Machine Learning Approaches for Fault Detection in Internal Combustion Engines: A Review and Experimental Investigation

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

This manuscript is in fact the revised version of the one I already reviewed. The Authors acknowledged all my suggestions and revised the manuscript in the appropriate manner. The main flaw of the previous version, the lack of new fault detection methods, has been overcome by introduction of two deep learning methods, and the scientific contribution of the paper is now obvious. The presentation, the quality of figures and language style have also been improved. Therefore, I have no further remarks. 

Author Response

We sincerely appreciate your time and effort in reviewing our manuscript. Your insightful suggestions and feedback were invaluable to us in enhancing the quality of our work.

We are grateful for your recognition of the improvements made, particularly with the introduction of the two new deep learning methods. We believe that these enhancements have significantly strengthened the scientific contribution of our paper.

Thank you once again for your thorough review and positive assessment. Your comments have been immensely helpful in refining our manuscript.

Reviewer 2 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

The authors followed my remarks, therefore I recommend the paper for publication.

Author Response

Thank you very much for your feedback and recommendation for publication. Your suggestions have significantly contributed to the improvement of our manuscript. We appreciate the time and effort you put into reviewing our work.

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The paper is a review paper for fault diagnosis for IC engine, with over 100 references, which seems to be a decent review. The followings would benefit the authors to improve the paper.

1) The paper is focused on a review on machine learning based fault detection. From general categories of the fault diagnosis methods pointed by the classic review papers:

A Survey of Fault Diagnosis and Fault-Tolerant Techniques—Part I: Fault Diagnosis With Model-Based and Signal-Based Approaches, IEEE Transactions on Industrial Electronics 62 (6), 3757 - 3767

A survey of fault diagnosis and fault-tolerant techniques-part II : Fault diagnosis with knowledge-based and hybrid/active approaches, IEEE Transactions on Industrial Electronics 62 (6), 3768 - 3774

The diagnosis approached can be categorized into model-based approach, signal processing approach and knowledge-based method and hybrid approaches. The machine learning method could belong to knowledge-based methods. Are there any model-based or signal based methods are used for IC engines? I feel the authors could add some discussions on this aspect. The classic references could be referred to if missing.

 

2) For the formulas, the order numbers such as (1), (2) etc should be added.

 

3) The resolutions of some figures such as figure 19 should be enhanced as some texts cannot be read. 

 

4) Further work should be much enhanced as a review paper should provide the insight about future direction of the field. 

 

5) I cannot see why Figures A and B should appear. A review paper should be focused on discussions. Too many simulated figures would distract the readers from the key points. 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report (New Reviewer)

Comments and Suggestions for Authors

This paper provides a comprehensive overview of methodologies for fault detection in internal combustion engines (ICEs), including both model-based and data-driven approaches. The review covers various diagnostic methods such as vibration analysis, thermography, acoustic analysis, and optical techniques, with an emphasis on the application of machine learning models like deep neural networks and convolutional neural networks. While the paper presents a broad array of diagnostic methods and introduces recent advances in deep learning techniques, it could benefit from further refinement in terms of literature coverage, methodological comparisons, and experimental depth.

1. The models and algorithms discussed in the paper are relatively outdated. The paper primarily focuses on traditional machine learning methods, which are no longer at the forefront of fault detection in ICEs. A more thorough review of state-of-the-art techniques, including comparisons with recent deep learning architectures like transformers or advanced convolutional networks, would significantly enhance the paper’s relevance. Additionally, including comparisons with recent methodologies would help demonstrate how the proposed approach fits into the current research landscape.

2. The experimental results presented in the paper lack depth and detailed analysis. The study mentions using machine learning algorithms under three load conditions, but there is insufficient explanation or comparison of results across these conditions. It would be beneficial to include a more comprehensive set of experiments, such as tests across different engine types, fault scenarios, and varying environmental conditions, to provide a more robust evaluation of the proposed diagnostic methods.

3. The literature review does not adequately address recent advancements in fault detection technologies, particularly in the application of deep learning. Including more contemporary studies would provide a clearer context for the paper’s approach. Works such as recent developments in Intelligent diagnosis method for machine faults based on federated transfer learning, uncertainty-aware deep learning methods for health monitoring in energy systems, should be considered. This would strengthen the positioning of the paper within the broader research community.

4. While the paper outlines several fault detection techniques, the description of the methodologies is somewhat vague. For example, the application of feature selection techniques and how they impact the diagnostic results should be more thoroughly explained. Furthermore, it is not clear how the different machine learning models are optimized or compared in terms of performance metrics. A more detailed explanation of the methodology would enhance the readability and technical rigor of the paper.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

I am happy with the revision, and the paper is ready to be accepted. 

Reviewer 4 Report (New Reviewer)

Comments and Suggestions for Authors

The manuscript was revised according to the suggested comments. The authors performed the review carefully, provided a detailed response to the comments in the previous review, and the quality of the article improved substantially. As they addressed the major concerns, I recommend publishing this manuscript.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The mention of "Machine Learning Approaches" in the paper's title is very appealing; however, upon reading, it was found that the content of the paper did not fully meet this expectation.

The paper provides a comprehensive review of traditional model-based and data-driven methods, including vibration analysis, thermography, acoustic analysis, and optical methods. These methods indeed hold significant value in the field of ICE fault diagnosis, but the paper fails to cover the latest deep learning algorithms.

It is recommended that the section on data-driven methods should not only include statistical analysis but also emphasize the study of machine learning, especially deep learning algorithms. Discuss how to utilize these deep learning algorithms to extract deeper features from sensor data and improve the accuracy of fault detection.

Comments on the Quality of English Language

non

Reviewer 2 Report

Comments and Suggestions for Authors

The paper gives a review of state-of-the-art in the area of IC engines fault detection, covering both model-based and data-driven approaches. In addition, the paper presents an experimental investigation of misfire detection in IC engine, using different classifiers and feature selection/extraction techniques.

The methodology of the paper is appropriate. The style and language are clear, although there are some parts of the text where sentences are not finished or some parts of sentences appear randomly (e.g. p4, line97; p5, lines 123-124; p10, lines 253-254; and so on). Therefore, the text needs additional language and grammar revision.

The main issue, however, is the content of the paper. In the first part, the latest and most advanced techniques for fault detection are overviewed. Then, in the experimental part, a common statistic features were used to detect misfire, along with some well-known feature selection/extraction and classification techniques. It remains unclear, if this is an overview or regular article, because there is no novelty in the experimental part of the paper. The analysis of the experimental results is deep, wide and detailed, but it still gives no specific recommendation on which classification is preferred and why. In my opinion, a substantial revision is necessary regarding this issue, to find balance between these two parts of the paper.

The particular issue I would like to point to are as follows:

-          Figure 1 is not necessary; these systems can be numbered in text, defining the proper order of following chapters

-          Section 2.8 does not fit in other sections before and after it; other sections describe fault detection methods for different parts or systems of ICE, and the section 2.8 refers to hybrid methods for fault detection, without targeting particular systems. Similar is with section 2.10. Also, the order of sections needs to be revised.

-          P14, lines 405-406; the parameters given in Table 2 are repeated in the text, which is not necessary

-          Why did the Authors apply only low load (0%, 10% and 15%)? Most fault detection techniques achieve the best performance only in high load level. Pease comment this.

-          Figure 14 a, the units on the vertical axis are not marked

-          Table 3, what is the meaning of the column Preferred?

-          Figures 18, 19 and 20 are not readable; there are too many similar line colors

Comments on the Quality of English Language

Slight language revision is necessary.

Reviewer 3 Report

Comments and Suggestions for Authors

The paper presents an extensive review of machine learning techniques used in fault detection of combustion engines. The Authors also performed their own experiment aiming at the evaluation of a selected, broad spectrum of ML methods. Each of the techniques was preceded by feature selection (importance) and preprocessing. The paper presents a rather well-known combination of methods like neural network combined with classical ML and standard approach for model validation, but due to the thorough number of models compared, it is worth publishing as a review paper. The reviewer has overall a good impression, although a few things need to be addressed before possible publication. All my remarks are listed below:

1. The quality of the figures is terrible. Please increase resolution and font, especially in heatmaps, or consider the use of classical tables. If necessary, reduce the number of data to the most relevant ones to maintain clarity and readability.

2. The same applies to metric (f1-measure, etc.) plots - consider bar charts or other clearer ways for data presentation.

3. I feel like the list of abbreviations is a must in this paper. It would greatly improve the reader experience.

The above comments yield the review to recommend the paper for publication after revisions.

Comments on the Quality of English Language

The language is ok; minor spell check is required.