Advances in adsorption processes driven by machine learning

This paper systematically reviews the research progress and application of machine learning in adsorption processes. By virtue of outstanding nonlinear modeling and data mining capabilities, machine learning has been successfully applied to adsorbent design and high-throughput screening, adsorption parameter prediction and process optimization, reactor design and digital twin simulation, as well as interpretable model development. The reviewed studies demonstrate that machine learning enables highly accurate prediction of adsorption performance, greatly accelerates material discovery and process optimization, and significantly reduces experimental and computational costs. This work also summarizes mainstream machine learning models, public adsorption databases, common simulation methods, and current challenges including insufficient model interpretability, data scarcity, and insufficient coupling with physical mechanisms. Finally, future directions such as deep integration of mechanistic models and AI, intelligent adsorption 2.0, and AI-assisted 3D printing for reactor manufacturing are prospected. This review is expected to provide a comprehensive reference for promoting the intelligent, efficient, and green development of adsorption separation technology.

Adsorptive separation is a core energy-intensive process in the chemical industry, with separation and purification accounting for 45%–60% of total industrial energy consumption. Conventional experimental and theoretical methods struggle to handle the strong nonlinearity, multicomponent coupling, and complex interactions in practical adsorption systems, leading to low efficiency, high cost, and poor scalability in material development and process optimization. Against this background, researchers led by Jiangtao Yu (Tarim University) and Wenshuai Zhu (China University of Petroleum-Beijing) have systematically reviewed the integration of machine learning (ML) with adsorption science and engineering. This work provides a unified framework for data-driven intelligent modeling, high-precision prediction, and rational optimization of adsorption processes.

As emphasized by Professor Yu, machine learning enables powerful nonlinear modeling and high-throughput data mining that overcome bottlenecks in traditional methods. It supports accurate adsorption performance prediction, rapid adsorbent screening, and intelligent parameter optimization while maintaining physical consistency and operational safety.

This review classifies ML models including ANN, SVM, random forest, CNN, GNN, and Transformer according to their applications in adsorption: nonlinear process prediction, microscopic image analysis, time-series kinetics, adsorbent design, and reactor optimization. Typical models achieve R² > 0.90 in adsorption capacity, selectivity, and kinetic parameter prediction, accelerating computation and material screening by up to 10⁴ times.

A key challenge addressed in this review is the “black-box” limitation of ML models. The authors highlight the importance of interpretable artificial intelligence (XAI) tools such as SHAP and the fusion of ML with fundamental adsorption mechanisms. This combination quantifies feature contributions, reveals structure–property relationships, and ensures predictions agree with thermodynamic and kinetic principles. The work systematically summarizes applications including high-throughput screening of MOFs and porous carbons, adsorption isotherm and energy prediction, CFD‑assisted reactor design, digital twin simulation, and real‑time process monitoring. Validations against experimental data confirm reliable, stable, and highly consistent performance across gas separation, wastewater treatment, and CO₂ capture systems. As commented by Professor Zhu, this review promotes the development of intelligent chemical engineering systems and provides a solid foundation for safer, more efficient, and low‑carbon adsorption technologies. The methodologies can also be extended to green separation, environmental remediation, energy storage, and AI‑assisted 3D‑printed adsorption reactors.

In summary, machine learning‑driven adsorption achieves high precision, low cost, and high interpretability, establishing a new benchmark for intelligence and sustainability in adsorption separation. Although challenges remain in small‑sample learning, data standardization, and multiscale mechanism integration, this review represents an important step toward next‑generation intelligent adsorption technologies.

This paper “Advances in adsorption processes driven by machine learning” was published in AI & Materials.

Yu J, Tang M, Ajibek P, Gao F, Zhu W. Advances in adsorption processes driven by machine learning. AI Mater. 2026, 2(2), https://www.elspub.com/doi/10.55092/aimat20260004.