When a gear set fails early, four explanations compete: it was rated for less than it was asked to carry, it was not built to the design it was rated against, it was installed so the load never landed where the design assumed, or it was lubricated in a way the contact could not tolerate. Each implicates a different party and a different document trail. The evidence on the teeth constrains the field but rarely settles it. The separation is made by testing each candidate against the record that should exist for it.
What a rating calculation actually claims
ANSI/AGMA 2001 and ISO 6336 are load-capacity methods, not guarantees. Each computes a contact stress for surface durability and a bending stress at the tooth root, compares them against allowable values for the material, and applies factors — application, overload, dynamic, load distribution, size — that convert a nominal torque into a design condition.
The output is a stress margin at an assumed cycle count under an assumed load. When a set fails early, the question is seldom whether the arithmetic was correct but which inputs were chosen, because the factors carry more judgement than the equations.
The service factor is an assumption about someone else's machine
The application or service factor encodes what the driving and driven machines are expected to do — uniform load, moderate shock, heavy shock — and it is often picked from a table early in a project by someone who never saw the installation. A conveyor, a crusher and a reciprocating compressor are not the same duty.
Where a set failed inside its calculated life, the gap between the assumed duty class and the observed one is frequently the largest discrepancy in the file.
Bending or surface: the calculation says where to look
The two halves of the rating fail differently. A root crack propagating across the tooth section from a fatigue origin at the fillet engages the bending strength side, and with it fillet geometry, root finish and residual stress. A pitting flank engages the surface durability side, and with it contact stress and hardness.
A tooth lost in a single ductile overload, with no fatigue precursor, engages neither. That is a question about what produced the torque spike, and the answer is usually outside the gearbox.
Manufacturing: heat treatment and case depth
For a carburised or nitrided gear the rating assumes a case of specified depth and hardness over a core of specified strength. A case too shallow for the applied contact stress lets cracking initiate at the case-core transition rather than where the calculation contemplated. A case too deep or too brittle changes root behaviour.
This is measured, not argued: a metallographic section through flank and root, a microhardness traverse establishing effective case depth, and comparison against the drawing callout. Heat-treat certifications are the paper side of the same question.
Manufacturing: accuracy class and tooth geometry
The drawing calls out an accuracy class — under ISO 1328 or the equivalent AGMA classification — bounding involute profile, lead, pitch and runout deviations. Those tolerances are what permit the load-distribution factor in the rating to be as favourable as assumed.
Metrology on the failed set measures those deviations directly. The complication is that wear also changes profile, so a measurement taken after failure must distinguish a manufacturing deviation from one the failure produced. That is why the undamaged teeth and the original inspection chart both matter.
Installation: the contact pattern is the test
Misalignment, bore or housing error, soft foot and incorrect backlash all produce the same outcome: the load lands on part of the face width instead of all of it. The tooth records this. A wear band concentrated at one end of the face, or at the tip, is an installation signature independent of the gear itself.
Where a set was checked with marking compound at commissioning, that record is directly comparable with the as-worn band. Where it was not, the absence is itself relevant to the standard of care.
Lubrication as a specification, not a habit
The lubricant is part of the design. Viscosity grade, additive package, filtration target and sump temperature are specified for reasons the rating relies on, and field substitutions made for availability or cost are common and rarely documented.
The evidence is easy to obtain: the oil in the box, tested for viscosity and additive state; purchase records for what was bought; oil-analysis history for how it trended; and the manual for what was specified. Where those four disagree, the disagreement is the finding.
Duty cycle: what the machine actually did
Modern drives record far more than anyone reads. Motor current, torque, speed, starts per hour, trip history and process data together reconstruct a load spectrum that can be set against the one the rating assumed. Methods for service life under variable load exist within the ISO 6336 series.
A set run for years at a torque the rating never contemplated is a different case from one that saw a single severe event, and the distinction is usually recoverable from data rather than testimony.
Separating the candidates, and where it is challenged
No single test resolves attribution. The metallurgical section addresses manufacture, the metrology addresses geometry, the contact pattern addresses installation, the oil addresses lubrication, and the recorded duty addresses whether the rating was ever the right rating. Work that survives is work in which each excluded candidate was excluded on stated evidence.
The predictable challenges: that the rating was recalculated with inputs chosen after the fact, that profile deviations on a worn gear were attributed to manufacture, that the assumed duty came from a nameplate rather than a log, and that lubricant condition was inferred from the specification.
This article is general technical orientation, not a failure analysis, an engineering opinion, or advice on any specific matter. Determining the cause of a particular incident requires hands-on examination by a credentialed expert.