Prediction of Creep Rupture Life in Nickel-Base Superalloys Using Bayesian Neural Network

Abstract

The creep rupture life of nickel-base superalloys has been predicted using a neural network model within a Bayesian framework. The rupture life was modelled as a function of some 42 variables, including temperature, chemical composition: Cr, Co, C, Si, Mn, P, S, Mo, Cu, Ti, Al, B, N, Nb, Ta, Zr, Fe, W, V, Hf, Re, Mg, ThO₂, La, four steps of heat treatment (each has its own temperature, duration and cooling rate), sample shape, solidification method, yield strength, ultimate tensile strength and elongation. The Bayesian method puts error bars on the predicted value of the rupture life and allows the significance of each individual factor to be estimated. The scale of the error bars changes with the accuracy of the prediction: it is large when the prediction is uncertain, indicating that the whole prediction system is reliable.