A Method for Constructing Health Indicators of the Engine Bleed Air System Using Multi-Level Feature Extraction
, Xi Chen 1,2 and Lei Dong 1,2Round 1
Reviewer 1 Report
The paper proposes a methodology for deriving health indicators from aircraft flight data. The algorithm is successfully assessed on engine bleed air data from the B737 fleet. The work is very interesting and well written. I suggest that the manuscript can be accepted after the authors have addressed a couple minor points:
In figures 4 to 9, all HIs appear scattered with higher density at discrete levels (i.e. a lot of data points seem to arrange in discrete horizontal bands in the graphs). Is this phenomenon related to the distribution of the training dataset (that may be the sympthom of some bias in the data) or to the way the HIs are computed?
Are the flight cycles related to an individual aircraft or to the entire fleet? In the former case, how do you explain the global trend of the data, which sometimes exceeds the thresholds, to get again inside the tolerance band after some cycles?
Author Response
We would like to express our gratitude to the reviewers for their valuable feedback and insightful comments on our manuscript.
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper investigates a new approach to construct health indicators of the Engine Bleed Air System. This method obtains health indicators by unsupervised learning and multi-level feature extraction of QAR data using ResNet Deep Autoencoder. It is validated on the QAR data of an airline and proves the superiority of the proposed model. In general, this method is relatively new and the paper has some contributions.
Some suggestions to improve the paper:
1. I suggest the authors divide Section 2 of the paper into two parts, which are the “data-level feature extraction” and the “ResNet Deep Autoencoder”, and explain them more detailed.
2. The advantages of unsupervised learning should be discussed in Introduction and more relevant literature should be added.
3. In Fig 3, what is the “short cut”, more explanation should be given?
4. In the beginning of Section 3, more description about the data sets should be given.
5. On pages 9-11, how did the authors determine the performance baselines in Figure 4 to Figure 9?
6. On page 9, why there are red points between the HI baselines in Figure 4?
7. On page 11, please explain each variable in Eq. (2).
no comments
Author Response
We would like to express our gratitude to the reviewers for their valuable feedback and insightful comments on our manuscript.
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors have revised the manuscript thoroughly, I have no further comments.
