ML-Based Humidity and Temperature Calibration System for Heterogeneous MOx Sensor Array in ppm-Level BTEX Monitoring

Abstract

Recently, indoor air quality is an important issue for human health and high concentrations of toxic Volatile Organic Compounds (VOCs) gases such as BTEX (Benzene, Toluene, Ethylbenzene, and Xylene) are very harmful to our respiratory system and metabolism. To detect BTEX gases at indoors, Metal Oxide (MOx) sensors are widely used because of their low-cost and high sensitivity. MOx sensors are easily affected by temperature and humidity, hence it is difficult to detect BTEX gases accurately without additional calibration process. In this paper, we present the calibration system for heterogeneous MOx sensor array where machine learning (ML)-based techniques, Linear Regression (LR), Non-Linear Curve Fitting (NLCF), and Artificial Neural Network (ANN), are exploited to reduce the impact of temperature and humidity. For the performance evaluation, we have setup the gas concentration measurement system and recorded the sensor outputs from Temperature-Cycled Operation (TCO) responses of five heterogenous MOx sensors. The proposed calibration system with ANN-based calibration system shows the reduction of gas sensors variation due to temperature and humidity 73% on average, and presents maximum 92% reduction for benzene, 75% for toluene, 83% for ethylbenzene, and 91% for xylene gases, respectively.