PID temperature controller

[Prune]

I want to mod my espresso machine with PID temperature controllers, since I notice too much temperature variation during extraction, which affects the end result. For comparison, Caffe Artigiano downtown in Vancouver use a customized machine with fancy PID that costs more than a lot of new cars, but the espresso is fantastic (anyone in Vancouver reading this, if you haven't tried these guys, it's the best in town).

However, PID controllers are very expensive, like $100 on eBay -- and I need three (one to preheat incoming water to the boiler, another for the boiler itself, and a third for the grouphead). So my question is where do I get them cheap, OR how do I DIY the thing (programming microcontrollers is a problem as I don't have the setup for that).

[Moose]

You can do it with opamps, I will try and dig up some schematics for it. The complexity depends on whether you want P, I and D control. I'm not entirely familiar with expresso machine operation, but if a small margin of temperature error is acceptable, and the rate of change of the temperature is is low enough you can simply use just "P" or Position gain. This is very easy to do with op amps, and circuits could be construsted for much less than $100 each.

[leadbelly]

Using P only will be of no advantage, since you will always have an offset from your setpoint with P only control, you need PI as a minimum. Can you go the old cheapo PC route and hook up relays to the parallel port for output? Getting the temp into the PC is another issue, but there are various DIY ADC techniques for that.

[Prune]

Assuming I'd need to buy the programming interface, how much can I expect to spend on a microcontroller design?

Temperature variation of 0.5 K is a good target, but as new cold water is pumped from reservoir into the boiler, the controller must handle the large transient properly. Preheating incoming water is a mediocre solution. I think the best approach would be to build a boiler from scratch using a piston and two one-way valves, with the grouphead forming the bottom of the piston to help it maintain proper temperature. That reduces from three to one temperature controllers (plus it allows more precise pressure control).

Caffe Artigiano downtown have customized machines with PID and water remineralization, but they cost more than a lot of new cars, plus they need to be plumbed in and are quite huge. On the other hand, the end result (of course barista skill plays a good part too) is the best espresso I've ever had (including ones I've had in Italy). As I can't possibly get such a beast, I figure I ought to build one of my own

[gearheaddruid]

Hi, I do not know if these would work. I have some industrial temperature controls from a plastic molder. They have variable on duty cycle and adjustable Hysterisis. I was going to post some on E-bay. I saw the thread and thought if you could use them, I will be glad to work a deal with you. I can e-mail the pictures to you.
They were originally designed to plug into a multiple chassis but could easily be hard wired. Regards, Steve

[Moose]

Originally posted by leadbelly
Using P only will be of no advantage, since you will always have an offset from your setpoint with P only control, you need PI as a minimum. Can you go the old cheapo PC route and hook up relays to the parallel port for output? Getting the temp into the PC is another issue, but there are various DIY ADC techniques for that.

You can use P only:

Assuming the water will never grow hotter by itself, you can use the input of your system as the difference between your setpoint temperature Ts and your actaul temperature, then apply some gain to this difference and use this as your driving function (Vout).

Vout=G(Ts-Ta) <-- at setpoint output is zero

For this approach you might be able to get away with a thermistor, or a thermocouple, probably hooked into some sort of bridge to get the range right, and then your reference temp could just be set py a small potentiometer, use an opamp to get your Vout. Assuming the heater is on/off only, you can run this into a 555 chip set up for pwm duty. This will give you simple control of the heater to turn on when the temperature drops below the setpoint.

Its not ideal, as derivative gain would allow you to sense the rate at which the temperature is dropping, and as compensate for this. I gain can be eliminated with enough gain on P. I find with all the PID systems I have tuned that I gain usually leads to system instability.

[Prune]

My target is something not worse than the graph over a 27 second extraction time (pump forces cold water into boiler @ around 13 atmospheres), up to 2 oz will flow through. That's a variation of less than a quarter Kelvin. Now, this is the grouphead temperature, not the boiler temperature, so it's not as easy to do.



What's the simplest controller design that I can use to accomplish this? I see no reason I can't drive the heater variably through transistors, as it's just a resistive element (though resistance does vary with temperature). Also, I don't know about choosing a temperature sensor, and how to calibrate it right, as I don't have a thermometer that shows sub-degree accuracy.

[Moose]

A simple NTC thermistor is $0.30 at digikey

http://ca.digikey.com/scripts/DkSear...120846&Site=CA

You will need some way of calibrating the circuit though. Thermistors are very accurate, but only as accurate as the circuit they are contained in. Maybe get a thermocouple reader on ebay?? That would give you good accuracy.

[Prune]

You said I tends to instabilities. How do I do the D then?

I was looking at this generic opamp design:
http://www.ecircuitcenter.com/Circui...pid/op_pid.htm
So I could use something like this, but omit the I? One question is how I can get useful data out of this for tuning. Are there any IC controllers I can use?

Also, the heater is in the boiler, but the temperature I want to control is at the grouphead, which means there's a significant delay in controller response, so either some predictive behavior is needed, or somehow using more than one sensor. It's not clear for me how to best do it. Additionally, the heater has limited output, so controlling large transients is a problem.

[Moose]

I was gonna post the same link. Thats a good site that displays the basics, and explains it in a fair amount of detail. I would leave the I in, as it may be needed, and you can always set its gain to zero if it does make the system unstable. Really that explains it all, and with the addition of power stage for driving the heating element, it could be used almost as is.

As for useful data, it really depends on what equipment you have to work with. A simply multimeter will be able to read the Verr voltage, so that will give you a good indication of the deviance from the setpoint. Really it can be done non-scientifically by trial and error to see how the system reacts given different gain values.

As for IC controllers, I am not entirely sure what you mean?

The predictive behaviour is nearly impossible to do without using a PIC, I'm sure it could be done, but the amount of effort it would take to get it right would be enourmous IMO.

Could you maybe send water throught he grouphead first to preheat it to approx the right temperature? I've never used an espresso machine before so I might be way out of line with that suggestion. Just to be clear, where are the heating elements located and how many are there?