Near-infrared dual-gas measurement of ethylene and ethane based on Noise-Suppressed NMF

Dissolved gas analysis (DGA) is a critical method for transformer fault diagnosis. Medium-to-high-temperature thermal fault is a key focus in transformer condition monitoring, with its characteristic gases primarily dominated by ethylene (C₂H₄) and ethane (C₂H₆). Traditional detection methods based...

Full description

Saved in:
Bibliographic Details
Published in:Sensors and actuators. B, Chemical Vol. 449; p. 139161
Main Authors: Gao, Huifang, Zhang, Zhirong, He, Chongjun, Lu, Yuangang, Wang, Le
Format: Journal Article
Language:English
Published: Elsevier B.V 15.02.2026
Subjects:
Cross-interference
Ethylene ethane mixed gas
Laser absorption spectroscopy
Non-negative Matrix Factorization (NMF)
Laser absorption spectroscopy
Cross-interference
Ethylene ethane mixed gas
Non-negative Matrix Factorization (NMF)
ISSN:0925-4005
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Dissolved gas analysis (DGA) is a critical method for transformer fault diagnosis. Medium-to-high-temperature thermal fault is a key focus in transformer condition monitoring, with its characteristic gases primarily dominated by ethylene (C₂H₄) and ethane (C₂H₆). Traditional detection methods based on mid-infrared (MIR) band absorption spectroscopy face limitations in manufacturing maturity and high costs. C₂H₄ and C₂H₆ exhibit weak and severely overlapping absorption cross sections in the near-infrared region (NIR). The small absorption cross-sections leads to significant noise, which in turn causes large errors in multispectral separation and makes it difficult to achieve high-precision cross-interference suppression. This study pioneers the application of Non-negative Matrix Factorization (NMF) for cross-interference suppression. By decomposing mixed-gas secondary harmonics (2f) into products of the 2f of individual component gas and their concentrations, this method isolates individual gas signals from overlapping spectral bands. Furthermore, to address the issue that noise affects the accuracy of cross-interference suppression, a Noise-Suppressed NMF model is established through rigorous noise-propagation analysis. Experimental results demonstrate that, using a 3-meter absorption cell at 1683.25 nm, Noise-Suppressed NMF achieves average relative errors of −0.57 % for C₂H₄ and −0.005 % for C₂H₆ across 15 calibrated C₂H₄ and C₂H₆ binary mixtures. Additionally, under severe cross-interference, absolute errors of 1.5 ppm for 40.6 ppm C₂H₄ and 2 ppm for 73.72 ppm C₂H₆ are attained. This work establishes a novel paradigm for resolving weak, highly overlapping NIR absorption bands, thereby enabling high-reliability, low-cost C₂H₄ and C₂H₆ quantification in industrial gas monitoring. [Display omitted] •Proposed Noise-Suppressed NMF resolving gas cross-interference in weak NIR bands.•Developed matrix decomposition model linking 2f signals to gas concentrations.•Achieved −0.57 % C₂H₄ and −0.005 % C₂H₆ errors in 15 calibrated mixtures.•Constrained errors to 1.5 ppm C₂H₄ and 2 ppm C₂H₆ under severe interference.
ISSN:0925-4005
DOI:10.1016/j.snb.2025.139161