Assessment of friction stir spot welding of AA5052 joints via machine learning

Abstract

In this study, successful joints were fabricated on 4-mm-thick aluminum alloy 5052 sheets by using friction stir spot welding (FSSW) method. This research thoroughly investigated the impacts of welding parameters, specifically dwell time (DT) and rotational speed (RS), on the microstructure, and joint efficiency mechanical characteristics of the joints. The finding of this study highlighted the importance of optimization of process parameters to achieve superior weld joints. The most noteworthy achievement of this study was the attainment of maximum tensile shear load TSL of 2439 N with 19.4% joint efficiency at DT of 2 s and RS of 1300 rpm. A remarkable 48% improvement was observed in the obtained results at lower RS of 850 rpm and longer DT of 5 s. Simultaneously, maximum microhardness was 37.2 HV which was attained in thermal-mechanical affected zone at DT of 2 s and RS of 850 rpm, which was about 51% higher than the condition involving lower RS. Microstructure examination unveiled the significant influences of process parameters on hook deformation and penetration around the pin area. Additionally, in this study, a novel prediction model was introduced to estimate the temperature evaluation and tensile shear load of the samples. The model was constructed employing various machine learning techniques, multi-linear regression (MLR), support vector machine (SVM), adoptive neuro-fuzzy inference system (ANFIS) and including artificial neural network (ANN). The results obtained using this model served as a pioneering approach to predict the tensile shear load and temperature evaluation of welded samples. Remarkably, ANFIS model surpassed the other models due to its accuracy in perdition. The average error of this model for tensile shear load was only 4.3%, and for temperature evaluation, it was only 0.803%. The outcome of this study revealed that this predictive model could be a milestone in this field, enabling more precise and reliable prediction of key welding process parameters which significantly enhanced the efficiency and quality of welding processes.