An analysis of formability of aluminium preforms using neural network

Cold upsetting experiments were carried out on sintered aluminium preforms in order to model and analyze the formability of aluminium preforms by using Neural Network (NN). A NN model has been explored with a radial basis neural network algorithm. The training data used was obtained from the experimental set up in the laboratory for the sintered aluminium with various preforms densities and different aspect ratios by using MoS2 as lubricant. The network was trained to predict the formability index (β), triaxial stress ratio parameters and axial strain. It is deduced that a lower aspect ratio and higher initial fractional density preforms exhibits improved formability index values compared to that of higher aspect ratio preforms. In addition to that, instantaneous stress ratio parameter coefficients Bi and Ai of the aluminium preforms were mathematically evaluated and simulated. It is found that it increases rapidly at the earlier stages of deformation and followed by a gradual increase with further increase in true axial strain. Regression analysis has confirmed a good coincidence between predicted and experimental data with small error and hence this approach helps to facilitate a knowledge base in order to generate advice for the designer at the earlier stages of design in the metal forming industries by forecasting and preventing the surface defects at the computer design stage itself and avoids the cost of prototyping or trailing.