Stacked gated recurrent unit-based prediction of impact energy response in cross-ply composite laminates under low-velocity impact

Composite materials offer a high strength-to-weight ratio, stiffness, and corrosion resistance. However, their layered structure makes them vulnerable to matrix cracking, delamination, and fiber damage under impact loading. Low-velocity impact is a major damage concern in fiber-reinforced composite laminates because it can generate internal defects that are difficult to detect by surface inspection. Accurate prediction of the impact-energy response is therefore important for evaluating impact resistance and improving composite structural design. In this study, a machine learning framework was developed to predict the impact-energy response of cross-ply IM7/Cycom 5320-1 carbon/epoxy laminates under low-velocity impact loading. A finite element model of the low-velocity impact setup was developed in Abaqus and validated against an existing experimental drop-weight impact result. After validation, the finite element model was used to generate impact-energy histories at 5, 10, 15, 20, and 25 J. These simulated responses were then used to train a stacked gated recurrent unit (GRU) model. The trained model was subsequently used to predict the impact-energy response at an unseen 30 J impact condition, and the prediction was validated against the corresponding Abaqus 30 J simulation result. The model achieved good agreement with the Abaqus 30 J simulation result, with a mean absolute error of 0.79 J, root mean square error of 0.89 J, coefficient of determination of 0.99, and mean absolute percentage error of 6.42%. These findings show that the stacked GRU model was able to predict the impact-energy response at an unseen impact condition with good agreement with the finite element result.