Real-Time Modeling for Design and Control of Material Additive Manufacturing Processes
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors
Dear author,
Thank you for this very interesting work related to the development of a hybrid ROM-ML model for real-time modeling of Wire Arc Additive Manufacturing process. However, please find a list of comments that have to be addressed. To begin with, it could be beneficial for the manuscript to review the text with a native English-spoken person and improve it accordingly since in many sentences it is difficult to understand what the author is trying to say.
Title and Abstract
The title is confusing since the “Controlling of material Additive Manufacturing Processes” part of the sentence is not clear. Especially, the term” Material Additive Manufacturing processes” does not stand in the standard terminology of AM processes.
Line 8. The term cyber manufacturing processes is not very accurate and descriptive since it is not explained. It is suggested, at first, to introduce somehow what is included in this term and then compare with physical manufacturing processes. This comment applies also to the rest manuscript.
Lines 13-14. In order to support the argument about the transformation of processes with various means such as digital technology, hybrid physical data-driven modelling and fast reduced models, lot of introductory material is needed which is not optimal decision for the abstract section. In this section try to be concise, focusing on the outputs of this work and the challenges.
Line 20. The author should probably elaborate what is the meaning of transient manufacturing processes and what process are included.
Introduction
Line 31. In this line the author talks about the “process data paradigm”. However, it is not very clear how this phrase is interpreted within the sentence. It is advised to elaborate on that and describe the paradigm.
Lines 46-50. The author introduces that in this work the modelling techniques will be applied on the WAAM process. However, it is not clear the background of this decision. Is this modelling technique applicable only in WAAM process? If yes why, and please describe how the approach of this work would change in the case of another AM process. As it regards the various modelling techniques, please find hereafter different sources that discuss the data-driven and physics-based simulation of DED AM processes, leveraging process data and creating process knowledge.
· Data analysis to assess part quality in DED-LB/M based on in-situ process monitoring. Procedia CIRP, 111, 345–350. https://doi.org/10.1016/j.procir.2022.08.036
· A computationally efficient Multi-Scale Thermal Modelling Approach for PBF-LB/M based on the Enthalpy method. Metals, 12(11), 1853. https://doi.org/10.3390/met12111853P
At this point and based on the description of the aforementioned sources, the author could talk about the significance of selecting the key variables for this study as they are mentioned briefly in line 49. Additionally, the author could talk about the competitive advantage of WAAM over other AM processes while also comparing their process mechanisms in an attempt to depict the significance of data-driven approaches on correlating process variables with process performance. This information is also vital because the reader would be able to understand better how the proposed models of sections 3.1 and 3.2 are implemented.
Simulation of AM Processes -Methodology
Line 56. It is suggested to the author to follow the correct terminology about AM processes. It is advised to refer to AM, as “AM processes” instead of “AM process technologies”.
Line 97. In the title please remove the work subsection.
Lines 115-129. It is not clear how often this loop is repeated during the deposition process. Also it is suggested to include a representative schematic of this sequence of steps.
Figure 1b) The content of this subfigure is not readable. Therefore, if the author considers this information as necessary it is advised to split Figure 1 in two or three figures so as to improve the readiness of the content.
Methods for AM Reduced Models
Line 231. The author presents an interesting way of identifying the equation-based link of variables and process outputs, however, the error of the proposed technique is not quantified compared to the other presented methods.
Lines 239-240. “In many … data”. This is not true in many cases. Data can be acquired by many monitoring devices in AM processes. Line 283. The author needs to clarify why is the temperature selected as the input parameter of the model?
Lines 303-309. The author should elaborate on the following: selection of the specific geometry, pre-heating strategy, selection of material, technique and monitoring device. Also is it possible to provide figures from the actual experimental setup as long as of the parts produced? Do the parts require post processing, which is very common in metal additive manufacturing and especially in WAAM?
Figure 2. (a) The parameter selection is also not justified, why were these set of parameters tested and not others?
Figure 3 and 4. These plots are not explained in detail and they are just introduced. For example, in figure 4(a) the author should explain why the RMS error is lower when using 1HL rather than 3HL. Also in fig 4(b) the ”all’’ label is not needed because the reader cannot see any values that are in this category in the plot. If the author wants to have it, it has to be supported by visible values.
Discussion
Line 358-359. What does the author mean by “However, … discussions”?
Figures 5 and 6 are not aligned and in the whole paper the figures are not aligned with the text.
Conclusions
The limitations or potential challenges that the hybrid ML-ROM model might face in real-time application to WAAM processes, are not mentioned by the author in this section. This would give a more balanced and critical view of the proposed model.
Thank you
Comments on the Quality of English Language
Moderate editing of English language required.
Author Response
Comment 1: The title is confusing since the “Controlling of material Additive Manufacturing Processes” part of the sentence is not clear. Especially, the term” Material Additive Manufacturing processes” does not stand in the standard terminology of AM processes.
Line 8. The term cyber manufacturing processes is not very accurate and descriptive since it is not explained. It is suggested, at first, to introduce somehow what is included in this term and then compare with physical manufacturing processes. This comment applies also to the rest manuscript.
Lines 13-14. In order to support the argument about the transformation of processes with various means such as digital technology, hybrid physical data-driven modelling and fast reduced models, lot of introductory material is needed which is not optimal decision for the abstract section. In this section try to be concise, focusing on the outputs of this work and the challenges.
Line 20. The author should probably elaborate what is the meaning of transient manufacturing processes and what process are included.
Response 1: Thanks for the comprehensive review of the manuscript. The manuscript title is supposed to give a brief representation of the content and since the aim of these data models is to predict and correct in real-time, they are enabling more optimised process controlling. Furthermore, the role of these data models on material designs for AM processes and the optimisation of process parameters make them an effective tool for AM design & optimisation.
The cyber manufacturing processes which sometimes described as the combination of physical, cyber tools/digital replica and twinning systems are well known. However, here in this manuscript, the only aspects of digital replica and twinning systems were intended. Although aspects of physical testing and experimental DOEs were carried out for calibration, as stated in the manuscript, only the data tools for the digital replica have been discussed. Nevertheless, to further clarify this point, text has been added to the manuscript to elaborate on the definition of the digital replica for AM processes.
Considering the first part of the abstract, as the whole section is about two hundred words, and the intention here is to give as much information as possible within this limit, the first part of the abstract has been written in a way to give a very brief overview of the state-of-the-art for industrial digitalisation.
On the definition of transient manufacturing processes, the manufacturing processes with generative nature in time were intended herein (like, AM, casting…). For the AM and other time-dependent generative manufacturing processes, the physical and simulation domains are continuously growing during these processes.
Comment 2: Line 31. In this line the author talks about the “process data paradigm”. However, it is not very clear how this phrase is interpreted within the sentence. It is advised to elaborate on that and describe the paradigm.
Lines 46-50. The author introduces that in this work the modelling techniques will be applied on the WAAM process. However, it is not clear the background of this decision. Is this modelling technique applicable only in WAAM process? If yes why, and please describe how the approach of this work would change in the case of another AM process. As it regards the various modelling techniques, please find hereafter different sources that discuss the data-driven and physics-based simulation of DED AM processes, leveraging process data and creating process knowledge.
- Data analysis to assess part quality in DED-LB/M based on in-situ process monitoring. Procedia CIRP, 111, 345–350. https://doi.org/10.1016/j.procir.2022.08.036
- A computationally efficient Multi-Scale Thermal Modelling Approach for PBF-LB/M based on the Enthalpy method. Metals, 12(11), 1853. https://doi.org/10.3390/met12111853P
At this point and based on the description of the aforementioned sources, the author could talk about the significance of selecting the key variables for this study as they are mentioned briefly in line 49. Additionally, the author could talk about the competitive advantage of WAAM over other AM processes while also comparing their process mechanisms in an attempt to depict the significance of data-driven approaches on correlating process variables with process performance. This information is also vital because the reader would be able to understand better how the proposed models of sections 3.1 and 3.2 are implemented.
Response 2: The point about data-process paradigm is taken and the topic has been elaborated.
The WAAM case study in this research work has been carried out due to the availability of the experimental data for model calibrations (as described in sub-section 3.3). There is no obstacle to apply similar data techniques for other types of AM processes, since the technique is quiet general.
As the intention here in this work is to elaborate on the data real-time models no extensive description has been presented for WAAM process and its advantages and short comes over the other popular AM processes. For more elaboration on different AM processes and their potential manufacturing features more references have been added to the reference list.
Comments 3: Line 56. It is suggested to the author to follow the correct terminology about AM processes. It is advised to refer to AM, as “AM processes” instead of “AM process technologies”.
Line 97. In the title please remove the work subsection.
Lines 115-129. It is not clear how often this loop is repeated during the deposition process. Also it is suggested to include a representative schematic of this sequence of steps.
Figure 1b) The content of this subfigure is not readable. Therefore, if the author considers this information as necessary it is advised to split Figure 1 in two or three figures so as to improve the readiness of the content.
Response 3: The text has been revisited and some minor corrections have been carried out.
For the lines 115-129, process flowchart has been shown in figure 1a.
For the figure 1b, the graph is representative and more detailed discussions about the evolving domain technique can be found in the author’s previous publications (listed in the references).
Comments 4: Line 231. The author presents an interesting way of identifying the equation-based link of variables and process outputs, however, the error of the proposed technique is not quantified compared to the other presented methods.
Lines 239-240. “In many … data”. This is not true in many cases. Data can be acquired by many monitoring devices in AM processes. Line 283. The author needs to clarify why is the temperature selected as the input parameter of the model?
Lines 303-309. The author should elaborate on the following: selection of the specific geometry, pre-heating strategy, selection of material, technique and monitoring device. Also is it possible to provide figures from the actual experimental setup as long as of the parts produced? Do the parts require post processing, which is very common in metal additive manufacturing and especially in WAAM?
Figure 2. (a) The parameter selection is also not justified, why were these set of parameters tested and not others?
Figure 3 and 4. These plots are not explained in detail and they are just introduced. For example, in figure 4(a) the author should explain why the RMS error is lower when using 1HL rather than 3HL. Also in fig 4(b) the ”all’’ label is not needed because the reader cannot see any values that are in this category in the plot. If the author wants to have it, it has to be supported by visible values.
Response 4: The error for reduction techniques depends on different parameters, including, data quality, size and dimensions of data space… and for the current work, the normalised errors are shown in figures 5 to 8.
For the argument in lines 239-240, as it is explained in the text, the large number of influential parameters and their variations and interactions can make the data predictions difficult and inaccurate. It is not just the amount of data, rather it is the complicated patterns of data for different input variations which make the data processing task cumbersome. Although the temperature, stress and deformation data were processed for the work herein, only the temperature data has been presented due to the space limitations.
As explained earlier, the focus of this paper is on the performance of data real-time models for manufacturing processes and hence, for the details of WAAM processes and their setup readers are referred to industrial literature.
For creating snapshot matrix in figure 2, sampling techniques are used to create scenarios with varying process parameters (as discussed in section 4).
For the figure 3 and 4, the schematic architecture of the NN framework (with single hidden layer) and tree shape diagram of GASR used in this study are shown. A brief description of NN framework has been given in subsection 3.2, however, more elaborations on these data science techniques requires more detailed mathematical and numerical discussions which are out of scope for this paper. Readers can find more in-depth discussions about NN and GASR techniques in the provided references.
For figure 4b, no such a label is found in the text or the figure.
Comment 5: Line 358-359. What does the author mean by “However, … discussions”?
Figures 5 and 6 are not aligned and in the whole paper the figures are not aligned with the text.
Response 5: The results presented for the case study section (at the end of the manuscript) is limited to temperature results since the rest of the mechanical and deformation results require a much larger sections and sub-sections to sufficiently elaborate on. Hence, to keep the paper content balanced and avoid having a very lengthy manuscript, the rest of the results are going to be presented in future publications.
The alignments of figures throughout the whole manuscript are double checked against the journal template. However, the process of printing the document in pdf format might cause some minor alteration in the formatting.
Comment 6: The limitations or potential challenges that the hybrid ML-ROM model might face in real-time application to WAAM processes, are not mentioned by the author in this section. This would give a more balanced and critical view of the proposed model.
Response 6: Although the details of challenges facing data real-time models for AM processes have been given in the discussion section, some text modifications have been carried out in the conclusion section to reflect these realities again.
Reviewer 2 Report
Comments and Suggestions for Authors
The presented study entitled "Real-Time Modeling for Design and Controlling of Material Additive Manufacturing Processes" presents a new insight into the hybrid techniques of modeling based on physical data and reduced order modeling (ROM) for digitizing AM processes within the digital twin concept. The main contribution is to show the advantages of ROM and machine learning (ML) technology for process data processing, optimization\controlling and their integration in real-time evaluation of AM processes. Therefore, a new combination of efficient data processing technology and an architecturally designed neural network (NN) module was developed for transient manufacturing processes with high heating/cooling rates. In addition, a real-world case study was presented where a combination of hybrid modeling along with ROM and ML schemes is used for an industrial wire arc AM (WAAM) process.
The presented study solves a very interesting problem in a "modern" way, and at the same time, as the author states, a real case study is also used in it, which I greatly appreciate. The study is divided into 5 basic chapters, and their further division has a logical structure and makes it easier to follow the text.
Overall, I appreciate the author's efforts and his approach to solving the problem. I have no objections to the study in any aspect and I recommend publishing it in its current form.
Author Response
Comment 1: The presented study solves a very interesting problem in a "modern" way, and at the same time, as the author states, a real case study is also used in it, which I greatly appreciate. The study is divided into 5 basic chapters, and their further division has a logical structure and makes it easier to follow the text.
Overall, I appreciate the author's efforts and his approach to solving the problem. I have no objections to the study in any aspect and I recommend publishing it in its current form.
Response : Many thanks for the thorough evaluation of the manuscript. I am grateful for the time and efforts spend by you and the Journal editors on careful review of the manuscript and the detailed resulting final comments.
Reviewer 3 Report
Comments and Suggestions for Authors
This paper presented a hybrid physical-data driven and ROM technique and used a WAAM process as a case study. The idea is interesting and novel. Therefore, I suggest that we publish it after minor revisions. To meet the publication requirements, the authors need to address the following issues:
1. The introduction to the background in the abstract is too lengthy. It is recommended to shorten it.
2. Using a yellow background for Fig.4 to Fig.8 looks ugly. It is recommended to change it.
3. Please adjust the format of the references according to the journal's requirements and ensure consistency throughout.
Comments on the Quality of English Language
Minor editing of English language required.
Author Response
Comment 1: The introduction to the background in the abstract is too lengthy. It is recommended to shorten it.
Response 1: Thanks for the kind review and advices. The abstract is about two hundred words, and the intention here is to give as much information as possible within this limit. Hence the first part of the abstract has been written in a way to give a very brief overview of the state-of-the-art for industrial digitalisation.
Comment 2: Using a yellow background for Fig.4 to Fig.8 looks ugly. It is recommended to change it.
Response 2: Thanks for the recommendation. The beauty and ugliness of the colouring schemes for these graphs are a matter of personal opinion/choice and no scientific merits can be generalised for these types of styles.
Comment 3: Please adjust the format of the references according to the journal's requirements and ensure consistency throughout.
Response 2: The point has been taken, and the references’ format and styles were double checked to make sure that they comply with the journal’s instructions.
Reviewer 4 Report
Comments and Suggestions for Authors
1. The conclusions should be simplified.
2. Author mentioned that the accurate modelling of deposition processes are still a cumbersome task. However, the established simulation model and the simualtion results were not illustrated. Furthermore, what was done by simulation in this work?
3. What is the input data for the reduced model? Did you reserve data without joining training process for prediction verification?
4. What are the errors of these reduced models?
5. What is your application of the predication results?
Author Response
Comment 1: The conclusions should be simplified.
Response 1: Thanks for the detailed review of the manuscript. The conclusion section has been revisited again and no complex aspects which might be cumbersome to readers has been found. Nevertheless, some texts have been added to make the final outcomes clearer.
Comment 2: Author mentioned that the accurate modelling of deposition processes are still a cumbersome task. However, the established simulation model and the simualtion results were not illustrated. Furthermore, what was done by simulation in this work?
Response 2: As it has been mentioned by many authors and researchers, the multi-physical and multi-scale nature of material processes are difficult to model and although today’s simulation techniques have come a long way to model some of these inherited aspects, there are still many challenges ahead (as stated in the references 11 to 14). Readers are referred to these published works and many more available research outcomes for further elaboration. In light of this comment, some extra references were added to give the reader more detailed discussions on numerical simulation of material processes.
Comment 3: What is the input data for the reduced model? Did you reserve data without joining training process for prediction verification?
Response 3: As it has been clearly stated in the manuscript, the data for the reduced model is came from the database which has been generated out of the snapshot matrix (please see the section 3 and figure 2). The verified simulation results were post processed and collected according to scenarios in the snapshot matrix and validations are done using extra scenarios (DOEs) at normal, near boundary and beyond boundaries (as detailed in section 4).
Comment 4: What are the errors of these reduced models?
Response 4: The normalised errors for the data models compared to verified simulation results were calculated and shown in the figures 5 to 8 for different validation scenarios.
Comment 5: What is your application of the predication results?
Response 5: As stated in the manuscript, these real-time data models and their predictions can be integrated into the digital twin and shadow frameworks. They can also play a crucial role in advisory systems for design and optimisation of material processes.
Round 2
Reviewer 4 Report
Comments and Suggestions for Authors
1. Author mentioned that the accurate modelling of deposition processes are still a cumbersome task. How about your own model? Is it accurate? How to make it accurate?
2. What were your simulation result data and contours?
3. What is the input data for the reduced model? Please illustrate in details. Did you reserve data without joining training process for prediction verification? As we know, the data for prediction verification can not join training process.
4. Reviewer can not believe the potential application. It is essential to illustrate at least one application.
5. In the simple word, this work use some data from simulations and training a model, and there are not any experiments. Consequently, this work can not make me convinced.
Author Response
Comment 1: Author mentioned that the accurate modelling of deposition processes are still a cumbersome task. How about your own model? Is it accurate? How to make it accurate?
Response 1: Thanks for the thorough review of the manuscript. The issues of accuracy, generality and applicability have always been one of the major discussions when developing modelling and simulation techniques. Although new modelling techniques, including data real-time modelling, can improve the accuracy and applicability of existing methods, as mentioned in manuscript’s section 3 and 4, there are also some challenges facing these techniques. Nevertheless, some more text has been added to the manuscript, especially the conclusion section to clarify these challenges.
Comment 2: What were your simulation result data and contours?
Response 2: As it has been stated in the manuscript sections 2 and 3, the verified numerical simulations are used to generate response data for the data real-time models. Since the discussions in the paper is focused on the data modelling and its application for the additive manufacturing predictions, a brief discussion about the numerical evolving domain and its mesh generation techniques were presented in sections 2.2, 3.3, figures 1 and 2. More comprehensive discussions about these simulation techniques and their results can be found in the reference 5, 12 and 20.
Comment 3: What is the input data for the reduced model? Please illustrate in details. Did you reserve data without joining training process for prediction verification? As we know, the data for prediction verification can not join training process.
Response 3: As it has been presented in the sections 3 and 4, the data for the generation of data real time models are obtained using verified simulation results. To validate the data models, as it stated in the section 1, part of available data has been left out of the database, to be used for validation after the generation of data models. Moreover, the validation of these models, as stated in the section 4, were carried out at internal, near boundary and extreme condition (extrapolation) for the given data search space.
Comment 4: Reviewer can not believe the potential application. It is essential to illustrate at least one application.
Response 4: To prove the applicability of data real time models, an additive manufacturing case study has been carried out herein, to examine the accuracy and efficiency of these models (please see sections 3 and 4). Furthermore, there are many applications of these data techniques for other manufacturing processes which have been reported by the author and other researchers in the literature. Please check out the reference section for more publications on this topic.
Comment 5: In the simple word, this work use some data from simulations and training a model, and there are not any experiments. Consequently, this work can not make me convinced.
Response 5: As it has been stated in the section 3, some initial experimental deposition works have been carried out to validate and calibrate the numerical finite element simulations. The simulation results were only used after thorough validations for accuracy and stability using experimental measurements. Finally, as it has clearly been explained in the introduction and conclusion sections, these data models are not intended to replace other experimental and\or detailed numerical process simulations, rather they can be valuable assets within the digitalization framework for process controlling and optimizations.
Round 3
Reviewer 4 Report
Comments and Suggestions for Authors
Almost all my comments have been responded, it can be recommended to be accepted.
