Intelligent Computing On Atmospheric RF Discharges Coupled With Kinetic Data From PIC-MCC Model

Atmospheric RF discharge plasma has a wide range of applications in material surface modification,biomedicine,energy and environment,etc.However,the micromechanisms of its production process and application process are very complex,so further research is necessary to investigate the influence of discharge parameters on plasma properties.Traditional experimental methods or numerical simulation tools always have many limitations.In recent years,thanks to the booming development of computer equipment and machine learning algorithms,deep learning algorithms have provided many revolutionary tools and methods for modern science.It has been demonstrated that deep learning algorithms also have promising applications in the field of plasma science.In this thesis,a deep neural networks(DNNs)with multiple hidden layers are designed to predict the fundamental properties of atmospheric RF discharge plasma based on PIC-MCC data.The training set of DNNs is constructed by the data generated from the PIC-MCC model,and the DNNs based on this training set are able to predict various kinetic behaviors of atmospheric RF discharges with high accuracy.The main contents of this thesis are as follows:(1)A DNNs model is constructed based on the physical data characteristics of atmospheric RF discharge plasma given by the PIC-MCC model,and the effectiveness of the DNNs model in predicting the basic physical properties of atmospheric RF discharge plasma is verified by comparing the computational results of both PIC-MCC and DNNs models.The computation time of the conventional PIC-MCC model is between 10 and 100 hours,while the trained DNNs model takes only 0.01 second,which is 106～107 times more efficient,and thanks to the generality of the DNNs model,the traversal range of the discharge plasma for a given parameter can be greatly expanded.(2)In the study of the frequency effect of atmospheric RF discharge plasma based on DNNs,the electron density distribution of atmospheric RF discharge plasma is always saddle-shaped when the other discharge parameters are kept constant and only the excitation frequency of the power supply is changed,the electron density and the peak discharge current increase linearly with the increase of the excitation frequency.Along with the contraction of the sheath region,the bulk plasma region in the discharge space gradually expands,and the electric field in the sheath region gradually increases,while the electric field in the intermediate plasma region is always kept at a small value.The EEDF results predicted by DNNs show that with the increase of excitation frequency,the percentage of low-energy electrons gradually decreases while the percentage of medium-energy electrons increases,which leads to the gradual evolution of the EEDF curve from a three-temperature distribution to a Maxwell distribution.(3)In the study of the spatial scale effect of atmospheric RF discharge plasma based on DNNs,the discharge is always in a uniform discharge state and the electron density distribution is a single-peak distribution structure dominated by the intermediate plasma region when the other discharge parameters are maintained constant and only the discharge gap is changed.With the increase of the discharge gap,the plasma potential in the discharge space slightly increases,while the electric field intensity in the sheath layer on both sides slightly decreases.The heating of electrons mainly occurs in the sheath layer region,and the secondary electrons emitted at the boundary are ionized when they are close to the bulk plasma after being heated by the electric field in the sheath layer,and then gradually relax and lose energy under the action of the RF electric field in the bulk plasma region.With the increase of the discharge gap,the percentage of low-energy electrons in the EEDF curve gradually increases,while the percentage of medium-energy and high-energy electrons gradually decreases,accompanied by the decrease of the medium-energy part and high-energy tail of the EEDF curve with the increase of the gap.Based on the training data provided by PIC-MCC,the DNNs model exhibits considerable accuracy and extremely high computational efficiency in predicting the properties of atmospheric RF discharge microplasmas.Compared with the time-consuming PIC-MCC,the DNNs model can predict the plasma electron density distribution,electric field spatial distribution,EEDFs and other characteristics in a given parameter range in a very short time,which provides a kinetic explanation and theoretical support for the industrial application of atmospheric RF discharge plasma,provides a new computational method for the numerical simulation of atmospheric pressure microplasma,and expands the deep learning algorithms for further applications in the field of plasma science.