The industrial partner of the Novosibirsk State University, the Sigma-Pro company, turned to the engineers of the Laboratory of Applied Digital Technologies of Novosibirsk State University with a request to develop software for a new class of problems in experimental studies in the field of fluid and gas mechanics. For example, to develop anti-icing agents for aircraft, to study aerosol drug delivery systems, to create spray cooling systems, to apply paintwork, etc.
Ksenia Borisova and
Timur Garipov, students of the Engineering School of the Department of Mathematics and Mechanics of NSU, who are doing an internship in the laboratory, were involved in automating the process of object recognition using machine vision methods.
The head of the laboratory, Doctor of Physical and Mathematical Sciences
Rustam Mullyajanov said:
Our laboratory is engaged in the development and implementation of algorithms for solving applied problems of mechanics, hydrodynamics and heat and mass transfer, bioengineering and natural language processing by combining the results of supercomputer modeling, experiment and modern machine learning methods. At the moment, Sigma-Pro is developing its own software for experimental research and testing in the field of fluid and gas mechanics. The guys from the Engineering School are regularly involved in our projects, and this case was no exception.
Under the guidance of laboratory employee
Ivan Plokhikh, the students learned to recognize drops of water spray from a nozzle in a supersonic flow using neural networks and process the data obtained, namely, to collect information about the centers of mass, sizes, velocities and movements of droplets in paired images.
Timur Garipov commented:
My role in the project was to mark up the data for training the model, as well as to develop and modify a neural network for segmenting drops in the image. Initially, the customer provided data in the form of a set of pairs of sequential images. Then, together with Ivan, I was engaged in the selection of architecture and modification of the solution to obtain a suitable result. I had a positive impression of working with Sigma-Pro. I think it was an important and valuable experience to work with an industry representative while I am still a student. I am extremely glad that I was able to apply my knowledge in an applied task.
At the moment, the developed algorithm has been successfully implemented in the ActualFlow software for the hardware base of the POLIS complexes, which was previously launched jointly with the Kutateladze Institute of Thermal Physics SB RAS, and is already expected in some research organizations and scientific divisions of commercial companies.
