Abstract
Optical sensors provide a fast and real-time approach to detect benzene, toluene, ethylbenzene, and xylenes (BTEX) in environmental monitoring and industrial safety. However, detecting the concentration of a particular gas in a mixture can be challenging. Here, we develop a machine-learning model that can precisely measure BTEX concentrations simultaneously based on an absorption spectroscopy gas sensing system. The convolutional neural network (CNN) is utilized to identify the absorbance spectra for each volatile, along with their concentrations in a mixture. A synthetic data set is generated using a series of physics-based simulations to create the predictive model. The data set consists of the overall absorbance of numerous random BTEX mixtures over time, based on various percentages of the permissible exposure limit (PEL). It is worth noting that benzene has a negligible absorbance (very low PEL, 1-5 ppm) compared to other volatile gases, which makes it difficult to detect. To address this challenge, we introduce a 3-stage solution to accurately discriminate between all BTEX species, regardless of their concentration levels. As a result, the R-squared above 0.99 for toluene, ethylbenzene, and o-xylene, and the R-squared above 0.96 for benzene, is achieved, indicating the model's capability to predict BTEX concentrations.