Machine learning lattice constants for spinel compounds

Predicting mechanical performance of starch-based foam materials

Variations in mixture proportions of plasticizers, additives, and crosslinking agents have significant impacts on mechanical performance of starch-based foam materials. In particular, starch/ethylene-vinyl acetate (EVA) foam materials have been developed with improved mechanical strength by optimizing the formulation. There is a lack of numerical correlations that could help analyze the effects of components and provide a predictive method for future research. In this study, we develop simple and accurate predictions for tensile strength and resilience based on mixture proportions of components for starch-based/EVA foam materials. The models constructed might be used to help design mixture proportions of starch-based foam materials. By combining optimization results from the Taguchi method and machine learning approaches, it is expected that more quantitative data can be extracted from fewer experimental trials at the same time.

Modeling of lattice parameters of cubic perovskite oxides and halides

Perovskites having the chemical formulae of ABX3 are promising candidates for various electronic, magnetic, and thermal applications. One of the important structural factors is a (the lattice constant), which represents the unit cell size. The variation in the lattice constant is a combined result of interactions between different ions, determined by valence electrons and ionic radii. The size and stability of unit cells have important influences on structural stabilities, bandgap structures, and therefore performance of materials. To obtain the lattice constant of cubic perovskites without going through experimental efforts such as synthesis and measurements, we construct a model based on Gaussian process regressions for cubic perovskite lattice constant predictions. The model utilizes the number of valence electrons as well as ionic radii of alloying elements as predictors. A total of 149 cubic perovskites containing fluorides, chlorides, and bromides with cation combinations of A1+B2+, as well as oxides with cation combinations of A1+B5+, A2+B4+, and A3+B3+ are explored. The model demonstrates good performance in terms of stabilities and accuracy, and thus could be a rapid approach to estimate lattice constants.

Machine learning glass transition temperature of styrenic random copolymers

For styrenic random copolymers, the glass transition temperature, Tg, is an important thermophysical parameter, which is sometimes difficult to measure and determine by experiments. Approaches based on data-driven modeling provide alternative methods to predict Tg in a fast and robust way. The Gaussian process regression (GPR) model is investigated to present the statistical relationship between important quantum chemical descriptors and glass transition temperature for styrenic random copolymers. 48 samples with Tg that have been measured experimentally are explored, which range from 246 K to 426 K. The modeling approach demonstrates high accuracy and stability, and provides a novel and promising tool for efficient and low-cost estimations of copolymer Tg values.

Solubility predictions through LSBoost for supercritical carbon dioxide in ionic liquids

The LSBoost model is developed to predict the solubility of supercritical carbon dioxide in 24 ionic liquids by using critical properties and biphasic system parameters as descriptors. The model is highly accurate and stable.