Machine learning-assisted constitutive modeling of a novel powder metallurgy superalloy

Abstract

The flow behavior and microstructure evolution during thermomechanical processing (TMP) of powder metallurgy superalloys are very important for their service behavior. In this contribution, a machine learning-assisted physical model was proposed to predict the flow stress, dynamic recrystallization (DRX) kinetic, and grain size evolution of a novel FGH4113A superalloy during the TMP. The work hardening and dynamic softening stages were modeled separately to predict the flow stress. The novel aspect of the proposed model is that the strong coupled effects of deformation temperature, strain rate, true stain, and primary γ' precipitates on the DRX kinetic were quantitatively described by the genetic algorithm-based artificial neural network (GA-ANN). Besides, the grain size modeling considered the pinning effect by primary γ' precipitates and the deformation-induced DRXed grain growth. Comparing the experimental and predicted flow curves, DRX fraction, and average grain size reveals that the proposed model has a preferable prediction accuracy than the conventional model. Then the specific DRX kinetic and grain size evolution characteristics were revealed. The promoting effect of strain rate on the DRX degree was gradually increased with the deformation temperature. Finally, the 3D processing map based on the developed model was established to guide TMP's optimized design and microstructure control. Three domains were divided and discussed: grain refinement, constant, and growth. The optimal processing window is defined as 1000–1050 °C/0.001–0.006 s−1 and 1075–1100 °C/0.008–0.01 s−1.