engleski

Application of neural networks in microstructure prediction of aluminum DC cast ingots

Artificial neural network (ANN) methodology has been used in the present study in order to predict microstructure (i.e. grain size) in DC cast aluminum ingots. A feed forward ANN model with the resilient back-propagation (Rprop) learning algorithm and weight decay regularization has been developed to relate the grain size (in AA1200 and AA5754 aluminum alloy ingots) to casting speed, meniscus level, casting temperature, cooling water flow and speed of wire for master alloy AlTi5B1 addition. The ANN was trained on data measured in the real industrial conditions during the vertical DC process of casting aluminum ingots, to be able to describe complex dependencies between the microstructure and casting parameters. After the ANN training, a computer simulation has been conducted for the purpose of showing how grain size in Al-ingots depends on the particular casting parameter. The results of simulations show satisfactory agreement with the theoretical researches and practical experience from the foundry. Eventually the optimization has been conducted in order to find the optimal values of casting parameters for achieving the finest grain size in microstructure of AA1200 and AA5754 aluminum alloy ingots.

Grain size prediction ; AA1200 ; AA575 ; aluminum ingots ; Simulation ; Neural networks ; DC casting

nije evidentirano