Data-driven optimization of friction stir welding parameters of AA7075 aluminium alloy using ANN surrogates and genetic algorithms

Abstract

Background

This study proposes a data-driven hybrid framework for optimizing the process parameters of friction stir welding (FSW) of AA7075 aluminium alloy by integrating artificial neural network (ANN) surrogate models with genetic algorithm (GA) optimization.

Methods

A Design of Experiments (DoE) approach comprising 27 systematically designed trials was employed to investigate the combined influence of rotational speed, welding speed, plunge depth, and shoulder-to-pin diameter ratio on the ultimate tensile strength (UTS) and microhardness of FSW joints. The experimental data were used to train and validate ANN surrogate models capable of capturing the nonlinear and interdependent relationships between the process variables and mechanical responses.

Results

The developed ANN models demonstrated high predictive accuracy, with R2 values exceeding 0.94 and 0.96 for ultimate tensile strength (UTS) and hardness, respectively, accompanied by low root mean square error (RMSE) and mean absolute error (MAE) values. Experimental validation confirmed the robustness of the models, yielding prediction errors ranging from 0.18% to 11%, with an average deviation of 3.6%. Coupling the trained ANN surrogate with a GA-based optimizer enabled the identification of parameter combinations that simultaneously maximize UTS and hardness. The optimal UTS of 345.76 MPa was obtained at a rotational speed of 814.13 rpm, welding speed of 51.53 mm/min, plunge depth of 0.213 mm, and a shoulder-to-pin ratio of 3.6, whereas the maximum hardness of 143.41 HV was achieved at lower rotational and welding speeds, combined with a higher plunge depth and shoulder-to-pin ratio.

Conclusion

The findings demonstrate that the ANN–GA hybrid approach provides a robust, accurate, and computationally efficient tool for predictive modelling and process optimization in FSW. This framework offers valuable potential for Industry 4.0 and intelligent manufacturing applications, particularly in the design of lightweight, high-performance aluminium structures.