A New Framework for Isolating Sensor Failures and Structural Damage in Noisy Environments Based on Stacked Gated Recurrent Unit Neural Networks

Round 1

Reviewer 1 Report

This manuscript presented a stacked gated recurrent unit neural network for isolating sensor failures and structural damage using noisy monitored structural response. The work will be reconsidered after the following comments are replied. 

1. For the settings of the gate recurrent unit, how to design the required parameters? How these settings of these parameters affect the overall results?

2. How to guarantee the quality of the training data and distinguish different fault sources so that a rational datum can be obtained? 

3. In the formulation of the S-GRU NN, please provide detailed criteria and explain the reasons for the selection of the architecture and activation functions. 

4. For the applied dropout, why 0.3 dropout rate in the stacked GRU, FC1 and FC2 layers were utilized? Please provide support to the selection of the dropout rate and the applied layers. 

5. In practice, the ambient conditions can induce significant influences on the structural response. How robust is the method to manage the variation of the ambient conditions? 

6. In this work, the locations of the mounted sensors play a critical role in the quality of the monitored information and hence the performance of the reconstructed output and structural damage diagnosis. Please discuss the effects of sensor placement on the performance of isolating sensor failure and structural damage. Some references provide related information to this issue.

Efficient Bayesian sensor placement algorithm for structural identification: a general approach for multi‐type sensory systems. Earthquake Engineering & Structural Dynamics, 44(5), 757-774.

Health monitoring sensor placement optimization for Canton Tower using immune monkey algorithm. Structural Control and Health Monitoring, 22(1), 123-138.

Sensor placement optimization on complex and large metallic and composite structures. Structural Health Monitoring, 19(1), 262-280.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

 

 

The manuscript develops a novel isolation framework for sensor failures and structural damage considering measurement noise through extensive numerical analysis and experimental validation studies. The results demonstrate that the proposed method is highly effective, and such a study would be a significant contribution to the literature. It is advised to be published in the Journal of Buildings once the authors address the following minor comments;

 

1.     L271. The mass density of the “mold” material… do you mean modeled material?

2.     L280. The elastic moduli of the ten beam elements … Why ten? The paragraph describes a linear beam element not a 3-D bridge model. Please clarify

3.     L316 - … using the method of literature. It should be “method in the literature.’’

4.     Figure 4 legends – Health should be healthy.

5.     The word ‘’substructure’’ is used over and over in the manuscript. Do you mean part of the structure? Substructure normally means foundation (below the structure level). E.g. Ln 378. It is confusing. Instead, you could say. These four sensors are located near the damage location.

6.     Section 4.3 – why do authors use 60 SNR but not 40SNR, which is more interesting and challenging to demonstrate the effectiveness of the proposed method. Looking at Figure 13, it is clearly evident that the noise level reduces its effectiveness. Nevertheless, L388 infers the opposite.

7.     L 373 – the severities of damage simulated in the numerical study are specified as 5,10,15%, whereas Figure 12 shows as 5,15,25%. The sentence in L280 also says 5,15, and 25% damage. What is the noise level introduced in this simulation? Check out Table 1 as well.

8.     L457. ‘’The excitation of the electromagnetic shaking table was a sinusoidal acceleration in the horizontal direction’’ Please be more specific. Do you mean in the bridge longitudinal or transverse direction? Or both?

9.     Please consider putting a figure showing the measured and reconstructed acceleration time history signal, similar to the one in Figures 8 or 9.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors replied the raised suggestions and comments satisfactorily. It is recommended for publication.