ViscNet: Neural network for predicting the fragility index and the temperature-dependency of viscosity

Viscosity ($\eta$) is one of the most important properties disordered matter. The temperature-dependence viscosity used to adjust process variables for glass-making, from melting annealing. aim this work was develop a physics-informed machine learning model capable predicting $\eta(T)$ oxide liquids. Instead itself, NN predicts parameters MYEGA equation: liquid's fragility index, glass transition temperature, and asymptotic viscosity. With these parameters, $\eta$ can be computed at any temperature interest, with advantage good extrapolation capabilities inherent equation. dataset collected SciGlass database; only liquids enough data points in high low regions were selected, resulting final 17,584 containing 847 different About 600 features engineered liquids' chemical composition 35 selected using feature selection protocol. hyperparameter (HP) tuning performed set experiments both random search Bayesian strategies, where total 700 HP sets tested. successful further tested 10-fold cross-validation, lowest average validation loss as best set. trained test 85 compositions than those training validating NN. $R^2$ dataset's prediction 0.97. This introduces three advantages: predict well index; designed focus on extrapolation; finally, available free open-source software licensed under GPL3.