Weld Defect Detection of a CMT Arc-Welded Aluminum Alloy Sheet Based on Arc Sound Signal Processing

Round 1

Reviewer 1 Report

1. Is the sound sensor absorbs other noise during welding other than the welding arc sounds? How can it differentiate the sounds?

2. What is the maximum distance between the sensor and torch? Is it sustainable to resist high temperatures to deliver reliable data?

3. Figure 4 should be improved by proving clear image with high resolution to identify the weld seam.

4. Is this process is spatter free process? Why there is no evidence of spatter formation and related sound peaks?

5. In CMT there are many hidden conditions such as arc length correction, dynamic correction, etc… based on the process stability, these conditions needs to be adjusted. Is the sound signals can able to detect the variation in sound caused by changes in above mentioned conditions?

6. Figure 11, is there any units for and Y axes?

7. The scientific quality of the manuscript must be improved.

8. CMT applications and defects detection methods needs to be included as suggested in the introduction.

https://doi.org/10.3390/app12094568 

https://doi.org/10.1007/s12666-022-02715-1

 

https://doi.org/10.3390/app13074232

 

Author Response

 Concern # 1: Is the sound sensor absorbs other noise during welding other than the welding arc sounds? How can it differentiate the sounds?

Author response:  Thank you for your comments. During the welding process, the sound sensor absorbs the arc sound as well as other noise. The sound sensor is located directly in front of the welding direction and moves synchronously with the welding torch to maximize the absorption of arc sound. The welding experiments in this paper were done in a laboratory with minimal ambient noise, which we consider negligible.

Concern # 2: What is the maximum distance between the sensor and torch? Is it sustainable to resist high temperatures to deliver reliable data?

Author response:  Thank you for your comments. The sound sensor is 15cm away from the torch and moves in sync with the torch. During the CMT welding process, the heat input is small and the thermal radiation from the molten pool and arc has negligible effect on the sound sensor.

Author action: In order to clearly show the position of the sensor in relation to the torch, we changed Figure 2 to a diagram of the welding experimental setup. In Figure 2, the position of the sound sensor in relation to the welding torch and the distance between them are clearly shown.

Concern # 3: Figure 4 should be improved by proving clear image with high resolution to identify the weld seam.

Author response: Thank you for your comments. We have revised the manuscript according to your comments.

Author action: We have revised Figure 4. We have provided clear pictures for each of the three different weld states.

Concern # 4: Is this process is spatter free process? Why there is no evidence of spatter formation and related sound peaks?

Author response: Thank you for your comments. CMT is spatter free process. In the short-circuiting phase of CMT process, the arc voltage changes into zero, arc energy is zero, arc extinguished. At the same time, the wire moves in reverse to help the molten droplet fracture transition to the molten pool, thus avoiding spatter.

Concern # 5: In CMT there are many hidden conditions such as arc length correction, dynamic correction, etc… based on the process stability, these conditions needs to be adjusted. Is the sound signals can able to detect the variation in sound caused by changes in above mentioned conditions?

Author response:  Thank you for your comments. CMT achieves precise control of the arc length by detecting the wire retraction length, making the arc more stable. And when the dry elongation of the CMT wire changes, it only changes the wire feed speed and does not lead to a change in welding current. The welding experiments in this paper were performed using robotic welding with a stable welding path and therefore a stable dry elongation of the wire.

We will investigate the relationship between CMT dry elongation variation and arc sound signal in our subsequent work.

Concern # 6: Figure 11, is there any units for and Y axes?

Author response: Thank you for your comments. We did 10 training tests for each classification method, and Figure 11 compares the accuracy of each test for different classification methods. In Figure 11, the x-axis is the training test number and the y-axis is the test accuracy; they have no units. However, the presentation in Figure 11 is not clear, and we have modified it.

Author action:  We have revised the relevant descriptions in the second paragraphs of section 4.2. In addition, to avoid ambiguity we have revised the description of the coordinate axes in Figure 11.

Concern # 7: The scientific quality of the manuscript must be improved.

Author response:  Thank you for your comments. We carefully checked the manuscript and have revised the introduction of the manuscript according to your comments.

Author action:  We have revised the manuscript to improve the scientific quality of the manuscript. We have combined the second and third paragraphs of the introduction into a second paragraph and added a third paragraph. The novelty of the proposed method in this paper is described in the third paragraph of the introduction.

Concern # 8: CMT applications and defects detection methods needs to be included as suggested in the introduction.

Author response:  Thank you for your comments. We have studied the CMT application and defect identification methods again in depth and have carefully studied the relevant literature. We have revised the introduction of the manuscript according to your comments.

Author action:  We highlight relevant and important research in the introduction. And we carefully studied the relevant references you provided and made reasonable citations. We have added references relevant to our paper in the first paragraph of the introduction. We have added a third paragraph to the introduction and cited relevant literature.

Reviewer 2 Report

Comments to Authors

            This study focuses on weld defect detection of CMT arc welded aluminum alloy sheets based on arc sound signal processing.

            The work was well planned and executed however there is a lack of presentations. The kinds of literature reviewed in this manuscript and in-depth knowledge of the field is to be required.

            The following clarification must be made to improvise the readability of the manuscript.

 1. The abstract of the article is not clear and concise.

2. In the introduction section add recent literature published.

3. Mention the novelty of the adopted review research in the introduction section.

4. English of the manuscript must be polished throughout the manuscript. Carefully correct the typographical mistakes in the entire manuscript.

5. The introduction section is minimal, so completely revised as per the journal requirement.

6. The materials and methods section will be completely revised per the journal requirement.

7. High-quality Figures to be provided for better readability with proper legend and labels.

8. To be provided with the detailed Al alloy sheet analysis in the summary in the text.

9.  Results and discussion must be supported by standard literature.

10.  The conclusion is needed to write more precisely with the application of this existing methodology to the proposed model.

 

The article may be acceptable for publication after clarifying the minor revision.

Author Response

Concern # 1: The abstract of the article is not clear and concise.

Author response:  Thank you for your comments. We have revised the abstract of the manuscript according to your comments.

Author action:  We have revised the abstract of the manuscript according to your comments.

Concern # 2: In the introduction section add recent literature published.

Author response: Thank you for your comments. We have carefully researched and analyzed the latest literature related to the study and made reasonable citations.

Author action: We have revised manuscript according to your comments. We added four references (9, 22, 23, 24) and justified the citations in the introductory section.

Concern # 3: Mention the novelty of the adopted review research in the introduction section.

Author response:  Thank you for your comments. We carefully study the processing method of arc sound signal and the method of defect identification, and describe the novelty of the method proposed in this paper. We have revised the introduction of the manuscript according to your comments.

Author action: We added a third paragraph to the introduction to show the novelty of the proposed method in this paper.

Concern # 4: English of the manuscript must be polished throughout the manuscript. Carefully correct the typographical mistakes in the entire manuscript.

Author response:  Thank you for your comments. We polished the throughout manuscript and carefully corrected the typographical mistakes in the entire manuscript.

Author action: We have revised manuscript according to your comments.

 

 Concern # 5: The introduction section is minimal, so completely revised as per the journal requirement.

Author response:  Thank you for your comments. We have revised the introduction of the manuscript according to your comments.

Author action: We have revised the introduction of the manuscript according to your comments. We have added a third paragraph in the introduction to describe the novelty of combining STFT with neural networks for CMT welding defect identification in thin aluminum alloy plates. And the relevant literature is reasonably cited. We also corrected statements in the introduction that were not clearly described.

 

 Concern # 6: The materials and methods section will be completely revised per the journal requirement.

Author response:  Thank you for your comments. We carefully checked the materials and methods section of the manuscript and revised the welding method handling procedures section to comply with the requirements of the journal.

Author action: We have revised the manuscript according to your comments. We have revised the soldering method handling procedure in section 2.1 of the manuscript.

 

Concern # 7: High-quality Figures to be provided for better readability with proper legend and labels.

Author response:  Thank you for your comments. We have revised the manuscript according to your comments.

Author action: We provide high-quality figures at 300dpi. We provide high quality graphics at 300 dpi with revised labels for Figure 3, Figure 5, and Figure 6.

 

Concern # 8: To be provided with the detailed Al alloy sheet analysis in the summary in the text.

Author response:  Thank you for your comments. Aluminum alloy sheets, due to the high thermal conductivity and low stiffness of the material, the traditional method of welding is prone to defects such as weld collapse, burn-through. CMT welding process can effectively reduce the probability of weld through occurrence for thin aluminum alloy plate welding. However, in actual production assembly errors lead to problems such as large or misaligned weld gaps, resulting in serious welding defects such as metal pool leakage and unfused welds, especially for thin aluminum alloy plates.

Author action:  We have revised the introduction of manuscript according to your comments. We have modified the first paragraph of the introduction to provide additional clarification on the analysis of aluminum alloys. We revised the third paragraph of section 3.2.1 to provide a summary analysis of the generation of welding defects in thin aluminum alloys.

 

 Concern # 9: Results and discussion must be supported by standard literature.

Author response:  Thank you for your comments. At present, there is less literature on the arc sound time-frequency analysis of CMT for defect identification. The welding process based on welding arc sound for defect identification is mainly focused on the GTAW welding process.

The GTAW arc sound signal consists of a pulse phase and a random variable phase. The frequency amplitude of the pulse phase is more active and is mainly concentrated in the low frequency part. However, However, the CMT arc sound signal correlates with the change in energy during the CMT cycle, with three peaks in the arc sound signal occurring during a CMT cycle.

The frequency band of CMT arc sound is wide and contains complex frequency components. This also illustrate the complexity of the CMT welding process.

Author action: We have revised the section 3.2.3 of manuscript according to your comments. We added a second paragraph where the CMT arc sound signal was compared with the GTAW arc sound signal.

 

Concern # 10: The conclusion is needed to write more precisely with the application of this existing methodology to the proposed model.

Author response:  Thank you for your comments. We have revised the conclusions of the manuscript.

Author action:  We added a conclusion and modified two existing conclusions.

Reviewer 3 Report

Artificial intelligence systems based on artificial neural networks show certain success in the analysis and classification of large data sets and allow not only to automate the analysis processes, but also to produce it in order to identify certain patterns of change in various features. The paper analyzes the mechanisms of generation of the sound of the SMT arc, as well as the relationship between the frequency-time spectrum of the sound signal of the arc and the quality of welding. The results of the experiment show the high accuracy of the proposed model. The research results are useful for science and practice, but there are several questions:

1. It would be useful to describe in more detail the principles of constructing a system for technical diagnostics of welding defects, including the selection of information frequencies and assessment of their significance, formation of a vector of diagnostic features.

2. In the spectral analysis of the time-frequency spectrum of arc sound, representing a mixture of both periodic and stationary random components, a significant amount of information contained in the characteristics of each of the components is lost. How did you choose the components that carry the maximum amount of information?

3. It is necessary to describe new features and feature spaces, determine their statistical and dynamic properties, methods for their transformation.

4. For which weld defect detection of CMT arc aluminum alloy sheet has this system been used? After all, for different types of defects, the accuracy of the system will be different.

5. How were sets of significant features formed from a set of incoming features, and what were the methods for ranking significant diagnostic features?

Author Response

Concern # 1: It would be useful to describe in more detail the principles of constructing a system for technical diagnostics of welding defects, including the selection of information frequencies and assessment of their significance, formation of a vector of diagnostic features.

Author response:  Thank you for your comments. In the welding process, welding current and voltage are the main factors affecting heat input, and their degree of adaptation to other parameters, such as welding speed, largely affects the quality of the weld. Due to assembly errors can make the welding arc unstable, which in turn produces defects such as burn-through, unfusion, etc., and is reflected in the arc sound. Therefore, a multi-information acquisition subsystem was constructed for current, voltage and arc sound. The frequency of the acquired signal is 32 kHz.

Author action: We have revised the Section 3.1 of the manuscript.

 

Concern # 2: In the spectral analysis of the time-frequency spectrum of arc sound, representing a mixture of both periodic and stationary random components, a significant amount of information contained in the characteristics of each of the components is lost. How did you choose the components that carry the maximum amount of information?

Author response:  Thank you for your comments. We analyzed the time-frequency mapping of the arc sound signal in section 3.2.3 and compared the time-frequency mapping of the arc sound for three different weld states. The time-frequency patterns of the three different weld states are clearly different. The time-frequency mapping of the arc sound can be used as an input for the identification of welding defects method.

We take the multi-channel time-frequency mapping as the input of the convolutional neural network, automatically identify the features of the input data by the convolutional layer, and finally input the feature vector automatically identified by the convolutional layer into the fully connected network for defect identification.

 

Concern # 3: It is necessary to describe new features and feature spaces, determine their statistical and dynamic properties, methods for their transformation.

Author response:  Thank you for your comments. We have revised the Section 4.2 of the manuscript.

Author action:  We have revised the first paragraph of Section 4.2 and added Figure 11 to describe the construction process of the statistical feature vectors according to your comments.

 

Concern # 4: For which weld defect detection of CMT arc aluminum alloy sheet has this system been used? After all, for different types of defects, the accuracy of the system will be different.

Author response:  Thank you for your comments. Welding is a complex process, due to a number of random and uncontrollable factors will lead to the occurrence of welding defects. In this paper, welding defects caused by assembly errors during CMT welding of thin aluminum alloy plates are identified. Three butt joint forms were designed and experimental results were obtained for two types of welding defects: melt pool leakage and unfused welds. We identify only these two types of welding defects.

 

Concern # 5: How were sets of significant features formed from a set of incoming features, and what were the methods for ranking significant diagnostic features?

Author response:  Thank you for your comments. We use convolutional neural networks for feature extraction of input data and weld defect identification. The output of each layer is the input of the next layer. Local features of the input data are extracted by convolutional operations in the convolutional layer, and the number of features is reduced by pooling layers, thus reducing the computational effort.

 

Reviewer 4 Report

Dear Authors,

Please make a clear explanation about how the chemical composition of the aluminum alloy could influence the detection of the defects inside of the bulk materials.

Author Response

Concern # 1: Please make a clear explanation about how the chemical composition of the aluminum alloy could influence the detection of the defects inside of the bulk materials.

Author response:  Thank you for your comments. The chemical composition of the base material determines which wire is selected to match it in order to achieve the required welding quality. The composition of the welding material, i.e. the wire, used for different aluminum alloys is also different, and improper selection of the welding material will produce welding defects. In this paper, the base material is 6061 aluminum alloy and the welding material is ER5356 wire, which is reasonable in terms of welding quality requirements. Under normal circumstances, no defect will be produced as a result. However, when the assembly gap is too large or the misalignment exists, a defect will also be generated, but the defect is generated by the latter and not by the chemical composition of the base material.

Author action:  We indicate the type of welding wire in section 3.2.1.

 

Round 2

Reviewer 1 Report

Revised version o the manuscript is improved and is good.