Constructing Condition Monitoring Model of Harmonic Drive
Round 1
Reviewer 1 Report
This research proposes a harmonic drive anomaly detection model by combiningdiscrete wavelet transforms, Log Mel-spectrogram, and EffientNetV25 network
architecture. The model uses wavelet transforms to separate the original sample audio
into audio components representing each frequency interval, then uses the Log Mel
spectrogram to extract features, and finally inputs features into the neural
network for training. The proposed detection model uses only the sound
of mechanical operation as the anomaly judgment.
This paper is well organized and written and can be recommended for publication.
Author Response
Please refer to the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In the paper anomaly detection model of a harmonic drive, which can detect whether the harmonic drive has a gear-failure problem through the sound recorded by the microphone was proposed.
In the research, multi-layer Discrete Wavelet Transform (DWT) was used.
Methodology used - correct.
References and conclusions from the research - correct.
Author Response
Please refer to the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors present a harmonic drive anomaly detection model based on discrete wavelet transformations, log mel- spectrogram and efficientnetv2s network architecture.
The work seems to be promising, but several changes are necessary before its publication. The current version contains errors in equations, figures and tables enumeration that make difficult follow the whole study.
English needs to be improved too.
General Comments
Avoid acronyms in the abstract whenever it is possible, e.g. DWT because it is not used.
Remember to use dataset or data set.
Redo the figures change the yellow with a different colour because the white text is not well readable.
Correct numeration of all figures and tables.
Feature Application
proposed --> proposes
Related Works
I will reorganize this section by rearranging the subsections order as follows 2.3, 2.2 and 2.1.
Table 1 is not referenced.
Subsection 2.3
analysis[7] --> analysis [7]
Subsection 2.4
You don't need to specify the full name for CNN acronym.
Section 3
Remove the first paragraph from line 136 to line 140.
Subsection 3.1
The first figure is 1, please correct the text in order to start from 1 and not 2.
You talk about two datasets, so please review the second sentence in the first paragraph. You repeat twice the same thing.
Subsection 3.2.1
, while the MIMI DC ... --> ; while the MIMI DC ...
Subsection 3.2.2
three second --> three seconds
Section 3.3
The current Figure 3 (that should be Figure 2) is not referenced in the text.
Improve the caption for the current Figure 4.
Reviews equations from 1 to 4 in subsection 3.3. Furthermore you should start the numeration from 3 and not 1.
Section 4
remove the first paragraph. from line 275 to line 279.
Subsection 4.1
Table numeration starts from 5 and not 14.
Subsection 4.2
Line 305: the proposed model one ... --> the proposed model ...
Table 17: Comparison of with ... --> Comparison with ...
Author Response
Please refer to the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors have applied almost all my previous comments, however there are still some minor changes to be applied.
When you refer to equation you have to consider the same term, such as Equation or equation.
When you introduce an equation and use :, you cannot start with capitol letter, as in line 218 and 219. I would rewrite lines 28, 219, 220 and 221, as follows:
This research used 15-level wavelet decomposition to obtain the coefficients set S ={cD1,cD2,cD3,cD4,cD5,cD6,cD7,cD8,cD9,cD10,cD11,cD12,cD13,cD14,cD15,cA15}. The corresponding equations, 218 Equations (3), (4), (5), and (6), are as follows:
Section 4 title is Methods, please correct the typo error.
Author Response
Please see attachement
Author Response File: Author Response.pdf
