Classification of Cellulosic Insulation State Based on Smart Life Prediction Approach (SLPA)

Round 1

Reviewer 1 Report

• Degree of Polimerisation (DP) of celulose insulation can be estimated on the basis of 2FAL measurement, which does not require transformer withdrawn from the service. The authors didn’t prove clearly why their method is better then well known method utilizing 2FAL concentration. The Statement: "In this work, the condition of the solid paper insulation is estimated based on easily measurable parameters i.e. temperature, moisture and the aging time." (lines 82-83) is debatable.

First of all, reliable estimation of the moisture content in paper is a difficult task in a real transformer. Also temperature measurement arise the question: what temperature should be measure (bottom oil temperature, top oil temperature or hot-spot temperature). These issues should be discuss in the paper.

• It is not true that proposed method allow to estimate absolute DP of cellulose. That is because the initial concentrations of 2FAL, CO and CO₂ are unknown and are not taken into account. In fact ANN estimates only the amount of aforementioned byproducts of insulation deterioration produced during particular aging time period.
• The temperature in the real transformer is a subject of continuous changes due to load variations. So how should the temperature given as an input parameter be understood? What errors can, the changing temperature, generate when estimating the concentration of 2FAL CO, CO₂ and DP?
• It is well known that correlation of CO nad CO₂ with DP is problematic especially in the case of "breathing" transformers. That is because CO₂ may enter into transformer from atmosphere and CO and CO₂ may realise from the transformer. Could the authors consider and discuss such a real case of transformer?
• Generally usefulness of proposed method should be validated using data from real transformers and not on the basis of Kraft paper samples aging in laboratory conditions. The article should be supplemented with such an analysis.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In the work, the authors presented the results of research on the current physical state of the cellulosic solid kraft paper, determined with the help of easily measurable parameters, like temperature, moisture and the aging time. Degree of deterioration of cellulosic solid kraft paper insulation was determined using an integrated insulation health assessment system. Authors developed a thermo-moisture-aging multi-layer feed-forward NN model to predict the concentrations of 2-FAL, CO2 and CO, which is further, correlated to estimate the degree of polymerization values adopting an SLPA. Authors developed a thermo-moisture-aging multi-layer feed-forward neutral network model to predict the concentrations of 2-FAL, CO2 and CO, which is further, correlated to estimate the degree of polymerization values adopting an smart life prediction approach.

The originality the concepts, the significance and the methods are good. The completeness and the organization of manuscript of the paper are good. The organization of the manuscript is good. In my opinion the technical treatment is plausible and free of technical errors. The paper is interesting, written in a transparent and legible way.

Below I presented some remarks that came to my mind during reading:

1. Line 40: There are also transformers filled with other insulating liquids (natural ester, synthetic esters) and dry transformers.
2. Lines 64-75: I think it is worth expanding these lines to describe the results obtained by the authors of the indicated publications (research).
3. Figure 4. Please add a better quality figure.
4. The Authors should highlight what are the advantages and disadvantages when comparing solutions from the scientific literature.

Author Response

Please find the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper shows a methodology for the condition assessment of the solid paper insulation. The condition of the cellulosic paper is estimated by non-destructive measurements: temperature, moisture, and the aging time. These measurements are the basis of the two-stage ANN model prediction. 

The introduction of the paper shows the importance of the topic, however the state-of-the art of similar non-destructive methods should be supported by some other non-destructive insulation design measurement techniques, such as voltage response methodology (doi.org/10.3311/PPee.16399), which technique is used not in transformers but in the diagnosis of the high voltage cables. 

What is the most important and interesting part of the paper is the selected type and the reason of the ANN methodology selection is not discussed, well in the introductory part of the paper. 
Please, discuss the applied algorithms and the selection criteria more in detail, with some new references. The problem is that due to the no-free lunch theorem of machine learning, there can be a lot of different methods selected for the same task, however, there is not exist which is outperforming the others for every kind of problem. Therefore,  the applicability of your model based on the selection of the right methodology, as this phenomenon discussed here: doi.org/10.3390/app10196653 for electrical machines, in the 2nd section.

Figure 3 shows very important information about the insulation health assessment system, however, it is very hard to read it, it would be nice to improve this image, but what is very important: please describe this part of your work more in detail. 

Other Comments:

Fig 2 looks like inserted from another scientific work, please cite it, or if it's possible please change this picture to a higher quality one. 

Please, improve the quality of Fig 4

Author Response

please find the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I would like to thank the authors for answering my questions. Not all answers are fully satisfactory and some raise further questions and willingness to discuss. However, according to the statement "The proposed novel method has been tested experimentally and has been found to be giving promising results and is open for validation using real transformer data as a future work", I would like to give a chance to a wide group of readers to get acquainted with the idea of the authors and assess its sense and possibilities of practical implementation.

Reviewer 3 Report

The authors answered  all of my questions.