Physics‐Informed Machine Learning for Predicting Phases and Density of Al ₂ O ₃ –TiB ₂ Composites

Predicting phase transformation in multicomponent, highly exothermic reactive systems remains a challenge. Variations in stoichiometry and processing parameters affect the phase purity and density of in situ manufactured composites. This work presents a novel physics‐informed machine learning (PIML) model to predict the composition and bulk density of Al ₂ O ₃ –TiB ₂ composites synthesized through furnace SHS and laser ignition chemical synthesis (LIChemS) from the Al–TiO ₂ –B ₂ O ₃ –H ₃ BO ₃ reactive system. The PIML model integrates physics rules across a multioutput mass‐fraction model (Model 1—V1, V2, V3), a bulk density model (Model 2), and a compositional–topological interpretation model (Model 3). Model 1 (V1, V2) achieved hold‐out testing R ² of 0.994, whereas the unconstrained V3 produced nonphysical outputs, including negative mass fractions. LIChemS composites yielded mAl ₂ O ₃  = 0.69 and mTiB ₂  = 0.22 (purity = 0.91), with the model deviating by 3 wt%. Model 2 achieved test R ²  = 0.927, with the ML and experimental densities being 3.68 and 3.71 g/cc, respectively. Pareto optimization identified optimal processing windows near ignition thresholds, targeting purity > 0.85 and bulk density > 3.5 g/cc. Ternary and tetrahedral mapping demonstrated that LIChemS yielded Al ₂ O ₃ –TiB ₂ outputs. The developed PIML model provides a time‐ and material‐efficient approach to optimizing reactive system outputs in industrial settings.