Evaluating Low Cost Sensor Performance for Use in Industrial Internet-of-Things Platforms: Distributed Sensing Solutions for Urban Area Air Quality Health Monitors

Abstract

The “Array of Things” project in Chicago, IL (U.S) is a ‘smart-cities’ joint initiative funded by the National Science Foundation and involving a partnership between Argonne National Laboratory, the University of Chicago, and the City of Chicago [1]. Several hundred of these platforms are distributed about the city of Chicago and each incorporates seven low-power electrochemical gas sensors for measuring the air quality health of the city. This collection of nodes is the foundation of an Industrial-Internet-of-Things, managed by a public institution, for the benefit of the surrounding communities. It presents a unique opportunity to develop smart sensor algorithms to calibrate these low-cost sensors, thereby providing meaningful and accurate data to map and identify sources of air pollution or to inform the communities of poor air quality days.To create the numerical calibration methods, ozone (O3), nitrogen dioxide (NO2), and carbon monoxide (CO) gas sensors are compared to reference grade monitors, with data collected between March 2018 and June 2019. The gas sensor sub-assembly board was calibrated in the labs at KWJ Engineering, Inc. (Newark, CA) in October 2017. The AoT node was installed and began operating at the EPA ComEd measurement site in Chicago (7801 S Lawndale Ave, Chicago, IL 60652), with help from Pinaki Banerjee of the Cook County Department of Environment and Sustainability. CO reference instrumentation was provided by Max Berkerlhamer of the University of Illinois – Chicago; NO2 and O3 data were from the EPA instruments run by Cook County. Sensor data is collected at 5-minute intervals, while reference data is reported at 1 hour (NO2/O3) and ½ hour intervals (CO). Over the course of 15 months, the reference data showed peak NO2 and O3 values of about 90 ppb, typically with opposite diurnal cycles. During the four months of CO reference monitoring, the range of recorded values were between 150 and 1,500 ppb. To reach these lower detection limits with low cost sensors, methods to remotely zero and span must be developed to overcome the effects of time, as well as outdoor temperature, humidity, and pressure extremes.The main advantage to EPA or research grade instrumentation is the ability to zero and calibrate the equipment at regular intervals. For instance, an EPA site may report an 8-hour moving average for O3 and/or NO2, and this equipment may be zeroed and calibrated on a weekly or daily schedule. For an air monitor mounted on a telephone pole, or internal to a publicly accessible structure (such as a bus stop), calibration methods requiring gas exposure are not feasible. As an alternative method, numerical zeroing using known or anticipated low values can enable an estimation of a zero. These low points can be timed to a weekly or sometimes a diurnal schedule. EC sensors have a typical baseline exponential sensitivity to temperature, which can change over time. Methods to predict this change for the CO sensor increased the R2 correlation from 0.14 when using a static calibration factor to 0.72 when using a time dependent dynamic factor. For determining the appropriate numerical span values ‘nearest neighbor’ methods can be implemented, using either a publicly available reference station, such as an EPA monitoring site, or similarly timing to a known or expected high point. Also important to the sensor stability are averaging methods that can respond quickly to environmental releases of pollutants. Approaches to these methods depend on the targeted gas as well as the sensors' chemistry including the effects of interference from other gasses typically present in air. For example, in this system the NO2 and O3 co-dependence is also shown to change with time. For the NO2 sensor the baseline (diurnal minimum) should decay over time, similar to the CO sensor. However, with increasing sensitivity to O3, the baseline will start to increase, and this pattern can be used to infer the increasing sensitivity to O3.While improvements to sensor design will mitigate some of the shortcomings of low power sensors, this work highlights important numerical methods, and will provide examples to improve the integrity of low-power sensing nodes for future distributed gas sensing networks, allowing for wider implementation of smart-city platforms at a lower cost to the public for implementation.[1] Catlett CE, Beckman PH, Sankaran R, Galvin KK. Array of things: a scientific research instrument in the public way: platform design and early lessons learned, Proceedings of the 2nd International Workshop on Science of Smart City Operations and Platforms Engineering, 26-33 (2017) Figure 1