Integrated analytical–FEA–ANN framework for predicting dynamic deflection and stress responses in graphite/epoxy laminated composite plates

Laminated graphite/epoxy plates are attractive for vibration-sensitive lightweight structures, yet their strongly anisotropic response makes repeated evaluation across fibre orientations and excitation frequencies computationally demanding. This study develops an integrated analytical–finite-element–artificial-neural-network (ANN) framework for predicting the harmonic deflection, normal stress, and shear stress of a simply supported eight-ply graphite/epoxy plate over 0–500 Hz. Classical laminate plate theory is used to establish an interpretable mechanics-based reference, while three-dimensional finite-element simulations provide the response database for supervised learning. Unlike conventional single-response FEA–ANN studies, the proposed framework treats three coupled response quantities within one consistent workflow and supplements predictive assessment with orientation sensitivity and uncertainty measures. The finite-element results show that the mean deflection rises from 3.67 × 1 0 − 4 m at 0° to 2.18 × 1 0 − 3 m at 90°, reflecting the pronounced loss of effective bending stiffness as the fibres rotate away from the principal load-carrying direction. The ANN models reproduce the finite-element trends with overall correlation coefficients of 0.9636, 0.9778, and 0.8404 for deflection, normal stress, and shear stress, respectively. The lower accuracy for shear stress is attributed to its stronger sensitivity to local gradients and through-thickness stress recovery. The main contribution lies in constructing a multi-response surrogate framework that is linked to laminate mechanics and accompanied by physically interpretable sensitivity information. Within the investigated material, boundary-condition, orientation, and frequency domain, the framework provides a computationally efficient basis for parametric design and surrogate-assisted optimization.