Mechanical Properties Control Of Low And Medium Carbon Steel Using Feed Forward Neural Networks

Abstract

Two feed forward multilayer neural network (FFMNN) architecture have been constructed. The first model has two hidden layers, 20 neurons for first layers and 30 neurons for second layers. The first model utilized to control the mechanical properties (yield strength, tensile strength and hardness) which are the inputs to this model. The model gives the carbon percentage and the heat treatment type and temperature as response to the required mechanical properties. The second model is implemented to predict the effect of carbon, quenching and tempering temperature on mechanical properties of both tensile strength and hardness of low and medium carbon steel. It has two hidden layers, 30 neurons for first layers and 60 neurons for second layers. In addition, the effect of different carbon percentages and heat treatment temperature for quenching were stud- ied on the mechanical properties of low to medium carbon steel. The results of Vickers hardness showed that hardness, yield strength and tensile strength were increased with increasing the car- bon percentage, quenching temperature and decreasing the tempering temperature. The neural networks models showed good agreements with experimental data. The correlation coefficients for the first model versus the experimental data are 0.9917, 0.9782 and 0.9954 for yield strength, tensile strength and Vickers hardness respectively. Higher correlation coefficients were obtained for the second modelling

Keywords

FFMNN, Vickers, ANNs