Continuous flow synthesis of N,O-dimethyl-N'-nitroisourea monitored by inline FTIR: comparing machine learning and kinetic modeling for optimization

An Su,Jiapeng Guo,Guihua Luo,Kejie Chai,Weike Su

DOI: https://doi.org/10.26434/chemrxiv-2024-dzrcd

2024-03-15

Abstract:The synthesis of N,O-Dimethyl-N'-nitroisourea, crucial intermediates in pesticide manufacturing, was explored through a substitution reaction between O-methyl-N-nitroisourea and methylamine within a novel continuous flow microreactor system, featuring FTIR inline analysis for real-time monitoring. This study embarked on a comparative analysis between two optimization approaches: the contemporary machine learning-based Bayesian optimization and the traditional kinetic modeling. Remarkably, both strategies obtained a similar yield of approximately 83 % under equivalent reaction parameters---specifically, an initial reactant concentration of 0.2 mol/L, a reaction temperature of 40 °C, a molar ratio of reactants at 5:1, and a residence time of 240 minutes. The Bayesian optimization method demonstrated a notable efficiency, achieving optimal conditions within a mere 20 experiments, in contrast to the kinetic modeling approach, which required a more laborious effort for model formulation and validation. Despite the long-standing reliance on kinetic modeling for its detailed insights into reaction dynamics, our findings suggest its relative inefficiency in optimization tasks compared to the machine learning-based alternative. This study not only highlights the potential of integrating advanced machine learning methods into chemical process optimization but also sets the stage for further exploration into efficient, data-driven approaches in chemical synthesis.

Chemistry

What problem does this paper attempt to address?

This paper mainly explores the optimization method for synthesizing N,O-dimethyl-N-nitrosoisourea through a continuous flow microreactor system, and utilizes online FTIR for real-time monitoring. Two optimization strategies were compared in the study: machine learning (based on Bayesian optimization) and traditional kinetic modeling. Under similar reaction parameters, both methods achieved approximately 83% yield. Bayesian optimization found the optimal conditions within 20 experiments, while kinetic modeling required more experiments to construct and validate the model. Although kinetic modeling can provide in-depth insights into reaction kinetics, it is less efficient in optimization tasks. The paper emphasizes the potential of integrating advanced machine learning methods into chemical process optimization and lays the foundation for further exploration of data-driven chemical synthesis methods. Additionally, the paper provides a detailed description of the experimental materials, methods, and result analysis, including the setup of the continuous flow microreactor system, sample analysis, and optimization process.