Neural network multi-component gas mixture analysis with broadband dual-frequency comb absorption spectroscopy

Abstract The cross interference of gas species in absorption spectroscopy is one of the most challenging obstacles for the analysis of multi-component gas mixtures with overlapping absorption features (blended spectra). We propose a multi-component gas mixture sensor combining a broadband absorption spectrum acquisition with a spectrum analysis algorithm. The sensor features a mid-infrared dual-frequency comb laser source enabling sensitive measurements in a broad spectral interval combined with a deep learning algorithm for spectral analysis to accurately identify the species and retrieve the concentrations of the gas components in the mixture. The sensor is tested with gas mixtures of three common gas species, namely methane, acetone and water vapor. The architecture tuning and model training are achieved by a physics-informed augmented dataset. The proposed spectral analysis model is evaluated firstly by comparison with two other state-of-the-art neural network algorithms (2L-ARNN and 1D-CNN). The performance of the complete sensor is then assessed by real-time measurements in realistic detection scenarios. In addition, we systematically analyzed and presented explicit visualizations explaining the inner working mechanism of the considered algorithms. The high performance of the proposed sensor suggests that it is feasible to realize schemes for more general gas mixture analysis by integrating broadband optical sensing and deep learning models.