Unsupervised machine learning techniques for the analysis of mass spectra from CHACE-2 aboard Chandrayaan-2 orbiter

Chandra’s Atmospheric Composition Explorer-2 (CHACE-2), a neutral gas mass spectrometer aboard the Chandrayaan-2 Orbiter, measures the mass-to-charge ratio (m/q) of collected gases to evaluate the composition of the lunar exosphere. The CHACE-2 instrument generates mass spectra, with peaks corresponding to the mass-to-charge ratio values of elements, atoms, or molecules detected in the sampled gas. Operating in its default mode, it produces approximately 94 mass spectra per hour of observation, resulting in the accumulation of an extensive dataset over several years.This study addresses the critical challenge of analyzing this vast and complex dataset to derive meaningful insights into the lunar exosphere’s composition. To achieve this, we employed artificial intelligence (AI)-driven methodologies, marking the first application of the deep learning-based clustering technique, Deep Embedded Clustering (DEC), in the analysis of lunar exospheric mass spectrometer data. Principal Component Analysis (PCA) was utilized for dimensionality reduction, revealing dominant patterns in the data, while clustering methods identified distinct group of elements present in the exosphere. Notably, DEC successfully uncovered intricate patterns and complex relationships within noisy datasets, outperforming traditional clustering techniques. The methodologies demonstrated here also provide a robust framework for analyzing mass spectra obtained from similar instruments in future space and planetary missions.