<p>Accurate vehicle damage assessment is fundamental to intelligent insurance claims and automated inspection systems. However, many existing methods overlook the structural context of the vehicle and rely on unstable global severity metrics that are sensitive to perspective and distance. This study proposes a part-aware damage assessment framework based on an enhanced YOLO11 segmentation architecture and a structured severity formulation. The model incorporates a receptive field aggregation module termed C3k2-RFAConv to improve contextual perception while preserving fine-grained details. In addition, an adaptive spatial feature fusion strategy known as ASF-YOLO is employed to enhance multi-scale feature interaction and boundary refinement. A dual-model design is adopted in which a vehicle part segmentation network provides structural priors for damage-aware analysis. Utilizing pixel-level masks, a Damage Severity Index is formalized as an interpretable heuristic severity indicator by combining the damage-to-part area ratio, confidence scores, and category-specific importance weights. A conditional normalization strategy is further applied to reduce sensitivity to viewpoint variations and imperfect part segmentation, establishing a regulated structural topology score rather than a validated actuarial metric. Experiments conducted on a self-constructed damage dataset and a public vehicle part dataset demonstrate improved segmentation performance over baseline YOLO11 models, particularly in mAP metrics at higher IoU thresholds. Qualitative results indicate that the proposed severity scores are visually interpretable and demonstrate steady alignment with observed damage characteristics across heterogeneous real-world scenarios.</p>