Electron beam welding of aerospace alloy (Inconel 825): A comparative study of RSM and ANN modeling to predict weld bead area

Electron beam welding (EBW) is an energy intensive beam welding method widely used for joining hard to weld metals found application in aerospace, nuclear, automotive industries. This work investigates the influence of EBW parameters on a popular aerospace alloy (i.e., Inconel 825). Welding speed (S), beam current (I), accelerating voltage (V) and beam oscillation (O) are considered as EBW process parameters to study the weld bead area (WA) of the weldments. Predictive modeling is developed using two different approaches viz., statistical approach based response surface methodology (RSM) and soft computing based artificial neural network (ANN) and their model effectiveness is compared. Parametric analysis shows effect of EBW variables on WA. Welding speed found as the dominant process parameter followed by accelerating voltage and beam current; the maximum WA found at welding speed of 900 mm/min having beam current of 46 mA with an accelerating voltage in the range of 54-60 kV. ANN model developed with 3 layers with 11 hidden layer neurons is found optimum network architecture to predict WA of the weldments. Predictive performance of ANN and RSM models exhibited excellent agreement with confirmation data sets with mean percentage error less 5%. ANN model is found statistically robust and accurate in prediction of WA with model accuracy of 98 %.