Surface morphology-driven inverse quantification of LVI energy in thermoplastic laminates via hybrid physics-data modeling

Low-velocity impact events in carbon fiber-reinforced thermoplastic laminates frequently produce barely visible internal damage that critically undermines residual structural strength, yet no practical field method exists to inversely quantify the absorbed energy without disassembly or specialized instrumentation. This study presents a hybrid physics-data framework that inversely determines impact energy directly from surface damage morphology in T300/PPS plain-woven laminates, eliminating the need for complex NDT procedures. A validated finite element model incorporating fiber-aligned meshing, plasticity-augmented Hashin damage criterion and cohesive delamination formulation achieves absorbed-energy errors within 8 % and compression-after-impact strength errors within 12 %. A hybrid dataset of 280 samples fuses experimental measurements with simulations, from which surface descriptors (dent depth, damage area, thickness) are automatically extracted. Six machine learning algorithms are benchmarked with systematic hyperparameter optimization, with LightGBM achieving optimal performance ( R ² = 0.9686, RMSE = 0.683 J, MAE = 0.486 J, MAPE = 9.65 %). Shapley analysis identifies dent depth and laminate thickness as dominant predictors, confirming indentation-thickness coupling as the governing mechanism. The framework enables accurate energy quantification using only surface-accessible measurements, offering a practical alternative to NDT-based characterization for in-service damage assessment.