Vendors don't like writing this article, because it admits the product is mortal. But you're sizing a joint for a machine that has to run for years, and "zero backlash" says nothing about year three. Here is what actually happens inside a strain wave gear as it ages, in the order you're likely to meet it.
Failure mode 1: wear, the clock
The dominant, expected, always-running failure mode is wear at the two sliding interfaces: the wave generator bearing working inside the flexspline, and the tooth flanks engaging and disengaging with every input revolution. Both live or die by the grease film between them.
In practice this means service life is largely grease life. As the lubricant ages and sheds from the contact zones, friction climbs, wear accelerates, and the first symptom you'll measure is a slow rise in lost motion and running torque, not a bang. Series in our catalogue carry rated lives of >8,000 to >10,000 hours at rated load; those figures assume the grease stays in its working window.
Three things shorten the clock, roughly in order of violence:
- Heat. Grease life falls fast with temperature. As a rule of thumb, plan on it roughly halving for every 10–15 °C above the design point. A joint run hard at the top of its −20 to 80 °C window ages years faster than the same joint at 40 °C.
- Contamination. Grit in the tooth mesh is a lapping compound. Check the sealing spec against the environment: our units run from IP40 bench-grade to fully-enclosed IP54, with higher ratings available on request.
- Sustained overload. Load above rated torque thins the grease film exactly where the contact pressure is highest.
Failure mode 2: flexspline fatigue, the cliff
The flexspline flexes twice per input revolution, forever. That sounds like a fatigue nightmare, but it's the best-understood part of the design: within rated torque, the alternating stress sits below the material's endurance limit, and the cup effectively doesn't fatigue. This is why wear, not fracture, is the normal end of life.
The cliff appears when sizing is wrong. Fatigue damage scales roughly with the cube of torque, so a duty cycle whose real mean load creeps 30% over what was assumed doesn't cost 30% of the life; it costs more than half. The failure, when it comes, is a crack at the cup's root or tooth rim, and it arrives without much warning.
The defence is arithmetic, not hope: compute the cubic-mean torque of the real motion profile — accelerations included, not just the steady state — and size against rated torque with that. Our sizing walkthrough works a full example.
Failure mode 3: ratcheting, the accident
Every strain wave gear has a momentary-peak-torque limit. Exceed it — a crash, a hard stop from full speed, a seized tool — and the flexspline deforms enough for its teeth to jump the mesh and re-engage somewhere else. That's ratcheting, and it's the one failure that happens in a millisecond.
A drive that has ratcheted is permanently altered: the tooth engagement is disturbed, backlash and accuracy are degraded, and wear accelerates from that day on. It may still turn perfectly smoothly, which is exactly why it's dangerous in a precision machine.
The design defence is making the worst case survivable: check your e-stop deceleration torque and crash scenarios against the momentary peak rating, and where the arithmetic is tight, use current limits in the drive or mechanical stops to cap what the reducer can ever see.
What to watch while it runs
All three modes telegraph through signals you already have:
- Motor current at constant speed and load. A slow climb is friction rising, which is wear talking.
- Housing temperature. Tracks both load reality and grease ageing.
- Backlash / lost motion, re-measured at maintenance intervals. The direct health metric, and the reason a logged as-shipped baseline matters.
None of this needs special instrumentation on an integrated servo joint: the encoder and current data are already on the bus. Trend them, and the clock, the cliff, and the accident all become visible before they become downtime.
Related: Harmonic drive efficiency · Nobody publishes a relubrication interval.