I _think_ i finally got my head around most of this...
Initial failures included -High side FET, FET driver and current sense
resistor in ground leg.
Tested with more capable FETs, FETs survived, driver and sense resistor
failed.
Much deliberation, added additional freewheeling diodes from phase to
ground (bypassing the sense resistor, that wasn't thought out and
deliberate, just the easy point to hook to)
Next failure took out the sense resistor but the FETs and and driver both
survived.
Datasheet says sense resistor is rated for 20A continuous, 85A peak....

After working through every register in the MCPWM module, I note that we've
initialized it in independent mode instead of complementary mode... FET
drivers do not include cross conduction prevention...

Revert to previous hardware and modify the init to complementary mode and I
can no longer replicate the failure.

It seems that once the sense resistor goes the failure then cascades
through the FET driver and possible the high side FET, since there's no
longer a path for the back-emf.

So, even though the sector table and the commutation logic say that it's
"not possible" to have a cross conduction, the evidence suggests that it is
happening under some unusual fault scenarios.
Still trying to figure out the underlying "why" and "how". Leading thought
is that some noise on the hall sensors is causing it to incorrectly
commutate (i.e. sector 1 to sector 3, skipping sector 2) Such that one
transistor hasn't turned off before the other comes on.

Still doesn't make sense that every failure to date has been controlled by
PWM channel 1.
Still doesn't make sense that the brushed motor version has similar
failures (no commutation, no motor reversals during that part of the
operation).


-Denny
--=20
http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist
.