STUDY ON UNDERWATER FRICTION STIR WELDED AA 2024-T3 PIPES USING MACHINE LEARNING ALGORITHMS

Underwater friction stir welding, a new variant of friction stir welding process in which the weld coupons and tool-specimen interface are completely immersed in the water, has been successful to achieve wide popularity among researchers recently. In most of the studies, the underwater friction stir welding process is limited to join the flat plates. The research conducted on the underwater friction stir welding of pipes is rare due to the complexity in the design of the fixture and setup. Therefore, the current research is aimed to investigate the scope of underwater friction stir welding process for producing quality welded pipe joints. Initially, the current research focused on developing a system with proper components and fixture attached to the vertical milling machine for underwater friction stir welding of pipes. Twenty-seven experiment runs with three intermittent levels of process parameters - spindle speed of milling machine (1000 rpm, 1400 rpm, 1800 rpm), travel speed (10 mm/min, 16 mm/min, 20 mm/min) and shoulder diameter of tool (10 mm,15 mm, 20 mm) are designed. Secondly, Al 2024-T3 pipes having an outer diameter of 30 mm and a thickness of 3 mm are welded using an underwater friction stir welding process for every combination of the process parameter. The elongation percentage, yield strength and tensile strength are experimentally evaluated from the tensile tests. Finally, the prediction capability of machine learning algorithms such as artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS) and adaptive neuro-fuzzy inference system with Harris hawks optimization (ANFIS-HHO) for 70% training data and 30% testing data was evaluated. The prediction capability of the machine learning algorithms was evaluated using the Mean Absolute Error, R² statistic and Root Mean Square Error. ANN was found to the best with the highest R² and least RMSE for predicting all three responses. Though the ANFIS exhibited the highest R² and highest RMSE for every response, the incorporation of Harris hawks optimization to the ANFIS slightly improved the prediction capability of ANFIS. The prediction accuracy for elongation percentage, yield strength and tensile strength is found to be in the increasing order of ANFIS, ANFIS-HHO and ANN. The underwater friction stir welding process, machine learning algorithms, methods and results discussed in the paper are promising and useful for experts in the industries.