Machine Learning Transforms Langmuir Probe Plasma Diagnostics

Research Background

Traditional plasma parameter inference relies on analytic approximation theories such as orbital motion theory (OMT), which are only accurate under specific physical conditions like OML regime. When probes operate under more general conditions, standard analytic theories become inaccurate. The research team generated large synthetic datasets by numerically solving the Vlasov-Poisson system and combined them with machine learning techniques including radial basis function (RBF) regression to build predictive models capable of handling broader operating conditions.

Key Findings

• Non-emissive Langmuir probes: Successfully inferred electron density (10^10 - 10^12 m^-3), temperature (0.05 - 0.2 eV), and plasma potential (-2 to +4 V) with relative errors below 2%
• Emissive probes: Plasma potential inference accuracy of 0.22-0.26 V (approximately 4% relative to full range), superior to traditional methods
• Introduced model skill metrics to quantify confidence intervals for inferred parameters, which traditional analytic methods cannot provide
• Inference accuracy on validation test sets matched training sets, demonstrating excellent model generalization

Technical Significance

This research provides practical tools for laboratory and space plasma diagnostics, particularly suitable for aerospace applications such as Hall thrusters and electric propulsion systems. Using pre-calculated synthetic datasets and multivariate regression models, experimental data can be processed in real-time, avoiding the computational costs of traditional iterative optimization methods. The research team is preparing to release open-source tools under GNU GPL license to promote widespread adoption of this technology in plasma diagnostics.