Machine learning-driven assessment of relationship between activator properties in phase change solvent and its absorption performance for CO2 capture

Abstract

Amine-based phase change solvents are recognized as an energy saving alternative to the benchmark solvent, mono-ethanolamine (MEA), for CO2 capture owing to their unique phase separation behavior and low energy penalty. However, screening of the phase change solvents is labor intensive via the experimental analysis. To address this issue, this work intends to develop a machine learning-driven assessment system to screen promising phase change solvents via exploring the relationship between activator properties and absorption performance. Based on the machine learning and quantum calculation, the relationships between 11 properties of 30 activators and 6 performance parameters were observed. Subsequently, the sensitive analysis was conducted via focusing on 5 key activator properties, such as alkalinity, hydrophobicity, molar refraction index, molar volume and the ADCH charge of N atom. Based on the proposed assessment system, 4 phase change solvents with potential excellent performance were screened, and kinetics and thermodynamics were further verified by experimental results. Among them, the regenerative heat of diethylenetriamine/N,N-Diethylethanolamine (DETA/DEEA) was around 2.44 GJ t−1, which is ∼ 34 % lower than that of benchmark MEA solvent. It is believed that the assessment system developed in this work can deepen the understanding of screening and practical application of phase change solvents.