Modeling of discharge characteristics and plasma chemistry in atmospheric CO2 pulsed plasmas employing deep neural network

In recent years, non-thermal plasma technology has emerged as one of the most promising candidates for decomposing CO (2). The fluid model, a powerful tool to investigate the plasma dynamics, is computationally costly in simulating complex CO 2 plasma with tens of particles and hundreds of reactions, especially driven by short pulsed voltages. In this paper, a deep neural network (DNN) is proposed to describe the discharge characteristics and plasma chemistry of CO (2) pulsed discharge at atmospheric pressure. The DNN is trained using the simulation data obtained from the fluid model and then continuously optimized by minimizing the loss function. The effectiveness and feasibility of the DNN are verified by comparing with the experimental measurement and the numerical simulation results. Compared to the time-consuming fluid simulations with tens of hours, the well-trained DNN typically requires only a few seconds to obtain the essential characteristics of CO 2 pulsed discharges with high accuracy, significantly improving the computational efficiency. The DNN prediction results show that increasing the pulse rise rate at a given voltage amplitude can effectively raise the discharge current and breakdown voltage, and the electric field in the sheath region also increases with the pulse rise rate. In addition, the density of the surface charge accumulated on the dielectric layer increases with the plateau duration, and then a strong induced electric field by the surface charges is established, which obviously improves the discharge current during the pulse fall phase. The predicted data also show that increasing the pulse rise rate and the plateau duration could effectively improve the density of product species, such as CO and O (2), leading to an increase in CO (2) conversion. This study demonstrates that the DNN method is a reliable tool for obtaining the essential discharge characteristics of atmospheric CO (2) pulsed plasma and provides a promising avenue for future applications of DNN-based methods in non-thermal plasmas.

Automated summary

Non-thermal plasma technology is a promising solution for decomposing CO (2), and this study proposes a deep neural network (DNN) to describe the discharge characteristics and plasma chemistry of CO (2) pulsed discharge at atmospheric pressure. The DNN is trained using fluid model simulation data and shown to provide accurate results in a fraction of the time compared to traditional simulations. The DNN prediction results indicate the influence of pulse rise rate and plateau duration on discharge current, breakdown voltage, electric field, and product species density. Overall, this study demonstrates the reliability and potential of DNN-based methods in non-thermal plasma applications.