For this post, I tried to measure the performance of my DIY air purifier, the Corsi-Rosenthal box. I wanted to compare its performance against natural ventilation, and commercial ones like the Levoit 400s home purifier. The main metric I needed to beat was the clean air delivery rate (CADR), which is how much ‘polluted air’ is filtered in an hour for a given type of pollutant (like smoke) - it has units in cubic metres of air per hour.
Again, this is not a lab grade experiment, but rather a minimal-trust investigation to learn about the subtleties of commercial air quality marketing and measurement.
The Leviot commercial purifier is 220 GBP and claims to have a CADR of 200m^3 / h, although the pollutant(s) used to test it is unclear, and I didn’t manage to dig up the reports for further inspection. If you have an air purifier with really detailed specs, get in touch - I’ve found commercial purifiers to be frustratingly opaque.
To test, I first needed to generate some PM2.5 in a controlled fashion. Thanks to the internet, I decided to fry some kale in a systematic way. In hindsight, I should’ve burned some matches, which I will do next time.
I did two trials. The first was to test the pollution reduction by just opening my window. The second was to close the window, and test my CR box on its highest setting. In theory, the initial level of PM2.5 doesn’t matter since we are measuring the rate of decay. But, I wanted to keep things consistent and controlled, in the spirit of ‘only changing one thing’ - so I designed a repeatable PM2.5 generating procedure!
For each trial,
I sealed my room.
I heated 30g oil on a fixed setting for 2.5 minutes.
I fried 120g of kale for 10 minutes
Each time, the PM2.5 stabilised at around 450 micrograms / m^3.
At that stage, I then either opened the window, or turned on my box at max setting. I then recorded the results with my PM2.5 / PM10 monitor at a fixed position.
In the plot below, the blue lines represent the particle concentration decay under natural ventilation and opening my window. The red line represents the particle concentration decay with my CR box on max setting. You can see that my CR box is winning by quite a big margin versus just natural ventilation.
Assuming a baseline level of PM2.5, I then took the log regression to get the decay factor.
Multiplying the decay factor by the volume of the room gives the CADR.
I think for most homes, measuring the size of their rooms is hard, especially if you are renting and have no access to planning documents. Most rooms also aren’t rectangular. One pretty good app I used is magicPlan, which takes a LIDAR scan and is able to export the scan as a plan with surface area. Then all you need to do is multiply this by the height (8ft for my ceiling).
After doing the multiplications, I measured the CADR for PM2.5 as 300 cubic feet per minute, or 550 m^3 / h. Surprisingly, this was larger than the commercial air purifier which has a CADR of 400 m^3 / h.
I learned that my CR box is surprisingly effective. But at a closer glance, doing this experiments have lead me to believe that CADR measurements are more subtle than people think, and I think we have reason to be careful when taking CADR values at face value.
Doing this exercise gave me some insight into why clean air delivery rate measurements aren’t as clear cut as I thought. A few things here that make me worry about how this CADR was measured.
In my case, I only measured the PM2.5 / PM10 at only one spot. For this to be valid, we would need to assume the room was well mixed. Otherwise, the concentration decay could just be attributed to the pollutant spreading out normally. One thing to fix this would be using multiple PM2.5 sensors around the room and taking an average of the decay factors. But buying pro air quality monitors are expensive, so I’m thinking about making an arduino version of this with the component parts and then calibrating it to my m2000.
Stuff like this:
Another way to fix this would be to install more fans in the measuring space so that the air in the room is more well mixed throughout the experiment.
Secondly, CADR as a measurement depends on the volume itself. This means that it adds an extra degree of freedom in the experiment. In this case, I used a LIDAR app to take measurements of my house, of which I expect a degree of error.
Also, CADR might change depending on the type of pollutant you are measuring. Here, we are measuring smoke from burning a vegetable. But you can’t expect the CADR measured from a different source of pollutant to be the same, especially if the particle size < 2.5 microns. I don’t know if my pollutant measurement for the air purifier that I built is the same as the one for the commercial
We also have real world imperfections. My DIY box is pretty loud, so I don’t put it on full strength. I also don’t put it on all the time. And I have no idea how this does with actual viruses or pathogens. The upshot of the commercial market is that the air purifiers tend to be quieter, and therefore are on most of the time.
I’ve learnt that there are loads of subtleties when trying to measure how effective a purifier is, but I hope to actually get some commercial ones to do the kale test to prove that the CR box can outmatch them on more improved tests.