Sparse Sensor Placement for a Cyclonic Burner under MILD Combustion Conditions

Abstract This study focuses on the development of a Reduced-Order Model based on a Sparse Sensing framework to reconstruct the thermochemical state of a cyclonic burner operating under MILD combustion with \((CH_4)\)/air reactant mixtures under atmospheric pressure. A RANS-based dataset, consistent with performed experiments, is decomposed via Proper Orthogonal Decomposition and coupled with QR Column Pivoting algorithm to optimize sensor placement under practical configurations (Unconstrained, Region- and Distance-constrained, Predetermined). The operating conditions consider a nominal thermal power 6-8 kW and \((Φ∈[0.6,1.1])\), where 33 simulations were performed of which 30 were used for training and 3 for testing. The performance to reconstruct and predict temperature and \((XCO_2)\) was evaluated for all the scenarios. The approach achieves \((≈)\)6% median relative reconstruction error, remaining typically \((≤)\)10% in the most challenging cases, while reconstruction time is \((≈)\)0.1-0.15 s, enabling near real-time monitoring. Furthermore, a trade-off analysis between the number of retained modes and sensors, quantification of the propagation of sensor uncertainty, and the computational costs for each constraint strategy was conducted. Results support the development of soft sensors and digital twin applications for MILD combustion, with validated performance at the outlet and in the chamber mid-plane, and methodical guidance for constrained sensor layout design.