Function–Structure–Synthesis: Machine Learning Enabled Closed-Loop Design of Biomass-Derived Porous Carbon Materials

Rational development of biomass-derived porous carbons (BDPCs) is vital for advancing fundamental research and addressing clean energy challenges such as hydrogen purification and supercapacitor energy storage. Current experimental designs are limited by empirical methods, and computational models struggle with the complex function–structure–synthesis relationships of biomass feedstocks. In this study, we introduce a data-driven closed-loop machine learning (ML) workflow integrated with literature-derived BDPC synthesis data sets, enabling inverse design from desired functionalities to optimal structures and synthesis conditions. Our ML framework predicts seven critical synthesis parameters─including precarbonation, physical state, type and amount of activator, activation temperature, time, and heating rate─based on targeted microstructural features. Experimental validations with various biofeedstocks confirm the efficacy of our approach. Notably, AI-S, a BDPC synthesized from sugarcane bagasse under ML guidance, achieves a CO2 working capacity of 12.3 mmol/g at 273 K and 10 bar and a CO2/H2 selectivity of 47.8─the highest reported for BDPCs to date. This ML-driven workflow bridges BDPC properties, microstructure, and synthesis, demonstrating a reliable and predictive method for functionally oriented porous material design and synthesis, thereby accelerating material discovery in energy and environmental applications.