About the authors
| First name, Middle name, Last name, Scientific degree, Scientific rank, Current position. Full and brief name of the organization, The organization address. | Arslan M. Ziganshin, doctor of technical sciences, associate professor, Kazan State University of Architecture and Engineering, Kazan, Russian Federation Email: This e-mail address is being protected from spambots. You need JavaScript enabled to view it , ORCID: 0000-0001-7335-7797 Vadim I. Shakirzyanov, HVAC design engineer, Stroyexpert Testing Center, Kazan, Russian Federation Email: This e-mail address is being protected from spambots. You need JavaScript enabled to view it , ORCID: 0009-0002-3045-5797 |
| Title of the article | Approximation of numerical simulation results for a “tee-sharp elbow” unit using a convolutional neural network |
| Abstract. | Although numerical simulation of ventilation system components using computational fluid dynamics (CFD) methods currently considered as a primary research and design tool, it has a significant drawback: it consumes significant computational resources. This slows both research and the implementation of results in design practice. The objective of this study is to apply a convolutional neural network to approximate velocity and pressure fields to accelerate the calculation of separated flows and aerodynamic drag in a tee-elbow ventilation unit. To achieve this goal, a method was developed for converting data from Ansys Fluent into a format suitable for neural network training; data was generated with information on the boundary conditions of the geometry and airflow parameters; a convolutional neural network architecture was implemented; a model was trained using the available data; and the approximation accuracy was assessed using a test sample of unit configurations. Results. An approach has been developed in which, in the first stage, numerical simulation results are converted into a matrix format and used for deep learning. The input data includes binary masks, signed distance fields, and boundary conditions. A network with residual and throughput connections was chosen as the architecture. Training was then performed on an expanded dataset using normalization and a combined loss function that accounts for field gradients. Subsequently, a study was conducted and hyperparameters were tuned, resulting in a neural network model that provides a fairly high accuracy in predicting velocity components: the determination coefficient exceeds 0.98, and the average error is 8–12%. The accuracy of the pressure field approximation is somewhat lower, at approximately 27.5%. Conclusions. The developed method allows for faster and more reliable determination of the pressure distribution and velocity components in tee-elbow units. The results obtained can be used in the calculation and design of energy-efficient ventilation system components. |
| Keywords. | convolutional neural network; flow field approximation; numerical modeling; computational fluid dynamics; ventilation units; engineering systems design. |
| For citations: | Ziganshin A.M., Shakirzyanov V.I. Approximation of numerical simulation results for a “tee-sharp elbow” unit using a convolutional neural network // News of KSUAE, 2026, № 2 (76), p. 100-121, DOI: 10.48612/NewsKSUAE/76.7, EDN: FUGVEE |
