Machine-learning-based high-temporal-resolution electron temperature evaluation from photomultiplier tube signals in...

In this study, we propose a new method of electron temperature (Te) measurements with higher temporal resolution using a photomultiplier tube combined with machine learning (ML). It has been demonstrated that Te can be evaluated from Balmer series emissions Dα, Dβ, Dγ, and molecular band emission (D2) [Nishijima et al., AIP Adv. 13, 055202 (2023)]. This method is applied to hydrogen plasmas in GAMMA 10/PDX, which is a large tandem mirror device. Balmer series emissions (Hα, Hβ, Hγ) and Fulcher band emissions are measured and correlated with Langmuir probe Te measurements in a divertor simulation experiment conducted in the GAMMA 10/PDX. Training datasets are constructed using these data. Two predictive models based on the random forest algorithm and the support vector regression algorithm are developed, enabling Te estimation directly from spectroscopic data. By employing PMTs with bandpass filters, we achieved Te measurement with higher temporal resolution than the fundamental method of Langmuir probes. Our results highlight the potential of ML-assisted spectroscopy as an alternative diagnostic tool under high heat load conditions and its applicability to feedback control in future fusion devices, including tokamaks.