Kovove Mater. 45 2007 41–49 41 The application of artificial neural network in the prediction of the as-cast impact toughness of spheroidal graphite cast iron Z. Glavaš 1 *, D. Lisjak 2 , F. Unki´c 1 1 Faculty of Metallurgy, University of Zagreb, Aleja narodnih heroja 3, 44103 Sisak, Croatia 2 Faculty of Mechanical Engineering and Naval Architecture, University of Zagreb, Ivana Luči´ca 5, 10002 Zagreb, Croatia Received 3 August 2006, received in revised form 6 September 2006, accepted 11 September 2006 Abstract This paper presents the application of artificial neural network (ANN) in the foundry process. Two-layer feedforward neural network which is trained using backpropagation al- gorithm that updates weights and biases values according to gradient descent momentum and an adaptive learning rate (Backpropagation Neural Network – BPNN) has been established to predict the as-cast impact toughness of spheroidal graphite cast iron (SGI) using the thermal analysis (TA) parameters as inputs. Generalization property of the developed ANN is very good, which is confirmed by a very good accordance between the predicted and the targeted values of as-cast impact toughness on a new data set that was not included in the training data set. K e y w o r d s: spheroidal graphite cast iron, impact toughness, artificial neural networks, thermal analysis 1. Introduction Spheroidal graphite cast iron (SGI), also known as ductile cast iron and nodular cast iron, is a kind of cast iron whose most important microstructural fea- ture is the presence of graphite nodules in the metal matrix. It is a specific engineering material, which pos- sesses good mechanical properties, castability, machin- ability, and particularly important, it has low produc- tion costs. Due to favorable combination of mechanical properties (high tensile strength and good ductility), SGI is used in many applications, such as pipes, vari- ous automotive parts etc. The mechanical properties of SGI are determined by the chemical composition, microstructural proper- ties and conditions during the solidification and the afterwards cooling [1–8]. The microstructure of SGI is determined in part during the solidification and in part during the following eutectoid (solid state) trans- formation. The shape of graphite is established during the solidification and it cannot be changed afterwards. In the as-cast condition, the typical metal matrix of SGI consists of ferrite and pearlite. *Corresponding author: tel./fax: +385 44 5333 78; e-mail address: glavaszo@siscia.simet.hr The most important factors which influence the impact toughness of SGI are: chemical composition, the shape and distribution of graphite, nodule count, the cooling rate during the solidification, the cooling rate through the eutectoid transformation range (solid state transformations) and the presence of other mi- crostructural constituents (for example: carbides, iron phosphide etc.). Chemical composition is one of the most significant factors in determining the metal matrix structure [1, 4–6]. The impact toughness of SGI depends strongly on the ferrite contents in the metal matrix. The SGI with ferritic metal matrix has lower tensile strength, but higher impact toughness and elongation. Si is a ferrite promoter, while elements such as Cu, Sn, Sb, Mn, Cr etc. are pearlite promoters. With the goal of producing as-cast ferritic SGI, the contents of pearl- ite promoters and carbides promoters (Cr, Mn, etc.) should be kept as low as possible. P is a very harmful element because it has a strong embrittling effect and should be kept as low as possible. Graphite nodularity has a significant influence on the impact toughness of SGI. Low graphite nodular-