Machine learning in gas separation membrane developing: Ready for prime time

Membrane technology is a promising next-generation gas separation technology and has drawn tremendous research interest during the past decades. Despite the advanced progress in membrane materials, the timelines for developing high-performance membranes still cannot be compatible with the urgent needs. Accurate prediction of membrane performance in advance can efficiently optimize the process of novel membrane development and economize the funds. As so, numerous empirical, semi-empirical, or completely theoretical prediction models were established over the past decades while still facing the challenges of poor universality and limited accuracy. Big-data science-based “machine learning” (ML) has made a great effort in many research communities including chemistry and material science. Using experimental or/and molecular simulation data as the input to train a suitable ML model enable to precisely predict and interpret many membrane properties including gas separation performance. This review aims to elucidate how machine learning, a promising data-centric strategy, accelerate the screening and design of gas separation membranes. First, the concepts and algorithms of ML are briefly introduced. Then, a comprehensive and critical review of the progress in gas separation membrane screen and design assisted by machine learning is made (consisting of polymer membrane, porous polycrystalline membrane, and mixed matrix membrane). Subsequently, the data ecology in gas separation membranes is summarized. Lastly, the current issues, along with the outlooks in improving the development of gas separation membranes through machine learning are proposed.