calibration approaches, once again

[Gerald Nelson]

In this email I want to first document what I understand about the calibration for the sensors in the AQE, get feedback on whether I have this right and then try out a potential solution that at least some of us can use.

My understanding of how the sensors work

Most of the sensors in the egg use a metal oxide technology. The relevant part of this technology is that the sensor has a current flowing through it and the resistance in the sensor depends on the characteristics of the air stream passing by it. This resistance is called Rs. So for example a sensor designed to measure ozone concentrations might have an Rs of 200,000 ohms when the ozone concentration is about 60 ppb. However, the relationship is not linear. The AQE developers took the approach of building a lookup table in the software. The table uses the sensor data sheet graph that a relates concentration levels to the ratio of Rs to R0. According to the data sheet, RO can have a range of values but the egg ships with software that has a single value.

Why does RO have a range of values? Because the sensors are not only sensitive to the gas in question but also temperature and humidity. And each has a somewhat different sensitivity to the gas and temp and RH. The graph in the data sheet is typically generated with temperature at 25C and 50% relative humidity. But my egg is likely to see temperature swings between 0C and 40C or more in a year and RH between 10 and 70 (I live in a dry area). One of the documents from SensorTech, the company that manufactures a couple of the sensors, says "We observe a change of 50% (reduced to half value) for a temperature increase of 25°C for MiCS-5521 CO/VOC sensor for example."

Sensortech sent me a very useful data set that reports ozone concentration values from a high end calibrated sensor, the local temp and RH value and the Sensortech Rs value. I used the R stat package to do a polynomial regression with Rs, Rs squared, Rs cubed, and temp and RH as explanatory variables. I'm happy to share the detailed results but the summary is that all explanatory variables are very statistically significant and with quite large numerical values. Note that the parameters are specific to the actual sensor used in the test so they can't be used with my egg.

How can this info be used for calibration?

There are three parts to the proposed solution.
1. Collect calibrated data (i.e. from a high end sensor) over a range of gas concentrations and ambient weather conditions. Put the sensor egg near the high end sensor exposed to ambient weather conditions. With the current software availability, it needs to have the EPA sketch installed and a serial port monitor (the Arduino software has one or you could use putty) operating to capture the Rs values. This means it needs to be hooked up to a computer of some kind. This process probably needs to run for a day to capture a range of values and sufficient data for step 2.

2. Do the regression described above to estimate parameters for the specific egg.

With the parameter estimates, and any temp, RH, and Rs value in the range of results from step 1, it becomes possible to generate 'calibrated' gas concentrations. This can either be done as a post-processing step, or someone could write a new sketch that would replace the look-up table approach.

So, comments please. Does this make sense? Has anyone tried anything similar to this? Is it something many people could do?

Regards,
Jerry

[Michael Heimbinder]

Hi Gerald.  This is a good description of an approach for calibrating MOx sensors.  But it's not sufficient to overcome all the challenges associated with using MOx sensors to accurately measure air pollutants in ambient air.  This report from Sonoma Technology, which includes an evaluation of MOx NO2 sensors, identifies some of the factors that need to be taken into account when utilizing MOx sensors for ambient air quality monitoring.  These include:

•Detection limits
•Range limits
•Response time
•Hysteresis
•Cross-sensitivity
•Relative-humidity effects
•Saturation effects
•Temperature effects
•Aging
•Air flow characteristics
•Poisoning

And that's just getting started.  Researchers at UC Boulder have developed an impressive method for calibrating MOx sensors but it requires serious resources and the sensors have to be regularly re-calibrated.  You can read more about their approach in their recently published paper on the topic.

We opted to invest our R&D efforts in developing an instrument built around a light scattering sensor because it's not susceptible to all the interference factors that plague MOx gas sensors.  To learn more about our instrument for measuring PM2.5, called the AirBeam, have a look at the evaluation we did in partnership with NYU School of Medicine, this independent evaluation conducted by Sonoma Technology, or check out our Kickstarter.  You can also get an idea of how it works in real life by reading over our latest Kickstarter update which describes using the AirBeam+AirCasting to measure PM2.5 while cycling in New York City.

-Michael-

--
You received this message because you are subscribed to the Google Groups "#AirQualityEgg" group.
To unsubscribe from this group and stop receiving emails from it, send an email to airqualityeg...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[NeilH]

Hi Jerry,
I think you have the basic idea of why the raw sensors values need calibrating, but need to have a more defined procedure for trace-ability of the physical quantity you are measuring - 'CO' - to the reference instrument.
Doing any basic calibration for the raw sensors is better than the raw value - however what is your decided target accuracy and how close do you think your method will meet it.
My engineering comment is that you have some undefined trace-ability in your statement

"Put the sensor egg near the high end sensor"

So how do you think what you are measuring is similar to what the reference sensor is showing.
Possibly if it is this within 5m, under no wind conditions for 30minutes with an ambient temperature variation to +-2C (in the shade) with sampling at 1min intervals, and measured changes are less than 20% on both instruments.
Someone else thinking of this from the ground up is:
http://publiclab.org/notes/kensanfran/11-02-2014/smart-sensor-board-for-electrochemical-gas-sensors-intro
http://publiclab.org/notes/kensanfran/11-04-2014/smart-sensor-board-for-electrochemical-gas-sensors-basic-info

I blogged on the subject of the need for traceability for the egg here
https://groups.google.com/forum/#!msg/airqualityegg/9a0GoVzRjLo/fO2lgMYFwOkJ

[anirudh uppal]

I want to add to what Michael said. I started measuring air pollution 4 years ago using the electro-chemical sensors and found that the resistance values changed with time i.e. degraded even when I tried to keep other factors constant. They were also very sensitive to voltage fluctuations.

When I found out about measuring particulate matter using light scatter - the results were a lot more repeatable and did not degrade with time. I have been following the same method Michael talks about to calibrate the PPD sensors with a more expensive co-located instrument. The correlation plots that AirBeam has shown are quite impressive.

thx,

A.J.

[Gerald Nelson]

I'm going to reply to three emails with one response. And I'm going to focus just on ozone. So Michael's comments about the AirBeam technology are not relevant for this email since it seems to be limited to particulates. Please correct me if I'm wrong about this.

1. The Sonoma Technology report has a critical flaw. It uses the O3 readings from the device that the sensor is embedded in and doesn't address how it gets from raw resistance values to O3 PPB values. And doesn't state whether the algorithm is calibrated for each specific sensor unit, or does as the AQE folks do, which is to have a standard value. The procedure I suggest corrects for these issues.
2. Michael's list of problems with the MOS are 'mostly' dealt with this this approach, depending on how frequently calibration takes place. It is important to point out that ALL sensors require periodic recalibration. The question is not if, but how frequently. Cross sensitivity is not dealt with in my methodology, but I have reviewed the specs for the sensors the AQE uses and for the relevant ranges of gasses it looks to me like this is not a serious issue.
3. Re Neil's comments about "Put the sensor egg near the high end sensor". I have in mind a location that is within 15 m or so of the high end sensor. My logic is that there is enough air mixing in that distance to have similar O3 concentrations.
4. For the statistics to work, I actually want a lot of variation in the temperature, RH, and O3 levels. The basic concept here is that a lot of things affect the Rs to O3 relationship but with enough data and a reasonable functional form you can capture all of them in the estimated parameters. It's also important to test this hypothesis by periodically recalibrating.