Autonomous Optimization of Continuous-Flow Clothianidin Synthesis via Dynamic Kinetic Modeling and Multiobjective Bayesian Optimization

Abstract

We report the continuous flow synthesis of clothianidin as a model system, in which a reaction kinetic model was rapidly established through dynamic experiments. However, kinetic modeling alone provides no strategy for resolving trade-offs between conflicting objectives. Multiobjective Bayesian optimization (MOBO) addresses this gap by mapping the Pareto front across the parameter space, with the kinetic model serving as a virtual experiment surrogate within the iterative cycle to replace physical experiments entirely. The coupled framework simultaneously minimized the E-factor and maximized the production rate, yielding a well-resolved Pareto front that quantifies the trade-off between these two objectives. Optimization was further extended to a pilot-scale continuous flow system incorporating a static mixer, with transfer learning employed to leverage prior knowledge and reduce experimental burden at the new scale. A maximum production rate of 1.32 g/min and a minimum E-factor of 1.53 were achieved via Meta-Learning-assisted MOBO (Meta-MOBO), with an average reaction yield of approximately 79.8% across the scale-up optimization. These results demonstrate that coupling kinetic modeling with MOBO and transfer learning offers an efficient and scalable strategy for bridging laboratory-scale synthesis and industrial production.