Exploring the constitutive description and workability of a Al_0.1TiZrNbMoTa_0.7 refractory high-entropy alloy during hot deformation using a novel SCSO-BP artificial neural network model

Thermal simulation is an essential physical method used to investigate hot deformation behavior and working performance. However, the complexity of refractory high-entropy alloys (RHEAs) poses significant challenges in accurately describing their hot deformation behavior using conventional constitutive models. Herein, a novel stress prediction model (SCSO-BP ANN) was developed by combining the back propagation artificial neural network (BP ANN) with the advanced sand cat swarm optimization (SCSO) algorithm, which was used to prediction for flow stress of Al_0.1TiZrNbMoTa_0.7 RHEA. The SCSO-BP ANN model exhibited superior predictive capability over the BP ANN and traditional Arrhenius models, as evidenced by its higher determination coefficient (R²), lower root mean square error (RMSE) and average absolute relative error (AARE). Hot processing maps for Al_0.1TiZrNbMoTa_0.7 RHEA at different strains were constructed based on SCSO-BP ANN model. The microstructure observations revealed that the instability resulted from the generation of cracks and deformation bands. The increase of energy dissipation factor (η) was accompanied by the increase of recrystallization volume fraction. In the instability, low η and medium η regions, the predominant high temperature softening mechanism was determined to be discontinuous dynamic recrystallization (DDRX). In the high η regions, DDRX and continuous dynamic recrystallization (CDRX) jointly dominated flow behavior. The regions in the processing maps developed using SCSO-BP ANN model exhibited good correspondence with deformation microstructure, enabling accurate identification of the optimum processing window.