point 4:In the method presentation, I can reasonably guess that lines 161-179 describe Figure 2. However, the text provides only some fairly obvious comments about fuel concentration and temperature, while omitting the proper context and the way how this type of representation is used

Response 4: We are grateful to the reviewer for pointing out these issues.Figure 2 is only a two-dimensional diagram using fuel and temperature. If different scalars are used for analysis, the distribution of scatters will show different characteristics. For example, in the preheating area (yellow scatters in Fig. 2) connecting the fuel inlet and the combustion zone, the concentration of CH2O is be very high, while it will be very low in other regions; the concentrations of major products, such as H2O, CO and CO2, are high in the region represented by orange in Fig. 2. Although the high-dimensional scalar distribution is difficult to be represented by a two-dimensional scatter diagram, the example in Fig. 2 presents that the multi-dimensional clustering。 We discuss the meaning of this diagram in more detail at the section 3.2 in the updated manuscript.

Point 5:Figure 3 - what is "Z" and its measurement units? How are these diagrams used? It is unclear what is presented on those plots, despite the provided captions. Similar issues exist for the plots in Figures 4 to 8, 9-10, 12.

Response 5: Z indicates the mixture fraction, and R indicates the radial length. We have added descriptions in the updated manuscript.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thanks to the authors for implementing the various comments with precision. The introduction is now much more complete and the purpose of the work is clearly understood. Acceptance of the manuscript is recommended. Beware of some inaccuracy as in line 75 where a space is missing between the quote and "have".

Author Response

Dear reviewer:

  Thank you very much for your willingness to recommend our manuscripts for acceptance. One reason for the missing spaces is because of the difference in version of word software, and another reason is because of our typographical errors. We have made further revisions to the manuscripts based on your comments. We have inserted spaces in line 31, line 47, line 76, line 86, line 89, line 102 ,line 108, line 121, line 213, and Figure 3, Figure 4, Figure 5, Figure 8, Figure 9, Figure 10, Figure 11, Figure 12, Figure 13, Figure 14. The word version we are using is 2019 and the missing space issue has been fixed. We would like to express our great appreciation to you for the comments on our manuscript.

Sincerely yours,
Hanlin Zhang

Author Response File: Author Response.pdf