Corrosion Hotspots Strongly Localize Crack Origins But Provide Conditional Timing Information

ABSTRACT Published specimen‐level data from 36 pre‐corroded structural‐steel fatigue specimens relevant to offshore applications were re‐analyzed to separate crack‐origin localization from crack‐initiation timing. Localization was available for 34 specimens and digital‐image‐correlation‐based timing for 21. Crack origins coincided with the global finite‐element hotspot in 20/34 specimens (58.8%) and with a global or near‐global hotspot in 31/34 (91.2%). For timing, adding maximum stress concentration to nominal stress range and stress ratio reduced leave‐one‐out cross‐validated root‐mean‐square prediction error from 0.2463 to 0.2114 log10 cycles (14.2%) and increased the predicted‐residual‐sum‐of‐squares coefficient of determination from 0.401 to 0.559. The pooled gain survived influence and within‐stress‐ratio sensitivity analyses but reversed in the largest homogeneous stress‐ratio subgroup ( R = 0.1) and depended on sparse upper‐tail support. Interaction models did not improve prediction. Mechanical hotspots strongly localize crack origins, whereas timing from a single maximum stress‐concentration value is conditional, motivating full‐field morphology tests.