Reliability assessment of friction stir welds in AA100 aluminium alloy using ANN and ANFIS predictive models

Abstract

Friction Stir Welding (FSW), a solid-state welding process, is used to weld AA 1100 alloy plates together, varying weld parameters such as spindle speed, feed rate, and axial load. The Taguchi L27 orthogonal array is employed to study the effects of multiple factors with minimal test runs. Tensile tests are conducted on welded samples to determine the time of failure. Using the Signal to Noise (S/N) Ratio to maximise the tensile strength, the influencing parameters and their effect are determined. Weibull analysis is used to calculate weld reliability. Soft computing techniques, such as Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Artificial Neural Networks (ANN), have been developed to predict weld reliability based on parameters. The ANFIS model demonstrates superior performance, with an average percentage error of 4.61% compared to 19.95% for ANN and a Root Mean Square Error (RMSE) of 0.002 versus 0.008 for ANN. Optimal FSW parameters for maximising joint reliability are 1200 rpm spindle speed, 18 mm/min feed rate, and 7 kN axial load, resulting in a predicted reliability of 0.87. The study highlights the potential of ANFIS as an effective tool for predicting and optimising FSW joint reliability in AA 1100 aluminium alloys.

Keywords: Friction stir welding, DOE, Soft computing, ANFIS and ANN models, S/N ratio