| Abstrakt: |
In recent years, non-thermal plasma (NTP) has received an increasing attention for in situ resource utilization of CO 2 in the Martian atmosphere. As an important approach to exploring the underpinning physics of NTP, fluid models with tens of species and hundreds of reactions are very time-consuming in simulating CO 2 plasmas under Martian conditions, especially driven by the nanosecond pulsed voltage. In this paper, a deep neural network (DNN) with multiple hidden layers is proposed as an example to replace the fluid model to accurately describe the essential discharge features of CO 2 pulsed discharge under Martian conditions. After trained by the data from the experimental measurements or numerical simulation and continuously optimized to minimize the loss function, the constructed DNN can achieve a satisfied prediction performance. Compared to the fluid model, the DNN takes only a few seconds to predict the discharge characteristics and profiles of the electric field and particle density, especially to show the spatial–temporal distribution of the given products in CO 2 plasmas, such as CO 2 + , CO 3 − , CO2v1. This study indicates that a DNN can efficiently yield the essential characteristics in CO 2 pulsed discharge even with plenty of species involved in seconds, strongly showing the potential ability to be a highly efficient numerical tool in NTPs with multiple temporal–spatial scales. [ABSTRACT FROM AUTHOR] |
