(372t) Data-Driven Approach to Automated Reaction Process Analysis

In the fields of reaction engineering and process systems engineering, there has traditionally been a focus on developing mechanistic models based on physical and chemical mechanisms. Focusing on organic reactions, the kinetic study approach is a useful method that comprehensively encompasses the elucidation of reaction mechanisms to the construction of models. Despite the potential effectiveness of the kinetic studies, their application in organic chemistry remains limited [1]. Nevertheless, starting from around 2005, the volume of research dedicated to the kinetic studies of organic reactions has been steadily increasing [2–5]. In addition, various new methods were introduced for kinetic analysis, such as reaction progress kinetic analysis [1, 5, 6] and variable time normalization analysis [7, 8]. Recently, there are several studies that integrate machine learning techniques to automate kinetic analysis, exemplified by initiatives like AI-DARWIN [9] and organic reaction mechanism classification using machine learning [10]. However, conventional machine learning approaches often necessitate extensive datasets for both training and evaluating the models.

This study presents a data-driven exploration aimed at automating the extraction of reaction mechanisms from time-series concentration data in chemical reactions, alongside the construction of corresponding physical models. We utilized dynamic mode decomposition (DMD) as our primary tool for determining reaction pathways and reaction rate coefficient values directly from the temporal data of reactants and product concentrations. Our research concentrated on the amination and Grignard reactions, both of which are widely used across various chemical processes. This study extended its scope by examining a diverse array of reaction mechanisms, diverging from the foundational reactions under consideration. For each reaction mechanism, we developed comprehensive mechanistic models and generated virtual experimental data through simulation. The implementation of the DMD technique allowed for the precise determination of reaction orders and kinetic parameters from this simulated dataset. As a progression of our work, we plan to undertake the validation of our developed methodology against actual experimental data.

[1] D. G. Blackmond, “Reaction Progress Kinetic Analysis: A Powerful Methodology for Mechanistic Studies of Complex Catalytic Reactions”, Angew. Chem. Int., 2005, 44, 4302.

[2] P. Sagmeister, C. Schiller, P. Weiss, K. Silber, S. Knoll, M. Horn, C. A. Hone, J. D. Williams, C. O. Kappe, “Accelerating reaction modeling using dynamic flow experiments, part 1: design space exploration”, React. Chem. Eng. 2023, 8, 2818.

[3] J. Kim, Y. Hayashi, S. Badr, K. Okamoto, T. Hakogi, H. Furukawa, S. Yoshikawa, H. Nakanishi, H. Sugiyama, “Mechanistic insights into amination via nucleophilic aromatic substitution”, React. Chem. Eng. 2023, 8, 2060.

[4] T. Cordero-Lanzac, A. T. Aguayo, A. G. Gayubo, P. Castano, J. Bilbao, “Simultaneous modeling of the kinetics for n-pentane cracking and the deactivation of a HZSM-5 based catalyst”, Chem. Eng. J. 2018, 331, 818.

[5] D. G. Blackmond, “Kinetic Profiling of Catalytic Organic Reactions as a Mechanistic Tool”, J. Am. Chem. Soc. 2015, 137, 10852.

[6] J. S. Mathew, M. Klussmann, H. Iwamura, F. Valera, A. Futran, E. A. C. Emanuelsson, D. G. Blackmond, “Investigations of Pd-Catalyzed ArX Coupling Reactions Informed by Reaction Progress Kinetic Analysis”, J. Org. Chem. 2006, 71, 4711.

[7] J. Bures, “Variable Time Normalization Analysis: General Graphical Elucidation of Reaction Orders from Concentration Profiles”, Angew. Chem. Int. 2016A, 55, 16084.

[8] J. Bures, “A Simple Graphical Method to Determine the Order in Catalyst”, Angew. Chem. Int. 2016B, 55, 2028.

[9] A. Chakraborty, A. Sivaram, V. Venkatasubramanian, Comput. Chem. Eng. 2021, 154, 107470.

[10] J. Bures and I. Larrosa, “Organic reaction mechanism classification using machine learning”, Nature, 2023, 613, 689.