Supervised and Unsupervised Machine Learning Algorithms for Forecasting the Fracture Location in Dissimilar Friction-Stir-Welded Joints

Round 1

Reviewer 1 Report (Previous Reviewer 2)

The information following answer given by the authors: "This work is different from previous published works, because in the previous work we just used supervised machine learning in algorithms without comparing their performance on the basis of AUC Score. But in this work, we have compared the performance on the basis of AUC score as well as we have used an Unsupervised Machine Learning for the first time in this work." needs to be included in the revised manuscript. On the other words, the authors must state, in the revised manuscript, the advances of the present manuscript when compared with their previous published works.

Author Response

Dear Reviewer, I have carried out the required amendment and further highlighted

Dear Reviewer, I have carried out the required amendment and further highlighted it.

Reviewer 2 Report (Previous Reviewer 3)

In this paper, the algorithm based on artificial intelligence is constructed to predict the fracture position of AA5754-C11000 alloy in heterogeneous friction stir welding. For the first time, an unsupervised machine learning algorithm, which is based on Self-Organizing Mapping (SOM) neural network, is implemented. The implementation of each interface layer in the model is described. According to the results of the article, the algorithm can predict the fracture position more accurately. It is meaningful for the developments of friction stir welding for dissimilar alloys. So this manuscript can be accepted and published under major revision. The modifications are shown below.

1. Page 3, line 133 to line 134: There is a redundant number “2” in the sentence “2. Tensile Specimens are classified on the basis of extracting the samples from 133 the different location of the weld line.”.

         2. Page 4, line 141: The figure title in the text should be similar to the format of the text, without capitalization in the figure title. For example, “Implementation of Supervised Machine Learning Algorithms” should be “Implementation of supervised machine learning algorithms”.

Author Response

Dear Reviewer, I have carried out the required corrections. 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

The topic of the paper is forecasting of fracture location in the dissimilar friction stir welded aluminum-copper alloys using different machine learning algorithms. The fractures to be detected are located either in the aluminum or in the copper zone, this is the output parameter of classification algorithms while input parameters are tensile sample and three process dependent characteristics of stir welding. Authors have developed four supervised and one unsupervised learning algorithm for forecasting the location of fracture and evaluated the performance of them. The elaborated methods are correctly demonstrated using detailed description of calculation steps except the last SOM method. Authors proved that the proposed Self-Organizing Maps Neural Network achieves better performance than supervised learning methods.

Some remarks for Authors:

·         The definition of abbreviations is missing sometimes (for example Row 41: CNN)

·         R107: “Friction stir welding has been shown to produce join aluminum alloys with greater mechanical properties…”

·         The first input parameter is not correctly defined. In R125 “tensile specimen” is mentioned, later in Table 1. tensile sample. It might be useful to show the different shapes of specimens by a simple sketch or more detailed description.

·         R139: “Pandas, NumPy, matlplotlib” correctly: Matplotlib

·         Table 1. Count should not be given by 3 decimal digits

·         Figure 6. It is impossible to read the boxes

·         R303: “The accuracy score obtained as result was 0.9692…” This statement should be better supported.

Author Response

Dear Reviewer, we thank you for pointing out the important amendments. So, we have carried out the following corrections:

1. The definition of abbreviations CNN is Convolutional Neural Network, we have highlighted the correction.
2. Tensile Specimens are classified on the basis of extracting the samples from the different location of the weld line.
3. Matplotlib is corrected.
4. Value of count is corrected
5. It is the obtained output after executing the code. We have decided to remove it.
6. It is due to the fact that competitive learning is at the heart of SOM. In competitive learning, the distance between input data and neuron weight influences neuron activity. The most learning occurs in an excited neuron, and as a result, its weights change.

Reviewer 2 Report

The same aprroach of using machine learning for stir welding has been published by the same authors as I point some of them below. At the moment, I'm against the publication of the manuscript due to the lack of novelty, unless the authors show the relevant contribution of the present manuscript in relation to their previously published works.

#1 Supervised machine learning classification algorithms for  detection of fracture location in dissimilar friction stir welded joints. Frattura ed Integrità Strutturale, 58 (2021) 242-253; DOI: 10.3221/IGF-ESIS.58.18

#2 Machine Learning Classification Models for Detection of the Fracture Location in Dissimilar Friction Stir Welded Joint. Applied Engineering Letters Journal of Engineering and Applied Sciences 5(3):87-93, 2020. DOI:10.18485/aeletters.2020.5.3.3

#3 Detection of Surface Defects in Friction Stir Welded Joints by Using a Novel Machine Learning Approach. Applied Engineering Letters Journal of Engineering and Applied Sciences 5(1):16-21, 2020. DOI:10.18485/aeletters.2020.5.1.3

Further important poits:

The authors must justify the interest on welding different AA5754-C11000 alloys. Otherwise an infinite number of works can be done just changing the joined materials.

Figure 6 is unreadable.

The trained algorithm must be validated with a practical condition, obviously different from the trained conditions (mandatory).

Author Response

Dear Reviewer, we are grateful for pointing out the amendments in the given work. We would like to mention few points about the corrections in the revision:

1. This work is different from previous published works, because in the previous work we just used supervised machine learning in algorithms without comparing their performance on the basis of AUC Score. But in this work, we have compared the performance on the basis of AUC score as well as we have used an Unsupervised Machine Learning for the first time in this work.
2. Aluminum (Al) and copper (Cu) dissimilar welding has several uses in the electrical power, electronic, and pipeline industries. These applications place a high priority on the weldments because of their ability to conduct heat, electricity, and corrosion. We have included this statement highlighted by the green colour in the revised manuscript.
3. As figure 6 is a full length image. It cannot be properly visible due to the template format as it is getting compressed. So, we have decided to remove it.
4. The purpose of using Machine Learning algorithm is to decrease the experimental work and reduce the time. So, it will be the future scope of this work to cross verify the results with other algorithms as well.

Reviewer 3 Report

Review comment

In this paper, the algorithm based on artificial intelligence is constructed to predict the fracture position of AA5754-C11000 alloy in heterogeneous friction stir welding. For the first time, an unsupervised machine learning algorithm, which is based on Self-Organizing Mapping (SOM) neural network, is implemented. The implementation of each interface layer in the model is described. According to the results of the article, the algorithm can predict the fracture position more accurately. It is meaningful for the developments of friction stir welding for dissimilar alloys. So this manuscript can be accepted and published under major revision. The modifications are shown below.

1. Page 3, line 100 to line 108: It is not necessary to introduce the principle of friction stir welding in the second section. It can be deleted or moved in the introduction.

2. Page 3, line 125 to line 127: In the second section, the author mentioned that the tensile sample is used as the input parameter. However, as we all know, there seems to be some ambiguity between the tensile sample and the input parameter. Please confirm it.

3. Page 5, line 182 to line 186: “heat map” appeared here for the first time, but the subsequent explanation failed to make the reader understand what the meaning of heat map is and its importance to the model studied in the article. I hope the author can explain it in more detail according to his own understanding.

4. Page 6, line 207 to line 210: Why should the F1-score and AUC score of the algorithm be evaluated? What are the effects of these two parameters on machine learning? Please explain these two indicators.

5. Page 6, line 213 to line 220: Please explain the function of Decision Tree plot.

6. The content of the article only focuses on the construction of the model. According to the theme of the article, the model is used in the friction stir welding scenario, but the uniqueness of the model in the friction stir welding process is not analyzed in the article.

Page 6, line 196 to line 199: “It is observed that the Tensile Sample parameter contributes towards the output most while Rotational speed (RPM) has a negligible impact on the output parameter.” Please explain why the influence of rotational speed on output can be ignored.

7. Page 6, line 182: The space between the words "heat" and "map" is missing.

Author Response

Dear Reviewer, we are grateful for pointing out the amendments in the given work. We would like to mention few points about the corrections in the revision:

1. We have moved the required amendment to the introduction part, highlighted by blue colour.
2. Tensile Specimens are classified on the basis of extracting the samples from the different location of the weld line. We have highlighted in the red colour.
3. These coefficients are displayed as a heat map to show the degree of association between various factors. It assists in identifying traits that are ideal for creating machine learning models. The correlation matrix is converted into color labeling via the heat map. We have included this statement highlighted in blue colour.
4. “An evaluation metric for a classification that is defined as the harmonic mean of recall and precision is the F1-Score or F-measure. It is a metric used in statistics to assess how accurate a test or model is. The capacity of a classifier to differentiate between classes is measured by the Area Under the Curve (AUC), which is used as a description of the ROC curve. The model performs better at differentiating between both the positive and negative classes the higher the AUC.” We have included this statement highlighted by blue colour.
5. We decided to remove the plot due to poor visiblility and low resolution as it was not visible to other reviewers.
6. In the present work, for the first time a neurobiological based unsupervised machine learning algorithm i.e., Self-Organizing Map (SOM) Neural Network is implemented for determining the fracture location in dissimilar friction stir welded AA5754-C11000 alloys. We have mentioned this in the abstract part.
7. We have indicated from the Feature importance plot that Rotational Speed is not contributing to the output parameter.
8. Space is inserted between “heat” and “map”.

Reviewer 4 Report

1-In all equations: The mathematical relationship must be shown clearly.

2-Enlarge Figure 6 until it becomes clear.

3-Clarify or correct the mathematical relationship in line 277.

4-In the word search, you must review the use of capital letters or small letters.

5-The number of references is few. Is it possible to increase the number of references through search?

Author Response

Dear Reviewer, we are grateful for pointing out the amendments in the given work. We would like to mention few points about the corrections in the revision:

1. We have pointed out the relation between the variables in the equations.
2. We have decided to remove figure 6, as on enlarging the pixelated image was not clear.
3. The minimization of the Euclidian distance means the maximization of the wTx which is the output (w is the weight matrix).
4. Use of capital and small letters are reviewed.
5. We have included more references.