Machine Learning-Based Classification of Dynamic Damage Modes in Rock–Concrete Composites Under Impact Loading

In tunnel engineering, rock–concrete composite structures may be subjected to dynamic loading induced by earthquakes, blasting, and rockbursts. This study investigated the dynamic response and damage evolution of granite- and sandstone-based composites with ordinary or fiber-reinforced concrete using split Hopkinson pressure bar (SHPB) impact tests at different impact velocities. Acoustic emission (AE) monitoring was combined with a Gaussian mixture model–support vector machine (GMM–SVM) framework to classify damage modes based on the rise-time-to-peak-amplitude ratio (RA) and average frequency (AF). Across the four specimen types, dynamic compressive strength ranged from 37.35 to 57.65 MPa under low-impact conditions and from 102.73 to 120.75 MPa under ultra-high-impact conditions. The corresponding dissipated energy values ranged from 34.19 to 64.12 J and from 192.98 to 266.81 J, respectively. Granite-based specimens exhibited a more pronounced double-peak response, whereas sandstone-based specimens showed more continuous post-peak softening. Fiber reinforcement improved post-peak load-bearing stability, although its effect on energy dissipation varied with impact velocity. AE activity was concentrated mainly during the early impact stage. Under medium-impact-velocity conditions, where damage-mode classification was performed, shear damage dominated in granite-based specimens (56.0% for GC-1 and 53.4% for GCF-1), whereas tensile damage dominated in sandstone-based specimens (52.6% for SC-1 and 59.9% for SCF-1). Fiber reinforcement increased the tensile-damage ratio in granite-based specimens from 44.0% to 46.6%. These findings clarify how rock type and fiber reinforcement influence the dynamic response and damage evolution of rock–concrete composites.