Comparison of ANN Training Algorithms for Predicting the Tensile Strength of Friction Stir Welded Aluminium Alloy AA1100

Abstract

Aluminium alloy AA1100 finds application in light weight structures due to its high strength to weight ratio. Friction stir welding is a solid state welding process, in which the materials are joined in the plasticized state. The quality of the friction stir welded joints depends on the process parameters used and tool parameters. In this study, four process parameters were varied at five levels and experimental trials were performed as per face centered central composite design. Artificial neural network model was developed with cascade forward propagation network architecture and trained with LM algorithm and BFGS QN algorithm. The models were used to predict the tensile strength of the joints and the error in prediction was used to judge the accuracy of the developed models. It is observed that BFGS QN algorithm trains the ANN efficiently and results in accurate predictions.