This study aims to integrate topology optimization, graded strut reinforcement and data driven surrogate modeling to tailor the compressive and energy absorption performance of fused deposition modeled lattice structures.
Five architectures (fluorite, Kelvin, isotruss, re-entrant, truncated octahedron), each representing a distinct mechanical archetype relevant to energy-absorbing and load-bearing applications, were compared under quasi-static compression (ASTM D695-23). The truncated octahedron, selected for optimal nodal connectivity, was further enhanced via pointwise strut-thickness grading (1.2 → 0.8 mm). A Taguchi L9(3³) array screened layer height, print speed and nozzle temperature (NT); analysis of variance and regression isolated significant factors. A Gaussian process regression (GPR) surrogate with an automatic relevance determination Matérn 5/2 kernel was trained on the experimental data set; Bayesian optimisation with an expected improvement acquisition function was used to predict specific energy absorption (SEA) and interrogate the strength–SEA Pareto front via dual-surrogate analysis.
The truncated octahedron exhibited the highest strength (7.43 MPa) in a statistically significant topology hierarchy (F4, 45 = 156.3, p < 0.001). Pointwise grading increased strength to 8.75 MPa, outperforming linear/stepped strategies by 71%/35%. Taguchi analysis identified NT as dominant ( = 0.63); optimal settings delivered 9.63 MPa. SEA decoupled from strength: Run S8 achieved the highest SEA (331.5 J·kg-1) at moderate stress (9.15 MPa), while Run S1 maximized strength (10.32 MPa) with slightly lower SEA. The GPR surrogate predicted SEA with root mean square error = 12.3 J·kg−1 and R² = 0.94 (bootstrap 95% CI: [8.7–17.1] J·kg−1; permutation p = 0.001); confirmation printing at the Bayesian-optimization-recommended parameters validated 0.9% prediction error.
We propose a validated framework for fused deposition modeling lattice crashworthiness that combines graded strut architectures with Gaussian process surrogate modeling. Our surrogate model cuts experimental effort by ∼70% while preserving accuracy, enabling efficient design-space exploration. The framework, validated against a physical confirmation experiment, is material- and geometry-agnostic and directly applies to aerospace, automotive and biomedical energy absorbers.
