Machine Learning‐Based Prediction of High‐Strain‐Rate Response in Additively Manufactured Lattice Structures Using Split Hopkinson Pressure Bar Experiments

Abstract

This study explores the predictive modeling of high‐strain‐rate mechanical behavior of additively manufactured lattice structures using machine learning (ML) and deep learning frameworks. Body‐centered‐cubic, gyroid, and honeycomb lattice architectures are fabricated using laser powder bed fusion with A286 superalloy and subjected to split Hopkinson pressure bar (SHPB) testing at impact pressures ranging from 2 to 6 bar. Experimental data are extracted for peak stress, max strain, max strain rate, and energy absorbed and structured into a multioutput supervised learning dataset. Four regression models (linear regression, support vector regression, random forest, and a fully connected neural network) are trained and compared. The neural network achieves the highest predictive performance with R 2 values of 0.9859, 0.9884, 0.9673, and 0.9815 for PeakStress, MaxStrain, MaxStrainRate, and EnergyAbsorbed, respectively, and corresponding root mean squared errors of 7.58 MPa, 0.0121, 131.59 s − 1 , and 1.59 J. Shapley additive explanation‐based interpretability analysis reveals that Pressure_Bar is the most influential predictor for MaxStrainRate, while Lattice_Type dominates the prediction of PeakStress and EnergyAbsorbed. Parity plots and error histograms confirm model fidelity, with the majority of predictions within ±5% of experimental values. The developed ML framework demonstrates strong generalization and offers a data‐driven alternative to traditional SHPB experiments for performance evaluation of architected materials. This methodology provides a foundation for rapid surrogate modeling, material screening, and AI‐assisted design optimization in high‐strain‐rate applications, with future potential for integration into physics‐informed ML systems.

Topics & Concepts