PINN-RANS
Description
This dataset supports the study of indoor airflow reconstruction using a compact single-network PINN-RANS architecture with R-function-based hard boundary constraints. The data were obtained from an Ansys Fluent simulation of a mechanically ventilated room exhibiting a two-vortex airflow structure. The file root_big2.csv contains the full-domain reference solution at 152,781 spatial nodes, including node identifiers, two-dimensional coordinates, pressure, and the x- and y-velocity components. The files point15.csv, point30.csv, and point54.csv contain three sparse observation sets comprising 15, 30, and 54 sampling points, respectively. Each observation file provides node identifiers, spatial coordinates, pressure, and velocity components. These datasets were used to train and evaluate the soft-constraint model (SC), the hard-constraint model without explicit global flow-rate balance (HC-V0), and the hard-constraint model with global flow-rate balance (HC-V1). They enable reproduction of the sparse-data configurations, comparison with the CFD reference field, and evaluation of velocity and pressure reconstruction accuracy.
Files
Institutions
- Harbin Engineering UniversityHeilongjiang, Harbin