Construction of extrusion processing window for P/M nickel-based superalloy using deep neural network and its validation

Abstract

Canning extrusion is an effective way to refine grains, eliminate defects from hot isostatic pressing consolidation and enhance mechanical properties of P/M nickel-based superalloy. The present research constructed the optimized extrusion processing window for superalloy in consideration of extrusion speed, initial billet temperature, and extrusion ratio to provide guidelines for the choice of extrusion parameters. The processing window was built by integrating the results of finite element (FE) simulations, extrusion experiments, and deep neural network (DNN). The peak extrusion load and temperature under different extrusion parameters were collected as input datasets by FE simulations. The DNN was trained by optimizing the learning parameters and then predicting the load and temperature over various extrusion conditions. Finally, the extrusion processing window was established, including an optimized processing zone and two instability zones. The extrusion experiments were performed to verify the reliability of the processing window, which could be able to promote the extrusion production of nickel-based superalloys with high efficiency and quality.