Deep neural networks for simultaneous BTEX sensing at high temperatures

Area of Science:

• Chemical kinetics and diagnostics
• Spectroscopy and laser-based sensing
• Artificial intelligence in chemical analysis

Background:

• Simultaneous detection of benzene, toluene, ethylbenzene, and xylenes (BTEX) is crucial for chemical reaction studies but challenging due to overlapping spectra.
• Existing methods lack the required sensitivity, selectivity, and fast response for complex, high-temperature environments.
• Trace detection of BTEX species in demanding conditions necessitates advanced diagnostic strategies.

Purpose of the Study:

• To develop and validate a mid-infrared laser sensor system for simultaneous and selective detection of BTEX species.
• To implement deep neural networks (DNN) for spectral deconvolution and species identification in high-temperature shock tube experiments.
• To achieve high sensitivity and fast time response for BTEX analysis in challenging chemical reaction environments.

Main Methods:

• Coupling a mid-infrared laser source with an off-axis cavity enhanced absorption spectroscopy (OA-CEAS) setup in a shock tube.
• Measuring absorption cross-sections of BTEX species at 1000-1250 K and ~1 atm.
• Developing a DNN model to resolve composite spectra into individual BTEX contributions for selective determination.

Main Results:

• The developed DNN model successfully split composite spectra, enabling selective BTEX determination with an absolute relative error of ~11% compared to manometric measurements.
• Minimum detection limits for BTEX species were achieved in the range of 0.73-1.38 ppm at 1180 K.
• Demonstrated the first successful implementation of multispecies detection using a single narrow wavelength-tuning laser in a shock tube via laser absorption spectroscopy.

Conclusions:

• The mid-infrared laser sensor combined with DNN offers a sensitive, selective, and rapid method for simultaneous BTEX detection in high-temperature shock tube experiments.
• This approach overcomes spectral overlap challenges, paving the way for advanced diagnostics in combustion and chemical kinetics research.
• The study highlights the potential of integrating laser spectroscopy with machine learning for complex chemical mixture analysis.