Machine learning-assisted design of porous carbons for removing paracetamol from aqueous solutions

To accelerate the design and production of porous carbons targeting desired performance characteristics, we propose to incorporate machine learning (ML) regression into pore size distribution (PSD) analysis. Here, we implemented a ML algorithm for predicting paracetamol adsorption capacity of porous carbons from two pore structure parameters: total surface area and surface area of supermicropores-mesopores. These structural parameters of porous carbons are accessible from the software provided with automatic volumetric gas adsorption analyzers. It was shown that theoretical paracetamol capacities of porous carbons predicted using the ML algorithm lies within the range of experimental uncertainty. Nanoporous carbon beads with a high surface area of supermicropores (997 m(2)/g) and mesopores (628 m(2)/g) had the highest adsorption capacity of paracetamol (experiment: 480 +/- 24 mg/g, ML predicted: 498 mg/g). The novel strategy for designing of porous carbon adsorbents using ML-PSD approach has a great potential to facilitate production of novel carbon adsorbents optimized for purification of aqueous solutions from non-electrolyte contaminates.

Automated summary

To expedite the design and production of porous carbons with specific performance characteristics, incorporating machine learning regression into pore size distribution analysis is proposed. A machine learning algorithm was implemented to predict the paracetamol adsorption capacity of porous carbons based on two pore structure parameters: total surface area and surface area of supermicropores-mesopores. The experimental and predicted results for paracetamol capacities using this algorithm were shown to be within the range of experimental uncertainty. This novel approach has the potential to facilitate the production of carbon adsorbents optimized for the purification of aqueous solutions from non-electrolyte contaminants.