Low-Cost Formaldehyde Sensor Evaluation and Calibration in a Controlled Environment

Abstract

Formaldehyde is a carcinogenic indoor air pollutant emitted from common wood-based materials. Low-cost sensing of formaldehyde is difficult due to inaccuracies in measuring low concentrations and susceptibility of sensors to changing indoor environmental conditions. Currently gas sensors are calibrated by manufacturers using simplistic models which fail to capture their complex behaviour. We evaluated different low-cost gas sensors to ascertain a suitable component to create a mobile sensing node and built a calibration algorithm to correct it. We compared the performance of 2 electrochemical sensors and 3 metal oxide sensors in a controlled chamber against a photo-acoustic reference device. In the chamber the formaldehyde concentrations, temperature and humidity were varied to assess the sensors in diverse environments. Pre-calibration, the electrochemical sensors (mean absolute error (MAE) = 70.8 ppb) outperformed the best performing metal oxide sensor (MAE = 335 ppb). A two-stage calibration model was built, using linear regression followed by random forest, where the residual of the first stage acted as a input for the second. Post-calibration, the metal oxide sensors (MAE = 154 ppb) improved compared to their electrochemical counterparts (MAE = 78.8 ppb). Nevertheless, the uncalibrated electrochemical sensor showed overall superior performance hence was selected for the mobile sensing node.

Remote Sensor untuk Monitoring CO dan NO2 Terpublikasi Online

“…Pemantauan kualitas udara (CO, CO2, PM2.5, suhu, dan kelembaban) dalam bentuk pemantau di dalam ruangan menggunakan internet of things dan jaringan sensor nirkabel [5] [6] [7] . Penelitian lain tentang pemantauan kualitas udara adalah penelitian yang berkaitan dengan perbandingan beberapa sensor murah [8]- [10] [9] [10] . Dibandingkan dengan penelitian sebelumnya, kami menggunakan sensor dengan harga yang relative murah, menempatkan sensor di titik tertentu yang berpotensi menimbulkan gas berbahaya yang masuk dalam jenis gas buang hasil pembakaran sesuai dengan kadar pencemar yang ada pada ISPU.…”

Laboratory Performance Evaluation of a Low-Cost Electrochemical Formaldehyde Sensor

“…In addition to, some other studies of formaldehyde sensor performance have presented mean absolute error (MAE) [26]. MAE is given by:…”

“…Some laboratory studies have evaluated the performance of low-cost formaldehyde sensors. Chattopadhyay et al [26] evaluated the performance of two electrochemical sensors and three metal oxide sensors in a laboratory chamber over the concentration range of 10-800 ppb, a temperature range of 22-50 • C, and a RH range of 8-85%. Deng et al [27] developed a metal oxide sensor based on hierarchical flower-like CuO nanostructure and evaluated the sensor response to a formaldehyde concentration range of 50-1000 ppb.…”

“…First, most of the studies focused on sensor fabrication and provided limited sensor performance metrics. Most of the studies reported the linearity of response [28][29][30], but only a few studies reported limit of detection (LOD) [27,30], sensor accuracy [26], or sensor repeatability [28,29]. Second, most of the studies focused on formaldehyde concentration ranges that were higher than typical indoor (~17 ppb) [31] or ambient (~3 ppb) [32] levels in the United States.…”

The development of a meta-learning calibration network for low-cost sensors across domains

“…Post-calibration refers to the calibration process undertaken after sensor deployment, devised to eliminate error sources that could destabilize sensor performance in actual deployment environments, thereby augmenting the measurement precision and data quality of the LCSA. The calibration models currently use a variety of methods ranging from simple statistical learning methods [15][16][17][18][19] to complex deep learning techniques [20][21][22][23], from a "one-to-one" approach where the measured values at the current moment are input to output the calibrated values at that moment, to a "many-to-many" approach where the measured values at a set of consecutive moments are input to output the calibrated values at the corresponding moment [20]. Most calibration models require the LCSA to be co-located with a reference station for a long period of time to obtain sufficient data for training [23] and to adapt to current patterns of environmental pollutant change.…”