I have essentially finished the big 10kW unit.
An in-progress shot, showing the unusually cramped chassis I built from some leftover pieces. It's just barely enough to fit everything. The cap bank fits in the rear, copper tubes going past the output transformer, which drives the tank in series resonance. 240V AC power comes in from the rear, to the contactor. Precharge resistors to the left of the contactor prevent turn-on surges. Brown wires come around from the bottom of the contactor to the rectifier, a GBPC5008 50A 800V bridge. The rectifier feeds a bank of 16 × 470μF 200V capacitors, arranged in series-parallel for a total of 1880μF 400V, with somewhat more ripple current capacity than an equivalent capacitor in, say, Computer Grade style. Wires come around the front of the divider, bringing 320VDC to the Inverter Board, which contains a mess of coupling capacitors and the output transistors. Four FGH80N60DTU IGBTs are wired in parallel pairs, half bridge, to drive the output transformer. Not visible, on the back side of this divider, a water cooled heatsink keeps the transistors and rectifier cool. On the far right, the Control Board supplies signals to the Gate Drive Boards, which are floating loose in this picture.
This is how the chassis looks, put together. I cut up two aluminum trays, which came from network hubs I think, and made a sort of clamshell structure, with front and rear cover panels. Unfortunately, they have these circular indentations, which look really ugly on a front panel, and make it difficult to mount pots and LEDs there. They can be pounded flat (one obstructed the tank cap, which I had no choice but to pound out), but this leaves hammer marks.
I had been having some problems, the main reason I took my time with this model. Here's an example of the output waveform:
Typical values: inverter output (top), 300Vp-p square wave, 20kHz. Inverter current (bottom), 80A peak, sinusoidal, nearly in phase (depending on how close to resonance it is). This is all nice and fairy-tale perfect. Buuut...
This is the same thing, zoomed out, triggered on current spikes. Unfortunately I don't have a storage scope to get a proper look at whatever's going on here.
At any rate, what it seems to be is, the high side gate drive is sticking on, causing weirdness to happen. The sudden change in duty cycle causes a large current spike, introducing an offset to the voltage waveform as the coupling capacitors charge back up (possibly saturating the output transformer as it returns to zero).
The reason the high side gate drive might stick on is because I made the unfortunate choice of 6N136 for the optoisolator driving it. This is a moderate speed opto, composed of a photodiode driving a regular transistor. The advantage of this over a phototransistor (e.g., 4N35), is you can bypass the photodiode's supply, cutting Miller capacitance significantly, so it works faster. The disadvantage is, the base node is incredibly sensitive to noise. The circuit diagram shows an internal shield protecting it, but it hardly does anything. Part of the problem is probably having the base brought out on a pin, which makes it very vulnerable.
The problem is common mode immunity. The switching speed of this circuit is around 100ns to swing 300V, or 3kV/μs. Fairchild rates their 6N136 for 10kV/μs, which should be enough. But the test is only for a 10V step, which is completely absurd! No kidding it can withstand 10kV/μs, it's injecting only 0.6pF × 10V = 6pC of charge, as long as the edge is faster than the time constant. The transistor needs more to change state. If they had tested with a more realistic step, like 100 or 500V, the dV/dt would've been embarassingly small, but at least a useful measurement.
So what did I do? I shielded the hell out of it. I wrapped the opto with copper foil and extended the ground on the circuit board. Miraculously, I haven't seen any further problems, so testing goes on.
This is literally the very first time I powered it up on the 240V, 50A circuit -- at the other end of the basement, because that's where my welder is, and where its circuit was installed. Not rehearsed, what you see is the whole thing!
The steel pipe in the coil reached a peak temperature around yellow hot in this test. All my previous videos suck instantly: this thing reaches a much higher temperature in seconds, which the 1kW model takes a minute to reach! On a later test (I don't have it on video), I kept heating this pipe. A seam melted right down the side -- that's right, I can melt steel now.
Using a much larger coil (about 7" dia., wound with 3/8" copper tubing), I can easily demonstrate skin effect with this steel slab. Video:
This coil is big enough to fit a #4 crucible, so I shall soon be melting -- that is, as soon as I get the money to buy a 50A extension cord (no way in hell am I melting steel indoors!) and a more powerful water pump (I'm still using the same old submersible pump, which is no longer enough even to keep the water from boiling in the pipes, let alone at the proper operating temperature)...
Web page maintained by Tim Williams. All rights reserved.