Hello All,
We've been developing a product with another company's tVOC sensor. In recent trials, we have found that their sensor-to-sensor variance is wildly different to the published/claimed values. The product datasheet claims a 15% sensor-to-sensor difference. We're seeing anywhere from 20% to 300% or more, and that is after following the manufacturer's guidelines to the letter.
We are now looking at alternative sensors. We decided to test the Renesas product since it has good on-paper specifications, attractive power consumption in low power mode, and has been available for a couple of years.
We created a simple test scenario with three Arduino boards. Each Arduino has a ZMOD4410EVZ attached, and we used the example code from Renesas. The Arduinos are connected to a laptop via USB. Each Arduino records a timestamp, IAQ, tVOC, EtOH, eCO2 and logRcda to a CSV file every couple of seconds. I'm using Grafana to plot the % variance between sensors for each of the measurements. All three sensors are physically within a few centimeters of each other. Each setup is effectively identical.
The first time I ran two sensors. The readings were quite different for the first 48 hours and then they all rapidly settled to be within +/- 15% (the green shaded area in the charts below). To be clear, the charts show the percentage different between the sensors, not the raw values from the sensors. The first chart shows the full 48 hour period, the second shows the last few hours when the numbers came within tolerance.
So far so good. The datasheet mentions a 48 hour run-in time on first use, so this seemed very promising. We stopped the test, hooked up the third sensor and restarted the test. This time, the first two sensors took longer to get within 15%, but they did eventually get there. Again, the first chart shows the full period of time it was running for, the second chart shows the last few hours.
However, the third sensor has yet to get within the +/-15% range.
And the last few hours:
You can see that IAQ and logRcda are within range, but the other values are not.
The only thing that changed between the first test run and the second is that the room being used for a test now has a window fan (it started getting warm here this week), so the indoor air in the room was probably more changeable than when the first test was run. Also, we are not currently providing temperature or humidity data back to the ZMOD for calibration and the weather has changed during the past few days. While running these tests, the temperature has been between 19C and 24C, and relative humidity has been between 55% and 84%.
My questions:
The results are already considerably better than we've seen from the other manufacturer's product, but after our bad experience with the other manufacturer, I'd like to be very sure about our expectations and the performance of the ZMOD chip.
Any advice or feedback would be most welcome.
-Nick
Hi Nick,
Thank you for evaluating our gas sensor and sharing your questions. The information is very well prepared so I can jump right into failure analysis.
Quote: “All three sensors are physically within a few centimeters of each other. Each setup is effectively identical.”
A few centimeters can mean quite a lot for gas sensors. You may have local VOC sources or turbulences and therefore different VOC concentrations in a room. Our sensors are quite sensitive to this. A better control is possible if you have them in a box (without lid) to have a better and equal distribution. Still, if the sensors are not in a professional gas laboratory it is impossible to achieve really equal conditions for all sensors.
Quote: “The first time I ran two sensors. The readings were quite different for the first 48 hours and then they all rapidly settled to be within +/- 15% (the green shaded area in the charts below).”
Note that this settling is a MOx-specific behavior and will happen at every MOx gas sensor (although some competitors will filter this out). You can imagine the sensor material like a muscle which is trained with sensor operation. When it is not trained it will lose its trained state over time. Especially very new sensors will need more time of operation to be trained. Usually this takes several days. Our algorithm breaks it down to a maximum of 48 hours.
A further influence can be the baseline calculation. The sensors need to see some relatively fresh (ambient clean) air within some days to find their baseline and report accurate values. These two factors (stabilization phase and fresh air event) together with set-up-related deviations will increase part-to-part consistency a lot.
Let me reply to your questions:
Question 1. Our main output parameter is IAQ and if not stated otherwise the statements are related to IAQ. For TVOC we can achieve +/-25% in a controlled environment and +/-15% in very ideal conditions.
Questions 2&3: Without seeing the raw data it is hard to tell. But I assume that the testing conditions (no fresh air event ever seen) caused all the devices to deviate a little more.
Question 4. We cannot influence the sensor material behavior. But taking the muscle comparison in consideration, the good news is the material memorizes its state for long time without being in operation and very slowly forgets it. The more the sensors are operated the better they will memorize their state. And even longer pauses (days) will not influence the results very much after power-on. So best advice is to let the sensors run as often as possible and they can come back to their stable status after power-on (seconds to minutes).
Question 5. From what I assume the testing conditions are (office-like environment with sensors on a table/cupboard) the results are reasonable, yes.
Note that we are aiming to release a new generation algorithm end of this summer 2022. It will contain the possibility to include RH/T measurements from an external sensor to increase further accuracy and consistency. This will be the same sensor product/hardware and only the firmware will be upgraded with additional features (e.g. sensor health status). We are constantly improving ZMOD4410 algorithms to have a long-term available and reliable product for you.
Thank you and best regards,
Anna from the Renesas Gas Sensor Application Engineering Team