Machine learning-powered prediction of argon thermophysical properties for hot isostatic pressing

Argon is the primary pressure-transmitting medium in Hot Isostatic Pressing (HIP), and its thermophysical properties under high-temperature, high-pressure conditions strongly affect control accuracy and component quality. However, accurate and efficient property prediction remains difficult because conventional equations of state (EOS) are computationally expensive and may become numerically stiff, while experiments are costly and limited in range. This study develops a machine-learning framework to predict argon thermophysical properties, including density, enthalpy, heat capacities, and transport coefficients, over a wide thermodynamic domain. Six regression models (XGBoost, LightGBM, Random Forest, Extra Trees, Support Vector Regression, and Multi-Layer Perceptron) were trained, among which Extra Trees showed the best overall robustness for the evaluated tasks. The proposed framework avoids iterative EOS solving during inference and achieves R2 values above 0.998 for core energy-related properties. For thermal conductivity, comparison with independent Green–Kubo molecular dynamic simulation showed good agreement on the test set and reproduced the main temperature- and pressure-dependent trends. These results indicate that the framework is a robust and numerically stable tool for argon property prediction in HIP and supports process monitoring and digital-twin applications within the sampled domain.