Automatic Interpretation of Ion Beam Measurements of Walls in Fusion Machines

Abstract: The purpose of this study is to investigate whether it is possible to automatically interpret theresults of Time-of-flight Elastic Recoil Detection Analysis (ToF-ERDA). And if so, find out if the automaticinterpretation is quicker and/or more accurate than the current approach that consists of manualanalysis. As an added bonus, it is hoped that using an automated technique will enable the identificationof previously undetectable elements.Two different methods were tested, a well-known clustering algorithm known as Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) for its capacity to handle high noise datasets. The other was an image segmentation algorithm known as You Only Look Once (YOLO) for itspattern recognition in abnormal shapes and how straightforward it is to train the algorithm withcustomized image sets.Even though the data was too noisy for HDBSCAN to handle, YOLO was able to quickly and effectivelyidentify most of the elements with just a set of 30 photos to train on. The flaw with YOLO was that inorder to gain the necessary data for the algorithm to train on, a manual input similar to the previousmethods was required. The methods used to navigate these problems and make use of all the data isdiscussed along with possible ways to enhance the clustering based method.