Purpose This study aims to establish an interpretable machine learning framework for predicting high-strain-rate mechanical responses of laser powder bed fusion (LPBF)-fabricated A286 steel lattice structures subjected to split Hopkinson pressure bar (SHPB) testing. By integrating multi-output regression with explainable AI (XAI) tools, the research seeks to quantify the influence of impact pressure and lattice topology on key dynamic response metrics. The objective is to develop accurate, physics-aligned predictive models that can inform data-driven lattice design strategies for impact-resistant applications in aerospace and defence. Design/methodology/approach Three lattice topologies: body-centred cubic (BCC), honeycomb and gyroid were additively manufactured and subjected to SHPB testing under dynamic pressures ranging from 2–7 bar. Key response variables, including peak stress, maximum strain, maximum strain rate and energy absorbed, were extracted. Multi-output regression was performed using CatBoost, XGBoost, Random Forest, Gradient Boosting and Extra Trees algorithms, with impact pressure and lattice type as input features. Findings The CatBoost model yielded the highest accuracy for peak stress prediction (R² = 0.9848), XGBoost for maximum strain (R² = 0.9877), Gradient Boosting for strain rate (R² = 0.9659) and Random Forest for energy absorption (R² = 0.9839). Shapley additive explanations analysis revealed nonlinear interactions between impact pressure and lattice type, particularly beyond 6 bar. Surrogate trees and local interpretable model-agnostic explanations identified interpretable design rules, such as BCC lattices at ≥6 bar leading to optimal energy absorption. The integrated framework provided transparent and high-fidelity predictions with strong alignment to physical deformation mechanisms observed during dynamic testing. Research limitations/implications The primary limitation of this study lies in the relatively small dataset size due to experimental constraints of SHPB testing and LPBF fabrication cycles. While the models demonstrate high predictive accuracy and explainability, their generalizability to other alloys, lattice topologies, or loading regimes remains to be validated. Additionally, the current framework does not account for temperature effects, anisotropy, or microstructural evolution during impact. Future research should focus on expanding the dataset, incorporating microstructural inputs and applying the framework to broader materials and geometries for more robust and transferable design guidance. Practical implications The interpretable machine learning framework developed in this study provides engineers and designers with predictive tools to optimize lattice geometries for high-strain-rate applications. The identified pressure–geometry performance thresholds and extracted surrogate rules can directly inform the design of energy-absorbing components in aerospace, automotive and defence systems. By offering a transparent model architecture with feature attributions, the approach enables accelerated design iteration, risk mitigation and enhanced material performance under dynamic loading, reducing reliance on exhaustive physical prototyping. The framework is modular and can be adapted to new materials and use-cases with minimal customization. Social implications Improving the impact resilience of structural components using data-driven lattice design can contribute to safer transportation, aerospace and defence systems. The ability to design energy-absorbing structures with minimal weight enhances fuel efficiency and reduces environmental impact. Furthermore, the integration of XAI into materials engineering promotes ethical and trustworthy adoption of machine learning in critical industries, ensuring that decisions are transparent and grounded in physical behaviour. The methodology fosters knowledge democratization by enabling domain experts, not just data scientists, to interpret and act upon model insights in socially impactful applications. Originality/value This is the first study to apply a comprehensive, explainable machine learning approach for high-strain-rate prediction of LPBF-fabricated A286 lattice structures using SHPB data. Unlike prior black-box models, this framework emphasizes interpretability through XAI, providing actionable design insights rooted in physics. It establishes practical thresholds for impact mitigation and identifies geometry–pressure interactions that influence energy dissipation. The findings support the development of AI-informed lattice design guidelines for protective structures, marking a significant advancement in the deployment of interpretable AI for additive manufacturing in defence and aerospace contexts.
Bobbili et al. (Wed,) studied this question.