Playing with low-cost gas sensors

As a scientist, I work with the characterisation of gases. For this purpose, I use professional analysers of various types: fourier transform infrared spectrometers, mass spectrometers, gas chromatographs with a plethora of detectors and tunable diode lasers. These instruments are expensive, typically in the range between 15-150 k€ and the number of instruments available to me is limited. The instruments performance are well established: accuracy and precision, sensitivity and selectivity. Most of the instruments I work with are multicomponent analysers; the ability to deconvolute interference from gas species has been vastly improved over the years by signal resolution and multivariate calibration.

While industrial process monitoring normally are discrete measurements, fugitive emissions and environmental monitoring generally improve with spatial resolution. This has been an important driving force for the development of low cost gas sensors. While the analytical performance often is compromised, this can be compensated for by redundancy: clustering of sensors is one way to improve accuracy and precision. The ability to analyse large amounts of data also makes it possible to correlate gas sensor data with other physical (e.g. meteorology)  sensor data.  EU regulations have opened for reporting of industrial emissions by using by-proxy methodology: by establishing a correlation between an online measurable parameter and the emission to be reported, the online measurement can be used.

For a couple of years, I've had a Netatmo weather station running at home. More recently, I've expanded the units to include wind speed and direction and rainfall.  Last summer, I duplicated the setup for our cabin in Tynset. What caught my attention was the difference in carbon dioxide levels between home and the cabin: while at home levels higher than 2000 ppm are rarely seen, the sensor at the cabin reaches 5000 ppm (maximum sensor reading)  rather quickly with four persons present. I started looking into the sensor used by Netatmo. The sensor recalibrates itself weekly by using atmospheric concentrations, but this of course requires weekly ventilation to that level. I guess the difference between home and cabin is that the house was built in 1968 whereas the cabin was built in 2017. With the recommended level for carbon dioxide indoor is 1000 ppm we have started more frequent ventilation of both locations.

Non-dispersive infrared spectroscopy (NDIR) is not the cheapest technology, but still sensors are below €100. My experience with the Netatmo sensor got me curious, so I wanted to check the accuracy of a commercial NDIR carbon dioxide sensor. I ordered a SVCD30 sensor from Sensirion, and used their sensor test kit to connect the sensor to a computer with their data logging software.

I used a calibrated gas mixer from Alytech to dilute a 2 % vol mixture in nitrogen with Nitrogen 5.0 (99.999%). I programmed two sequences to be run: 0 to 250 ppm in steps of 50 ppm, and subsequently 0 to 2000 ppm in steps of 400 ppm. For the first series the step time was 30 minutes. This was reduced to half for the second series. Between and after the series, the sensor was exposed to atmospheric carbon dioxide concentrations.

The unfiltered sensor data is shown in the chart. Using the data in both series, a method evaluation fuction is described by y = 1.10x-22.0: A concentration dependent error of about +10 % and with a concentration independent error of -22.0 ppm.  The linearity was excellent: an adjusted R² value of 0.999 was estimated.



I wasn't sure what to expect, and my first assessment was that an error of about 10 % is not too bad.
Next up is to test the sensor for interference from varying humidification levels in the gas.