An Improved Fault Diagnosis Approach Using LSSVM for Complex Industrial Systems
Round 1
Reviewer 1 Report
This paper deals with an approach for fault diagnosis based on the combination of particle swarm optimization with wavelet mutation and the least square support(WMPSO-LSSVM)
Although the tools used are known, their association is interesting and can be considered as a contribution. Indeed, the filtering of signals by wavelet packet transform makes it possible to reduce the risk of falling on local optimum when using the PSO, and the use of PSO in support of Lagrange multipliers can improve the optimization of LS-SVM parameters.
The paper is well written, it is easy to read and understand. Obtained results, using the database from the Guangdong Provincial Key Laboratory of Petrochemical Equipment Fault Diagnosis of China, show the effectiveness of the proposed method compared to other existing approaches.
The paper can be improved on the following points:
1) There are many parameters in the equations and the proposed algorithms, although these parameters have been defined for each equation, a nomenclature table would, in my opinion, allow readers to easily find the meaning of the parameters and variables.
2) The accuracy obtained is interesting (over 95%). There are decision-making methods in the literature, which are easy to implement and which allow to achieve 100% accuracy, it would be interesting to mention them in the introduction, such as the Temporal based SVM which allows decision-making on an observation window instead of decision making on each single observation. See for example: A temporal-based SVM approach for the detection and identification of pollutant gases in a gas mixture. Applied Intelligence (2021). https://doi.org/10.1007/s10489-021-02761-0.
2)The choice of an LS-SVM was not justified by the authors? the choice of the kernel function either? these choices should be justified.
3) The introduction can be enhanced with review papers on fault diagnosis and prognosis to give readers an overview of the existing methods. As for example: A survey of fault diagnosis and fault-tolerant techniques; part i: fault diagnosis with model-based and signal-based approaches, IEEE Trans. Ind. Electron. 62 (6) (2015)
Author Response
Dear Reviewer:
Thanks to your advice, the revised article is clearer and provides more valuable information for readers in related fields.
In the following, we provide a detailed statement describing the changes made to the paper in this revision. We would like to thank the editor and all reviewers for their thoughtful review and constructive comments on our manuscript. We believe that the additional changes, which we have made in response to the reviewers’ comments, can greatly help us to improve the results and presentation of this paper. Below is our point-by-point response to the reviewer’s comments.
your sincerely
Shuyue Guang , Darong Huang, Shenghui Guo, Hongtian Chen
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In general, this is a rather well organized manuscript. It could be recommended for publication if the following issues are properly addressed.
General comments:
The improved fault diagnosis approach using LSSVM is interesting, but not revolutionary approach. The authors just modify the existing algorithms and optimize some of their parameters. The novelty of the introduced approach should be discussed in more detail.
The authors use the wavelet packet decomposition algorithm to decompose the vibration signals into layers. Then, each signal component is used to generate a set of indexes (labels). The further decision making is done according to those sets of indexes. The role of deep learning should be explained in more detail. For example, the authors do not confusion matrices which are an integral part of the characterization of any deep learning classifier. Therefore, the authors cannot present the sensitivity and the specificity of their classifier. These issues should be addressed in detail.
The authors could expand the literature overview. An alternative approach to the index-based characterization of the vibration signal is the feature extraction approach. For example, vibration signals can be converted into two-dimensional digital images representing the patterns of permutation entropy of those signals. Later, deep learning algorithms are used for the classification of faults by means of automatic analysis of those digital images. A typical reference: Permutation entropy based 2D feature extraction for bearing fault diagnosis. Nonlinear Dynamics (2020) vol.102, 1717-1731.
Particular comments:
The title of the first panel in Figure 13 is: "The frequency spectrum of the original signal". The authors should double-check Figure 13. It looks like a time signal, not its frequency spectrum.
Equation 21 - the authors mention that the Morlet base function was selected for the sake of simplicity. The authors are required to elaborate on the simplicity versus accuracy. What is the gain is the simplicity and losses in the accuracy.
Author Response
Dear Reviewer:
Thanks to your advice, the revised article is clearer and provides more valuable information for readers in related fields.
In the following, we provide a detailed statement describing the changes made to the paper in this revision. We would like to thank the editor and all reviewers for their thoughtful review and constructive comments on our manuscript. We believe that the additional changes, which we have made in response to the reviewers’ comments, can greatly help us to improve the results and presentation of this paper. Below is our point-by-point response to the reviewer’s comments.
your sincerely
Shuyue Guang , Darong Huang, Shenghui Guo, Hongtian Chen
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors did manage to perform a proper revision of the manuscript. This article now can be recommended for publication in the Journal.
