Machine learning-guided design of crosslinked cellulose acetate ultrafiltration membranes for sustainable water purification.

The accumulation of harmful organic pollutants in water resources hazards the ecosystems and human health, making the access to clean drinking water a major challenge. Biodegradable cellulose acetate (CA) based ultrafiltration (UF) membranes are regarded as ideal materials for removing these pollutants, while their practical applications are limited by poor compaction resistance and antifouling performance. Herein, we propose a machine learning (ML)-guided membrane design strategy to prepare crosslinked CA-based UF membranes with excellent compaction resistance and antifouling property by introducing sodium lignosulfonate (SL) into the CA matrix through the crosslinking reaction mediated by isocyanate groups. Benefiting from the crosslinking among CA chains, the elastic modulus of fabricated membrane exhibits 5.8 times higher than pristine CA membrane and its pore structure maintains intact after 5 h of filtration at 3 bar. Meanwhile, the SL improves the antifouling property of the membrane, leading to a high rejection rate of 98% and flux recovery ratio of 98%. Notably, the membrane shows excellent pollutants removal capacity and reusability in real surface water purification, superior to the commercial UR030 membrane. This ML-guided design strategy of crosslinked membrane provides an effective approach to develop high-performance CA membranes, facilitating their practical application in sustainable water purification.