Intelligent Computation Of Discharge Plasma Based On Fluid Simulation

In recent years,with the development of gas discharge technology at atmospheric pressure,the application of low temperature plasma has drawn widespread concern in plasma medicine,pollution prevention,energy conversion and other fields.At atmospheric pressure,low temperature plasma has more diverse boundary processes and the interaction between the plasma and the material to be treated is more complex,often involving hundreds of particles and thousands of chemical reactions.The multi-particle species and frequent collisions will impose a great computational burden on the numerical simulations at atmospheric pressure.For these reasons,atmospheric plasma requires a more efficient computational method for simulation.Deep neural networks are computational tools that can analyze and utilize large data to solve different target problems and can solve the discharge characteristics of atmospheric plasma accurately and efficiently.The main contents of this thesis are as follows:(1)Atmospheric dielectric barrier discharges and radio frequency discharges are widely used to produce low temperature plasma in various applications,which are usually numerically investigated by fluid model.In this thesis a deep neural network together with a universal hidden layer structure is developed to explore the characteristics of atmospheric pure helium dielectric barrier discharges and radio frequency discharges.At the same time,one-dimensional fluid models for atmospheric dielectric barrier discharges and radio frequency discharges are constructed,and the computational data from the fluid model is used as the initial training sets.(2)Deep neural networks are used to study the current density,electron density,ion density,and electric field of the atmospheric dielectric barrier discharge with voltage input,and the comparison with simulated data from the fluid model demonstrates that deep neural networks can be accurately and efficiently applied to the study of atmospheric plasma.Meanwhile,the practicality of deep neural networks for solving low temperature plasma is demonstrated by solving the dielectric barrier discharge model with multiple control parameters to show that the constructed deep neural networks have good generalization performance for multiple input properties as well.(3)For the current-input radio frequency discharge model,the effectiveness of the algorithm is demonstrated by solving its macroscopic and microscopic discharge characteristics such as the gap voltage,the average electron density,ion density,electric field,and electron temperature in one cycle,and by comparing it with the existing experimental findings and simulation results to prove that deep neural networks can be applied to a wide range of discharge types and their different stages.Numerical simulations are an important way to study the discharge characteristics of low temperature plasma at atmospheric pressure.In this thesis,deep neural networks are constructed to replace the fluid model to accurately and efficiently study and calculate the atmospheric dielectric barrier discharge and radio frequency discharge characteristics as an example to demonstrate the superior capability of deep neural networks to study low temperature plasma.The training set of deep neural networks in this thesis consists of fluid simulation data,but deep neural networks can also be trained using experimental data.The computational results show that,given a suitable training set,the constructed deep neural networks can describe the characteristics of atmospheric plasma with almost the same computational accuracy(error less than 1%)as the fluid model,while the computational efficiency is much higher than that of the traditional fluid model,with an average time of 0.01 s for solving a set of data,which is more than 5 orders of magnitude more efficient.The examples in this thesis show that the combination of deep neural networks and fluid model will greatly improve the efficiency and effectiveness of atmospheric discharge plasma simulations and deepen the understanding of discharge plasma.