I'm using a MAX31865 RTD measurement IC, and a standard PT100 RTD to measure maple syrup temperature as it's boiling (219 degF). At ambient, the temperature measurement is reasonably stable. But, when boiling, I get a substantial amount of drift. I need stability preferably in the range of ±0.1 degF. In my data below, I'm already applying a pretty heavy rolling average. And, the data you see was taken in seconds, so it's fair to say I don't have an electrical noise issue (and there's really no frequency component). This looks like random drift. Do I have a measurement problem, or is this a physics problem? Does boiling liquid drift like this?

enter image description here

I took the suggestion of @sstobbe and replaced the RTD with a resistor of similar resistance that would give me a 219 degF reading. The reading is solid as a rock. The stability is actually within the precision of the device. I see a very infrequent change of around 0.016 degF (I should mention that I'm using an oversampling routine to increase the resolution). So, I am convinced that this is a physics problem.

Next, I think I will use use the suggestion of @10ppb and try to increase the thermal mass by using a metal block of some sort. Unfortunately, agitating the syrup, or giving it some sort of recirculation, is not an option. Any other mechanical suggestions on how to overcome the measurement volatility induced by the boiling?

UPDATE 2: I tried a couple things last night with a boiling pot of water. I wrapped the end of the RTD with about a half inch of aluminum foil. There was perhaps a 10-20% improvement. Then I put a test tube upright in the water bath, added a little bit of vegetable oil in the bottom of the test tube, then put the tip of the RTD in the oil. The thought being, the oil wouldn't boil (and cause measurement fluctuations), but would be a good conductor of heat. While the water was at 212 degF, I couldn't get the oil past 204 degF. I waited at least 20 minutes. I'm sure it would have reached equilibrium at some point, but it was clear this isn't going to work for my application.

Also, someone had mentioned that they were surpised that my data wasn't more erratic. It actually is quite erratic. The data shown has a heavy rolling average applied.

  • \$\begingroup\$ Comments are not for extended discussion; this conversation has been moved to chat. \$\endgroup\$
    – Voltage Spike
    Mar 17, 2021 at 16:41
  • 1
    \$\begingroup\$ Is the liquid mechanically stirred or just by convection? You'll need to stir at a rate that the mass flow rate, times the heat capacity of the liquid, is greater than 10 x the heating power. Try playing with this 2D CFD simulation to see how heat flows. 0.1 F stability is hard to imagine in a system not in equilibrium, if it's boiling there's probably a 5 or 10 F difference between bottom and top of the vessel. \$\endgroup\$
    – tomnexus
    Mar 18, 2021 at 3:22
  • 1
    \$\begingroup\$ Anyone who's ever tried to make a small batch of caramel will tell you that an even temperature is non-existent. You spend the whole time stirring frantically just to try and keep it from being half burnt and half not yet caramelized. (Wet caramel is basically the same ingredients as maple syrup. But it could be worse, dry caramel is possibly the hardest one-ingredient recipe in the world.) \$\endgroup\$ Mar 18, 2021 at 18:24
  • \$\begingroup\$ I use the same chip and it's quite solid, even with 3m long cables and lots of noise. Try another PT100, the one you have probably is of low quality, perhaps shaking around on the housing. \$\endgroup\$
    – Damien
    Mar 19, 2021 at 8:15

5 Answers 5


That's a pretty mediocre chip to be expecting 0.1°F stability from. At 1mA 0.1°F is about 8uV. The chip has typical noise of 150uV over the bandwidth, and thermal EMFs can cause errors of some uV here and there with little trouble.

If you are running the probe at high current you might also be getting variation from bubbles and heat loss, but less from the noise and air currents. Depending on construction self-heating will be an issue even at 1mA when you are looking for +/-0.055°C.

Your reference resistor should be a precision type 25ppm/°C or better.

If you have a reasonably good digital multimeter (preferably 4-wire) you can just hook it directly across the sensor (remove the other circuitry, obviously) and confirm one way or the other.

  • \$\begingroup\$ Do you have experience with any other RTD measurement chips? Anything you could recommend? Would self-heating cause the back-and-forth drift I'm seeing? I started thinking about the ref resistor. It does not have a low TCR. But, again, would that cause the back and forth drift? Good idea on checking sensor directly. I do have access to a good 4 wire milliohmmeter. \$\endgroup\$
    – gtetil
    Mar 17, 2021 at 13:16
  • \$\begingroup\$ This all-in-one thing is not great. More accurate solutions generally involve separate signal conditioning and ADC. If you poke around in application notes from the erstwhile Linear Technology, AD and TI (former Burr-Brown) you can find circuits that others have designed. If you don't really need 0.1° though, 1° is a lot easier. Thermistors are more sensitive so the circuitry is simpler/cheaper, especially for a narrow temperature range, but the stability is not likely as good as a platinum sensor. \$\endgroup\$ Mar 17, 2021 at 13:24
  • \$\begingroup\$ Ok. Thanks for the tips! Unfortunatley, I do need 0.1 accuracy. I am suspicious of my ref resistor, but do you think this could cause what I'm seeing, or would it just cause general measurement error? \$\endgroup\$
    – gtetil
    Mar 17, 2021 at 13:28
  • \$\begingroup\$ Without seeing your setup and knowing how long it's been running it's hard to be sure, but there would typically be a component that varies with air currents. I would also expect a sort-of linear-looking drift as it warms up. \$\endgroup\$ Mar 17, 2021 at 13:30
  • 4
    \$\begingroup\$ If you think the fluctuations might be due to the bubbles, put the RTD in a block of something with some heat capacity (like a block of aluminum) and put that into the liquid. The block will average out fast fluctuations in the actual temperature at the RTD. \$\endgroup\$
    – 10ppb
    Mar 17, 2021 at 15:22

The physics side of the problem:

Boiling as a process arises from local overheating of the liquid and cooling the place by evaporating inside a growing bubble. The temperature needed to create a new bubble is higher than the temperature needed to grow the bubble and can be quite a few of degrees above the equilibrium "boiling point".

Boiling a mixture is even more complex from boiling a pure liquid. The constituents of the mixture (sugar and water in your case) constantly change concentration. This changes both the overheating needed to create a new bubble and the equilibrium boiling point. The bubbles carry amounts of water between the bottom and the top, making the solution weaker at the top. Bubbles also mix everything at some rate.

In short, you are lucky to get fluctuations that low and this is probably because of the sensor not being quick enough.


I haven't tried measured boiling liquids to this accuracy, so curious how this turns out.

SUMMARY / UPDATE To try to see if effects of a hard boil make a difference on the unwanted 0.01-0.1 Hz order-of-magnitude signal components, perhaps try the following (all together)

  • slow boil (i.e. bubble nucleation only on surfaces)
  • boiling stones (create better surface nucleation sites than pot surface)
  • stirring the liquid, fast swirl (avoid hot/cold spots)
  • make sure sensor stays in fixed position and depth (reduce depth-pressure effect, if container is tall)

These may not be practical for making maple syrup, but if I wanted to learn how to get smooth boiling temp data with limited measuring hardware, that is what I would try.

FWIW there is a pressure dependence for the boiling point (and also dependence on dissolved substances, as the OP clearly noticed). At first glance, those are "DC" issues.

For pure water at standard conditions, the coefficient vs pressure is approx. 1 deg C / 20mbar (eyeballed from this data).

Movement of the liquid will influence the local pressure. Depth will influence it slightly too, approx. 1cm / mbar.

I would very much wonder if a "gentle" boil, heat flow just barely above the point of homogeneous nucleation, would be much preferable to a "hard" boil if you need measurement stability...

As a debugging tactic, maybe try bringing the liquid just below homogeneous nucleation (bubbles on the walls only [*better: add boiling chips/stone]) and see how steady the reading is . . . Maybe that will help produce clues as to whether the issue is electronic or physical.

[update2] Watching some actual boiling just now, hot spots on the bottom probably would dominate all other factors. More reason to stir.

  • \$\begingroup\$ It's a wonderful suggestion, but one concern is that the graph shown by OP is over a period of almost 1000 seconds. Even if these phenomenon had an impact, to me they appear to be very fast and I don not think they would appear on such a long acquisition and time average signal like this. Of course, the standard of care, once you have determined that a real phenomenon is likely, would be to compare your test to a well calibrated precision measurement system. \$\endgroup\$
    – crasic
    Mar 18, 2021 at 1:31
  • 3
    \$\begingroup\$ Also RTDs are typically very slow, even in a liquid, so there will be effectively a low pass filter in there already. If the sensor was a fast thermocouple and the variation sub 1 second I could see that. That said, I've seen surprisingly large superheating in boiling liquids (several degrees C). The boiling point is not as consistent as some think. \$\endgroup\$ Mar 18, 2021 at 2:11
  • 1
    \$\begingroup\$ Other effects that may be involved: once it starts to boil, the syrup is no longer uniform in concentration, i.e. boiling point. This is obvious in making jam, where the goal is to get the whole lot to 105°C (220°F); even if the total pressure is constant, the partial pressure of water over the surface is likely to change. This at least will affect the rate of evaporation, and therefore the heat flow. Taking this further, condensation dripping off a lid will provide local cooling \$\endgroup\$
    – Chris H
    Mar 18, 2021 at 7:22

Do I have a measurement problem, or is this a physics problem? Does boiling liquid drift like this?

I know that you can use boiling liquids as a reference if the mixture is the same chemical. I don't know if the converse holds true because you have several chemicals each with different temperatures (water, sugar, and other volatiles each with a different boiling point). Since there isn't a lot of information about boiling molasses outside of the boiling point, I'd think you'll have to do your own testing.

A good way to test your probe would be to use water which should be pretty close to 100C (depending on your elevation) and is commonly used as a reference. Or just to get a reality check buy a reference probe with an accuracy of 0.1C to validate your probes measurements in the material.

  • \$\begingroup\$ I actually did a bunch of testing using water, with the same instability. \$\endgroup\$
    – gtetil
    Mar 17, 2021 at 17:10
  • \$\begingroup\$ Then it's probably the sensor, do you really need the accuracy? \$\endgroup\$
    – Voltage Spike
    Mar 17, 2021 at 17:17

This is mainly in response to the updates added since several other answers, but would be too much for comments.

The metal (copper/aluminium, does it need to be food safe?) block's thermal mass will provide time averaging. Its size will provide spatial averaging, and if you have packets (you do, I'm sure) of hotter/cooler fluid it should help by sampling several of those.

Another way would involve multiple sensors in different places. If you have the ability to read several, seeing how they're correlated as a function of separation may be instructive (in 3d but especially vertically and radially if the vessel is a cylinder). Pack them together and they should follow each other fairly well, confirming fluid effects rather than electrical. Measuring at several points and averaging may in fact be a solution if you want a representative sample of something that's not uniform.

BTW The nice easy DS75 I've used (integrated SPI thermometer) probably doesn't have the absolute accuracy you need, but would be an alternative during development and reading several on a cheap/simple device is trivial.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.