Anomaly classification by inserting prior knowledge into a max-tree based method for divertor hot spot characterization on WEST tokamak

Abstract

The divertor of WEST (W Environment in Steady-state Tokamak) is the main component for plasma control and exhaust. It receives high heat fluxes, which can cause damage to plasma facing units above the allowable heat flux. Improving the operation safety on the actively cooled tungsten divertor is being researched in place at WEST, toward providing divertor monitoring solution for ITER. Divertor operation safety relies on detecting, monitoring, and classifying all hot spots on the divertor surface using infrared (IR) cameras. In this paper, a method based on max-tree representation and attributes of IR images is used to classify normal from abnormal strikelines on the divertor. The proposed method requires only high-level prior knowledge of abnormal temperatures and divertor structure but does not require any labeled data, unlike existing methods, such as support vector machines (SVMs) or convolutional neural networks (CNNs). The max-tree classifier method is tested on real IR images from the WEST tokamak and shows that 88% of hot spots are accurately classified with a small enough calculation duration that can be performed between two pulses.