Dummy Load II

[jocanon]

I have a question that has been bugging me for a while now. I would think that since each FET circuit is basically independent of the other that once I blew out some and disconnected them, I would still turn the dial on my pot to the same place and just get proportionately less current running through the system. For example, if I had to turn it 3/4 on when all 10 FETs were working to get 47 amps and I would think that now that I only have 8 FETs if I turned it 3/4 of the way I would get 8/10ths of 47 amps, or 37.6, but it doesn't seem to be working that way. I only have to turn it barely on and it starts to go up fast. If I were to turn it 3/4 of the way on now, it would be above 47 amps, so I don't do that. Is this behavior to be expected, or is there something wrong?

 

[ronv]

Sounds like there is something wrong. It should work as you said. It almost sounds like you are missing the feedback from one of the .13 ohm resistors to the - side of one or more of the op amps. Or something is wrong with the voltage on the center terminal of the pot that it jumps up when you turn it.

 

[jocanon]

arrg..if I just wouldn't have shorted the op-amp...oh well, no use crying over spilt milk right? It seems that one little slip up has cascaded into a whole slew of problems lol! I don't know how much time I really want to spend trouble shooting this dummy load since time is such a premium right now...maybe I should just quickly get dummy load II done and let this one RIP.

 

[()blivion]

I would think that since each FET circuit is basically independent of the other that once I blew out some and disconnected them, I would still turn the dial on my pot to the same place and just get proportionately less current running through the system.

If I'm remembering the schematic correctly, then physically yes, that is exactly how the FET's and control should work.

However, how exactly are we/you measuring the current? If I remember right we are taking the voltage from one current sense resistor, scaling it, then sending it to a voltage meter as indicated current in volts. Here's the thing... with such a design ronv would have had to take into account that each resistor is some fraction of the whole current, and he would have had to hard wire that into the scaling circuit accordingly. In other words, 1 amp on one channel, indicates 10 amps total current, but that is because we are expecting 10 channels. If we lose a channel, then the true current really does go down, but the indicated current would remain the same.

I think similar effects could possibly make it seem that the same knob position is using more current. But that would be very odd still.

 

[jocanon]

()blivion, that is a good point, but since the FETs bit the dust I have been using a seperate watt meter that I had laying around for measuring current in RC plane systems, so it is correct, or at least within a margin of error.

 

[()blivion]

I have been using a seperate watt meter

Ah, I was wondering about that. OK then, scratch that off the list. Something is definitively off with the thing. Possible one of the other FET's is damaged and is acting odd. Or as ronv said, one of the controls is messed up.

 

[Toe Cutter]

A few ideas.

Hi. I am a bit late to this thread, but found it interesting because I recently prototyped and debugged a current sink like the one being discussed, (but much smaller). You really do want an oscilloscope to debug these things. They don't always behave as simple theory would suggest. Simulators probably won't simulate the problems accurately. The problems are when the circuit oscillates. Sometimes the circuit will be stable, sometimes it oscillates, it seems to depend on the control setting. Any stability issues or unusual readings on your multimeter are probably the result of oscillations.

The first oscillation is in the op-amp control loop. This oscillation might be in to 10-100 khz range. I think this is one of the stability issues that was mentioned. It looked like the remedies offered were not quite right. In my circuit, I added a resistor and capacitor in the feedback. This turns the opamp into a classic integrator circuit, which is also a low-pass filter. I used R=1k and C=0.01uF. Gain will be reduced at high frequencies, but DC gain will still be very high, which preserves regulation accuracy. I'll attach a quick sketch.

The second oscillation that might occur is when MOSFETs are connected in parallel. This is oscillation between the MOSFETs. It tends to produce RF oscillations, maybe 50-100 mhz. The solution to this problem is to add components to each gate lead to dampen the oscillation. A series gate resistor (10-100 ohms) with a ferrite bead slipped over the gate lead works well. The ferrite bead looks transparent at low frequencies, but appears as a resistance at high frequencies.

As for the power handling of the TO-220 transistors. They were rated at, what? 300 watts. They can only do that max power _IF_ the transistor case is maintained at 25*C. The thermal resistances have to be taken into account and the transistor derated. Putting a thermocouple on the front plastic face of the transistor can give a pretty good measure of the junction temperature.

One way to find the upper limit is destructive testing. Disconnect all but one transistor. Then set the control so you get 50 watts dissipation in the mosfet (Vds*Id). Wait several minutes until the temperature stabilizes. Does it fail? No? Try 60 watts, 70 w, 80 w, etc. Until it fails. Then you can figure your absolute max dissipation and how much margin you have.

It seems like you have very effective water cooling. I'd be interested to hear what the failure point is and how that compares to the mosfet datasheet.

 

[ronv]

Hi TC, I think your right something is oscillating. So here's the latest.

I added a voltage regulator so the 12 volts will come up after the 24. This makes the reference more stable and it would let us use Mr. RB's idea if we like. The +12 is really +9.5 now - I forgot to relabel everything.

Raised current sense to .2 ohm.

Rescaled divider for 15 Fets - 8 op amps.

Lowered the temperature cut off to 50C.

Added the current shunt.

Changed to Ducey's op amps for less offset and better phase margin.

Changed FETs for slightly better ones.

Added roll-off to the op amp and to the FET gate to hopefully improve margin due to layout etc.

I added a bead, but have no idea which one to specify. I need to read up.

I didn't show the fuses, but we can add them. I'm a little concerned that the path from the FET source to ground is going to have a lot of inductance in it now. We may want some clamp diodes for when the fuse blows. I also want to model it.

 

[Toe Cutter]

Looking at your most recent post, I'm now not sure how the opamps and FETs are configured. You say 15 Fets and 8 (dual?) opamps. Does this mean that one opamp drives one Fet? From looking at earlier schematics I thought 1 opamp was driving 10 Fets. In the later case, you are likely to have a parasitic oscillation between Fets.

If each opamp is driving 1 Fet, then you might be able to omit the ferrite beads. But there is still some possibility that there is oscillation between Fets. He would need a scope to be sure. In your schematic, I would swap the position of the gate resistor and the bead. Slip a bare ferrite bead on the gate lead of the transistor.

Thinking about it more, a ferrite bead may or may not be necessary. Just adding enough gate resistance might do it. But without a scope there's no way to know what is required. This isn't a high speed switching circuit. The time to charge the gate isn't so important. And besides, the opamp current limits at 10 mA.


C2 and C3 can be omitted. They just cause more work for the opamp and might decrease stability.

Your idea about the fuses and resistor inductance is interesting. What is the inductance and how fast does the fuse interrupt? You might be able to estimate peak voltage from V = L*di/dt. But if the fuse blows, something else already died and it might not matter.

 

[()blivion]

Edit: I was thinking "integrator"... but actually it is just a low pass filter. Disregard.

 

[ronv]

Yes, 15 FETs and 8 dual op amps. Since they are running in the linear region they can't just be put in parallel because of the potential difference in threshold voltage, so each needs its own closed loop then they share the current.
Yes, I think the ferrite should be last in line. They are cheap so I think they are a good idea since we are going for the gusto before Jeremy needs to buy a scope.
I added the 2nd rolloff for a couple of reasons. One was if the 24 comes on or off quickly the drain to source capacitance couples it to the gate. Since in the come on situation the 12 volts isn't up yet I was worried about the op amp latching up. If the big supply (DUT) crowbars the negative transient may go past the gate source breakdown voltage. It also adds a second rolloff that brings the bandwidth down to 100Khz from 300Khz. Below:

 

[fernando_g]

. You really do want an oscilloscope to debug these things. They don't always behave as simple theory would suggest. Simulators probably won't simulate the problems accurately. The problems are when the circuit oscillates. Sometimes the circuit will be stable, sometimes it oscillates, it seems to depend on the control setting. Any stability issues or unusual readings on your multimeter are probably the result of oscillations.

.

I'm not happy to see anyone struggle to prove my point, I don't want to be an obnoxious poster, and neither I'm relishing on an "I told you so" attitude ....... But I suggested a scope to troubleshoot weird behavior back in thread #3.

 

[()blivion]

I'm not happy to see anyone struggle to prove my point, I don't want to be an obnoxious poster, and neither I'm relishing on an "I told you so" attitude ....... But I suggested a scope to troubleshoot weird behavior back in thread #3.

Yes... yes you did make that suggestion. And yes... I agree... an O-Scope would be the preferred method for doing this kind of work. And YES, without one we are likely going to be playing a long game of trial and error in the dark. Not that we already haven't.

However, the problem with getting a scope is that this project, like many others, has some unfortunate but perfectly reasonable constraints. Namely the investment and the fact that the OP is not interested in becoming a regular electrical engineer. Beside that point, there is thus far no reason that we can't come up with a working design without needing to resort to scoping anything. Unwanted oscillation is a well understood problem that has many text book fixes. Just knowing that there are likely problems with such things and where they are likely to be is more than enough to conclude that we should make this a part of the design calculations. Putting the whole problem another way, would you go and spend $200+ on a tool to diagnose a $100 device? Especially considering you expect to only ever have to do it the task the one time? Because... from where I'm sitting... that's about what you're suggesting. He would most likely end up shelving the scope after this project is finished and never dig it out again. Finally, consider that the solution will probably be found for free by simple examining the problem. No further investment required.

Unfortunately, one can't always use the best tool for the job, sometimes you only get to use the best tool for the dollar. And sometimes that's OK.

 

[fernando_g]

I'll repeat myself...my post meant no offense.

Sorry for the misunderstanding.

 

[()blivion]

I didn't take what you said to be of any offence, there is nothing to apologize for. Maybe I'm the one that came off as abrasive? Blanket pardons all around?

 

[jocanon]

I'm not offended, I understand both sides...in fact if I end up needing a scope I will break down and get one, but like ()bilivion said, I would only be using it for very limited purposes. I do have a natural curiosity though and like to learn new things, so that would be a good side benefit if I were to get one. One thought I have is, I would like to get rid of or minimize the oscillation, BUT even if it does oscillate some, is it really a deal breaker for the load. I mean, it was working fine for my purposes, oscillations and all, before I dumb thumbed it and broke it, ever since then parts have been failing (I have an eerie suspicion it all relates to me shorting it out though). So I guess I am actually asking a question of you all here, and that question is, other than the oscillation being an annoyance, as long is it does not oscillate too wildly, can I still use it for my purposes of testing power supplies? I don't see why not, but someone please enlighten me if there is a problem (i.e. harm to the PSU, or it might just blow up on me at some point...I don't know). Seems to me it would still work as is, but we are just trying to improve it, am I right?

 

[Mr RB]

... Does this mean that one opamp drives one Fet? From looking at earlier schematics I thought 1 opamp was driving 10 Fets. In the later case, you are likely to have a parasitic oscillation between Fets.
...

The FETs can't oscillate if the large RC integrator is added on each FET gate as we have been discussing. That gives the greatest level of safety regarding FET oscillation, and also guarantees a slow ramping of load current up or down when switched etc.

 

[Toe Cutter]

I'm not offended, I understand both sides...in fact if I end up needing a scope I will break down and get one, but like ()bilivion said, I would only be using it for very limited purposes. I do have a natural curiosity though and like to learn new things, so that would be a good side benefit if I were to get one. One thought I have is, I would like to get rid of or minimize the oscillation, BUT even if it does oscillate some, is it really a deal breaker for the load. I mean, it was working fine for my purposes, oscillations and all, before I dumb thumbed it and broke it, ever since then parts have been failing (I have an eerie suspicion it all relates to me shorting it out though). So I guess I am actually asking a question of you all here, and that question is, other than the oscillation being an annoyance, as long is it does not oscillate too wildly, can I still use it for my purposes of testing power supplies? I don't see why not, but someone please enlighten me if there is a problem (i.e. harm to the PSU, or it might just blow up on me at some point...I don't know). Seems to me it would still work as is, but we are just trying to improve it, am I right?

You might be able to get away with the circuit oscillating. It might work good enough that way. The downside might be lower reliability and less accurate measurements. The FETs will be operating differently, for example: 50% on-time at double the expected current, then %50 off-time during one cycle, which averages out to your set current level. Also, you will be testing the power supply in a different way: a rapidly changing load, instead of a constant load. I don't know what that might do to your PS under test. More ripple current on the filter capacitors? I just seems to me that a testing tool should operate as expected and do what it is designed to do.

Re: Fernando's post. I'm not trying to go out of my way to prove this point about the scope. But just share my experience building circuits like the PS tester. I thought I had a good circuit on paper. Then when I had it built and running, the multimeter connected, I found wasn't working as expected. Sorry to rehash old posts, I may not have read them all in detail. No offense taken or anything.

Anyway. If Jocanon isn't ready to invest in a scope, I think he can still get it running well.

 

[ronv]

The FETs can't oscillate if the large RC integrator is added on each FET gate as we have been discussing. That gives the greatest level of safety regarding FET oscillation, and also guarantees a slow ramping of load current up or down when switched etc.

No disagreement. Do you see any more holes like we had when the +12 was on before the 24?

 

[ronv]

Here is another meter.

**broken link removed**

It will run off the 12 or 24 volt DUT. Also needs a shunt, but might make a nice package.