FYI... you can't just slap things together and expect them to work.
The ATX supply controller isn't a terrible starting point, but you need to provide
all the interfaces required:
1. Auxiliary supply
2. Voltage sense/feedback
3. Current fault (or sense)
4. Gate drive
1. Supply is usually from the motherboard 5V standby supply, alongside the main converter. Usually a smaller transformer and a single-transistor driver (either an old fashioned 1T blocking oscillator circuit, or the more modern equivalent with an offline regulator IC which may be transistor-sized, or DIP or SOIC).
To power the gate drive, you will need a bigger supply. Standby is typically 5V 1A or something like that; expect to need 12V or more at a few 100mA.
2. This should be okay, wired in the same way (a resistor from the output(s) to the feedback node, with common ground tied to the output). But beware of voltage and current surges even over short lengths of that ground -- use star-grounding and lots of filtering, TVS protection, and shielding preferably, to keep the controller comfortable.
3. This is your first complication. If the original circuit has a current fault (the usual case), it won't be very useful for you as such. It's usually a latching function, so as soon as you spark up (or maybe during one of the myriad sparks that makes up a running bead -- GMAW draws intermittent load current), it just stops.
This should be changed to a current limiting function, so that if current rises over the limit, PWM throttles back, maintaining the arc without destroying the power supply.
Bypassing or omitting this function will lead to destruction. You may still be able to run some beads, but the bucketful of burnt transistors and other components won't be very economical.
4. This is your second complication. You can't simply replace BJTs with IGBTs and expect anything to happen at all, let alone survive more than one switching cycle without exploding.
So the original driver circuit and transformer is useless to you. Well, you could perhaps rewind the transformer as a GDT (gate drive transformer), but you still need something to drive it in the way IGBTs expect to be driven.
You will also be using quite large IGBTs (or MOSFETs), to deliver the required kilowatts. These have continually improved over the years, but you're still looking at 100-200nC of gate charge required.
What's a nC? A nanocoulomb is a unit of charge. Charge is current * time. It takes current to change the gate voltage -- the gate has a capacitive characteristic, so gate voltage and charge are proportional -- and to move the gate in some amount of time, requires current. For a switching frequency of 100kHz or so, you need switching under 200ns or so.
So we calculate the required peak gate current as:
Qg (gate charge) / t_r (rise time) = Ipk
Actually peak current may be a bit higher than that, because waveforms, but it's in the right ballpark. An ampere peak, in fractional microseconds, also requires low inductance.
What is inductance? Inductance is the voltage drop associated with a change in current:
V = L * dI/dt
So if we're driving an amp of current into the gate, in much less than 200ns (because that current needs to rise, and then fall), and we can only afford to drop say 5V in the process, we need much less than L = (5V) * (200ns) / (1A) = 1uH.
This sets the maximum leakage inductance for the GDT.
So we already know we need a GDT of some level of performance.
5. The last, and major, concern you will have, is constructing the new inverter and rectifier entirely from scratch. There isn't anything you can reuse from the original circuit -- the BJTs are too small, the rectifier and transformer are too small, the capacitors... You will be using larger components, so even if you copy the layout and scale it for the new components, the layout will be larger.
Stray inductance limits us again here. When one transistor turns off, all the current it was carrying, has to drop to zero, and the load current has to find another route (usually a clamp diode). This change in current, must not produce an excessive voltage drop -- else transistosplosions -- and so we again can find a maximum inductance.
Say the inverter is supplied by 360VDC and delivers 20A peak, and turns off in 200ns; to use 600V transistors, we only have 240V of headroom, and so L < (240V) * (200ns) / (20A) = 2.4uH. (Probably, a time of ~50ns would be more representative -- transistors sharpen risetimes, so if you're getting 200ns at the gate, the collector will be less. We also don't want to run with zero headroom, so a voltage excess of 50 or 100V would be much more comfortable. Which gives a much safer limit of L < 0.3uH.)
What gives inductance? Length. Simple as that. The inductance of free space is 1.257 uH/m.
Because of geometry effects, the inductance/length of a typical wire is lower, maybe 0.3 to 0.6 uH/m. Circuits also contain loops -- there's no meaningful notion of the inductance of a free wire -- so we need to count the full loop path length.
For an inverter, the loop length starts at one transistor, through the catch diode (again, usually the opposing inverter transistor -- note you'll need co-pack IGBTs here), through the supply bypass/filter cap, back to the one transistor.
So this is why even merely scaling up the existing layout, isn't a sure thing. You're scaling up all the lengths, so Lstray goes up proportionally. (Incidentally, a proportional scaling keeps the geometric factors constant, so this is actually a very accurate representation. Whereas using different sized traces or spacings would change the geometric factors some.)
You likely won't succeed by bolting transistors to a heatsink and wiring them with hookup wire -- even with short leads, you'll incur loop lengths of fractional meters, and 0.1's of uH.
Example from an early project I did:
Two pairs of IGBTs, half bridge configuration (two in parallel, each side). The two copper plates are +/- DC bus (the three red caps, and two blue ones, bypass them together). The little plate in the middle is the output node (leading to the bank of many red caps, and the load, out of frame). The stray inductance of this configuration was around 20 or 30nH (0.02-0.03uH), pretty good.
Another good method is to use a 4-layer board, with solid pours: DC+ on top, DC- on bottom, and OUT in the middle. This can get Lstray in the single digits nH range. Not an extreme you need to go to, but it's pretty cheap (< $200) considering how effective it is.
Then there's the transformer. You can't use the ATX transformer, or core. You might be able to use a dozen of them wired together, if you happen to have identical supplies to gut. More likely you will be winding one new? Big cores like ETD59 and 87+ mm toroids are easily purchased, so that's fine. The wire needs some attention. You can't use solid wire at this frequency -- it'll get way too hot, skin effect. Preferably you'd use Litz cable, assembled from a gazillion strands of 30 AWG or finer. You can also use foil/tape/strip/sheet, but how well depends on the ratio (if you have lots of turns stacked up, foil will be worse than round wire; it has to be interleaved, always primary facing secondary, never primary facing primary, to work out).
Here again, we have length -- winding wire length -- and so, stray inductance, also called leakage inductance. This matters to the rectifier, depending on how you're arranging it.
If you opt for a FWCT configuration, then the leakage between ends of the secondary impacts the peak voltage the diodes will see. That is, when one diode turns off, its current flow has to divert to the other diode, and this can only be done through the length of the winding between them.
You can easily exceed the 400V limit of those diodes, even for a mere 50V nominal output, with a careless transformer winding. So it's very important.
You have some savings with a FWB configuration: there's always a pair of diodes to grab either polarity of flyback voltage. Downsides: a. you're burning twice the voltage drop in the diodes -- a big deal at 100A+; b. the diodes are clamping into a filter choke, which doesn't really clamp at all, it's just another inductance, right?... So for this reason, you may need a large TVS (or some means of hard voltage clamping) at the rectifier output, or across the diodes.
Other clamping means would include an RCD clamp snubber -- a catch diode dumps the flyback into a beefy capacitor, and a resistor bleeds away the accumulated charge. Note that diode-capacitor loop needs to be very short, otherwise the series inductance will defeat its purpose.
If you have enough voltage overhead, an MOV can be used. For a 50V output and FWCT configuration, the nominal reverse voltage per diode is somewhat over 100V. A 100VDC-nominal MOV could be used, which will have a peak breakdown voltage around 300V (MOVs are not actually very good at clamping, they're just economical), and so 400V+ diodes could be used safely.
And obviously, all the output connections need to be beefy -- you'd probably need a custom fab PCB to do this (heavy copper). A regular proto fab PCB could be used with a lot of soldered-on strap to increase ampacity. Takes a bit of faffing to do. Or it can be constructed from cut and formed sheet, insulated with plastic sheet and clamped with plastic hardware.
6. Other things.
If you're confident in modifying a circuit so extensively, why settle for the -12 and -5V undervoltage circuits? Why not bypass or remove them? Or reuse the comparator for something else, perhaps? (I...dunno what; some dumb timers or something?..)
What filtering network will you choose? A traditional choke-input, then capacitor filter is what the controller is expecting. The capacitor may work against your load, however (the arc has a negative resistance and tends to oscillate with capacitors and cable inductance). You could use just a choke, but the controller will have to be adjusted for that; it also won't be able to regulate the voltage very well, because that's what a choke does, it allows voltage to vary while resisting changes in current. In particular, the flyback voltage when the arc breaks, will destroy pretty much anything directly connected to the output node.
You could put some capacitance on there, or some R+C damping, to provide some filtering and dampening of ripple and flyback, but this will dissipate quite a lot of heat (even if it's handling 10% of full output power, that's hundreds of watts continuous!).
Probably a modest amount of electrolytic capacitance will behave anyway, but be careful here too, because those caps will heat up quickly if you are expecting a lot of ripple current from them. Again, if you're running a ~10% ripple fraction, that's still hundreds of VA, and dozens of watts the capacitors will dissipate -- you'll need many more capacitors than are found in an ATX supply to handle that.
So, all in all, this is an ambitious project -- it is not an impossible one, but it will lead to the discovery of much new arcane knowledge, and it will take a long time to complete (years).
If you're really looking for a learning experience with a shorter payoff, consider something smaller in scale. Or, if you can't push yourself to concentrate on a single project for
years -- I certainly can't, I still have some of that above-referenced project floating around, over a decade later...
Or consider something closer to completion already -- pick up a shitbox inverter welder, and see how it's made. Tweak it for better performance (water cooling for high duty cycle?), or add control features. Once you have more experience, you'll be able to take on a project like this in much less time (maybe months?). Maybe not a welder from scratch, since you'll already have one then, but maybe something related like an induction heater, or play with pulsed high voltage (Tesla coils) for not much practical purpose but much amusement instead.
Good luck!
Tim
Home
Help
Search
Login
Register
