Machine Learning-Empowered Plastic-Derived Porous Carbons for High-Performance CO ₂ Capture

Abstract

Conspectus Plastic pollution and climate change are interconnected global environmental challenges. Conventional methods (incineration and landfills) exacerbate these issues by emitting greenhouse gases and releasing micro/nanoplastics. To simultaneously address these two critical environmental issues, we upcycle plastic waste into porous carbon materials, enabling high-performance postcombustion CO 2 capture in a transformative and practical manner. This strategy tackles environmental pollution, aligns with circular economy principles, and supports several of UN Sustainable Development Goals (SDGs). We conduct systematic studies, including experimental validations, numerical simulations, and machine learning (ML)-empowered optimizations, to provide detailed guidelines for upcycling plastic waste into porous carbons with high-performance CO 2 capture. Synthesis routes vary in their environmental benefits and economic feasibility. Different activating agents (e.g., steam, potassium hydroxide, and urea) are comprehensively compared. Experimental operating parameters (e.g., activation temperature, activating agent type, and loading mass ratio) are optimized to produce micropore-dominated carbon materials that exhibit excellent CO 2 adsorption performance. Our main results show that plastic-derived porous carbons achieved high specific surface areas (up to 2,060 m 2 /g) and micropore volumes (∼1.02 cm 3 /g), demonstrating great potential in CO 2 adsorption capture. Functional groups like C═O and O–H further enhance the CO 2 adsorption capacity due to their strong dipole–quadrupole interactions with CO 2 molecules and the formation of hydrogen bonding. Based on experimental investigations, Grand Canonical Monte Carlo (GCMC) simulations reveal that narrow micropores (<0.8 nm) and optimal isosteric heat (23–28 kJ/mol) favor CO 2 physisorption. In this work, interpretable ML techniques, such as feature importance ranking and SHAP analysis, reveal the key structural and chemical features that dominate CO 2 uptake performance. These include the pore size distribution and surface chemistry, which provide valuable guidance for the rational design of plastic-derived porous carbons. Building on these insights, we also apply ML, particularly active learning and particle swarm optimization (PSO) approaches, for iteratively identifying optimal synthesis parameters, thereby enhancing CO 2 adsorption capacity by up to 2-fold relative to seed experiments. This strategy offers a more efficient route to performance improvement compared to conventional trial-and-error approaches. To ensure industrial applicability, we assess the cyclic performance of CO 2 capture by temperature swing adsorption (TSA), pressure swing adsorption (PSA), and vacuum swing adsorption (VSA) processes and then scale these technical routes via process simulations. Multiobjective optimization achieves an excellent 35.13% exergy efficiency, aided by artificial neural network (ANN)-based surrogate modeling and genetic algorithms. This work was also studied from perspectives of both environmental benefits and economic feasibility to explore its potential for enabling sustainable development. This holistic and multidisciplinary strategy, combining materials science, AI algorithms, and environmental engineering, offers a carbon-negative and economically viable path to simultaneously mitigate climate change and achieve a circular plastic economy. Our future work will focus on data set expansions, intelligent optimizations, and large-scale deployment for real-world impacts.

Topics & Concepts