Inverse prediction of Friction Stir Welding parameters using Artificial Neural Networks

Abstract

Friction Stir Welding has become an invaluable joining process in aerospace and automotive industry. It is often required that the independent input parameters (traverse speed, pin diameter, rotational speed etc.) in Friction Stir Welding (FSW) be predicted from response values such as tensile strength and hardness. This will enable the use of input parameters that gives the desired results. If this is attained, near optimal results can be achieved without use of many resources. It also allows the selection of the closest input parameters available on the machine. Artificial Neural Networks (ANN) have been successfully applied in determining the input parameters in Friction Stir Welded materials when given the tensile strength. This procedure is however problematic at times since there may be several combinations of input parameters that gives the same result.In this research ANNs were used to predict the input parameters required to give a tensile strength of 300, 340, and 345 MPa of an aluminium alloy AA6082-T6. The predicted speeds were rotational speeds of 532.7 rpm at a traverse speed of 11.8 mm/min to obtain a tensile strength of 300 MPa. For tensile strength of 340 and 345 MPa, 437.1 rpm at a traverse speed of 13.6 mm/min were predicted as the input parameters.