Modelling Thermal Ageing Embrittlement in Turbine Rotors Using Neural Networks

Abstract

An artificial neural network-based algorithm, which adequately captures the complexity of the temper embrittlement phenomenon in NiCrMoV steels has been developed in order to predict the FATT50% value of material as a function of the chemical composition and serviced time. The model, validated using published data, relies for its training on a very large experimental data set of serviced rotors, aged up to 88,000 hours, and it captures the interactions between input parameters using complex non-linear functions. The results agree with considerations reported in literature and provide analytical relations between evolution of the FATT50% value at various ageing time for a specific alloy compositions, as well as the influence of the main elements in the alloy composition.