Deep learning-based modeling of the strain rate-dependent thermomechanical processing response for a novel HIPed P/M nickel-based superalloy

Abstract

Understanding and modeling the thermomechanical processing (TMP) response of a novel HIPed powder metallurgy (P/M) superalloy is crucial for its industrial application. This research characterizes the TMP-responding behavior of a novel HIPed FGH4113A P/M superalloy at 1000–1150 ℃ with strain rates of 0.01–5 s−1. The results depict that the adiabatic temperature rise during high-strain-rate TMP will accelerate the stress softening. A two-stage deep learning (DL)-based constitutive model framework, coupled with the deformation-induced temperature rise and its influence on flow stress, was first proposed and compared with the traditional Arrhenius-type model. Two types of DL approaches including the backpropagation (BP) and long short-term memory (LSTM) neural networks were trained and compared separately. It is noted that the LSTM neural network, considering previous deformation history in its unique gate structure, depicts the best prediction accuracy and generalization ability under large strains than other models. Then the strain rate-dependent hot workability and flow softening mechanisms were modeled and discussed. High-strain-rate deformation (5 s−1) in certain temperatures (1110–1150 ℃) can avoid unstable defects and transform the generated deformation heat into the driving force for the activation of multiple dynamic recrystallization mechanisms. Therefore, the near fully recrystallized microstructure can be achieved similarly to that at 0.01 s−1, indicating the feasibility and superiority of the high-strain-rate TMP of P/M superalloys in a certain temperature range.