Estimation of Cyclic Strain Hardening Exponent and Cyclic Strength Coefficient of Steels by Artificial Neural Networks

Abstract

The objective of this paper is to demonstrate the applicability of artificial neural networks on estimation of the cyclic strain hardening exponent and cyclic strength coefficient of steels on the basis of monotonic tensile tests properties. In order to demonstrate this applicability, steels tensile data was extracted from the literatures and two separate neural networks was conducted. One set of data was used for training networks and remaining of data for testing them. The regression analysis was used to check the system accuracy for training and test data at the end of learning. Comparing results of neural networks with values obtained from direct fitting of experimental data was indicated that cyclic strain hardening exponent and cyclic strength coefficient, which characterize the stable curves of true stress amplitude versus true plastic strain amplitude, were predicted reasonable. It was concluded that predicted stable cyclic true stress-strain curve properties by trained neural network are more accurate compared to approximate relations based on low-cycle fatigue properties.