A Novel Feature for Fault Classification of Rotating Machinery: Ternary Approximate Entropy for Original, Shuffle and Surrogate Data
Round 1
Reviewer 1 Report
I recommend improving the captions of the following figures and graphs:
Figure 3: Improve the labels of the four types of gearbox vibration data (a), (b), (c) and (d) and assign corresponding graphs in the figure.
Figure 10: Specify which two detail components of the nine detail components and one approximate component are depicted in the figure.
Figures 11 and 12: If possible, specify which data displayed in the figures are the original, shuffle and surrogate versions for all four types gear vibration data.
Try to improve the following figures 20., 21., 22. and 23. in a similar way.
For the 3D plots in Figures 12 and 23, it would be useful to add a comment in the text of the paper to describe the displayed result. In its present form, the contribution of the third dimension in the graph is not entirely obvious.
Author Response
Response to Reviewer 1 Comments
Point 1: I recommend improving the captions of the following figures and graphs:
Figure 3: Improve the labels of the four types of gearbox vibration data (a), (b), (c) and (d) and assign corresponding graphs in the figure.
Response 1:
Four types of gearbox vibration data in Figure 3 have been assigned labels (a)~(d).
Point 2: Figure 10: Specify which two detail components of the nine detail components and one approximate component are depicted in the figure.
Response 2:
Captions in Figures 9 and 10 have stated which two IMFs (detail components) are selected for use.
Point 3: Figures 11 and 12: If possible, specify which data displayed in the figures are the original, shuffle and surrogate versions for all four types gear vibration data.
Response 3:
In Figure 11, labels of X-axis and Y-axis are “ApEn of original data” and “ApEn of shuffle data”, respectively.
In Figure 12, labels of X-axis, Y-axis and Z-axis are “ApEn of original data”, “ApEn of shuffle data” and “ApEn of surrogate data”, respectively.
Point 4: Try to improve the following figures 20., 21., 22. and 23. in a similar way.
Response 4:
Captions in Figures 20 and 21 have stated which two IMFs (detail components) are selected for use.
In Figure 22, labels of X-axis and Y-axis are “ApEn of original data” and “ApEn of shuffle data”, respectively.
In Figure 23, labels of X-axis, Y-axis and Z-axis are “ApEn of original data”, “ApEn of shuffle data” and “ApEn of surrogate data”, respectively.
Point 5: For the 3D plots in Figures 12 and 23, it would be useful to add a comment in the text of the paper to describe the displayed result. In its present form, the contribution of the third dimension in the graph is not entirely obvious.
Response 5:
Comments are added in the end of Sec. 3.1 and Sec. 3.2 to describe the displayed results in Figures 12 and 23.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Check that the style of writing is in the third person throughout. Don’t use ‘we’.
Check that the abstract provides an accurate synopsis of the paper. It is very vague in its present form.
The methodology of the proposed model must be illustrated by a clear flowchart.
The figures are of poor quality, not labeled properly, and not cited in the text.
Besides, the writing of the paper, including contributions, and methodologies, should be clearer and highlight the innovation of methods & principles.
Insufficient literature is presented to support the aim of the study. This point still needs further revision. You may refer to papers such as ‘A machine learning approach for vibration-based multipoint tool insert health prediction on vertical machining centre (VMC)’, ‘Application of bayesian family classifiers for cutting tool inserts health monitoring on CNC milling’ etc. These papers use time-domain-based statistical features of vibration signals to classify cutting tools.
This paper doesn't detail the data collection, preparation, processing, and modeling procedure.
It is unclear whether the faults made artificially while acquiring data are actual.
How to deal with the possibility of misclassification of a normal condition as faulty depending on the degree of fault?
How to deal with the data diversity of the present moment and moment in the future?
How to ensure the robustness of the model in a highly noisy environment?
Was the data normalized/ standardized?
There is no discussion on results. The results are also included considering limited examples.
The original signal in the time domain and decomposed signal using wavelet transfer should be included to visualize the residual signal.
The manuscript is more like a report than a research paper failing in solid discussion. Revise results and discussion part by critically examining results and including inferences drawn.
Overall writing of the paper is in colloquial or oral language. The grammar is improvable. There must be a thorough proofreading of the paper.
Author Response
Response to Reviewer 2 Comments
Point 1: Check that the style of writing is in the third person throughout. Don’t use ‘we’.
Response 1:
The use of “we” has been eliminated and the writing style has been made in the third person.
Point 2: Check that the abstract provides an accurate synopsis of the paper. It is very vague in its present form.
Response 2:
Authors have tried the best to further refine the abstract.
Point 3: The methodology of the proposed model must be illustrated by a clear flowchart.
Response 3:
The flowchart of the proposed method has been illustrated in Figure 1.
Point 4: The figures are of poor quality, not labeled properly, and not cited in the text.
Response 4:
Four types of gear vibration data in Figure 3 have been assigned labels (a)~(d). Authors have gained the copyright permission of Figure 2 and Figure 13 from Elsevier and Taylor & Francis, respectively, whose resolutions cannot be improved further. Also, authors have carefully checked this manuscript for ensuring that all figures have been cited in the text.
Point 5: Besides, the writing of the paper, including contributions, and methodologies, should be clearer and highlight the innovation of methods & principles.
Response 5:
Authors have tried the best to improve the writing.
Point 6: Insufficient literature is presented to support the aim of the study. This point still needs further revision. You may refer to papers such as ‘A machine learning approach for vibration-based multipoint tool insert health prediction on vertical machining centre (VMC)’, ‘Application of bayesian family classifiers for cutting tool inserts health monitoring on CNC milling’ etc. These papers use time-domain-based statistical features of vibration signals to classify cutting tools.
Response 6:
These two references have been cited as Refs. [2] and [3] in this manuscript.
Point 7: This paper doesn't detail the data collection, preparation, processing, and modeling procedure.
Response 7:
This manuscript have detailed the experimental rig, data collection and processing procedures. In addition, this manuscript does not involve modelling procedures.
Point 8: It is unclear whether the faults made artificially while acquiring data are actual.
Response 8:
All the data used in this manuscript are provided by authors of Refs. [25] and [27].
Point 9: How to deal with the possibility of misclassification of a normal condition as faulty depending on the degree of fault?
Response 9:
In this manuscript, the proposed method can reliably distinguish a normal condition from faulty ones. In the future, if unreliable in some circumstances, the proposed method should been improved further.
Point 10: How to deal with the data diversity of the present moment and moment in the future?
Response 10:
Dynamic vibration signals from defective rotating machinery are non-stationary and nonlinear. In this sense, statistics of vibration signals are time-varying. Fortunately, the proposed method has the capability to process non-stationary and nonlinear signals. As demonstrated in Figures 12 and 23, the ternary ApEn remains almost constant for an identical condition at different temporal stages.
Point 11: How to ensure the robustness of the model in a highly noisy environment?
Response 11:
If the proposed method performs badly in analyzing highly noisy data, some de-noising methods can be considered as preprocessing procedures.
Point 12: Was the data normalized/ standardized?
Response 12:
These data are not normalized/ standardized in this manuscript.
Point 13: There is no discussion on results. The results are also included considering limited examples.
Response 13:
A discussion on the results has been set up in Sec. 4. In authors’ opinions, any method cannot be applicable everywhere. In this manuscript, the proposed method works well in classifying different machinery conditions. In the future, if performing poorly in some circumstances, the proposed method should be improved further.
Point 14: The original signal in the time domain and decomposed signal using wavelet transfer should be included to visualize the residual signal.
Response 14:
In the gear case, there are four types of gear vibration data. In the rolling-bearing case, there are five types of rolling-bearing vibration data. If decomposed results of these data by both EMD and WT are visualized, at least (4+5)*2=9*2=18 figures are provided. Obviously, the number of these visualized figures is enormous, which will dilute the theme of this manuscript. Alternately, readers can reproduce the decomposed results by using datasets analyzed during the study, which are available from the corresponding author on reasonable request.
Point 15: The manuscript is more like a report than a research paper failing in solid discussion. Revise results and discussion part by critically examining results and including inferences drawn.
Response 15:
The principal contribution of this paper is to develop the ternary entropy by integrating ApEn of original, shuffle and surrogate data into a three-dimensional vector for characterizing properties of complex vibration data. Additionally, the ternary ApEn is compared with conventional temporal statistics, conventional ApEn, two-dimensional energy entropy based on empirical mode decomposition or wavelet decomposition and binary ApEn using both gear vibration data and roller-bearing vibration data containing different types and severity of faults. The results suggest that the ternary ApEn is superior to the others in identifying conditions of rotating machinery.
Authors have tried the best to revise discussion parts.
Point 16: Overall writing of the paper is in colloquial or oral language. The grammar is improvable. There must be a thorough proofreading of the paper.
Response 16:
This manuscript has been carefully checked for avoiding colloquial or oral expressions.
Round 2
Reviewer 2 Report
The authors tried to address my comments
