Physics-Informed Neural Operators for Real-Time Multiscale Simulation and Control of Turbulent Multiphase Reactive Flows in Petrochemical Systems

The accurate simulation and real-time control of turbulent multiphase reactive flows in petrochemical systems represent one of the most formidable challenges in computational engineering. These flows involve tightly coupled phenomena spanning orders of magnitude in spatial and temporal scales from molecular-level combustion kinetics to reactor-scale hydrodynamics rendering conventional numerical approaches computationally intractable for online control applications. This article presents a comprehensive framework based on Physics-Informed Neural Operators (PINOs) that embeds governing physical laws directly into the operator learning architecture, incorporating the Navier Stokes equations, species transport, turbulence closure models, and multiphase interfacial dynamics. By training simultaneously on sparse observational data and physics residuals, the proposed framework achieves high-fidelity surrogate modeling capable of real-time inference at a fraction of the cost of direct numerical simulation (DNS) or large eddy simulation (LES). The framework is demonstrated on three industrially relevant petrochemical benchmarks: fluid catalytic cracking (FCC) risers, steam cracking furnaces, and gas–liquid bubble column reactors. Results show speedups of three to four orders of magnitude relative to high-fidelity solvers, with prediction errors consistently below ۲% for key quantities of interest. Integration with model predictive control (MPC) loops demonstrates the framework's potential for closed-loop process optimization under uncertainty.

کلیدواژه ها:

Physics-Informed ، Neural Networks ، Neural Operators ، Turbulent Reactive Flows ، Multiphase Flows ، Petrochemical Systems ، Real-Time Control ، Multiscale Modeling