A Metallic Fracture Estimation Method Using Digital Image Correlation

Round 1

Reviewer 1 Report

The research presents fracture estimation using a machine learning algorithm in a DIC setup. Following recommendations are made to the authors:

1.  Has the crack length been measured using other methods for comparison with the proposed algorithm to ascertain the error? Please include any relevant data. It says that the estimation error is 7.7um, what method was used to measure the actual crack length? 

2. It is a bit unclear whether the SDGs used to train the model are from real experiments or simulated images as presented in figure 2. 

3.  How do the results compare to other relevant works? You may want to include papers like  10.3390/s21175894 to aid your discussion.  

4. There are a few grammar/sentence structure errors that can be removed after a proofread.  

Author Response

Thank you for your helpful comments. The responses are as follows:

1. Has the crack length been measured using other methods for comparison with the proposed algorithm to ascertain the error? Please include any relevant data. It says that the estimation error is 7.7um, what method was used to measure the actual crack length? 

A comparison between our work and Yuan’s work (10.3390/s21175894) has been added in line 410-426. Please see the response to comment 3.

The ground-truth crack length values are obtained by manual measurement from the source images acquired by the DIC camera. We count the pixels of the crack and make the conversion by the pixel-length rate. It is introduced in line 308-311.

 

2. It is a bit unclear whether the SDGs used to train the model are from real experiments or simulated images as presented in figure 2. 

We train the model by the SDGs from our fatigue experiment. The simulation is only an auxiliary proof showing that the SDG varies with the crack propagation. It now has been clarified in line 51-52.

 

3. How do the results compare to other relevant works? You may want to include papers like 10.3390/s21175894 to aid your discussion.  

Thank you for your recommendation. We compare Yuan’s result (10.3390/s21175894) with ours, and conclude that our method achieves lower maximum absolute error and similar maximum relative error. In addition, Yuan’s test only inspected 13 test examples, while we test more than 4000 examples including cracks propagating in different directions, so we believe our result can be more convincing. See line 410-426

 

4. There are a few grammar/sentence structure errors that can be removed after a proofread.  

Thank you very much for the kind suggestion. Multiple errors have been corrected and marked by red.

Reviewer 2 Report

This paper discusses a method to estimate metallic fractures. The paper is well-organized. However, what is the exact novelty of the proposed method? Please clarify it. 

There are also the following moments to be considered:

1. Please avoid abbreviations in the abstract. The text must be brief and specific (without extra information) to deliver the main idea of the manuscript.

2. In the introduction, please provide a sufficient literature review by citing relevant and recent research works.  

3. Please proofread the text again. Grammar and readability must be checked.

Author Response

Thank you for your helpful comments. The responses are as follows:

This paper discusses a method to estimate metallic fractures. The paper is well-organized. However, what is the exact novelty of the proposed method? Please clarify it. 

The novelty of the paper is to propose a noncontact, nondestructive, real-time approach that automatically detect the occurrence and estimate the length of cracks. The method can detect a crack propagating in any direction at its initiation, and estimate the length precisely. The novelties and advantages of the proposed method is discussed in line 133-145.

 

1. Please avoid abbreviations in the abstract. The text must be brief and specific (without extra information) to deliver the main idea of the manuscript.

Abbreviations are removed from the abstract, and the text is updated.

 

2. In the introduction, please provide a sufficient literature review by citing relevant and recent research works.  

We have now added description and discussion on some latest research works in 2. Related Work, the purple text.

 

3. Please proofread the text again. Grammar and readability must be checked.

Thank you very much for the kind suggestion. Multiple errors have been corrected and marked by red.

Reviewer 3 Report

In the manuscript titled "A Metallic Fracture Estimation Method using Digital Image Correlation", the authors use digital image correlation (DIC) to estimate fractures in metallic components by convolutional neural networks (CNN). They used a camera with a resolution reaching 0.005 mm/pixel and considered Griffith’s model for the calculation of strain energy when becomes non-uniform near the crack.

however, the paper needs to be revised with minor corrections before being accepted. The authors should take into account the following points:

3.1. DIC-based Fracture Estimation: Figure 2: The numbers in the figure are very small and it is very difficult to read.

This problem occurs too in figures 3, 4, and 10.

6. Conclusions and Further Discussions: The conclusions must be improved. The conclusions are unclear and do not demonstrate the most relevant results shown in the article

 

Author Response

Thank you for your helpful comments. The responses are as follows:

3.1. DIC-based Fracture Estimation: Figure 2: The numbers in the figure are very small and it is very difficult to read.

This problem occurs too in figures 3, 4, and 10.

The text in the figures is now enlarged.

 

6. Conclusions and Further Discussions: The conclusions must be improved. The conclusions are unclear and do not demonstrate the most relevant results shown in the article

This section now has been modified. The work and result are described in detail in this section, as marked by blue.

Reviewer 4 Report

Based on machine deep learning, the author uses digital image correlation (DIC) to estimate the cracks in metal parts through convolutional neural network (CNN), which is a novel metal fracture detection method. The results show the precision as well as the practicality of the proposed method. There are some suggestions listed below.

1. In the part of literature review, it is necessary to supplement the advantages of the combination of DIC technology and CNN technology, and briefly describe its practical application to prove that the research method in this paper is sufficient and feasible.

2. Material, temperature, shape and other factors will have an important impact on the fracture of metal. Whether these factors will have an impact on the test results can be supplemented with relevant instructions.

3. The spot on the sample is an important measurement variable. The camera uses the spot distribution to identify and detect, so it is necessary to briefly explain the making of the spot.

4. Figure 11 in the text lacks corresponding explanation and annotation, which needs to be briefly supplemented.

5. During the corresponding process of strain distribution diagram and DIC diagram, a certain correlation is required. Please specify the method to establish an effective correlation between them.

Author Response

Thank you for your helpful comments. The responses are as follows:

1. In the part of literature review, it is necessary to supplement the advantages of the combination of DIC technology and CNN technology, and briefly describe its practical application to prove that the research method in this paper is sufficient and feasible.

Thank you for the kind suggestion. The description is now added at the end of 2. Related Work, line 133-145.

 

2. Material, temperature, shape and other factors will have an important impact on the fracture of metal. Whether these factors will have an impact on the test results can be supplemented with relevant instructions.

Those factors do affect the fracture. The strain distribution is the comprehensive reflection for the factors of fatigue. Therefore, the trained model is suitable for the fracture estimation in a certain situation. Our fatigue test is performed under the condition defined by the ASTM standards to prove the feasibility. For a specific application, the test should be performed under the corresponding condition and train the corresponding CNN model.

The discussion is now added, in line 54-61.

 

3. The spot on the sample is an important measurement variable. The camera uses the spot distribution to identify and detect, so it is necessary to briefly explain the making of the spot.

The description on how we made the speckled specimens are now in line 250-259. In fact, the form of speckles is strictly required by the DIC technique. It can resolve the surface with heavy dust, blots or scratches.

 

4. Figure 11 in the text lacks corresponding explanation and annotation, which needs to be briefly supplemented.

Explanation for Figure 11 is now added, in line 369-373.

 

5. During the corresponding process of strain distribution diagram and DIC diagram, a certain correlation is required. Please specify the method to establish an effective correlation between them.

The processing from DIC images displayed in Figure 10 to strain distribution graphs displayed in Figure 11 is explained in line 364-368. The color regions in the DIC images are an illustration of different strain values. We crop the key area, linearly normalize the strain values into [0, 1] and rotate the graphs. Figure 11 is the greyscale illustration which is what we feed into the ResNet model.

Round 2

Reviewer 2 Report

Dear authors,

Thank you for your contribution. I do not have anymore comments.

Regards