In ultra-high vacuum systems, obtaining the composition of a mass spectrum is often a challenging task due to the highly overlapping nature of the individual profiles of the gas species that contribute to that spectrum, as well as the high differences in terms of degree of contribution (several orders of magnitude). This problem is even more complex when not only the presence but also a quantitative estimation of the contribution (partial pressure) of each species is required. This paper aims at estimating the relative contribution of each species in a target mass spectrum by combining a state-of-the-art machine learning method (multilabel classifier) to obtain a pool of candidate species based on a threshold applied to the probability scores given by the classifier with a genetic algorithm that aims at finding the partial pressure at which each one of the species contributes to the target mass spectrum. For this purpose, we use a dataset of synthetically generated samples. We explore different acceptance thresholds for the generation of initial populations, and we establish comparative metrics against the most novel method to date for automatically obtaining partial pressure contributions. Our results show a clear advantage in terms of the integral error metric (up to 112 times lower for simpler spectra) and computational times (up to 4 times lower for complex spectra) in favor of the proposed method, which is considered a substantial improvement for this task.
Reconstruction of Ultra-High Vacuum Mass Spectra Using Genetic Algorithms