Automated and robust Langmuir sweep analysis using machine learning

Swept Langmuir probes are used to deduce temperature, density, and electric potential in laboratory plasmas. Traces from swept probe measurements can be difficult to interpret using existing hand-tuned heuristics, and are restricted to using only analytical probe models. Using an unsupervised hybrid model of neural networks and analytical theory, I constructed an automated sweep analysis routine that is robust to noise and provides plasma parameters for a semi-infinite planar probe and a Maxwellian plasma. The model was trained on over a million swept Langmuir probe measurements from the Large Plasma Device (LAPD) and the Small Plasma Device (SMPD), and was validated on data from a smaller device. This model can be easily expanded to accommodate any theoretical probe model and an arbitrary velocity distribution function. An overview of the model and a demonstration of its capabilities will be presented. The source code will also be provided.