A neural network prediction for the graphite nucleation and morphology in modified cast irons

Abstract

The present research is an attempt to provide a possible way of predicting the characteristics of graphite in modified cast irons. Using a simple melt solidification thermal analysis and some data related to charge chemical composition it could be possible to make a relatively good prediction on two main characteristics of graphite; the number of particles and the morphology. The method involves a stochastic mathematical model, input and output data being related through back-propagation neural networks that were especially designed and trained. The result of this investigation consists on a set of contour diagrams that can predict for hypoeutectic modified cast irons the above mentioned characteristics as functions of holding time prior to pouring, the amount of eutectic recalescence, the level of sulphur addition and the residual magnesium content.The experiment was carried out using hypoeutectic cast iron due to its particular sensitiveness on graphite shape modification, being possible to complete the entire graphite shape morphology transition from nodular to flake for settled parameters like residual magnesium, level of sulphur addition or holding time.Last but not least, this paper can be an argument for a further involvement of neural networks in materials science engineering as a reliable investigation tool.