Optimization of Refill Friction Stir Spot Welded AA2024-T3 Using Machine Learning

Abstract

The Refill Friction Stir Spot Welding is an innovative spot like solid state process befitting of overlap joint configurations of similar and dissimilar materials. This process caught the interest and is rapidly growing in the aerospace sector due to its potential to substitute traditional mechanical fasteners, surpassing their mechanical performance while maintaining the so desired lightweight “rationale.” In the current study, process parameters, namely plunge depth, plunge time and rotational speed, are optimized in order to obtain the highest Ultimate Lap Shear Force (ULSF) of 2024-T3 Aluminum Alloy similar joints. The optimization campaign was carried out using a second order multivariate polynomial regression machine learning (ML) algorithm. The trained ML model was able to generalize and accurately predict the Ultimate Lap Shear Force on the holdout set, having a R2 of 88.0%. Moreover, the model suggested an optimum parameter combination (Rotational Speed = 2,310 rpm, Welding Time = 5.3 s and Plunge Depth = 2.6 mm) from which the predicted maximum ULSF was computed. Confirmation tests were carried out to evaluate the agreement between the predicted and the experimental values.

Exploring a novel chamfered tool design for short duration refill friction stir spot welds of high strength aluminium

“…Although comparable welding times (< 2 s) have been successfully employed for 5xxx and 6xxx series aluminium [9,[14][15][16], this has not been widely reported on for high strength aluminium alloys such as AA2024-T3. Several studies have achieved high quality AA2024 joints, although at the cost of welding times in excess of 5 s [5,10,17].…”

A State‐of‐Art Review on Prediction Model for Fatigue Performance of Welded Joints via Data‐Driven Method

“…[ 44–46 ] With reference to the development of data‐driven method in biology, medical, energy, and other industries, the analysis and prediction methods for fatigue performance based on data‐driven method can also be well applied in practice after further development. [ 47–51 ]…”

“…[44][45][46] With reference to the development of data-driven method in biology, medical, energy, and other industries, the analysis and prediction methods for fatigue performance based on data-driven method can also be well applied in practice after further development. [47][48][49][50][51] The fatigue performance analysis and prediction methods based on data-driven methods can be divided into four main aspects: data acquisition, data preprocessing, data analysis, and prediction. [52] Therefore, the process of data acquisition and preprocessing are presented in Section 2.…”

Neural Networks Applied for Predictive Parameters Analysis of the Refill Friction Stir Spot Welding Process of 6061-T6 Aluminum Alloy Plates

“…The study resulted in Effertz PS. et al [14] aimed to optimize the refill friction stir spot welding process parameters using a multivariate polynomial regression (MPR) machine learning algorithm. The study found that the model exhibited significant dependence on the quadratic features of the studied parameters, with welding time and rotational speed having a detrimental effect on the ultimate lap shear force (ULSF).…”