Recent Advances in Data-Driven Methods for Degradation Modeling Across Applications

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper reviews recent advances in data-driven approaches for cross-disciplinary degradation modeling, with a particular focus on statistical methods, dynamic prediction models, and machine learning techniques, and illustrates their applications in materials science, engineering, and medicine. The article covers a wide range of literature and holds certain academic value. However, it tends to emphasize descriptive introductions of principles and applications, while lacking critical evaluation, quantitative comparison, and conclusive guidance. As a result, the work falls short in both academic depth and practical significance. The main issues are as follows:

1. The correspondence between methods and application scenarios is unclear. Although the paper discusses multiple domains, it does not specify which methods should be prioritized under different operational environments or data conditions, nor does it provide a rationale for method selection.

2. The search strategy and literature screening criteria are insufficiently transparent. While the authors claim to have adopted a systematic review process, no details are given regarding database sources, search terms, timeframes, or inclusion/exclusion criteria, thereby undermining the objectivity, transparency, and reproducibility of the review.

3. The paper lacks quantitative evidence synthesis and performance comparison. Despite citing numerous studies, it fails to provide systematic comparative tables or performance metrics, making it difficult for readers to discern the relative strengths and weaknesses of different methods. The review remains primarily descriptive.

4. The discussion of frontier issues is underdeveloped. Although Section 6 mentions uncertainty analysis and physics-informed modeling, the discussion is overly general and does not elaborate on specific technical pathways, challenges, or potential solutions.

5. The conclusion section largely reiterates earlier content without distilling clear practical implications or forward-looking research directions, resulting in weak summarization and limited guidance.

Author Response

Responses are in the attached file

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Overall questions and comments

The 32-page manuscript reviews items related to the modeling of degradation across three applications: materials science, engineering, and medicine.

The paper is well-written and well-organized, making it a pleasant and interesting read, even if it raises significant questions and concerns.

Section 1

1. The literature gap is not identified in the introduction. Are there existing reviews on degradation modeling? What are their key findings? Why are they insufficient? What is the novelty of this review? See, for example, 10.23940/ijpe.17.03.p6.299314.

Section 2

1. Three families of models are described: physical models, data-driven models, and knowledge-based models. Are there mixed approaches that merge two or more of those? See, for example, 10.3390/jmmp4010027.
2. The difference between what belongs to materials science vs. engineering is quite fuzzy. For example, the authors mention fatigue in both categories. This is because the degradation of materials would likely belong to a subsection of engineering (i.e., materials engineering) rather than a specific separate category. Therefore, the difference between engineering and materials sciences in this context is not clearly established.
3. Likewise, the degradation itself is defined differently from one field to another, which is understandable. But the selection of the types of degradation modeled should be questioned or discussed. What about the degradation of databases in information technology? What about the degradation of chemicals in chemical industry processes, e.g., due to contamination, oxidation, and UV-related degradation? The authors should better identify the borders of their review and acknowledge the limitations that stem from these choices.
4. In medicine-related degradation, what about the patients' mental state degradation modeling for example? Or their cognitive functions? Even with the broad definition of degradation in the medical context provided by the authors, the health or function of biological systems should also encompass the cerebral functions. Further, literature exists on both domains. See, for example, among a large number of possible references, 10.3390/brainsci12091132.
5. The authors should also consider updating the list of ISO standards, as ISO 55000 has an updated version as of 2024, incorporating important changes with respect to the 2014 version. Likewise, ISO 15156 has a new version (2020), as well as ISO 11469 (2016).
6. In this table, the ISO 9001 standard is raised. Should the degradation of quality be modeled also? Is there literature on the subject?

Section 3

1. Stochastic processes, which have been the core of degradation modeling in many reliability-centered applications, are completely ignored in the review. This is unacceptable given the essential impacts of these models in degradation modeling and reliability engineering. See, for example, 10.1016/j.ress.2007.03.019 on gamma processes. 
2. What is the interest of Figure 2? Time-related evolutions are only relevant in systematic reviews, but since there is no clear description of the criteria for inclusion/exclusion in the review, Figure 2 only shows the time evolution of the authors' selections rather than actual trends in the literature. There is no accountability for selection bias.

Section 4

1. In all presented models, it should be made clear where the degradation value/prediction appears in the formulae.
2. For all presented models, the advantages, limitations, and usage are simplistic. The reviewer questions the relevance of this information, as it is not comprehensive.
3. Regarding the regression model in particular, only the OLS estimator for the parameters β is presented; however, there are other approaches (e.g., Bayesian methods, maximum likelihood).
4. Continuing with the regression model, the limitations mention the reliance on linearity, which assumes that only linear regression is used. As the authors mentioned, it is only the simplest form of regression models. This exemplifies the care necessary in establishing limitations of models (see question 2 regarding section 4).
5. Section 4.3.2 concerns deep learning. But shallow ANNs are not mentioned. See, for example, 10.1115/1.4047636.
6. In Section 4.3.4, K-Means clustering and Hierarchical clustering need to be visually distinct from the surrounding text as they are titles of sections that follow. Please consider formatting them as items in a list.

Section 5

1. Similar to the previous comment regarding Figure 2 (see question 2 regarding Section 3), the reviewer questions the relevance of Table 4 and Figures 4, 5, 6, and 7, which are not representative of the literature or its evolution, but rather of the authors' selection.
2. Similar to a previous comment regarding the necessity of clarifying where the degradation appears in models (see question 1 regarding Section 4), the methods presented in Section 5.1 do not clearly explain how these methods are applied to degradation modeling. I.e., what are the inputs, what is the output, and in what context it may be applied.
3. Are there contexts that favor one or the other method? In this question, context is not a broad term like "materials science", "medicine", or "engineering", but rather application-specific cases. E.g., for the roll bearing degradation estimate, is there an outstanding method? Same for electrical motors wiring? Etc.

Section 6

1. Table 8 needs to be revised in light of the previous questions, particularly those regarding Section 4.

 

Author Response

Responses are in the attached file. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author has made comprehensive revisions to the paper, significantly improving its quality. However, some obvious errors in expression and formatting still exist. For example, line 220 contains "Table ??", and line 223 contains phrases such as "In the revised version of this manuscript," which should not appear in the main text. The author is requested to carefully review and revise these revisions.

Author Response

Authors are very grateful to the reviewer for their kind words. Paper was significantly improved thank to reviewer comments, which was also mentioned in acknowledgements. Regaring the remark:

Comment 1: However, some obvious errors in expression and formatting still exist. For example, line 220 contains "Table ??", and line 223 contains phrases such as "In the revised version of this manuscript," which should not appear in the main text. The author is requested to carefully review and revise these revisions.

Response 1: This was an obvious editing error, and we are very grateful to the reviewer for catching it, entire paragraph starting with words "We note that several of the standards listed in(...)" was originally intended as part of response and was accidentally left in the submitted revision. Entire paragraph is now removed. No other missing references to either figures or papers remain. 

 

Reviewer 2 Report

Comments and Suggestions for Authors

The authors responded to my remarks and questions. In the gamma process equation (line 381), Gamma is usually written as the uppercase Greek letter Γ. It could also be defined mathematically or in the text.
The reviewer acknowledges the authors' huge work in revising the article.

Author Response

Authors are very grateful to the reviewer for their kind words. Paper was significantly improved thank to reviewer comments, which was also mentioned in acknowledgements. Regaring the remark:

Comment 1:  In the gamma process equation (line 381), Gamma is usually written as the uppercase Greek letter Γ. It could also be defined mathematically or in the text.

Response 1: We propose a compromise solution, in order to avoid confusion between gamma function and Gamma distribution (which uses said function in its definition) we have retained capitalized upright Gamma as the name of distribution, but we provide exact formula for the density function and also define the gamma function with  Γ under the formula using standard integral definition.