Abstract: Using welding tools with different pin diameter, the butt-welding tests of 2024-T4 aluminum alloy with a plate thickness of 3 mm were conducted via friction stir welding, the effects of pin diameter on the surface morphology and the tensile properties of the welded joints were studied. Based on the acquisition of experimental data, Logistic chaotic mapping and adaptive weights were introduced to optimize the whale algorithm, and then the improved whale optimization algorithm (LIWOA) was utilized to optimize the BP neural network. The tensile strength prediction model of FSW joints was established by using LIWOA-BP neural network, and the tensile strength prediction system of FSW joints was designed by using mixed programming of C# and Matlab. The results show that the tensile strength of the joint shows a trend of increasing and then decreasing with the increase of the pin diameter, and the tensile strength of the joint reaches 325.54 MPa when the stirring pin diameter is 4 mm. The average absolute percentage error(MAPE) of the LIWOA-BP neural network-based FSW tensile strength prediction model is 1.26%, which is a high prediction accuracy. The FSW joint tensile strength prediction system realizes the visualization of prediction, which is conducive to the regulation of friction stir welding process parameters and provides a new idea for the process optimization of FSW.