Machine Learning-Driven Insights into Defects of Zirconium Metal–Organic Frameworks for Enhanced Ethane–Ethylene Separation

Structural
defects in metal–organic frameworks (MOFs) have
the potential to yield desirable properties that could not be achieved
by “defect-free” crystals, but previous works in this
area have focused on limited versions of defects due to the difficulty
of detecting defects in MOFs. In this work, a modeling library containing
425 defective UiO-66 (UiO-66-Ds) with a comprehensive population (in
terms of concentration and distribution) of missing-linker defects
was created. Taking ethane–ethylene separation as a case study,
we demonstrated that machine learning could provide data-driven insight
into how the defects control the performance of UiO-66-Ds in adsorption,
separation, and mechanical stability. We found that the missing-linker
ratio in real materials could be predicted from the gravimetric surface
area and pore volume, making it a useful complement for the challenges
of directly measuring the defect concentration. We further identified
the “privileged” UiO-66-Ds that were optimal in overall
properties and provided decision trees as guidance to access and design
these top performers. This work offers a general strategy for fully
exploring the defects in MOFs, providing long-term opportunities for
the development of defect engineering in the adsorption community.