Inversion of electron mobility in atmospheric radio-frequency plasmas by physics-informed neural networks

Plasma simulation can effectively capture experimental observations in atmospheric radio-frequency (RF) plasmas. Nevertheless, it still poses a challenge to control and optimize experimental processes directly using simulation data. In this study, the Physics-Informed Neural Networks (PINNs) approach is proposed to solve inverse problems, in which the governing equations and observed data are incorporated as constraints into the loss function, and the key physical parameters, such as the electron mobility, are treated as a trainable parameter to be inferred. Based on the given data, the method leverages PINNs to inversely estimate a refined value of electron mobility, and the influence of data location and noise levels on the accuracy of the inferred parameter is also discussed. Since diagnostic data generally contain measurement errors and noise, and computational models are typically based on certain assumptions and approximations, the proposed PINNs framework provides a preliminary and model-consistent way to fine-tune critical parameters in numerical models by integrating physical constraints with sparse observational data. This proof-of-concept study offers a basis for future extensions toward experimental diagnostics and model-driven analysis of atmospheric RF plasmas.