Abstract :
In a Friction Stir Spot Welding (FSSW) process, welding parameters (the
tool rotational speed, tool plunge depth, and stirring time) a ect the nugget formation
in high-density polyethylene (HDPE) sheets. The size and microstructure of the nugget
determine the resistance of the joint to outer forces. The optimization of these parameters
is vital to obtaining high-quality welds. Feed forward back-propagation arti cial neural
network models are developed to optimize the FSSW parameters for HDPE sheets. Input
variables of these models include tool rotation speed (rpm), the plunge depth (mm), and
the stirring time (s) that a ect lap-shear fracture load (N) output. Prediction performances
of 6 models in di erent speci cations are compared. These models di er in terms of the
training dataset used (80%-100%) and the number of neurons (5-10-20) in a hidden layer.
The best prediction performances are obtained using 20 neurons in a hidden layer in both
training dataset. There is good agreement between developed modelsʹ predictions and the
experimental data.
