An Interpretable Deep Learning Method for Identifying Extreme Events under Faulty Data Interference

Round 1

Reviewer 1 Report

 

This paper presents a deep learning-based method with visual interpretability is proposed to identify seismic data under sensor faults interference.

The results are presented analytically and show that the proposed method can  effectively identify seismic data mixed with various types of faulty data while providing good interpretability. 

An interpretable deep learning method for identifying extreme events under faulty data interference

 

1. I believe that this research applies an analysis method to identify seismic data under sensor faults interference

 

2. Τhe topic of the research is original in the field to the monitoring of  the operational state of structures

 

3. Deepens the research into the identify seismic data under sensor  faults  interference

 

4. The conclusions consistent with the evidence and arguments presented and do they address the main question posed

 

5. The authors use appropriate and recent bibliography mainly in the introduction section and the section methods.

Author Response

The authors would like to thank the reviewer for his interest in the manuscript, and for the constructive comments and suggestions. The authors are glad to hear that the reviewer thinks that the proposed method can effectively identify seismic data mixed with various types of faulty data while providing good interpretability.

Based on the comments and suggestions from the reviewer, the authors will continue to explore in-depth in this research direction, and aim to make contributions in three aspects. Firstly, to generalize this method to other structures and extreme events. Secondly, to improve the accuracy of extreme events identification and sensor faults detection. Thirdly, to further study the application of interpretability in anomaly diagnosis.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper proposes an interpretable deep learning method for identifying extreme events in structural health monitoring systems, even when data interference occurs. 

[Main contributions]

1. The proposal of a novel interpretable deep learning method for identifying extreme events in structural health monitoring systems.

2. The use of transfer learning to improve the performance of the proposed method.

3. The demonstration of the effectiveness of the proposed method through experiments on real-world data.

[Improvements]

This paper presents a novel approach using the ResNet34 architecture and Grad-CAM for improving the interpretability and accuracy of image classification models. The proposed method is relatively simple, but the authors could provide more detailed information on how it compares to existing methods in terms of performance. The authors should include a comprehensive comparison of their method with other state-of-the-art techniques for image classification, highlighting the strengths and weaknesses of each approach.

The authors have provided a good description of its classification error in Section 4 Discussions. However, to make the paper more comprehensive and impactful, the authors could also discuss the potential limitations and challenges of implementing the method in real-world scenarios and its potential impact on different application domains.

[Typos]

4 DISCUSSIONS  -> 4 Discussions

Overall, this paper is well-written, but polishing the wording of the sentences could further enhance its flow and clarity. By using simpler language, straightforward sentence constructions, and logical transitions between sections, the authors can improve the readability and impact of their work.

Author Response

The authors would like to thank the reviewer for his interest in the manuscript, and for the constructive comments and suggestions. The manuscript has been revised according to the reviewer’s comments. Please find the attached pdf file for the reply.

 

 

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper discusses the challenges of identifying extreme events from monitoring data due to the interference of faulty data. A deep learning-based method with visual interpretability is proposed to identify seismic data under sensor faults interference. The transfer learning technique is employed to learn the features of seismic data and faulty data with efficiency. A posthoc interpretation algorithm is embedded into the neural networks to uncover the interest regions that support the output decision. The proposed method is verified using in-situ seismic responses of a cable-stayed long-span bridge, and the results show that the method can effectively identify seismic data mixed with various types of faulty data while providing good interpretability.

 

Some questions :

What types of sensor faults were considered in the study, and how were they introduced into the seismic data?

How was the transfer learning technique employed to learn the features of seismic data and faulty data, and what was the efficiency achieved?

What are the limitations of using the Grad-CAM algorithm for interpretability, and how were they addressed in the study?

Was the proposed method compared to existing methods in terms of performance and interpretability? If not, what are the potential drawbacks of not doing so?

How generalizable is the proposed method to other types of structures and extreme events, and what are the potential challenges in applying it to different scenarios?

To enrich your introduction, added following paper to REf and introduction.

1-Barkhordari, M. S., Armaghani, D. J., & Asteris, P. G. (2022). Structural damage identification using ensemble deep convolutional neural network models. Comput. Model. Eng. Sci, 134(2). https://doi.org/10.32604/cmes.2022.020840

2-Barkhordari, M. S., Barkhordari, M. M., Armaghani, D. J., Rashid, A. S. A., & Ulrikh, D. V. (2022). Hybrid Wavelet Scattering Network-Based Model for Failure Identification of Reinforced Concrete Members. Sustainability, 14(19), 12041. https://doi.org/10.3390/su141912041

 

Author Response

The authors would like to thank the reviewer for his interest in the manuscript, and for the constructive comments and suggestions. The manuscript has been revised according to the reviewer’s comments. Please find the attached pdf file for the reply.

 

Author Response File: Author Response.pdf