Neural network modelling of flow stress in Ti-6Al-4V alloy with equiaxed and Widmanstatten microstructures

Abstract

In the present study, artificial neural networks (ANNs) were used to model flow stress in Ti-6Al-4V alloy with equiaxed and Widmanstatten microstructures as initial microstructures. Continuous compression tests were performed on a Gleeble 3500 thermomechanical simulator over a wide range of temperatures (700-1100 degrees C) with strain rates of 0.001-100 s(-1) and true strains of 0.1-0.6. These tests have been focused on obtaining flow stress data under varying conditions of strain, strain rate, temperature, and initial microstructure to train ANN model. The feed forward neural network consisted of two hidden layers with a sigmoid activation function and backpropagation training algorithm was used. The architecture of the network includes four input parameters: strain rate (epsilon) over dot, temperature T, true strain epsilon and initial microstructure and one output parameter: the flow stress. The initial microstructure was considered qualitatively. The ANN model was successfully trained across (alpha + beta) to beta phase regimes and across different deformation domains for both of the microstructures. Results show that the ANN model can correctly reproduce the flow stress in the sampled data and it can predict well with the non-sampled data. A graphical user interface was designed for easy use of the model.