I'll go for a twofer.
Pretty but flawed:
Lead acid battery charger, almost 200W (nominal 100W, but it limits closer to 200, who's counting?).
Pretty: PCB; enclosure; printed label; SMPS; the transformer is wound on a proper bobbin, with proper yellow polyester tape; EMI is quite low (extra filters in the enclosure, besides what's on the board).
Flaws: resistors, diodes and capacitors tack-soldered onto flying leads (the TO-220F is a snubber diode, formerly a UF5406 that desoldered itself; plopping in an SiC schottky shut it up!); the LED wires lack strain relief; the fan (and output side LED) is powered from a series dropping resistor from the output, so it's parasite-powered when left wired to the battery; etc.
Ugly but flawless:
Ugly: just look at it. It's been sitting out on the project pile for years, neglected. It's a mashup of PCB and deadbug construction.
Flawless: it's a textbook example of the circuits used (and not in a bad way), it's efficient, it works great, and it's not very noisy (I wouldn't say silent, but more filtering can be added if that turns out to be necessary). At least, last I recall from working on the project, again it's been a while. Though, I may run out of available supply current once I build the next section, which would then be a pretty clear flaw. Not sure.
Why is it a mashup? What the heck is it?
It's the power supply to my old tube scope project. So, rough specs:
120VAC input
CRT filament: 6.3V 0.6A (HF AC), HV isolated
CRT cathode and bias: -2000V 1mA, zener regulated
Heaters: 6.3V 10A
B+, B-: +/-250V 100mA
So, a bit over 100W total (maybe 2-3A mains fuse would be adequate).
The original circuit dating from 2010 was a forward converter,
https://www.seventransistorlabs.com/Images/Tubescope_Supply2.pngwhich I underestimated the transformer size required to pack in all the windings (at adequate wire ampacity plus room to keep the HV windings' impedance high). Result, nasty overshoot on the HF windings. The UF4007 melted right away. Take two, doubled up diodes (two UF4007 in series for each diode shown). Better, but one diode always got meltingly hot -- reverse recovery loss is always a runaway condition, it's just a matter of keeping temperature low enough that it doesn't actually tip over. These... weren't. Take three: with 5.6k (1W) + 100p dampers in parallel with each diode pair. Helped reduce the peak voltage, but burned a lot of power, and the diodes still weren't feeling it. (You can try to mitigate a bad transformer, but there's only one solution to the reactive power stored in its parasitic capacitance and stray inductance, and that is: not storing it in the first place.)
And yes, that was built on a hand made, single side PCB. The biggest I had made to that point (4 x 8 inches), and on the only PCB stock I had at the time, which was embarrassingly thin. So this big heavy power supply was also very floppy, out of its mounting bracket...
Much later, I picked up the project again, said
fuck it, and literally cut the board in two, saving only the resonant HV circuit and mains rectifier and filter circuit. With new PCB stock in hand, and a newly learned fondness for deadbug, I built a UC3843 based flyback supply, as basic as can be. STW11NM80 switch (which I had a few leftover/salvaged from an earlier induction heater project), and learning lessons not just from the previous disaster transformer but also general design lessons (TLT theory is the most general and helpful approach), I designed the transformer windup so that no* common mode voltage appears between single-layer windings, giving maximum bandwidth -- sharp edges, so that when the transistor turns off, the secondary diodes catch almost instantly, without current flowing through common mode capacitance, bouncing around winding wire lengths.
*Still some, but only due to leakage, not the whole winding voltage.
And this worked great, except it had a tendency to 1. overheat the HV rectifier diodes anyway, and 2. run away and flame out the transistor.
Which was also fun for a secondary reason: the fuse took so long to clear (allowing the transistor to flame out, not just pop) because of the high impedance of the isolation transformer I was working on. The lesson being: higher available fault current can actually be safer, when fuses are involved. Limiting current to a low value, where a fuse may not blow at all, is rather more dangerous than leaving it unlimited!
So for my final step, I added a TC4420
at the gate to drive the shit out of the transistor, and a dV/dt snubber to ensure controlled commutation time and ringing. I also changed the HV diodes to SiC schottky, which despite their higher capacitance (mainly at low voltages, which looks very much like reverse recovery, but over a wider voltage range, and which isn't lossy), have far lower losses, and better dynamics (less ringing). Recovery really is just that much of a problem, even on fast recovery (~30ns) types.
Also you'll note, the HV filter caps are film (8.2uF 250V PE), mainly because I happened to have a bunch on hand, but they filter nicely, too.
Tim