You're probably getting tired of this but I've come up with a simulation =
that seems like it should reasonably match real-world performance. I = added a=20
0.5 ohm precharge resistor which is bypassed with an NMOS FET when the=20 output exceeds about 200VDC, and it greatly reduces the initial inrush=20 current and high power levels seen previously. I found that it worked = better=20 by eliminating the inductor at the output and relying only on the = leakage=20 inductance, which I simulated with two 20nH inductors in series with the =
primary legs. I tried various ways of dealing with the high voltage=20 transients but two RCD snubbers seemed to do the job. At close to 50% = duty=20 cycle the rectified square wave is almost pure DC and the 40uF output=20 capacitor smoothes the output well enough during the approximate 2 uSec = dead=20 time at 16 kHz.
I had another design which used a MOSFET and some zeners as a 66V high=20 current shunt regulator and it seemed to work well also as a snubber. I=20 could also possibly use some TVS diodes.
For a practical design I might use a relay in place of the precharge = MOSFET=20 switch, but it would need to handle over 100A and a couple of MOSFETs = (or=20 even one) should work. There are lots of inexpensive devices with 2 mOhm = or=20 less and voltage of 30-40V which should be enough. So here are the=20 simulation images and ASC file:
formatting link
formatting link
formatting link
Now I suppose I'll need to build the thing and see how it works. I'll = try to=20 take better inductance readings on the transformer first, and then = re-run=20 the simulation if needed. I'll also probably add some current feedback = and=20 shutdown in case of overload.
That's what the marketing types say. In reality one has to contend with sending all the current across a tiny source pin. In your case I see an average current of almost 50A. So if you absolutly have to use this brute force method use many FETs.
Also, take a look at the dissipation in M3 peak around 5.1msec. This peak is half a millisecond wide and goes to almost a kilowatt. That can greatly exceed the SOA of many devices.
Why can't you do the soft-start with the controller? Just make short pulses in the beginning. There should also be an inductor between D10 and C3, right now you are cramming large current spikes into C3 which causes increases losses in that cap and the diodes.
Take a look at the output current waveform, before it hits the zener/cap/load.
Take a look at the primary current waveform.
All energy transfer occurs in the first 15uSec of the switching phase period. No energy transfer occurs in the later part of the switching phase, as primary current ramps linearly.
You've made a push-pull flyback. Yeesh.
This is where you hope that the simulation reflects real hardware, after you've gone over all the nodes to check for realistic stress levels.
Those 120V drain spikes might be possible, if the fets's withstand them without clipping - if they do clip, thats an unanticipated fet loss and a reduction in the di/dt of the associated circuitry.
The ringing during actual energy transfer may be more lossy than expected, and include unanticipated harmonics, if any clipping or snubber/rectifier current reversal actually results.
re: volts per turn. This is a very loose expression but usually applies as rms/N. You are unlikely to come across a practical example of 60V/turn in your work. RMS and average values of a displayed period are easily extracted by your current simulator software GUI by cntrl/clicking any plotted item.
Higher volts per turn will generate core losses in an exponential relationship. Peak flux, and proximity to saturation will exhibit an inverse relationship. Below 40KHz, most ferrite cores are saturation limited, in commercially available sizes.
Actually I had meant to do this simulation using the tape wound silicon=20 steel toroid that I used in my original design. The 60 V/turn was=20 extrapolated from that and I was trying to see if a transformer with one = or=20 two primary turns of bus bar might be feasible.
So I changed the transformer to my actual measured toroid which I have=20 tested at 16 kHz, and I ran the simulation again, with 10 nH as the = leakage=20 inductance. The results were actually much better. The power into the = load=20 went up to almost 1200 watts and the efficiency was about 99%. Here are = the=20 plots and ASC files:
formatting link
.asc
formatting link
formatting link
ition.png
formatting link
c.png
I do see some problems at the transition, but otherwise things seem to = be=20 OK. But I thought 10nH might be too low for the transformer, so I used=20
100nH. It seemed even better, perhaps. And I slowed down the transition = by=20 increasing the gate capacitor on M3, and it reduced the peak power = spikes to=20 more reasonable levels. I'll probably use two or more MOSFETs for that=20 switch, as well as the push-pull drivers. But I was very surprised to = find=20 that they apparently dissipate only about 3 watts each. Here's an = updated=20 plot over 40 mSec:
formatting link
_40mSec.png
I still need to look more closely at the waveforms at critical points = and=20 see what I may need to do to reduce the high power transients which will =
I admire your consistency of getting the inverter to work. If more people would put that much effort in their work, the world would be a better place! :)
But may I suggest something? I have made such an inverter in the past and suffered as you are with the high wheeling voltage using power fets.
That type of Push Pull inverter works well as long as you keep some current in the primary field between alternations. This saturation current or where ever the current maybe at the time helps to reduce the effects (dump), have you played with over lapping gate signals?
I made one before that simply had delayed trigger on the reference where each ending cycle would trigger the start of the next for the other phase however, I used a PIC chip to control that because I also was doing current monitoring via a ADC to set output demands for soft ramps and the frequency naturally varied. As current demand drops off, the overlap got narrow and as current demands increased, I would increase the overlap time. This made it so that when the secondary side was up where it belong (low PI error), the current damned on the primary was low and there was no need for a over lap to speak of, which makes for low quiescence current.
Have you considered using a "Buck Boost" type of supply? We made one that had to operate from a 36 volt system and generate ~480 volts DC for a AC drive on a battery operating equipment. The AC motor was small,only 1/4 so we didn't need a lot there however, we used high voltage fets,
500V (Yes, very close but they worked) types that handles 18 Amps.
The system at 36 volts needed no more than 10 amps from the battery. We used 4 of those MOSFETS on a widely spread heat sink to keep things cools. It was most likely over kill but it worked find. Btw, we used 4 separate coils, one for each MosFet and a switching diode on each one all tied to gather to the HV cap output side filter.
We also used current mode sensing on that, too. The only trick there is, to make sure your full magnetizing current has a maximum time limit before it switches, incase the feed back loop error is greater than the supply can deliver. Cases like low battery and out of parameter range, otherwise, you'll get a latch up. Ect..
So you're back into straight DC-DC transformer territory.
If all of your inrush is due to the output filter cap, as you propose, then why not stick the switched inrush limiter in series with 'it' ? That way it only has to deal with this specific inrush path, and doesn't have to deal with continuous power train current, just the ripple.
Such a control can be on the low side, tailored by whatever small signal bells and whistles you desire. This isn't an isolated application, after all.
That is a good idea, and I tried the simulation you provided in the = other=20 post. The only problem is that the output needs to connect to a VFD, = which=20 normally contains some hefty capacitors on the DC link. But that just = means=20 using a high-side MOSFET. Actually it may be easier to use a simple = relay=20 which can fairly easily handle 320VDC at 3 to 5 amps. There may be some=20 inrush on the battery side but I really don't need large capacitors. The =
40=20 uF capacitor was chosen for low ESR and high surge currents.
formatting link
603
It is 40 uF 400V with 1.4 mOhm for about $14 and handles 29A RMS:
formatting link
I really think the DC-DC transformer topology is the way to go. The = major=20 problem was the high current, high voltage, and high power transients. I =
still don't know if the steel core toroid will really work well with a =
16=20 kHz square wave, but my earlier testing showed reasonably low core = losses of=20 just about Vin^2/16, so 9W at 12V and estimated 36W at 24V.
But the transformer worked OK at 2kHz with 24V, so I would think it = would=20 work about as well at 16kHz with 1/8 the number of turns, which is where = I=20 got the idea for the single turn primaries. Wouldn't that reduce the = core=20 losses? It would reduce the magnetizing inductance but the current = should=20 stay about the same at the higher frequency. Let's see, it has about=20
3.2uH/sqrt(T) so 1 turn is 3.2uH which is 0.32 ohms at 24 volts is 75A.=20 Oops! But 8 turns is 180uH and at 2kHz that's 2.26 ohms and about 12A at =
24V. Looks like I'd need 2 turns for 12.8uH and 1.28 ohms or 24A, or go = to 3=20 turns and 28.8uH and 2.89 ohms and about 8.3A.
I see that I've not been figuring this correctly. I had always thought = V/t=20 was constant for a given core (yes) and was linear with frequency (no).=20 Actually it must go by the square root of frequency, so the transformer = that=20 had 3 volts/turn at 2 kHz would have 8.5 volts/turn at 16 kHz. My = initial=20 estimate was based on about 0.25 V/t at 60Hz and thus 2.5 V/t at 600Hz. = I=20 should have used 0.8 V/t, and 1.44 V/t at 2kHz. In that case, it would = be 4=20 V/t at 16 kHz and my 8 turns is just about right for 24V. AHA!!! That = makes=20 more sense, so that's my theory and I'm sticking to it!
If You want versatility in choice of load, you'll need to use a smarter converter than unregulated DC transformer.
As previously mentioned, filtering requirements of the unregulated DC transformer re modest - that includes ripple current rating of the capacitors. Ideally, no capacitors are needed.
You've been exposed to other opinions re DC-DC transformer and have spent some time dealing with the issues involved. If this isn't enough to convince you otherwise........
At lower voltages, the siren call of the single turn winding is hard to ignore. You've already run into the mechanical and coupling issues it raises. These are separate from issues related to the use of the topology and are a side-tracking obfuscation.
As turns are reduced incrementally, approaching one (why stop there - why not half or other fractional turns?), it's like moving to another planet, due to the specific problems involved and differing solutions available in each iteration. You can end up having to build the core around the winding, rather than vice-versa - and this is notoriously less flexible. Check out 'Matrix transformer'.
Not for me to make the horse drink. I might have done it, but couldn't find a triac model that reflects real life. Using an FET.driven as shown, produces a nasty dissipation burst in M3, as it goes through its active region, at 2.5ms. Not helped by Cgs.
Depends on how well you model things. I always check a modeled curve trace of a new device model against a real life one. I sometimes check device capacitances, too. I've known models where they were way out.
Useful for proof-of-concept. No substitute for hardware.
--
"For a successful technology, reality must take precedence
over public relations, for nature cannot be fooled."
(Richard Feynman)
- but didn't do the 'work' for him. 'fait accomplis' tends to sell.
The whole simulation business gives me the squitters, when no hardware is being built. It takes effort to make a simulation represent real hardware, even after it's built.
Only really useful to demonstrate basic priciples or to predict iterations of subcircuits - and even then you need to add salt.
Well, in this case I actually have built the hardware, and it worked for = a=20 while but suffered from certain problems such as high voltage spikes and =
extreme current surges. So with the help of some more experienced people =
here, I was able to see where my design had issues, and now after = another=20 round of simulation, I think I am ready to rebuild the hardware and try=20 again.
There are some things that do not lend themselves well to simulation. I = plan=20 to use a PIC, probably the PIC16F684 or PIC16F616, but their PWM modules = are=20 not suited to a push-pull topology except at 50% duty cycle. I may be = able=20 to use a low duty cycle PWM to reduce the high current surges from the=20 output capacitance, and I don't think the small 2HP VFD has a precharge=20 built in. Some, I think, use SCRs as part of the input rectifier bridge, = but=20 that obviously will work only on AC and not where the DC link is being=20 powered directly.
There were certainly clues to the problems of the original circuits. = When I=20 powered them using a 5A current limited supply, the current limit always =
kicked in until the output came up to near the target voltage. But I was =
expecting a surge of up to 50 amps when using the battery. The first=20 circuit, using a 12V battery, seemed to work OK and the 30A circuit = breaker=20 did not usually trip when turned on. But it used the output of an LM324 = to=20 drive the MOSFETs, so the gate drive was "soft" and that probably = reduced=20 some of the high voltage spikes. Also, for 12V I was using a doubler = output=20 configuration, which may or may not have reduced the stress on the=20 components. And I was also using 500 Hz.
I think it was when I added proper gate drivers and went up to a 24V = supply=20 and 2 kHz that one of the MOSFETs failed. But that could have been = because=20 of a software bug where I had forgotten about the change to inverting = gate=20 drivers and the start-up code turned both ON for maybe 500 mSec. They = might=20 have survived the current limited supply, but maybe not. I had a 5,000 = uF=20 capacitor on the output of the supply so the surge currents were = essentially=20 unlimited.
There is still much work that can be done using the simulator. I can try =
running a low duty cycle for start-up and see if there are any high = current=20 or high power transients. I could use several gate drive sources and set = the=20 delays so that they use increasing duty cycle as the output voltage = ramps=20 up. Or I could make a variable duty cycle driver using a ramp function = and a=20 comparator and perhaps use the current through the MOSFETs to cut the = PWM=20 short, cycle by cycle. Even with the PIC, this may be the best approach, = and=20 I might be able to use the PIC's comparator and an interrupt to end the = ON=20 cycles. This will probably work OK at 16 kHz.
The main objective at this point is to ascertain that the steel core = toroid=20 design will work at the power level I want. Once that has been = accomplished=20 I want to be able to "button up" this unit and just use it for my = electric=20 tractor project as I take other measurements to see what power levels = are to=20 be expected under various conditions. For that, I have made a PIC = project=20 that displays various parameters on an LCD display and streams serial = data=20 to a laptop computer for later analysis.
Eventually, I want to come up with a complete solution, with battery = charger=20 and monitor, DC-DC converter, VFD, and an overall vehicle controller and =
dashboard display. I want to use 2 or 3 lead-acid deep-cycle batteries, =
105=20 A-H, 62 lb and $85 each, which should give me a safely usable 1800 W-Hr, = and=20 able to operate at an average of 1-2 HP for up to two hours.
But another option is to use LiPo batteries, and get enough of them for = a=20
300V nominal link voltage without the DC-DC. They are expensive, but I = found=20 some where 30 pieces at $240 would give me 732 W-Hr. Since they are more =
efficient and can be discharged and charged up to 10C or even 30C, they=20 might give me 45 minutes at 1 HP and 5-10 HP peak if needed.
formatting link
OK, enough discussion for now. Gotta get stuff done.
Well, in this case I actually have built the hardware, and it worked for = a=20 while but suffered from certain problems such as high voltage spikes and =
extreme current surges. So with the help of some more experienced people =
here, I was able to see where my design had issues, and now after = another=20 round of simulation, I think I am ready to rebuild the hardware and try=20 again.
There are some things that do not lend themselves well to simulation. I = plan=20 to use a PIC, probably the PIC16F684 or PIC16F616, but their PWM modules = are=20 not suited to a push-pull topology except at 50% duty cycle. I may be = able=20 to use a low duty cycle PWM to reduce the high current surges from the=20 output capacitance, and I don't think the small 2HP VFD has a precharge=20 built in. Some, I think, use SCRs as part of the input rectifier bridge, = but=20 that obviously will work only on AC and not where the DC link is being=20 powered directly.
There were certainly clues to the problems of the original circuits. = When I=20 powered them using a 5A current limited supply, the current limit always =
kicked in until the output came up to near the target voltage. But I was =
expecting a surge of up to 50 amps when using the battery. The first=20 circuit, using a 12V battery, seemed to work OK and the 30A circuit = breaker=20 did not usually trip when turned on. But it used the output of an LM324 = to=20 drive the MOSFETs, so the gate drive was "soft" and that probably = reduced=20 some of the high voltage spikes. Also, for 12V I was using a doubler = output=20 configuration, which may or may not have reduced the stress on the=20 components. And I was also using 500 Hz.
I think it was when I added proper gate drivers and went up to a 24V = supply=20 and 2 kHz that one of the MOSFETs failed. But that could have been = because=20 of a software bug where I had forgotten about the change to inverting = gate=20 drivers and the start-up code turned both ON for maybe 500 mSec. They = might=20 have survived the current limited supply, but maybe not. I had a 5,000 = uF=20 capacitor on the output of the supply so the surge currents were = essentially=20 unlimited.
There is still much work that can be done using the simulator. I can try =
running a low duty cycle for start-up and see if there are any high = current=20 or high power transients. I could use several gate drive sources and set = the=20 delays so that they use increasing duty cycle as the output voltage = ramps=20 up. Or I could make a variable duty cycle driver using a ramp function = and a=20 comparator and perhaps use the current through the MOSFETs to cut the = PWM=20 short, cycle by cycle. Even with the PIC, this may be the best approach, = and=20 I might be able to use the PIC's comparator and an interrupt to end the = ON=20 cycles. This will probably work OK at 16 kHz.
The main objective at this point is to ascertain that the steel core = toroid=20 design will work at the power level I want. Once that has been = accomplished=20 I want to be able to "button up" this unit and just use it for my = electric=20 tractor project as I take other measurements to see what power levels = are to=20 be expected under various conditions. For that, I have made a PIC = project=20 that displays various parameters on an LCD display and streams serial = data=20 to a laptop computer for later analysis.
Eventually, I want to come up with a complete solution, with battery = charger=20 and monitor, DC-DC converter, VFD, and an overall vehicle controller and =
dashboard display. I want to use 2 or 3 lead-acid deep-cycle batteries, =
105=20 A-H, 62 lb and $85 each, which should give me a safely usable 1800 W-Hr, = and=20 able to operate at an average of 1-2 HP for up to two hours.
But another option is to use LiPo batteries, and get enough of them for = a=20
300V nominal link voltage without the DC-DC. They are expensive, but I = found=20 some where 30 pieces at $240 would give me 732 W-Hr. Since they are more =
efficient and can be discharged and charged up to 10C or even 30C, they=20 might give me 45 minutes at 1 HP and 5-10 HP peak if needed.
formatting link
OK, enough discussion for now. Gotta get stuff done.
And that's exactly where the main problems is. Every mundane switcher IC worth its salt has a soft-start feature. Other than a little ceramic cap it requires zero in additional parts.
I don't know what it is with these PICs. They are ok as a uC but why are people using them as switcher chips? It makes no sense. I've seen designs where they were used (against my advice) and very painful compromises resulted. These things cannot do proper current mode control, there is no good dead time control, there is no easy soft-start solution, any loop that was being attempted was bog-slow, and to top it all off the uC cost a lot more than a switcher chip that could have done a much better job.
The one time I considered it (with an AVR) the AVR was for something else. And it was hey, look, I can take this bit away, and that bit, and do this in software, and even the power supply chip could go if i wanted.
But micro controllers can't. That's the problem. Unless you have a nearly infinite horsepower in the thing which will results in high cost, a fat chip and wasteful power consumption.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.