TY - JOUR

T1 - PAINeT: Implementation of neural networks for transport coefficients calculation

AU - Истомин, Владимир Андреевич

AU - Кустова, Елена Владимировна

N1 - Conference code: IX

PY - 2021/7/1

Y1 - 2021/7/1

N2 - In the present study, a possibility of neural networks implementation for evaluation of transport coefficients in atomic gases taking into account electronic excitation and in molecular gases with electronic, vibrational and rotational degrees of freedom is discussed. Atomic nitrogen N and oxygen O, molecular nitrogen N 2 and oxygen O 2, as well as mixtures (N2, N, O2, O) and (N2, N, O2, O, Ar) are considered in the one-temperature approach of the kinetic theory. The results of exact calculations are compared to the neural network-based simulations. It is shown that for single-component gases, the proposed approach yields good accuracy and calculation speedup up to 3 times for atoms and up to 19 times for molecules. The speedup is significant for multi-component mixtures and increases with the mixture complexity, attaining for four- A nd five-component mixtures from 597 to 1196 times correspondingly. Ways to improve the accuracy of neural-network predictions of multi-component mixtures transport coefficients are discussed.

AB - In the present study, a possibility of neural networks implementation for evaluation of transport coefficients in atomic gases taking into account electronic excitation and in molecular gases with electronic, vibrational and rotational degrees of freedom is discussed. Atomic nitrogen N and oxygen O, molecular nitrogen N 2 and oxygen O 2, as well as mixtures (N2, N, O2, O) and (N2, N, O2, O, Ar) are considered in the one-temperature approach of the kinetic theory. The results of exact calculations are compared to the neural network-based simulations. It is shown that for single-component gases, the proposed approach yields good accuracy and calculation speedup up to 3 times for atoms and up to 19 times for molecules. The speedup is significant for multi-component mixtures and increases with the mixture complexity, attaining for four- A nd five-component mixtures from 597 to 1196 times correspondingly. Ways to improve the accuracy of neural-network predictions of multi-component mixtures transport coefficients are discussed.

UR - http://www.scopus.com/inward/record.url?scp=85111984958&partnerID=8YFLogxK

UR - https://www.mendeley.com/catalogue/4998bb5e-1d79-3642-8c5d-526279642d80/

U2 - 10.1088/1742-6596/1959/1/012024

DO - 10.1088/1742-6596/1959/1/012024

M3 - Conference article

VL - 1959

JO - Journal of Physics: Conference Series

JF - Journal of Physics: Conference Series

SN - 1742-6588

IS - 1

M1 - 012024

Y2 - 9 March 2021 through 12 March 2021

ER -