AAAR 32nd Annual Conference
September 30 - October 4, 2013
Oregon Convention Center
Portland, Oregon, USA
Abstract View
An Efficient Algorithm for Very Low Cost Personal Particulate Monitors
MICHAEL TAYLOR, Nourbakhsh Illah, Carnegie Mellon University
Abstract Number: 557 Working Group: Portable and Inexpensive Sensor Technology for Air Quality Monitoring
Abstract In this paper we demonstrate that a weighted estimation algorithm can be effectively used to process noisy or unreliable data from inexpensive mass-produced sensors. Given the rising interest in low-cost portable environmental monitoring, we can expect the availability of higher-quality and lower-cost sensors in the next decade. In the meantime, we propose a method for processing and filtering currently available low-cost sensors to acquire data with higher accuracy and precision than are typically achieved using static formulae or simple time averages and low-pass filters. Specifically, we present an algorithm for estimating 2-micron particle concentrations from an inexpensive dust sensor in addition to a simple calibration method. The concentration estimate is adjusted at every time step by increasing or decreasing depending on whether the sensor is or is not triggered, respectively. The rate of change is weighted according to the probabilistic behavior of the sensor. This takes into account the significantly higher probability of false negatives than false positives. In other words, because we know that we see only a fraction of the particles passing through the inexpensive sensor, we give a much greater weight to detection than absence. We compare our estimate versus the formula given in the datasheet and a scaled time average of the raw signal against reference 2-micron particle counts from the Hach HHPC-6. We show how this dust sensor is incorporated into Speck, a $100 desktop PM monitor for commercial deployment, and AirGo, a wearable PM and carbon monoxide monitor for the EPA's My Air My Health Challenge. We discuss design decisions regarding the cost of including various additional features and the theoretical price points of adding additional functionality. We also outline future plans for developing mobile and wearable monitors based around this inexpensive dust sensor and data processing algorithm.