A component passes static strength with margin, looks clean in FEA, and still fails in service after a few months. For many engineering teams, that is the moment fatigue moves from a checkbox to a design driver. This guide to fatigue life prediction is written for engineers who need defensible results, not just colorful contour plots.
Fatigue analysis sits at the intersection of loading realism, material behavior, stress quality, and modeling discipline. If any one of those is weak, the life estimate can be off by orders of magnitude. That is why fatigue life prediction is rarely about a single solver setting. It is about building a workflow that connects measured or credible loads, appropriate fatigue theory, and validated finite element results.
What fatigue life prediction is actually estimating
At its core, fatigue life prediction estimates how many cycles a part can withstand before crack initiation or failure under repeated loading. The detail that matters is which part of fatigue life you are trying to predict. In most CAE workflows, analysts are estimating initiation life based on local stress or strain response. That is not the same as total life after a crack has formed and grown.
For high-cycle applications where elastic behavior dominates, stress-life methods are often appropriate. For low-cycle applications with meaningful plasticity, strain-life methods usually provide a better basis. Crack growth methods become important when fracture-critical structures, inspection intervals, or damage tolerance requirements govern the program. Choosing the wrong framework at the start can make the rest of the analysis look precise while being fundamentally misaligned with the physics.
A guide to fatigue life prediction starts with loading
Most fatigue errors begin before meshing. The first question is not which fatigue model to use. It is whether the loading history reflects how the part really operates.
Constant-amplitude loading is useful for screening and early design studies, but many field failures come from variable-amplitude service histories, transient events, startup and shutdown cycles, vibration, or mixed loading states that were simplified too aggressively. If the duty cycle includes overloads, dwell periods, temperature shifts, or out-of-phase loading, those details can influence mean stress effects, damage accumulation, and local hot spots.
When test data exists, engineers should resist the temptation to reduce everything to a single peak load. A realistic spectrum, even if condensed, is often more valuable than a neat but unrepresentative static equivalent. If measured loads are not available, the loading assumptions need to be documented with the same rigor as material properties and boundary conditions.
Selecting the right fatigue method
There is no universal fatigue model that works best in every situation. The right method depends on material behavior, stress state, geometry, and the quality of available data.
Stress-life approach
The stress-life or S-N method is common for welded structures, machinery components, and many high-cycle applications. It is efficient and widely supported in commercial fatigue tools. It works best when local plasticity is limited and when material S-N data or design-code curves are available.
Its limitations matter. S-N methods can become less reliable near notches if the stress calculation is poor, and they do not capture plastic strain effects directly. Mean stress corrections such as Goodman, Gerber, or Soderberg may improve realism, but each introduces assumptions that need to match the material and failure mode.
Strain-life approach
The strain-life or E-N method is typically better when local yielding, notch effects, or low-cycle fatigue are significant. It captures the combined elastic-plastic response more directly and is often preferred for components subjected to startup transients, thermal cycling, or severe local stress concentration.
The trade-off is that strain-life analysis requires better material data and more care in estimating local strains. If the input cyclic properties are weak or borrowed from unrelated materials, the result may carry a false sense of accuracy.
Fracture mechanics approach
When the design problem centers on defect tolerance, inspection intervals, or crack propagation, fracture mechanics is usually the better framework. Instead of asking when a crack starts, it asks how an existing flaw grows under cyclic loading.
This is essential in aerospace, offshore, and other high-consequence sectors, but it demands a different level of material data, flaw definition, and validation. It also requires engineers to be clear about whether the governing requirement is initiation life, crack growth life, or both.
Material data is not a formality
Fatigue life prediction is highly sensitive to material inputs, and generic handbook values should be treated cautiously. Surface finish, heat treatment, residual stress, plating, size effects, temperature, corrosion, and manufacturing route can shift fatigue performance substantially.
For metals, one of the most common pitfalls is using polished specimen data for as-machined or welded production parts. For welded assemblies, the local base-metal strength may matter less than weld class, geometry, and fabrication quality. For additive or cast components, defect distribution and anisotropy can dominate life more than nominal strength values suggest.
If test data is unavailable, engineering judgment has to be conservative and explicit. A good fatigue workflow does not hide uncertainty. It identifies it, bounds it, and shows where testing would reduce program risk.
Why FEA quality controls the quality of fatigue results
Fatigue post-processing cannot rescue a weak stress solution. If the underlying finite element model has unrealistic constraints, poor contact behavior, mesh distortion, or singular stress hot spots, the life contour will simply convert those issues into misleading damage predictions.
Mesh and stress extraction
Local fatigue predictions depend heavily on stress gradients and stress extraction method. A coarse mesh may smear a notch response and overpredict life. An excessively refined mesh at a geometric discontinuity can create nonphysical peak stresses that underpredict life. The solution is not just more elements. It is an appropriate mesh strategy tied to the fatigue method and the physical location of damage initiation.
Engineers also need to distinguish between structural stress, nodal stress, element stress, and stress linearization concepts. For welded structures, for example, using nominal or structural stress methods may be more defensible than chasing a peak notch stress that has little correlation to available fatigue data.
Boundary conditions and contacts
Small changes in constraint assumptions can shift load paths and alter fatigue-critical regions. Bolted joints, interfaces, and preload conditions deserve particular care because fatigue failures often initiate where load transfer is more complex than the model assumes.
In Nastran-based environments, disciplined setup, contact validation, and subcase definition make a major difference. Teams that treat fatigue as a late-stage post-process often miss that the real work was in getting the load path and local stress field right from the start.
Mean stress, multiaxiality, and cumulative damage
Real components rarely experience clean fully reversed uniaxial loading. Mean stress can either shorten or extend predicted life depending on material and mode of loading. Multiaxial stress states can further complicate matters, especially near notches, fillets, weld toes, and contact regions.
This is where fatigue software settings can become dangerous if applied mechanically. A default mean stress correction may not be appropriate for the material system. A uniaxial equivalent stress reduction may miss the governing damage mode under combined torsion and bending. Miner’s rule for cumulative damage is useful and widely used, but it remains a simplification, especially under load sequence effects and overload interaction.
The practical lesson is simple: fatigue theory should match the failure mechanism you expect in service. If the result conflicts with field behavior, test evidence, or basic engineering intuition, that tension should be investigated rather than explained away.
Validation is part of the analysis, not an optional extra
The most credible fatigue predictions are anchored to some level of correlation. That may mean coupon data, strain gauge measurements, shaker test results, durability testing, or comparison to known field performance.
Validation does not require perfect agreement, but it does require traceability. Engineers should be able to explain why the selected fatigue model, material data, stress measure, and loading history are appropriate for the component and service environment. That is often the difference between an analysis that supports a design review and one that can withstand customer scrutiny or certification demands.
For organizations building repeatable CAE workflows, fatigue should be handled as a process capability, not a one-off calculation. That means standardizing modeling practices, documenting assumptions, and training analysts to recognize where solver output stops and engineering judgment begins. That is also where experienced support from groups such as eNastran Engineering can materially improve both model credibility and team productivity.
Common reasons fatigue predictions fail in practice
Most bad fatigue predictions are not caused by exotic theory. They come from familiar mistakes: unrealistic duty cycles, borrowed material data, overinterpreted peak stresses, weak mesh strategy, ignored residual stress, and no correlation plan.
Another frequent issue is asking fatigue tools to answer design questions that were never framed clearly. Is the target infinite life, warranty life, inspection interval, or relative ranking between concepts? Each goal can justify a different level of model fidelity. Without that clarity, teams either overbuild the analysis or trust a simplified result beyond its useful range.
A strong guide to fatigue life prediction therefore has to be more than software instructions. It has to connect business risk, test strategy, and simulation discipline. When that connection is made, fatigue analysis becomes a practical decision tool for reducing prototypes, prioritizing design changes, and improving confidence before release.
The best fatigue models do not pretend uncertainty is gone. They narrow it enough that engineering decisions become faster, clearer, and much harder to second-guess.