A welded bracket on a machine that cycles 20 times per minute. It cracked at the weld toe after 6 months. The weld looked fine. The static strength was fine. It failed in fatigue. The customer thought the welder was bad. The welder was fine. The joint design was wrong for a cyclic load. This is how I design welded joints for fatigue life.
Why welds fail in fatigue
A weld toe is a stress concentration. The weld bead transitions into the base material at an angle. The sharp transition creates a local stress that’s 2-3x the nominal stress. Under a cyclic load (the bracket flexes every cycle), that local stress drives a crack from the toe. The crack grows each cycle. After enough cycles, it reaches critical size and the bracket breaks. Static strength doesn’t predict this — you can have a weld with 2x the required static strength and still fail in fatigue at 10,000 cycles.
The fatigue categories
Eurocode 3 and IIW classify weld details by fatigue strength. The category is a number that represents the stress range at 2 million cycles. For example, a category 80 detail can take 80 MPa stress range for 2 million cycles before failure.
| Detail | Category (MPa @ 2M cycles) | Use for |
|---|---|---|
| Continuous full penetration weld, ground flush | 100 | Critical fatigue, high cycle |
| Continuous fillet weld, as-welded | 80 | Moderate fatigue |
| Intermittent fillet weld | 50 | Low fatigue, static loads |
| Weld with start/stop at the toe | 36 | Poor — avoid |
| Weld penetration defect (under 1 mm) | 25 | Worst — rejects |
The example
The bracket was a 6 mm plate welded to a 10 mm column. The weld was an intermittent fillet (20 mm weld, 10 mm gap, repeat). The cyclic load at the weld toe was 60 MPa stress range. Category 50 (intermittent) means failure at 50 MPa. We’re at 60 MPa — over the limit. The weld fails in fatigue. After 6 months (about 2 million cycles), it cracked. That’s exactly what happened.
What I changed
1. Continuous weld instead of intermittent. I made the fillet weld continuous around the joint. The category jumped from 50 to 80. At 60 MPa stress range, the category 80 detail survives for 2 million cycles at 80 MPa. We’re at 60 — 75% of the limit. It doesn’t fail.
2. Ground the weld toe. I had the welder grind the toe of the weld smooth. The transition from weld to base material is no longer sharp. The stress concentration drops. The category jumps from 80 to 100. The fatigue life doubles. Grinding takes 5 minutes per weld but prevents a $500 field repair.
3. Reduced the stress range. The 60 MPa stress was from the bracket flexing. I added a gusset (a triangular brace) that stiffened the joint. The stress dropped from 60 to 30 MPa. At 30 MPa on a category 80 detail, the fatigue life is enormous (100+ million cycles). The gusset costs $10 of material. It prevents the crack from starting.
The cyclic load, not the static load
Fatigue depends on the stress range (max stress minus min stress), not the maximum stress. If the bracket sees 100 MPa max and 80 MPa min (range = 20 MPa), it’s fine. If it sees 50 MPa max and 0 MPa min (range = 50 MPa), it fails faster. A bracket that sits unloaded between cycles has a wider stress range than one that’s preloaded. I check the range, not the peak.
The weld quality matters
A good weld (no undercut, proper penetration, smooth toe) is 2x stronger in fatigue than a bad weld (undercut at the toe, porosity, incomplete fusion). I specify the weld quality on the drawing: “fillet weld, 6 mm leg, continuous, toe ground smooth.” The welder knows what’s expected. A sloppy weld on a cyclic joint will fail regardless of the design category.
The number I check: the stress range at the weld toe, compared to the detail category. For cyclic loads (over 10,000 cycles), use continuous welds, ground the toe, and add stiffeners to reduce the stress. Intermittent welds are for static loads, not fatigue. The bracket that cracked after 6 months had an intermittent weld on a cyclic load — the design was wrong, not the welder.
