Work

Three-dimensional reconstruction of low-pressure supersonic flows via physics-informed neural network enhanced laser-induced fluorescence imaging

Year: 2025

Type: article

Abstract: Accurate three-dimensional flow diagnostics in low-pressure supersonic environments remain a critical challenge for vacuum-relevant applications. In low-pressure supersonic flows, traditional particle image velocimetry encounters limitations in velocity measurements due to tracer particle inertia effects, while planar laser-induced fluorescence lacks direct velocity measurement capabilities. Additionally, the vast amount of accumulated fluid mechanics data remains underutilized due to its complexity in practical applications. To address these issues, this study proposes a physics-informed neural network enhanced planar laser-induced fluorescence technique for reconstructing three-dimensional velocity fields from concentration data in low-pressure supersonic flows. By integrating experimentally measured concentration fields, auxiliary datasets, governing equations, and entropy-based physical constraints, the method achieves full-field reconstruction. The domain decomposition framework based on compressible flow physics and entropy constraints are introduced to address high-gradient discontinuities. Normal Shock Tables serve as auxiliary references for estimating properties near the Mach disk, where direct measurements are infeasible. Validation on an extremely under-expanded jet at 520 Pa demonstrates a relative L2-norm error of 7.56 % for the reconstructed velocity field, with localized errors of 5.41 % within the isentropic barrel shock region and 14.92 % at the shock front. This method provides a non-intrusive, data-physics fusion approach for flow characterization in low-pressure supersonic environments.

Source: Flow Measurement and Instrumentation

DOI: 10.1016/j.flowmeasinst.2025.103123

Authors: Tingxuan Fan, Tingting Liu, Tianliang Zhao, Shuqin Jia, Jiaxu Li, Ying Huai

Institutions: Dalian Institute of Chemical Physics, Chinese Academy of Sciences, University of Chinese Academy of Sciences

Language: English

Topic: Model Reduction and Neural Networks

Subfield: Statistical and Nonlinear Physics

Field: Physics and Astronomy

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $3,080

Funder: National Natural Science Foundation of China