Time-resolved three-dimensional temperature field reconstruction of soot flames using physics-informed neural networks

In this study, a novel time-resolved three-dimensional (3D) temperature field reconstruction algorithm based on Physics-Informed Neural Networks (PINNs) was proposed, in which a custom-developed 3D tomographic flame imaging system is employed. By synchronously acquiring full-field emission images at dual wavelengths (550 nm and 650 nm) through a nine-channel optical fiber bundle and a high-speed camera, the system provides non-intrusive, time-resolved temperature measurements of unsteady sooting flames. By integrating the two-color pyrometry principle, the algorithm accurately reconstructs spatial temperature distributions and captures their temporal evolution. Numerical studies demonstrate the accuracy and superiority of the proposed algorithm. Compared with traditional algebraic iterative reconstruction methods, the proposed approach reduces computation time from 233 s to 79 s while achieving a lower mean relative error of 6.07%. Experimental validations demonstrated that the reconstructed temperature results in the laminar stable flame showed good agreement with the previously proposed CMAE method, whilst in the swirling unstable flame, the results effectively illustrate both the spatial distribution and dynamic variations of the flame temperature. Methodologically, the proposed framework integrates physical constraints, sparse-view tomography, and data-assisted completion into a unified reconstruction paradigm. This method provides a novel and efficient approach for 3D full-field diagnosis in combustion and reacting flows to achieve on-line diagnostics in real combustion equipment.