Approach to constitutive relationships of a Ti-5AI-2Sn-2Zr-4Cr-4Mo alloy by artificial neural networks

Abstract

In the present paper, artificial neural networks (ANNs) have been applied to acquire the constitutive relationships of a Ti-5Al-2Sn-2Zr-4Cr-4Mo (wt-%) alloy at elevated temperature, using the data obtained from experiments carried out on a Thermechmastor-Z hot simulator. In establishing the neural network model for the constitutive relationship of the present alloy, deformation temperature, equivalent strain rate, and equivalent strain, were taken as the inputs, flow stress was taken as the output, and three neurons were used in the hidden layer. The activation function in the output layer of the model obeyed a linear function, while the activation function in the hidden layer was a sigmoid function. The neural network became stable after 32 500 repetitions in training. Comparison of the predicted and experimental results shows that the ANN model used to predict the constitutive relationship of the Ti-5Al-2Sn-2Zr-4Cr-4Mo alloy has good learning precision and good generalisation. The neural network methods are found to show much better agreement than existing methods with the experimental data, and have the advantage of being able to treat noisy data or data with strong non-linear relationships.