Comparison of Modified Arrhenius, Modified Zerilli–Armstrong, and Long Short-Term Memory Models for Predicting Hot Deformation Behavior of NiAlCrFeMo High-Entropy Alloy

The NiAlCrFeMo high-entropy alloy (HEA) exhibits exceptional high-temperature strength but presents significant machining challenges due to its dual-phase (FCC + BCC) structure. Systematic hot compression tests were conducted over a range of temperatures (1123.15-1323.15 K), strain rates (0.1-10 s^-1), and a strain of 0.7. The predictive accuracy of a long short-term memory (LSTM) neural network was compared with two enhanced traditional models: modified Arrhenius and modified Zerilli–Armstrong. The LSTM model achieved superior predictive accuracy with an average absolute relative error (AARE) of 1.79%, a correlation coefficient (R) of 0.999, and a mean absolute error (MAE) of 2.51, significantly outperforming the modified Arrhenius model (AARE: 6.31%, R: 0.992, MAE: 5.90) and the modified Zerilli–Armstrong model (AARE: 8.77%, R: 0.974, MAE: 8.68). The hot working diagram was established based on the predictions from the LSTM model. SEM observations reveal numerous voids in the instability zone, while defect-free morphologies were observed under optimal conditions (1223.15-1323.15 K, 0.1 s-¹), confirming the model’s reliability. This work establishes a new paradigm for HEA process optimization by synergizing data-driven and physics-based approaches, offering a robust alternative for precise prediction and enhanced material performance.