كيف توقف التنغستن عن الخوف من البرد سريع

الملخص الأصلي

كما كتبه المؤلفون، بالإنجليزية arXiv:2606.26501

Tungsten-based refractory high-entropy alloys (RHEAs) are promising materials for fusion applications but often remain brittle at room temperature because of tungsten's high ductile-to-brittle transition temperature (DBTT). To identify alloying strategies that improve ductility, we compiled a curated dataset of experimentally reported tungsten-containing alloys and developed composition-based, physics-informed descriptors for machine learning. Three classifiers were evaluated using nested cross-validation, and a support vector classifier (SVC) showed the best generalization for this sparse dataset. Shapley additive explanations identified the exchange-correlation parameter, valence electron concentration, pressure field, and electronegativity mismatch as the most influential features governing ductile-brittle behavior. Synthetic compositions generated within the convex interpolation domain of the training data were evaluated and visualized on pseudo-ternary diagrams to map predicted ductility trends. The model predicts that moderate additions of Ti, Ni, and Co increase the likelihood of room-temperature ductility, whereas high Cr contents promote brittle behavior. These predictions agree with published experimental observations, including Ti-assisted ductility through solid-solution and electronic effects and embrittlement in Cr-rich refractory alloys. Agreement between DFT-derived elastic descriptors and SVC decision-function margins further supports the physical relevance of the learned classification boundary. Although the available mechanical dataset remains limited and heterogeneous, the model captures experimentally consistent trends and provides an interpretable, physics-informed framework for screening tungsten-based RHEAs for targeted simulation and experimental validation in fusion environments.