Optimization of the Weld Pool Boundary Calculated by the LSTM-Based Measurement Method in GTAW

Round 1

Reviewer 1 Report

1. Is the weld pool repeated the same when it was repeated because there are several reasons to form the asymmetric bead. Moreover, 4 parts are not enough to obtain accurate results. It is better to increase divisions.

2. The weld pool length varies by welding current and speed, is the developed model can detect variable molten pools accurately?

3. During welding smoke and spatter would be generated, and how the LSTM could can avoid such obstacles to find the accurate pool dimensions.

4.   Missing points and more points, using this phenomenon is it possible to estimate the boundaries if the points were missed.

5. What are minimum and maximum points are required to confirm the accuracy of the measurement?

6. Is there any validation tests were conducted after considering the missing points to measure the pool boundaries.

7. Conclusions needs to be improved.

 

8. please provide the still images of the practical results for the Missing points and more points.

Author Response

Dear Reviewers:

Thank you for evaluation and comments on our manuscript entitled “Optimization of the weld pool boundary calculated by the LSTM-based measurement method in GTAW” (manuscript number: metals-1824871). We have tried our best to provide the best possible answers and corrections and wish this can be well received by the editor and the reviewers. The revised manuscript with the correction sections is attached. The typographic mistakes were also corrected in the revised manuscript.

 

Best regards

Corresponding author: Shaojie Wu

First author: Yingchao Feng

2022.07.27

 

The point-by-point responses are as follow

 

Respond to Reviewer #1

We would like to say thank you to Reviewer #1 for the constructive criticism and suggestion. We addressed all the points raised by Reviewer #1, as summarized as below.

 

Comment #1

Is the weld pool repeated the same when it was repeated because there are several reasons to form the asymmetric bead? Moreover, 4 parts are not enough to obtain accurate results. It is better to increase divisions.

 

Respond to Comment #1:

Thank you for your reasonable question. This paper is a subsequent work based on our previous work which a novel LSTM-based measurement method of 3D weld pool surface in GTAW based on the structured light sensing system is developed. This measurement method introduces LSTM neural network to achieve the mapping calculation from the 2D imaging points on the imaging plane to the 3D reflection points on the weld pool surface, the scheme of the LSTM-based method is shown in Fig. 1. Hence, the simulation samples which calculated through the twin simulation model of the structured light sensing system depend the range and precision of the obtained LSTM-based model. In other words, the LSTM-based model can handle the asymmetric weld pool as the simulation model considers the according situation no matter what reason occurred. However, the simulation model in this paper don’t consider the asymmetric weld pool because the weld pool is extremely stable as the torch is not moving. Therefore, we believe 4 parts is enough to obtain accurate results while more divisions will add in the subsequent research of the moving torch.

 

Fig. 1 Scheme of the LSTM-based measurement method of 3D weld pool surface

 

Comment #2

The weld pool length varies by welding current and speed, is the developed model can detect variable molten pools accurately?

 

Respond to Comment #2:

Thanks for your question. We believe the proposed novel LSTM-based measurement method of 3D weld pool surface in GTAW can handle the variation of the weld pool length caused by welding current and speed. As we responded in comment #1, the range and precision of the obtained LSTM-based model basically depended by the simulation samples which calculated through the twin simulation model of the structured light sensing system. If the simulation model simulate a variety of weld pool surfaces, then the obtained LSTM-based model can reconstruct an accurate weld pool surface.

Fig. 2 describes a simulation model of a moving weld pool with a compressed surface. The simulation model also used to create a LSTM-based model suit for the moving GTAW weld pool with a welding current above 100A. An ideal consistency between the actual weld width and the calculated one is achieved even as the welding current is changed during a weld, as shown in Fig. 3.

 

Fig. 2 Geometric model of the weld pool

 

Fig. 3 The calculated weld pool surfaces and their cross sections

 

Comment #3

During welding smoke and spatter would be generated, and how the LSTM could can avoid such obstacles to find the accurate pool dimensions?

 

Respond to Comment #3:

Thanks for your constructive suggestion. Because the structured light sensing system is based on the mirror-like weld pool surface, in other words, the weld pool is better to be stable during the welding. Therefore, we only focused on the GTAW process which barely generated welding smoke and spatter right now. We will study the flexibility of the LSTM neural network on the situation of the reflected pattern on the imaging plane is really disordered in the future.

 

Comment #4

Missing points and more points, using this phenomenon is it possible to estimate the boundaries if the points were missed?

 

Respond to Comment #4:

Thanks for your question, the boundaries can be estimated in the situations of missing points and more points. As shown in Fig. 4 (as Fig. 13 in the manuscript), the Fig. 4(a) and the Fig. 4(c) represented scattered missing 15% points and concentrated missing 15 points respectively. The calculated weld pool surfaces and their boundaries are shown in Fig. 4(b) and Fig. 4(d) by using the LSTM-based model proposed in our previous work (Li, L.D., Cheng, F.J., Wu, S.J.: An LSTM-based measurement method of 3D weld pool surface in GTAW. Measurement. 171, 108809 (2021)) and the optimization method proposed in this manuscript (metals-1824871). The results have an acceptable robustness as described in Table 1 (as Table 7 in the manuscript). It can be observed that the  is 0.54% and 0.66% for the scattered missing 15% points and concentrated missing 15 points, which has no difference with the ideal result (0.66%) in Table 1. However, the  is 7.55% and 6.86% for the scattered missing 15% points and concentrated missing 15 points, which is obviously increased according to the ideal result (2.52%). This is because the LSTM-based model has been well trained to reconstruct the curved weld pool surface when the reflection points are enough, the  will increase when the reflection points are lacked. A forward-reverse united reconstruction optimization method is also proposed in the manuscript to solve this problem.

 

Fig. 4 The Samples with the phenomena of missing points and more points and their reconstructed surfaces. (a) The reflection image of scattered missing points (15% lost points, 12 points lost in total); (b) The reconstructed surface of Figure. 13(a); (c) The reflection image of concentrated missing points (15 points lost in focus); (d) The reconstructed surface of Figure. 13(c); (e) The reflection image of random adding points (add 15 points totally); (f) The reconstructed surface of Figure. 13(e).

Table 1. The detailed reconstructed surface characteristics of the three examples.

Scattered missing 15% points	0.3302	0.54	26.81	7.55	8.15	6.86	8.45	3.43
Concentrated missing 15 points	0.3342	0.66	27.01	6.86	8.83	0.91	8.46	3.31
Random adding 15 points	0.3518	5.96	28.82	0.62	9.41	7.54	9.47	8.23

 

Comment #5

What are minimum and maximum points are required to confirm the accuracy of the measurement?

 

Respond to Comment #5:

Thanks for your question. Most researchers around the world believed that the in-situ GTAW weld pool surface contains a lot of information about the underneath penetration of the weld pool because the skilled welders can adjust the parameters just by observing the weld pool surface. Hence, the structured light sensing system is proposed to measure the shape of weld pool because the weld pool surface is a mirror-like surface. The laser generator projects structured light pattern into the weld pool and the reflected pattern will contain the information about the weld pool surface. Therefore, the accuracy of the reconstructed weld pool is increased with the number of the reflected points. However, the GTAW weld pool normally small than 15mm in length and 10 mm in width. In most cases, the laser generator can only project about less than 100 points even into the largest weld pool. According to our research, the accuracy will obviously decrease if more than 25% points are missing.

 

Comment #6

Is there any validation tests were conducted after considering the missing points to measure the pool boundaries?

 

Respond to Comment #6:

Thanks for your suggestion. Fig. 5 and Fig. 6 (Fig. 8 and Fig. 9 in the manuscript) are the reflection images and their image processing results taken continuously in the 70A current test (the frame rate of the camera is 26Hz, and the first frame corresponds to the time of 15.46s after arcing). It can be seen that all these photos has the situation such as missing points and more points. The samples in Fig. 7 (Fig.13 in the manuscript) are used for testing the robustness of the LSTM-based model and also the optimization method just proposed in section 3. We sorry that the verification of the proposed optimization method is not well announced in the manuscript. The revised manuscript will emphasize it.

 

Figure 5 The phenomena of “missing points”. (a) A reflection image taken in the 70 current test; (b) The image processing result of Figure. 5(a).

 

Figure 6 Missing points caused by the formation of floating oxides on the surface of the molten pool. (a) A set of reflection images taken in the 70A current test (the camera frame rate is 26Hz, the first frame corresponds to 15.46s after arcing); (b) Corresponding image processing results of the three reflection images in Figure. 6(a).

 

Figure 7. The Samples with the phenomena of missing points and more points and their reconstructed surfaces. (a) The reflection image of scattered missing points (15% lost points, 12 points lost in total); (b) The reconstructed surface of Figure. 13(a); (c) The reflection image of concentrated missing points (15 points lost in focus); (d) The reconstructed surface of Figure. 13(c); (e) The reflection image of random adding points (add 15 points totally); (f) The reconstructed surface of Figure. 13(e).

 

The revised content in Page 10 Line 261:

Although the proposed LSTM-based measurement method avoids the complicated procedure of point recognition, it is of great significance to analyze the influence of missing points and more points on the LSTM-based model after boudary extension by the proposed optimization method.

The revised content in Page 11 Line 266:

A reflection image reflected by a simulation surface (created through MATLAB) which contained 140 reflection points is taken as an example, the reflection image is shown in Figure. 12(a) and the reconstructed surface (after boundary extension by the proposed optimization method) of the sample without missing points and more points is shown in Fig. 12(b). The simulated surface height , radius of curvature , weld width  in the X direction and weld width  in the Y direction are shown in Table 5.

Comment #7

Conclusions needs to be improved.

 

Respond to Comment #7:

Thanks for your suggestion. We have rewritten the conclusion as you request.

 

The revised content in Page 16 Line 385:

In this paper, a boundary extension method is designed first to optimize the boundary profile of the interpolated weld pool surface calculated by a LSTM-based weld pool measurement method specialized for the structured light sensing system. Then the phenomenons of missing points and more points which are observed in physical welding are analyzed. The robustness of the LSTM-based method to the phenomenons is also studied To solve the problem caused by the missing/adding the first few imaging points, a forward-reverse united reconstruction optimization method is designed at last. The followings are concluded.

(1) A boundary extension method is proposed to realize the boundary optimization of the weld pool effectively. After boundary extension, the errors of the left and right part of the boundary of the weld pool are obviously reduced.

(2) The LSTM-based method has good robustness in the cases of missing scattered points (no more than 15%), missing concentrated points (no more than 10%) and more points randomly (no more than 10%), but shows certain inaccuracy in the phenomenon of missing/adding first few imaging points.

(3) The designed forward-reverse united reconstruction optimization method effectively overcomes the inaccurate prediction of the missing/adding first few imaging points. After optimization by this method, the boundary error of the reconstructed weld pool surface is obviously reduced. The weld width error is reduced to 1.62% in the X direction and 3.94% in the Y direction.

 

 

Comment #8

please provide the still images of the practical results for the Missing points and more points.

 

Respond to Comment #8:

Thanks for your suggestion. The Fig. 8, Fig. 9 and Fig. 10 (Fig. 9, Fig. 10 and Fig. 11 in the manuscript) are both the images of the practical results for the missing points and more points.  

 

 

 

Figure 8. Missing points caused by the formation of floating oxides on the surface of the molten pool. (a) A set of reflection images taken in the 70A current test (the camera frame rate is 26Hz, the first frame corresponds to 15.46s after arcing); (b) Corresponding image processing results of the three reflection images in Figure. 9(a).

   

Figure 9. More points phenomenon caused by the presence of bright noise. (a) A reflection image taken early in the formation of the molten pool; (b) The image processing result of Figure. 10(a).

   

 

Figure. 10. More points phenomenon caused by splitting of imaging points. (a) A reflection image taken in a 70A current test; (b) The image processing result of Figure. 11(a).

Author Response File: Author Response.pdf

Reviewer 2 Report

- It is necessary to indicate which is the base metal.

-It is not mentioned if filler metal is used (please, indicate which one) or not. 

-The values of the welding variables used (welding current, voltage, welding speed, and gas flow rate) are not specified.

-Caption of Fig. 17 is incomplete.

-Fig. 18 is missing and is not mentioned in the text.

Author Response

Dear Reviewers:

Thank you for evaluation and comments on our manuscript entitled “Optimization of the weld pool boundary calculated by the LSTM-based measurement method in GTAW” (manuscript number: metals-1824871). We have tried our best to provide the best possible answers and corrections and wish this can be well received by the editor and the reviewers. The revised manuscript with the correction sections is attached. The typographic mistakes were also corrected in the revised manuscript.

 

Best regards

Corresponding author: Shaojie Wu

First author: Yingchao Feng

2022.07.27

 

The point-by-point responses are as follow

 

Respond to Reviewer #2

We would like to say thank you to Reviewer #1 for the constructive criticism and suggestion. We addressed all the points raised by Reviewer #2, as summarized as below.

 

Comment #1

It is necessary to indicate which is the base metal.

It is not mentioned if filler metal is used (please, indicate which one) or not. 

The values of the welding variables used (welding current, voltage, welding speed, and gas flow rate) are not specified.

 

Respond to Comment #1:

Thanks for your suggestion. The material and the welding parameters all add in the section 2.

The revised content in Page 2 Line 79:

Figure. 1 is the experimental system used in this paper. A 100 mW laser with a wavelength of 660 nm is selected to project a pattern of 17×17 dot-matrix rays. The GTAW torch keeps stationary to easily obtain a complete weld pool surface such that the accuracy and the robustness of the LSTM-based measurement method could be tested by comparing with the reconstructed weld pool. The 304L stainless steel is chosen as the base metal. Because the GTAW torch is not moving in this study, their is no need to use filler metal. The welding current is 70A, the welding voltage is 10.5V, and the flow rate of the Ar shielding gas is 10L/min.

 

Comment #2

Caption of Fig. 17 is incomplete. Fig. 18 is missing and is not mentioned in the text.

 

Respond to Comment #2:

Thanks for your suggestion. It is our mistake that the Fig. 18 should be modified as the Fig. 17(b). we have already modified it in the revised manuscript.

The revised content in Page 16 Line 371:

 

Figure 17. The optimization results of the weld pool surface in Fig. 2(a). (a) the weld pool surface after boundary optimization and the forward-reverse united reconstruction optimization method; (b) the boundary comparison between the measured weld and the reconstructed weld pool surface.

 

 

 

Author Response File: Author Response.pdf