Fatigue accounts for an estimated 50 to 90 percent of all mechanical failures in service — a remarkably wide range that reflects how often the precise mechanism goes undiagnosed. When a part fractures under loads it has handled many times before, the instinct is to blame overload, material defect, or assembly error. In most cases, the correct explanation is fatigue: the accumulation of microscopic damage over repeated load cycles until a crack propagates to critical size.
Understanding fatigue is fundamental to both preventing new failures and diagnosing existing ones. The physical process is well understood; the variables that control it are largely within the engineer's influence at the design stage. This post covers the mechanism, the diagnostic features, and the design decisions that prevent fatigue failures from occurring.
Fatigue is damage accumulation in a material under cyclic stress. Unlike static overload, which requires that the applied stress exceed the material's yield strength, fatigue damage occurs at stresses well below yield — sometimes as low as 30 to 50 percent of the ultimate tensile strength. The mechanism is not a single catastrophic event but a progressive process: initiation of a crack at a local stress concentration, followed by stable crack growth per cycle, followed by unstable fracture when the crack reaches a critical size.
The reason fatigue damage occurs below yield strength is the behaviour of stress at discontinuities. A smooth, uniform cross-section under nominal stress σ has a maximum stress equal to σ everywhere. But a notch, a hole, a surface scratch, or a change in cross-section creates a stress concentration where the local peak stress is Kt × σ, where Kt is the stress concentration factor. A sharp keyway in a rotating shaft might have Kt = 2.5 to 3.5, meaning the local stress at the keyway root is two and a half to three and a half times the nominal shaft stress. If the nominal stress is well within the material's rated allowable, the local stress at the concentration may not be — and under cyclic loading, this local stress drives crack initiation.
Fatigue Is a Local Phenomenon
Fatigue damage begins at the location of highest local stress — which is
almost always a stress concentration rather than the nominal cross-section.
This is why fatigue failures almost always initiate at geometric features:
keyway roots, hole edges, thread roots, shoulder radii, surface defects, and
corrosion pits. The nominal stress can be well within the design allowable
while the local stress at a concentration exceeds the fatigue limit.
Crack initiation begins at the surface or just below it, at the location of highest local stress. In metals, the mechanism involves cyclic plastic deformation at the microscopic level. Under repeated stress reversals, slip bands form in individual grains oriented favourably with respect to the stress direction. These slip bands eventually develop into micro-cracks, typically along persistent slip planes in the grain.
Initiation is promoted by any condition that increases local stress or reduces local material resistance:
Surface stress concentrations: machining marks, scratches, stamping marks, corrosion pits, and grinding burns all act as initiation sites
Residual tensile stress: tensile residual stress from machining, welding heat-affected zones, or incorrect heat treatment accelerates initiation by adding to the cyclic mean stress
Surface condition: rough surfaces have more sites for slip band formation than smooth surfaces; this is why fatigue specimens are highly polished and why production components are specified with ground or polished finishes at stress-critical locations
The initiation stage typically consumes 10 to 40 percent of the total fatigue life in high-cycle fatigue. In low-cycle fatigue (high stress, short life), initiation is rapid and propagation dominates.
Once a micro-crack has formed, it propagates under continued cyclic loading. Crack propagation follows the Paris Law relationship:
da/dN = C × (ΔK)^m
where da/dN is the crack growth increment per load cycle, ΔK is the cyclic stress intensity range at the crack tip, and C and m are material-specific constants. The implication is that crack growth rate is strongly sensitive to stress — doubling the stress range at the crack tip increases crack growth rate by a factor of 2^m, and m for most structural metals is between 2 and 4.
During stable propagation, the crack grows incrementally with each load cycle, leaving a record of its progress. Each time the loading is interrupted (by a shut-down, a load change, or a period of reduced amplitude), an arrest line is left on the fracture surface. These are the beach marks visible on fatigue fractures — they are the macro-scale record of the crack front position at different points in time.
The propagation zone typically shows a smooth, almost polished texture compared to the final fracture zone. This smoothness results from the crack faces rubbing together under compressive phases of the load cycle, burnishing the surface over thousands of cycles.
When the crack has propagated to the point where the remaining uncracked cross-section can no longer support the applied peak load, unstable fracture occurs. This happens rapidly — often in a single load cycle — and produces the rough, irregular final fracture zone visible on fatigue fractures.
The relative size of the propagation zone and the final fracture zone communicates information about the stress level. A large propagation zone (the part cracked through most of its section before fracture) indicates that the final fracture load was not much higher than the fatigue crack driving force — the nominal stress was relatively low. A small propagation zone and large final fracture zone indicates that the final fracture occurred at a high fraction of the section capacity — the nominal stress was relatively high.
Several factors reduce the fatigue life of a component below what the material's baseline properties would predict.
Stress concentration. The most significant factor. Fatigue life is inversely proportional to the effective stress at the initiation site. Reducing Kt from 3.0 to 1.5 at a critical location can double or triple the fatigue life, depending on the material's notch sensitivity.
Surface roughness. Rough surfaces reduce fatigue life by providing more initiation sites and by creating micro-stress-concentrations. The reduction in fatigue limit from a machined surface compared to a polished laboratory specimen can be 10 to 30 percent; the reduction from an as-cast or as-forged surface can be 30 to 50 percent.
Residual tensile stress. Tensile residual stress from welding, grinding, EDM, or aggressive machining adds to the mean stress at the initiation site, reducing the cyclic stress range the material can sustain before fatigue damage begins. Shot peening, surface rolling, and careful control of machining parameters can introduce beneficial compressive residual stress that delays initiation.
Corrosive environment. Fatigue in a corrosive environment (corrosion fatigue) is significantly more damaging than fatigue in a dry environment. The corrosion accelerates slip band formation, prevents crack tip blunting, and dissolves the oxide layer that would otherwise reduce crack opening displacement. Components operating in marine, chemical, or humid environments have substantially reduced fatigue life compared to laboratory specimens.
Size effect. Larger components have more material volume and more surface area exposed at high stress, increasing the probability that a critical initiation site exists. Fatigue data from small laboratory specimens is non-conservative for large cross-sections. Size correction factors are applied in design calculations to account for this.
Welded Joints Have Lower Fatigue Resistance Than Base Metal
Welded connections in structural applications typically have fatigue limits 40
to 60 percent lower than the base metal. The reasons are geometric stress
concentration at the weld toe, residual tensile stress from the weld thermal
cycle, and potential undercut or porosity defects. When designing welded
structures for cyclic loading, use fatigue design standards (such as AWS D1.1
or BS 7608) that account for weld category and apply appropriate reduction
factors — do not use base-metal fatigue data.
The S-N curve (also called the Wöhler curve) plots cyclic stress amplitude against the number of cycles to failure. It is the primary tool for fatigue life prediction in design.
For steels and some other iron-based alloys, the S-N curve becomes horizontal at a stress level called the endurance limit (or fatigue limit). Below this stress, the material theoretically will not fail under cyclic loading regardless of the number of cycles. The endurance limit for steels is typically 40 to 50 percent of the ultimate tensile strength for rotating beam specimens — but this is the baseline value for a polished, unnotched specimen with no surface treatments. Real components have lower effective endurance limits due to surface finish, stress concentration, and size effects.
For aluminium alloys and most non-ferrous metals, the S-N curve does not flatten out — it continues to decline gradually at high cycle counts. These materials do not have a true endurance limit; instead, a fatigue strength at a specified number of cycles (typically 10^7 or 5×10^8) is used as the design allowable.
The Goodman diagram extends the S-N concept to account for mean stress (non-zero mean loads superimposed on cyclic loads). A non-zero mean tensile stress reduces the allowable stress amplitude; a compressive mean stress increases it. The Goodman line connects the endurance limit on the alternating stress axis to the ultimate tensile strength on the mean stress axis. Design combinations of mean and alternating stress that fall below this line are considered safe for infinite life.
When examining a part suspected of fatigue failure, the visual features described in the previous post provide the primary diagnostic evidence. The confirmation checklist:
Beach marks visible on the fracture surface, spreading from one or more origin sites
Initiation site located at a stress concentration (notch, hole, radius, surface defect)
Final fracture zone size consistent with the expected load level
Absence of significant plastic deformation in the vicinity of the fracture (fatigue fractures in moderate-strength materials are typically flat with little distortion of the surrounding geometry)
Part has been in cyclic service — rotating shafts, reciprocating mechanisms, vibrating structures, pressurised systems with cyclic pressure variation
The absence of beach marks does not rule out fatigue — in high-frequency applications or when the loading is very uniform, arrest lines may be too closely spaced to resolve visually. Scanning electron microscopy at 500–5,000× can reveal fatigue striations (micro-scale beach marks corresponding to individual cycles) in these cases.
Count the Initiation Sites
The number of fatigue crack initiation sites visible on the fracture surface
is a direct indicator of the stress level relative to the material's fatigue
limit. A single initiation site typically indicates a local defect (scratch,
pit, inclusion) in a component operating at moderate stress. Multiple
initiation sites around the perimeter of a shaft, for example, indicate high
nominal stress — the stress concentration is the geometry itself rather than a
localised defect. High-stress fatigue with multiple initiation sites requires
a redesign to reduce stress, not just a surface quality improvement.
A stainless steel mixing shaft was failing by fatigue fracture at the root of a drive keyway after approximately 2,000 hours of service. The shaft was 50 mm diameter, rotating at 120 rpm under a bending moment from off-centre impeller loading, yielding a calculated nominal alternating bending stress of 58 MPa.
Visual examination confirmed fatigue: beach marks, single initiation site at the keyway root corner, final fracture zone covering approximately 30% of the cross-section.
Material check: the shaft was 316L stainless steel with a literature endurance limit of approximately 170 MPa for polished specimens. The keyway was machined with an end radius of approximately 0.1 mm (estimated from examination). For this geometry, Kt was estimated at 3.1 from standard reference charts.
Effective fatigue stress at initiation site: 58 MPa × 3.1 = 180 MPa. This exceeds the endurance limit of the base material; fatigue was expected.
Two corrections were evaluated. Option A: increase the keyway end radius to 1.5 mm, reducing Kt to approximately 1.6. Effective stress: 58 × 1.6 = 93 MPa — below endurance limit. Option B: reduce the nominal bending stress by revising the impeller mounting geometry to centre the load.
Option A was implemented: the keyway was re-machined with a 1.5 mm radius and the shaft was shot-peened to introduce beneficial compressive residual stress at the keyway root. Estimated endurance limit after shot peening: approximately 210 MPa for the surface condition. No subsequent fatigue failures were reported over the 4,000 additional service hours of the follow-up period.
Fatigue failure prevention operates through three main levers at the design stage:
Reduce stress concentration. Use generous radii at section changes, avoid sharp corners at reentrant features, minimise stress concentrating features at high-stress locations. A radius increase from 0.5 mm to 2.0 mm at a shoulder transition can reduce Kt from 2.8 to 1.5 — a substantial improvement in fatigue life with minimal geometry change.
Reduce nominal stress. Increase the section modulus at the critical location, reduce the applied load through balanced design, or relocate the stress-critical feature to a region of lower nominal stress. Moving a keyway from the highest-stress zone of a shaft (midspan under distributed load) to a lower-stress zone (near a support) reduces the nominal bending stress at the feature without changing the geometry.
Improve surface condition. Specify appropriate surface finish at fatigue-critical locations. Ground or polished surfaces have fewer initiation sites than machined surfaces. Shot peening, surface rolling, and nitriding introduce compressive residual stress that delays crack initiation. Avoid machining operations that introduce tensile residual stress (aggressive grinding without cooling, EDM) at critical surfaces.
Fatigue is the dominant failure mode under cyclic mechanical loading. The next post examines corrosion failure — a different mechanism, with different visual signatures, but often interacting with fatigue to produce corrosion fatigue at stress levels that would be safe in a non-corrosive environment. Understanding how to recognise and distinguish corrosion mechanisms is essential for any engineer working with equipment exposed to moisture, chemicals, or outdoor conditions.
Fatigue failures initiate at stress concentrations well below the material's yield strength, then propagate incrementally before final fracture
The three stages — initiation, propagation, final fracture — each leave characteristic visual evidence on the fracture surface
Stress concentration factor (Kt) is the most significant variable controlling fatigue life; reducing Kt through geometry improvements has more impact than upgrading material strength
The S-N curve and endurance limit define the cyclic stress a material can sustain indefinitely; real components require correction factors for surface finish, stress concentration, and size
Fatigue prevention at the design stage: reduce stress concentration, reduce nominal stress, and specify appropriate surface condition at fatigue-critical locations
