Structural damage diagnosis and prognosis based on in-service monitoring data are critical for condition-based maintenance and flight safety assurance. However, most existing studies focus on individual aircraft, overlooking inter-aircraft structural similarity within a fleet. This paper proposes a fleet-level digital twin framework for structural damage diagnosis and prognosis by leveraging similarity among aircraft. A physics-decoded variational neural network (PDVNN) is developed to efficiently extract and quantify structural features and damage states. A Copula-based probabilistic model is further introduced to capture joint dependencies among damage states across the fleet, enabling collaborative updating of damage estimations. Case studies on representative aircraft panels demonstrate that the proposed approach significantly improves diagnostic accuracy and prediction consistency at the fleet level, thus providing a foundation for fleet-scale digital twins and condition-based maintenance strategies.