Fatigue of Welded Joints: Mechanisms, Assessment, and Design

The majority of in-service structural fatigue failures — in pressure vessels, offshore platforms, bridges, and rotating equipment — do not originate in the base metal: they initiate at the weld. By the time a crack is visible, it has typically been growing silently for years — initiated at a stress concentration that is an unavoidable consequence of the welding process itself.

O'Donnell Consulting Engineers, Inc.  |  Technical Resources

Why Welds Are Fatigue-Critical

The weld toe — the junction between the weld face and the base metal surface — is geometrically abrupt. Even a well-executed fillet weld transitions at an angle that produces a local stress concentration factor of two to five times the nominal value. Combine this with tensile residual stresses that frequently approach yield strength (a consequence of constrained thermal cycling during fabrication) and microstructural discontinuities in the heat-affected zone, and you have conditions that are intrinsically hostile to cyclic loading.

In practice, a large proportion of in-service structural fatigue failures initiate at welds rather than in the base metal — not because welding is inherently defective, but because the physics of melting, solidification, and constrained cooling under load produce a predictable set of crack-initiating conditions at the weld toe and root.

S-N Curves and Detail Classification

The S-N curve relates applied stress range (Δσ) to cycles to failure (N). For welded joints, design S-N curves are statistical — typically mean minus two standard deviations, representing approximately 97.7% survival probability — and are applied to a classified joint detail. The slope exponent m = 3 governs the finite-life regime for most weld categories in IIW and BS 7608; below the constant amplitude fatigue limit (CAFL, typically at 10⁷ cycles), crack propagation is assumed to arrest under constant amplitude loading.

The IIW system uses FAT classes, where the FAT number is the characteristic stress range in MPa at 2 × 10⁶ cycles. Representative values:

Assessment Methods

Three analytical approaches are available, each offering progressively finer resolution at greater computational cost:

For variable amplitude service loading, Rainflow cycle counting combined with Palmgren-Miner linear damage accumulation (D = Σ n_i/N_i) is the standard framework. Cycles below the CAFL require special treatment — most codes apply a reduced second slope (m = 5) rather than truncating sub-threshold cycles, which would non-conservatively overstate service life.

Post-Weld Treatment

When the as-welded fatigue performance is insufficient, several proven treatments can extend life:

• Post-Weld Heat Treatment (PWHT) — Slow heating to 550–650°C reduces residual stresses to 10–20% of yield strength. Standard for pressure vessels and critical pipe welds; improves HAZ toughness and reduces stress corrosion susceptibility.
• High-Frequency Mechanical Impact (HFMI) — Plastically deforms the weld toe to introduce compressive residual stress and improve toe geometry. Recognized in IIW recommendations with FAT class improvements of two to four classes for high-strength steels.
• Burr Grinding & TIG Dressing — Geometry-focused treatments that remove undercut and sharpen the weld toe radius. One to two FAT class improvements when properly applied and inspected.

Fatigue-Resistant Design Principles

The most effective mitigation is geometric. The following principles apply across pressure vessels, structural frames, and rotating components:

Route welds away from high-stress regions whenever the geometry permits. A weld that carries no primary structural load is not a fatigue detail.

Avoid abrupt thickness changes and offset misalignment in butt welds. Specify full-penetration welds at primary load paths in preference to partial-penetration or fillet welds; taper cover plate and attachment terminations to reduce stress concentration.

Specify appropriate weld quality and NDE. Undercut, overlap, and lack of fusion are not cosmetic defects — they are pre-existing crack-like notches at the most stressed location:

ISO 5817 Level B or C for fatigue-critical details, combined with phased array UT (PAUT) or TOFD where detection sensitivity matters, is not conservatism for its own sake. In fracture mechanics assessment, assumed initial flaw size is directly coupled to NDE detection capability: better inspection translates directly into longer calculated inspection intervals.

Account for the full load spectrum. Variable amplitude loading requires Rainflow cycle counting and Miner's rule damage summation. Truncating sub-CAFL cycles from the damage sum can non-conservatively overstate life — particularly relevant for wind turbines, cranes, and vessels subject to pressure cycling.