A neural network for the analysis of Langmuir-probe characteristics

Langmuir probes have been widely used in the field of plasma diagnostics for the characterisation of plasma properties. These probes play a crucial role in understanding the behaviour of a diverse range of plasmas, e.g.  edge plasmas in fusion experiments. The measurement of electron density (ne) and electron temperature (Te) provides valuable insights into the plasma's state, stability, and confinement properties. Conventionally, this analysis involves post-experiment fitting methods to calculate plasma properties from the measured current–voltage curves obtained from Langmuir probes. This work introduces a neural-network approach for analysing probe data from the TJ-K stellarator, allowing for fast associative plasma characterisation. The results show a reliable performance on test data within the domain of the training set, predicting both ne and Te within the 10 % intrinsic error. Performance on unseen data outside the domain of the training set was on average within a 26 % and 21 % error on ne and Te, respectively. The network's further abilities, including the identification of low-quality and falsely-labelled data, were also explored. The use of neural networks (NNs) offers fast predictions, enabling further research into real-time applications and live feedback control. This paper highlights the promising role of NNs in enhancing the analysis of Langmuir-probe characteristics.