Application of a neural network in modelling of hardenability of constructional steels

Abstract The paper presents a new, original method for hardenability modelling of the alloy constructional steels, both carburized and heat-treatable. The automatic steel classification based on steel chemical composition was employed in this method. Three neural network based models of hardenability curves pertaining to each of the investigated steel groups were employed to obtain calculation results as close as possible to experimental results. The tests performed on about 1500 experimental hardenability curves gave, for the various heats indicated, a significant reliability of calculations made according to the presented method. Knowledge of the hardness distribution, depending on the distance from the specimen face and the distribution of the cooling rate of specimen cooled from the face, described by the Jominy hardenability curve, makes the rational selection of the alloy constructional steels possible for the heat-treated or thermo-chemically treated machine parts. Therefore, the new method of the hardenability curve modelling presented here may be an important element of the computer based systems for selection of alloy constructional steels for machine parts.