A variance-based fixed-smoothing-scale SPH protocol with h = 2 σ and kernel-dependent support H = κ h is implemented in OpenRadioss to decouple kernel/layout effects from resolution. Four kernels (CS, W2, W4, W6) and two ordered layouts (SC, FCC) are compared at equal per-particle volume/mass on two benchmarks: Taylor impact in 4340 steel with pure SPH and spherical projectile perforation of an AA2024-T3 plate with FE–SPH coupling. The investigated smoothing-scale/particle spacing ratios are in range h / d = 1 . 10 –2.74. Accuracy is assessed using residual bar length and cap diameter and residual projectile velocity, together with wall-clock runtime. For Taylor impact, errors are typically minimized at intermediate smoothing scale, where the numerical results are close to reference experimental data (between 1% and 10% depending on the quantity of interest); the smallest h can exhibit free-surface neighbor deficiency and tensile-instability-driven particle ejection, while the largest h over-smooths cap geometry. Errors generally decrease with kernel smoothness, and FCC often outperforms SC at fixed h . For plate impact, residual velocity is governed primarily by parameter h , with larger h yielding smoother deceleration, lower plateau velocities, and reduced kernel sensitivity. Runtime is jointly affected by the choice of kernel and smoothing scale h . The fixed- h protocol establishes a common-resolution framework for cross-kernel comparison andprovides a transferable baseline for balancing accuracy, stability, and cost in explicit SPH impact simulations.
Li et al. (2026) studied this question.