Neural digital twin-enabled interpretation and optimization of high-strain-rate deformation in LPBF A286 lattice structures

This study presents a high-fidelity Neural Digital Twin (NDT) framework for predictive modelling, mechanistic interpretation, and multi-objective optimization of the high-strain-rate response of laser powder bed fused (LPBF) A286 lattice structures. Honeycomb, body-centred cubic (BCC), and Gyroid architectures were fabricated and experimentally evaluated using Split Hopkinson Pressure Bar (SHPB) testing over an impact pressure range of 2-6 bar. The resulting responses were used to train a deep neural network, establishing a continuous, topology-sensitive mapping between loading intensity and mechanical performance. The NDT demonstrated excellent predictive accuracy, with coefficients of determination of 0.996 for peak stress, 0.926 for maximum strain, 0.957 for strain rate, and 0.941 for absorbed energy, effectively capturing elastic–plastic transition and progressive collapse behaviour. The Honeycomb topology exhibited superior load-bearing capacity and energy dissipation, achieving peak stresses of approximately 480-500 MPa and energy absorption of ∼12.0 J at 6 bar, outperforming BCC and Gyroid structures. BCC showed moderate strength with progressive collapse, while Gyroid enabled greater deformation accommodation at lower peak stress. Sensitivity analysis identified a critical deformation corridor between 3.5 and 5.0 bar, marking the transition from buckling to densification. An energy efficiency index (EEI) quantified performance normalized by stress demand, revealing topology-dependent efficiency trends. Multi-objective optimization indicated an optimal operating window of 4-6 bar, with Pareto analysis defining trade-offs between strength and energy absorption. The proposed NDT framework enables data-driven, uncertainty-aware design and optimized lattice selection for high-strain-rate impact applications.