Parameter identification for gradient porous media flow using sparse data

作者：Haoyun Xing, Guice Yao, Dongsheng Wen · 发表于：Physics of Fluids · 年份：2025 · DOI：10.1063/5.0295332 · 被引用次数：4 · 研究领域：Hydraulic Fracturing and Reservoir Analysis、Lattice Boltzmann Simulation Studies、Enhanced Oil Recovery Techniques

Parameter identification for flow and heat transfer in gradient porous media plays a critical role in engineering design and anomaly detection in applications such as transpiration cooling. In this work, our recently developed TB (trunk-branch)-net physics-informed neural network (PINN) is combined with sparse pressure sensor data to enable the identification of porosity and permeability in dual-domain and multi-domain gradient porous flow, achieving satisfactory results. Meanwhile, the forward inversion results of pressure and velocity fields also exhibit accurate flow trends. Compared with traditional regularization techniques, the use of TB-net PINN eliminates the need of multiple forward solution iterations, enabling complex parameter identification in multi-domain scenarios.