Determination of the beta-approach curve and beta-transus temperature for titanium alloys using sensitivity analysis of a trained neural network

A feed-forward neural-network (FFNN) technique with a back-propagation-learning algorithm was used to estimate the beta-approach curve and beta-transus temperature for alpha/beta titanium alloys. The input parameters were the alloy composition (Al, V, Fe, O, and N) and heat-treatment temperature, and the output parameter was the beta-phase volume percentage. The model was trained using selected data from the literature as well as new measurements. The trained model was used to predict the beta-phase volume percentage for the remaining data and to perform a sensitivity analysis to estimate the beta-transus temperature for other titanium alloys. The sensitivity analysis showed that a trained neural network can be used for extrapolated predictions (outside the range of measurements) unlike previous neural-network techniques used primarily for interpolation or approximation. Comparisons between model predictions and experimental data indicated that the NN model thus holds promise for estimating the beta-transus temperature of titanium alloys.