Hot deformation behavior of novel nickel-based powder superalloys: Modified modeling, machine learning prediction, and multiscale simulation

ABSTRACT This study systematically investigates the isothermal hot deformation behavior of a novel powder-metallurgy nickel-based superalloy and proposes an integrated modeling and prediction framework that combines experimental modification, data-driven approaches, and multiscale coupling. First, Uniaxial isothermal compression tests were conducted on a Gleeble simulator at temperatures ranging from 1065 to 1140 ℃ and strain rates from 0.001 to 1 s⁻¹. To improve the accuracy of rheological data, the experimental flow curves were corrected for end friction and adiabatic heating, and the latter was further validated by finite element analysis. Under conditions of 1085 ℃ and 1090 ℃ at 0.1 s⁻¹, the calculated temperature rises were 13.10 ℃ and 12.80 ℃, respectively, in good agreement with the simulation results. Based on the corrected sequential data, three types of time-window regression models (SVR, MLP, GBR) were developed, and their extrapolation performance was evaluated using independent training/testing datasets. The results revealed that the SVR model with time-windowing achieved the best performance in both overall fitting and extrapolation tasks (overall R²=0.9999). The stress–strain curves predicted by the machine learning model were subsequently incorporated into a finite element–cellular automaton (Finite Element–Cellular Automaton（FE+CA）) coupled simulation, which successfully reproduced the nucleation and growth of dynamic recrystallization (DRX). The predicted average grain sizes exhibited a relative error of less than 5% compared with experimental measurements. Furthermore, the processing map constructed from the corrected flow curves showed that the unstable regions were reduced by approximately 25% relative to the uncorrected case, thereby improving the reliability of hot working window identification. The proposed correction–data-driven–multiscale coupling strategy provides a transferable technical route and a high-accuracy predictive approach for optimizing hot-working parameters, processing simulation, and microstructural control of nickel-based powder superalloys.