Hmm... lots of small issues, and a couple big ones. Comments from left to right:
I'm assuming this is supposed to run somewhere around 30-50kHz.
The flyback load (D11) is fine, but it needs to be rated for about a watt, BZX84 won't do. Alternately, use the same two-switch configuration as the inverter, and recycle that energy into the 24V supply. Drive could be HIP2101 or the like.
Interesting, K3 has been set (even emphasized with the asymmetry of L5) but, K2 is unrealistic. I think you will find it is also important!
Vge(on) ~ 18V is rather high, and those poor zeners, current limited only by circuit resistances and leakage?
A bigger transistor (Q3, Q4) can be used for faster turn-off, but it's ultimately limited by termination resistance and GDT magnetizing current, which in general don't look to be very much (470 ohms and up to 50mA). Turn-off is critical for low switching loss.
Speaking of which, the transformer needs to be well damped to prevent accidental turn-on; if the dominant capacitive load is M1 (~500pF), then 4mH resonates with that, with an impedance around 3kohms. R27 || R31 is much less than that, so that should behave.
No idea what CMRR of the transformer is going to be like; that would require a more detailed model (including LL and Cp).
Not sure why 900V IGBTs, but they are adequately rated. The conduction and switching losses are marginal, needing most of the rating -- and considering IR's power dissipation test consists of a pool of boiling Freon at 25°C, that's not a practical limit to push. There are faster IGBTs now, and also it may be worth looking at SuperJunction and SiC MOSFETs. (With Si MOSFETs, preferably a resonant topology, as they still have quite high capacitances which make them not as attractive in hard switching applications like this, but with a resonant inductive load, that goes away.)
I like the dV/dt snubber; that will help save some turn-off loss. C15/C16 can even be smaller, I think? They're competing with C_oes which is under 1nF at most voltages, and R35/R37 dissipate ~30W as shown. There's a compromise between switching and snubbing losses, usually with a local minima between the two extremes (none vs. oversized snubber); I don't know how to calculate that minima, but it would be nice to approach it, experimentally if nothing else.
Hmm, why two different 15V 600A diodes? -- MUR1560 is relatively old perhaps, but I don't think there's anything particularly wrong with it. Maybe it has poor recovery at high temperature, the datasheet doesn't show that unfortunately. Package? The HFA is D2PAK, so despite its ratings, it'll never serve as a main rectifier, anywhere near that current rating, but that makes it fine as a snubber. How much current needs to be rectified, anyway? Well...
- LL(pri) = 37.8uH + 1uH. At ~30A peak and 310V, this discharges in 3.7us. Hmm, that's... pretty long, you might want to check that. (Note that the transformer's EMF is forced to zero during this phase, due to the output filter choke. The magnetizing inductance is unable to discharge during this phase, so is delayed by leakage.)
- Lpri (magnetizing inductance) charges up to about 2.8A in 17us, and discharges in the same time; maximum duty cycle is 50%, of course. (This phase takes over when LL discharges to below 2.8A, at which point both discharge effectively in series, and of course Lpri is dominant in this case, so the timings aren't very much in error -- just so long as the two times are added together, in this case limiting duty cycle to more like 40%.)
- Note that it likewise takes as long, for current to ramp up during turn-on, due to LL. So duty cycle is shaved off both the rising and falling edges, effectively reducing overall inverter gain.
Note that you want a few C13s spread around, to close the loop between U3/D18 and U2/D17. Keep this loop short, and keep the snubber loop even shorter; you may find it's beneficial to use a smaller (SMC?) diode for snubbing, so as to keep stray inductance down (which is more or less proportional to lead and component body length). And the snubber caps should be C0G ceramic of course.
Okay, output side now. Your inverter is half wave, it is not a full bridge. Your rectifier however is full wave -- this is a problem. The only current that can be driven negative, is that due to the transformer magnetizing inductance -- which isn't much, as the inductance is quite high. This configuration can be used for flyback, but I wouldn't recommend using that here. (I would recommend full bridge.)
As shown, the FWB just shorts out the transformer magnetizing inductance, ratcheting it up until it can finally exceed the load current and develop voltage; likely this would take ten cycles while the transformer saturates in only two, and then you're driving a saturated transformer in parallel with a heavy load. NFG. So, I'm continuing my analysis assuming the most likely remediation: a half wave forward rectifier.
For this configuration, a double diode is used, one from the winding, one from GND, into the filter choke (or into the output, with the choke in the GND return, same thing). This way, during the off phase, output current is conserved through the GND catch diode. Since duty max is <50%, the on phase needs to deliver at least as much flux as is lost during the off period -- in other words the secondary needs to deliver twice the nominal output (DC) voltage.
Which likewise means the primary needs to draw double the current. So, what you wanted to be a 70V 130A supply, is actually only a 35V 130A supply; or if the turns ratio is adjusted, the primary now needs 60A switches, and a hell of a lot more filter caps to handle the ~30A RMS ripple!
With a full bridge, current is delivered twice as frequently, and at high duty cycle (approaching 100%), using much less ripple current in the primary filter caps, and a lower turns ratio. The savings in capacitors alone makes this worthwhile; transistors are cheap. (Okay, all the drive circuitry getting duplicated, gets a bit annoying, and GDTs aren't particularly cheap. You may want to consider a pair of half bridge bootstrap drivers instead. Also, note that co-pack IGBTs can be used now. The dV/dt snubbers can still be used, or they may simply be omitted.)
Anyway, continuing with the half wave circuit: the challenge is taking a unidirectional secondary, and obtaining both polarities from it. Easiest method again is to just slap an H-bridge on it. If you have two secondaries, you can use one in reverse, but really since it can be completely isolated, you can simply duplicate the whole circuit and get +/-70V 130A output, and just select from one. Rub: each one needs its own filter choke; you can't connect them in parallel, the catch diode from one side shorts out the load current from the other. NFG. If wired in series, the filter choke can be on the center-tap between them, so long as only one side or the other is active at a time, and some means is provided to dissipate the choke's charge during switching -- notice U1/U4 make a half bridge (and do contain co-pack diodes), and switching the inductive load (the filter choke) from one side to the other, means its full load current from one direction, acts to supercharge the opposite direction -- it's a buck-boost supply, from the right perspective. But the half wave rectifier doesn't permit current flow in that direction, so the whole thing explodes in one step...
So, if that voltage is constrained using a big fat TVS or MOV, and as long as it isn't switched often enough to burn that up, it can be okay. Welders typically pulse at, what, ~100Hz tops?
L8 charged to 130A, is 84mJ. Switched at 100Hz, it would deliver 8.4W average. A modest array of TVSs would handle that okay (say, a stack of ten SMCJ's with a thermal pad and heatsink?).
Note that IRGP4068D isn't rated for nearly the current; I'm assuming it's just a placeholder. And 80EBU04 will have to be beefed up.
Note that R3, C7, R12, and probably C6, C9, do essentially nothing. Compared with a fractional-ohm load resistance, these values are negligible. I don't think I would worry about snubbing the half-bridge here, it can switch slowly enough that much care isn't required. Adjust gate resistance and series source inductance accordingly. Some protection may still be needed to deal with EMI (the load is sparking, after all, let alone if HF start is added), which can consist of TVS or MOVs and modest filter C or RC.
Also, note that you're still duplicating things, namely the secondary and rectifier, this way; that seems like a pretty bad deal to me, and I would be more than happy to do an output H-bridge instead. Note that it should be shorting mode commutation: the loop is inductive and current-sourcing, so the safe thing to do is to short it out momentarily, setting voltage to near-zero (low power dissipation) before reversing the load. Peak voltages will depend on turn-off rate and the effective inductive divider between internal filter chokes and welding cables, and a TVS or MOV should still be provided to handle that.
As for controls:
This is a simulation of course, so you might not have much shown here about controls. The big problem is going to be sensing filter choke current bidirectionally, and accounting for that polarity in the control. (I'm assuming average current mode control, which is the easiest, most reliable, and most suitable for this sort of application.) You may want some blanking time as well, during output commutation, to avoid odd voltages or currents.
I highly recommend desat protection, for the primary side IGBTs (and it still works nicely for MOSFETs as well). This is partly a prototyping protection, and partly an operational reliability thing. Turning off the transistors, within a few microseconds of a short circuit condition -- including not just a load short but a short across either leg of the inverter itself -- can save many transistors. It's a cost saving measure, if you like. You can literally shove a screwdriver into the inverter, and... "tick"-- it just stops, no harm no foul. This is easily added, and there are also gate drive ICs which include it as a feature.
Tim
Home
Help
Search
Login
Register
