Sunday, April 21, 2013

Thermals and Calibration


Toaster oven meet test setups. Test setups, toaster oven. Now that introductions are done, let’s get down to preliminary business.

The above picture shows two test crank arms – one from V2, and one built up for this testing.. Each crank arm has two different sensor setups. Four in total and all very different. There is a thermocouple inside the oven. This is the first test measuring four sensor setups. Five tests have been conducted. The first three are really for figuring out how to deal with setting up an actual experiment and the last two are the real data experiment. I used an insulated tape on the thermal couple to measure each crank arm independently during the last two tests. More after the break.

As far as I can tell, most companies are using one single temperature measurement to compensate. I’ve decided that I need to sense at each sensor setup because they are so far away from one another. For SRM to do the same that would mean they would require 4 temp sensors, Quarq would require 5, Rotor would need 2 or 4 depending on how they are setup, but really they probably get away with 2. My experiments are showing that even though they are theoretically thermally compensated there is still a very small effect from temperature.

Example (Celsius):

Of the test setups, the calibration for the least sensitive to force and one of the greatest to thermal shows approximately 1500 counts per N-M in 24 bit. Over a temperature range of 25 – 50 this setup in 24bit mode experienced an offset slope of –36.2 counts per degree. So if you went outside at 25 degrees and it was 30 degrees out you could theoretically have a -0.121 N-M offset after it warms up! Sounds like nothing!

But lets add a steady rotational velocity into the mix. At 90RPM this would be a –1.13 watt offset. At 40 degrees this would grow to –3.41 watts. Over two pedals this becomes –6.82 watts. This is enough to be compensated in my opinion.

Why am I not ignoring this? It’s cheap to fix. How much? 4 Dollars + coding. And believe me the coding isn’t hard. I have to switch from the ADS1247 to the ADS1248. Slightly larger physically, but it gives me more inputs which I need.

So the new design will have two Platinum RTD thin film sensors located within close proximity to each strain gauge sensor to get a good, direct measure of what the actual temperature at the location is. Each sensor setup will therefore be compensated individually. Otherwise there can be thermal lag, causing offsets.


The only small downside to this is the offset value transmitted via ANT+ – It can only show one value for zero offset. They are unit-less, so they can mean literally anything. N-M, bits, truncated bits, scaled values, etc.

If one company uses a 24bit ADC and another uses a 16bit ADC, the 24bit ADC might show offsets ranging from –20000 to + 20000, but the 16bit version of this would be –78 to + 78. So, one might assume one power meter is more stable… when in actuality you cannot make any inference at all, unless you were the person who programmed these or have experimented with forced offsets.

I’m expecting I’ll do a larger post on my results. The short is that:

  1. Digital thermal compensation is needed even in "”thermally compensated” analog designs.
  2. Accurate compensation depends strongly on thermal measurement location
  3. Clever thoughts are required on both when and how to incorporate the thermal measurements. Every rotation, every five rotations, five times a rotation?

1 comment:

  1. interesting blog. It would be great if you can provide more details about it. Thank you...

    Temperature calibration