Development of an improved deep network model as a general technique for thin film nanocomposite reverse osmosis membrane simulation

Machine Learning for Polymer Design to Enhance Pervaporation-Based Organic Recovery

“…PV performance prediction has traditionally relied on theoretical models and simulation methods, such as the pore-flow model, solution-diffusion model, density functional theory, and molecular dynamics, which have limited applicability and are computationally intensive. − Machine learning (ML) has emerged as a data-driven approach to solve complex, multivariate problems with high computational efficiency. , ML models have proved to be promising to facilitate the discovery of new materials and achieve the targeted design of materials. − ML models have been successfully employed to predict polymer properties (e.g., glass transition temperature, thermal conductivity) based on polymer structure. − In addition, ML models showed great potential to reveal the relationship between materials’ structure and their properties and could assist the design of membranes for different processes, but their use in PV membrane design is limited. − Previous studies linked 15 representative chemical functional groups with PV membrane performance, but the impacts were limited due to the small sample size (681 samples) compared to the total number of samples in the literature . Using only 15 chemical functional groups to describe a wide range of polymers runs into high risks of missing topological features, and lack of seed randomness and data leakage management (DLM) also compromise the model’s robustness and accuracy.…”