Digital Twin for Experimental Data Fusion Applied to a Semi-Industrial Furnace Fed with H₂-Rich Fuel Mixtures

Round 1

Reviewer 1 Report

The paper focus on building a Digital Twin of a semi-industrial scale furnace using Gaussian Process Regression coupled with dimensionality reduction via Proper Orthogonal Decomposition. It is successfully demonstrated that the same approach can be applied on heterogeneous datasets, obtained from experimental measurements. The paper is well organized; however, this paper needs a little revision to be readable and understandable:

1.         As for abstract, the introduction gave a satisfactory literature survey on the similar topic, but the proposed method is not outlined enough. Please try to set the problem discussed in this paper in a clearer way, especially in aspect of development and application of Digital Twins.

2.         The description of experimental setup is not clearly displayed in text. It is recommended to use charts to show the parameter of experiment in section 2 about the semi-industrial facility. 

3.         More detailed descriptions should be included for explaining the prediction for NO2, which is slightly worse than others, and how to fix it.

In summary, the presented work is useful in developing a GPR-based DT capable of predicting the temperature, chemiluminescence and species concentration of a semi-industrial furnace fed with an H2-rich fuel mixture. The paper is rich in content and clear in organization, and I believe if the author makes a serious effort to account for all the comments provided, the paper will be ready for publication. I suggest some minor revision before publication.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

This paper is well-written and organized. The authors adopted Gaussian Process Regression coupled with Proper Orthogonal Decomposition to develop a surrogate model of a Digital Twin applied to a semi-industrial furnace. The following issues need to be addressed in the manuscript.

Line 188: "The testing cases randomly selected from the complete dataset are cases 2, 8, 11, 15, 16 and 29. The remaining cases are used to train the model."

Line 221: "The predictions are generally good, especially for the temperature field. However, the model tends to overpredict the OH∗ signal, while it underpredicts the CH∗ signal."

According to Table 2, there are 36 cases in total, and each contains six features. In this work, the above 6 cases are selected as the testing data, and the rest are training data. The single run on the split data could lead to local optimization and a lack of generalization.

Performance of k-fold Cross Validation on the dataset is suggested to demonstrate that the model is robust and repeatable. For example, k = 6 can be conveniently set here since there are 36 cases.

Furthermore, Nested Cross-Validation can be used to tune the model's hyperparameters to reduce local optimization or overfitting. The performance of the predictions will be the average outcome.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

I do not find the technical quality of the article meets the standard that would be required for publishing as an academic paper; I would suggest further research is required to produce a robust methodology and meaningful results. 

" it is not always possible to rely on the availability of numerical data" This is strong and contentious statement without proof? If this is your own conclusion, it is out of place here in same paragraph with statement?

"CFD simulations do not produce satisfying" you should motivate why you say that, because it does not clearly follow out of the literature discussion that you have given.

references [2-4] The authors need to be very specific here. The reader needs to know what exactly was based on these studies, and which studies in particular.

Why did you use such a position for the sensors? The type of sensors?

"The species concentration depends mainly on the equivalence ratio" it is not clear to the reader!

Please explain why the temperature control is set to 130 °C?

"methodology used to build the Digital Twin" What are the other feasible alternatives? What are the advantages of adopting this particular metric over others in this case? More details should be furnished. Used software should be presented; simulation studies? PU time should be mentioned. 

Moreover, the manuscript could be substantially improved by relying and citing more on recent literatures about studies on numerical and CFD modelling such as the followings:

https://doi.org/10.1016/B978-0-323-90521-3.00019-3

https://doi.org/10.1016/j.engfailanal.2020.104548

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

N/A