Numerical Simulation of Inductively Coupled Plasma Based on Machine Learning

Plasma stealth technology effectively reduces the radar cross-section of a target by utilizing the electromagnetic scattering properties of plasma, with its performance primarily determined by electron density. However, traditional experimental and numerical simulation methods often struggle to rapidly acquire large volumes of data, which limits the efficiency of parameter design. This paper investigates the electron density distribution of inductively coupled plasma (ICP) using deep learning methods and employs the Newton-Raphson-based optimizer (NRBO) algorithm to optimize the hyperparameters of the deep neural network (DNN), achieving high-precision predictions. The model demonstrates low loss values for both the training and test sets, thereby verifying its reliability. Furthermore, the model was utilized to further explore the influence of power, pressure, and coil position on electron density. The results indicate that electron density significantly increases with rising power and pressure, while the position of its peak varies with changes in coil position. The prediction results of the NRBO-DNN model under multiple sets of varying parameter conditions are highly consistent with the simulation model, demonstrating that the model can effectively regulate the magnitude and distribution of electron density by altering power, pressure, and coil position. This capability meets the requirements for the rapid design of electron density distribution in plasma stealth applications.