Machine learning assisted analysis of visible spectroscopy in pulsed-power-driven plasmas

Aug 3, 2026

Upload by Anda

Machine learning models predict ion density and electron temperature directly from visible emission spectra produced in a high energy density pulsed-power-driven aluminum plasma generated by an exploding wire array. Radiation transport simulations using PrismSPECT-generated spectral emissivity and opacity determine line-of-sight spectral intensities. Al II and Al III lines show density- and temperature-dependent variations in line ratios and widths. The approach yields a synthetic dataset of intensity spectra to train and compare regression models, with AutoGluon achieving about 98% R2 accuracy and simpler models above 90%, enabling rapid or real-time spectral diagnostics.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

Emission spectroscopy and plasma parameter inference

Collisional-radiative modeling (emissivity and opacity)

Radiation transport and optical thickness

Motivation for ML-based fast spectral analysis

Machine learning assisted analysis of visible spectroscopy in pulsed-power-driven plasmas Rishabh Datta, Faez Ahmed, and Jack D. Hare arXiv :2308 . 16828v1 [physics .plasm-ph] 31 Aug 2023 Abstract—We use machine learning models to predict ion density and electron temperature from visible emission spectra, in a high energy density pulsed-power-driven aluminum plasma, generated by an exploding wire array. Radiation transport simulations, which use spectral emissivity and opacity values generated using the collisional-radiative code PrismSPECT, are used to determine the spectral intensity generated by the plasma along the spectrometer’s line of sight. The spectra exhibit AlII and Al-III lines, whose line ratios and line widths vary with the density and temperature of the plasma. These calculations provide a 2500-size synthetic dataset of 400-dimensional intensity spectra, which is used to train and compare the performance of multiple machine learning models on a 3-variable regression task. The AutoGluon model performs best, with an R2-score of roughly 98% for density and temperature predictions. Simpler models (random forest, k-nearest neighbor, and deep neural network) also exhibit high R2-scores (> 90%) for density and temperature predictions. These results demonstrate the potential of machine learning in providing rapid or real-time analysis of emission spectroscopy data in pulsed-power-driven plasmas. I. INTRODUCTION Spectroscopy is a powerful technique for inferring plasma parameters from emitted electromagnetic radiation. For instance, line widths and line ratios can be used to determine electron density and temperature [1]–[3], velocity can be determined from the Doppler shift of spectral lines [1], [4], and magnetic field strength can be inferred from the Zeeman splitting of line radiation [5], [6] . The wide applicability of spectroscopy makes it an attractive tool for implementation ina variety of laboratory plasmas [1], [4], [7]–[9] . In emission spectroscopy, a typical intensity spectrum can contain several peaks (called emission lines) overlaid on a continuum [1] . The lines correspond to bound-bound electron transitions in the ions of the plasma, while the continuum emission results from free-free (Bremsstrahlung emission) and free-bound electron transitions (recombination radiation) [1],[4] . Line radiation generated by the plasma arises due to either collisional or radiative processes [1] . Collisional processes, such as electron impact excitation/de-excitation and three-body recombination, change the energy levels of bound electrons via collisions with other electrons [1], [4] . Similarly, radiative processes, such as photoexcitation/de-excitation, induce energy transitions due to the interaction of bound electrons with photons [1], [4] . Collisonal-radiative models balance the rates of excitation (and ionization) against that of de-excitation (and deionization), to determine the spectral emissivity and opacity of radiation emitted from the plasma [2] . R. Datta and J. D. Hare are with the Plasma Science and Fusion Center, Massachusetts Institute of Technology, Cambridge, USA F. Ahmed is with the Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, USA The typical approach to determining the ion density ni from emission spectra is to identify lines dominated by Stark (collisional) broadening, and then to compare the line widths with tabulated data [1], [9], [10], or with the predictions of CR codes, such as PrismSPECT [11], [12] . Similarly, for the characterization of electron temperature Te , we typically compare the intensity ratios of two or more lines (typically, inter-stage lines for which density changes have a small effect) with the predictions of CR models [1], [3], [10] . When the plasma is not optically thin, radiation transport, which describes how the energy distribution of radiation changes as it propagates through an absorbing, emitting, and/or

FAQs

How does the method predict ion density and electron temperature from spectra?

It uses machine learning models trained to regress ion density and electron temperature from visible emission spectra intensity features produced by the plasma.

What role does radiation transport modeling play in the workflow?

Radiation transport simulations compute spectral intensity along the spectrometer line of sight using emissivity and opacity values generated by the collisional-radiative code PrismSPECT.

Which model performs best and how accurate are the predictions?

The AutoGluon model performs best, reaching roughly 98% R2-score for density and temperature predictions, while other models like random forest, k-nearest neighbor, and deep neural networks still exceed 90%.

Aug 3, 2026

Upload by Anda