Automatic Crack Classification by Exploiting Statistical Event Descriptors for Deep Learning

Round 1

Reviewer 1 Report

The authors of this manuscript propose a method for classifying the modes of concrete damage under load by the acoustic emission data in the test structure. Deep neural networks with Bidirectional Long-Short Term Memory are used as a classifier, and the training sample is sets of specific statistics derived from time series of responses. The proposed model is analyzed from the point of view of the optimality of its parameters and classification accuracy. This study is of interest for the problems of standardizing SHM systems and embedding them into real engineering structures.

While the manuscript is potentially interesting, it is written with confusing describing, so there are some comments:

The first block of remarks concerns the description of formulas

1) The authors designate the maximum of the refined signal in different ways, for example, in lines 154 and 452, as well as in Figures 2 and 3. Uniform notation needed. Also, relative to Figure 3, it is not clear what the indexing of the intrinsic mode functions.

2) It should also be noted that the various notation of time series under study is used in blocks III.1 and III.2 of the framework for damage classification. Compare the spelling of the time series under study in lines 241, 263 and 326. The designation for the time series, used as the input of block III.1, is specified in line 154, so it makes sense to write the whole proposed algorithm in these designations.

3) Formulas (7) and (8) are identical in content, and there is no need to give both. Also, not all matrices and vectors in these formulas are described. In line 348, there is a reference to the nonexistent equation.

The second block of remarks relates directly to the implementation of the algorithm.

4) High-quality training of the model requires well-labelled data. However, the authors have not sufficiently covered this issue, so some questions arise. What was the numerical criterion for distribution by three classes of the different crack events? In how proportion were the 1500 acoustic emission events distributed among these classes?

5) Are all classes equally well defined for the test sample? Intraclass accuracy data is preferable.

6) It was stated that 15,000 digitized samples were collected, and 1,650 were used for testing purposes. That is, the test sample is only 11% instead of usual 20%. Does the quality of classification decline with increasing test sample size?

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Reviewer 2 Report

 

There are some weaknesses through the manuscript which need improvement. Therefore, the submitted manuscript cannot be accepted for publication in this form, but it has a chance of acceptance after a major revision. My comments and suggestions are as follows:

1- Abstract gives information on the main feature of the performed study, but some details about the considered cracks must be added.

2- Authors must clarify necessity of the performed research. Objectives of the study must be clearly mentioned in introduction.

3- The literature study must be enriched. In this respect, authors must read and refer to the following papers: (a) https://doi.org/10.1016/j.jclepro.2020.123074 (b) https://doi.org/10.1016/j.jmrt.2021.07.004 and other research works.

4- It would be nice, if authors could add some real figures to show conditions of experimental practice.  

5- The main reference of each formula must be cited. Moreover, each parameters in equations must be introduced. Please double check this issue.

6- Although some figures are illustrate in a high quality, there are figures must be presented in a high quality. Also, figures should not be too small, and legends must be legible.

7- Standard deviation is the presented curves must be discussed. In addition, error in calculation must be considered and discussed.

8- Details of error calculation must be presented.

9- In its language layer, the manuscript should be considered for English language editing. There are sentences which have to be rewritten.

10- The conclusion must be more than just a summary of the manuscript. List of references must be updated based on the proposed papers. Please provide all changes by red color in the revised version.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript is dealing with a supportive classification tool for SHM for cracks detected via acoustic emission. In Introduction, the authors firstly discussed the performance of an application of acoustic emission (AE) in SHM and further its combining with deep neural networks. The performed literature review provide good state-of-the-art for the investigated problem as well as strong justification of the topic of investigation. However, some minor extensions are recommended. Further, in section 2, the authors described in detail the experimental setup with well-justified parameters defined by the authors. Then, in section 3, the authors described the method of processing of the acquired signals, including characterization of AE events, description of three features used in the study, and the implementation of a deep learning approach. This section presents a very well described methodology supported with numerous references. In section 4, the authors presented the results of classification of AE events and a deep discussion on the obtained results.

The manuscript is original and provides meaningful insight into the investigated problem of classification of cracks in concrete structures. Before a publication, some minor revisions are recommended.

 

1) To strengthen the originality of this study, it is recommended to discuss other classification approaches of AE signals using various methods.

2) It is important to provide details on the tested concrete specimen, including type and manufacturer.

3) It is recommended to remove the heading of subsection 2.1, since there are no other subsections in section 2.

4) According to the worldwide accepted scientific style of writing, all variables need to be written with italic font, please make appropriate corrections.

5) Line 455: to avoid confusion, it is recommended to substitute sign “x” with a special character “times”.

6) Line 465: “to but to avoid …” – please correct.

7) In Figures 8 and 9, since the descriptors do not represent a common function, it is recommended not to connect the points on the plots, representing particular values.

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Round 2

Reviewer 2 Report

The paper has been improved and corresponding modifications have been conducted. In my opinion, the current version can be considered for publication.