Selective and Sustainable Oxidation of Allyl Alcohol in Water Using a TS-1 Catalyst in Continuous Flow: A Machine-Learning-Driven Approach

Page view(s)
0
Checked on
Selective and Sustainable Oxidation of Allyl Alcohol in Water Using a TS-1 Catalyst in Continuous Flow: A Machine-Learning-Driven Approach
Title:
Selective and Sustainable Oxidation of Allyl Alcohol in Water Using a TS-1 Catalyst in Continuous Flow: A Machine-Learning-Driven Approach
Journal Title:
ACS Sustainable Chemistry & Engineering
Keywords:
Publication Date:
05 December 2025
Citation:
Mehta, K. H., I Made, R., Agrotis, S., Parkin, I. P., Sankar, G., & Handoko, A. D. (2025). Selective and Sustainable Oxidation of Allyl Alcohol in Water Using a TS-1 Catalyst in Continuous Flow: A Machine-Learning-Driven Approach. ACS Sustainable Chemistry Engineering. https://doi.org/10.1021/acssuschemeng.5c07402
Abstract:
A titanosilicate possessing an MFI structure was synthesized and characterized using a suite of ex situ and in situ techniques to ensure that the produced catalyst has isolated Ti4+ centers in a tetrahedrally coordinated environment. First, the catalyst performance was evaluated using a batch process for allyl alcohol oxidation in the presence of H2O2. Subsequently, we developed an automatable flow reactor system coupled with a machine learning optimization algorithm to improve the production of glycidol through this catalytic route. Importantly, the reaction was carried out using water as the sole solvent, demonstrating that high selectivity can be maintained in an aqueous medium without the need for organic solvents typically required to suppress hydrolysis side reactions. An optimized turnover frequency of 4.33 h–1 was achieved, representing a significant improvement over the batch process (3.08 h–1). The favorable mass transfer and catalyst-to-substrate ratio in the flow reactor allowed for efficient use of the active sites, achieving 14% conversion in just 12.6 s, with selectivity control achieved through simply tuning the residence time. Our modified autoencoder model proved to be an effective predictor for the sparse data set and outlier-prone results, outperforming conventional machine learning frameworks through simultaneous dimensional reduction and reaction yield prediction. This dimensional reduction approach also facilitated the identification and interpretation of divergent solutions offered by the optimization algorithm. These findings reinforce the potential of integrating machine learning with flow catalysis to develop more efficient and sustainable chemical processes.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the A*STAR - Industry Alignment Fund - Pre-Positioning
Grant Reference no. : A20G9b0135

This research / project is supported by the A*STAR - NA
Grant Reference no. : C231218004
Description:
This document is the Accepted Manuscript version of a Published Work that appeared in final form in ACS Sustainable Chemistry & Engineering, copyright © American Chemical Society after peer review and technical editing by the publisher. To access the final edited and published work see 10.1021/acssuschemeng.5c07402.
ISSN:
2168-0485
2168-0485
Files uploaded:

File Size Format Action
aapaper-manuscript-20251126-finalised.pdf 495.99 KB PDF Request a copy