A Joint Stacked Autoencoder Approach with Silhouette Information for Industrial Fault Detection

Round 1

Reviewer 1 Report

In this paper, a silhouette stacked autoencoder (SiSAE) model is constructed for the process data by considering both global/local information and silhouette information, which depicts the link between local/cross-local. Experimental results verified the effectiveness of the method. The work is interesting, some comments are as follows:

1. How does the hierarchical clustering partition the raw data into many blocks?

2. What’s the effect of the silhouette loss function in the framework?

3. The literature review about fault diagnosis methods should be enhanced. Such as: An intelligent diagnosis framework for roller bearing fault under speed fluctuation condition; Hybrid distance-guided adversarial network for intelligent fault diagnosis under different working conditions.

4. Some future works are suggested to add in the conclusion.

5. Please proofread the English language throughout the manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This manuscript presents a silhouette stacked autoencoder (SiSAE) model for fault detection and validated the efficiency of the model with two case studies. Overall, the manuscript is well written and easy to follow, but below please find several comments that the authors may consider. 

(1) For the Tennessee Eastman process, the fault detection rate of different methods is given in Table 1. For these results, the performance of the SiSAE model is not as good as the classic KPCA model for faults 3, 9, and 15. The authors may comment on this.

(2) For the validation of the algorithm with semiconductor data, the authors mentioned that 16 spectra in each layer are available and 7 of them were used for fault detection purpose. The authors may comment on how these 7 spectra were selected.

(3) The authors may consider re-organize Fig. 3, some of the captions are too small to read.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf