An integration of RSM and ANN modelling approach for prediction of FSW joint properties in AA7178/AA5456 alloys

The selection of optimal process parameters in Friction stir welding is vital for achieving high-quality welds. This study utilises Response Surface Methodology (RSM) to statistically analyze the influence of process parameters on the AA7178/AA5456 joint. The study focuses on weld qualities such as tensile strength (TS), elongation (EL), and nugget zone hardness (HN). Five significant process parameters are considered: Tool Rotational speed (TOD), Welding speed (WED), Tool tilt angle (TTE), Pin depth (PDT), and Tool Pin profile (TPP). Through RSM, the optimal experimental conditions were determined, resulting in the highest tensile strength. These conditions include TOD (1200 RPM), WED (300 mm/min), TTE (3 degrees), PDT (0.5 mm), and TPP (straight pin). To enhance the prediction accuracy of tensile strength and hardness, an Artificial Neural Network (ANN) model is developed using the Levenberg-Marquardt (LM) algorithm. The optimised ANN with a feed-forward structure (5-10-3), demonstrated high precision with an R2 value of 0.98902 and Mean Squared Error (MSE) within the range of 0.3754 for the experimental dataset. The regression analysis confirmed the ANN model's ability to accurately predict mechanical properties. Additionally, the joint strength and microstructure are assessed using Optical microscopy and Scanning Electron Microscopy (SEM) respectively.