Abstract

This study presents a deep learning framework for multi-parameter gas sensing by integrating laser absorption spectroscopy with physics-informed neural network. The network is trained with simulated absorption spectra, and no experimental data is required for calibration. The absorption spectroscopic physics is incorporated into the neural networks by the integration of a full-physic spectra generator and a physic-informed loss function, which together enhance the retrieval of gas states from multifold spectral information. This method enables kHz-rate simultaneous measurements of H2O volume fraction, temperature and pressure directly from the absorption spectra near 1343 nm and 1392 nm, achieving respective uncertainties of 1.23×10^-4, 0.22 ℃ and 0.007 atm at room environment. Furthermore, the present framework recovers absolute absorbance from the detection-reference dual-path signals and provides uncertainty quantification for spectroscopic data, demonstrating strong potential for application in complex scenarios.

Related Publications

Publication Info

Year
2025
Type
article
Citations
0
Access
Closed

External Links

View on DOI.org

Social Impact

Altmetric
PlumX Metrics

Social media, news, blog, policy document mentions

Citation Metrics

0
OpenAlex
0
CrossRef

Cite This

Tengfei Jiao, Zhongxing Wan, Kin-Pang Cheong et al. (2025). Enhanced extraction of multiple gas parameters from laser absorption spectroscopy with physics-informed neural network. . https://doi.org/10.1364/opticaopen.30833165.v1

Identifiers

DOI
10.1364/opticaopen.30833165.v1

Data Quality

Data completeness: 66%