Failure is rare. This implies an unusual tolerance stack occurring. Normally the stacks work. On a very rare occasion, an extreme stack is occurring. This is statistical.
This leads to the conclusion that the failure is either due to a) a load sharing problem among the planet gears or b) a vibration problem. Solutions are to improve load sharing if necessary; and reduce vibration by adding damping coatings/rings to the gears. An all new helical planetary (assuming it is spur) would be beneficial. Use damping as an interim solution, and the helical planetary as a long term solution.