Integrating data assimilation and sparse sensing for updating a digital twin of a semi-industrial furnace

Abstract

This study presents a data assimilation (DA) framework that combines a simulation-based digital twin (DT) with a sparse sensing (SpS) strategy using experimental data. This approach continuously enhances the DT model with newly available data from numerical simulations and experiments. The DT, built by coupling Proper Orthogonal Decomposition (POD) and Gaussian Process Regression (GPR), is based on 49 Reynolds-averaged Navier–Stokes simulations of a semi-industrial combustion furnace, covering a range of operating conditions in terms of fuel inlet mixture, equivalence ratio, and air inlet velocity. The experimental campaign utilizes Laser Rayleigh Scattering (LRS) to map the temperature field in the combustion furnace. The SpS model is employed to project the experimental data into a low-dimensional manifold. Afterwards, DA is carried out to obtain an updated set of coefficients within that manifold. The assimilated solution leads to a DT with enhanced predictive capabilities. The findings highlight the potential of this approach to improve the accuracy of DTs through the integration of experimental and numerical data.