Abstract:

In this study, Ramberg-Osgood parameters, which are the cyclic strength coefficient (K’) and the cyclic strain-hardening exponent (n’) belonging to various steels, are estimated by means of artificial neural networks (ANN). The effects of input parameters and hidden neuron numbers on estimation accuracy are investigated. The ANN structure consists of nine input parameters, one hidden layer, and two output parameters. In order to investigate the effect of the parameters on the estimation accuracy, first the results for nine input parameters were found, then each input parameter was excluded from the estimation, and finally new results were obtained with eight input parameters. The number of hidden neurons, which is another factor affecting the results was varied in the range of 1-30, and its effect on the estimation results was observed. To reduce the randomization of the estimation results, 100 replications made for each parameter. As a result, while estimating the cyclic strength coefficient (K’), the best neuron number was 1, the most effective parameter on the result was yield strength (σy), and the least effective parameter on the result was Brinell hardness (BHN). While estimating the cyclic strain-hardening exponent (n’), the best neuron number was 17, the most effective parameter on the result was fatigue strength exponent (b), and the least effective parameter on the result was Young’s modulus (E). Also, the output parameters were estimated with high estimation accuracy.