Machine Learning‐Assisted Prediction of Low‐Velocity Impact Resistance in CFRP /Steel Hybrid Laminates

ABSTRACT Fiber metal laminates are promising lightweight structural materials, but remain susceptible to hidden internal damage under low‐velocity impact. This study investigates the impact resistance of CFRP/DP590 hybrid laminates with and without adhesive‐film interlayers. A non‐adhesive configuration and an adhesive‐toughened configuration were fabricated and tested following ASTM D7136. Multi‐scale characterization revealed energy‐dependent damage evolution in Structure A, from matrix cracking and interfacial delamination at 5–10 J, to resin shear failure and fiber fracture at 20–30 J, and pronounced delamination accompanied by a plateau‐like force response at 40 J. In contrast, Structure B showed markedly improved impact tolerance, with a 41% increase in peak load (10.6 kN) and a 75.5% reduction in damage area at 40 J. The adhesive interlayers suppressed delamination and redirected energy dissipation toward progressive fiber fracture. A temporal neural network was developed to reconstruct impact force–time histories of Structure A. Trained on data at 5, 20, 30, and 40 J and evaluated across different impact energies, the model showed good agreement with experiments, with the most balanced accuracy at 20 J (MAE = 278 N, MAPE = 13.7%, R 2 = 0.976). A comparison with non‐temporal regression approaches suggested the advantage of explicitly representing temporal evolution in force–time responses. Overall, the study demonstrates that interface engineering critically governs damage mechanisms and impact tolerance of CFRP/steel hybrid laminates, and that combining adhesive interlayers with data‐driven temporal modeling offers a practical framework for impact assessment and design in automotive and rail applications.