Physics-informed neural networks enhanced planar laser-induced fluorescence technique: Reconstructing flow fields of supersonic gas flow into the low-pressure region

Abstract

Acquiring flow field information under supersonic low-pressure conditions poses a critical challenge. In such conditions, traditional particle tracer measurement methods struggle to provide sufficiently small tracer particles to track the flow accurately. However, planar laser-induced fluorescence techniques can capture concentration data but typically do not provide enough insights into parameters such as velocity and pressure, hindering the understanding of flow characteristics in these extreme environments. We enhanced the planar laser-induced fluorescence technique via deep learning methods in this paper. The proposed approach employs physics-informed neural networks to process planar laser-induced fluorescence measurement data, thereby inferring data that traditional methods are unable to measure, and reconstructing the complete flow field. This method was demonstrated for the flow field under extreme conditions with a peak Mach number of 9.8 and a back pressure of 213 Pa. The relative L2-norm error of the inferred velocity data was 23.76 % compared to the results of a computational fluid dynamics simulation. The findings of this research demonstrate the capability of the deep learning-based method to capture challenging data under extreme conditions, thereby offering novel insights into fluid dynamics.