Convection reflow oven

For anyone doing electronic assembly at home, keeping pace with circuit manufacturing technology can be a challenge. More and more parts only come in fine-pitch and/or leadless packages. In my own experience the progression has been something like this:

  1. Point-to-point wiring, perfboards
  2. Printed circuit boards with through-hole components
  3. Surface-mount components with leads (SOIC, QFP), wire solder
  4. Surface-mount components without leads (QFN), wire solder
  5. Leadless surface-mount components with mandatory ground pads, paste solder
  6. Ball grid arrays

where I am currently somewhere between levels 4 and 5. Interestingly, after mastering the skills at a given level, I have little nostalgia for the previous levels. This is particularly true for surface-mount versus through-hole. I know some very accomplished amateurs who insist on sticking with through-hole components, but I would never dream of going back. What a pain!

Faced with some components available only in BGA packages (e.g. the latest generation of FPGAs), and an ever-increasing number of QFN-style packages with ground pads, I decided the time had come to put down the soldering iron and move decisively into level 5 and then hopefully level 6. The distinguishing technology here is the use of solder paste applied by stencil or syringe, and “solder reflow”—the process of heating up an entire board, components, and paste, soldering everything at once.

Reflow can be done on a hot plate or skillet, inside an oven, or by vapor-phase immersion into a fluid with its boiling point at the target reflow temperature. I chose the oven method, which is also the most commonly used in mass production today.

Finding a reflow oven candidate

Ovens can be further subdivided by the dominant method of heat transfer, infrared radiation or convection, conduction being negligible. Infrared ovens seem to dominate the low-end market, and most amateur toaster-oven reflow also fits in this category. IR reflow must be adequate in most cases, or it wouldn't be so popular, but it has some fairly obvious disadvantages. Dark components like ICs, which may be the most sensitive parts on the board, absorb more IR and heat up quickly, while bright shiny massive components like connectors heat up slowly. Large components may cast shadows, leading to cold spots, unless the infrared sources are evenly distributed above the board.

The modified-toaster-oven route has been followed many times by now, and there are some really nice write-ups like Dan Strother's available. I decided it would be worthwhile to try something different, even if it failed, so when I found this commercial convection oven for sale on Craigslist, I decided to attempt turning it into a true convection reflow oven:

Commercial convection oven

Commercial convection oven

In a previous life this oven baked brownies at a Ben & Jerry's franchise in Seattle. Notice the “Sodir” brand: we're halfway there already!

Ovens like this have the heating element in a rear compartment. A fan circulates the hot air into the main cooking chamber. With the back opened up, we see a circular calrod-type element and fan blade. Behind this compartment, separated by a half-inch layer of insulation, is the shaded-pole AC motor.

Rear oven compartment with heating element and fan

Rear oven compartment with heating element and fan

Any hope of using the oven as-is with a temperature controller was dashed by my initial thermocouple measurements. It took nine minutes to reach 217 °C, by which time the temperature slope had flattened out to 0.2 °C/second—much too slow for any reasonable reflow profile. The obvious solution was more power and more air.

Upgrades!

Warning: The oven modification described here is definitely NOT SAFE. If you try something similar, you assume the risks, which include electrocution, burning down your house, etc.

The souped-up oven now runs on 220V. This is a standard voltage for 5-kW duct heater (electric furnace) coils, and it's easy to find replacement kits on Ebay. The nichrome coils must be carefully insulated to avoid shorting onto the chassis, a task compounded by the fact that they will expand and sag when hot! This is a major electrocution risk if you don't take the right precautions, which may not be obvious due to the variety of 220V house wiring. I used ceramic posts arranged in a circle and tied the heating coils in place with baling wire using holes in the posts.

New heating chamber showing the 5-kW nichrome element and fan blade

New heating chamber showing the 5-kW nichrome element and fan blade

More heat requires more airflow (otherwise the heating element will burn out), which the oven needed anyway. I was worried at first that air blasting over my circuit board might dislodge components, but it turns out the surface tension of the melted solder is strong enough to prevent this. Suitable PSC-type motors and centrifugal fan blades are made for furnace draft inducers. It's important to get a reversible motor, or be absolutely certain of the rotation sense of the fan blade ahead of time. I used a Fasco D404 from Ebay.

New blower motor

New blower motor

The D404 has a fairly long shaft, so I used an extra inch of rock-wool insulation behind the heating chamber. The motor still gets hot, both from heat conducted down the shaft and the lack of airflow around the motor itself. To mitigate this I have a table fan blowing crosswise against the motor, and it never runs more than 5-10 minutes between cooldowns.

The back of the cooking chamber was opened up with a hole saw to match the diameter of the centrifugal fan blade, which is almost exactly flush with the panel. Air is sucked in here, blown across the coils, and returned through the existing rectangular holes at the sides.

Back of the oven showing air inlet for fan

Back of the oven showing air inlet for fan

In actual operation the fan makes a satisfying roar, but the airflow is rather chaotic and I haven't quite figured out why. The added turbulence should only help the heat transfer, though. The main concern is keeping the heating coil cool enough to not risk melting it. In my case, I think I could have used an even larger fan blade as the coils are quite yellow at peak power.

Temperature controller

AVR32 oven controller, baling-wire version

AVR32 oven controller, baling-wire version

The 220V circuit to the heating coil passes through a 25-amp solid-state relay (SSR). A Stackfoundry Copper AVR32 dev board PWMs the SSR at 2 Hz and reads temperature through a MAX31855 thermocouple interface. The code implements PID control with a couple of modifications. When the temperature reaches a programmable threshold, the power is set to maximum, unconditionally, until the temperature reaches a second programmable threshold, at which point the power is shut off. This ensures the maximum possible dT/dt in the final ramp up to reflow, which for my oven is about 2 °C/sec.

I wrote some MATLAB code based on this model to simulate the enhanced oven and its controller. Choosing parameters to match the observed thermal time constants, I was then able to tune the PID coefs for good tracking of the desired temperature profiles.

Simulated oven profile

Simulated oven profile

Unfortunately, applying the simulation to the real oven was not so simple. For good response you really want some derivative control (the D term in PID), but thermocouple measurements are noisy, and its derivative is noisier still. You can try to smooth or filter out the noise, but this incurs extra phase lag around the loop, making the whole system prone to oscillation. In the end, I read the thermocouple at the fastest allowed rate of 10 Hz, ran this through a single-pole IIR filter, ran the control law at 2 Hz, and used much less aggressive PID coefs compared to the model.

Results

This graph shows the desired profile, actual profile, and temperature error for the oven's inaugural run on the “scopebox” circuit board:

Measured oven performance

Measured oven performance

I'm not happy about the oscillations, but the performance is certainly good enough for now. If I revisit the derivative noise reduction strategy, these low-group-delay IIR filters designed for motion tracking might be a good place to start.

Scopebox PCB

Scopebox PCB

Next up: BGAs!

Source

  • MATLAB-compatible simulation code: oven.m
  • AVR32 C code (not in good shape; use for inspiration only): ovenpid.c

CC-SA-3.0
This work is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.