Estimation of Dependent Parameters of Saw Process

This paper is an attempt to develop a model to predict the output responses of Submerged Arc Welding (SAW) process with the help of neural network technique. Also a mathematical model has been developed to study the effects of input variable (i.e. current, voltage, travel speed) on output responses (i.e. reinforcement height, weld bead width, metal deposition rate). SAW process has been chosen for this application because of the complex set of variables involved in the process as well as its significant application in the manufacturing of critical equipments which have a lot of economic and social implications. Under this study the neural network model is trained according to the actual inputs and outputs. When the training is completed then the desired inputs are given to the model and it gives the estimated output value. And according to this we can also estimate the error between the actual and predicted results. Neural network is implemented here because of having remarkable ability to derive meaning from complicated or imprecise data and can be used to extract patterns and detect trends that are too complex to be noticed by either humans or other computer techniques. Hence a trained neural network can be thought of as an “expert” in the category of information it has been given to analyses. quality and to evaluate the interaction effect of bead parameters and to optimization the bead quality to obtain a high quality welded joint at a relatively low cost with productivity. The most common neural network model is the multilayer perception (MLP). This type of neural network is known as a supervised network because it requires a desired output in order to learn. The goal of this type of network is to create a model that correctly maps the input to the output using historical data so that the model can then be used to produce the output when the desired output is unknown. Set. No. Travel Speed (Cm/Min) Voltage (In Volts) Current (In Ampere) Reinforcement Height (Mm) Bead Width (Mm) Metal Deposition Rate (Gm/Sec) 1 25 25 300 3.2 13 0.81 2 28 26 320 2 12.5 0.909 3 24 28 280 2.6 13 0.759 4 27 24 300 3.5 13.6 0.55 5 20 20 400 5 11.7 1.5 6 16 20 280 1.7 10 1.25 7 17 19 380 2 12 0.909 8 21 38 450 2.6 21.9 0.792 9 21 30 320 2 15.5 0.681 10 17 24 300 2.3 13.8 0.869 11 16 24 280 1 16.3 1 12 12 23 300 0.6 23 0.877 13 17 22 320 2.3 18.7 0.869 14 20 25 340 2.4 13.8 1.224 Table 1: Experimental Data. Journal of Applied Mechanical Engineering J o u r n al o f A pp lied ical Eninee r i n g


Introduction
Submerged arc welding (SAW) is commonly used for joining plate because of high deposition rate welding process. Here an arc is formed between a continuously fed wire electrode and the work piece, and the weld is formed by the arc melting the work piece and the wire. To shield the arc from atmospheric contamination, the arc is completely submerged under a blanket of granular, fusible flux.
During welding the intense heat of the arc simultaneously melts the tip of the bare wire electrode and a part of the flux. The electrode tip and the welding zone are always surrounded and protected by molten flux, while all of them are covered by the top layer of unfused flux. As the arc progresses along the joint, the molten metal settles down while the lighter molten flux rises from the puddle in the form of slag. The weld metal, having a higher melting point solidifies first while the slag above it takes some more time to freeze. The solidified slag continues to protect the weld metal while it is still hot, and is capable of reacting with the atmospheric oxygen and nitrogen. After the weld has solidified, the unfused flux is removed manually or by a vacuum pick up system, to be screened and reused [1].
The process can be fully automatic or semi-automatic. The submerged-arc welding (SAW) process is popular because of its ability to match the chemistry and physical properties of the base material. This ability allows a multitude of possible weld-wire and flux combinations. These combinations can be easily sorted and matched to specific applications. SAW process is one of the major fabrication processes in industry because of its inherent advantages, including deep penetration and a smooth bead. In this parameters yield characteristics of SAW process have been predicted through Neural Network [2], which help to reduce the cost and time as well as to obtain the required information about the main and interaction effects on the yield characteristics.
This model is useful for selecting correct process parameters to predict desired weld bead parameters so this model facilitate optimization of the process and sensitivity analysis, also help to improve the understanding of the effect of process parameters on bead

Abstract
This paper is an attempt to develop a model to predict the output responses of Submerged Arc Welding (SAW) process with the help of neural network technique. Also a mathematical model has been developed to study the effects of input variable (i.e. current, voltage, travel speed) on output responses (i.e. reinforcement height, weld bead width, metal deposition rate). SAW process has been chosen for this application because of the complex set of variables involved in the process as well as its significant application in the manufacturing of critical equipments which have a lot of economic and social implications. Under this study the neural network model is trained according to the actual inputs and outputs. When the training is completed then the desired inputs are given to the model and it gives the estimated output value. And according to this we can also estimate the error between the actual and predicted results. Neural network is implemented here because of having remarkable ability to derive meaning from complicated or imprecise data and can be used to extract patterns and detect trends that are too complex to be noticed by either humans or other computer techniques. Hence a trained neural network can be thought of as an "expert" in the category of information it has been given to analyses.
quality and to evaluate the interaction effect of bead parameters and to optimization the bead quality to obtain a high quality welded joint at a relatively low cost with productivity.
The most common neural network model is the multilayer perception (MLP). This type of neural network is known as a supervised network because it requires a desired output in order to learn. The goal of this type of network is to create a model that correctly maps the input to the output using historical data so that the model can then be used to produce the output when the desired output is unknown.

Experimental Procedure
The output variable was decided based on the guidelines in the reference material. The bead parameters were considered for measuring the response output. Three output parameters considered are (figure 1): P-penetration (mm) H-reinforcement height (mm)w, width(mm).

Sample preparation for the experiment
Now the first thing for conducting the experiment was to prepare the sample for experiment. For this purpose a long mild steel piece was selected and was brought to the welding shop for cutting it to the required dimensions and shape. The mild steel piece was having thickness more than 10 mm. Now here with the help of oxy-acetylene flame it was cut in the desired length. After cutting it into pieces it was left for cooling. After few hours of cooling the pieces were welded in pairs so that they remain together during the experiment. This was done with the help of welding. These samples were having a V-groove over which submerged arc welding was to be done [3]. Now the initial weight of each sample was taken and noted. Also the Submerged Welding Machine was started. Flux was filled in the machine so that there may be no problem in between the experiment.
The machine was fully automatic. Now the welding was done on each sample. The welding parameters travel speed, current [4], voltage were varied each time. Total time taken during the welding was noted and weight after each welding was also taken.

Conducting the experiment
This experimental study was conducted at ISM workshop (Dhanbad). For this study MEMCO semiautomatic welding equipment with constant voltage rectifier was used. The welding parameters were noted during actual welding to determine the fluctuations if any. The slag was removed and the job as allowed to cool down. The values of the reinforcement height, and width were taken using venire caliper of least count 0.02 mm. And metal deposition rate was also calculated with the help of stop watch and weighing machine. Finally the readings obtained are tabulated in table 1

Alternative approach-multi regression method
Reinforcement height, Weld bead width, Metal deposition rate are denoted by RH, W and MDR for this purpose.

Result and Discussion
Prediction of weld bead parameters Model was developed using neural network. This model was first trained with the help of actual experimental data. And when the training is completed then it is used to predict the values of the output on the same input data to estimate the percentage of error between the experimental result and the result predicted with the help of neural network model.
Also mathematical model has been developed which gives 95% correct result (Table 4). From these two models, we can get a comparative idea of predicted output responses of saw process.

Conclusion
From table 1 we can say that travel speed has a negative effect on all the three bead parameters. An increase in travel speed substantially reduces the heat input resulting in lower burn off rate. This reduced burn off rate decreases the metal deposition at the weld joint thereby lowering all the bead parameters. It has been seen that the penetration increases with increase in the current, the reinforcement height also increases marginally with an increase in the current but the width decreases i.e. it has a negative effect. It is generally observed that as the arc voltage increases, the weld bead becomes wider and flatter and the penetration decreases .Also understood that there are interactions present among the input parameters for output parameters. A soft computing model was developed in C with the help of neural network. In which we first feed our actual experimental values which includes both inputs and outputs. And on the bases of these values the model is train. And when the training is completed then the desired value of inputs parameters is given to the model and it gives the predicted value of the outputs according the trend. On the bases of these actual and predicted values we can estimate the optimum value of the welding parameters. Also we can estimate the % of error (Table  2). Side by side a mathematical model, comparison of the predicted and observed values of bead parameters revealed the models' accuracy above 95% (Table 4), has been developed by which we can also predict output responses of saw process.