When I started building DIY air purifiers in Beijing to protect my lungs from this…
…I wanted to test the air in my home, but particle counters were really expensive. Most cost around $1,000 or $2,000 dollars.
But then there was the Dylos, which I bought for about $260.
Just like air purifiers, the technology behind particle counters is actually really simple and cheap. The detector is just a laser and a photo receptor—nothing super expensive. Fortunately for data nerds like me, the market has capitalized on this basic fact and exploded with even-more-affordable air quality monitors.
The machines correlated highly (rs ≥ 0.90) with official PM2.5 measurements.
On average, the three machines deviated from official PM2.5 statistics by less than 10 micrograms.
But now there are even cheaper machines coming out, such as this Huawei Honor华为荣耀 particle counter. This sells for 429 RMB (US$63) in China.
But is it any good?
With the help of Smart Air volunteer LQ in Beijing, we tested the Huawei against official PM2.5 numbers over two days in March, 2018. The data collection was small, just 25 hourly data points.
However, the range in outdoor PM2.5 was wide, from 28 to 213 micrograms per cubic meter. (For reference, the average in Beijing for 2017 was 58 micrograms).
Overall, the Huawei (blue line) tracked the official PM2.5 numbers (orange line) pretty well. However, the Huawei consistently undercounted. At worst, the Huawei undercounted by 52 micrograms. That error is 5 times the WHO annual limit for PM2.5, so that’s pretty big.
On the whole, the Huawei correlated highly (r = 0.88) with the official numbers, although it performed on the low end of all the particle counter Smart Air has tested so far. I should warn that the Huawei test was done at a different time and location than the earlier tests, so some of the differences could be due to the testing.
Next I analyzed the average deviation—on average, how many micrograms of discrepancy was there between the particle counter and the official numbers? This is where the Huawei was clearly the worst performer. It had double the average error of the other machines.