A DNN Surrogate Framework with Mode-Aware Scoring for Blast-Resistant Design of CFRP-Steel Fiber Metal Laminates

Fiber metal laminates (FMLs) combine the ductile energy absorption of metals with the high specific strength of composites, making CFRP–steel laminates promising for blast protection. Their near-field blast resistance depends on charge mass, stand-off distance, and thickness ratio, making conventional finite element (FE) optimization costly. This study proposes a rapid framework integrating small-sample deep neural network (DNN) surrogate modeling with mode-aware scoring. A validated explicit dynamic FE model generated 63 cases covering maximum displacement, residual plastic displacement, and absorbed energy. With scaled distance Z as a physically guided input, the DNN pipeline achieved R 2 ≥ 0.988 on 13 in-domain test samples and exceeded the single-output XGBoost baseline by 20 percentage points in mean R 2 under the adopted protocol. Ablation showed that Z halved absorbed-energy RMSE, while predictions displayed partial consistency with the Hopkinson–Cranz cube-root scaling trend. As thickness ratio k increased, maximum displacement generally decreased, whereas absorbed energy exhibited an interior minimum, indicating distinct response trends. An 8 mm residual-plastic-displacement threshold partitioned the design space into small- and large-deformation regions and justified mode-dependent weighting. The resulting domain-bounded preliminary optimization chart for Z ∈ [0.149, 0.646] recommended k * ∈ [6, 8] and [3, 5] for the two regions, respectively. The surrogate provided a 10 6 -fold speedup over direct FE analysis, enabling rapid thickness-ratio selection within the validated domain for vehicle armor, ship bulkheads, and blast-proof doors.

Hali tarjima qilinmagan