Weld toe and weld root failures dominate the fatigue failures of the loadcarrying cruciform fillet welded joints. Compared to weld toe failure, the weld root failure is strongly affected by its inherent weld quality and throat size, which is more unpredictable and should be eliminated. This work proposes a weld sizing criterion in terms of two weld leg sizes and weld penetration to avoid root failure. The criterion is obtained by a data-driven approach with the help of the mechanical-directed data augmentation technique. The tractionstress-based method is applied for the data augmentation to identify the fatigue failure mode, which is statistically tested against 372 experimentally obtained fatigue data.cruciform joint, data augmentation, data driven, weld root failure, weld sizing | INTRODUCTIONThe load-carrying cruciform fillet welded joint (LCFWJ) is broadly applied to connect secondary components to main members in metal structures. [1][2][3] Two failure modes dominate the fatigue fractures of LCFWJ, namely, weld toe and root failure, depending on the load conditions and the joint geometry, as illustrated in Figure 1. Fatigue crack initiating from the weld root and propagating into the fillet weld is referred to as weld root failure, which should be eliminated during the design stage of the welded structures because of two reasons. (1) Fatigue lives according to weld root failure is usually much lower than that associated with weld toe failure. As stipulated in EN-1993-1-9, 4 the S-N curve recommended for weld root failure falls into the worst possible category, namely, FAT36, which is lower than that of the weld toe failure. 5 Ziqi Zhao and Tairui Zhang contributed equally to this work and should be considered co-first authors.

No abstract