The first step towards intelligent wire arc additive manufacturing: An automatic bead modelling system using machine learning through industrial information integration
Introduction
Additive Manufacturing (AM) has gained popularity worldwide over the past decades. It is often interchangeable when referring to the AM process with terms of 3D printing, layered manufacturing, rapid prototyping, and rapid manufacturing [1]. AM technology covers many processes in which consumable material is deposited or jointed under the control of certain Computer Aid Manufacturing (CAM), depositing three-dimensional object layer by layer [2]. Recently, the focus of AM has changed to fabricate expensive metal components such as titanium [3] and nickel alloys [4] in the aerospace industry where such components often suffer an extremely high buy-to-fly ratio [5]. Commonly, AM technologies for metallic parts include: i) wire arc additive manufacturing (WAAM); ii) Electron beam freeform fabrication (EBF3); iii) Electron beam melting (EBM); iv) Selective laser sintering (SLS); and Selective laser melting (SLM) [6]. amongst these, WAAM, which combines electric arc as the power source with welding wire as a feedstock, is able to fabricate large scale parts with high efficiency and low cost.
As shown in Fig. 1, the WAAM process consists of several major processing steps: (i) bead modelling, the relationship between the welding parameters and the bead geometries; (ii) 3D slicing, a 3D model is sliced into a set of 2.5D layers with predetermined thickness; (iii) 2D path planning, the optimal deposition path is designed to fill the 2.5D cross-sections; (iv) deposition process, the welding process parameters are selected according to the weld bead modelling and the molten material is deposited along the path layer upon layer; and (v) post-process machining, the part is machined for removing extra support structure and/or to meet a desired surface quality [7]. In the WAAM process, each layer is deposited with a large number of single weld beads side by side. The offset distance of adjacent beads has a great impact on the geometric accuracy and surface quality [5]. Besides, the height of the weld bead, to some extent, is also required to be the same as the layer thickness for 3D slicing. To realize a practical WAAM process, the relationship between weld bead geometry and welding process parameters is critical to select optimal welding process variables. However, for current WAAM systems, the welding variables are still determined manually or partially automated based on a database and operator experience that may affect the accuracy and the efficiency of the WAAM process [8]. For example, in the study by Ding, et al. [7], the authors proposed a fully automated WAAM system, which only requires the 3D profile as system input and creates the final part without human intervention. However, the bead modelling module is only capable of predicting bead geometry with the given welding parameters. Similarly, in the WAAM system proposed by Prado-Cerqueira, et al. [9], the collected experimental results were used as a database to provide information for weld settings. Therefore, it is imperative to develop an automated bead modelling system to achieve an intelligent WAAM process.
Currently, many modelling methods have been proposed for creating a forward model for weld bead formation, such as traditional regression model [10], [11], [12], Taguchi approach [13], and Artificial neural networks (ANNs) [14], [15], [16]. The forward model is referring to the predictive model that is able to predict the bead geometry based on the given welding parameters. However, it is difficult to produce accurate weld bead geometry with the limited welding parameter sets. This is considered a drawback of the forward model. In practical industrial production, parts may have complex geometries, therefore the weld bead geometries are always required to be varied slightly when fabricating parts with complex structures. Ideally, the weld bead model should be used in reverse, which means the welding parameters can be obtained based on the weld bead geometry generated in the path planning process. Xiong, et al. [8] used ANNs method to develop both forward and backward model for predicting welding parameters. The backward model is referring to the predictive model that is able to predict the welding parameters based on designated bead geometry. The authors conclude that the accuracy of the backward model is unsatisfactory, and the forward model was used to examine the accuracy of the backward model. If the error between the predicted and desired bead geometry is unacceptable, the welding variables are slightly adjusted for the next iteration until the error becomes smaller than a threshold. Note that the parameters are adjusted based on the influence of welding variables on the bead geometry according to the reference [10].
Compared to conventional ANNs based model, Support vector machines (SVMs), an alternative machine learning algorithm, has the potential to provide a more accurate and efficient solution for the bead modelling process in WAAM. In machine learning, SVMs are supervised learning methods with corresponding learning algorithms that process data used for regression analysis [17]. In many studies, the SVMs are identified to be superior to ANNs, as they avoid some major weaknesses of ANNs: (i) SVMs often converge on global minimum rather than local minimum, which means ANNs model may miss the optimal result [18]. (ii) ANNs has the limitations on generalization giving rise to models that may lead to overfit the data. (iii) The training time of SVMs is substantially less than that of ANNs [19]. In the current literature, SVMs have been demonstrated a good capability of solving regression issues for various applications [20], [21], [22]. For the modelling problem in welding, Chen, et al. [23] proposed a modelling method to predict weld penetration using SVMs in gas tungsten arc welding. The experimental results indicate that the SVM-based model shows a higher level of accuracy in predicting welding penetration by comparing with the ANN method. To date, little has been done to create a highly accurate weld bead model for WAAM using SVMs.
This study aims to develop an automated bead modelling system from weld bead deposition, data collection, and processing to welding parameters prediction in WAAM process. Conventionally, the system would be developed by manufacturing engineer. However, knowledge on a single discipline can not meet the industrial requirements of high level of accuracy, product quality, and system automation, the integration of technologies from a number of disciplines should be considered. Industrial information integration engineering is carried out to solve complex industrial problems by combining methods [24]. As comprehensively reviewed in [25,26], industrial information integration engineering is an emerging subject attracting much attention in academia. Moreover, Chen [25] also pointed out that research on the manufacturing category was the second biggest category in this industrial information integration. Thus, the bead modelling system was proposed through integrating information and knowledge from computer science, material engineering, mechanical engineering, and industrial system engineering. As the first step of the WAAM process, the fully automated WAAM system can improve product quality and reduce the time required for the design and manufacturing cycle. In addition, Machine learning is firstly used for weld bead modelling in WAAM process. A novel algorithm using SVM was proposed for the system which provides users with a set of accurate welding parameters for the deposition of weld bead with designate geometry.
In the following sections, the workflow of the proposed system is described in Section 2. In Section 3, the data collection and processing system is presented and the SVMs based modelling algorithm is detailed in Section 4. The effectiveness of the proposed system is examined through the experimental validation and a case study in Section 5, and followed by a conclusion in Section 6.
Section snippets
System overview
This paper proposes a novel Computer Aided Manufacturing (CAM) system for bead modelling process using arc welding-based AM technology. Weld beads are deposited side by side, as demonstrated in Fig. 2, the sliced layer height and offset distance are the same as the desired bead high (BH) and overlapping distance (OD). Thus, the proposed system aims to generate optimal welding parameters and setups to produce the user preferred bead geometry.
Data collection and processing
Generally, the bead modelling process identifies the relationship between weld settings and bead profiles. The data generation module is used to provide essential weld bead geometry data for creating the weld bead model. The module includes three major steps, bead on plate deposition, data collection, and data processing.
Modelling of the weld bead
This section includes the model building process and parameter prediction process. The purpose of this modelling process is to generate a model that can represent the relationship between welding parameters and bead profile parameters, and furthermore, is to provide the welding parameters that meet the path planning state requirements, as discussed in Section 2. Therefore, a successful model should be able to predict the welding parameters by the given bead profile parameters which are OD and
Experimental validation and discussion
The performance of the proposed algorithm and system was evaluated through the fabrication of a real-world workpiece, whose 3D model is shown in Fig. 12(a). The detailed fabrication process and relevant discussion are provided in this section.
Conclusion
This paper developed an automated weld bead modelling system for the WAAM using machine learning. The system is a novel backward model, which generates the optimal deposition parameters according to the desired weld bead geometry under various temperatures (IPT). An innovative predictive model had been established by using the SVR method for predicting welding parameters with minimum human operation. By using the proposed SVM based algorithm, weld bead models can be created with a high level of
CRediT authorship contribution statement
Donghong Ding: Formal analysis, Validation, Investigation, Resources. Fengyang He: Software, Writing - original draft, Data curation. Lei Yuan: Conceptualization, Writing - review & editing, Supervision. Zengxi Pan: Supervision, Project administration, Writing - review & editing. Lei Wang: Methodology, Writing - review & editing. Montserrat Ros: Writing - review & editing, Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
This work was supported in part by the National Natural Science Foundation of China under Grant 51805085 and in part by China Scholarship Council under Grant 201708200016, 202008200004.
Reference (41)
- et al.(2011)
- et al.
Shaped metal deposition of a nickel alloy for aero engine applications
J. Mater. Process. Technol.
(2008) - et al.
A multi-bead overlapping model for robotic wire and arc additive manufacturing (WAAM)
Robot. Comput. Integr. Manuf.
(2015) - et al.
Enhanced beads overlapping model for wire and arc additive manufacturing of multi-layer multi-bead metallic parts
J. Mater. Process. Technol.
(2018) - et al.
Towards an automated robotic arc-welding-based additive manufacturing system from CAD to finished part
Comput. Aided Design
(2016) - et al.
Preliminary development of a wire and arc additive manufacturing system (WAAM)
Proc. Manuf.
(2017) - et al.
3D welding and milling: part II—optimization of the 3D welding process using an experimental design approach
Int. J. Mach. Tools Manuf
(2005) - et al.
Bead modelling and implementation of adaptive MAT path in wire and arc additive manufacturing
Robot. Comput. Integr. Manuf.
(2016) - et al.
Artificial neural networks and support vector machines with genetic algorithm for bearing fault detection
Eng. Appl. Artif. Intell.
(2003) - et al.
Practical selection of SVM parameters and noise estimation for SVM regression
Neural Netw.
(2004)
Online SVM regression algorithm-based adaptive inverse control
Neurocomputing
Forecasting electricity consumption: a comparison of regression analysis, neural networks and least squares support vector machines
Int. J. Electr. Power Energy Syst.
Industrial information integration—a literature review 2006–2015
J. Indus. Inform. Integr.
The effects of forced interpass cooling on the material properties of wire arc additively manufactured Ti6Al4V alloy
J. Mater. Process. Technol.
Modeling of bead section profile and overlapping beads with experimental validation for robotic GMAW-based rapid manufacturing
Robot. Comput. Integr. Manuf.
Financial time series forecasting using support vector machines
Neurocomputing
Application of support vector machines in financial time series forecasting
Omega (Westport)
Fast cross-validation algorithms for least squares support vector machine and kernel ridge regression
Pattern Recognit
Weld appearance prediction with BP neural network improved by genetic algorithm during disk laser welding
J. Manuf. Syst.
3D printing for mathematical visualisation
The Mathematical Intelligencer
Cited by (56)
A comprehensive review of process planning and trajectory optimization in arc-based directed energy deposition
2024, Journal of Manufacturing ProcessesWire arc additive manufacturing: A review on digital twinning and visualization process
2024, Journal of Manufacturing ProcessesOnline distortion simulation using generative machine learning models: A step toward digital twin of metallic additive manufacturing
2024, Journal of Industrial Information IntegrationIn situ identification of laser directed energy deposition condition based on acoustic emission
2024, Optics and Laser TechnologyMulti-objective robust parameter optimization using the extended and weighted k-means (EWK-means) clustering in laser powder bed fusion (LPBF)
2024, Expert Systems with Applications
- #
The authors contributed equally to this work.