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

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.