Study on the Optimal Double-Layer Electrode for a Non-Aqueous Vanadium-Iron Redox Flow Battery Using a Machine Learning Model Coupled with Genetic Algorithm
Round 1
Reviewer 1 Report
This manuscript reported the study of a double-layer electrode design and optimization for non-aqueous vanadium-iron redox flow battery. The optimization of the ratio between the two electrode components is an interesting topic. There are a few issues should be addressed before accepted for publish.
1. Is the goal of this study finding the optimal double-layer thickness ratio for different flow rate and current density? The study already ran 620 FEM simulations. It looks these simulation results already can provide the information of optimal thickness ratio.
2. For training ANN and all the other ML approaches, what is the training data, validation data, and testing data size? Is the MSE listed in the manuscript for the testing data?
3. There is no validation for the FEM simulation results.
4. Is the proposed coupled GA approach an alternative for the commonly used optimization method for training the neural network? such as Adam optimizer and SGD optimizer?
5. The manuscript shows that the proposed coupled GA approach can train the ANN to better accuracy. Is it because the size ANN used in this study too small?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The presented manuscript “Study on the optimal double-layer electrode for non-aqueous vanadium-iron redox flow battery using machine learning model coupled with genetic algorithm” needs minor changes before taken into the considerations for publication:
· Abstract is generic. Need to address a) need of this work, b) novelty, c) methodology suggested, d) results obtained (numerical).
· Introduction: Add more recent literature work to justify the novelty of your work and selected conditions of the study.
· What are the advantages of genetic algorithm compared to other metaheuristic techniques?
· Figure 4: Scale bars are not visible. Improve the quality of the figures.
· What is the basis of the selection of mentioned input conditions? Specify them in detail.
· Line 356-358: Justify the sentence with proper technical reason and relevant literature source.
· Give technical reasons behind the obtained result with relevant references for section 3.2.
· Mention the limitations and further scope of improvement in last section.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
It has been reported that the double-layer porous electrode (carbon paper plus graphite felt) can boost the operational performance of non-aqueous DES electrolyte-based V-Fe RFB. However, it still lacks an optimal design method for the electrode architectures under different operational conditions. This manuscript develops a multi-layer ANN model to predict the relationship between RFB performances and electrode structural characteristics. Besides, GA is coupled into the ANN regression training process for optimizing the model parameters. The results show this ML model can reach satisfactory predictive accuracy, and the well-trained ML algorithm can be used to estimate whether a double-layer electrode should be applied in non-aqueous V-Fe RFB and determine the appropriate thickness ratio of this double-layer electrode.
However, the discussion and explanation should be further improved. I suggest giving a minor revision and the authors need to clarify some issues or supply some more experimental data to enrich the content.
1. For grammar issues, it is suggested that the author double-check the small grammar errors in the full text, especially the lack of and redundant use of definite articles.
2. For the Keywords, 'operational performance', 'DES electrolyte', 'three-dimensional finite-element numerical simulation', and 'vanadium-iron' should be added in order to attract a broader readership.
3. Page 1, "To further boost the energy storage density of RFB’s electrolyte, non-aqueous solvents were proposed to applied in RFBs, on account of the used non-aqueous solvents have the broader voltage window width relative to the aqueous electrolyte, thus improving the theoretical energy storage density of RFB [3]."
The description of this sentence may be correct. However, in order to give readers a more intuitive concept of energy density, it is recommended to briefly compare the energy density of DES-based RFB and traditional RFB with specific values. For example, the VRFB processes quite a low energy density (ca. 15–25 Wh L−1), while the energy density of zinc-based RFBs was typically higher than 50 Wh L−1 {Batteries 8.11 (2022): 202}.
Therefore, it has been tried to increase the energy density by replacing the VRFB system with zinc-based RFB. Is the energy density of DES-based RFB similar to that of zinc-based RFB? Or can it break through 100 Wh/L? Therefore, giving a brief comparison will be more straightforward and highlight the significance of this work.
4. Page 2, “This composite electrode has a heat treated CP electrode assembled near the membrane side, that can boost the electrochemical reversibility and hydrophilicity within this high reactive rate region.” Does only the CP electrode need heat treatment, and GF does not need heat treatment to improve hydrophilicity? It was reported that "The thermal oxidation method can decrease the activation polarization loss by enlarging the surface area and increases the number of oxygen functional groups in GF" {10.1002/er.5179}. Hence, I think the pretreatment processing of CP and GF should be explained clearly, otherwise, it will cause confusion for readers.
5. Page 2, “Here, this paper proposes a multi-layer ANN coupled with genetic algorithm (GA) to optimize the double-layer gradient electrode architectures for improving the discharging efficiency of DES electrolyte-based vanadium-iron RFB.” What does it mean by 'discharging efficiency'? Maybe discharging capacity, or CE/EE/VE?
6. Page 4, Figure 2 is not very accurate. Normally, the thickness of CP should be 280 um (0.28mm), which should be much thinner than that of GP (3-5mm). Now it seems their thicknesses are exactly the same, which does not reflect reality. I suggest making certain modifications to avoid confusing the readers. I understand the ratios and layers can be adjusted, however, it is still quite strange at the first glance of this figure.
7. Page 8, "For all electrode architectures, the polarization losses of all electrode architectures rise observably with the increase of discharging current density, which means RFB undertakes a worse energy efficiency under the higher current density operation condition."
It is correct the polarization losses will increase with current density, but this will influence voltage efficiency, rather than energy efficiency. Normally, CE increases with current density due to the shorter time for species permeation through membranes. Hence, EE as the result of CE*VE, it is hard to judge whether it increases or decreases with current density {10.1016/j.electacta.2021.138133}. It depends on whether the increase of CE is dominant or the decrease of VE is dominant.
8. Have the conclusions obtained by the model in this manuscript been verified by experiments? Why not add the results of experimental verification to add convincingness?
In addition, the model only gives the thickness of the membrane material and the thickness of the electrode material, but if the membrane material is changed, for example, Nafion is replaced by SPEEK/PBI or even a porous membrane, but the membrane uses the same thickness as the model, does the conclusion still stand? The same problem applies to electrodes.
For grammar issues, it is suggested that the author double-check the small grammar errors in the full text, especially the lack of and redundant use of definite articles.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The revised manuscript and authors' response letter addressed all my concern. The revised manuscript is good for accept.
