Sciencemadness Discussion Board

DIY melting point apparatus

brubei - 27-5-2024 at 06:32

Hi, i aim to build an opensource melting point measurement apparatus.

Any help is welcome.

here are the spécifications:

compatibility with classic micro capillary test tube

1 arduino-like chipset

1 thermal probe

1 heat source

1 cooling fan

should provide 1 to 20W (already test the easy fusion of some products with about 300°C mp with a device of this power)

heat source may be a aluminum block with:1 hole for test tube, 1 hole for temp probe, 1 hole for magnifer lens.

1 potentiator that should regulate speed rate

1 lcd screen that should show the temperature evolution

idk if phase transition could be detected by laser as powder/liquid interaction with laser is obvioulsy different

Sulaiman - 27-5-2024 at 18:27

Illumination?

Mateo_swe - 29-5-2024 at 03:00

I think it´s important that the temperature measuring element is very close to the sample and of cause that the temperature at suspected melting point range is increased very slowly.
Many use an oil filled testtube (or thiele-tube) with the thermomether and micro capillary tube with sample put side by side (rubber bands) in the oilfilled test tube.
If you make a apparatus please show it off here, it´s always nice with homemade devices.

Sulaiman - 29-5-2024 at 04:07

You could consider incorporating a usb camera,
or a camera modulewith a microcontroller interface for auto-melting-detection?

[Edited on 29-5-2024 by Sulaiman]

RedDwarf - 29-5-2024 at 11:00

My thoughts :

Using PWM (with a reasonable frequency) with an increasing duty cycle for the heating element rather than just on/off will generate smoother more reliable heating curves. Remember to insulate the block and don't make the block too small (too low a heat capacity) or too big.

For manual detection a usb camera is a great idea (beats bending over the device peering through a lense). For auto detection I think teaching a camera to recognise the point of melting might be a step too far. Using a laser (or even just an led) with a photodiode or photo transistor to measure transmitted light levels such that you recorded a transmission curve might be a better approach. You could similarly record the curve during cooling, although I've personally never had much success in determining melting point during cooling. Beware of using too high powered a light source as that could heat the capillary above the block temperature.

I would arrange my aluminium block with a hole for the heater on one side and three holes on the far side. Use the middle of the three for the capillary tube and use two thermistors in the other two (one between the heater and the sample and the other on the far side of the sample). That way you can adjust for any temperature profile across the block. You'll need to consider differential expansion of the glass and aluminium in deciding what size hole to use for the capillary and also whether you need to use any liquid heat transfer compound between the capillary and the block (and what impact that might have on light transmission in an automated setup.)

Good luck!

Rainwater - 29-5-2024 at 12:33

I think the simplest setup with be the heater, thermometer described by RedDwarf, and use tye camera to make hyperlapse video with a time/temperature overlay included. Teaching a computer to see is not your goal. With controlled heating you can quickly ramp your tempature up to a starting point, then slowly increase the temperature until you see a melt.

With enough insulation and something like MAX31856, and a properly filtered dc power supply, you can get about (1/128)K of resolution. My little setup doesnt have enough filtering or a properly shielded thermocouple, i only get about 1/85K of precision and the rest is noise. Running off a battery and wrapping the thermocouple leads in foil, i can get about 1/100 (millikelven).

With a large enough sample, and proper heat transfer to the sample, and accurate monitoring of heating power, and a lot of math. that would be enough thermal resolution to calculate the extra energy needed for a phase transition.
Going a few steps deeper into the math, you could draw some calibration curves and calculate the thermal mass and specific heat without a sample,(analog to zeroing out your scale) and then with a sample(analog to weighting a sample) and know another property. Could be useful. Anyway I do something similar with my distillation setup to try and calculate the distillation rate and composition. Still havent got it to work. Thermodynamics is not as easy as the books make it out to be

Sulaiman - 30-5-2024 at 00:15

Quote: Originally posted by Rainwater  
With enough insulation and something like MAX31856, and a properly filtered dc power supply, you can get about (1/128)K of resolution. My little setup doesnt have enough filtering or a properly shielded thermocouple, i only get about 1/85K of precision and the rest is noise. Running off a battery and wrapping the thermocouple leads in foil, i can get about 1/100 (millikelven).
out of curiosity and possible interest,
Have you calibrated against any mp references?
What do you estimate your errors to be.?

PS since a microcontroller is used,
I would want some kind of calibration/correction table/equation in non-volatile memory
and an inbuilt callibration procedure.
(similar to doing a multi-point re-calibration of an analytical balance)

[Edited on 30-5-2024 by Sulaiman]

Rainwater - 30-5-2024 at 06:44

I use distilled water, and saturated NaCl solution, then methanol and ethanol 195 proff based on a hydrometer hydromoter, dam. Glass thingegy that bobs up and down when you catch an alcoholic.
This gives me a good 4 point calibration that is within my normal range.
The chips are very consistent with the datasheet. About 0.3 lsb per degree K, the thermocouples are the major source of offset, each needing to be independently calibrated. But as far as i can see, within a 200 degree range, a linear calibration will get you within 0.010 ~ 0.020 degrees of accuracy. Thats within swings of barometric pressure.(misspelled that one to).
Pointers -
every connection increases the amount of offset error and linear deviation, by creating a new cold junction. When you have to make more than 1 connection, its best to use the exact same material. Took me hours to figure out that the aluminum spring inside a wirenut connecting 2 copper wires was the cause of my offset not being linear. Same goes for connections to pcb boards. Anytime you change material, you create another cold junction. Like lead solder, onto tin plated through hole on to copper trace. And thats just to mount the ic chip. You get another one mounting the terminal adapter, and again connecting the thermocouple. So there are a lot of itty bitty things that throw off the ideal conditions the chip will work with. All that said, for 20 bucks and 0.01 resolution at a sample rate of about 90ms. Thats a lot of bang for the buck. It also works with about 10 different types of thermocouples. I mainly use type k, b, w5. Type B is more accurate, w has a higher range, but the chip doesnt support correction curives for ether of those so programming is a little more involved. You can access the raw data from the amplifier/adc and calculate the temperature yourself.

[Edited on 30-5-2024 by Rainwater]

RedDwarf - 30-5-2024 at 10:41

I think it would be easy to get carried away worrying about the absolute accuracy of the temperature measurement. Firstly a far greater inaccuracy in the process is introduced by the determination of at what point the material is melting (especially with impure materials), and secondly melting point determination is generally a presumptive analysis (I think I've got X, does the data support that?). I'm not going to feel any more confident that I have material X because my apparatus told me the melting point was 181.5 C rather than 180.0 C , if I want to determine whether I have material X or material Y which have melting points within a few degrees of each other I would be turning to an alternative form of analysis.

Twospoons - 30-5-2024 at 13:29

Given the inherent non-linearity of thermocouples I'd be skeptical of claiming 0.01K accuracy with just a 4 point calibration.
You'd do better to use a Pt100 or Pt1000 sensor, in a 4 wire configuration.

Rainwater - 30-5-2024 at 13:52

I have you agree with you Twospoons. I get accuracy and precision mixed up. I really dont know how accurate the setup is.
It would be better for me to word it as, i can get a stable reading down to 2 decimal places.
Edit:
Now im curious.
Non linear numbers from the datasheet.
Screenshot_20240530_180901_Drive.jpg - 326kB
Type K, under normal operating temp
-200°C to +1372°C
has an error between -0.13 +0.12c.
After the chip performs its magic, I apply a non linear offset using a 4 point polynomial, the c code has been my toolbox for years, i dont think i wrote it but cant remember, probley came from a book. But it is just an offset correction
Taking 10 samples a second, i fill a ring buffer then take the rms average of the dataset.
Yep. Worth the 20 bucks.

[Edited on 30-5-2024 by Rainwater]

[Edited on 30-5-2024 by Rainwater]

Twospoons - 30-5-2024 at 18:29

Quote: Originally posted by Rainwater  
I get accuracy and precision mixed up.
[Edited on 30-5-2024 by Rainwater]


You are not alone. Plenty of engineers have fallen into the trap of thinking 24 bit precision is the same thing as 24 bit accuracy. Its pretty easy to build a 24 bit DAC where the 6 least significant bits are rubbish. Thats where the term ENOB comes in - short for "Effective Number Of Bits"
Metrology is a science all by itself.

wxyz - 30-5-2024 at 21:46

Sure thermocouples can be inaccurate. Just build a correction table into from output voltage to temperature into your arduino software (like linear interpolation of several calibration points). Depending on your patience you can then make it very accurate. The calibration can be right from the published thermocouple data, or if you can borrow a Pt100 reference you can make a table yourself right on the setup you have. As long as your thermocouple gives repeatable results (hint: get a good one) you can make the readout as accurate as the device is repeatable.

Other points: As someone else said def should use PWM power source. Not a rheostat. You want to be able to select deg C per minute. My Metler Toledo FP62 allos for 5 to 0.5 deg per minute, a good range. Note that the rise rate needs to be monitored and adjusted by the arduino as it will slow as the temp increases.

You can simplify your project a great deal by buying one or more non-working instrument for parts and the overall chassis.

Twospoons - 31-5-2024 at 14:47

Quote: Originally posted by wxyz  
As long as your thermocouple gives repeatable results (hint: get a good one) you can make the readout as accurate as the device is repeatable.


And this is the biggest issue. Many system factors can alter a thermocouples response - even bending the wires (strain can affect the Seebeck coefficient).

If you want accuracy its always best to start with an accurate sensor, rather than try to compensate for an inaccurate one. There are always other considerations of course - finding a Pt RTD that works at 1200C is going to be challenging.

Rainwater - 31-5-2024 at 19:37

Even with the thermocouples non-linearity, for this application, it might provide enough information to generate a power vs temperature graph, that will show a phase transition. Thermocouples may not be accurate, but can be extreamly precise and fast acting.

I really like the idea of using the graph to try and detect a phase transition reguardless of the technology used.

bnull - 1-6-2024 at 04:42

Don't forget the sounds. A bip when the device is turned on, a bip when it is turned off, that kind of thing. Not really important but they give a nice touch.

RedDwarf - 1-6-2024 at 06:10

Unless you want the device to measure melting points over 300C I would opt for a thermistor over a thermocouple. Just as accurate and easy to calibrate without any of the connection issues that affect thermocouples, a simpler and more reliable circuit and solution overall imo.

monolithic - 13-6-2024 at 15:56

I disassembled a Mel-Temp 3.0 a while ago. They used PT100 RTDs for increased accuracy over thermocouples. Ordinary type K thermocouples typically have an accuracy of greater of 2.2°C or 0.75%. A class A PT100 is +/- 0.15°C.

Sulaiman - 13-6-2024 at 22:50

I consider the most important aspect of temperature measuring devices to be Repeatability.
Everything else can be compensated for.
For a mp determination device my first choice would be Pt100 or Pt1000 (3- or 4-wire), second would be Thermocouples.
Whatever the temperature sensor is,
I would try to calibrate the device vs. reference mp samples.