An Improved Fault Diagnosis Approach for Pumps Based on Neural Networks with Improved Adaptive Activation Function

Round 1

Reviewer 1 Report

This paper proposed an improved adaptive activation function and apply it to five types of neural networks. The structure of the paper is relatively complete, and the results are good. However, there are still some questions that need be further explained. Main points are as follow:

1.    The introduction needs to be improved. The literatures review should not simply list what each piece of literature has done but analyze them and highlight the motivation for the methods proposed in this paper.

2.    Figure 1 is not clear. The author needs to replace the figure with a clearer one.

3.    LSTM appears only once in the article and there is no need to abbreviate it.

4.    The abstract and the article are two separate parts. In the abstract, the full name of ReLU needs to be given.

5.    The description of the simulation results is rudimentary, which needs to be further improved.

6.    In this paper, a new adaptive activation function is designed and applied to five models of neural networks. However, there is no comparison with other adaptive functions. If possible, please give some comparisons in the simulation section to highlight the advantages of the proposed method in this paper.

7.    There are some language errors and typos. The authors should check this paper carefully to avoid these issues.

8.    In the Reference section, the authors should cite more recent literature. Though the fault diagnosis approach is proposed by authors, the main objective is served as the fault-tolerant control. Some fault-tolerant control methods should be added and described briefly, such as DOI: 10.1109/TR.2018.2886278; DOI: 10.1016/j.jprocont.2018.09.003, etc.

9.    Some more discussions on the future work need to be provided.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

1. Suggesting reorganizing the introduction. first, as what described in Introduction, deep learning neural networks are used in this paper. So you'd better show the keyworks in your title. Second, you just proposed a new activation function. So, you'd better tell readers the disadvantages of current activation functions systematically.

2. Figure 1 and 2 should be improved to read. They look not good.

3. Are the results showed in Table 2 obtained by five models with your proposed activation functions? If yes, how can we know the results of the fived models with other activation functions? It's better to compare them.

4. It's very hard to see the advantages of your proposed activation function in Figure 8. Please make them clearly. 

5. Why did not mention the results of your proposed methods in fault classification based on CIFAR10?

Authors should continue to improve their English, although there is no mistake in grammar. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I have no further comments.

Reviewer 2 Report

Thank authors for revising. No comments