Series Arc Fault Detection under Vibration Condition Based on N-M-M-B

Response to Reviewer X Comments

1. Summary

Thank you very much for reviewing our manuscripts in your busy time. By carefully reading your review opinions, we found our own shortcomings and carefully revised the paper. In this process, we benefited a lot. Thank you again for your valuable comments on the paper.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Can be improved

We have added references to the article

Are all the cited references relevant to the research?

Yes

Is the research design appropriate?

Yes

Are the methods adequately described?

Can be improved

We have modified the description of the content of the article in the introduction in order to clearly express the research content of the article.

Are the results clearly presented?

Can be improved

We can modify the conclusion of the article so that we can clearly express the conclusion of the article.

Are the conclusions supported by the results?

Can be improved

We add data description to the conclusion so that the conclusion can be fully supported by the data.

3. Point-by-point response to Comments and Suggestions for Authors

1、Comments 1: Please highlight the novelty of the proposed method in comparison to the other methods in the field.

Response 1: Thanks for your suggestions, we have added a section in the introduction to compare with other algorithms in this field, so as to emphasize the novelty of the algorithm proposed in this paper. The details are as follows :

To address the above problems, an arc fault generator with adjustable vibration frequency and amplitude is designed in this paper, which can simulate the SAF under different vibration conditions. The noise-assisted multivariate empirical mode decomposition (NA-MEMD) arc feature extraction method is constructed by combining the noise channel signal and the multivariate empirical mode decomposition, so as to better extract the frequency domain characteristics of the SAF current under vibration condition and solve the problem of feature extraction modal aliasing. In the meanwhile, the arc voltage and circuit are simultaneously analyzed to avoid misjudgment caused by single information. Finally, the multi-scale technology is combined with the traditional CNN and bidirectional long short-term memory network (BiLSTM) to construct MSCNN-BiLSTM arc fault detection model, so that BiLSTM module can receive more and more abundant multi-level features from CNN and improve the accuracy of the detection model.

Comments 2: Please increase quality of the figures.

Response 2: Thank you for your comments. We are very sorry that the figures in our paper do not meet the clarity requirements. We have modified the unclear figures in the paper to improve the quality of the figures.

Comments 3:Please reorganize the conclusion and enrich it by data.

Response 3: Thank you for your comments on the paper, which gives us a deeper understanding of how academic contributions and innovations should be condensed. We have modified the conclusion part and enriched it by using the data. The results are as follows :

6.Conclusion

In this paper, the back-end arc fault experiment of frequency converter under vibration condition is carried out. The SAF generated by the electrical contact point at a specific vibration frequency under the mechanical fault state of the motor is simulated. Compared with the fault simulation at the front-end of the frequency converter, the scheme proposed in this paper is more suitable for the actual SAF under the mechanical vibration of the motor. In this paper, noise-assisted multivariate empirical mode decomposition is used to extract the features of arc current and voltage, which reduces the time of feature extraction. Compared with CEEMDAN and other algorithms with high detection accuracy, the feature extraction time of single sample is shorter, which reduces the calculation amount for subsequent detection and improves the detection accuracy.

Finally, this paper proposes a CEEMDAN-MPE-MSCNN-BiLSTM model based on the SAF detection. The accuracy of this model for SAF detection at the back-end of frequency converter can reach 99.69 %. Compared with the detection methods in other papers, this paper not only improves the detection accuracy, but also controls the detection time of a single sample at 0.044 ms. While improving the detection accuracy, the detection time is controlled, which provides the theoretical basis for the development of subsequent arc fault detection devices and realizes the preliminary judgment of the cause of motor mechanical faults. In addition, the detection accuracy of the model proposed in this paper can reach 99.37 %, 98.75 % and 95.94 % respectively under 10dB, 20dB and 30dB noise signal interference, and it still has high detection accuracy under different fault conditions. Therefore, the model proposed in this paper has good noise immunity and generalization.