A BAYESIAN DATA DRIVEN MULTI-FIDELITY MODELLING APPROACH FOR EXPERIMENTAL UNDER-SAMPLED FLOW RECONSTRUCTION

This work showcases a multi-fidelity approach, which couples experimental techniques with numerical models within a Bayesian data-driven framework, thus enabling a direct uncertainty assessment. With such an approach, a substantial reduction of the instrumentation effort can be achieved, making it particularly interesting for high intrusiveness applications. A CFD model is employed to provide an initial belief of the flow, which is then updated based on random undersampled experimental observations in a multi-fidelity Gaussian Process framework. The approach is applied to a small aspect ratio axial compressor for which an experimental database is available. With a 70% random undersampling of a complete measurement test, the results exhibit a correct mean flow reconstruction with clear signatures of the typical flow features. The reconstructed flow error is comparable to the experimental measurement uncertainty. An assessment of the approach's robustness shows an influence of the location of the measurements on the flow reconstruction that decreases with increased flow measurements.