CAFE modeling, neural network modeling, and experimental investigation of friction stir welding

Abstract

The main aim of the present work is to develop a cellular automata finite element –artificial neural network (CAFE-ANN) hybrid model to predict the evolution of grain size and yield strength during friction stir welding. The CAFE model was developed by linking CA cells with elements in ABAQUS, a finite element code. Then a neural network was developed and trained appropriately by using the data obtained from the validated CAFE model that predicts the grain size and yield strength. The ANN results are validated. Finally it has been demonstrated that ANN can be used as a ‘virtual machine’ to examine the effect of rotational speed, welding speed, axial force and shoulder diameter on the variation in grain size and yield strength. The temperature and strain-rate distribution predicted by the thermal and strain-rate models are consistent with the existing results. The CAFE model grain size predictions are also accurate as compared to experiments. The grain size and yield strength predictions from ANN coincide well with the experimental values and CAFE model predictions. It has been demonstrated that CAFE-ANN hybrid model can be used as a ‘virtual machine’ to predict and analyze the effect of above said parameters on the grain size and yield strength evolution. The predicted influence is found to agree with some of the available results. In few cases, a slight deviation in the trend is witnessed as compared to literature results.