LSTM-Based Condition Monitoring and Fault Prognostics of Rolling Element Bearings Using Raw Vibrational Data

Round 1

Reviewer 1 Report

In this study, the authors present an effective fault prognostics system for rolling element bearings using univariate time series analysis with Long Short-Term Memory (LSTM). However, to improve the quality of the manuscript, several queries need to be addressed and incorporate the following comments:

1. Highlight the contribution and novelty of the paper in the abstract and introduction sections. 

2. Clearly mention what sets your methodology apart from existing techniques. Emphasize any innovative aspects or advantages of your approach.

3. The authors should consider addressing the literature survey more comprehensively by incorporating a wider range of relevant studies and providing a more detailed overview of the existing research landscape.

4. I recommend that the authors explore additional relevant literature to further enrich their study. Here are a few recommended papers:

https://doi.org/10.1007/s10489-022-03344-3

https://doi.org/10.1016/j.ymssp.2019.106587

DOI: 10.3390/s20216356

DOI: 10.1109/ACCESS.2021.3128669

DOI: 10.1109/TIA.2019.2895797

DOI: 10.1109/TIA.2017.2661250

DOI: 10.3390/math9182336

5. Consider adding a comparison with other machine learning and deep learning models: Although the LSTM model performed well, it would be beneficial to provide a comparison with other algorithms such as CNNs, GRUs, or even traditional machine learning models like Support Vector Machines. This would give readers a better understanding of the advantages and disadvantages of each approach.

6. Provide more details on hyperparameter optimization: The paper should discuss the process of hyperparameter optimization in more detail, including the range of values considered, the optimization method used, and any challenges encountered. This would enable readers to better understand the model's performance and potentially reproduce the results.

7. Evaluate model performance using additional metrics: While RMSE is a useful metric for model evaluation, incorporating other metrics such as precision, recall, F1 score, or area under the ROC curve would provide a more comprehensive assessment of the model's performance.

8. Assess model generalizability: The paper should provide more information on the model's ability to generalize to other types of bearings or rotating machinery. This could involve testing the model on data from different machines or conducting transfer learning experiments.

9. Analyze computational complexity and processing time: The paper could benefit from an analysis of the computational complexity and processing time associated with the LSTM model, especially considering its application in real-time monitoring and decision-making.

10. Investigate early fault detection capabilities: The paper should discuss the model's performance in terms of early fault detection and the lead time provided for maintenance actions. This is crucial for practical applications where timely interventions are required to prevent catastrophic failures.

11. Are there any limitations in the proposed methodology that could affect the performance of the model when applied to real-world scenarios, such as sensor noise, data loss, or varying sampling rates?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors of this article invistigate the problem of prognosing the remaining useful life of bearing elements using Long Short Term Memory networks, raw data were injected as inputs for training the model and prognostics results were compared with results from the literature. The subject is interesting but this research work methods and results are insufficient for publication.

Indeed, this research lacks indepth analysis and discussion (practically two pages and half 9 to 11 including results and discussions). Also, most of the parts of this article are well established knowledge in the literature (LSTM structure development and dataset explanation), and there is no details on how the data was normalised and cleaned before training the LSTM model.

In order to this research to be publishable; I advice authors to details what really they contribute (for example: data cleaning and normalisation), also, the model should be tested with several datasets (since no deep processing of the data is necessary and the method seems to be general); In addition, a random sampling cross validation scheme can depicts more on the robustness of the proposed method.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper proposed a fault prognostic system using long short-term memory for rolling element bearings. Some comments are as follows:

1. In Section 1.1, the authors give the innovations in this paper, but I think they are very limited. 

2. The authors state “this research concentrated on conducting model training and testing in the time domain by taking the actual sensor data as an input to the model.” Please give details of the authors' reasons for this treatment, supported by some literature.

3. In the model comparison section, only RMSE was used as an evaluation metric. It is suggested to add other metrics to better highlight the superiority of the proposed models. Additionally, more model comparisons and result analysis are needed.

4. The author used traditional LSTM to model time series and only provided point estimates for predictions. More advanced methods such as Bayes-based deep learning, transfer learning, and attention mechanisms could be adopted and reviewed, for example:

A prognostic driven predictive maintenance framework based on Bayesian deep learning. Reliability Engineering & System Safety, 2023, 234, 109181.

5. Inconsistent formatting of literature.

6. It is recommended that the results in Table 3 be retained to a uniform 4 decimal places.

In summary, the manuscript should be revised to make it clearer before further consideration for publication in machines.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The manuscript qualifies for publication after the revision. 

Author Response

Thanks for your positive response.

Reviewer 2 Report

Authors of this article have considerably updated the paper in the revised version for considering the comments suggested at the first stage of reviewing. The adopted methodology is better explained now and the LSTM model prediction results are more trustful sine it was tested for varying datasets. 

Some minors remarks can be found in the attached file in order to finalize the paper.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have revised the manuscript well.

Author Response

Thanks for your positive response.