Predicting the Life of Varistors via a Nonlinear Coefficient Based on a Small-Scale Data Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

A life prediction method based on BiLSTM is proposed for the cumulative effect of the degradation process of varistors caused by surge impact. This method uses nonlinear coefficients to construct life parameters, which can better characterize the degradation process of varistors under surge impact, and uses BiLSTM network to build a life model, which can effectively use historical data to judge the life of varistors.

1.Provide a more detailed comparison of the proposed method with existing life prediction methods, including their performance metrics, limitations, and computational complexity?

2. Figures quality are not up to the mark please modify all, can y-axis scale be modify above 1 and please technically explain the graphs in text?

3. Can additional factors, such as varistor manufacturing process and environmental conditions, into the life prediction model be incorporated in the model?

 

4. Clearly outline the potential application areas of the proposed method?

Comments on the Quality of English Language

Many grammar and syntax error please proofread

Author Response

Thank you for your ssuggestions.

1. At present, most of the studies on the life of varistor focus on the direction of material science, and few research on the deterioration process. Therefore, there are not many relevant methods proposed at present. An alternative model is outlined in Section 3.4, and the comparison results are shown in Table 1.
2. The nonlinear coefficient is intuitively expressed as the slope of the curve when the breakdown current is reached in the voltage-current diagram on the logarithmic axis. So in general, the nonlinear coefficient is a number between 0 and 1. So in some figures the Y-axis is capped at 1. The legend has been revised to ensure that each diagram has a corresponding annotation and description.
3. This is a very good suggestion, and will be one of the future research directions of the author.
4. The conclusions have been revised.

Reviewer 2 Report

Comments and Suggestions for Authors

Discuss how the sample data is collected and how it has been trained.

Discuss how the proposed model handles noise and outliers

Discuss how the parameters of the nonlinear coefficient model are optimized to achieve the best performance.

 

Discuss the predictions made on the nonlinear coefficient models to interpret data.

Author Response

Thank you for your suggestions.

1. Data collection is described in Section 3.1. The data processing mode is explained in section 2.3 Data preprocessing. Both parts have been revised. In addition, in network training, it is actually the parameters of the network that are trained, and the description of trainable parameters is provided in Part 2.2.
2. Noise and outliers are unavoidable in practical engineering and are generally caused by measurement errors or errors. But their percentage is usually very low. Therefore, training with normal data will not lead to distortion of the model, but will enhance the robustness of the model.
3. The best parameters are selected by comparing the test results under different parameters. This is discussed in Section 3.2.
4. The model predicts the variation of the nonlinear coefficient, which has a strong correlation with the remaining life of the varistor. It is explained in part 2.1.

Reviewer 3 Report

Comments and Suggestions for Authors

I read the manuscript several times. In this work, it is difficult to connect the proposals and catch the ideas behind the proposals because. The Life prediction of Varistors using a novel tool, sounds quite interesting. But, even when various figures showing results are depicted, and some briefly described, confusion arise after reading paragraph after paragraph. The reason of this is due the manuscript is badly structured as well as the writing has many deficiencies. The paper is barely understandable.  

 

Also,

 

The abstract does not summarize the most relevant sections of the work neither highlight the importance, difference, advantages of the proposals reported in the manuscript

 

Regarding the introduction, problem statement is not clear enough, as well as contributions, benefits, ...etc. In addition, few references are reviewed, and these are not critically analyzed but just described.

 

Any reliability analysis is carried out? Mean time to failure has any meaning in this work? Please discuss.

Comments on the Quality of English Language

Every section of the manuscript necessitates an in-depth edition. It is hard to understand the writing.  

Author Response

Thank you for your suggestions.

1. The abstract has been revised.
2. This paper aims to establish a life prediction model for varistor. There is very little research in this area. The introduction focuses on the origin and background of the problem, the first paragraph indicates that the nonlinear coefficient is not paid attention to in the current life prediction research, and the second paragraph indicates that the nonlinear coefficient research is focused on the direction of materials science rather than engineering applications.
3. There is reliability analysis in section 3.3, where the mean square error represents the difference between the predicted model and the actual data.
4. The manuscript has been revised to improve its expressiveness.

Reviewer 4 Report

Comments and Suggestions for Authors

This paper presents a method for life prediction of varistors using non-linear coefficients using data based model approach.  The presented idea and the BiLSTM algorithms are explained clearly. However, the presentation and writing of the paper need to be improved. The  queries to the authors are listed as follows:

1.       Please provide a brief description of the non-linear coefficient and its significance in estimating varistor lifetime in the introduction section. 

2.       Please improve the quality of all figures and legend descriptions. (Fig 14.)

3.       It is hard to understand the equation 1 and 2, use proper terminologies and define them. The nominal discharge voltage and discharge current terminologies used are different in the introduction and in the experiment section, please use the same terminologies to indicate the same variables, similarly, the variable A is also misleading in many places, please rectify this.

4.       The intermediate state parameters of the LSTM network mentioned in the text and in Figure 5 are different.

5.       Please improve the writing in the experimental data section.

6. please provide a description of the experimental procedure along with the impact of current waveforms along with Figure 11 on the image of the damaged varistor.  Also, it will be interesting to know the time duration and the number of impact currents, it took for the varistor to be completely damaged.   

Comments on the Quality of English Language

This paper presents a method for life prediction of varistors using non-linear coefficients using data based model approach.  The presented idea and the BiLSTM algorithms are explained clearly. However, the presentation and writing of the paper need to be improved. 

 

Author Response

Reviewer 4

Thank you for your suggestions.

1. The second paragraph of the introduction has been revised.
2. Figures quality have been improved.
3. Formulae 1 and 2 are supplemented. The variables mentioned have also been revised.
4. The parameters mentioned in Figure 5 and its description are trainable parameters of the network, which are automatically adjusted by the network during the training process. The preset parameters of the experiment part are other parameters.
5. The experimental data section has been revised.
6. The description in Section 3.1 has been revised. It is a great pity that the appearance of the varistor after each impact was not photographed during the experiment, because its appearance did not change greatly until it was completely damaged. The author will record the experiment in more detail in future studies.

Reviewer 5 Report

Comments and Suggestions for Authors

Dear authors, please accept my comments as recommendations only:

 

1. The literature review could be extended. I would suggest an analysis and comparison between the surge protection methods to be included and estimated.

 

2. Please clarify the aim and novelty of your work. The problem you solve is not clear.

 

3. The origin of Fig. 1 – is this simulation or experimental study? Also, the Y-axis should be labelled according to the voltage and current. The X-access “impact time” requires additional explanation.

 

4. You may want to include a presentation of the industrial standards for surge protection testing. Respectively, to show how they are accommodated in your study.

 

5. The LSTM network (fig. 5) and its components require a better explanation.

 

6. The math model of the varistors characteristics is well developed with equations 1 – 6, but this part looks separated from the rest of the paper. It is not clear how this model is applied in the life prediction model (part 2.3).

 

7. The experimental setup could be better supported with the experimental circuit. A photo of the experimental setup is also an option. Also, you may want to publish oscillograms showing the measured voltages and currents through the varistor, particularly depicting the moment of failure.

 

Thank you for the interesting paper.

Author Response

Thank you for your suggestions.

1. Appropriate revisions have been made to the relevant parts. The purpose of this paper is to establish the life prediction model of varistor, different surge protection methods are not the focus of this paper.
2. The abstract and introduction sections have been revised to highlight the problems the authors are trying to solve.
3. Figure 1 shows the experimental study. Because the range difference is too large, the data in the comparison chart has been normalized for convenience. Section 2.1 has been revised.
4. It is supplemented in Section 3.1.
5. Section 2.2 has been revised.
6. The calculation of the nonlinear coefficients is done directly by the experimental equipment and is included in the data collection part of the life prediction model. The description has been revised.
7. The experimental part has been revised and the figures have been added,

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has many good amendments. This version of the paper is better than que previous one.  

Comments on the Quality of English Language

light improvements are needed

Reviewer 4 Report

Comments and Suggestions for Authors

The experimental section has been rewritten well now. The reviewer's comments are clarified now.

Reviewer 5 Report

Comments and Suggestions for Authors

The paper has been improved. I would recommend the article to be published in the journal.