image credit: @donblair
See and contribute to the up-to-date Wiki page made from this note.
What I Want to Do
We are in the midst of a boom of low-cost air sensing devices, and open source particulate sensing projects have really taken off in the past year. In this note I’ll look at four in-development projects using almost identical sensors (Shinyei PPD42NS, Syhitech DSM501A, Shinyei PPD60PV ) and a low-cost commercial competitor with a similar design, the Dylos.
I'll also lay out a plan for comparing and calibrating these sensors to each other and to commercial reference equipment. Current efforts have correlated low-cost sensors with commercial reference sensors, but never to each other. AQICN, AirBeam, Sonoma Tech Gao et al. 2015, Holstus et al 2014. Thanks to @Willie for many of those links.
How the Sensors Work
All these dust sensors are optical sensors. @DonBlair explains how optical dust sensors work:
There are issues calibrating optical sensors for not just water particles (humidity), but also for the albedo (or brightness) of the particles themselves, as Tim Dye succinctly documents. That said, the biggest differences between different devices is the way they control airflow. Passive monitors rely on convection to draw in air, while active monitors blow a controlled stream of air across the sensor’s path. Which of these strategies is best? I don't know, so I propose co-locating all of these sensors and testing them against benchmark equipment.
Airflow strategies comparison:
Passive airflow: a system that doesn't actively control for airflow, either using natural or thermal convection to drive particulates through.
Active airflow: The use of a fan or pump to control and/or meter airflow through a sensor.
Dylos DC1100
The Dylos uses a small fan. A great teardown is here:
Speck
by Carnegie Mellon’s CREATELab, my previous notes.
the Speck uses a small fan:
hooked up to their open source fluxtream platform and designed for home use, it's being used by the Southwest Pennsylvania Environmental Health Project.
Airbeam
developed by the HabitatMap, it is designed to hook to an android device. It is open source hardware and uses a small fan.
Dustduino
in development by @schroyer and @willie passive airflow via a convection current, it sends data to xively
There are a few more systems using these sensors too, see “other platforms” below.
Benchmark systems
The idea is to use a single Raspberry Pi to read all the sensors in their enclosures simultaneously, along with the reference system.
Optical Benchmark
Currently there are only proprietary commercial services available for commercial sensors, such as Netronix, which costs around $100/month.
I'll connect to a TSI Dusttrak I or II or Thermo Scientific pDR 1500, as per @rjstatic's directions, updated for the Open Pipe Kit.
Filter system benchmark?
Both Dusttrak and pDR-1500 have a filter for "gravimetric analysis" (weighing the quantity of dust collected) as an additional calibration. There are extra questions about characterizing an optical sensors' response to local dust via fingerprinting the different dust particles. That may require an extra system, and I'm looking to source some low cost passive monitors as well.
[image of turtle pump + tube]
23 Comments
@jefffalk @marlokeno interested in your thoughts on this plan.
Reply to this comment...
Log in to comment
This is a great overview, thanks @mathew!
Reply to this comment...
Log in to comment
This is a wonderful overview-I'm going to have to read it more than once to grasp all of this. I wonder if you could explain more for me as a non-scientist about this. Specifically, I'm wondering about light scattering and how that works with 2.5 u silica particles, as the small silica dust is the most dangerous. When I've looked at bigger silica particles under magnification, they are so rounded, have so much sphericity (API 0.7 or better), that I keep wondering whether they'll scatter light as other dust does.
I know most chemicals have signature scattering pattern, and I'm sure dust detector makers have taken this into account So 2 questions, as I look out my glass window-is there any difficulties with the wavelengths that silica scatters, and do micron sized "crystal balls" have any optical properties that an irregularly surfaced particle doesn't? Thanks for any clarification you can offer.
And I have a raspberry pi that I haven't yet put to use; should some builder want one, I'll donate it
p.s- I have some chunks of snowy white Jordan sandstone that I could mail if it would help any device refinements. It's from a roadside, and crumbles easily. Also, I have a ziploc bag of what used to be a piece of Wonewoc sandstone that crumbled to dust-it has more impurities, as it's orange. Plenty of tiny silica.
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
have you come across this one before? http://www.mytzoa.com/#homepage do you know what sensors they use?
It is maller and cheaper than I would imagine is possibly accurate
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
I don't see any images of them running the device-- seems like they demoed it under glass in January. it looks like a non-functional prototype. is it for sale somewhere? they ran an unsuccessful kickstarter but it looks like they're still working: https://www.kickstarter.com/projects/tzoa/meet-tzoa-the-worlds-first-enviro-tracker/
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
I think they are trying a new Indie go go https://www.indiegogo.com/projects/tzoa-wearable-enviro-tracker#/story They've raised 10 grand in the first 7 hours.
They really stress their custom built exposure chamber. Its no fishtank but I guess its okay.
Reply to this comment...
Log in to comment
@nshapiro @mathew @marlokeno Hey, Kevin here from TZOA. I met Shannon Dosemagen of PublicLab at an EPA event last year. Happy to answer any questions. We currently have research units out in the wild of the custom particulate sensor we engineered. We are doing a new production run which will be available in September. Would love input from everyone here. Cheers! -K
Reply to this comment...
Log in to comment
It's my understanding that the Dylos is the only monitor of those mentioned that is calibrated by its manufacturer. That implies that some comparison and adjustment is done to each Dylos so that within limits under the same conditions each one will give the same output. And presumably this is not just a question of correlation. I would wonder about checking one against the others before checking one against multiple samples of itself. I just looked at the article on using the PUWP monitors in China and each one was separately calibrated against a Dusttrak before being positioned. It would also be interesting to hear just what method you are going to use to make the comparisons.
Reply to this comment...
Log in to comment
I pasted Grapherate live graphs for 4 sensors Mathew set up into this page: http://publiclab.org/wiki/open-air -- not sure if these are those used on a live test at the moment, but it's easy to paste in the public keys of Phant streams to generate the graphs. I can also help clean up the graph presentation a bit.
Reply to this comment...
Log in to comment
Or try to get them all on the same graph?
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
@warren @mathew thought this group might enjoy the ongoing testing of another set of low cost PM sensors going on in China! Co-located with the official government monitors in Beijing. Seems to be updated every minute.
http://aqicn.org/sensor//
Reply to this comment...
Log in to comment
kevinvivergy's referred to website is a great example of technology in the hands of techies. They need to learn that more data doesn't always translate into more useful information.
Reply to this comment...
Log in to comment
This is an amazing and incredibly useful page! Thanks all!
Though there is one thing I'm confused about. I've been working with the Shinyei PPD42NS and the datasheet and all examples I can find talk about low pulse occupancy. The hand-drawn diagram above - its seems to me - describes counting the 'high pulses'. I know its some minute detail but I'm just wondering if it makes any difference? The goal is to count the ratio between low and high pulses to calculate the number of particles? I'm a bit confused here.
And has anyone come across any python code to integrate the PPD42NS with a raspberry pi - as well as a simple schematic to drop the +5v from the sensor to the rpi's 3.3 digital read pin? (and can the rpi supply enough power?) I can't seem to find much.
Thanks!
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
I want to note that while Open Pipe Kit development continues, running this comparison has been a low priority since AQICN is running effectively the same thing: http://aqicn.org/sensor//
Reply to this comment...
Log in to comment
I work with scientific dust monitors and have written and published calibration protocols for them. I have been intrigued by these low cost sensors for a while and was wondering how exactly they work. I was intrigued by your description of the duration of the peak determining the size of the particles. For the instruments I work with the height of the peak is measured to determine the particle size instead. The peaks are approximately Gaussian because the light source is a laser and as the particle enters the dim edge of the beam they generate a low signal compared to when they are in the bright beam centre. I wondered if these low cost sensors cap the signal? This would mean that a bigger particle (which scatters more light) would saturate the signal for more time during its transit through the light source.
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
I came across this article about testing the ShinyeiPPD42 in a laboratory environment. I think their evaluation is very insightful and encouraging. @philrosenberg - I believe you are right about the larger particles saturating the signal, the previously mentioned article describes this as happening with particles around ~20µm. A possible solution could be to apply some filter before particles enter the device.
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
Thanks for sharing that, Nicholas. I've started keeping a list of sensor evaluations on the wiki page for optical pm monitors: https://publiclab.org/wiki/optical-pm
Reply to this comment...
Log in to comment
@mathew- As a follower of this discussion, asking couple questions-
First, does electric charge of the particles play a role in dust collection? This question first occurred to me when thinking about dust from sand mines. Silica particles are crushed, and sometimes blasted, they may have a piezoelectric charge. Also, when reading about cloud formation, there are often references to the charge of the particle around which the water condenses.
Most articles discuss sulfates and organic compounds. There's an interesting recent article concerning mineral dusts- http://www.rsc.org/chemistryworld/2013/05/mineral-dust-plays-key-role-cloud-formation-chemistry
2 questions from this- 1] does ionization of the dust particles affect the collection process-how does the charge of the particle interact with the collection instrument? Plus, I'm wondering about static electricity. Dust explosions are another issue that raised my curiosity.
2] In high humidity, is moisture adhering to a fine/ultra-fine particle?
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
@marlokeno Your questions are very good ones. I do not know the details about this instruments, but I do know details based on research grade equipment. In a number of ways the output from these instruments may be so uncertain that these are not important.
However, in research grade equipment yes we worry about static charge on particles causing the particles to adhere to instrument inlets and pipework. This is minimised by using electrically conductive material for pipes such as metal or conductive rubber-like material.
In high humidity you can also get growth of particles, however it is actually bigger particles that adsorb water most easily. Google for Kohler curve for more details. High end instruments dry the air before measuring particles or at least measure the relative humidity for quality control purposes or to try to correct for this effect.
Another issue that you haven't mentioned is known as inlet efficiency. Basically particles bigger than a micron or so have enough inertia that they can deviate from the streamlines of the air entering the instrument. Depending on if the wind is blowing faster or slower than the instrument is sucking and depending on the angle between the wind and the inlet this can cause large particles to be enhanced or diluted in the instrument.
However with these low cost sensors, especially those without active air movement I think that there is likely to be so many possible uncertainties that they are "calibrated" for typical conditions and if they are within a factor of a few most of the time then that's probably all they are aiming for.
There is a really good primer on some of these things at http://aerosol.ees.ufl.edu
Reply to this comment...
Log in to comment
@marlokeno material and charge probably matter quite a bit. There are several research projects using the charge of certain particles for detection. One such passive monitor (for detecting nitrates and sulfates) is : https://publiclab.org/wiki/passive-pm#Ferm+Passive+Monitor
These charge effects on deposition are something we have to test 's something we have to control for in the passive monitors. We have made several prototypes from plastics (much easier fabrication) and are going to compare to aluminum prototypes. I'm still getting the housings cut but I brought one to WI to show. https://publiclab.org/wiki/pm-dev#Problem+Definition
@philrosenberg, Thanks for linking to UFL's primer! I wish I'd seen it before. real nice. I'd love to see your thoughts on the page I illustrated trying to explain the issue of droplet mode particles and humidity: https://publiclab.org/wiki/pm#Droplets
The problems to control for on these electronic sensors are significant, from inlet efficiency, lack of impactors, variation in flow issues and calibration issues from 'wet particle counts. I linked to a few reviews of these systems in my wiki page version of this note: https://publiclab.org/wiki/optical-pm
Reply to this comment...
Log in to comment
@marlokeno "2] In high humidity, is moisture adhering to a fine/ultra-fine particle?"
Good question! At TZOA, we have seen humidity increase the perceived size of a fine particle through an optical device in 100% humidity by as much as 500%. We don't see drastic effects from 0%-70% humidity. The U.S. EPA does education on these types of devices and are actively telling people to only purchase devices that include accurate humidity compensation. We've implemented one of the defacto compensation algorithms that translated into way better R2 correlation to gold standard reference devices. Best, Kev.
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
Just in case anyone on here is interested, I work for the Institute of Climate and Atmospheric Science at Leeds Uni. We actually have some of the Dylos instruments (googling about them is how I found this page) and we are hoping to have a summer student placement this year for someone to do some characterisation work on one to see how good they really are and if we can get anything better out of them. If anyone here is a current undergraduate student in the UK and is interested then feel free to check out http://www.nercdtp.leeds.ac.uk/research-experience-placements/. Please note that there are some eligibility conditions for applying though so please read carefully.
Reply to this comment...
Log in to comment
cool opportunity, @Philrosenberg! you may want to share it out with the Public Lab Air Quality list: http://publiclab.org/lists
@Marlokeno: The question of what type of fine particles are wetted by humidity is one i have too. it certainly effects 'droplet mode' particles significantly, and they make up the bulk of fine particles. Whether it also effects 'dust mode' particles directly, I'm not sure.
This issue cuts to the heart of the difficulties in monitoring silica dust emissions-- the silica can be a fraction of the fine particles, and yet also a health risk. Days with high measurements fine particles (PM2.5, for instance) won't necessarily correspond with days of high silica. Some minimal speciation, identifying between dust and droplets (even if silica isn't positively identified) would be illuminating.
Reply to this comment...
Log in to comment
Login to comment.