Interpretable Recurrent Variational State-Space Model for Fault Detection of Complex Systems Based on Multisensory Signals
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1. There are too many language problems in the current version, which reduce the readability of the paper. Please polish the paper carefully.
2. For the simulation of the state responses, it is better to show the difference in normal case and faulty case.
3. After the main theorem, some remarks are needed.
4. Please discuss the limitations of the derived conditions.
5. The authors could enhance the introduction with some recent related works such as Hybrid-triggered and fault-tolerant observer-based control for neural networks under malicious attacks.
6. The future direction of this manuscript should be included in the conclusion section.
Comments on the Quality of English Languagenil
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper proposed a recursive variational state space model for liquid rocket engine fault detection. Although the paper is interesting trying to combine state space models with variational inference, it is written poorly, and the reader must guest a lot of things. Also, simulation experiments should be included since lack comparisons with state-of-the-art models. Also, in my opinion the structure of the paper has to be redesigned to improve clarity and overall presentation.
- b_t and ε_t in equation #1 are not defined. Also, the input vector is missing in the first equation. Explain why.
- Please explain the inference network of VAE. Perhaps, the term should be replaced with encoding network (line 120).
- In figure #1 or figure #2, the use of VAE should be presented.
- The training process is presented poorly. The revised version should contain information about the training of the model. Also, questions arise, such as the number of phases in the training, the hyper-parameters, the optimization algorithms, and others.
- In lines 189–199, the well-known Evidence Lower Bound (ELBO) should be explained better.
- Why is constant in equation #8?
- Is the input of the ARD network sensor data s_t(i) of the model? Please explain.
- The paper includes simulation comparisons with simpler versions of the proposed model. Although, this could be useful, comparisons with state-of-the-art models in the same application are needed.
A minor editing is needed
Author Response
Please see the attachment.
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsAlthough the manuscript has been improved with the revision, the overall presentation of the paper is poor and needs further improvement.
Some drawbacks that must be resolved follow:
1. Why were the state-space model equations removed? These equations, in my opinion, are crucial for the reader to comprehend the model architecture and develop intuition.
2. Correct equations #5, #7.
3.Furthermore, the reader is left to make assumptions regarding the final step of fault diagnosis due to unclear presentation.
4. The absence of simulation comparisons with some other comparative model is important to the scientific significance of the presented paper and its applicability.
Comments on the Quality of English Language
No comments
Author Response
Please see the attachment.
Author Response File: Author Response.pdf