Periodic Reporting for period 1 - SmartDeZIgn (Smart Design Tool of High Performing ZIF Membranes for Important CO2-Related Separations)
Reporting period: 2020-05-01 to 2022-04-30
Separation processes constitute a highly important tool in our efforts to mitigate CO2 emissions. However, today’s separations are tremendously energy intensive, (separating various gases accounts for ~15% of the total global energy consumption), and they don’t achieve the desired performance for CO2-related separations. Zeolitic imidazolate frameworks (ZIFs) in the form of membranes, are studied towards developing cheaper, and better performing separation methods. Their most fascinating aspect is their modification and/or functionalization capabilities: ZIFs can be modified on the molecular level, and this can potentially provide control on their macroscopic properties. In other words, someone could tailor their separation performance by applying the appropriate changes in their framework (i.e. by using optimal variants of the ZIF building blocks). However, the correlation between the molecular structural changes and their subsequent ZIF separation efficiency is a very complex correlation that is yet to be resolved. Unveiling this correlation is very important, since it is expected that proper implementation of such materials in separation processes can have a huge impact on the global economy as well as on our effectiveness in reducing CO2 emissions.
In this project the main objective was to unmask the missing modification-separation correlation in ZIFs. Modification comes in the form of replacing the structural sub-units of the framework. Regarding the separation performance, we focused on the diffusivity of gases ZIFs, since this is the main driving force in separations with such materials. First, a novel artificial intelligence (AI) tool was developed, that is the first-reported tool that can predict the diffusivity of any gas in any known or un-known ZIF, as long as a researcher-user knows simple and readily available information about the ZIF structure under consideration. Then, with the help of this AI tool we were able to identify highly performing ZIFs for three highly challenging, CO2-related separations: H2/CO2, CO2/N2 and CO2/CH4. Finally, a new approach based on genetic algorithms was used and a software that can construct and propose optimal ZIF structures was developed.
In this project the main objective was to unmask the missing modification-separation correlation in ZIFs. Modification comes in the form of replacing the structural sub-units of the framework. Regarding the separation performance, we focused on the diffusivity of gases ZIFs, since this is the main driving force in separations with such materials. First, a novel artificial intelligence (AI) tool was developed, that is the first-reported tool that can predict the diffusivity of any gas in any known or un-known ZIF, as long as a researcher-user knows simple and readily available information about the ZIF structure under consideration. Then, with the help of this AI tool we were able to identify highly performing ZIFs for three highly challenging, CO2-related separations: H2/CO2, CO2/N2 and CO2/CH4. Finally, a new approach based on genetic algorithms was used and a software that can construct and propose optimal ZIF structures was developed.
The work carried during the project was purely computational. Through Density Functional Theory (DFT) calculations and molecular simulations, we designed more than 50 new ZIFs, by varying the structural sub-units that build up the material’s framework, and we carried out simulations in each ZIF to measure structural changes and to estimate the diffusivity of a big variety of gas species. Chemical and structural information on each ZIF’s structural sub-unit, along with the gas diffusivities in the ZIFs, constituted a big data-set, which was used to train Machine Learning (ML) models, in order to predict the diffusivity of any gas in any possible ZIF, either known or unknown. Our ML models helped us understand better the proper replacement patterns that can lead to new ZIFs with exceptional performance for targeted separations. Thus, we extended the initial database of 50 ZIFs, with 20 additional (optimized) structures. We identified top-performing ZIFs for three challenging CO2-related separations: H2/CO2, CO2/N2 and CO2/CH4. In order to validate the performance of these three ZIFs, we calculated again their performance through simulations that are close to the experimental membrane permeation experiments after using Non-equilibrium molecular dynamics (NEMD).
The nature of our work advanced the state of the art in multiple ways. High-throughput screening simulations in nanoporous solids are focused on sorption properties, while diffusivity, which is the driving force in molecular sieving membranes, is completely omitted. Our work reported the first-ever high-throughput screening for the calculation of diffusivities. Also, we reported the first AI tool for the prediction of diffusivities in nanoporous solids. This way a researcher can skip the time-costly traditional simulations steps and get an accurate first prediction for a ZIF of their design in seconds. Moreover, an outstanding outcome of this work, which was not initially planned in the proposal, was a novel method to design ZIFs for targeted separations. This approach works in the opposite direction of the traditional ML predictive models. It is based on genetic algorithm optimizer approaches and can automatically design optimum ZIFs targeted for specific separation processes. More specifically, a user can input a set of desired diffusivity values for any binary gas mixture and the our tool can propose optimal ZIF structures which are expected to exhibit enhanced performance .