Remaining Useful Life Prediction of Rolling Bearings Based on an Improved U-Net and a Multi-Dimensional Hybrid Gated Attention Mechanism

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article corresponds to the subject of the journal. Extensive changes to the article are indicated.

 

Instructions:

-The abstract does not describe the purpose of addressing the validity of the topics. No mention is made of boundary and real conditions. Model validation should be explained.

-When discussing the method used in point 2, the values ​​of the tested signals should be indicated in the table, the extracted ones should be indicated (explain on what basis this set is analyzed), and the frequencies should be mentioned. Presenting the scheme in the form of a diagram is not legible and is predictable.

 

It is unnecessary to provide real data and compare them with experimental data.

In the described steps 1-5, please base on digital data. The same remark applies to point 3.3. The mathematical model is correctly written, obvious dependency: no digital data that should simulate the given range.

Point 4 is the presentation of simulation results. At this stage, the scientific nature and innovativeness of the conducted research should be included.

The point devoted to results and conclusions should be clearly defined and separated.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This reviewer has the following comments:

  Lines 125, 129, etc. have missed characters   Line 200 RUL was previously defined.   How do you verify that correlation, monotonicity, and robustness are effective indicators for your problem? Weren't other indicators tested to verify reference 18?   Why did you choose the CNN-BiGRU and TCN-BiGRU methods to compare with your proposed method? Elsewhere, you mention these as traditional methods; could you also provide more references to these?   Computationally, which methodology is the most expensive: CNN-BiGRU, TCN-BiGRU, or your method?

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Overall, this paper is sound and covers a useful topic. There are only small issues that need to be fixed. To begin with, some of the text in Figure 1 looks like it's in a foreign language; please fix this. In steps 1 and 2, in the equations that are in the text, there is a rectangular box; is this correct, or is it a display issue? There is some weird text in the y axis label of Figure 8 as well.

In Section 3.2, there is a comma and space in the middle of the word "specific".

Also, it would be helpful to explain Figure 4 a little bit, and to maybe have a zoomed-in view of Figure 7, as it's hard to read and interpret as it is.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The paper presents a methodology to predict the remaining useful life of rolling bearings. The presented methodology is compared with other methods (CNN-BiGRU and TCN-BiGRU) and the results are evaluated according to accepted and validated metrics for this type of application. The new methodology proposes improvements in the data pre-processing algorithm to reduce noise and reconstruct the signal from the samples and uses a multi-dimensional hybrid gated attention mechanism followed by a fusion model to generate a health index. This health index is used to predict the remaining useful life of rolling bearings.

The three methods in comparison are applied to the FEMTO-ST bearing data set, full-life accelerated degradation, provided by the IEEE PHM 2012 Data Challenge. Experimental tests were also carried out, from which a new dataset was constructed to confirm the performance of the new method with data obtained under other conditions and by other equipment.

The newly proposed methodology presents the best performance in predicting the remaining service life of bearings.

 

Comments:

The subject addressed is current and has interest.

The work is clearly laid out, but I consider that the numbering of sections and subsections must be revised.

The methodology is clear, but I consider that the process of Health Indicator Evaluation should be better described.

The results are clearly presented.

The conclusions are supported in the presented results, but I consider that it presents some inconsistencies.

 

I suggest the following improvements:

• I consider that in Figure 4 the input and output of the processed signal are represented and not the Signal Processing Schematic Diagram as referred to in its caption.
• I consider that the evaluation of the health indicator is not properly described in subsection 3.3 – in fact, this subsection only describes the metrics that this indicator must satisfy, without specifying which features are extracted from both the time and frequency domains of bearing vibration signals, and the corresponding weight factors utilized to construct the health indicator.
• In line 233, when is written “the average score (Score) defined by the IEEE PHM2012 Challenge [20].” It seems to me that the reference is incorrectly indicated – it should be reference [21] instead of [20].
• In line 274, when is referred the Table 2, it should be Tables 2 and 3.
• I consider the x-axis of the graphs represented in Figures 7, 9, 11, 12, 13, 15 and 17 should be clarified (Time/10s and Time/15s) - does it represent the successive samples collected?
• The DRSN acronym (Deep Residual Shrinkage Networks) should be defined at their first use.
• The Figures are not numbered in order of their appearance in text – Figure 11 appears in text (line 326) after Figures 12 and 13.
• On the x-axis of Figure 14, "bearing" is written in Mandarin.
• In Section 4, after subsection 4.4, in line 368, is initiated subsection 1. Introduction to the Test Platform – it is not clear if this is the first subsection of a missing section 5 or if it belongs to section 4, and it should therefore be numbered as the 4.5 one; similarly for subsection 5.2 in line 386.
• In the conclusions, in lines 419 to 421, two distinct datasets are referred, which correspond to only one dataset; On the other hand, no reference is made to the dataset that was constructed based on experimental results – described in subsections 5.1 and 5.2..

 

Comments on the Quality of English Language

Also, some typos should be corrected ant the English should be revised, e.g.:

• In lines 200 to 201 please revise the English.
• In line 251, it should be “…PRONOSTIA”, instead of “…PRONO-STIA”.
• In line 421 should be “…IEEE”, instead of “IEE”.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I appreciate the article's detailing with a correct description of predictions and signals, it explains a lot. The novelty aspect must be addressed in the work, it is worth presenting an example of real data. After these corrections, you can publish

Author Response

Please see the attachment

   

 

 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The authors responded well to the comments this reviewer provided.

Just try to standardize the same font in figures and throughout the text. Also, they don't include a space in Table 4.

Author Response

Please see the attachment

Author Response File: Author Response.docx