Sparse data recovery of tomographic diagnostics for ultra-large-area plasmas

Abstract

As the size of plasma systems reaches multiple square meters, several phenomena lead to non-uniform plasma production; as a result, the requirement for real-time monitoring of plasma uniformity is increasing, particularly for industrial plasmas. Although non-intrusive diagnostics (e.g. optical emission spectroscopy) are preferred to monitor such plasmas, line (or volume)-integrated measurement is indispensable. Thus, special attention has been paid to the tomographic reconstruction technique; however, limitations remain for tomographic diagnostics for ultra-large-area plasmas. Here, we report an inversion technique that remarkably improves the performance of tomographic reconstruction for large-area industrial plasmas. A computational tool for the tomographic reconstruction with Phillips–Tikhonov regularization was prepared, and reconstruction performance tests were conducted using artificial (phantom) images (namely, expected plasma emission distributions). While very sparse line-of-sight and projection data result in poor reconstruction performance, two suggested operators, which constrain the reconstruction solution, noticeably improve the reconstruction performance and reduce the overall reconstruction error. Although the projection data are contaminated by random noise (<5% of data), the reconstruction error was improved by 43.8% and 40.5% for flat- and hollow-shaped phantoms, respectively. The computational analysis clearly demonstrates that tomographic plasma diagnostics with the proposed inversion technique can be used as a real-time tool for detecting arcing and monitoring plasma spatial uniformity in large-area industrial plasmas.