bigmoose
1 MW
A few days ago I commented about being called in to trouble shoot some problems a client was having on a custom 3 phase brushless dc controller that they designed in house. I thought some of you that are contemplating rolling your own might learn from the layout and the fix. They were OK with the pix. I did not design this controller.
First this custom controller was meant for a telepresence mobile sensor platform. That's geek speak for a tracked robot with special sensors on it for measuring special stuff. They have three sizes of these. Think baby bear, moma bear, and papa bear. This is mama bear. Weight about 150 lbs. Powered by 14 AHr 36 Volt LiFePO4 battery.
When mama bear ran into an obstacle at full speed, no problem. When she stubbed her right track on something, just crawling very slow, she blew the 3 phase driver chip for one leg. ALWAYS the same chip on the right driver board. Never on the left track driver.
View attachment 2
This is the controller in it's cute, custom plastic case. Notice that there is only one wee little capacitor sticking it's head up to be chopped off. Where are all his friends I asked? Well the spec sheet says he is a good cap, and doesn't need any friends, said a voice in the lab. We shall see, I said.
View attachment 1
Now this is what the controller looks like with it's lid off. It is built on two boards. The micro board that contains a Cygnal micro, DC-DC converter, and telemetry link via CAN bus, and the FET board, with drivers.
The software load is very neat. You are able to select 3 phase brushless DC, 3 phase Sine Commutation, and Parallel all three legs for Brushed DC motor use. It is a "universal" driver. Nominal battery current is like 20 amps. Lots of data in telemetry, all the typical stuff and odometry so it can autonomously track itself back home.
View attachment 3
Now lets look at what is blowing. Note: This is a previous version, but very similar. I didn't have my camera when I was there with the new version. Note that there ARE NO BUS CAPS.
Now the designer understood grounding and built this very nice 4 layer board around a single star point ground as shown. That circled point is what Ground will be defined as. Note that he beefed up his traces just like we do. Second phase about an inch of wire, and the third phase is no more than 2 inches from Ground.
Now this driver has a digital ground and an analog ground. THIS IS VERY IMPORTANT for what comes next. The digital ground is tied to the star point, and the analog ground is tied to the lower FET's source. Sounds great, right? The source is only 1 or 2 inches from the star point via a 12 AWG (10 AWG on the new one). How far can "ground" move?
The fact that it was the outermost driver ALWAYS blowing, gave it away. I went for coffee and said, I have to think on this. I told a bad joke or two to fill the time up.
Are you getting ideas yet? Things starting to correlate with all my old posts?
Here is the FET side of the first generation board. Note the 8 transorbs they used to try to control spikes. Well they said, "something" is blowing those nice driver chips. Why always the right track. Must be the wiring on the chassis they said.
No, I said. It is the motor. No two are alike. We pulled out the good LCR bridge and measured. Sure enough the right track motor had slightly lower R and L... Imagine that, I said. I went for another coffee.
View attachment 6
I grabbed the schematic and started questioning the design. What PWM period, 8 KHz they said. What deadtime on the H bridge 2.5 uSec. Why so much I asked, because we are blowing FETs. What gate resistor, 5 Ohms. WHAAAAT, I said! WHY! Because we are blowing FETs and think it is shoot through, we have to shut them off quickly to be sure.
Did you ever think about dI/dt effects, I asked. No was the answer, there is 12 AWG ground traces, it can't matter, they said. Really, I said, and asked for a good scope and went for another cup of coffee. I shot the breeze with their secretary for a few minutes, and took the long way back to the lab. This is going to be easy, I said to myself, but we have to make them think a bit to learn!
We got the controller running on the bench with no load on the motor, and we scoped the source lead of the second phase to star point ground. 10 Volt spikes, WOW they said, must not be real. It IS real, I said. You can't switch 20 amps off in 200 nSec and not have "effects." But the motor inductance slows the turn on, I am not talking about turn on, but turn off I said. Then I moved the scope lead to the source of the driver that was blowing... want to guess?
Over 20 volts in the spike!! The driver is rated only for digital ground to be 10 volts offset from analog ground. No wonder the poor silicon die is exploding! I went for my 4th cup of coffee.
What do we do they asked. CAPS I said, and lots of them! Stuff all the good caps you can between the drain of the upper FET to the source of the lower FET on each leg of the bridge. Then we need to raise the gate resistor up to at least 10 ohms, they suggested 15 to 22 ohms to slow down the dI/dt. That should work, but might not be optimum, I said. Then we need a good 15 volt zener between the gate to source of each FET no matter what the driver data sheet says. Why are you driving the FET's off of a 15 volt rail, I asked? I got a "shrug" in return. We need to change that to 12 volts. You are wasting time and charge going that far above the Miller Plateau on the FETs, they are already as on as they are going to be. You are just slowing the turn off.
Here is the beginning of my solution. Inside this cooling pocket are the 6 surface mount FETs. See the red caps? They were the biggest we could find in their lab. Just 0.68 uF and 250 V rated polypropolene I think, maybe polyester. It was what was in the drawer. Just a little under 3 uF total across the rail. That won't make a difference, they said, there is MR super duper 1000 uf electrolytic on the BUS they said.
View attachment 5
Here is the original circuit board removed from the case. Note where the only original BUS capacitor was. Note the layout well. That old design had REAL 20 volt spikes across a 2 inch piece of 12 AWG wire.
View attachment 4
With the modifications specified above, the spikes were reduced to under 0.4 Volts. The controller has been running flawlessly. The motors are so strong they destroyed the gearbox yesterday. There was no controller failure. Story slightly embellished to make a good read.
I write this to show just how important BUS capacitors are to the successful operation of a three phase brushless DC controller. They are as important as the FET selection.
Also DO NOT switch your FETs faster than you need to. Sub 1 uSec dI/dt events make bad things happen. Faster is not better. It is the holestic system approach theat meets success.
First this custom controller was meant for a telepresence mobile sensor platform. That's geek speak for a tracked robot with special sensors on it for measuring special stuff. They have three sizes of these. Think baby bear, moma bear, and papa bear. This is mama bear. Weight about 150 lbs. Powered by 14 AHr 36 Volt LiFePO4 battery.
When mama bear ran into an obstacle at full speed, no problem. When she stubbed her right track on something, just crawling very slow, she blew the 3 phase driver chip for one leg. ALWAYS the same chip on the right driver board. Never on the left track driver.
View attachment 2
This is the controller in it's cute, custom plastic case. Notice that there is only one wee little capacitor sticking it's head up to be chopped off. Where are all his friends I asked? Well the spec sheet says he is a good cap, and doesn't need any friends, said a voice in the lab. We shall see, I said.
View attachment 1
Now this is what the controller looks like with it's lid off. It is built on two boards. The micro board that contains a Cygnal micro, DC-DC converter, and telemetry link via CAN bus, and the FET board, with drivers.
The software load is very neat. You are able to select 3 phase brushless DC, 3 phase Sine Commutation, and Parallel all three legs for Brushed DC motor use. It is a "universal" driver. Nominal battery current is like 20 amps. Lots of data in telemetry, all the typical stuff and odometry so it can autonomously track itself back home.
View attachment 3
Now lets look at what is blowing. Note: This is a previous version, but very similar. I didn't have my camera when I was there with the new version. Note that there ARE NO BUS CAPS.
Now the designer understood grounding and built this very nice 4 layer board around a single star point ground as shown. That circled point is what Ground will be defined as. Note that he beefed up his traces just like we do. Second phase about an inch of wire, and the third phase is no more than 2 inches from Ground.
Now this driver has a digital ground and an analog ground. THIS IS VERY IMPORTANT for what comes next. The digital ground is tied to the star point, and the analog ground is tied to the lower FET's source. Sounds great, right? The source is only 1 or 2 inches from the star point via a 12 AWG (10 AWG on the new one). How far can "ground" move?
The fact that it was the outermost driver ALWAYS blowing, gave it away. I went for coffee and said, I have to think on this. I told a bad joke or two to fill the time up.
Are you getting ideas yet? Things starting to correlate with all my old posts?
Here is the FET side of the first generation board. Note the 8 transorbs they used to try to control spikes. Well they said, "something" is blowing those nice driver chips. Why always the right track. Must be the wiring on the chassis they said.
No, I said. It is the motor. No two are alike. We pulled out the good LCR bridge and measured. Sure enough the right track motor had slightly lower R and L... Imagine that, I said. I went for another coffee.
View attachment 6
I grabbed the schematic and started questioning the design. What PWM period, 8 KHz they said. What deadtime on the H bridge 2.5 uSec. Why so much I asked, because we are blowing FETs. What gate resistor, 5 Ohms. WHAAAAT, I said! WHY! Because we are blowing FETs and think it is shoot through, we have to shut them off quickly to be sure.
Did you ever think about dI/dt effects, I asked. No was the answer, there is 12 AWG ground traces, it can't matter, they said. Really, I said, and asked for a good scope and went for another cup of coffee. I shot the breeze with their secretary for a few minutes, and took the long way back to the lab. This is going to be easy, I said to myself, but we have to make them think a bit to learn!
We got the controller running on the bench with no load on the motor, and we scoped the source lead of the second phase to star point ground. 10 Volt spikes, WOW they said, must not be real. It IS real, I said. You can't switch 20 amps off in 200 nSec and not have "effects." But the motor inductance slows the turn on, I am not talking about turn on, but turn off I said. Then I moved the scope lead to the source of the driver that was blowing... want to guess?
Over 20 volts in the spike!! The driver is rated only for digital ground to be 10 volts offset from analog ground. No wonder the poor silicon die is exploding! I went for my 4th cup of coffee.
What do we do they asked. CAPS I said, and lots of them! Stuff all the good caps you can between the drain of the upper FET to the source of the lower FET on each leg of the bridge. Then we need to raise the gate resistor up to at least 10 ohms, they suggested 15 to 22 ohms to slow down the dI/dt. That should work, but might not be optimum, I said. Then we need a good 15 volt zener between the gate to source of each FET no matter what the driver data sheet says. Why are you driving the FET's off of a 15 volt rail, I asked? I got a "shrug" in return. We need to change that to 12 volts. You are wasting time and charge going that far above the Miller Plateau on the FETs, they are already as on as they are going to be. You are just slowing the turn off.
Here is the beginning of my solution. Inside this cooling pocket are the 6 surface mount FETs. See the red caps? They were the biggest we could find in their lab. Just 0.68 uF and 250 V rated polypropolene I think, maybe polyester. It was what was in the drawer. Just a little under 3 uF total across the rail. That won't make a difference, they said, there is MR super duper 1000 uf electrolytic on the BUS they said.
View attachment 5
Here is the original circuit board removed from the case. Note where the only original BUS capacitor was. Note the layout well. That old design had REAL 20 volt spikes across a 2 inch piece of 12 AWG wire.
View attachment 4
With the modifications specified above, the spikes were reduced to under 0.4 Volts. The controller has been running flawlessly. The motors are so strong they destroyed the gearbox yesterday. There was no controller failure. Story slightly embellished to make a good read.
I write this to show just how important BUS capacitors are to the successful operation of a three phase brushless DC controller. They are as important as the FET selection.
Also DO NOT switch your FETs faster than you need to. Sub 1 uSec dI/dt events make bad things happen. Faster is not better. It is the holestic system approach theat meets success.
