A new nondestructive test technique for predicting the temper embrittlement of turbine rotor steel with genetic programming

The genetic programming approach for predicting temper embrittlement of rotor steel (30Cr2MoV) is proposed. Two independent data sets are obtained experimentally: training data and verifying data. Peak current density of reactivation, temperature of electrolyte, the general chemical composition parameter (J-factor), chemical composition of Cr and S, hardness and the grain size parameter of the material are used as independent variables, while fracture appearance transition temperature as dependent variable. On the basis of training data, the best model was obtained by genetic programming, and the accuracy of it was verified with the verifying data. The prediction error of the model is within the scatter of ±20 °C. The results suggest that, the prediction model obtained by genetic programming is feasible and effective.