Heheh, but do you know any ring terminals that can handle 100A? This should be capable of 100A. I'll use them until I find something better.

Is that liquid electrical tape denoting polarity? Or did the insulation just kind of melt into the wire braid?

 

Is that liquid electrical tape denoting polarity? Or did the insulation just kind of melt into the wire braid?

The wire insulation melted while soldering.

 

So, I got my MOSFETs today. I bought surplus stock for cheap, as on-resistance doesn't matter too much. They are IRFZ44 devices. 0.012ohm Rds(on) and 60V Vdss. I bought 30 - I thought if you're going to do something you might as well do it properly. If each fet dissipates 30W (which I'm confident I can get them doing if I get the cooling right) I'll be able to load up to 900W, or 75A on 12V, which should be more than enough for the power supplies I deal with.

I'm probably going to mount 25 for now and leave the 5 remaining as spares. I should still be able to get 900W with just 25. If I want to increase output power, I'm going to have to replace all 25, as otherwise the gate drive would be different for all of them and some will end up conducting all of the current and others will conduct little or nothing, leading to localised heating and eventually, failure.

Each group of 5 FETs will have a ~150°C thermal fuse in series with the gate mounted adjacent to one of the group; this will blow if the group gets too hot and will hopefully protect them (as they will blow in sequence if there is a fault.) I'm also considering adding a 30A fast fuse for each group.

 

First three FETs fitted.

The green wire is the gate drive. The resistors are pull-down to source (39k each); they make sure the FETs are off when the tester is switched off or the thermal fuse for the gate blows. The yellow wire is the source, which goes to the shunt resistor. The yellow wire is two AWG18 in parallel. It has to carry a maximum of 20A (the wire rating should give me 38A.)

I initially added a mica pad between the FET case and the metal heatsink. But then I realised my metal bolt would short it out anyway. Therefore I decided to make the heatsink hot (in other words, it's connected to the drain, so up to 55V) - this is okay as long as everything else is insulated.

Still more work to do; I've got to connect the drain wires. Still waiting for fuses. Once I have the drains and fuses wired up, I'll build a simple breadboard and try it at a low 3A current level (but at about 105W, at 35V D-S.) I'll need to see how well this heatsink performs.

The current design won't fit with the amplifier case on. That means it's going to be an open air design... I will probably build an air guide for it though, for the fans. This will have to sit on top of all my other equipment. But that is to be expected as it will get very hot.

 

Those thick wires are a bit nasty looking. If you are using 25 FETs they will run about 4A each for your 100A total?

I would have used small neat wire to each FET source pin, and run the 25 wires to a common star ground pin (which could have a large messy solder blob).

Or run a neat copper strip "busbar" style (capable of 20A) across the source pins of 5 FETs with a solder joint to each FET pin. Then 5 wires to join the 5 busbars to a common star ground. You don't need much to carry 25A when it has an airgap (no insulation) and forced air cooling.

 

Or run a neat copper strip "busbar" style (capable of 20A) across the source pins of 5 FETs with a solder joint to each FET pin. Then 5 wires to join the 5 busbars to a common star ground. You don't need much to carry 25A when it has an airgap (no insulation) and forced air cooling.

That's what I'm doing. Each group of 5 FETs will have these two wires joining each the source and the drain. Then, the wires will meet up with the shunt resistor or fuse, depending on whether or not they are source or drain. No more than four wires will ever leave each FET.

 

Try isopropyl alcohol (rubbing alcohol) to get the thermal paste off your fingers. It has a way of "getting away from you" the moment even the slightest bit hits your finger.

--ETA Your thermal fuse opening the gate connection won't shut the MOSFETs off right away unless you have a pulldown on that bus for the gate line. Just a thought.

 

That's what I'm doing. Each group of 5 FETs will have these two wires joining each the source and the drain. Then, the wires will meet up with the shunt resistor or fuse, depending on whether or not they are source or drain. No more than four wires will ever leave each FET.

Aren't the drains all screwed down onto one massive heatsink? Why bother connecting wires to the drains?

 

Aren't the drains all screwed down onto one massive heatsink? Why bother connecting wires to the drains?

He has a mica insulator between the MOSFET and the heatsink in the photo above, though that is easy enough to fix.

 

The mica insulator can easily push up the junction temperature by about 30 deg C, assuming 1 deg C/W for mica and 30W dissipation per MOSFET. I would prefer mounting directly, if it can be helped.

 

--ETA Your thermal fuse opening the gate connection won't shut the MOSFETs off right away unless you have a pulldown on that bus for the gate line. Just a thought.

That's what the 39k resistor soldered from source-gate is for .

Aren't the drains all screwed down onto one massive heatsink? Why bother connecting wires to the drains?

Ever tried to solder to a steel heatsink? It's not easy... Maybe I could clamp wires to it, though...

He has a mica insulator between the MOSFET and the heatsink in the photo above, though that is easy enough to fix.

The mica insulator can easily push up the junction temperature by about 30 deg C, assuming 1 deg C/W for mica and 30W dissipation per MOSFET. I would prefer mounting directly, if it can be helped.

I did have a mica insulator, but I've since removed the mica insulator, as the bolts I have are not insulated.

 

...
Ever tried to solder to a steel heatsink? It's not easy... Maybe I could clamp wires to it, though...
...

Not to one that big!

Seriously, it would be easy enough to use a threaded bolt and a large circle lug. Or a couple of smaller bolts and lugs if it improves the current path.

 

To make the hooks, I bundled together stripped wire and covered it in a liberal amount of solder. Unfortunately, I couldn't quite finish the other one, because I ran out of solder! Ordered some more...

Edit: Anyone know how much current solder can handle? These connectors are more solder than copper! (I'm using tin lead 63/37.)

Good god Tom..

U could do better if you have boiled solder and dipped the hook in it.

 

U could do better if you have boiled solder and dipped the hook in it.

I don't have nearly enough solder to boil it, but I might do that next time.

 

I don't have nearly enough solder to boil it, but I might do that next time.

Get a solder melting pot

Thow in all your old solder , after a couple years you end up with quite a lot of "used" solder from solder suckers, etc.

Apply flux to your wire, and dip in the molten solder.

Save for the next time you need it. You can also buy bars of solder used in wave soldering machines to get a head start and a cleaner mix. Solder pots and solder bars can be found on eBay cheap if you look.

 

So, I've designed a few parts of the control board.

The most important part is the control loop, which consists of some op-amps. A differential summing amplifier measures and amplifies the current. I'm using LM324s here, because error can be compensated for in software, but any precision quad op-amp in DIP14 could be easily substituted. There are two gain ranges, selected by a relay (who doesn't like the click of a relay when a setting is changed?), one from 0-10A (249.5mV/amp) and another from 0-100A (24.95mV/amp). The output drive is normally connected to ground to keep the devices off; only a high signal from the microcontroller turns on the output drive, through another relay.

Another part is the current calibrator. This generates a constant current which is passed across the main shunt resistor. The current is adjustable, DAC controlled, 2.5V/amp, up to 1A. This allows us to find both terms of error; the constant offset, and the percentage error. An LM317T is used here (a common NPN transistor is a drop in substitute if necessary), but the LM317T has current limit and thermal protection built in. Output is switched by relay. This circuit has been tested in LTspice. It works well, the only limitation is a small amount of oscillation (about 5mAp-p), but as current measurements are averaged, it shouldn't matter too much.

Power is 12V DC @ 1000mA from a switching wall wart (I've got a few.) A buck converter supplies the 5V rail which powers the logic, as well as the calibration current supply.

 

Designed a fan speed controller. There will be two or three of these on the main board - one for each fan.

This is a single layer board with 10/10 rules. Only one wire jumper.

 

One of the main 150A fuses has been added, I'll post a picture later.

Have also designed the differential voltage measurement circuit. This compensates for the ground being offset.

The power supply for this device is a switch mode wall wart 12V power supply - current draw at most 1A.

However being a switcher, it's quite noisy, so I opted to use a 8V regulator. There's a negative rail too, provided by a TC7662 (inverts +12V to -12V) which is then regulated to -8V. The negative rail just allows the offset to be more easily compensated for. But I may just end up using a precision op-amp so I don't need to do this.

0.1% resistors are expensive. And I don't actually need 0.1%; I need *matched* resistors. So I'll procure some low-tempco 1% resistors 5k and 100k, and match them to one particular value.

The op-amp in this design is operated with a gain <1, so I'll need to double check stability.

I'm considering adding a true RMS converter for measuring ripple voltage, for characterising power supplies. But there will also be an ac-coupled voltage output for a scope.

 

...
0.1% resistors are expensive. And I don't actually need 0.1%; I need *matched* resistors. So I'll procure some low-tempco 1% resistors 5k and 100k, and match them to one particular value.
...

I agree, trying to use 0.1% resistors is silly, and things wile solder, wires, traces etc will influence the reading anyway. I would suggest you use sense resistors that will be most practical, and calibrate in software by comparing against good test equipment.

 

The only problem is this unit will get *hot*. Probably 80°C in the case in parts. (Which is why for the design, especially the electrolytic capacitors, I've chosen high reliability, high temperature, low ESR capacitors, and I've got multiple ones in parallel.) Those resistors will drift as a result. It might be possible to compensate for this, by compensating for temperature changes (if we know component temperature coefficient we can do this.) To be decided.

Meanwhile, the negative fuse has been added, and soldered in. I've not bothered with a connector because I think anything which blows the fuse is likely to require more than replacing the fuse. The fuse is held in with cable ties. Some more fets have also been added, but not soldered in yet. I'm going to order gate fuses soon but it's Boxing Day.

Note soldering is not yet finished on the fuse; I ran out of solder!