Fooled by a thermocouple: Temperature sensing gone awry
What can explain an infrared temperature sensor's unique "proximity-sensing feature," whereby the measured temperature rises whenever someone approaches or touches the system?
By Ken Whiteleather, Sparton Corp -- EDN, April 12, 2007
During the development of a medical product requiring noninvasive temperature sensing of fluid passing through 3/8-in. medical plastic tubing, the design team I was working with selected a miniature infrared optical temperature sensor. The cylindrical sensor measured 1/4 in. in diameter by 1 in. long. The sensor had a 1-to-2 field of view (Note 1). The tubing we needed to sense was within the disposable component of the system. The sensor, spring-loaded to maintain slight pressure against the tubing when the disposable was clamped to the device, was centered on a U-shaped channel on the device to align with and "receive" the tubing.
We used the "heat-balance" method that the vendor recommended to accomplish temperature sensing. This method requires pressing the tubing against the sensor, which permits the sensor to convert the infrared energy that the fluid emits but ignores the effects of the tubing material or the disposable housing. To our surprise, this method seemed to work well in early breadboarding experiments to track the actual temperature, which we measured using standard thermocouples in contact with the fluid within a ±1°C tolerance of error. The method also tracked rapid changes in the fluid's temperature with only a few seconds of delay.
|
The vendor advertised that the sensor behaves as a K-type thermocouple at 37°C and is relatively accurate within our temperature range of interest: 10 to 50°C. In other words, its output should resemble the output of a contact thermocouple at the same temperature. We implemented a "cookbook" input circuit for a standard K-type thermocouple, expecting that it would perform perfectly. The breadboard prototype performed well, requiring only the addition of an offset adjustment to compensate for variations in components. We used the same conditioning circuit in the final design for both the optical infrared sensors and the standard-contact thermocouple sensors.
Once we implemented the design in the device, we noticed some odd behavior. With all sensors reading correctly and temperatures stabilized throughout the system, the optical thermocouples' output would rise significantly to as much as 5°C higher if anyone approached or touched any of the exposed metal parts on the device. The manufacturing operators also had a difficult time of adjusting the offset circuit for the infrared sensors with any repeatability, a fact that was no doubt related to the sensor's undocumented "proximity-sensing feature." The standard-contact thermocouple outputs did not change. This situation was, of course, unacceptable. A lot of head-scratching ensued!
After some investigation, we discovered that the optical infrared sensors had a measured impedance across their leads of nearly 20 kΩ! A standard thermocouple would normally appear as a short circuit. Apparently, this mismatch of impedance at the output of the infrared sensor and the input of the conditioning circuit was amplifying any minute sources of noise—in this case, induced ground noise—to an untenable level.
The cure was to place a 20-kΩ resistor across the input leads of the conditioning circuits of only the optical infrared sensors. The proximity-sensing feature and the difficulty in adjustment of the offset circuits miraculously disappeared! A review of the optical-sensor data sheets confirmed that they never mentioned this "output impedance" (Note 2). I suppose, in this case, a K-type "thermocouple" wasn't really a K-type thermocouple.
Ken Whiteleather is a senior electrical engineer for Sparton Corp. Like Ken, you can share your Tales from the Cube and receive $200. Contact Maury Wright at mgwright@edn.com.
Editor's notes
Correction (4/23/2007): As originally published, the article incorrectly stated that the sensor had a "1-to-1" field of view.Update from the author (4/23/2007): Upon further review, the data sheet did include an "impedance" spec, but the design team overlooked it in favor of the "K-type thermocouple output" spec. In addition, the project team would like to make it clear that the sensor vendor provided helpful information to help resolve the difficulties with the application.
-
This story illustrated no technical solution - only the foolishness of the author.
Especially the nonsense about impedance mismatch amplifying noise:
"Apparently, this mismatch of impedance at the output of the infrared sensor and the input of the conditioning circuit was amplifying any minute sources of noise..."
Glad the noise divider worked okay for you - frightened that the product may actually be in use....
DFL - 2011-11-4 05:37:44 PDT -
The below link goes to an app note that talks about rfi and noise pick up with low level signals. The note is about strain gage sensors, yet the noise issue is the same w/ thermocouple sensors. If you go to the below link, please see
"Verifying that RFI is not inducing ... Errors"
and
"3b. Noise from RFI "
Here is the link:
www DOT instrunet.com/applications/fm/sg_theory.html
Bill - 2007-24-4 14:29:00 PDT -
We work with thermocouples, and rfi can be a problem. In your 20K ohm case, it is a good thing in some ways because this means you can place caps (e.g. 0.1uF) across the tc+ and tc- leads, and between the tc+ lead and ground, and filter out rfi. With a regular tc, the R is 0 ohms (approx) and the R*C in an rc filter is not too good whereas here, you get 20K * 0.1uF rc filter. Nice !
Ted - 2007-24-4 14:20:00 PDT -
Or maybe this is an April fools joke?
Rob MacLachlan - 2007-19-4 10:19:00 PDT -
Well, yeah, It presumably did work for some reason, but not "impedance mismatch". This is DC here. It may also be that the true sensor output impedance under no load was a lot higher than the 20K they measured with their handy DMM, as I gather this is some sort of thermopile device, and probably highly nonlinear. What is the output impedance of a diode? Depends on the bias current. Zero current means high impedance.
Rob MacLachlan - 2007-19-4 10:14:00 PDT
