A Generalised Method for Friction Optimisation of Surface Textured Seals by Machine Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this study, the authors investigate a generalized method for optimizing friction in surface-textured seals using machine learning. The analysis in this work is grounded in experimentally derived data concerning surface texture parameters, which are characterized by dimple diameter, distance, and depth. Moreover, the study encompasses the measurement of friction data between the seal and a pneumatic cylinder across various friction regimes, spanning from boundary to mixed and up to hydrodynamic lubrication.

A noteworthy innovation in this study is the introduction of a generalized method that facilitates the entire machine learning process, starting from raw data acquisition and culminating in model prediction. Unlike previous approaches that commit to a limited set of learning algorithms, this study explores a broad array of 26 regression learning algorithms. These algorithms are employed in building machine learning models through supervised learning, allowing for an evaluation of their suitability in the specific application context. To identify the most appropriate model, mathematical metrics and tribological relationships, such as Stribeck curves, are applied and compared with each other.

Revisions and Additions:

1. More Comprehensive Literature Review: To strengthen the foundation of this study, it is advisable to include a more comprehensive literature review. This should encompass prior research on surface-textured seals, machine learning applications in tribology, and relevant studies on friction optimization in mechanical systems. A well-structured review will provide a better context for the current work.

2. Increased Reference Inclusion: Expanding the list of references is essential to support the claims and findings in the study. Additional citations should be included to connect the research with the existing body of knowledge and to acknowledge the work of other researchers in the field. This will enhance the academic rigor of the paper.

3. Bullet-Pointed Conclusion: The paper should conclude with a clear and concise summary of the key findings and implications of the study. Presenting the conclusion in bullet points will make it easier for readers to grasp the main takeaways from the research. This structured approach will enhance the paper's overall clarity and impact.

By incorporating these suggested revisions and additions, the paper will not only provide a more robust analysis but also make its findings more accessible and valuable to the scientific community.

Comments on the Quality of English Language

Should be modified by a  native speaker.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors proposed quite interesting work that can be useful from a scientific and practical point of view. However, there are a number of questions for the authors, which are listed below.

 

1. It is not clear what is the main scope and subject of the study, the cylinder-seal tribological system itself and its optimization, or their data-driven models and corresponding methods? In the introduction, please formulate more clearly the subject of the research, its novelty, main findings and contribution of the paper.

 

2. The introduction provides almost no information about the current state-of-the-art in the field of study (which, as noted above, is also not entirely clear). The authors made conclusions in the introduction based on just a couple of references. Refs 1-11 are not commented on at all. The provided justification for the problem being solved is insufficient. The authors did not present any scientific background or achievements in the field of seal texturing. 18 references are completely insufficient to characterize the background of such interdisciplinary research.

The introduction should be significantly improved and expanded. It should give readers a clear and informed understanding of the problem being addressed and of the latest achievements of other researchers in the areas covered.

 

3. Regarding the statements in the sentences in lines 31-33 and 35-36: if certain selected methods, such as ANN, do a good job (we can also see this in the presented results), then why do the authors present this as a problem? What difficulties are associated with the fact that several algorithms are enough to solve problems?

 

4. How was the thermal stability of the system controlled and ensured during the experiment, taking into account possible heating during repeated movement of the piston? What was the greatest increase in the temperature? Were the actual dimensions of each of the 11 seals tested measured and was their size variation assessed? Without this information, it cannot be ruled out that there were other factors influencing friction than those listed in Table 2. Please provide relevant data.

 

5. Please add specific results (graphs, tables, etc.) of the analysis using DQR, IQR, SPLOM mentioned in Section 4 so that readers can also independently evaluate the dataset.

 

6. The authors stated 26 algorithms in Table 3, but provided results for only a few of them in Figure 6. Moreover, even Figure 6 identifies categories, but not specific algorithms. What results did each algorithm demonstrate?

 

7. Again, the authors tested many methods. They also note overfitting of some models, for example the Kernel Approximation Model, Figure 6. Was there any optimization of the hyperparameters of the models to adapt them to the features of the dataset? For example, perhaps the mentioned Kernel Approximation Model could show better results.

 

8. It is not clear why the authors consider GPR to be the most appropriate. Figure 6 shows that, for example, ANN provided almost the same fitting, and MSE and R2 metrics are even superior to GPR. Again, the authors did not provide data on other algorithms in principle. Please explain your choice.

 

9. How were the optimal solutions in Table 5 obtained and which optimization methods were used?

 

10. The explanation in lines 335-339 is overly simplified and poorly supported. Many studies, for example regarding hydrodynamic bearings and seals, show that texturing leads to a reduction in viscous friction due to micro-hydrodynamic lubrication, and not simply a local increase in film thickness. The authors should look into this issue in detail and revise Section 6 taking this into account.

 

11. Multiple typos in the text must be corrected.

 

As I already noted, the topic is quite interesting, but clarification of the experimental design is strongly required to ensure the correctness of the data (point 4 above). If this is all right, then a significant improvement in the quality of the presentation is also mandatory.

Comments on the Quality of English Language

In general English is clear, only minor revisions and correcting typos are required.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper presents a systematic approach to building regression models using experimental data, showcasing the importance of data analysis and preparation in machine learning applications. It demonstrates a rigorous evaluation process with the use of various machine learning algorithms and cross-validation techniques. The paper's clarity and organization make it accessible to both experts and readers with a general interest in the topic. However, addressing the highlighted points would enhance the paper's technical depth and provide a more comprehensive understanding of the research methodology and findings.

-The abstract provides a clear overview of the research focus and methodology, setting the stage for the study. However, it lacks specific results or findings obtained from the study. Including a brief summary of the outcomes would enhance the abstract's comprehensiveness and give readers a glimpse of the research findings.

- Expand the introduction part, see the following references:

http://dx.doi.org/10.24874/PES01.01.029
https://doi.org/10.1007/s00170-021-06954-2
10.24874/ti.2018.40.03.11

What are troubles with surface texturing?

-Why is this research interesting and why it should be published?

- The paper mentions the use of 26 regression learning algorithms but does not discuss their comparative performance, or the rationale behind choosing them. Providing this information is crucial as it adds depth to the methodology section and offers insights into the authors' decision-making process.

- The flowchart illustrating the general process of machine learning model development is informative, but it could benefit from more detail. Specific techniques used for data analysis, model evaluation, and feature selection should be included to provide a comprehensive understanding of the research methodology.

- Key steps involved in data analysis, data preparation, and model development are outlined in the paper, emphasizing the importance of meticulous data processing. However, the paper lacks specifics on the techniques used for handling outliers and reducing dimensionality. Providing these details would enhance the transparency of the research methodology.

-Emphasize the reasons for selecting specific machine learning algorithms utilized for building the regression models.

-The significance of principal component analysis (PCA) in feature extraction and dimensionality reduction is briefly mentioned. Providing a more detailed explanation of how PCA was applied and its impact on the model's performance would enhance the technical clarity of the paper.

-The paper discusses the impact of surface texture parameters on friction coefficients, providing valuable insights into tribological interactions in different lubrication states. However, there is a lack of discussion on the limitations of the study, such as potential biases in the data collection process or assumptions made during model development. Addressing these limitations would provide a more comprehensive view of the research outcomes.

-The paper would benefit from exploring potential challenges faced during experimentation and how these challenges were overcome. This information would provide readers with a deeper understanding of the practical aspects of the study.

- While the paper offers valuable insights, providing practical examples or real-world applications of the developed models would enhance its relevance and demonstrate the potential industrial implications of the research.

- Explore what are potentials for future research, especially in addressing the observed increase in friction coefficients in certain conditions, this would add depth to the conclusion section and inspire further studies in the field.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors responded acceptably to many of the comments, however still there are some questions.

 

1. The authors have made some useful changes to the introduction, however not all the mentioned problems are solved. The main problem is that the authors still have not presented an adequate review on the recent achievements in the field of seals/bearings texturing and its optimization, though the scientific background is quite reach and voluminous, multiple studies can be easily found by a simple search, including the review papers, e.g.:

- doi.org/10.1016/j.triboint.2020.106792 

- doi.org/10.1016/j.triboint.2022.108010.

Based on the review, readers would like to know the current state of affairs, as well as which problems and gaps exist there now, and why they should pay attention to this study.

 

2. Since the authors attempt to provide an overview of the factors influencing friction (lines 25-31) and common methods for reducing it (lines 37-40), then they should make the lists more complete. Thus, regarding the methods for reducing friction in tribological systems, some additional possible ways should be mentioned:

- optimization and adjusting the lubricant feeding conditions:

- doi.org/10.1016/j.triboint.2013.11.016

- doi.org/10.3390/lubricants11050218;

- optimization of the geometric parameters of the system: 

- doi.org/10.3390/lubricants3030569

- doi.org/10.3390/lubricants6010021;

- modification of operating modes of the system as a whole, i.e. doi.org/10.1016/j.ijrefrig.2021.01.016, etc.

 

3. Supplement Table A2 with the data for last epochs to demonstrate the changes in the models.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf