The Estimation of the Remaining Useful Life of Ceramic Plates Used in Iron Ore Filtration Through a Reliability Model and Machine Learning Methods Applied to Industrial Process Variables of a Pims

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The introduction is suggested to rewrite. A lot of paragraphs here, so authors may consider to grouping as same aspect ones.

The motivation of the objective should be more clarified even though  Authors already mentioned the previous works before but how importance of this work beside of working with large volumes of data should be more emphasized. 

What is the description of Fig. 4 .... Source Author (2025)?  Better, specify in the standard approach (way to take: photos, how to obtain : data and so on).

Please verify and discuss more on Results in Section 4.2. 

How reflecting of the orthogonal values of the correlation matrix in Table 3? Comparing and discussing with the maximum in the column should be collaborated. 

On each iteration (increasing), how effective results relevant here? 

Please try to recheck writing issues such as language in the figures. 

In Fig. 15, is it any reason for displaying the negative on y-axis? Authors may consider this point along with the following figures such that 17 and 18. 

Conclusion should be given to emphasize more founding with respect to contributions. In this work, including accuracy of the prediction possibility for general for large class of applications seems appropriate. 

 

 

 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1. The abstract indicates that the model resulted in the implementation of a "strategy of replacing these components based on condition rather than on a predetermined schedule." Could the authors detail the economic and operational advantages (e.g., cost reductions, enhanced uptime, diminished waste) of transitioning from fixed-time replacement to condition-based replacement for ceramic plates in this industrial context?
2. This study utilizes "data that has already been collected and processed in automation systems such as PIMS and MES, eliminating the necessity for supplementary sensors." What are the benefits and possible drawbacks of only using current sensor data for Remaining Useful Life (RUL) prediction, especially with data quality, granularity, and the detection of various failure modes?

 

3. The abstract reports an R² of 78.32% for the linear regression model and 63.7% for the random forest regression model. Then it asserts, "The final model for predicting the remaining useful life had an R² of 99.6%." The increase in R2 is substantial. Could the authors elucidate the definition of this "final model" and explain how it attains a high R2, particularly when the individual component models have lower R2 values?
4. The introduction states that Industry 4.0 produces substantial data, "many of which are accessible in real-time to end users; however, this data is frequently not utilized to its fullest potential." What are the common obstacles (e.g., insufficient analytical tools, skill deficits, data integration challenges) that hinder companies from properly using this data for predictive maintenance?
5. The authors assert that "Prior studies utilized analogous methodologies... yet the majority of these investigations focused on sensors specifically designed for this purpose." Although this underscores your innovation, could the authors elaborate on particular problems or complexity encountered when modifying techniques traditionally used with specialized sensors to accommodate broad process variables from PIMS/MES?
6. Section 2.1 delineates three maintenance strategies: reactive, preventative, and predictive. Could the authors provide a succinct illustration of how the failure of a ceramic plate may be addressed under each of these ways to elucidate the distinctions more concretely?
7. The theoretical approach provides a succinct explanation of the exponential distribution in the context of reliability analysis. The authors state that "the absence of memory is a significant constraint that must be acknowledged" for the exponential model. Considering this constraint, what prompted the selection of the exponential model for analysis in their work, particularly in light of the abstract's reference to Weibull survival models? How can they mitigate the restriction of "memory deficiency"?
8. Section 2.2 presents CRISP-DM as a framework for data analysis. Although the authors highlight its practical focus, agility, and adaptability, could you elaborate on any recognized limits or critiques of the CRISP-DM technique that may pertain to its use in this particular industrial predictive maintenance context?
9. In Section 2.4, regarding Random Forest Regression, it is said that the approach is "nonparametric, comprehensible, and readily adjustable, even for large-scale problems, and does not necessitate extensive statistical expertise." In addition to these benefits, could the authors also address any downsides or issues associated with using Random Forest Regression for Remaining Useful Life (RUL) prediction in this particular context, such as interpretability beyond feature significance and sensitivity to hyperparameters?
10. Table 1 delineates the "Business Understanding" phase, indicating that "Ceramic plates... account for roughly 40% of the operational expenses of an industrial filtration facility." What methodology was used to get this cost percentage? Was it determined by direct replacement prices, downtime, or other factors?
11. In the "Data Acquisition and Data Understanding" phase, a selection of 12 sensors was made based on the feedback of process experts. What criteria were used in the selection process for these 12 sensors? Were other sensors evaluated and subsequently rejected, and if so, what criteria were employed?
12. In the section on "Data preparation," the authors indicate the process of addressing data to "eliminate outliers and missing values." What precise approaches or algorithms were used for outlier identification and the management of missing data? What was the magnitude of the outliers and the extent of missing data observed?
13. In the "Identification of health condition indicators" phase, "a temporal trigger is established when the reliability of the plate sets attains 70%." What is the rationale for establishing the reliability level at 70%? Is this an industry standard, or was it established via empirical evidence?
14. The "Modeling" step indicates that "Model creation utilizes only a subset of the partitioned data, referred to as the test set." This wording is quite atypical; often, a training set is used for development and a test set for assessment. Could the authors elucidate the partitioning of your data (e.g., training, validation, test sets) and the application of each subset throughout the modeling and assessment phases?
15. For the case study, the authors examine an "industrial iron ore filtration facility situated in Brazil featuring 14 ceramic filters." Was data obtained from all 14 ceramic filters or only a subset? In what manner do the number of filters and the magnitude of the operation influence the generalizability of your results?
16. The authors state that "Ceramic plates possess a manufacturer's designated useful life of 2-3 years of continuous operation." Nonetheless, "the complete assembly of ceramic filter plates is replaced at varying time intervals," and "malfunctions may arise, necessitating corrective replacements." How do they include these discrepancies in actual operating life and the incidence of corrective replacements when developing your reliability models and predicting Remaining Useful Life (RUL)?
17. The document asserts that "The primary failure modes are delineated in Error!" Reference source not found. This serves as a placeholder. Kindly ensure that the real figure illustrating the failure modes (e.g., cracked plate, fractured plate, leaking near fasteners) is provided and represents these modes for enhanced comprehension.

Comments on the Quality of English Language

The English might be enhanced to more effectively explain the study. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Authors already considered all comments. 

Author Response

Thank you very much for taking the time to review this manuscript. Please find the corresponding revisions/corrections highlighted/in track changes in the re-submitted files. We found your comments very useful for improving our manuscript. We made the revisions regarding real context for R(t) fitted values higher than 100%.