A sparse sensing and Chemical Reactor Network based framework for the development of physics-based digital twins of combustion devices

This paper introduces an approach that leverages sparse sensing techniques to construct digital twins of combustion systems using models grounded in fundamental physical principles. These models possess varying levels of fidelity. Specifically, we employ CFD simulations of an ammonia-fuelled multi-stage rich-lean combustor, from which an equivalent Chemical Reactor Network was built and able to replicate the input–output behaviour of the system. The individual reactors, each corresponding to a specific zone of the combustor, operate as soft sensors, offering temperature and composition measurements at distinct points within the combustor domain. By applying linear sparse sensing techniques, we successfully reconstruct the CFD fields for selected variables, leading to the development of a reduced-order model for the system. The reactor network-sparse sensing method was compared against a reduced-order modelling approach available in the literature, coupling Proper Orthogonal Decomposition and Gaussian Process Regression. The reactor network-sparse sensing framework showed improved extrapolation capabilities towards scarcely explored operating conditions thanks to its physics-based nature. The proposed approach represents an attractive solution to develop digital twins of combustion systems enabling broad design space explorations, real-time predictions under time-varying operating conditions, faster optimization and control strategies, and informed decision-making.