Machine learning density prediction and optical properties of calcium boro-zinc glasses

Abstract

In the present work, 20CaO–60B2O3 – 5Li2O–15ZnO – xEu2O3: 0 ≤ x ≤ 0.5 mol% glass samples have been synthesized via the melt-quenching technique. The role of the Eu2O3 oxide on the physical and optical properties of different europium (III) borate glasses through artificial intelligence techniques is investigated. The experimental density (ρ) of the proposed glasses enhanced from 2.231 g/cm3 for Eu0.0 sample to 3.338 g/cm3 for Eu0.5 sample. The molar volume (VM) varies between 20.643 cm3/mol and 20.506 cm3/mol. Artificial intelligence (AI) density model is proposed based on the composition of the glass for a dataset of 10000 oxide glass samples to predict density of present glasses using various AI techniques like gradient decent, Artificial Neural network, and Random Forest regression. Random forest regression RFR model fit the glass data with highest R2 value 0.979 for density prediction compare to other models. The R2 is regulated to 0.885 for gradient decent with minimum cost function for both density prediction which is lowest value compare to RFR. Artificial neural network with different activation function has been applied and the highest R2 value 0.970 for density prediction obtained in the case of Tanh activation function compare to other activation functions. The optical energy gap (Eg) of the proposed glasses was varied from 2.98 eV to 3.07 eV, while the values of Urbach energy are within 0.19 and 17 eV. Refractive index was in the range of 2.38 and 2.40.