Three-dimensional hot processing map of a nickel-based superalloy (Alloy 925) established by modified artificial neural network model

Abstract

The hot deformation behavior of Alloy 925 in the temperature range of 900–1150 °C and strain rate range of 0.01–10s−1 was investigated through the isothermal hot compression tests. Based on the artificial neural network model, the stress distribution in the three-dimensional (3D) space consisted of deformation condition parameters was reconstructed. Afterwards, the 3D distribution maps of strain rate sensitivity (m) and power dissipation efficiency (η) were proposed to reveal the effect of deformation parameters on the hot workability. Furthermore, a novel 3D processing map was established by combining the traditional dynamic materials model and the corresponding data set. Through the microstructural characterization, it can be found that the instability regions exhibit inhomogeneous microstructure with massive distorted grains and few dynamic recrystallization (DRX) grains, while the stable regions are characterized by various DRX mechanisms, including discontinuous dynamic recrystallization, continuous dynamic recrystallization, and twin-DRX. Noteworthily, the relatively low η-value (<22%) was revealed in the high temperature-high strain rate domain caused by substantial occurrence of DRX. Finally, the optimum processing parameters at the true strain of 0.85 can be determined to be 970–1150 °C and 0.01–1.08s−1 with the η-value of 30.6–42.3%. In contrast, the construction method of novel 3D processing map can effectively avoid data distortion and also demonstrate a better validity for predicting hot workability. Through the present work, the feasibility of artificial neural network for learning complex hot deformation behavior has been illustrated. It not only has better prediction of typical microstructure, but also can reproduce the adiabatic heating law, which is convenient to be used in actual industrial production.