2021/03/05 by Mohd Zaki, Zaki, Mohd, Vineeth Venugopal +12
Materials Science · #Data Analysis #FOS: Physical sciences #Glass properties and applications #Materials Science (cond-mat.mtrl-sci) #Optics (physics.optics) #Statistics and Probability (physics.data-an)
paper · pdf · doi:10.48550/arxiv.2103.03633
openalex publication_date 2021/03/05 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Due to their excellent optical properties, glasses are used for various\napplications ranging from smartphone screens to telescopes. Developing\ncompositions with tailored Abbe number (Vd) and refractive index (nd), two\ncrucial optical properties, is a major challenge. To this extent, machine\nlearning (ML) approaches have been successfully used to develop\ncomposition-property models. However, these models are essentially black-box in\nnature and suffer from the lack of interpretability. In this paper, we\ndemonstrate the use of ML models to predict the composition-dependent\nvariations of Vd and n at 587.6 nm (nd). Further, using Shapely Additive\nexPlanations (SHAP), we interpret the ML models to identify the contribution of\neach of the input components toward a target prediction. We observe that the\nglass formers such as SiO2, B2O3, and P2O5, and intermediates like TiO2, PbO,\nand Bi2O3 play a significant role in controlling the optical properties.\nInterestingly, components that contribute toward increasing the nd are found to\ndecrease the Vd and vice-versa. Finally, we develop the Abbe diagram, also\nknown as the "glass veil", using the ML models, allowing accelerated discovery\nof new glasses for optical properties beyond the experimental pareto front.\nOverall, employing explainable ML, we discover the hidden compositional control\non the optical properties of oxide glasses.\n