Combinatorial Analysis of Sparse Experiments on Photocatalytic Performance of Cement Composites: A Route toward Optimizing Multifunctional Materials for Water Purification.

Blending TiO2 and cement to create photocatalytic composites holds promise for low-cost, durable water treatment. However, the efficiency of such composites hinges on cross-effects of several parameters such as cement composition, type of photocatalyst, and microstructure, which are poorly understood and require extensive combinatorial tests to discern. Here, we report a new combinatorial data science approach to understand the influence of various photocatalytic cement composites based on limited datasets. Using P25 nanoparticles and submicron-sized anatase as representative TiO2 photocatalysts and methyl orange and 1,4-dioxane as target organic pollutants, we demonstrate that the cement composition is a more influential factor on photocatalytic activity than the cement microstructure and TiO2 type and particle size. Among the various cement constituents, belite and ferrite had strong inverse correlation with photocatalytic activity, while natural rutile had a positive correlation, which suggests optimization opportunities by manipulating the cement composition. These results were discerned by screening 7806 combinatorial functions that capture cross-effects of multiple compositional phases and obtaining correlation scores. We also report •OH radical generation, cement aging effects, TiO2 leaching, and strategies to regenerate photocatalytic surfaces for reuse. This work provides several nonintuitive correlations and insights on the effect of cement composition and structure on performance, thus advancing our knowledge on development of scalable photocatalytic materials for drinking water treatment in rural and resource-limited areas.

Deep Learning to Speed up the Development of Structure-Property Relations For Hexagonal Boron Nitride and Graphene.

Structure-property maps play a key role in accelerated materials discovery. The current norm for developing these maps includes computationally expensive physics-based simulations. Here, the capabilities of deep learning agents are explored such as convolutional neural networks (CNNs) and multilayer perceptrons (MLPs) to predict structure-property relations and reduce dependence on simulations. This study contains simulated hexagonal boron nitride (h-BN) microstructures damaged by various levels of radiation and temperature, with the objective to predict their residual strengths from the final atomic positions. By developing low dimensional physical descriptors to statistically describe the defects, these results show that purpose-specific microstructure representation can help in achieving a good prediction accuracy at low computational cost. Furthermore, the adaptability of the trained deep learning agents is explored to predict structure-property maps of other 2D materials using transfer learning. It is shown that in order to achieve good predictions accuracy (≈95% R2 ), an agent that is training for the first time ("learning from scratch") requires 23-45% of simulated data, whereas an agent adapting to a different material ("transfer learning") requires only about 10% or less. This suggests that transfer learning is a potential game changer in material discovery and characterization approaches.