Sandpiper Optimization with a Deep Learning Enabled Fault Diagnosis Model for Complex Industrial Systems

Round 1

Reviewer 1 Report

The manuscript 'Fault diagnosis method for differential inverse gearbox of crawler combine harvester based on order analysis' is focused on actual problem оf condition monitoring. The authors use the combination of the known methods such as CWT, BLSTM, sandpiper optimization algorithm. The study has potential of well results. Nevertheless, the manuscript has significant problems with quality and clarity of presentation.

Some questions and suggestions are following:

1. The 'Introduction' and 'Literature Review' do not explain to the reader the limitations of the existing methods. Why is a new method of signal processing proposed? What is a research statement of problem?

2. Why did the authors choose the CWT? The decision of method choice must have some description. 

3. The Fig.1 is not descriptive for the article.

4. The size of training, validation and test datasets of the proposed method is not described.

5. Comparative methods have no any description. The size of training, validation and test datasets of the comparative method is not described.

6. Did the authors train comparative methods or did they get the characteristics from some sources?

7. The article doesn’t have description of the epochs number for datasets. The figures 5,6,9,10 do not answer about training epochs number.

8. What is mean of ‘Class-1’, ‘Class-2’, ‘Class-3’…. Each class must have some description.

9. Abbreviations do not always have full designations.

Author Response

The manuscript 'Fault diagnosis method for differential inverse gearbox of crawler combine harvester based on order analysis' is focused on actual problem оf condition monitoring. The authors use the combination of the known methods such as CWT, BLSTM, sandpiper optimization algorithm. The study has potential of well results. Nevertheless, the manuscript has significant problems with quality and clarity of presentation.

Some questions and suggestions are following:

1. The 'Introduction' and 'Literature Review' do not explain to the reader the limitations of the existing methods. Why is a new method of signal processing proposed? What is a research statement of problem?

As per the reviewer comment, the drawbacks of the existing models and research gap is provided in the revised manuscript. Kindly refer Page 3, Section 2, Paragraph 4.

2. Why did the authors choose the CWT? The decision of method choice must have some description.

Thank you for the comment. The CWT is a time-frequency domain transform of the original signal and can contain most of the information of the vibration signals.. In this method, CWT is formed by discomposing vibration signals of rotating machinery in different scales using wavelet transform. Then the DL model is trained to diagnose faults, with CWT as the input.

3. The Fig.1 is not descriptive for the article.

As per the reviewer comment, the figure 1 is removed from the revised manuscript.

4. The size of training, validation and test datasets of the proposed method is not described.

As per the reviewer comment, the experimental details are provided in the revised manuscript. Kindly refer Page 7, Section 4, Paragraph 1.

5. Comparative methods have no any description. The size of training, validation and test datasets of the comparative method is not described.

We have made a detailed literature review of stating the existing methods and a detailed comparison study is also made in the revised manuscript.

6. Did the authors train comparative methods or did they get the characteristics from some sources?

We have referred the literature for comparative methods.

[24]      Yang, Y., Zheng, H., Li, Y., Xu, M. and Chen, Y., 2019. A fault diagnosis scheme for rotating machinery using hierarchical symbolic analysis and convolutional neural network. ISA transactions, 91, pp.235-252.

[25]      Surendran, R., Khalaf, O. I., Andres, C. (2022). Deep Learning Based Intelligent Industrial Fault Diagnosis Model. CMC-Computers, Materials & Continua, 70(3), 6323–6338.

7. The article doesn’t have description of the epochs number for datasets. The figures 5,6,9,10 do not answer about training epochs number.

The parameter settings are given as follows: learning rate: 0.01, dropout: 0.5, batch size: 5, epoch count: 50, and activation: ReLU.

8. What is mean of ‘Class-1’, ‘Class-2’, ‘Class-3’…. Each class must have some description.

As per the reviewer comment, the dataset details are given in the revised manuscript. Kindly refer Page 7, Section 4, Paragraph 1.

The experimental validation of the SPOAI-FD technique is carried out by means of the automotive gearbox and bearing fault dataset [23]. The former dataset comprises seven classes whereas the latter dataset includes 10 classes. The first dataset holds 7 types of health status like an outer race bearing fault, a minor chipped gear fault, a missed tooth gear fault, three types of compound faults (Normal, Minor chipped tooth, Missing tooth (0.2 mm), and Missing tooth (2 mm)). The second dataset has both normal and fault data. The types of bearing fault have Inner race (IF), Outer race (OF), and Ball faults (BF). There-fore, there are a totally of 10 kinds of bearing health status under varying loads.

9. Abbreviations do not always have full designations.

As per the reviewer comment, the usage of abbreviations and acronyms are corrected in the revised manuscript.

Reviewer 2 Report

In this article, a new deep learning-based fault diagnosis method is proposed to solve the problem in complex industrial systems. The topic is interesting, and this study contains some materials to publish. An improvement of the manuscript's quality would be better. In the reviewer's opinion, specific comments are given as follows:

1. The language expression of the article needs to be further improved.

2. The article is not innovative enough, and related research is more common in the literature in recent years.

3. What is the innovation of this research, what is its significance, what is improved or improved, this should be described in more detail in the article.

4. The description in the introduction is too brief, and many advanced methods are not discussed, which is a serious problem. I hope that the authors can add some new references in order to improve the reviews and the connection with the literature. Explainable intelligent fault diagnosis for nonlinear dynamic systems: From unsupervised to supervised learning; Fault detection for nonlinear dynamic systems with consideration of modeling errors: A data-driven approach.

5. The conclusion part is short and not refined enough, and should be further developed and given in detail for future efforts.

6. It seems that the actual engineering data can better verify the effectiveness of the proposed method.

Please kindly revise the manuscript according to these suggestions. We look forward to receiving your revision.

Author Response

In this article, a new deep learning-based fault diagnosis method is proposed to solve the problem in complex industrial systems. The topic is interesting, and this study contains some materials to publish. An improvement of the manuscript's quality would be better. In the reviewer's opinion, specific comments are given as follows:

1. The language expression of the article needs to be further improved.

As per the reviewer comment, we have improved the language quality of the manuscript and thoroughly proofread for grammatical as well as typographical errors.

2. The article is not innovative enough, and related research is more common in the literature in recent years.

As per the reviewer comment, the novelty of the paper is defined in the revised manuscript. Kindly refer Page 1, Abstract.

3. What is the innovation of this research, what is its significance, what is improved or improved, this should be described in more detail in the article.

Based on the reviewer comment, we have provided the paper contribution in the revised manuscript. Kindly refer Page 2, Last paragraph.

4. The description in the introduction is too brief, and many advanced methods are not discussed, which is a serious problem. I hope that the authors can add some new references in order to improve the reviews and the connection with the literature. Explainable intelligent fault diagnosis for nonlinear dynamic systems: From unsupervised to supervised learning; Fault detection for nonlinear dynamic systems with consideration of modeling errors: A data-driven approach.

Based on the reviewer comment, we have given necessary information by citing the above mentioned references in the revised manuscript. Kindly refer Page 16, References.

5. The conclusion part is short and not refined enough, and should be further developed and given in detail for future efforts.

Based on the reviewer comment, the key findings and possible future works are included in the conclusion section. Kindly refer Page 16, Section 5.

6. It seems that the actual engineering data can better verify the effectiveness of the proposed method.

Please kindly revise the manuscript according to these suggestions. We look forward to receiving your revision.

As per the reviewer comment, the experimental details are provided in the revised manuscript. Kindly refer Page 7, Section 4, Paragraph 1.

 

Round 2

Reviewer 1 Report

Thank you for your efforts to improve the manuscript. However, the manuscript is needed additional corrections to increase clarity for the reader.

1. Reference [23] does not lead to a datasets.

2. The added description of studied defects does not accordance a 'class number', which is used the manuscript as before.

3. The figures 4,5,8,9 don't shown the intersection moment of training and validation curves or validation extremum as before.

4. The manuscript don't have parameters of compared FFTKNN, FFTSVM, FFTDBN, FFTSAE, CNN, CNN2, and IIFD-SOIR. as before.

Author Response

Thank you for your efforts to improve the manuscript. However, the manuscript is needed additional corrections to increase clarity for the reader.

1. Reference [23] does not lead to a datasets.

As per the reviewer comment, the references of the dataset are given in the revised manuscript.

• Saravanakumar, N. Krishnaraj, S. Venkatraman, B. Sivakumar, S. Prasanna et al., “Hierarchical symbolic analysis and particle swarm optimization based fault diagnosis model for rotating machineries with deep neural networks,” Measurement, vol. 171, no. 108771, pp. 1–8, 2021.
• A. Smith and R.B. Randall, 2015. Rolling element bearing diagnostics using the Case Western Reserve University data: A benchmark study. Mechanical systems and signal processing, 64, pp.100-131.

2. The added description of studied defects does not accordance a 'class number', which is used the manuscript as before.

Based on the reviewer comment, the dataset details with proper class labels are defined in the revised manuscript. Kindly refer Table 1.

Dataset	Class number	Class label
Dataset-I	Class 1	Outer Race Bearing Fault
	Class 2	Minor Chipped Gear Fault
	Class 3	Missed Tooth Gear Fault
	Class 4	Normal
	Class 5	Minor Chipped Tooth
	Class 6	Missing Tooth (0.2 mm)
	Class 7	Missing Tooth (2 mm)
Dataset-II	Class 1	Outer Race Bearing Fault
	Class 2	Minor Chipped Gear Fault
	Class 3	Missed Tooth Gear Fault
	Class 4	Normal
	Class 5	Minor Chipped Tooth
	Class 6	Missing Tooth (0.2 mm)
	Class 7	Missing Tooth (2 mm)
	Class 8	Inner Race (IF)
	Class 9	Outer Race (OF)
	Class 10	Ball Faults (BF)

3. The figures 4,5,8,9 don't shown the intersection moment of training and validation curves or validation extremum as before.

Thank you for the comment. From the figures, it is apparent that the training and validation curves reached saturation point with an increase in epochs.

4. The manuscript don't have parameters of compared FFTKNN, FFTSVM, FFTDBN, FFTSAE, CNN, CNN2, and IIFD-SOIR. as before.

We have made a detailed comparison study with existing models such as fast Fourier transform (FFT) with k-nearest neighbor (KNN), FFT with support vector machine (FFT-SVM), FFT with stacked autoencoder (FFTSAE), convolutional neural network (CNN), CNN2, and Intelligent Industrial Fault Diagnosis using Sailfish Optimized Inception with Residual Network (IIFD-SOIR) using the following citations:

[25]      Yang, Y., Zheng, H., Li, Y., Xu, M. and Chen, Y., 2019. A fault diagnosis scheme for rotating machinery using hierarchical symbolic analysis and convolutional neural network. ISA transactions, 91, pp.235-252.

[26]      Surendran, R., Khalaf, O. I., Andres, C. (2022). Deep Learning Based Intelligent Industrial Fault Diagnosis Model. CMC-Computers, Materials & Continua, 70(3), 6323–6338.

 

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Thank you for your efforts to improve the manuscript. The manuscript is much better now.