bg
News
18:22, 22 November 2026
views
1

Penza Researchers Move Away From Computational Grids in Fluid Dynamics

Researchers at Penza State University have developed and registered a software package for solving fluid dynamics problems using physics-informed Kolmogorov–Arnold networks.

The development makes it possible to work without computational grids and is aimed at the aviation industry and the oil and gas sector. To set up a problem, users need to specify the mathematical model, the geometry of the domain, the properties of the medium, and the initial and boundary conditions. After training, the neural network provides a continuous approximation of the velocity and pressure fields.

The research is being carried out under a state assignment from the Russian Ministry of Science and Higher Education. The package was tested on the Kovasznay flow, the Taylor–Green vortex, and flow in a square cavity with a moving lid. The absolute error did not exceed 0.001, training took about a minute, and velocity and pressure values are computed in fractions of a second.

"The main feature of Kolmogorov–Arnold networks is that the usual numerical weights of connections between neurons are replaced by learnable functions of a single variable," explained Vladimir Gorbachenko, Doctor of Technical Sciences, Professor, and Head of the Computer Technologies Department at Penza State University.
like
heart
fun
wow
sad
angry
Latest news
Important
Recommended
previous
next