An Intelligent Fault Diagnosis Approach for Multirotor UAVs Based on Deep Neural Network of Multi-Resolution Transform Features

Round 1

Reviewer 1 Report

- The title uses the word "Drones" while the abstract and keywords use the word "UAV". It is better to use the same for all of them.

  - Please avoid long sentences in the abstract and keep it as simple as possible.   - Line 25: "the proposed model" ===> "The ....."   - Lines 39-41": "Consequently, to improve a fault diagnostic model for helicopters ...." ===> The meaning of this sentence is ambiguous (since the target of this work is UAVs and not helicopters).   - The authors are invited to summarize their contributions in the form of a list of short sentences in the introduction.   - It is better to split the content of the introduction into two parts and place the study of related works in a separate section.   - The way in which Table 1 is presented makes it a bit difficult to understand.   - The authors are invited to add a new paragraph in the introduction or the related work section about the security aspects related to UAV communications.   - For this purpose, they are invited to consider/insert the following inserting references (and others): https://ieeexplore.ieee.org/document/9842403; https://link.springer.com/chapter/10.1007/978-3-319-94496-8_7    - Line 75: DWT = ???   - It would be useful to add a new section directly after the introduction that gives an overview of the possible types of faults and some examples of them, along with some preliminary information about UAVs in general.   - I believe the writers provide far too much theoretical material regarding the Discrete Wavelet Transform and Deep Neural Networks. I invite them to reduce the details for these two concepts.   - It is not clear which Deep Learning technique was adopted in this work.   - Figure 10 may be split into two figures to make it easier to understand.   - Line 412: "Relief algorithms are efficient and general-purpose attribute estimators" ===> The authors need to provide more arguments about the use of this technique.   - Line 426: "The χ2 statistical optimal feature selection approach was employed" ===> similarly the authors are invited to provide more arguments about the choice of this approach.   - Section 4 is too long and needs to be split into subsections.    - The authors are invited to share the dataset and code they developed in this work.     - The conclusion section is too long and needs to be shortened.

 

Author Response

Please see the attachment. 

Author Response File: Author Response.docx

Reviewer 2 Report

-The paper should be interesting ;;;

-it is a good idea to add a block diagram of the proposed research/review (step by step);;;

-What is the result of the analysis?;;;

-figures should have high quality. ;;;;; Fig. 7, 8,12

-labels should be used added;;;

-text should be formatted;;;;

-please add photos of the application of the proposed research, 2-3 photos ;;; 

-what will society have from the paper?;;

-labels of figures should be bigger;;;;

-Is there a possibility to use the proposed research/methods/classification methods, neural network for other topics;;;

"Thermographic Fault Diagnosis of Shaft of BLDC Motor";;;

-references should be from the web of science 2020-2022 (50% of all references, 30 references at least);;;

-Conclusion: point out what have you done;;;;

-please add some sentences about future work;;;

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This article proposed a hybrid-based transformed discrete wavelet and DNN scheme to compensate for it. Due to the availability of wide-scale, high-quality, and comprehensive soft-labelled data extracted from a selected hovering quad-copter incorporated with an accelerometer sensor via experimental work, a data-driven intelligent diagnostic strategy was investigated. The topic of the article falls within the scope of the journal, and the article is organized very well. Herein are some minor comments:

*Write the contributions in bullet points in the Introduction section.

*I think Fig. 1 is not needed, you can remove it.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors took into consideratoin my remarks. I have no other suggestions to make. Goog luck.

 

Author Response

Response to Reviewer 1 Comments

 

Point 1: The authors took into consideratoin my remarks. I have no other suggestions to make. Goog luck.

 

Response 1: Thank you for checking our edited version based on your first-round comments. Best of luck to you too.

Reviewer 2 Report

-Fig. 9 damages can be bigger;;

-Fig. 12 please add SI units [m/s^2]

-Is there a possibility to use the proposed research, neural network for other topics, for example, thermal imaging?;;;

"Thermographic Fault Diagnosis of Shaft of BLDC Motor";;;

Author Response

Point 1: Fig. 9 damages can be bigger;;.

 

Response 1: Acknowledged. The picture is re-uploaded with a higher resolution and bigger format.

Highlighted in line: [355]

 

 

Point 2: Fig. 12 please add SI units [m/s^2].

 

Response 2: Thank you for the suggestion; The units are added in the Y-axis of each figure.

Highlighted in lines: [383-386]

 

 

Point 3: Is there a possibility to use the proposed research, neural network for other topics, for example, thermal imaging?;;;

"Thermographic Fault Diagnosis of Shaft of BLDC Motor";;;.

 

Response 3: Yes, there is.

A recently published paper: https://www.mdpi.com/1424-8220/22/21/8537 adopted a methodology of using thermographic images and convolutional neural network for fault diagnosis of BLDC motor. This can also be achieved by implementing our introduced methodology by using a vibration accelerometer to acquire vibration signals where the accelerometer is fixed onto the BLDC motor. The same proposed deep neural network can be utilized.

However, the discrete wavelet transform will not be recommended as it provides better results in non-stationary operational conditions. Time-domain statistical features can be used instead of time-frequency domain statistical features.

We would like to thank you for this question. It opened future work doors.

The article is cited among another one for clarification purposes.

Highlighted in lines: [464-468]