Optical sensor for BTEX detection: Integrating machine learning for enhanced sensing.

Area of Science:

• Environmental Science
• Analytical Chemistry
• Machine Learning

Background:

• Optical sensors offer rapid detection of volatile organic compounds (VOCs) like BTEX for environmental monitoring and industrial safety.
• Accurately quantifying individual gases within a BTEX mixture presents a significant analytical challenge, particularly for low-concentration species such as benzene.

Purpose of the Study:

• To develop and validate a machine learning model for simultaneous, precise quantification of BTEX concentrations in gas mixtures.
• To overcome the challenge of detecting low-absorbance gases like benzene within complex mixtures.

Main Methods:

• Utilized an absorption spectroscopy gas sensing system integrated with a convolutional neural network (CNN).
• Generated a synthetic dataset via physics-based simulations, encompassing diverse BTEX mixture compositions and concentrations relative to permissible exposure limits (PEL).
• Implemented a novel 3-stage approach to enhance discrimination between individual BTEX components.

Main Results:

• Achieved high predictive accuracy with R-squared values > 0.99 for toluene, ethylbenzene, and o-xylene.
• Demonstrated strong performance for benzene detection, with R-squared values > 0.96, despite its low absorbance.
• The model accurately predicted BTEX concentrations across various mixture scenarios.

Conclusions:

• The developed CNN-based model precisely quantifies simultaneous BTEX concentrations from absorption spectra.
• The 3-stage solution effectively addresses the challenge of detecting low-absorbance gases like benzene.
• This approach significantly advances real-time BTEX monitoring capabilities for environmental and industrial applications.