Heat Source Modeling for FCAW of Fillet Joints using Artificial Neural Networks

Abstract

The present work deals with the prediction of heat source parameters of Goldak's double ellipsoidal model for flux cored arc welded fillet joints through an artificial neural network (ANN). Extensive experiments were carried out on low-carbon mild steel plates of thickness ranging from 3 mm to 10 mm. In each case, welding current (I), voltage (V), and speed (υ) were recorded as the welding parameters. The flux cored arc welding of fillet joint was carried out using 100% CO2 as the shielding gas. This work is based on the simple and fundamental presumption that the weld pool is the primary source of heat which is delivered from the welding arc. Therefore, the magnitude and distribution of heat can be defined through the weld pool dimensions. Hence, these dimensions were directly considered for determining the heat source ellipsoid parameters. These parameters, namely, front double ellipsoidal length (Cf), rear double ellipsoidal length (Cr), half-width of the heat source (a), and depth of the heat source (b) were measured from the welded specimens. Furthermore, the ANN model was trained and tested for the prediction of the heat source parameters. The ANN predicted data were in good agreement with the measured values. Subsequently, the ANN model was used to predict the parametric dimensions of the double ellipsoidal heat source model for a set of welding parameters. The temperature history numerically computed with the ANN-predicted parametric dimensions of the double ellipsoidal heat source model compared very well with the measured data. The proposed ANN structure, therefore, can be gainfully used as a tool for the prediction of parametric dimensions of the double ellipsoidal heat source model in case of gas metal arc welding with 100% CO2 as the shielding medium for a plate thickness range of 3–10 mm. 1. Introduction Flux cored arc welding (FCAW) is extensively used in various fabrication industries, including the ship building industry. It is widely used in semiautomatic mode for fabrication of, among other things, stiffened flat and curved panels. These are fabricated by butt welding of the plates followed by fillet welding of the longitudinal and transverse stiffeners. In thin panel fabrication, it may lead to weld induced buckling. Hence, to assess the possible behaviour during fabrication of the stiffened panels, numerical simulation is often carried out. The computation of temperature field is at the heart of numerical simulation of weld-induced distortion and residual stress. Thus, it is very important to have a model for accurate prediction of the temperature field. The temperature field again depends on the distribution and quantum of heat that gets delivered from the welding arc to the weld joint. This calls for developing a heat source model that represents the heat distribution and its magnitude as accurately as possible.