Accelerated discovery and formation mechanism of high-entropy carbide ceramics using machine learning based on low-cost descriptors

The vast compositional space of High-entropy carbide ceramics (HECCs) renders traditional experimental trial-and-error and computational simulations inadequate for rapid traversal and screening. In this work, leveraging retrievable and low-cost physicochemical parameters as feature inputs, we have successfully established an excellent predictive machine model (precision = 0.938, recall = 0.994, F1 score = 0.953) for HECCs, which is comparable to excellent model currently reported that was trained on density functional theory data, utilizing machine learning techniques. Further, we have identified compositional threshold conditions for the formation of (TiZrHf)C-based HECCs. Compared against literature data and experimental results, the constructed model is capable of reliably predicting the formation probabilities and phase diagrams for 5–9 component (TiZrHf)C-based HECCs with both equiatomic and non-equiatomic ratios those most have not been reported so far. Moreover, it has been discovered that the inability of equiatomic ratio formulations to form (TiZrHf)C-based HECCs does not necessarily imply that formulations across a global scope are incapable of forming (TiZrHf)C-based HECCs. Our developed model presents a highly efficient and low-cost technological approach for the swift screening of HECC phases.