Three planet gears in a planetary reducer. The catalog said the unit handles 120 Nm continuous. The application was a 3 kW servo on a rotary index table, peak torque 90 Nm. The gearbox failed at the planet bearings after eleven months. The customer ran the numbers again. 90 Nm against 120 Nm rated. They were at 75% of rating. It shouldn’t have failed. The rating assumes something the real gearbox wasn’t doing — equal load sharing between the three planets.
A planetary stage carries the torque across however many planets are in the design. Three planets means each should carry one third. But they don’t, not automatically. The planets sit on pins pressed into the planet carrier. The pins are located by machining tolerances. If the carrier holes are 20 microns off, one planet carries more than its share. The gearbox manufacturer compensates by letting the sun gear float. The sun gear centers itself between the planets and equalizes the load. That only works if the sun gear can actually move.
Where the load actually goes
The sun gear in a standard planetary stage is splined to the input shaft. The spline lets it float a little — the clearance between the sun gear bore and the spline teeth allows radial movement. With that float, the three planets share load within a few percent. Without it, or if the float is small, the stiffest planet path takes the load first. The planet pin holes in the carrier are ground to a tolerance of maybe 15 microns. That sounds tight, but the planet bore clearance is only a few microns. The machining error is bigger than the bearing clearance. One planet binds, the other two take the load, and the planet bearing on the loaded side runs hot until it fails.
The failure on the index table gearbox: the planet carrier had a 25 micron position error on one pin. That planet carried roughly 55% of the load instead of 33%. At peak torque of 90 Nm, that planet saw the equivalent of 165 Nm worth of planet load. The bearing overloaded. The wear pattern on the failed bearing showed the classic single-side loading. The sun gear float wasn’t doing its job because the spline had a tight fit from the factory.
The real selection rule
Planetary gearbox ratings assume perfect load sharing. Real gearboxes share between 70% and 95% of perfect, depending on the manufacturer’s machining and the sun gear float design. The practical rule that has held up: take the catalog continuous rating, multiply by 0.7, and use that as your continuous torque. For shock loads, use 0.5. This sounds like a huge derate. It isn’t. The gearbox manufacturers that design floating suns properly, with hardened and ground planet pins and lapped carrier holes, do achieve close to perfect sharing. The cheap ones don’t. You can’t tell from the catalog which is which, so the derate covers both.
On the index table, the replacement was one size up — 150 Nm catalog rating. The peak load of 90 Nm is now 60% of the derated continuous (105 Nm). The gearbox has run for three years without issue. The extra cost was about $180. The failed gearbox cost $420 plus a day of downtime. The derate paid for itself in one failure.
Checking load sharing on an installed gearbox
There’s a quick check that doesn’t need special equipment. Run the gearbox at full load for an hour, then stop and measure the planet carrier temperature through the housing with an infrared thermometer. If one planet position is more than 10°C hotter than the others, that planet is carrying more than its share. The gearbox will fail. This check finds the problem early, before the bearing fails. On gearboxes that pass the check, the temperatures read within 3°C of each other.
Temperature checking is worth doing on every new planetary installation, and again after a year of service. Gearboxes that start with even sharing can drift out of it as bearings wear unevenly. The annual check is a five-minute job with a $30 IR thermometer.
Peak torque and the duty cycle question
The other misread number is peak torque. The catalog lists a peak rating, often 2x continuous, with a note like “for less than 5% duty.” The 90 Nm peak on the index table happened 12 times per minute, each for 0.3 seconds — that’s 6% of the time. Over the catalog limit. Even with perfect load sharing, the gearbox was borderline on thermal. The derated continuous (84 Nm) was below the 90 Nm peak, which means the gearbox was running above its derated capacity for 6% of every minute. The bearing was getting hot even on the healthy planets. The one over-loaded planet pushed it over.
For indexing applications with frequent peaks, the selection should be based on the RMS torque over the cycle, not the peak. The RMS for the index table worked out to about 65 Nm — fine for a 120 Nm gearbox derated to 84 Nm. But the planet sharing error pushed one planet past its limit. The derate rule catches this. The RMS rule catches the thermal side. Use both.
Planetary ratings assume perfect load sharing. Real gearboxes don’t share perfectly, and the error shows up as one hot planet. Derate the catalog rating by 0.7 for continuous duty, 0.5 for shock. Temperature-check the planet positions on new installs. The index table failure wasn’t undersizing — it was one planet doing 55% of the work.
