Low-cycle fatigue parameters and fatigue life estimation of high-strength steels with artificial neural networks

Abstract

Fatigue life estimation is essential for life safety and cost reasons. Fatigue parameters at low cycles must be estimated with high accuracy to correctly estimate fatigue life. Conventional equations for parameters at low cycles are inadequate and unable to calculate the values correctly. Artificial neural networks (ANNs) are used to estimate and optimize the parameters and are more accurate than traditional equations. This study aims to estimate the fatigue parameters at low cycles and fatigue life by ANN with basic tensile properties of high-strength steels (HSSs), which can be easily obtained from the literature. In particular, the fatigue strength exponent and fatigue ductility exponent primarily characterize the strain-life curve, and the estimation of these parameters is extremely important. To improve the accuracy of the estimation, activation functions, epoch numbers, training functions, elapsed times of training functions, and the number of hidden neurons is compared and determined.