Raspberry Pi Heating Things

Raspberry Pint, 27th February 2018

Ben Clifford
benc@hawaga.org.uk
www.hawaga.org.uk/ben

Flat heating: background

Radiator slow ramp up

Flat heating: units

Measuring temperature

Flat heating: fan heater

Fan heater: units

Flat heating: X10

Flat heating: energenie

https://energenie4u.co.uk/catalogue/product/ENER002-2PI

Flat heating: python thermostat

Flat heating: summary

Fan heater
Energenie
Python thermostat
ESP8266
One-wire thermometer

Cooker: sous vide

Now the next project. I quite like cooking. There's a particular cooking technique called "sous vide" which some people are very enthused about, and which you can buy all sorts of hardware for.

The basic principle is that for meat to be cooked, it should be at a certain temperature all the way through - when you're doing a roast and stick a meat thermometer in, for example, the meat probably should be about 65 degrees centigrade in the middle.

In an oven you usually stick the oven on much higher than 65 degrees, eg 180 degrees, and wait for the heat to propagate in. If you leave it too long, the whole thing ends up at 180 - way too hot - it gets burned.

The idea of sous vide is that you heat the whole of a piece of meat to the target temperature - say 65 degrees - by putting it in a waterbath that is at that target temperature (in a bag - else you'd get some kind of weird soup) and then wait a long time.

The timing for this can be much less accurate: although the meat will overcook if left in too long, that is much less pronounced. So I could have steak in for 45 minutes or for two hours.

What comes out then is a piece of steak that is (for example) medium rare all the way through -- even on the edges!

So then you stick it in a really hot frying pan for say 45 seconds a side. This is to crisp up the outside, without further cooking the insides.

It's a pretty common food hobbyist project to build a sous vide cooker. And that's what I did.

Slow cooker

http://www.argos.co.uk/product/7242524

Slow cooker controls

Waterproof thermometer

https://www.sparkfun.com/products/11050

sous vide: temperature control

sous vide: more accuracy

PID controller

proportional
integral
derivative

Proportional

The first component is the proportional output. Rather than having the heater completely on if the temperature is too low, and completely off if the temperature is too high - instead, the heat output is proportional to the error: the difference between the target and the measured temperature. So if the measured temperature is really low, the heater *will* be completely on; and if the measured temperature is really high, the heater *will* be completely off; but in the the heater power is in proportion to the error.

Now, these energenie devices only go on or off. They don't do "proportional" - I can't ask it to supply 100 volts to the slow cooker to get a half power cooker. But a standard technique for turning a high frequency digital output into a low frequency analogue output is pulse width modulation, PWM. It's how, for example, you control the brightness (or colour) of LEDs. You pulse your output on and off in the appropriate ratio: so 25% power means power on 25% of the time, and off for the rest of the time.

With an LED that might happen at say 100Hz - fast enough that your eye can't see the pulsing. With the sous vide cooker, I have it approximately one pulse per minute - about 0.02Hz - fast enough that the cooker bowl/water can't "see" the pulsing.

The main limiting factor here is that the energenie box is a mechanical relay which I don't want to switch too often, so as to not wear it out.

There's a more serious problem with proportional control: this graph shows proportional control settling nicely on a temperature, just below 61 degress. But the target temperature is actually 62 degrees! We need to keep supplying power even when we're at the target temperature to maintain temperature - but proportional control won't do that: being at the correct temperature means no error which means no power. This 61 degrees point is where the actual equilibrium occurs.

Integral

Derivative

The third term is derivative. This isn't used quite so widely and I only implemented it out of interest. The derivative term looks at the slope of the temperature and uses that to predict linearly what the temperature will be in, say, a minute's time.

Then the proportional control acts based on that prediction, rather than the current temperature.

The practical effect is that this can flatten out oscillations: in this graph, the temperature overshoots once, and then really quickly stays on target.

The main practical problem I had here was deciding how to compute the differential: you can't just keep comparing the output of the thermometer with the previous reading, because that has this very sharp increases of a fraction of a degree. So I do an exponential average over the last minute or so.

Cooking an omelet

So that's getting the water bath to a stable temperature without any disturbances.

But in real world use, once it's at temperature I'm going to introduce a big disturbance: I'm going to put food in to cook. That's the ostensible purpose of this device.

I'm going to cook a bag-omelette at 73 degrees. So I heat the water to 73 degrees - nice and stable. Then I decided that as I knew the egg would bring a sharp drop in temperature, I'd heat the water a bit extra and change the setpoint back to 73 when I put the egg in. So there's this period here when I put the setpoint up, and then it stabilises.

At which point I put the egg in and drop the setpoint back to 73.

The rest of the graph is the egg cooking, with me occasionally moving stuff around and fiddling. No where near as smooth, but it kind stays in the right area. Every time I fiddle with the egg/water, there's some disruption - even just taking the lid off causes a noticeable disturbance.

Eating an omelet

sous vide: software

image: carritech

I initially implemented this in python as a really lame script, without much error handling. It crashed one day and turned my sous vide chicken into a bag of soup.

I'm interested in different programming languages, and I've been waiting a few years for a vaguely relevant project for erlang - which is a language originally for building telecoms systems.

It's not a "traditional" raspberry pi language, but it is included in raspbian - so easy to get started.

It's intended for fault tolerant, soft realtime systems. Just what i need! The model is that you have lots of concurrent processes which work asynchronously - so I have a process controlling the PWM, and another process measuring temperatures, another process doing the calculations, another one in charge of logging data.

I think the basic PID controller could run on a tiny arduino, rather than a whole Pi - but as always it's nice to have the extras of running on a unix system - for example, it logs in CSV format and so these graphs were easy to generate, and it's useful to control it over the network.

... and I wanted to learn erlang

Futher work

accuracy
circulation

sharing radio control

If I do more work on this, what would I do?

I think there's a big problem with water temperature at the moment: those graphs show the temperature being controlled to an almost perfectly flat line. But actually that's not what they show: they show the output of the temperature sensor being controlled to a perfectly flat line. That's something similar but not the same.

First, temperature sensor is only rated to an accuracy of half a degree - so I could be getting the temperature perfectly half a degree out.

Secondly, the temperature sensor is reporting the temperature of the water where they are in the water bath. A professional sous vide device circulates the water around, and people on the internet have reported a 1 degree difference in water temperature at different points in their bath. I think this is likely to be especially pronounced right at the time that it really matters: when the food has just gone in and we really need the water to be the right temperature. Some kind of circulation mechanism would be good - I don't really know what: the hobbyist community seems to use tropical fish tank pumps with mixed results.

The third problem is that although I have two energenie plugs, there's one radio transmitted controlling them both. The fan heater and slow cooker code can't both be driving the GPIO pins. Probably solution is a network protocol - eg HTTP for widespread support.

Raspberry Pi Heating Things

Flat heating: background

Radiator slow ramp up

Flat heating: units

Measuring temperature

Flat heating: fan heater

Fan heater: units

Flat heating: X10

Flat heating: energenie

Flat heating: python thermostat

Flat heating: summary

Cooker: sous vide

Slow cooker

Slow cooker controls

Waterproof thermometer

sous vide: temperature control

sous vide: more accuracy

PID controller

Proportional

Integral

Integral

Derivative

Cooking an omelet

Eating an omelet

sous vide: software

Futher work

The End

blank slide