An Unsupervised Learning Approach for Analyzing Unsafe Pilot Operations Based on Flight Data

Round 1

Reviewer 1 Report

Based on QAR data, this study used K-means clustering algorithm to classify and analyze pilot’s unsafe operation behavior patterns during flare phase. After the key characteristic parameters are determined by expert method, different types of operation patterns are obtained by clustering with K-means algorithm. Then this method is applied to analyze 86504 flights of QAR data, and the results show that there are four types of unsafe operation behavior patterns in the flare, and the relationship between airport and pilot and operation behavior patterns is also analyzed. My comments on this article are as follow.

Introduction：

1.     The term of“pattern” is not clear. It is suggested to make additional explanations on this word.

2.     Some words like“four main patterns”、“types of adverse event” and “the causes for different types of adverse events” need to be expressed more clear for avoiding reader’s confusion.

 

Methodology：

1.     In the Section 2, it is recommended to supplement and summarize the relevant research on pilot operation characteristics.

2.     It is suggested to supplement the quantity and qualification of flight experts and prove the rationality of the selected parameters.

3.     In 4.3. Experiment Results and Interpretation（279-351），the authors described the causes for different types of adverse events”as four types. But after reading the description of these four categories in the article, I think the above description (especially The second type (High pitch influence), The third type (Long flare influence)) only stays at the description level of "different types of opposite events", and does not refer to the interpretation level. It is suggested to explore deeper reasons of events.

4.     In 4.4. Further Analysis on Impacting Factor，the authors pointed out that the design problem of runway and made some analysis. However, this analysis does not fully support the viewpoint. This analysis is more inclined to explain that the first type (Headwind influence) is related to the wind speed of the airport. The wind speed of coastal airports is high, while the wind speed of downwind landing is relatively low, so the proportion of the first type (Headwind influence) is small, and that of inland airports is reverse. This part of the analysis is not very logical and persuasive, so it is suggested to adjust or supplement arguments for discussion and explanation.

5.     In part of 4.4.Further Analysis on Impacting Factor，only two pilots are taken as examples for demonstration, which is highly inconvincible It is suggested to increase the sample size.

6.     In Figure5. k-means clustering result, it is recommended to adjust the line color to make the graph more readable.

7.     In the Section 4.4, it is recommended to delete the model analysis or expand the data set. The two types of models listed in the study are A320 series, so the differences of the operation modes brought by the models cannot be better reflected.

 

Others:

It is suggested that the authors should make professional English language revisions to their articles, paying special attention to English grammar and sentence structure, so that readers can clearly understand the objectives, contents and results of the study.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

According to the QAR data, this paper studies the issue of pilots’ unsafe stick operations based on K-means clustering algorithm. The reasons for different types of unsafe pilot stick operations are analyzed. My comments addressing the technical and editorial content of the paper are summarized below.

1. There are some grammatical errors and typos that should be corrected in the paper.

2. In the introduction, the main research results and contributions are introduced too much, and the research significance is not introduced enough. For example, the specific results of the paper should not be given, such as lines 75 to 87.

3. In the related work, this part should summarize the research progress of K-means algorithm since the application of the algorithm is one of the important contributions of this study.

4. The relevant research literature is insufficient, and the latest literature, such as 2021 and 2022, is relatively few.

5. It is recommended to introduce the advantages, disadvantages and application fields of K-means in Section 3.3. There are many relevant research results in this context, such as  K-Means-Based Nature-Inspired Metaheuristic Algorithms for Automatic Data Clustering Problems: Recent Advances and Future Directions DOI10.3390/app112311246;

Estimation of the Mixed Layer Depth in the Indian Ocean from Surface Parameters: A Clustering-Neural Network Method. DOI10.3390/s22155600

6. The experience of flight experts has been mentioned many times in the paper. Can you explain it in detail？This can help readers better understand the research methods.

7. It is suggested to explain whether the data of this study can be published or obtained.

8. The references format should be unified. Please check all the references.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

1.      When the Airbus A320 landed, the automatic landing (manual landing) system has been revised to a simpler release method, not the complicated procedure of the past. The authors looked at the past 86,504 flights to see if aircraft with simple automatic landing procedures had been ruled out. This is related to the reference value and innovation of this paper.

2.      The effects of aircraft landing in crosswinds were not discussed in the study in Section 4.3, and are suggested to be included in the study results.

3.      The study sample was only for two different airlines, and I think the study limitations should be clearly stated.

4.      For an aircraft, it is best to choose a headwind during takeoff and landing. Taking off against the wind is to increase the lift of the aircraft, and landing against the wind is to increase the drag.

The author and reference 34 propose that the runway design is based on the principle that the aircraft takes off against the wind and lands with the wind. Therefore, the argument should be made clear.

5.      The findings should be more clearly stated, and a discussion section is suggested.

6.      The conclusions are insufficient to support the statement of the results.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The authors have improved their manuscript according to the reviewers' comments. It can be accepted for publication in this version.

Author Response

Thank you for the positive comments on the revised version. Following your suggestion, we have carefully gone through the paper and made our best efforts to improve the English language and style. Please kindly refer to the new version for detailed revisions which were yellow highlighted.