An integral regularization method was applied to the Gurson-Tvergaard-Needleman (GTN) damage model to address the plastic deformation localization at the crack tip. The parameters for the constitutive model were calibrated with an evolutionary algorithm. The model was applied to predict ductile fracture of a tensile specimen, allowing mesh convergence and geometry independence. Fatigue crack growth was predicted through a node release strategy, obtaining a very good approximation to the experimental results in the upper part of the Paris Regime. Mesh independence was verified for medium/high Δ K  levels, but near the threshold regime no mesh independence was obtained and da/dN was overestimated. The characteristic length for the non-local model was defined as the reverse plastic zone size, which indicates that this should be the fatigue process zone. Finally, at the accelerated regime, final fracture loci predictions agree with the experimental results, but mesh independence was barely achieved.