Paul & Sabrina's cheap DIY 144v motor controller

[MPaulHolmes]

If you were evaluating a bunch of robots, then I would say it was bionic. A little toooo bionic, ya I really do think. And if it was a new disease that you were helping some doctors study, then it would be a little too bubonic, ya I really do stink, because I'm diseased. hahaha.

Oh ya, it was a dspic30f4011. I might swap it out to have the atmega168 in there, and dump the CAN right into the recycling bin, or you could use DJ Becker's sneaky way of including it still. One thing I like about the dspic is that you can do 4 A/D conversions simultaneously. Throttle, current, temperature, voltage in my case. The specs are that it is basically a regular old control board, but now surface mount, dspic30f4011, with the driver section removed, and voltage monitoring included. There are 2 CAN ports since CAN is usually a network of things chained together, so that it makes it easier to be a link in the chain if one is in, and the other is out.

Putting in the atmega168 would take about 1 hour, so it's no big deal. That would be nice because the code is so well documented and developed.

EDIT: Oh ya, and a 1"x1.5"x10.5" or something heatspreader, which should help conduct heat away OK, and should be about proportional of an increase from 500amp to 1000amp... 0.75"x1.5"x8" to 1"x1.5"x10.5". OK, maybe not proportional. haha.

[DJBecker]

Originally Posted by MPaulHolmes

Hey DJ Becker! I was just thinking that because I'm limited to 3 and 4 ounce copper right now, that I shouldn't push too much current through too short of a board. I was thinking that 100 amps per leg was asking for trouble with the solder joints and the pcb tracks. I thought if I could just make the board a bit longer and use maybe 14 of the irfp4668's instead, which would keep the current down. Man, there are schottky 200v diodes in a TO-264 package that are rated for 250 amps. An almost perfect match to the TO-264 gigamos mosfets that are rated for 230amps. My main reason for doing SR before was because I couldn't find any diodes that matched the current of the mosfets.

I can see that as an issue. We are wiring directly to the power device pins, which puts extra material high up on the leg of the device. That reduces the resistance and provides a heftier thermal path so I haven't been concerned with the lead heating. The bigger packages have heavier leads, but that doesn't help you if the limit is the foil immediately around the solder pad.

On the plus side, these are hand-assembled boards. It would be easy enough to solder a copper washer to the board as a heat spreader. Or a copper wire with a loop formed at the device end. The tricky part would be writing a description that would let first-timers get it right.

For those that don't know, the reason that circuit boards can carry high current with a tiny copper cross-section is that the large surface area keeps things cool. Increasing the foil thickness doesn't let you get rid of any more heat. Doubling the foil thickness only gets you about a 1.4 (root-2) increase in allowable current because the electrical resistance drops.

Normally a device lead will cool the circuit board. In the case where the lead is generating the heat, there is a critical ring around the solder pad where there is a minimum of surface area and thermal conductivity. Increase the thickness of the wire coming out of the device won't change that ring size. The only thing you can do is move the critical ring to a larger diameter with solder or a reinforcement.

Back to the device topic: the larger devices still seem to have an edge. They have a lower on resistance, and are in a bigger package with heavier leads. You won't be able to take advantage of their full power capability, but you wouldn't want to do that anyway. The "on-resistance per inch" is only slightly better, but the thermal coupling to the heatsink is much better. I think the packages are rated at 520 vs 1670 watts.

Another consideration is the gate drive. The bigger devices are almost 50% more difficult to drive. You probably only want to put 30% more current through them. But at some point the number of individual devices becomes more of a problem then providing enough current in the gate drive pulse. I don't know where that point is, but 14 devices on one side might be getting there.

Finally, the device count is a bit of a hassle for hand assembly. It's more clips, more holes to drill, etc. We are using a "live" heatspreader/bus-bar where we can't put a nut on the back, so every hole has to be tapped. Not a huge deal, but every extra device is a little more work.

BTW, I didn't mean for any of this to sound critical. I was really just interested in what pushed your decision to that side of the trade-off. You never really know the best choice until you build both. And then everything changes the next week when a new device is available.

[mrbigh]

Originally Posted by MPaulHolmes

If you were evaluating a bunch of robots, then I would say it was bionic. A little too bionic, ya I really do think. And if it was a new disease that you were helping some doctors study, then it would be a little too bubonic, ya I really do stink, because I'm diseased. hahaha.

Oh ya, it was a dspic30f4011. I might swap it out to have the atmega168 in there, and dump the CAN right into the recycling bin, or you could use DJ Becker's sneaky way of including it still. One thing I like about the dspic is that you can do 4 A/D conversions simultaneously. Throttle, current, temperature, voltage in my case. The specs are that it is basically a regular old control board, but now surface mount, dspic30f4011, with the driver section removed, and voltage monitoring included. There are 2 CAN ports since CAN is usually a network of things chained together, so that it makes it easier to be a link in the chain if one is in, and the other is out.

Putting in the atmega168 would take about 1 hour, so it's no big deal. That would be nice because the code is so well documented and developed.

EDIT: Oh ya, and a 1"x1.5"x10.5" or something heatspreader, which should help conduct heat away OK, and should be about proportional of an increase from 500amp to 1000amp... 0.75"x1.5"x8" to 1"x1.5"x10.5". OK, maybe not proportional. haha.

If you keep the DSpic, all the software will have to be rewritten?

[MPaulHolmes]

Hi DJ! You don't sound critical at all. I actually wasn't trying to make a permanent decision. The super super main reason I was using the irfp4668's was that I have them in the garage, and I know for sure that the driver can do 10 of those in practice, so I figure 2 drivers can do 7 each. Technically I'm over the limit of the total capacitance rating for the mic4451 driver with 10 of the irfp4668s, but it works with 22 Ohm gate resistors, so I was just trying to get something that works. According to the datasheet, I should only drive about 2.7 of those gigamos mosfets with a single mic4451, so I wasn't sure if they would be so forgiving like the mic4451 with the irfp4668. The mosfets are from different companies, and it's hard to know the way they both rate their parts.

[MPaulHolmes]

Well, the code is mostly similar. It's all just C and is already written. Even the CAN code is tested and works. The pwm period triggers all 4 A/D conversions and generates an interrupt, you do a bunch of stuff inside the interrupt, and do the serial communications when not inside the interrupt. It's really pretty simple.

[MPaulHolmes]

Originally Posted by jackbauer

Today i connected the car's tacho sender to the controller via 2k2 resistor and 5v1 zener. Fran kindly put together some code last year for measuring rpm and this was the first test in vehicle. Works very well indeed. Now i just need to implement rpm limiting and motor idle for a future project.
...
edit: thinking of calling it revoliton 0.9 on the grounds the hass 300 can only go to 900amps

That is just so dang awesome I'm going to be sick! I'll get the melexis 25mT linear range current sensor calibrated. The guy that wrote the datasheet on it told me that they use those parts to handle thousands of amps in some applications

[DJBecker]

Originally Posted by MPaulHolmes

Oh ya, it was a dspic30f4011. I might swap it out to have the atmega168 in there, and dump the CAN right into the recycling bin, or you could use DJ Becker's sneaky way of including it still.

My MCP2515 CAN-OBD code "works great for me", and I wrote that section so that it could drop right into the Cougar firmware.

It took a while to get working, but it should be easy enough to understand and extend the message part of the code.

Most of my debugging problems were related to the clock. I was prototyping on a solderless breadboard. I thought I would avoid crystal capacitive loading problems by using a clock from the AVR. Hilarity ensued.

You can avoid all of that, and many hours of debugging, by just using a 16MHz crystal.

(I initially thought that I could clock the chip with the SPI clock, avoiding a real clock until we needed to transmit on the wire. Not even close. The chip seems to need an independent, symmetrical clock to work correctly. The Arduino board dead-ends the CLKOUT pin, so I generated a 8MHz clock with a timer "counting up to zero". But I initialized with 0, so it counted to 0xFFFF the first time, resulting in the CAN controller not being detected but running fine later(!). I built a whole elaborate interactive debug program that timed each operation. It only told me that a reset reliably took about 3.2 usec. First time. Every time. No matter what. Only then did I figure out what was happening.)

[DJBecker]

Originally Posted by MPaulHolmes

That is just so dang awesome I'm going to be sick! I'll get the melexis 25mT linear range current sensor calibrated. The guy that wrote the datasheet on it told me that they use those parts to handle thousands of amps in some applications

Don't these parts require factory programming?

Do they sell a version with general purpose settings that you can buy a single unit of, rather than just get a free sample. Free samples sound great, but they usually take much more time than you expect.

[MPaulHolmes]

Oh by calibrated, I just mean figure out the distance from the bus bar to the current sensor such that -1000 amps to 1000 amps corresponds to the linear range of the chip. It's a little SOIC-8 chip, that just sits there right on the bar. With a pcb track going right below the chip on the same side of the pcb, the range is about +/- 30 amps. The range gets wider if you increase the distance. For example, to get a -1000 to 1000 range on a 7mT linear range device, you needed a distance of 0.75" from the current sensor to the bus bar. That was a bit annoying, so I think the 25mT one will be better suited.

[DJBecker]

Originally Posted by MPaulHolmes

Technically I'm over the limit of the total capacitance rating for the mic4451 driver with 10 of the irfp4668s, but it works with 22 Ohm gate resistors, so I was just trying to get something that works. According to the datasheet, I should only drive about 2.7 of those gigamos mosfets with a single mic4451, so I wasn't sure if they would be so forgiving like the mic4451 with the irfp4668. The mosfets are from different companies, and it's hard to know the way they both rate their parts.

My reading of the MIC4451/TC4451 datasheet doesn't show that as being over the limit. The limit is not the capacitance driven. It is really the instantaneous and average power dissipation in the device. Attaching a big capacitor without a resistor means all of the power is burnt inside the gate driver package. It doesn't matter how good the gate driver is -- all of the heat stays there. Using external resistors quickly shifts most of the heat there, and using a lower resistance gate driver starts paying off.

We bought a 4451 in the TO-220-5 package thinking that we could switch more/better/faster without being concerned about heat. And have it conveniently mounted on the same heatsink as the power devices. But at 20KHz with 10 ohm gate resistors we shouldn't be at the limits of the PDIP package. (OK, it was really that having Vs on the tab was more of a hassle than we wanted to deal with, and there wasn't enough motivation to use the IXYS gate driver that has a different pinout.)