Use of a Machine Learning based method to detect anomalous Thermoluminescence Glow Curves (TL-GC) in routine Dosimetry Services

This contribution describes the development of a set of numerical methods based on Machine Learning algorithms to generate an automated classification of experimental Thermoluminescence (TL) Glow Curves obtained routinely by Dosimetry Services. This classification will use experimental data historically recorded by Thermoluminescence Dosimeter (TLD) devices and will be based on the search for possible anomalies in the curves. The classifier tool will ease the labelling of experimental data and the detection of anomalies without previous supervision, implying an improvement in the control evaluations in Quality Guarantee Systems often implemented by Dosimetry Services. Furthermore, this study shows that each curve provides information about the status of each dosimeter, and can be used to perform unsupervised classifications of the measurements. • Machine Learning numerical methods for solving technological problems. • DBSCAN algorithm for classifying experimental thermoluminescence glow curves. • Dosimetry Services quality control.