AI-Driven Predictive Maintenance in Modern Maritime Transport—Enhancing Operational Efficiency and Reliability

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This work has proposed a practical working method to analyze and monitor sensor data in ship operations. The KNN method is used during the working procedure, which is used to classify faults during operations. The work is practical to improve the efficiency for shipping industry, but the current form is not acceptable for a scientific publication due to its writing, logic and confusion statements. My comments are listed below.

1.    The title should be updated. There should not be including any dot in the title.

2.    The use of abbreviations in the whole article should be updated. Only the first time it appears, the full term and the abbreviation will present. For other places, you don’t need to present the full term.  There are also wrongly-use of terms such as, from line 48 in paragraph 1 from page 2, Machine Learning Techniques (ML) ?

3.    In the paragraph 3, section 1, the 3 primary categories in PHM are listed.  I think it is wrong or unsuitable to use model-based approaches as one category.  In general, it should be physical-based model.

4.    In section 1, the description of the motivation of the work is not clear enough based on your listed data and literature. It should be fully restructured.

5.    English writing is hard to understand for some sentences, using too long sentences and clauses in a single sentence.

6.    The logic and writing of section 2: Materials and Methods are not logical, which needs to be reformed totally. It should be, at least, spilt to 2 different relevant sections about literature review and your methodology.  There are a lot of repeating and confusion structures in this section. Sub-sections are needed to address this.

7.    In this work, the authors state that they have developed a new algorithm, as seen in Figure 3. But in my opinion, it is not a new algorithm, it is just a method with a working flowchart, which can be served for better practical operational practice. It should be updated according to this.

8.    I didn’t see any necessity to put Figure 1 in the paper, since it is too general.

9.    Figures 3,4 6,7 and others are not clear with low resolution. They should be updated.

10.  There is no need to put the pseudocode in the main content of the article, for instance Equation (10),  (12) and (13). Maybe you can put them into the appendix.

11.  In the Discussion and Conclusions, the authors stated that “ this research offers a significant contribution to the field, ….” , “it is also scalable, accommodating the complexity of installations and …  ”. But I did not see any results to support such a statement.  It is a realizing of the workflow, but the resulted of PdM should be demonstrated or validated through comparisons.  This part needs to be updated.

Comments on the Quality of English Language

required to update

Author Response

Comments and Suggestions for Authors:

This work has proposed a practical working method to analyse and monitor sensor data in ship operations. The KNN method is used during the working procedure, which is used to classify faults during operations. The work is practical to improve the efficiency for shipping industry, but the current form is not acceptable for a scientific publication due to its writing, logic and confusion statements. My comments are listed below.

General response:

In order to respect the deadline, we upload the revised version of the article, fully restructured, with improvements in writing, form and content, according with the reviewers comments. Red lines represent the revised sections. We also make a request for English editing to MDPI and after receiving English language revised version, we will upload on the platform.

Below we respond punctually to each of the reviewer's comments.

 

Comment 1: The title should be updated. There should not be including any dot in the title.

Response 1: The title has been updated to "Enhancing Operational Efficiency and Reliability in Modern Maritime Transport through AI-Driven Predictive Maintenance"

Comment 2: The use of abbreviations in the whole article should be updated. Only the first time it appears, the full term and the abbreviation will present. For other places, you don’t need to present the full term.  There are also wrongly-use of terms such as, from line 48 in paragraph 1 from page 2, Machine Learning Techniques (ML)?

Response 2: The abbreviations have been updated.

Comment 3: In the paragraph 3, section 1, the 3 primary categories in PHM are listed.  I think it is wrong or unsuitable to use model-based approaches as one category.  In general, it should be physical-based model.

Response 3: The paragraph 3 have been updated.

Comment 4: In section 1, the description of the motivation of the work is not clear enough based on your listed data and literature. It should be fully restructured.

Response 4: The entire section has been restructured to highlight the motivation based on the literature review.

Comment 5: English writing is hard to understand for some sentences, using too long sentences and clauses in a single sentence.

Response 5: The full paper has been updated in terms of English language. We will also submit a request to a paid editing service provided by MDPI ( https://www.mdpi.com/authors/english ) and we will update the revised article as soon as possible.

Comment 6: The logic and writing of section 2: Materials and Methods are not logical, which needs to be reformed totally. It should be, at least, spilt to 2 different relevant sections about literature review and your methodology.  There are a lot of repeating and confusion structures in this section. Sub-sections are needed to address this.

Response 6: The section has been restructured and split in two relevant sections, named Literature review and Methodology.

Comment 7:  In this work, the authors state that they have developed a new algorithm, as seen in Figure 3. But in my opinion, it is not a new algorithm, it is just a method with a working flowchart, which can be served for better practical operational practice. It should be updated according to this.

Response 7: We want to keep the term algorithm because we think of the total approach as steps to solve problems, as we stated in the introduction.

Comment 8:  I didn’t see any necessity to put Figure 1 in the paper, since it is too general.

Response 8: In our opinion, we believe that figure 1 should be kept in the article because it reflect the idea of building this approach which will be continued with a program aimed at identifying the faults of on-board systems.

Comment 9: Figures 3,4 6,7 and others are not clear with low resolution. They should be updated.

Response 9: The figures have been updated.

Comment 10: There is no need to put the pseudocode in the main content of the article, for instance Equation (10), (12) and (13). Maybe you can put them into the appendix.

Response 10: We removed the Equation 10 to 13.

Comment 11: In the Discussion and Conclusions, the authors stated that “this research offers a significant contribution to the field, ….”, “it is also scalable, accommodating the complexity of installations and …”. But I did not see any results to support such a statement.  It is a realizing of the workflow, but the resulted of PdM should be demonstrated or validated through comparisons.  This part needs to be updated.

Response 11: We have updated this section between lines 559 and 568 to support the observation.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This paper is generally good enough for publication in Applied Science. The followings are some specific comments to be addressed by the authors:

1、 In Figure 1, the word "Exploatare" seems to be a spelling error, and the formula numbers are incorrect in lines 120 and 322.

2、 The flowcharts in Figures 3 and 4 are presented in code form, which makes it difficult to intuitively understand what the author intends to convey. It is recommended to briefly describe the functions implemented in clear language.

3、 All formulas provided by the author are set in computer language, but they should be expressed in mathematical terms. For example, in Formula 2, the author proposes a linear model, which should be mathematically described as a specific type of mathematical model. In Formula 10, computer language is used, but variables such as "min_max" and "GraphData" are not explained in detail. It is recommended that the author use pseudocode to present the method, and a writing style for pseudocode can be referenced in [1].

4、 The author mentions the concepts of life cycle and time degradation in lines 234–236 and 277–284, but these concepts are not reflected in the designed algorithm.

5、 In Formula 5, the author does not explain what (-1, 0, 1) represent, which makes it difficult to understand the confusion matrix in Figure 7, and it is unclear how the defect occurrence rate in Figure 8 is calculated.

6、 The author does not provide detailed information on the experimental data nor does he demonstrate whether the data used is correlated with the types of faults being analyzed. (For example, why these 20 parameters can be used as criteria for fault diagnosis.)

7、 In the confusion matrix for operation at nominal parameters (or with minimal defects) in Figure F12, the accuracy is only 78.3%. This suggests that the algorithm may frequently cause false alarms in daily use, classifying normal states as fault states. Moreover, due to the variability of real-world situations, sensor data collection may encounter some disturbance or significant interference. Can the algorithm designed by the author operate correctly under such interference?

8、 The author uses 20 basic operational parameters commonly found in seawater cooling systems as a monitoring array. Due to the nature of the KNN algorithm, when there is not much data, KNN can train quickly. However, as shown in Figure 1, as time progresses, historical data will continuously accumulate, and the amount of data for KNN to process will grow exponentially. In this case, how does the author handle the issue of large-scale data?

9、 Since the algorithm is designed to be general-purpose, other systems may have more than 20 parameters. How can the algorithm ensure good performance in general-purpose scenarios?

[1] Cui, Zhichao, Hui Cao, Zeren Ai, and Jihui Wang. 2023. "A Multi-Adversarial Joint Distribution Adaptation Method for Bearing Fault Diagnosis under Variable Working Conditions" Applied Sciences 13, no. 19: 10606. https://doi.org/10.3390/app131910606

Author Response

Comments and Suggestions for Authors:

This paper is generally good enough for publication in Applied Science. The followings are some specific comments to be addressed by the authors:

General response:

In order to respect the deadline, we upload the revised version of the article, fully restructured, with improvements in writing, form and content, according with the reviewers comments. Red lines represent the revised sections. We also make a request for English editing to MDPI and after receiving English language revised version, we will upload on the platform.

Below we respond punctually to each of the reviewer's comments.

 

Comment 1: In Figure 1, the word "Exploatare" seems to be a spelling error, and the formula numbers are incorrect in lines 120 and 322.

Response 1: We have corrected the spelling mistake in Figure 1 and all the numbers in the equation.

 

Comment 2: The flowcharts in Figures 3 and 4 are presented in code form, which makes it difficult to intuitively understand what the author intends to convey. It is recommended to briefly describe the functions implemented in clear language.

Response 2: We have replaced the figures 3 and 4 with corresponding pseudocode defined by relations (2) and (3).

Comment 3: All formulas provided by the author are set in computer language, but they should be expressed in mathematical terms. For example, in Formula 2, the author proposes a linear model, which should be mathematically described as a specific type of mathematical model. In Formula 10, computer language is used, but variables such as "min_max" and "GraphData" are not explained in detail. It is recommended that the author use pseudocode to present the method, and a writing style for pseudocode can be referenced in [1].

Response 3: We have explained the methodology in pseudocode. Variables such "min_max" and "GraphData" are described in relation (3) which represent the pseudocode for graphical representation of operating data values according to the user’s preference set over time intervals. The deviation of these parameters lies between the minimum and maximum values accepted for the optimal operating regime.

Comment 4: The author mentions the concepts of life cycle and time degradation in lines 234–236 and 277–284, but these concepts are not reflected in the designed algorithm.

Response 4: We have revised the Figure 1 and we explained the concept in lines 321-324.

Comment 5: In Formula 5, the author does not explain what (-1, 0, 1) represent, which makes it difficult to understand the confusion matrix in Figure 7, and it is unclear how the defect occurrence rate in Figure 8 is calculated.

Response 5: We have revised the lines 378-382 to explain what (-1, 0, 1) represent and how to interpret in confusion matrix.

Comment 6: The author does not provide detailed information on the experimental data nor does he demonstrate whether the data used is correlated with the types of faults being analyzed. (For example, why these 20 parameters can be used as criteria for fault diagnosis.)

Response 6: We have provided the missing information in lines 431 to 442. It was an error from out part, because we did not explain the origin of these parameter and how are used in the proposed approach. 20 basic operational parameters are provided by sensors at the inlet and outlet for each component and the algorithm compute other 10 and 11 additional parameters for an improved prediction using kNN classifier for ANLYSIS module.

Comment 7: In the confusion matrix for operation at nominal parameters (or with minimal defects) in Figure F12, the accuracy is only 78.3%. This suggests that the algorithm may frequently cause false alarms in daily use, classifying normal states as fault states. Moreover, due to the variability of real-world situations, sensor data collection may encounter some disturbance or significant interference. Can the algorithm designed by the author operate correctly under such interference?

Response 7: We run the program and we discovered ways to improve the accuracy. The article was updated with necessary information in lines 503 to 518.

Comment 8: The author uses 20 basic operational parameters commonly found in seawater cooling systems as a monitoring array. Due to the nature of the KNN algorithm, when there is not much data, KNN can train quickly. However, as shown in Figure 1, as time progresses, historical data will continuously accumulate, and the amount of data for KNN to process will grow exponentially. In this case, how does the author handle the issue of large-scale data?

Response 8: To address these challenges, we have provided information in line 286 to 301, 365 to 368 and relation (4).

Comment 9: Since the algorithm is designed to be general-purpose, other systems may have more than 20 parameters. How can the algorithm ensure good performance in general-purpose scenarios?

Response 9: With the response for question 8 and additional information from lines 559 to 568 and 628 to 633 the proposed algorithm will ensure good performance in general-purpose scenarios.

Author Response File: Author Response.pdf