Prediction and Control of Asymmetric Bead Shape in Laser-Arc Hybrid Fillet-Lap Joints in Sheet Metal Welds

The shape and size of a weld bead - consisting of outer weld surface and inner fusion boundary - are important quality and strength attributes in sheet metal welds. The asymmetricity coupled with additional controlling parameters makes it challenging to predict the bead shape in laser-arc hybrid fillet-lap joints with use of lower order nonlinear analytical mathematical functions. An artificial neural network is designed to address the challenge, considering the welding speed, wire feed speed, voltage, current, and laser power as inputs. The experimentally obtained weld bead profiles are digitized in polar coordinates (r, θ) and thereby many input-output pairs are made available for training even with a limited number of experiments. An optimized neural network topology is presented with an assessment of reliability of simulation results. A rational approach for determining the number of coordinate points needed to accurately map the weld bead profile is an important contribution from the present investigation. The parametric study elucidates the effects of input parameters on geometry of the weld beads. The neural network exhibits the capability of capturing the process physics - demonstrated through the analysis of the weld dilution obtained from the simulation results. The welding speed and wire feed speed signifyingly affect the bead shape while the laser power has a minor impact. The laser, even though with less power, improves the weld dilution due to preheating of the base plate and stabilization of the welding arc.