A Long Short-Term Memory Network for Plasma Diagnosis from Langmuir Probe Data

Area of Science:

• Plasma physics
• Artificial intelligence in diagnostics

Background:

• Traditional electrostatic probe diagnosis methods face challenges with accuracy due to various influencing factors.
• Accurate measurement of plasma parameters like electron density (N) and temperature (T) is crucial for understanding plasma behavior.

Purpose of the Study:

• To develop a more accurate and rapid method for plasma probe diagnosis using a Long Short-Term Memory (LSTM) network.
• To assess the LSTM network's ability to derive electron density (N) and temperature (T) from Langmuir probe data.

Main Methods:

• Utilized a Long Short-Term Memory (LSTM) neural network, feeding it current-voltage (I-V) characteristic curves from Langmuir probes.
• Trained the LSTM network on a portion of collected data and validated its performance on a separate test set.
• Evaluated prediction accuracy using Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE).

Main Results:

• The LSTM approach demonstrated reduced impact from probe surface contamination compared to traditional methods.
• Accurate diagnosis of underdense plasma was achieved.
• The LSTM network provided more accurate results for electron density (N) compared to electron temperature (T).

Conclusions:

• LSTM networks offer a robust and accurate alternative for plasma probe diagnosis, applicable across diverse discharge environments, including space ionospheric applications.
• The developed LSTM model effectively overcomes limitations of traditional diagnostic techniques, enhancing the reliability of plasma parameter measurements.