The Hot Log

Project: April - May 2021

Writeup: July 2021

Freshly warmed hot log

This dangerous, shiny log is an electric muffle furnace. I made it to give me better control over the various heat treatment processes involved in making blades. Hardening steel with a blowtorch is certainly doable, but you're greatly limited in terms of the kinds of steels that you're able to use. Most steels outside of simple carbon steels require more precise temperature control and hold times to reach the high performance they're designed for. I've had a lot of success using simple carbon steels like 1095, but I'm keen to try out some steels with higher potential performance. Some simple stainless steels (which often need to be held at their hardening temperature for ten minutes or more) will be first on my list.

Lots of the DIY "heat treatment ovens" you see around are made from fire bricks encasing some kind of coiled wire element, but the design has always seemed somewhat inefficient to me, certainly for knife making. With this cylindrical design, the internal area is greatly reduced in comparison to a more traditional oven chamber, but the internal temperature is extremely even as the element surrounds the entire chamber symmetrically.

Expensive tube + expensive wire = crunchy hot log center

This is a 450mm tube of Aluminous Porcelain with a 60mm ID. I believe they're generally used to make honking great rheostats and high-power fuses, but it's ideal here because of its stability at extremely high temperatures (1500°C and above). It's wrapped in 20m of 1mm thick Kanthal-A1 heating wire, an iron-chromium-aluminium alloy designed for similarly high temperatures. That gives a coil resistance of 35Ω for a total heating power of 1600W. It's all glued together with fireplace cement. The hardest part of making this assembly was drilling the small holes in each end of the tube to give me somewhere to anchor the wire. The porcelain is ferociously hard, and ate three carbide drill bits to get through just 10mm of it.

Hot.

Once the coil was dry, it was wrapped in 100mm of ceramic insulation and then crammed in a galvanised chimney tube. I thought I'd be able to compress the roll enough to fit inside the 250mm pipe I'd ordered, but the ceramic was too dense. I had to cut a slice out of the tube and rivet it all up!

The end of the chamber was sealed with a little disc of tile, with an opening for the ceramic-covered K-Type thermocouple to pass through and protrude about 150mm into the chamber (see above image). This was done so we're reading as close as possible to the center of the chamber. The probe and power wires snake out the back of the sealed outer shell to the controller box.

Gooey!

There are plenty of PID controllers available to control exactly this kind of system, but I decided to do things the hard way. I saw Marco Reps'* video on the Tinkerforge system and my mind was immediately full of ideas.

*An absolute legend, and one of my only two Patreon subscriptions. I just have a thing for dorky German dudes ok?

So I picked up a thermocouple reader, LCD touchscreen, solid-state relay and a Pi Zero control brick and started putting stuff together! The sheer power you get from being able to use this cherry-picked, validated, documented hardware in a language of your choice is just incredible. Everything worked like a charm, and I had a working unregulated control system with a simple GUI up and running in an evening. Absolute madness. The full controller code is here: here, but the long and short of it is: you set a temperature and the PID control loop sets the heater power as needed. There's even a little graph you can use to monitor stability! Hours of fun.

The Pi Zero running the whole show, display, thermocouple reader, relay with heatsink and a fan were all crammed into this cute little enclosure. The neons on the front indicate power coming in, and power going out via the relay. The heater indicator is particularly cool as it's paralleled with the heater coil itself so flashes with the same PWM frequency for some very satisfying live feedback! I did spend a non-zero amount of time planning the layout and wiring internally, so it's not too bad. But I always end up wishing I'd done a PCB everything could just go on instead of connecting it all with spade connectors or whatever, no matter how neat they are. Maybe in V2 eh? (Put that on my gravestone)

After I'd finished reinventing that wheel, then I had to tune the thing. This is the process of finding the PID parameters that give you the desired performance. It's a balancing act between heat-up speed, risk of overshoot and stability. I eventually settled with a fairly laid-back proportional term, heavy derivative term to kerb the intense thermal momentum of the system, no integral, and a slight positive bias to let it sit right on the setpoint in the middle of the usable range. I figured stability was more important than needing to stick to the setpoint perfectly.

"The only difference between science and screwing-around is writing it down"

The heating graph illustrates the tuning pretty well: no overshoot, no oscillation. At the 500°C setpoint, the stable temperature is bang on. But at the lower setpoint (230°C) it's slightly above and at the top end (900°C) it's slightly under. I guess this is due to the losses being non-linear, so a similarly non-linear bias term would probably sort it out. But I'm happy enough with this performance!

Cool it down now.

This less interesting graph just illustrates the impressive performance of the insulation. After a full power off from 900°C, we were still above 100°C after four hours. Now that I think about it, that non-linear bias term I want is probably proportional to the derivative of this curve. (TODO: Check the curve is the same from different starting points, implement something.)

The next challenge was calibrating the thermocouple. Sure, it's reading 200°C, but how do you know that's right? I picked up a range of temperature-indicating sticks which are like little crayons which melt at finely-specified temperatures. They were a bit finnicky to use, but after various tests I was happy enough that the indicated temperatures are within ~30°C of reality across the range. This seems high, but it's relatively linear and so simple to compensate for. It's also easy enough to dial in the temperatures required for any given steel based on measured properties (hardness etc.), so the absolute accuracy of the dial isn't the be-all and end-all.

So after about two months of fiddling I finally felt comfortable putting a knife in it. And NO, the 1095 heat-treat won't be any better than what I was getting with the blowtorch setup, but it's now quantifiable and repeatable. Also I can use it silently in the middle of the night I suppose? Perks.

No, I didn't do the "1" upside down and then redo it...