Automatic mass spectra recognition for Ultra High Vacuum systems using multilabel classification

Abstract

In Ultra High-Vacuum (UHV) systems it is common to find a mixture of many gases originating from surface outgassing, leaks and permeation that contaminate vacuum chambers and cause issues to reach ultimate pressures. The identification of these contaminants is, in general, done manually by trained technicians from the analysis of mass spectra. This task is time consuming and can lead to misinterpretation or partial understanding of issues. The challenge resides in the rapid identification of these contaminants by using some automatic gas identification technique. This paper explores the automatic and simultaneous identification of 80 molecules, including some of the most commonly present in this kind of environment by means of multilabel classification techniques. The best performance is drawn from a dependent binary relevance method trained by extreme gradient boosting. We obtain a Hamming loss of 0.0145 in the test set. The mean binary AUC for the test set was 0.986, and the minimum test AUC was higher than 0.89. A public interactive web app has been developed to allow vacuum users to test the model with their own data.