Optimal sensor placement for flow field classification in an H2 combustor: Foundations and Experimental results

Abstract

Stricter requirements on efficiency, emissions, safety, opera tional flexibility, and costs demand improved monitoring of com bustion in gas turbines. In the research project “KI-PIRINHA” 1, we investigate methodologies for health monitoring of hydrogen flame stability in recirculating combustors. Undesired flow conditions can lead to unstable flame con tainment (flashback), threatening nozzles, combustor integrity, and sensors, while also reducing efficiency and increasing emis sions. To support future detection of unintended combustion regimes, sensor locations must be chosen such that the available measurements are maximally informative. A reduced-order model of the flow is constructed from tran sient CFD simulations, and Laser Doppler Velocimetry (LDV) measurements are performed in a prototype combustion cham ber. Modal decomposition via Singular Value Decomposition (SVD) is used to construct an optimal basis, which is truncated to the most informative modes for sparse sensing. In a second step, linear discriminant analysis (LDA) is applied to further reduce the dimensionality andenableclassification in alow-dimensional discriminator space. Each class corresponds to one of two CFD-based flow regimes defined by distinct inlet conditions. We present the math ematical foundations of the approach, compare optimized and random sensor placement, and demonstrate the applicability of the method in combustion-chamber flow experiments. The results establish the suitability of the proposed methodology as a proof of principle for future investigations under reacting conditions and provide a general framework that can be adapted to a wide range of user-specific applications.