Prediction of bead geometry in pulsed GMA welding using back propagation neural network
Introduction
The GMAW-P process has recently gained wide attention in the welding industry, owing to its comparatively low heat input and precise control over the thermal cycle. This is because, with the pulsed current gas metal arc welding process, spray transfer or more precisely controlled droplet transfer is obtained at a low average current. This provides a smaller and well-controlled weld pool, which allows cladding of thin materials in all positions.
The superiority of the GMAW-P process is mainly related to the nature of the metal deposition, which is governed by the pulse parameters, namely peak current, pulse frequency, wire feed rate, wire feed rate/travel speed ratio, etc. Bead geometry in the arc welding process is an important factor in determining the mechanical characteristics of the weld (Srinivasa Rao, 2004, Srinivasa Rao et al., 2005, Rajasekaran, 1999, Senthil Kumar and Parmer, 1986). Bead geometry variables, such as bead width, bead height, and penetration depth, are greatly influenced by welding process parameters such as wire feed rate, welding current, welding speed, plate thickness etc. Weld shape and size are represented by bead width (W), bead height (H) and depth of penetration (P) as shown in Fig. 1. Convexity index (CI) is defined as the ratio of bead reinforcement height to the bead width. Selection of the most suitable combination of pulse parameters is very difficult owing to the complex interdependence of the above parameters on the pulsed current (Srinivasa Rao, 2004, Srinivasa Rao et al., 2005).
However, costly and time-consuming experiments are required in order to determine the optimum welding process parameters due to the complex and nonlinear nature of the welding process. Therefore, a more efficient method is needed to determine the optimum welding process parameters.
Technique of neural networks offers potential as an alternative to standard computer techniques in control technology and has attracted a widening interest in their development and application (Rajasekaran, 1999). The advantages of the neural networks is that the network can be updated continuously with new data to optimize its performance at any instance, the networks ability to handle a large number of input variables rapidly, and the networks ability to filter noisy data and interpolate incomplete data.
The back propagation network (BPN) system is one of the family of artificial neural network techniques used to determine welding parameters for various arc welding processes The network is a multi layer network that contains at least one hidden layer in addition to input and output layers. Number of input layers and number of neurons in each hidden layer is to be fixed, based on the application, the complexity of the problem, and the number of inputs and outputs. Use of nonlinear log-sigmoid or tan-sigmoid transfer function enables the network to simulate non-linearity in practical systems. Due to its numerous advantages, back propagation network is chosen for present work (Kim et al., 2001, Kim et al., 2002, Kim et al., 2005, Chan et al., 1999, Lee and Um, 2000, Krose and Van der Smagt, 1996).
Section snippets
Feed forward back propagation network
Back propagation learning is a supervised learning where it needs to know the inputs and the desired outputs in advance. This network is well established as a method for data mapping. In this work, the welding parameters are mapped to the weld dimensions through the internal representation of BPNs. Data obtained from the experiments and regression models are provided to a network at the learning stage, e.g., welding parameters and weld bead geometry dimensions. During network learning, the
Process parameters used in the study
Five input process parameters, plate thickness (A), pulse frequency (B), wire feed rate (C), wire feed rate to travel speed ratio, i.e. WFR/TS ratio (D) and peak current (E) are used in the present study to predict two output parameters, the depth of penetration (PD) and the convexity index (CI). Experiments have been carried out for different combinations of inputs and the bead penetration depth and convexity index have been found (Srinivasa Rao et al., 2005).
The process parameters included in
Network experimentation
In this study, an attempt is made to construct a single multi output BPN model to predict both weld dimensions with one network, as it is believed that the welding parameters and weld dimensions are interrelated in such a way that the solution should always be considered as a set (Manikya Kanti, 2006). Therefore solving all with one network is a more logical approach in this case. The neural network development and the training are carried out using MATLAB 6.1 application tool.
A schematic
Results
A back propagation neural network (BPN) model to predict bead geometry is developed in this work. To ensure the accuracy of the BPN model developed to predict bead penetration depth and the convexity index, the experimental results and the predicted results using the developed BPN model are compared. A correlation coefficient is calculated to measure the relationship between experimentally measured output and the output predicted by the BPN model. However, a high correlation coefficient is not
Conclusions
The effects of process parameters on bead penetration depth and convexity index in pulsed GMA welding using artificial neural networks has been studied, and the following conclusions have been reached:
- •
The neural network model developed in this work from the experimental data (for bead penetration depth and convexity index) can be employed to control the process parameters in order to achieve the desired bead penetration depth and convexity index.
- •
It is observed from the results that a
References (16)
- et al.
Modeling gas metal arc weld geometry using artificial neural network technology
Can. Metall. Quart.
(1999) - et al.
A study on prediction of bead height in robotic arc welding using a neural network
J. Mater. Process. Technol.
(2002) - et al.
An investigation into an intelligent system for predicting bead geometry in GMA welding process
J. Mater. Process. Technol.
(2005) - et al.
A prediction of welding process parameters by prediction of back bead geometry
J. Mater. Process. Technol.
(2000) - et al.
Neural Network Design, Thomson Learning
(1996) Working with neural networks
IEEE Spectrum
(1993)- et al.
Development of an intelligent system for selection of the process variables in gas metal arc welding processes
Int. J. Adv. Manuf. Technol.
(2001) - et al.
An Introduction to Neural Networks
(1996)
Cited by (111)
Conditional generative adversarial network for welding deformation field prediction of butt-welded plates
2023, Journal of Constructional Steel ResearchArtificial neural Network-Based approaches for Bi-directional modelling of robotic wire arc additive manufacturing
2022, Materials Today: ProceedingsThe first step towards intelligent wire arc additive manufacturing: An automatic bead modelling system using machine learning through industrial information integration
2021, Journal of Industrial Information IntegrationCitation Excerpt :The backward model is referring to the predictive model that is able to predict the welding parameters based on designated bead geometry. Although lots of studies [38–40] mentioned the method to predict the welding bead geometry and also achieved acceptable accuracies, there is currently no opportune backward model for forecasting the welding parameters by expected bead geometry, which has practical significance in industrial manufacturing. The backward model cannot be established by simply reusing the forward predictive model due to the dimension difference between the input and output parameters.
End-to-end prediction of weld penetration: A deep learning and transfer learning based method
2021, Journal of Manufacturing ProcessesCitation Excerpt :The artificial neural network (ANN) is one of the most widely intelligent algorithms which can be used in the prediction of the bead geometry and penetration problem [7]. And the back propagation artificial neural network (BPNN) is applied on predicting penetration depth, back-bead width and the bead geometry size [8,9]. However, all these methods have the issue that the prediction accuracy is not high often due to the missing of some characteristic information.
Prediction of welding residual stresses using Artificial Neural Network (ANN)
2021, Materials Today: ProceedingsCitation Excerpt :This paper uses an ANN Regression model to predict residual stresses during the butt-welding process of Two mild steel plates. The ANN model was developed in an anaconda environment using libraries such as Keras and Scikitlearn [15]. The data collection was done by doing thermal and stress analysis using general-purpose finite element analysis software like ANSYS.