Neural network in predicting steel properties - artificial neural network as a tool in predicting mechanical and thermal properties of steels

Over the past few years a detailed study of artificial neural network application in material engineering has been performed at the Department for Materials of FMENA. This paper presents the applied methods and results in predicting different steel properties. Hardenability Jominy curves, i.e. hardness at different Jominy distances are predicted for different groups of nonboron constructional steel grades for hardening and tempering, and for carburising. Input parameters were the results of chemical analysis of melt. A set of different heats was used to test the neural network efficiency in predicting J- hardness. Furthermore, a similar method was used to predict the tempering curve of tool steels, using the given chemical composition and austenising temperature. The mean error and standard deviation for the learning and for the testing dataset were small and acceptable. Another very important physical property that was determined was the heat conductivity. Heat conductivity versus temperature, which is needed in calculation and simulation of heating and cooling processes, is predicted using both the regression analysis and the neural network method.