Measuring temperature with high precission - Questions

[J_Nichols]

I'm trying to build a circuit to control and measure the temperature of a metal.
The accuracy I want to measure is 0.1 degrees Celsius.

I'm building a circuit to control the temperature of a electric stove like this:
**broken link removed**

using pulse width modulator to allow/stop the electrical flow to the stove. Basically is a circuit that opens/close the relay.

In the other part, I was asking if this bought would be ok or there is something better to measure the temperature.
**broken link removed**

The temperature I want to work with is around 320 degrees C.

 

[jpanhalt]

One of the problems you might have with that approach is that the metal itself will have temperature gradients in it. You might be able to measure one spot accurately, but on a hot plate like that and depending on the shape/size of metal, some of the gradients might be quite large. Just to convince you, try a liquid crystal temperature stick or IR temperature device and see how much variation you get.

John

 

[J_Nichols]

But I thought about it and I decided to use sand above and below the metal, to prevent the direct contact of the metal with the surface of the stove. Also, I would use small flat pieces of metal, to prevent the temperature gradients as much as possible.

 

[crutschow]

I'm trying to build a circuit to control and measure the temperature of a metal.
The accuracy I want to measure is 0.1 degrees Celsius.

I'm building a circuit to control the temperature of a electric stove like this:

using pulse width modulator to allow/stop the electrical flow to the stove. Basically is a circuit that opens/close the relay.

For such precision control you would need the PWM on/off cycle to be several times a second to avoid ripple in the temperature, so you should be using a solid-state relay controlled from the PWM. A mechanical relay would rapidly wear out.

Be aware that such a temperature control loop can be subject to low frequency oscillations due to the thermal lag in the system. You may need some sort of lead-lag compensation to stabilize it (PID).

 

[Jaguarjoe]

That $19 eBay thermometer looks nice for the price but at 320C it's only accurate to 1 degree, not 0.1 as you want.
You may want to build an expanded scale thermometer with a range of 310 to 330 or so.
If you use a quartz thermometer, its output will be a frequency which can be read very, very accurately.
Honeywell makes "LN" series linearized thermistors. You can make your own linearized thermistors for your temperatures of interest using the formulas readily available on the web.
How will you calibrate this thermometer? Generally, you need something at 10x better than what your calibrating.

 

[Reloadron]

Here are the accuracy specifications for the measuring device you linked to:

50 ℃ to 199.9℃...0.2% rdg + 1 ℃
-50 ℉ to 199.9 ℉ ...0.2% rdg + 1 ℉
-50 ℃ to 1000 ℃... 0.3% rdg + 1 ℃
-50 ℉ to 1999 ℉ . ..0.3% rdg + 2℉
1000 ℃ to 1300 ℃..0.5% rdg + 1 ℃
(apply to mainframe only at 23℃±5 ℃)

Despite a .1 degree C resolution the accuracy in the range you desire well exceeds .1 degree C. Then you have another problem with the sensor, in this case a Type K Thermocouple:

LIMITS OF ERROR
(whichever is greater)
Standard: 2.2°C or 0.75% Above 0°C
2.2°C or 2.0% Below 0°C
Special: 1.1°C or 0.4%

Even a Type K Thermocouple made with special limits certified wire won't begin to come close to what you want. The only sensor / indicator combinations I have seen that will do as you wish for accuracy and resolution are PRT (Platinum Resistance Thermometers)

Platinum sensors care manufactured to comply with a standard curve within a specified tolerance. The international standard, IEC 751, is published in the UK as BS EN 60751. The Class A tolerance for a PT100 sensor is ± 0.25 ºC over the range 0 ºC to ±200 ºC. However, better uncertainties can be achieved if the sensor is suitably protected in a steel probe and calibrated individually.

For the highest accuracy, special glass-sheathed standard PRTs, usually of 25 ohms at 0 °C, are calibrated at the fixed points of the International temperature scale 1990 (see above). The ITS-90 specifies equations to relate the resistance to temperature and, using these, uncertainties can be achieved of 0.001 °C or better. Standard PRTs can be used from temperatures as low as 259 °C up to 660 ºC, or even, 962 ºC, with some increase in uncertainty and of loss of reproducibility.

Just the measuring accuracy you are looking for will not be simple.

As to control I believe PID is about the only method of temperature control that will work. Even with PID control there is another problem. You have pictured what amounts to a hot plate. If a vessel (pot) containing a liquid is placed on the plate you will have thermal inversion layers and need some form of agitation or circulation to achieve thermal equilibrium in the solution.

Achieving a tight temperature control of 0.1 degrees C is not an easy task, beginning with the sensor.

<EDIT> I see you are heating metal. Some of the above still holds true in that where the sensor is placed in the metal is important for thermal equilibrium to be achieved. When heating a metal part rather than a hot plate a small furnace or oven is a better choice. Since your control point will be a direct function of sensor set point the sensor accuracy and placement are very important. </EDIT>

Just My Take
Ron

 

[Jaguarjoe]

I like this:

.....better uncertainties can be achieved......

If I had a nickle for every uncertainty I achieved I'd be rich

 

[Reloadron]

I like this:

.....better uncertainties can be achieved......

If I had a nickle for every uncertainty I achieved I'd be rich

I don't like it, I am just well..... a little uncertain about it.

Ron

 

[cachehiker]

The only sensor / indicator combinations I have seen that will do as you wish for accuracy and resolution are PRT (Platinum Resistance Thermometers)

Even then he's going to need a better than standard and possibly a metrology grade Platinum RTD. I hope he's prepared for a shock.

 

[Reloadron]

Even then he's going to need a better than standard and possibly a metrology grade Platinum RTD. I hope he's prepared for a shock.

Absolutely, you would be looking at a lab grade PRT. A PRT supported with a chart showing deviation from mean over the range he wants. You want to maintain temperature +/- a few degrees C or F not a big deal but when you start looking at .1 degree it's a whole new ball game. Pretty expensive game too.

The OP doesn't mention the application beyond heating metal so who knows.

Ron

 

[carbonzit]

I'm trying to build a circuit to control and measure the temperature of a metal.
The accuracy I want to measure is 0.1 degrees Celsius.

Can I ask why you think you need such a high degree of accuracy?

A lot of the problems I see here with peoples' projects are unrealistically high expectations of accuracy/performance/quality. Does your application actually require that degree of accuracy, or are you just aiming as high as you think you can possibly go in that respect?

 

[DirtyLude]

I'm guessing sous-vide, maybe? In that case measuring the temperature of what you are heating is more useful than measuring the cooking element.

 

[carbonzit]

I'm guessing sous-vide, maybe?

Not being a Francophone, I had to look that one up; apparently it means "cooking vacuum-sealed food".

Which makes the O.P.'s desired accuracy really ridiculous. Why in the world would you need (or even want) 0.1 degree accuracy if all you're doing is cooking some food?

 

[Mr RB]

I agree with carbonzit that 0.1'C is ridiculous resolution at 320'C, I can't imagine any process that needs those 2 specs.

Also, that hotplate will probably not get to 320'C even at full bore, they usually rate the element wattage to top out at around 200'C in normal use.

Thirdly, there will be an internal temperature fuse that would blow long before 320'C even if you were to wrap enough insulation around the thing to get it to 320'C.

 

[DirtyLude]

Not being a Francophone, I had to look that one up; apparently it means "cooking vacuum-sealed food".

Which makes the O.P.'s desired accuracy really ridiculous. Why in the world would you need (or even want) 0.1 degree accuracy if all you're doing is cooking some food?

You didn't look it up very well then, because the cooking is also temperature controlled to a fairly high degree. Not .1deg, but still within 1 degree and less. Though I think 320deg is much too high for sous-vide, so I think that's out.

 

[gary350]

This works great accurate to .1 deg F or C completely adjustable with unlimited options.

**broken link removed**

 

[J_Nichols]

yes gary, I will take a deeper look at the temp controller. But I know those temp controller controls the temperature using PWM and sometimes the accuracy is 1%. That means 300ºC in a 1% system accuracy is 3ºC of possible error. That is too much. maybe would be better controlling the temp by a variable power supply? and use a better temp sensor?
How can I obtain the best accuracy at 300ºC?

 

[carbonzit]

Do you mind my asking what it is you're doing that requires such a high degree of precision? And are there existing controllers used to do whatever that is that you can buy, or copy, or reverse-engineer?

 

[Reloadron]

While thinking about this I agree it would/could go much better if the entire application were explained. However, I'll share this again in a nutshell. Strict tight temperature control is done using what is called a PID (Proportional Integral Derivative) control. You can find an overview here.

Any temperature measurement or control system is only as good as its sensor(s). Earlier I showed why a thermocouple would not be practical for measuring or controlling temperature accurately to an uncertainty of +/- 0.1 degree C. I also touched on control sensor placement. If a thermocouple were to be used it would only be possible to do so at a few temperatures. The thermocouple would need to be calibrated and charted for a few specific temperatures around its working temperature. Not a cheap or inexpensive process and done by a lab.

The uncertainty you want to achieve is a lab grade tolerance. There is no easy way or inexpensive way to achieve it. It is not going to be done using a basic temperature controller, even a basic PID controller. Just to read accurately (sans any control) temperatures with .1 degree C uncertainty you start with units like these. Then you add a Platinum PRT for a sensor. You are at about $3,000 plus USD real quick.

Now if you are absolutely sure you need this uncertainty it doesn't come cheap which is why this would go much better if the ultimate goal were known. I will tell you right now this is not going to happen with the hotplate you originally linked to.

Ron

 

[killivolt]

How about this one. Maybe your doing something in a lab that needs more result though? Fluke handheld inferred

FLUKE
FLUKE-566
799

Temp. Range -40D to 1200DF
Focus Spot Size and Distance 1" @ 30"
Accuracy Greater Than32DF +/-2.0DF or +/-1%
Spectral Response 8 to 14 uM
Laser Sighting 1 Dot
Response Time 500 msec
Emissivity Adjustable 0.10 to 1.00
Repeatability +/-0.5%
Display Dot Matrix LCD
Alarm High/Low
Data Logging 20 Points
Data Hold Yes
Max./Min. Plus AVG and Differential
Memory Recall Max/Min/Avg/Dif
Ambient Temp. Range 32D to 122DF
Safety Rating Class II Laser
Trigger Yes
Battery Type AA
Includes Battery, K Type Bead Probe, and Hard Case
Manufacturers Warranty Length 2 Years