Calibrating Particulate Matter Sensors

Particulate Matter (PM), and more specifically PM with aerodynamic diameters less than 2.5 micrometers, is a commonly measured indoor and outdoor pollutant since PM has been linked to a variety of health outcomes. Like with carbon dioxide, we calibrate the PM sensors in two environments:

1. controlled, environmental laboratory chamber
2. mock home environment

For both these environments, the process is the same and shown below:

Environmental Chamber

The environmental chamber doesn't represent a real-world environment, but helps us calibrate these devices with greater ease and confidence.

Each of the four experiments (3 for model generating, 1 for testing) were conducted in the following manner:

1. Start data collection on the Beacons and reference monitor
2. Run for 5 minutes
3. Inject particles (PTI Arizona Test Dust A1) with median diameters between 3 and 5 micrometers from a hand-powered nebulizer until the reference monitor reads at least 50 micrograms per cubic meter (typically 2 squeezes from the nebulizer).
4. Allow the concentration to slowly decay for the remaining experiment time

During the experiment, one rotating house fan is used to induce mixing and the ventilation system is not active. The only ventilation provided to the space is by infiltration through small cracks between the double doors.

Univariate, linear least-squares regression models are created for each of the first three experiments and the two model parameters -- intercept and slope -- are averaged across these three experiments. The models are then applied to data from a fourth experiment to assess the model's performance. The image below represents an example outcome from the PM2.5 calibration efforts in the environmental chamber.

Mock Home Environment

We perform a similar experiment in the UTest House. Beacons are setup in the open kitchen area in close proximity to the reference monitor.

Each of the four experiments (3 for model generating, 1 for testing) were conducted in the following manner:

1. Start data collection on the Beacons and reference monitor
2. Run for 5 minutes
3. Inject particles (PTI Arizona Test Dust A1) with median diameters between 3 and 5 micrometers from a hand-powered nebulizer, squeezing three times to generate particles. Particles were injected above an angled box fans which would blow and disperse particles over the devices shown in the image above.
4. Allow the concentration to decay to normal conditions -- typically around 30 minutes -- and again generate particles. This time, use the nebulizer four times, injecting over the box fan again.
5. Allow the concentration to decay back to normal for the remaining experiment time.

During the experiments, four box fans spread around the combined kitchen, living, and dining area to help induce mixing. The ventilation system is turned off so "fresh" air is provided through infiltration only. Due to the size of the space, particle concentrations decrease rapidly and generating concentrations similar to those in the chamber is challenging. Ideally, the concentrations we generate in both spaces should be similar so that our models are fit to similar data.

Univariate, linear least-squares regression models are created from data gathered during each of the first three experiments and the two model parameters -- intercept and slope -- are averaged across these three experiments. The models are then applied to data from a fourth experiment to assess the model's performance. The image below represents an example outcome from the PM2.5 calibration efforts in the UTest House.