Artificial Neural Networks (ANNs) for Prediction and Optimization in Friction Stir Welding Process

Abstract

Worldwide, researchers have been making many attempts to optimize process parameters and predict various properties for different manufacturing techniques, including friction stir welding (FSW). And this is achieved by using different optimization tools, including artificial neural networks (ANNs). Friction stir welding is a solid-state, environmental-friendly joining technique utilized to weld similar and dissimilar materials. FSW has successfully been used to join several materials, including aluminium, copper, and magnesium. Artificial neural networks (ANNs) are computational modelling tools and are inspired by the human or animal biological nervous systems. ANNs have widely been used in various engineering fields due to their captivating structures, including learning, faster computation, and easiness in the execution. This chapter provides an overview of the usage of the ANNs in friction stir welding to optimize the process and predict joint properties. Not many researchers have reported the usage of ANNs to optimize FSW process parameters such as rotation and traverse speeds. On the other hand, properties such as microhardness, surface roughness, and tensile strength have been predicted using ANNs. And this was achieved by using various algorithms, including backpropagation, gradient descent with momentum, and Levenberg–Marquardt algorithms. It was observed that the usage of ANNs to optimize process parameters in FSW and predict properties should be extended and the extensive use of ANNs should be applied to as many as possible joint configurations. This would reduce the need for conducting many experiments and the discarding of joints with extensive defects. Furthermore, the optimization of the FSW process parameters could extend its industrial applications and has a positive impact on the environment.