Hybrid composite structures (e.g., carbon–glass laminates, fiber–metal laminates, and multi-material sandwich panels) offer superior stiffness-to-weight performance but exhibit complex, multi-mode damage mechanisms such as matrix cracking, fiber breakage, delamination, and interface debonding. These damage modes are often difficult to detect early and expensive to simulate at full structural scale with high fidelity. This paper proposes an integrated framework that combines multi-scale progressive damage simulation with machine learning (ML)–assisted damage inference and load optimization. At the microscale and mesoscale, damage initiation and evolution are captured using established composite failure criteria and degradation laws (e.g., Hashin-type mechanisms), while structural-scale response is computed using reduced-order surrogates calibrated from multi-scale results. On the data side, guided-wave/shock-response features and simulated strain-field descriptors are mapped to damage states using supervised and uncertainty-aware ML models. Finally, a load optimization module minimizes peak interlaminar stresses and damage growth rate under service constraints. A case study on a hybrid laminate panel demonstrates that the proposed pipeline can (i) identify early delamination and matrix cracking signatures with high classification performance, and (ii) reduce damage-driving stress metrics through ML-guided load redistribution.