Prediction of Weld Bead Geometry in Chromium-Manganese Stainless Steel Gas Tungsten Arc Welded Plates Using Artificial Neural Networks

1. Introduction

Gas tungsten arc welding (GTAW) is used for welding hard to weld metals like aluminium, stainless steel, magnesium and titanium (Kumar & Sundarrajan, 2008). With the increased use of mechanized welding, the selection of welding process parameters and welding procedure must be more specific to ensure that the weld bead parameters of good quality are obtained at minimum cost with high repeatability (Gridharan and7 Murugan, 2008). GTAW quality is strongly characterized by the weld bead geometry. This is because weld bead geometry plays an important role in determining the mechanical properties of the weld (Tarang and7 Yang, 1998). Hence the input welding variables which influence the bead width must therefore be properly selected to obtain an acceptable bead width and hence a high quality joint. It is difficult to obtain analytical solution to predict depth of penetration and weld bead width. Costly and time consuming experiments are required to determine the optimum welding process parameters due to complex and non linear nature of the welding process. Therefore, a more efficient method is needed to determine the optimum welding process parameters. The 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 (Manikya Kanti et al., 2008). The advantage of neural networks is that the network can be updated continuously with new data to optimize its performance. The network has the ability to handle large number of input variables rapidly, filter noisy data and interpolate incomplete data (Manikya Kanti et al., 2008). Li et al. (2000) has proposed a neural network for on-line prediction of quality in GMAW. Kim et al. (2003) compared multiple regression and back propagation neural network approaches in modeling top bead height of multi pass GMAW. They reported that back propagation neural network was considerably more accurate than multiple regression techniques. Nagesh and Datta (2002) applied a back propagation neural network to predict weld bead geometry and penetration in Shielded Metal Arc Welding (SMAW). They reported that artificial neural networks are powerful tool for the analysis and modeling of weld bead geometry and penetration. The prediction of laser butt joint parameters using back propagation and learning vector quantization networks was done by (Jeng et al., 2000). The work piece thickness and welding gap were used as input parameters. The optimal focused position, acceptable welding parameters of laser power and welding speed, and welding quality were used as output parameters. They concluded that both networks are very useful in selecting welding parameters. Siva et al. (2009) optimized the weld bead parameters of nickel based over lay deposited by plasma transferred arc surfacing. The results showed that penetration is increased when the welding current is increased and decreased when travel speed is increased. Tarang et al. (1998) optimized the weld bead geometry in GTAW. They employed Taguchi method to formulate the experimental lay out and analyzed the effect of process parameters on weld bead geometry. Ghazvinloo et al. (2010) studied the effect of electrode to work angle, filler diameter and shielding gas type on weld penetration of HQ 130 steel joints produced by gas metal arc welding. They showed that increasing of electrode to work angle increased the depth of penetration and increase in filler diameter resulted in decrease in weld penetration. Gridharan and Murugan (2008) have investigated the pulse GTA welding process parameters for the welding of AISI 304L stainless steel sheets. They developed mathematical models by regression analysis to predict penetration, bead width and bead area. They concluded that weld bead parameters predicted by the models were found to confirm observed values with high accuracy. Rayes et al. (2004) studied the influence of various hybrid welding parameters on bead geometry. They conducted experiments on 316 L austenitic stainless steel work piece. They varied arc and laser power to study their influence on various bead dimensions. They found that arc power has a great influence on bead width. Ugur Esme et al. (2009) optimized weld bead geometry in TIG welding process using Grey relation analysis and Taguchi method. They investigated the multi – response optimization of TIG welding process for an optimal parametric combination to yield favorable bead geometry of welded joints using Taguchi method. They conducted sixteen experimental runs based on an orthogonal array of Taguchi method to optimize the objective function i.e. bead width. Thao and Kim (2009) developed an interaction model for predicting bead geometry for lab joint in gas metal arc welding process. They conducted experiments based on full factorial design with two levels of five process parameters to obtain bead geometry using a gas metal arc welding process. They developed linear, curvilinear and interaction models for the process and found that prediction capabilities of interaction models are reliable than the linear and curvilinear models. From the study it was inferred that welding voltage, arc current, welding speed and welding angle have large significant effects on bead width. III – Soo Kim et al. (2001) optimized bead width for multi pass welding in robotic arc welding process. They developed mathematical model for predicting bead width for multi pass welding by both neural networks and multiple regression methods. They found that both models are able to predict bead width with reasonable accuracy, but the neural network model performed better than the empirical model. Kim et al. (1996) compared experimental data obtained for weld bead geometry with those obtained from empirical formulae in GMAW. Correia et al. (2005) explored the possibility of using GA as a method to decide near optimal settings of a GMAW process. Their problem was to choose best values of three control variables on four quality responses i.e. deposition efficiency, bead width, depth of penetration and reinforcement inside a previous delimited experimental region. From their study they concluded that GA was able to locate near optimum conditions, with relatively small number of experiments.