This study introduces an AI-driven, multiscale modeling approach to improve catalytic performance in reactions on metal catalysts, with a focus on propane dehydrogenation (PDH)—a critical process for sustainable propylene production. Using machine learning, we predict adsorption energies, activation barriers, and reaction pathways on a range of mono- and bimetallic surfaces, traditionally calculated with density functional theory (DFT). Through microkinetic modeling, we identify key reaction pathways and rate-determining steps, streamlining this complex model into a simplified Langmuir-Hinshelwood kinetic model for practical application. The AI-driven, scaling relationship-based microkinetic modeling enables rapid catalyst screening, pinpointing surface properties that influence propane activation and selective C-H bond scission. This approach advances the efficient design of catalysts for PDH, supporting sustainable propylene production with enhanced selectivity and stability.

At the mesoscale, reinforcement learning optimizes catalyst nanostructures and support materials, enhancing transport properties and reducing side reactions like cracking and coke formation. This modeling provides insights into reaction kinetics and diffusion in catalyst pores, crucial for maintaining selectivity and prolonging catalyst life.

At the macroscale, AI-driven process modeling enables reactor-level simulations, optimizing temperature, pressure, and feed composition to support stable and efficient propylene production. The model further integrates real-time adjustments for catalyst deactivation, improving operational consistency and yield.

Our findings reveal critical descriptors—such as adsorption energy and electronic properties—that strongly correlate with catalytic performance, guiding the rational design of stable, selective catalysts for chemical processes such as propane dehydrogenation, and acetylene semi-hydrogenation. By minimizing computational and experimental demands, this AI-driven, multiscale modeling approach provides a robust, scalable framework for advancing chemical technologies. Ultimately, this study contributes to the development of efficient, energy-saving catalytic processes, enabling sustainable innovations in catalytic science through AI-driven multiscale insights.