A component that fails before its stated life invites a straightforward inference: the metal was not what it should have been. Sometimes that is right. More often the stated life was a statement about a set of loading assumptions, and the loading the part actually saw was not the loading assumed. Fatigue life is not a property of a material in the way that yield strength is. It belongs to a material, a geometry, a surface condition, an environment and a load spectrum considered together, and a premature-failure claim is a claim about all five.
High-cycle and low-cycle fatigue are different problems
The distinction is not merely one of cycle count. In high-cycle fatigue the bulk material stays elastic and life is governed by stress amplitude; the laboratory basis is force-controlled testing of the kind described in ASTM E466. In low-cycle fatigue each cycle imposes plastic strain, life is governed by strain amplitude, and the relevant practice is strain-controlled testing, as in ASTM E606. Applying a stress-life curve to a component that is actually cycling plastically will overstate its life, sometimes by a wide margin.
The endurance limit is narrower than it is usually quoted
Many ferritic steels show a plateau in the stress-life curve below which laboratory specimens survive very large numbers of cycles, and that plateau is what people mean when they say a part will last indefinitely below a certain stress. Aluminium alloys and many other non-ferrous materials show no such plateau; their curves continue downward, so there is no stress that is safe without reference to a cycle count. Carrying a steel assumption across to a non-ferrous part is a recognisable design error.
Corrosion fatigue removes the plateau entirely
Where a corrosive environment acts at the same time as cyclic stress, the endurance limit effectively disappears and initiation shifts toward pits and other environmentally produced surface features. Corrosion fatigue is also frequency-sensitive in a way that purely mechanical fatigue is not: slower cycling gives the environment more time to act within each cycle, so data generated at high laboratory frequency can be optimistic relative to service. A life prediction is not conservative merely because it used a published curve.
Thermal fatigue counts events, not revolutions
Constrained thermal expansion produces cyclic strain with no external mechanical load at all. Start-ups, shutdowns, quench events and process upsets each impose a cycle, so the relevant count is of operating events over the life of the equipment, which may number in the thousands where a mechanical count would run far higher. That places most thermal fatigue in the low-cycle regime, and it means operating and maintenance logs, rather than run hours, are the primary source for the cycle count.
What a rated life is actually built from
A design life begins with specimen data and is then adjusted for the things specimens do not represent: surface finish, size, geometry, mean stress, temperature and the required reliability. Design codes that contain fatigue rules apply their own margins on both stress and cycles, and those margins exist to cover scatter and uncertainty rather than misuse. The number that emerges is a statement about a population under an assumed spectrum, not a warranty on an individual part under whatever spectrum it encountered.
Scatter is intrinsic, not a sign of a defect
Fatigue data scatter widely. Nominally identical specimens tested at the same stress amplitude routinely differ in life by a large factor, because initiation depends on the worst feature in the loaded volume rather than on average properties. One early failure in a fleet is therefore weak evidence of nonconformance on its own. A cluster of early failures sharing a common origin type is much stronger, which is why field-return data and the condition of unfailed sister parts belong in the analysis.
Counting the cycles that actually occurred
Real service loading is rarely constant amplitude, and comparing it against constant-amplitude data requires reducing an irregular history to equivalent cycles. ASTM E1049 sets out the cycle-counting practices used for this, rainflow counting among them, and the counted spectrum is then combined with a life curve through a cumulative-damage rule. Those rules are approximations that take no account of the order in which loads arrive, which is a known limitation rather than a hidden one.
Where no measured loading history exists it has to be reconstructed from process data, operating logs or strain measurement on a comparable unit in service. That reconstruction is usually the most contestable part of the exercise.
Premature failure is a claim about loading
Once the material has been verified against its certification by composition, hardness and tensile testing, the question becomes whether the service loading resembled the design loading. Resonance and insufficient damping, misalignment, a commissioning overload, transport damage, an unanticipated transient, or simply a duty heavier than the one specified will each produce failure well inside a correctly calculated life without any defect in the metal. Establishing which occurred is a loading investigation as much as a metallurgical one.
What settles it
In practice the answer comes from three sources read together: the physical evidence on the part, the verification of the material against its specification, and the reconstruction of what the component was actually asked to do. Where all three agree, the conclusion is stable. Where the loading history is thin, the honest position is that the metallurgy is consistent with more than one explanation, and saying so is more durable than choosing between them.
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.