Direct coupling of microkinetic and reactor models using neural networks

Microkinetic modelling facilitates a detailed description of chemical reactions and has been an important tool in heterogeneous catalysis research to resolve mechanisms at the molecular level. For industrial processes however, transport phenomena inside the chemical reactor can strongly influence performance. In order to understand the interplay between kinetics and transport properties, embedding of microkinetics inside reactor models is necessary yet inhibited by the computational cost of microkinetic simulations. To relieve this issue, we have developed artificial neural networks to substitute microkinetic models and improve computational performance. These networks are capable of accurately reproducing reaction rates across different regimes while obeying chemical conservation laws. A workflow is detailed here for the automated construction of the networks and the cost-efficient generation of the required training datasets using adaptive mesh refinement. These techniques allowed for construction of a reactor model for Fischer-Tropsch synthesis with full mechanistic detail. This model elucidated the undesired promotion of CO2 formation at higher conversions following re-adsorption of in-situ generated water, inhibition of this effect through formation of a carbon reservoir and lowered catalyst activity due to mass transfer limitations.

Automated summary

Microkinetic modelling is a crucial tool in studying heterogeneous catalysis but transport phenomena in chemical reactors affect performance. To overcome computational cost, artificial neural networks are used to substitute microkinetic models and improve efficiency. A reactor model for Fischer-Tropsch synthesis reveals the undesired promotion of CO2 formation due to re-adsorption of in-situ generated water.