Efficient numerical simulation of atmospheric pulsed discharges by introducing deep learning

Abstract

Plasma simulation is an important but sometimes time-consuming approach to study the discharge behaviors of atmospheric pulsed discharges. In this work, an efficient simulation method is proposed by introducing deep learning to investigate the discharge characteristics driven by very short pulsed voltages. A loss function is designed and optimized to minimize the discrepancy between the Deep Neural Network (DNN) and the verified fluid model. The prediction data obtained via well-trained DNN can accurately and efficiently reveal the key discharge characteristics, such as the waveforms of discharge current and gap voltage, spatial profiles of charged particles density and electric field. The spatial distributions of charged particles density and electric field obtained from DNN are also given to unveil the underlying mechanisms. Additionally, the predictions from deep learning and the formula analysis both highlight that the breakdown voltage and current density can be effectively reduced by increasing repetition frequency, which quantitatively agrees well with the experimental observations. This study provides a great potential promise for vastly improving the simulation efficiency by introducing deep learning in the field of atmospheric plasmas computation.

Plasma medicine: The era of artificial intelligence

“…Zhang et al emphasize the use of DL in the accurate numerical simulation of atmospheric pulsed discharges. [ 71 ] The authors present a novel approach that employs DL to get beyond the computational limitations of conventional numerical techniques. The scientists show that their method can greatly reduce the time that is needed to run simulations by training a neural network with data from numerical simulations, making it a promising tool for exploring complex plasma processes.…”

Review: Machine learning for advancing low-temperature plasma modeling and simulation

“…Data-driven surrogate modeling applied to different atmospheric discharges has been proposed by Zhang et al 47 – 49 A drift-diffusion model taking into account particle balance coupled to Poisson’s equation, and the electron energy conservation equation was used to simulate pulsed He and CO2 discharges.…”

Personalized Plasma Medicine for Cancer: Transforming Treatment Strategies with Mathematical Modeling and Machine Learning Approaches

“…Using information-rich datasets collected from simpler diagnostic methods like optical emission spectroscopy (OES) [129] and electro-acoustic emission and enabling fast, automated data processing and analysis using AI techniques in real-time monitoring and control improves the feedback control, ensuring effective and optimal performance of CAP sources [127,128]. Simulated data [130,131], although not capturing all the complexities in real-world scenarios, complement real data by providing a controlled environment for analysis and experimentation.…”

“…2023, 13, x FOR PEER REVIEW 10 of 19 effective and optimal performance of CAP sources [127,128]. Simulated data [130,131], although not capturing all the complexities in real-world scenarios, complement real data by providing a controlled environment for analysis and experimentation. The manual analysis of the datasets with traditional methods can be challenging due to the complex relations and patterns in the data.…”

“…ML and other computational approaches (Table 1, Figure 6) are designed to handle such complexity and can identify patterns and trends that may not be apparent to human analysis [128]. Plasma plume length Video frames of the plasma plume captured using a camera (iPhone 11) Computer vision algorithms [136] Temperature setpoint Simulated data from thermal dynamics model of plasma-substrate interactions Reinforcement learning [130] Self-adaptive plasma chemistry Gas input densities and energy levels OES Artificial neural networks (DL), gradual mutation algorithm [122] Pulse discharge characteristics (current density and gap voltage) Simulated fluid model data of time and pulse rise rate Deep neural networks (DL) [131] Plasma chemistry (tokamak) FTIR Physics-informed neural networks [137] Figure 6. Common ML components employed for the application of CAP in oncology, highlighting their respective advantages and disadvantages.…”