Numerical-Resolution Effects on FORM Reliability Estimates of Flexible Pavements Under Moving Loads: A Mode-Specific Assessment

Reliability estimates from dynamic pavement models can depend on numerical settings that appear adequate at nominal inputs. This study evaluates how Fourier truncation, cross-sectional mesh, time step, and interface recovery propagate into FORM estimates for fatigue and a rutting proxy in a four-layer pavement under a moving load. The proposed assessment separates recovery-location artifacts from discretization effects, tests numerical adequacy near each FORM design region, and permits mode-specific resolution. The SAFEM implementation reproduced benchmark response signs, ordering, timing, and a reported midspan interface stress within 2.8%. Interface-tied recovery removed a spurious depth-refinement trend. Across the numerical hierarchy, the signed fatigue index changed from 0.9099 to −2.1818, reversing the conditional classification; Fourier and transverse resolution controlled fatigue most strongly, whereas depth resolution controlled rutting. Direct half-step FORM reoptimization and an independent start agreed, and a 60-sample importance-sampling check provided evidence of consistency with direct sampling within the achieved sampling uncertainty. However, a nested half-to-quarter-step diagnostic exceeded the predeclared numerical target, so temporal convergence was not claimed and a diagnostic envelope was reported. A prospective Gaussian-process gate also required direct fallback. Nominal convergence alone therefore did not certify reliability-region accuracy; qualification had to combine response-consistent recovery, mode-specific refinement, design-point checks, and probability-level controls.