Using artificial neural networks to identify and optimise the key parameters affecting geometry of a GMAW fillet weld

摘要

Control of Gas Metal Arc Welding (GMAW) parameters is key to maintaining good quality and consistent fillet weld geometry. The external geometry of the fillet weld can be easily measured, however the internal geometry (i.e. penetration), which is critical in determining the structural integrity of the joint, is difficult to measure without destructively testing the workpiece. Consequently the most cost effective way to ensure adequate penetration is to maintain close control of the input parameters. Furthermore if we can demonstrate tight control of the parameters and interactions that affect the joint penetration then we can increase the confidence that sufficient penetration is being achieved.Previous studies have shown that the variation in set up parameters between welders and the guidance given by industry/suppliers can vary widely and in some cases be contradictory. Also in practice there are several characteristics of the manual/semi-automatic GMAW fillet weld process that are difficult to control (e.g. gun angle, travel angle and gap) but yet have an impact on the resultant geometry.This paper will document a programme of work which has used an Artificial Neural Network (ANN) to identify the parameters, and specific interactions that have an impact on the resultant fillet weld geometry. The variables that will be assessed in this paper will include current, voltage, travel speed, gun angle, travel angle. Further follow on studies will take place to understand the impact of gap, gas flow & nozzle diameters.