While I don't consider 2 weeks of trouble free use a definitive reliability
test, I'll gladly share my opinions.
Upon inspection I noticed a diode tacked on top of the PCB, presumably to
prevent future failures. I looked at where it was connected (to base of
relay drive transistor) and IMO it wouldn't do anything there, but used that
as a data point for the probable failure vector. It's common in relay
circuits to use a catch diode to protect against damage from the flyback
spike (when the relay opens). In this case the catch diode was a relatively
slow (1n4003) rectifier diode. I changed it to a much faster(1n4148)
switching diode. I also noticed that there was no low impedance capacitors
across the power supply (doesn't matter how fast the catch diode is if the
power supply won't hold the spike). So I added (.1uF) ceramic disc
capacitors to the PS in two places. This is pretty much speculation and
since the design is what's called a hot chassis (not isolated from 120v) I
decided not to probe around and look at waveforms.
I believe after the mods the unit actually functioned slightly differently
(more distinct hysteresis in heater control), as I could hear the blower
speed change every time the heater cycled on and off. However I only roasted
3 times before the first failure and am not interested in removing the mod
I can share specifics with anyone handy with a soldering iron, but I repeat
a few weeks is a pretty short life test. So this is not proven to be "the"