Gaussian process regressions on hot deformation behaviors of FGH98 nickel-based powder superalloy

Abstract

The hot deformation behaviors of FGH98 nickel-based powder superalloy were experimentally investigated and theoretically analyzed by Arrhenius models and machine learning (ML). Hot compression tests were conducted with a Gleeble-3800 thermo-mechanical simulation machine on the FGH98 superalloy at strain rates of 0.001–1 s–1 and temperatures of 1025–1175 °C. The peak stresses under different deformation conditions were analyzed via the Sellars model and an ML-inspired Gaussian process regression (GPR) model. The prediction of the GPR model outperformed that from the Sellars model. In addition, the stress-strain responses were predicted by the GPR model and tested by experimentally measured stress-strain curves. The results indicate that the developed GPR model has great power with wide generalization capability in the prediction of hot deformation behaviors of FGH98 superalloy, as evidenced by the R2 value higher than 0.99 on the test dataset.