Question: What are the advantages to using 2 particle sensors at the same time?

warren is asking a question about purpleair
Follow this topic

by warren | April 18, 2019 17:11 | #19107


Hi all, it makes sense that you get extra credibility/validity if you have 2 sensors (like the #plantower) and they agree (basically for any sensor!). But I wanted to try to break this down with as little jargon as possible -- are there other advantages? Does the Purple Air average the two, or how does it use the two data streams?

I also want to acknowledge that this has been addressed in parts in different questions at #purpleair and #plantower (and #simple-air-sensor) but it's come up from time to time, so I wanted to try to develop a simple answer in one place. Thank you all for helping make this information more accessible!



11 Comments

@CBarnes9 @samr @Tomp @OrionAllgaier @wu_ming2 @guolivar @BrandonFeenstra @kkoerner, any thoughts or info to share on exactly what the #purpleair does with it's 2 sensors' data?

Thank you!

Is this a question? Click here to post it to the Questions page.

So from my experience they're used primarily as a sanity check. I'm not sure of the math behind it, and since PA removed the R2 values that used to be on their old interface, it's hard to say what is "good enough" for how well they correlate. When a sensor is not performing well, the PA map will flag it. I've attached a picture of a flagged vs not flagged sensor (that red "no" indicator I think is what indicates questionable readings or other issues). It looks like the flagged sensor in this case was flagged because there is no second channel data.

This is nice for the reasons you mentioned. I don't think there is any averaging or algorithm that keeps the values close to each other, or else this sanity check wouldn't work. I've seen values that were wildly different between sensors with no response from one sensor vs the other.

The plantower 5003s I've used in my custom monitors have gotten very similar values when compared with a PA-II. So while there could be some adjustment, I don't think it's very noticeable. It very well might be a raw factory reading from the 5003. Though this is anecdotal, so we could probably do more testing.

There are adjustment factors for sensors based on a couple of studies, but I don't think they have to do with the two sensors and are mostly about correction vs. regulatory monitors due to under or over prediction.

Capture.PNG

Capture_2.PNG

Is this a question? Click here to post it to the Questions page.


This is great, thanks @kkoerner -- and interesting to see that they'll flag sensors. I wonder how much two would have to not correlate before it got flagged. Any interest in blocking one of your fan ports on the Purple Air to see if you can get it to flag itself? I may get one of ours out to try this too.

I wonder if it would get flagged immediately, or if they have to be reading different values for a while before it's flagged?

Is this a question? Click here to post it to the Questions page.


OK testing this out now. Here's our sensor, on Cromwell St at our office: https://www.purpleair.com/map?582419|582421#13/41.81161/-71.41883

Note that I covered the fan, but not all the holes in it, but it should substantially reduce how it's able to detect stuff. But I could redo it if need be.

Screen_Shot_2019-05-07_at_5.49.40_PM.png

Now, after lighting a bit of paper on fire near it, it doesn't immediately flag the sensor:

Screen_Shot_2019-05-07_at_5.51.22_PM.png

Here's a picture of the tape over the fan vent:

IMG_20190507_174838.jpg

Is this a question? Click here to post it to the Questions page.


OK, i covered it more thoroughly and will wait a while to report back on what happens.


i'll post more soon, but one is now totally blocked and it shows on the graph, but it's not rejecting the data.

I'll let this run for a day. Then i'll try covering the other one. Looks like I've covered "A"

image_(1).png

Is this a question? Click here to post it to the Questions page.


Aha! One day later, my sensor A has been "downgraded"! I'll move the tape to the other sensor and see what happens.

Screen_Shot_2019-05-08_at_5.27.56_PM.png

Is this a question? Click here to post it to the Questions page.


OK! After now blocking the B sensor instead, it's accepting A data once more, and hasn't yet flagged the B sensor. Interesting!

Screen_Shot_2019-05-08_at_6.23.52_PM.png

Is this a question? Click here to post it to the Questions page.


OK, update - since the B sensor was covered yesterday, it's now showed as "downgraded" just as the A was yesterday:

Screen_Shot_2019-05-10_at_11.44.55_AM.png

My question now, is apart from blocking, what if I put a filter over one sensor so it (presumably) reads lower but not "nothing"?

In this comment above, we saw it report "4" despite that being the Sensor A reading, and I'd blocked Sensor A. It didn't look to be averaging, but we should double-check.

But if we're right about this, does it always show the lower number of the two? Let's see once I put a filter over just one of the two sensors to artificially depress its readings.

Is this a question? Click here to post it to the Questions page.


Reply to this comment...


And just linking to here, where the wiki page documents the 2 channels, which i would guess are from the 2 sensors? https://publiclab.org/wiki/purpleair#Other+things+to+know+about+using+your+Purple+Air

@wu_ming2 notes (in this post:

PA-II includes two laser sensors. An unknown algorithm estimates particulate concentration. I wouldn’t exclude further processing by PurpleAir, maybe matching against an accuracy estimate obtained correlating nearby certified sensors where available. This before finally making PM2.5 averages available on the map and through JSON API.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


@julieta @nanocastro have either of you used 2 sensors in your DIY #plantower -based devices? #maca etc?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Log in to comment