A shaft is a passive component. It carries whatever the machines on either end of it impose, and a fatigue fracture at a keyway says only where the shaft was weakest, not what loaded it. Where material, geometry and finish all check out against specification, the investigation has to move outward: to alignment, to the coupling, to torsional dynamics, and to the operating record. That shift usually changes who the dispute is between, which is why it tends to be contested well before the engineering is.
Misalignment is a bending load
Parallel or angular misalignment between two coupled machines does not merely wear the coupling. It imposes a bending moment on the shaft that reverses once per revolution, which is the classic driver of rotating-bending fatigue. A flexible coupling accommodates misalignment within a stated range and transmits the reaction beyond it, and API 671 governs that duty for special-purpose couplings while stating the alignment assumptions its ratings depend on.
Alignment reports, shim histories and the condition of the coupling elements are the evidence. Where records are absent, coupling wear patterns and the distribution of fatigue origins around the shaft circumference often carry more weight than recollection does.
Torsional vibration is invisible to standard monitoring
Machine-protection systems overwhelmingly measure lateral vibration. Torsional oscillation, a cyclic twist superimposed on mean torque, produces little lateral signature, so a drivetrain can run for years inside every alarm limit while accumulating torsional fatigue damage at a keyway. Reciprocating drivers, variable-frequency drives, generator sets and any system with large inertias separated by compliant shafting are the recurring settings.
The asymmetry matters legally as much as technically. An assertion that the monitoring showed nothing is much weaker once it is established that the monitoring installed could not have shown this.
Resonance and the critical-speed question
Where a forcing frequency, running speed, vane or blade passing, gear mesh, or an electrical order from a drive, coincides with a system natural frequency, response amplitude and cyclic stress rise far above what a steady-state calculation predicts. The questions are whether a torsional or lateral analysis was performed at design, whether a barred speed range was identified, and whether the machine was later operated or re-rated into it.
Retrofits are a common route in. A new driver, a replacement coupling of different stiffness, a longer spacer, or a drive added to a machine that was previously fixed-speed all shift the natural frequencies of a system whose original analysis nobody revisited.
Transients count for more than their duration suggests
Starts, stops, reversals, breaker reclosures, sudden load rejection and jams impose torque excursions well above rated. They are brief and usually unlogged, but their contribution to cumulative fatigue damage is disproportionate, and a shaft sized comfortably for steady-state duty can be marginal against a population of transients nobody counted.
The environment as a load
Corrosion fatigue removes the endurance limit that fatigue design of steel shafting often assumes. In a corrosive environment cracks initiate from pits rather than from geometric features, and there is no stress below which life is indefinite. A shaft in a wet, chlorinated or process-exposed setting cannot be assessed against dry-air data, and pitting at the origin reframes the entire calculation.
What vibration records actually establish
Vibration data is valuable and bounded. ISO 20816 covers measurement and evaluation of machine vibration on non-rotating parts and provides the framework most industrial monitoring follows, and ISO 21940 governs rotor balancing. Together they can show a rising trend, a change in machine condition, or the date at which a fault first became detectable.
What they generally cannot do is quantify stress at the fracture origin. Converting a velocity or displacement reading taken at a bearing housing into a shaft surface stress requires a model, and the assumptions inside that model are where the opinion is attacked.
Duty cycle versus rating
Most shaft disputes eventually reduce to whether the machine was operated as specified. Rated torque, permitted starts per hour, allowable overload duration and permitted misalignment are each stated somewhere in the equipment documentation, and the operating record either matches them or does not. Where a machine was uprated, re-driven or repurposed, the original shaft sizing may never have been revisited at all.
The cascade obscures the sequence
A shaft failure rarely stops at the shaft. Couplings, bearings, seals and gearcases are damaged in the seconds that follow, and secondary damage is routinely mistaken for cause. Establishing sequence, which component shows progressive fatigue and which shows a single overload consistent with the release of a broken drivetrain, is the discipline that keeps a system-level attribution honest.
Where system attribution is challenged
Predictably: that the torsional model used assumed stiffness and inertia values rather than measured ones, that misalignment was inferred from wear rather than recorded, that the transient population was estimated, and that no strain measurement was ever taken on the shaft itself.
Work that states each input, its source, and how far the conclusion moves when that input is varied withstands the challenge. A single computed stress offered without its assumptions does not.
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.