Artificial neural networks to predict of liquidus temperature in hypoeutectic Al-Si cast alloys

Abstract

Determining the liquidus temperature of cast alloys is an important factor in considering the superheating temperature and melt treatment of aluminium-silicon cast alloys. In addition to experimental calculation, the liquidus temperature can also be determined using simulation software for more reliable results. In this study, Artificial Neural Network (ANN) with hyperbolic tangent was selected to predict the liquidus temperature of Al-Si alloys as a function of chemical composition. The neural network was trained with seven input parameters (Si, Fe, Cu, Mn, Mg, Zn and Ti) and one output parameter (liquidus temperature). Training and testing dataset has been chosen from different published works, any casting software and aluminium binary phase diagrams. The accuracy of neural network was verified using values reported in literatures. The result of this investigation has shown that the backpropagation feed forward neural network is accurate enough to predict liquidus temperature.