Improving the tensile strength of non-keyhole friction stir lap welding joint of 2024-T4 Al alloy by radial basis function neural network and improved particle swarm optimization algorithm

Abstract

The non-keyhole friction stir lap welding (N-KFSLW) technology assisted by the outer stationary shoulder and the inner upper half-thread rotating pin was proposed to obtain the welding joint without keyhole through one-time process. Choosing 2024 aluminum alloys as the research object, the formation, microhardness and tensile strength of N-KFSLW joint were investigated. The improved particle swarm optimization (IPSO) algorithm was newly developed and had the advantages of large convergence speed and strong search ability, by which the radial basis function (RBF) neural network was optimized to enhance its prediction accuracy. After that, the RBF and IPSO (IPSO-RBF) system was used to predict the joint strength and optimize the process parameters combination. The results showed that the lap joint had not only the SZ with the thickness almost equal to the thickness of upper sheet but also the cold lap with a very small height, thereby leading to the high tensile strength of joint. The optimized parameters of welding speed, rotating speed and pin type by the IPSO-RBF system were respectively 612 rpm, 80 mm/min, and upper half-thread pin, and the tensile strength of lap joint reached 11.88 kN/mm. The N-KFSLW technology assisted by upper half-thread pin provides an effective way to obtain the lap joint with high performance, and the IPSO-RBF system can be used to maximize the strength of welding joint.