Phase-Field Simulation and Machine Learning for Predicting Rafting Kinetics in Ni-Based Superalloys

Directional coarsening of the γ' phase (rafting) in Ni-based single crystal superalloys during creep at 1273 K was simulated by the phase-field method. The inelastic strain introduced in the γ phase was assumed to be composed of plastic strain (ε^p) and creep strain (ε^c). The simulations were performed with various sets of values of material parameters and the magnitude of external tensile stress. We let a feed-forward neural network learn the simulation data in order to enable fast and exhaustive prediction of the time to rafting, t_raft. From the analysis based on the trained neural network, it has been shown that t_raft becomes longer with increasing magnitude of γ/γ' lattice misfit, with decreasing creep coefficient, and with increasing yield stress of the γ phase (σγ_ys). The sensitivity of t_raft to σ^γ_ys is high when the ratio of ε^p to the total inelastic strain (ε^p + ε^c) is high.