针对金属搭接结构隐藏裂纹难检、传统风险评估高度依赖离线无损检测的问题，基于数字孪生框架提出了一种诊断-预测一体的动态风险评估方法。该方法以实测应变驱动物理-数据融合的裂纹扩展模型，并建立动态贝叶斯网络实现虚实闭环；通过CUSUM检测裂纹存在、KNN定位裂纹位置、MOGPR识别裂纹尺寸，结合DBN实时更新裂纹扩展参数C和m。在此基础上，采用蒙特卡洛仿真计算SFPOF。试验结果表明：CUSUM-SFPOF联合判据能及时实现裂纹预警，降低对高精度EIFS的依赖；随着监测数据累积，参数C和m标准差逐渐降低，寿命预测不确定性降低。该方法可实现金属搭接结构持续在线风险量化评估，为结构视情维护提供可靠支撑。

Hidden cracks in metallic lap-joint structures are difficult to detect, and conventional risk assessment methods rely heavily on offline nondestructive inspections. To address these challenges, an integrated diagnosis and prognosis dynamic risk assessment method within a digital twin framework has been proposed. The method employs an experimental strain driven physics and data fusion crack propagation model and implements a dynamic Bayesian network to close the loop between virtual and physical models. Crack initiation is detected using the CUSUM algorithm; crack location is determined by a KNN classifier; and crack size is estimated through a MOGPR model; subsequently, a DBN is employed to dynamically update the crack propagation parameters C and m in real time. Based on these updated parameters, the SFPOF is computed through Monte Carlo simulation. Experimental results demonstrate that the combined CUSUM and SFPOF criterion provides timely crack warnings and reduces dependence on high precision EIFS. As monitoring data accumulate, the standard deviations of C and m decrease, lowering uncertainty in life predictions. The proposed method enables continuous online quantitative risk evaluation of metallic lap-joint structures, offering reliable support for condition-based maintenance decisions.