Testing the Shinyei PPD42NS

Around this time last month, the haze (or what some people call smog) here set a record high level for the Pollutant Standards Index (PSI). This is what it looked like outside:

haze vs no haze

As our National Environment Agency only published 3 hour PSI averages, I thought it would be good if we could get our own measurements. The PSI used here is somewhat like the Air Quality Index (AQI) used in the US, and is made up of 5 components:

  1. PM10 particulate matter
  2. sulphur dioxide (SO2)
  3. carbon monoxide (CO)
  4. nitrogen dioxide (NO2)
  5. ozone

Note that the AQI includes PM2.5 particulate matter whereas PSI does not. From what we can see, I would think that a major contributor to the PSI is particulate matter (PM).

I took a brief look at the projects such as the Air Quality Egg and PACMAN. They used either the Sharp GP2Y1010AU0F or the Shinyei PPD42NS. These sensors generally operate based on the light-scattering principle, by measuring the amount of light that is scattered by particles.

The PPD42NS

Chris Nafis has done a great job documenting the use of both the GP2Y1010AU0F and the PPD42NS, compared against a Dylos DC1100 air quality monitor. As the GP2Y1010AU0F requires a certain pulse waveform to be supplied to its LED pin, I would say that the PPD42NS is self-contained and thus much easier to hook up.

PPD42NS (front)

On the front, it has 2 pots labelled VR1 and VR3 that have been already factory-calibrated. The IR detector is covered under the metal can. Interestingly there’s a slot by the side labelled SL2 which is unused. If you’d like to see what’s under the hood, Chris opened up the black casing and posted a photo here.

PPD42NS PCB

Looking at the date code grid on the PCB, the units look like they were manufactured in July 2012. The circuit consists largely of passives and an op-amp. RH1 is the resistor heater which, in theory, could be removed to save power if there was some other method of air circulation.

Sensor Placement

There are several points that should be observed when deciding how the sensor is placed.

  • The sensor must be placed in a vertical orientation. This is easy to understand as the heater is located below. Any other orientation would not achieve the desired airflow expected by the designers.
  • The sensor should be kept in a “complete [sic] dark condition”. I believe the unit is smart enough to not be affected by normal lighting as it uses infrared, but I guess it is still a good measure.
  • “Also soft cushioning material to seal the gap between sensor and housing is necessary.”
  • Sensor should be located away “from any winds or air stream effected by air purifier”.

I’m not sure if leaving the sensor “window” exposed will affect the readings by a lot but David Holstius did cover it with some stiff paper.

Sensor Output

The sensor output is normally high, but goes low in proportion to the PM concentration, hence by measuring what they call Lo Pulse Occupancy (LPO), the PM concentration can be determined.

This LPO is recommended to be measured over a unit time of 30 seconds, but I opted to record at 15-second intervals instead to get a higher resolution. To get the LPO time for a 30-second time period, we just need to average the two 15-second values – simple math.

As I do not need the exact calibrated figures, there was no need to fit the values to the equation derived by Chris (yet). With the raw values recorded, I can always do this at a later time.

Note that the sensor readings are very noisy. Here are the plots over 24 hours, without averaging and with a 50 value rolling average (12.5 minutes):

It is definitely easier to read the graphs with averaging enabled. Subsequent graphs that appear in this post have a 15 value rolling average applied.

Test Setup

Using the same parcel that the sensors were delivered in, I punched a couple of holes on both sides, made sure they lined up and tied 2 sensors on either side using raffia string.

Both sensors were hooked up to an Arduino that measured the LPO time of each sensor and output them via serial. The Arduino was connected via USB to my Linux server which runs a simple Python script that timestamps and logs the values.

PPD42NS setup

One of the sensors would produce a higher and sometimes spurious output occasionally and I couldn’t figure out why. However it would agree with the other sensor most of the time. This is also why I usually buy at least 2 sensors to make sure their readings are similar.

Notice the spike from one sensor occurring around 17:00 to 17:30:

graph 1 with one sensor showing spurious readings

Notice 2 spikes from only one sensor between 09:00 and 10:00, and the region from 06:00 to 08:00 shows a significantly higher reading:

graph 2 with one sensor showing spurious readings

Subsequently, I took the sensors out, used a blower to remove any dust in the assembly, replaced the raffia string with twist ties and hooked them up again. I also closed the front of the box, leaving just a small slit for the air to flow through. The problem went away and I didn’t manage to figure out what was the cause.

Anyway, here’s what my ambient indoor readings (when nobody is home) look like:

graph showing normal, ambient readings

The sensor is sensitive enough to pick up cigarette smoke. When nobody is at home and the doors and windows are closed, the PM concentration gradually goes down.

What’s Next?

Now that I’ve been measuring the level of particulate matter indoors, I will need to build a proper enclosure for outdoor use. I have an enclosure that looks very similar to this one that David Holstius used, but I plan to use a slightly bigger one so that I can house the electronics inside as well.

As I opted to ship the sensors by registered mail (it’s the cheapest method), it took about 3 weeks to arrive. By the time I received the sensors, the PSI had dropped back to normal or “safe” levels, which is no doubt a good thing. At least now I can start collecting data to get a baseline for “clean” air.

Advertisements

41 comments on “Testing the Shinyei PPD42NS

  1. Fabrizio says:

    Hi Darell
    very interesting post!
    I tried to reproduce your results, but I encountered a problem with the LPO time as read from Arduino. Very often it is 0 ! I get a lot of zero and some “spikes” some time to time.
    I used two PPD42 and two Arduino to be sure not to have some hardware failure, but always same results.
    Do you have any suggestion or hint to improve my results ?
    Thanks
    Fabrizio

    • darell tan says:

      Hi Fabrizio, it sounds like you may be reading the output directly. The idea is the count the total amount of time that the output is zero, for about 30s.

      You may want to try out this simple Arduino sketch here:
      http://www.howmuchsnow.com/arduino/airquality/grovedust/grovedust.ino

      • Fabrizio says:

        dear Darell,
        I measure the LowPulseOccupancy, as you suggest.
        In order to make a further test, I uploaded the grovedust.ino sketch into Arduino and left it run during the last night. I got 1800 readings, of which more than 1300 are zero. ( I mean 0 LPO, the concentration is 0.62 because of the offset in the formula ). The lowest non-zero reading is 2200. The sensor is inside an office, and I expect quite low concentrations ( about 20 ug/m3 )
        But so many zero values and the “step” between o and 2200 seems me quite suspect.
        Do you observe similar data ?
        Thank you in advances and ciao

        • darell tan says:

          Hi Fabrizio, the data doesn’t sound correct. I would suggest you plot it out to see how it looks.

          You can try hooking the output pin to an oscilloscope and observe what happens when you blow at the sensor opening. You should see the LPO time increase.

          Here are some sample readings (15s intervals): 0, 473, 294, 348, 0, 118, 0, 41, 601, 62, 143, 0, 290, 61, 0.

        • What formula are you using to convert particle counts/0.01 cu feet to concentration in ug/m3?

        • darell tan says:

          I’m not converting anything. These are the raw LPO times that are read from the sensor and plotted directly. I think it’s not very useful to convert them without having some reference to an accurate reading. If you are looking for a formula, see the excellent writeups by Chris Nafis linked within the post.

  2. Bob Rich says:

    I would suggest that anyone experiencing lots of noise cover the large hole over the detection area. With this open you are eliminating the chimney effect that draws the air up from the intake to the output (reducing the airflow), and you are going to be more exposed to turbulent drafts and any light noise in the environment.

  3. Walter says:

    FYI, on a standard Arduino, pin 8 is considered the Input Capture pin and uses a 16 bit timer to measure pulse duration. The other pins use an 8 bit timer. If you used pin 8 for one sensor and a different pin for the other, this may explain some of your discrepancies.

    • darell tan says:

      Hi Walter, the sketch I’m using is just recording the pin state at a predefined interval and it does not use the Input Capture unit at all. This will enable me to sample the signal from both sensors at about the same time.

      However, this setup is far from ideal. Sampling the sensors this way could miss out on some time when the signal goes low.

  4. Brian says:

    Hi, It’s nice to see this research. Could you share those figures in original post? It seems there’s problem the figure lost.

  5. enetnut says:

    Darrell,

    I have been trying to get a better response from the PPD sensors without averaging and found that adding the right fan in the right config made a lot of difference. I started with the PPD60PV-T2 and saw a massive (positive) difference that I documented here:

    http://indiaairquality.com/

    AJ

  6. […] For those who have worked with the 42 or one of its cousins (the DSM501 and the Sharp GP2Y thingy) know that the data gives many many zeros. As in this blog post from Darrell Tan in Singapore. http://irq5.io/2013/07/24/testing-the-shinyei-ppd42ns/ […]

  7. A.J. says:

    Hi Darell,

    I referenced your article on my post today on modifying the PPD42NS to improve its response at:
    http://indiaairquality.com/2014/12/14/measuring-the-pickle-jr-a-modified-ppd42-with-an-attached-fan/

    Would love your comments!

    thx,

    A.J.

  8. […] Testing the Shinyei PPD42NS […]

  9. igorfonseca83Igor says:

    I’m trying to use a PPD42NS dust meter with an Arduino Uno. I used the standard code from the wiki (http://www.seeedstudio.com/wiki/Grove_-_Dust_Sensor), but the values I’m reading are too erratic. I tested the sensor for several hours, in an office enviromnet (free of dust), but the sensor indicated a ratio around 30% and concentration of 45000!
    I also measured the voltage on signal pin and it’s constantly oscilating.
    Can anyone help me?

    • A.J. says:

      I have some tips on my blog on how to get the PPD42NS to work well. If you are getting erratic results, you need to a) verify the orientation b) smooth by averaging over several time periods (10 mins is good), c) make sure your power supply is clean.

      If the environment has very little dust, the sensor is going to show erratic but low results. You can increase the sensitivity (and get less erratic results) by adding a resistor and a fan as I have described. Good luck!

      • A.J.,
        Thanks for you advices. I have some comments:

        a) I tried in two positions (vertically and horizontally). The results were almost the same. I kept it vertical after the tests;
        b) Once the values indicated by the sensor are always high, the averaging of the values over a period also resulted in a high value;
        c) I measured my power source with a voltmeter. The voltage was between 4,79 and 4,85V. I think those values are according to the specifications, right?

        Thanks for your help!

        • A.J. says:

          Looks like there is light entering the sensor and resulting in a high reading constantly. Cover the front open part i.e. the triangular piece with a black non reflecting paper or tape. and also feel free to send a picture.

          PS: Don’t set the sensor horizontal. Your PS seems fine.

      • hugh says:

        Dear A.J.
        Thanks your advice.
        I got a solution from you!!!
        🙂

        • I covered the the front opening and even tried to measure in the darkness, but nothing changed. Regarding the potentiometers, is it possible to adjust sensor’s sensibility with them? I kept the configuration from factory.

        • A.J. says:

          You should not be getting such a high LPO if the sensor is shielded from light, and is upright with no vibrations. No, do not change the pots that will not help. Send your code and I will take a look. If the code is ok, you may have a bad sensor.

        • I’m using the standard code available on senser’s wiki page (http://www.seeedstudio.com/wiki/Grove_-_Dust_Sensor).
          I am trying something different. I read in a blog that the sensor have thwo channels: P1, which count particles between 1 and 10 microns, and P2, which counts particles betweeen 2.5 and 10 microns. Standard electrical connector has a wire for P1 channel only. I added another wire to measure P2 signal.
          There results are very different: channel 1 is indicating around 45000 pt/cf while channel two indicates around 2500 pt/cf. Even though, I believe those values are too high for an office environment. Am I right?

        • A.J. says:

          Not really. The two channels are really just different thresholds of voltage. At P1 the voltage is 1.0 volts for P2 the voltage is 2.5 volts. So much less number of pulses make it over P2, hence the LPO on P2 is lower. Therefore I had modified the threshold on P2 to be lower than P1, approx 0.5 volts. This significantly increases the sensitivity of the sensor.

          Also, using pulsein it is not easy to measure two pins. You need to move to interrupts if you want to measure both simultaneously. I assume you measured one and then the other i.e. not simultaneously ?

          Back to your main issue, the numbers you are getting are way too high. Can you post a printout of your results ?

        • Actually I’m just using one channel now (P2). I’m considering a sample time of 30s.Those are the values I obtained for period of test:

          low pulse occupancy: 1439962
          ratio: 4.80
          concentration: 2530.65

          low pulse occupancy: 1368226
          ratio: 4.56
          concentration: 2397.52

          low pulse occupancy: 1295275
          ratio: 4.32
          concentration: 2263.46

          low pulse occupancy: 1281659
          ratio: 4.27
          concentration: 2238.58

          low pulse occupancy: 1275442
          ratio: 4.25
          concentration: 2227.23

          low pulse occupancy: 1445482
          ratio: 4.82
          concentration: 2540.95

          low pulse occupancy: 1325400
          ratio: 4.42
          concentration: 2318.67

        • A.J. says:

          And what about if you use only P1 ?

        • If I use only P1, I have values like these ones:

          low pulse occupancy: 9234646
          ratio: 30.78
          concentration: 44490.77

          low pulse occupancy: 9105093
          ratio: 30.35
          concentration: 43035.06

          low pulse occupancy: 7286404
          ratio: 24.29
          concentration: 26149.17

          low pulse occupancy: 8102732
          ratio: 27.01
          concentration: 32946.49

          low pulse occupancy: 7949507
          ratio: 26.50
          concentration: 31578.32

          low pulse occupancy: 9407674
          ratio: 31.36
          concentration: 46491.93

          low pulse occupancy: 9446295
          ratio: 31.49
          concentration: 46947.64

          low pulse occupancy: 9551457
          ratio: 31.84
          concentration: 48205.27

          low pulse occupancy: 9155035
          ratio: 30.52
          concentration: 43591.92

          low pulse occupancy: 8365364
          ratio: 27.88
          concentration: 35395.64

        • A.J. says:

          Howdy, lets move the discussion over to my blog at indiaairquality.com so that we don’t take over Darrel’s blog 🙂

  10. j. says:

    Darell,

    Very nice article, however I have two comments:
    1. Math. Since you used the same formula for both sensors, the outcome of the comparison should not be affected, but if you would like to compare readings of your sensor to some calibrated source, do not use average of averages to compute the final value. If you do the math, and calculate the difference between proper average using raw counts over integration time, and value computed using your method, you will see that it is non-zero (meaning the results are different).

    2. Spikes: Make sure that the sensor is not exposed to sun, or artificial lightning. This could also be reflection of sunlight from a car parked on the street, or window in a building. Could be also light from a ceiling light, or desk lamp.

  11. […] To ensure that the sensing units were in dark condition and not affected by winds or air streams, I placed both units in a cardboard box with large apertures. The units were in vertical position to ensure the generated flow by the heater follows the expected path. […]

  12. evannak says:

    I have this ppd42 with arduino uno. but don’t know what to do with it yet. Do you have experiences with Arduino uno?

  13. Mathieu says:

    Hi! Great post. I was interested to see if people were always taking the raw output or filtering it.
    Do you happen to have a file containing the raw output for a really high sampling frequency? (like one value every second or even faster). People on the internet always display with one sample every 30 minutes so it is not possible to have an idea of the output of the sensor before buying one.

    Thank you!

  14. hammillbr says:

    https://airquality406.wordpress.com/
    Here is my work with the dust sensor. I use a Particle Photon to make my sensor web enabled.
    Just thought I would share or people looking for information.

  15. A.J. says:

    Mathieu, by raw input I take it you still mean from P1 or P2. The samples are taken several hundred to thousand times per second and then averaged. I am also modifying the sensor to measure “before” P1 and P2. That will be a raw output before the signal processing on the board. More details to follow at:
    https://indiaairquality.com/

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s