Search

Abstract: Hot compression tests of modified 2.25Cr–1Mo steel were conducted on a Gleeble-3500 thermo-mechanical simulator at the temperatures ranging from 1173 to 1473K with the strain rate of 0.01–10s−1 and the height reduction of 60%. Based on the experimental results, an artificial neural network (ANN) model and constitutive equations were developed to predict the hot deformation behavior of modified 2.25Cr–1Mo steel. A comparative evaluation of the constitutive equations and the ANN model was carried out. It was found that the relative errors based on the ANN model varied from −4.63% to 2.23% and those were in the range from −20.48% to 12.11% by using the constitutive equations, and the average root mean square errors were 0.62MPa and 7.66MPa corresponding to the ANN model and constitutive equations, respectively. These results showed that the well-trained ANN model was more accurate and efficient in predicting the hot deformation behavior of modified 2.25Cr–1Mo steel than the constitutive equations. [Copyright &y& Elsevier]

Abstract: The true stress–strain data from isothermal hot compression tests on Gleeble-3500 thermo mechanical simulator, in a wide range of temperatures (1323–1473K) and strain rates (0.01–10s−1), were employed to establish the constitutive equations based on modified Zerilli–Armstrong and strain-compensated Arrhenius-type models respectively, and develop the artificial neural network model to predict the high-temperature flow stress of T24 steel. Furthermore, a comparative study has been made on the capability of the three models to represent the elevated temperature flow behavior of this steel. Suitability of the three models were evaluated by comparing the accuracy of prediction of deformation behavior, correlation coefficient and average absolute relative error of prediction, the number of material constants, and the time needed to evaluate these constants. The results showed that the predicted values by the modified Zerilli–Armstrong model could agree well with the experimental values except under the strain rate of 0.01s−1. The predicted flow stress of the other two models shows good agreement with the experimental data. However, the artificial neural network model could track the deformation behavior more accurately throughout the entire temperature and strain rate range though it is strongly dependent on extensive high quality data and characteristic variables and offers no physical insight. [Copyright &y& Elsevier]