HARDNESS OF DUCTILE CAST IRON ESTIMATED BY ARTIFICIAL NEURAL NETWORKS

The paper gives the results of application of artificial neural networks in determining hardness of ductile cast iron. Data for 147 melts were collected in a Croatian foundry. The error back-propagation algorithm was used to train the multilayer feed-forward network. The optimal size of the hidden neuron layer was selected by analysing error parameters in the testing data set. There were 8 input parameters: liquidus temperature, lowest eutectic temperature, the recalescence, solidus, graphite factor 1, graphite factor 2, cooling rate at solidus, and eutectoid temperature. A statistical analysis of errors in estimating hardness of ductile cast iron was made.