Detection of Hazardous Gas Mixture in the Smart Kitchen Using an Electronic Nose Under Interference Gas Environment

Abstract

Introduction For a long time, due to various factors such as kitchen gas, house decoration and living habits, concentrations of toxic and harmful gases in the kitchen are likely to be much higher than national standards in a short period of time. High concentrations of toxic and harmful gases can cause extremely adverse effects on the human body, which can lead to physical discomfort and even serious accidents such as fires and explosions. In the past few years, some researches focused on the development of a more specific and intelligent gas detection system, including the detection of harmful gases in industrial environments [1], dairy industry [2] and living environment [3]. However, there are few studies on graded, low-cost, wide-range, cheap, multifunctional and simultaneous gas detection systems for mixed harmful gases in kitchen so far. E-nose System The potential toxic and harmful gases in the kitchen are mainly carbon monoxide and methane. However, formaldehyde is also present in the kitchen and acted as interference gas in this study. Therefore, a corresponding detection system was developed to evaluate the concentrations of the mixed harmful gases using a MOS gas sensor array including MP-9, MP-4, MP503, TGS821, TGS816, TGS2602. Nearly half of collected samples contained formaldehyde. Besides, a SHT20 sensor is used to monitor humidity and temperature. This system (Fig.(A)) consists of three parts: the gas blender module, the detection module and the concentration identification module. The gas blender mixes different gases and provides required mixture for detection. All mixed gas samples are diluted by the standard air in order to make the MOS sensors work properly. The standard air is directly discharged into the gas chamber for cleaning. After that, the detection module collects the response of the sensor array to the mixed gas under the control of a microcontroller unit. Here, we adopted a differential amplification and low-pass filter circuit to process the initial sensor voltage signal. Simultaneously, the signals are sent to the software on the computer for further analyzing. Method According to China‘s national indoor air quality standard, combustible gas detection standard and the detection range of MOS sensors, the concentration levels of target gases were set as Fig.(B). In this study, the actual concentrations of different gases in the mixed gas were calculated by the gas blender. The flow rate was set to 1000 mL/min constantly. The experiment process consists of several parts. First, the sensor array was preheated for more than 30 minutes in advance and cleaned in standard air for nearly 30 minutes until the voltage of each sensor was below the corresponding threshold. Next, gas blender injected the mixed gas sample into the gas chamber at a constant flow rate for 5 minutes. In the meantime, the feature parameters of each sensor’s response were extracted and calculated by the MCU in real time. Finally, the sensor array was cleaned for 5 minutes. The extracted features including baseline, maximum positive slope, maximum, peak area, temperature and humidity were all considered in the LDA model. A total of 316 samples were collected. 183 of them contain formaldehyde ranging randomly from 1-40 ppm. In addition, 90% samples for training, 10% for testing. Results and Conclusions During model training, the LDA model can effectively distinguish different levels with or without formaldehyde for both target gases (Fig.(C), (D)). The testing accuracies of LDA model for carbon monoxide and methane without formaldehyde samples are 91.89% and 97.30%, respectively. When there is formaldehyde interference, the accuracies are 80.95% and 92.06%, respectively. Therefore, the interference of formaldehyde has a greater impact on the identification of carbon monoxide. Besides, it is worth noting that carbon monoxide is difficult to distinguish at lower concentration levels, while methane is difficult to distinguish at higher levels. This may be due to the actual detection concentration is close to the upper or lower detection limits of the sensor, causing the sensor to be insensitive at lower levels and too saturated at higher levels. But this study still proves that the system can well meet the requirements of the detection of harmful gases in the interference environment. Figure 1