The approach
Sync to the shaft, max out the evidence
A loaded gearbox excites dozens of orders at once. Averaged over time they merge into a broad hill that tells you a gearbox is loud. Referenced to shaft angle, the same measurement separates into discrete lines you can attribute to a component.
One reference per mesh
Every load-carrying gear mesh gets its own coherent order analysis, so a line is read in the frame of the shaft that actually produced it — not in the frame of whichever shaft happened to carry the tacho.
Sidebands carry the location
The mesh order says which stage. The sidebands around it say which wheel on that stage — and that is usually the question worth money.
Evidence, stated as such
Every conclusion comes with the operating range it holds for and the separation quality it rests on. Where the data does not decide, it says so.
Why a gearbox model
A frequency is not a diagnosis
Two wheels in mesh produce one excitation order — and every shaft in the train produces its own. In a real gear train with clutches, planetary stages and several load paths, dozens of these coincide within a few percent of each other.
Which is why the analysis starts from the kinematics: the gear train is modelled, the orders each element can produce are derived from it, and only then is the measurement asked which of those candidates is actually present. That turns pattern spotting into hypothesis testing.
It also makes the result reviewable. A named cause can be traced back through the order it was seen at, the reference shaft it was resolved against, and the operating points where it held.
Model, measurement data and every analysis run are kept in s2m, the analysis platform behind this work. Where it helps a project, clients get their own access to the relevant parts.
Honest limits
What this method does not do
It does not invent information
If the run has no usable shaft-angle reference, or the speed range never crosses the suspected order, no analysis will recover the answer. That is a measurement problem, and saying so early is cheaper than saying it late.
It does not replace a subjective assessment
Whether a tone is a complaint is a question about people. This method tells you where it comes from, not whether it matters to your customer.
It does not settle every twin
Two elements can be kinematically indistinguishable in one operating point. Then the honest output is a ranked pair plus the run that would separate them.