Multi-Defect Identification of Concrete Piles Based on Low Strain Integrity Test and Two-Channel Convolutional Neural Network
Round 1
Reviewer 1 Report
The current study develops a method to classify defects in pile foundations using a two-channel convolutional neural network (TC-CNN) and low-strain pile integrity test (LSPIT). Overall, six types of multi-defect piles were simulated in Abaqus, and then, the python scripts for batch modeling were recorded to control the parameters of defects and to perform adaptive meshing on the multi-defect piles. Afterward, the numerical data of the LSPIT signal, after being verified using experimental results, were used as the input data for the TC-CNN model. The research paper shares valuable information with potential readers in the journal, and thus, it is recommended to be published in the journal. This paper clearly demonstrates the capabilities of CNN application in the supervision of piles with creative methodology. However, some adjustments should be made by the researchers to improve the quality of the paper, as follows:
· The sentences in lines 93-96 are recommended to be reorganized to avoid ambiguity and confusion.
· The word “COOK” in lines 164-165 is advised to be explained more in detail to avoid confusion.
· The explanation of the “experimental steps” subsection, starting from the 274th line, should be from an instructive manner to a more descriptive manner in the past simple.
· Fig. 7 should be explained more in detail by focusing on the defect types and their influence on the velocity-time curve.
· The reason for choosing layer 3 for denoising the waveform effect should be explained in detail, lines 343-346. According to the defined criteria in lines 330-334, the results of layers 1 and 2 also show a good match. For that reason, the reason behind choosing a 3-layer decomposition should be described in detail.
· Fig. 9 should be explained in detail to show the differences in the wavelet time-frequency diagram of 6 types of multi-defect piles.
· The misspelling of the word “fully connected layer” in Fig 10, line 411, should be corrected.
Author Response
Point 1: The sentences in lines 93-96 are recommended to be reorganized to avoid ambiguity and confusion.
Response 1: Many thanks for the comment.
The authors have reorganized this part of the content.
Yufeng Qin input the rolling bearing vibration spectrum signal and the corresponding generalized S transform time-frequency diagram into 1D-CNN and 2D-CNN, respectively. Then, the future fusion layer spliced feature vectors output of the two CNN models (See in line 104-107).
Point 2: The word “COOK” in lines 164-165 is advised to be explained more in detail to avoid confusion.
Response 2: Many thanks for the comment.
The authors have supplemented the full name of “ COOK “ and briefly introduced him.(See in line 176-177).
Point 3: The explanation of the “experimental steps” subsection, starting from the 274th line, should be from an instructive manner to a more descriptive manner in the past simple.
Response 3: Many thanks for the comment.
Many thanks for the comment on the writing.
The authors have turned the interpretation of the ' experimental steps ' subsection into a more descriptive way (See in line 294-310).
Point 4: Fig. 7 should be explained more in detail by focusing on the defect types and their influence on the velocity-time curve.
Response 4: Many thanks for the comment.
The authors have explained the performance of different pile foundation defects on the velocity-time curve in more detail (See in line 314-324).
Point 5: The reason for choosing layer 3 for denoising the waveform effect should be explained in detail, lines 343-346. According to the defined criteria in lines 330-334, the results of layers 1 and 2 also show a good match. For that reason, the reason behind choosing a 3-layer decomposition should be described in detail.
Response 5: Many thanks for the comment.
In the definition standard, in addition to calculating the root mean square error ( RMSE ) and signal-to-noise ratio ( SNR ), it is also necessary to observe the denoised waveform (See in line 357-359). The denoising waveforms of the first and second layers still have slight noise, while the waveform states of the fourth and fifth layers have been deformed. The authors have supplemented the details (See in line 366-371).
Point 6: Fig. 9 should be explained in detail to show the differences in the wavelet time-frequency diagram of 6 types of multi-defect piles.
Response 6: Many thanks for the comment.
The authors take defect type 1 in Figure 9 as an example to explain the specific meaning of the wavelet time-frequency diagram. It shows that the value and color of the frequency component can judge the difference and characteristics of six multi-defect piles (See in line 395-401).
Point 7: The misspelling of the word “fully connected layer” in Fig 10, line 411, should be corrected.
Response 7: Many thanks for the comment.
The authors have corrected this spelling error (See in line 440).
Author Response File: 
Author Response.docx
Reviewer 2 Report
This manuscript should be significantly revised, or show novelty, as it is similar to the following published article by the same authors. Some equations and figures are plagiarized.
Buildings 2022, 12(5), 664; https://0-doi-org.brum.beds.ac.uk/10.3390/buildings12050664
Unless it is significantly revised, this manuscript, or justified, is not fit for publication.
Author Response
Point 1: The This manuscript should be significantly revised, or show novelty, as it is similar to the following published article by the same authors. Some equations and figures are plagiarized.
Buildings 2022, 12(5), 664; https://0-doi-org.brum.beds.ac.uk/10.3390/buildings12050664
Unless it is significantly revised, this manuscript, or justified, is not fit for publication.
Response 1: Many thanks for the comment.
Since our team is working on combining low-strain integrity detection and deep learning technology, this paper is inevitably the same as our previous articles in some formulas commonly used in this field. However, this paper differs from that article in using the wavelet packet transform formula. In addition, the material parameters of the model will affect the simulation effect of the velocity response curve, including the amplitude difference, the apparent degree of pile bottom reflection, and the accuracy of pile length measurement (See in line 183-186). The length of the model pile used in this study ( 1.2m ) is only 0.2m different from the length of the model pile in the literature ( 1m ), so the parameters selected for finite element simulation and signal acquisition are not much different.
This manuscript differs from that in terms of the research object, secondary development method, data processing method, selection of deep learning technology, and wavelet packet transform formula. Moreover, the method used in this study performs better in the accuracy of defect type recognition. We will organize the differences with that paper in the following table :
|
|
Research object
|
The method of secondary development |
Data processing method |
The choice of deep learning technology |
Application of wavelet packet transform formula |
|
buildings12050664 |
Single-defect pile foundation |
Python loop control variables generate the finite element model of defective piles with different degrees. |
The one-dimensional time domain data is reconstructed into two-dimensional data by wavelet packet decomposition. |
2D-CNN |
For data reconstruction |
|
applsci-2227495(This manuscript) |
Multi-defect pile foundation |
Use the “macro manager” in ABAQUS to record, modify and reorganize the recorded script code.In addition, the grid division of a multi-defect pile foundation is more complicated than that of a single-defect pile foundation. This study proposes a new method of adaptive grid division of the multi-defect pile foundation model. |
Firstly, the time domain signal is denoised by a wavelet packet. Secondly, the 1D time-domain signals before and after denoising are transformed into a two-dimensional wavelet time-frequency diagram. |
TC-CNN |
For denoising |
In order to further distinguish the two papers, the authors explain the necessity and novelty of this study.
The necessity of research : Most research is about single-defect piles and rarely involves the type identification of multi-defect piles (See in line 36-37). When the pile foundation has multiple defects, the elastic wave will produce multiple reflections. It is much more challenging to identify the type of multi-defect pile only by velocity-time domain curve than by single-defect pile (See in line 51-54). The authors also added a necessity statement in the abstract (See in line 13-16).
The novelty of research :This paper proposes a new method to dynamically correlate the boundary conditions that control the size of defects with the cutting position of the mesh (See in line 257-260). Recently, other types of signal recognition have well used TC-CNN’s powerful and flexible feature extraction capability. However, the multi-defect identification of piles has not applied this feature extraction technology (See in line 102-104). The authors briefly describe the new method for identifying multi-defect types of concrete piles proposed in this paper (See in line 113-117).
Author Response File: Author Response.docx
Reviewer 3 Report
The authors have conducted a well and systematic study using experimental and analytical methods. The paper is well written with a novel research area may be useful for academic community and industry.
The authors can include some relevant applications in Introduction section
Give some explanation on the inputs are considered for ABAQUS model
Include some discussion on relevance and accuracy of ABAQUS results
How the proposed method is novel than the existing methods?
Tabulate the hyper parameters which are adopted for the CNN
How the ground truth is generated give detailed steps
Comment on speed
Limitation and scope for future studies can be included
Refer the following literature
Zhong, M., & Meng, K. (2023). Theoretical Analysis of Dynamic Response of Pipe Pile with Multi-Defects. Journal of Marine Science and Engineering, 11(1), 83.
Zhang, C., & Zhang, J. (2009). Application of artificial neural network for diagnosing pile integrity based on low strain dynamic testing. In Computational Structural Engineering: Proceedings of the International Symposium on Computational Structural Engineering, held in Shanghai, China, June 22–24, 2009 (pp. 857-862). Springer Netherlands.
Andrushia, D., Anand, N., & Arulraj, P. (2020). Anisotropic diffusion based denoising on concrete images and surface crack segmentation. International Journal of Structural Integrity, 11(3), 395-409.
Author Response
Point 1: The authors can include some relevant applications in Introduction section.
Response 1: Many thanks for the comment.
The authors added the application of non-destructive testing technology in the light rail construction site (See in line 44-46). Use deep learning technology to identify application cases of cast-in-place piles (See in line 55-57).
Point 2: Give some explanation on the inputs are considered for ABAQUS model.
Include some discussion on relevance and accuracy of ABAQUS results.
Response 2: Many thanks for the comment.
The material parameters of the input model will affect the simulation effect of the velocity response curve, including the amplitude difference, the apparent degree of the pile bottom reflection, and the accuracy of the pile length measurement (See in line 183-186). The selection of analysis step length is also discussed (See in line 199-200).
Point 3: How the proposed method is novel than the existing methods?
Response 3: Many thanks for the comment.
The novelty of research :This paper proposes a new method to dynamically correlate the boundary conditions that control the size of defects with the cutting position of the mesh (See in line 257-260). Recently, other types of signal recognition have well used TC-CNN’s powerful and flexible feature extraction capability. However, the multi-defect identification of piles has not applied this feature extraction technology (See in line 102-104).
Point 4: Comment on speed
Limitation and scope for future studies can be included
Refer the following literature
Zhong, M., & Meng, K. (2023). Theoretical Analysis of Dynamic Response of Pipe Pile with Multi-Defects. Journal of Marine Science and Engineering, 11(1), 83.
Zhang, C., & Zhang, J. (2009). Application of artificial neural network for diagnosing pile integrity based on low strain dynamic testing. In Computational Structural Engineering: Proceedings of the International Symposium on Computational Structural Engineering, held in Shanghai, China, June 22–24, 2009 (pp. 857-862). Springer Netherlands.
Andrushia, D., Anand, N., & Arulraj, P. (2020). Anisotropic diffusion based denoising on concrete images and surface crack segmentation. International Journal of Structural Integrity, 11(3), 395-409.
Response 4: Many thanks for the comment.
Regarding the comments on the velocity time domain curve and the research limitations, the authors have referred to the relevant literature and explained in detail. (See in line 314-324,65-67). In the part of future research space, the authors make further prospects in the conclusion part (See in line 507-512).
Author Response File: Author Response.docx
Reviewer 4 Report
The authors investigated a new method using neural network to detect damages and defects in concrete piles. The reviewer believes the paper is interesting and the topic can help future damage detection; however, there are a couple of comments that need to be addressed before it can be considered for publication. Please see my comments below:
1- The novelty and the importance of the study has not been clearly shown. Please provide more details to indicate why the research is necessary and worth investigation.
2- The paper mostly deals with damage detection in piles. Accordingly, the introduction of the paper needs to be strengthened by new articles on pile elements and NDT methods for damage detection. Please see below some of new papers suggested to be included in the paper:
For NDT and damage detection:
· "NDT methods for damage detection in steel bridges." In Health Monitoring of Structural and Biological Systems XVI, vol. 12048, pp. 385-394. SPIE, 2022.
3- Please provide the version of the software you have implemented.
4- The English language of the paper should be revised. There are several grammatical issues, such as mixing active and passive voice, spelling, and singular vs. plural, making it difficult for a reader to understand. A grammatical review is recommended to polish such issues.
Author Response
Point 1: The novelty and the importance of the study has not been clearly shown. Please provide more details to indicate why the research is necessary and worth investigation.
Response 1: Many thanks for the comment.
The necessity of research : Most research is about single-defect piles and rarely involves the type identification of multi-defect piles (See in line 36-37). When the pile foundation has multiple defects, the elastic wave will produce multiple reflections. It is much more challenging to identify the type of multi-defect pile only by velocity-time domain curve than by single-defect pile (See in line 51-54). The authors also added a necessity statement in the abstract (See in line 13-16).
The novelty of research :This paper proposes a new method to dynamically correlate the boundary conditions that control the size of defects with the cutting position of the mesh (See in line 257-260). Recently, other types of signal recognition have well used TC-CNN’s powerful and flexible feature extraction capability. However, the multi-defect identification of piles has not applied this feature extraction technology (See in line 102-104). The authors briefly describe the new method for identifying multi-defect types of concrete piles proposed in this paper (See in line 113-117).
Point 2: The paper mostly deals with damage detection in piles. Accordingly, the introduction of the paper needs to be strengthened by new articles on pile elements and NDT methods for damage detection. Please see below some of new papers suggested to be included in the paper:
For NDT and damage detection:
"NDT methods for damage detection in steel bridges." In Health Monitoring of Structural and Biological Systems XVI, vol. 12048, pp. 385-394. SPIE, 2022.
Response 2: Many thanks for the comment.
The authors have introduced literature and cases on non-destructive testing and non-destructive testing of pile foundations (See in line 532,39-42,536,44-46).
Point 3: Please provide the version of the software you have implemented.
Response 3: Many thanks for the comment.
The software version is ABAQUS 2020 (See in line 164).
Point 4: The English language of the paper should be revised. There are several grammatical issues, such as mixing active and passive voice, spelling, and singular vs. plural, making it difficult for a reader to understand. A grammatical review is recommended to polish such issues.
Response 4: Many thanks for the comment.
The authors have solved these problems through grammar review, please check. Thank you for your reminder.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The reviewer's comments were satisfactorily addressed by the authors. Thank you.
Reviewer 4 Report
The authors revised the paper significantly; hence, it can be accepted for publication.
