Hey Bill,
Thank you for the elaborate answer.
I will do my best to clarify the issue you have raised.
First, this is not a mower but a robot for cleaning solar panels but the idea may be the same.
There is a sequence for charging and stopping the battery charge - it is controlled by the MCU.
The charger is not working while we operate the robot but it is still part of the board.
There are also mechanisms to start and stop the motors, control the speed, identify over-current situations, stalling and to monitor the battery charge state.

The problem that I describe can happen so fast that the MCU cannot respond to that and the battery will shut down.
You asked what has changed, mainly the environment and the length of the operation.

When a shut down occurs, the motor will coast, but if it was stalled or even working in full load, it will not coast but rather stop.
In any case it will act as a generator.

I am looking for a way to respond to that, I was thinking of adding a resistor + diode on the (+) (-) of the board so the current will have a path to go through.

If you have other or better ideas I will be happy to hear them.

Not unless you have a DC-fed motor with internal commutation.
Do you have a link to the motor spec/datasheet?

Edit:
What provides the input to VIN_SUPPLY? It might be worth adding extra capacitance to that pin. At the moment the total for the various parts in the pdf which rely on that voltage seems to be about 30uF. I would have expected at least a magnitude greater.

Thank Alec,
I actually have a 1000uF bulk capacitor between Vin_supply and GND.
See page 3 (Rotor motor) on the bottom left side.

 

See page 3 (Rotor motor) on the bottom left side.

Ah. I missed that.

 

Thanks Bill,
The battery can go to "protection mode" for many reasons, such as over current and temperature so adding code may not help.
I would like to solve this with hardware only.
That is why I was thinking on a snubbed.
What do you think?

It appears that the damage happens because the power is cut off completely while the motor is still turning fast enough to generate power, but the control system is no longer powered enough to correctly control the motor drive stages. At least that is what it appears to be. If that is what is happening then the solution would be a scheme to keep the controls section powered and have it do the correct motor stop sequence. That may not be a simple task.
A better solution would be for the battery protection system to provide a signal prior to the switch-off that it was going to switch off. That signal could then give the drive system a motor stop sequence. This option may not be possible.
Another option will be to change the driver circuit so that at the instant of power supply loss all of the drive circuit transistors switch off. My guess is that when there is an instant power loss that somehow the drive transistors are switched on. That would account for the circuit burning. This option will require an understanding of exactly how the drive stage operates. That may be difficult.

 

I tried to simulate what happens when the power supply cuts off while the motor is still rotating. Here's the sim schematic for the situation with two half-bridges where initially phase A is high and phase B is low :

I've assumed some values which may or may not be anywhere near reality, namely voltages and BLDC coil parameters.
20V supply source V1 and the gate drives are assumed to decay at the same instant (0.5s) over a 10ms period. The BLDC coil is assumed to be generating 15V initially, dropping to zero over 0.5s as the motor slows.
Simulation shows that if there is no current path back through V1 once it cuts off, then the properties of decoupling cap C2 are critical as to whether or not the supply rail voltage spikes above 20V due to the combined effect of the coil's generated voltage and it's inductive kick-back. V(supply) goes above 20V if (a) C2=1000u and its ESR is > ~7Ω, or (b) C2 < ~220u

 

If the motor is observed " and a brutal stop of the motors. "then we can guess that somehow a very low resistance path is created so that a lot of current, generated by the inertia of the motors still spinning, is flowing through whatever components are being damaged. This is a reasonable guess based on the understanding that there is no other battery sourcing power to any portion of the circuit.

Is it possible to share the section of the circuit that is damaged when the failure happens? It is entirely possible that some of the very wise participants will immediately be able to discover exactly what is happening and suggest a solution.

 

Thanks, In my design I use MOSFETs so the diode is already there.

I was also thinking along the lines of Dick_C (#2), as a general rule, if hi back EMF is predicted, then higher value external HS switching diodes are recommended and used.
Even though the Mosfet's have them internal.

 

The unanswered question is what burned up in the failures? I am unable to read other people's minds and know which parts were damaged, and that portion of the circuit where the damaging current flowed, or the damaging voltages were.
Imagining what others are seeing, accurately, is not part of my skill set.

 

Thank you everyone,
The part that is getting damaged is the MPPT part.
Please ignore D21 - as it is not present and only harms the circuit.
Max - you are correct that there is a very low resistance path that is created.
We see many times that the capacitors are burned - mainly C66 and C68 and as a result go to short - in this case more parts on the MPPT get damaged.
Sometimes - the MPPT area get damaged even when the capacitors keep working.