结合逆非线性主成分分析和极值理论的桥梁损伤检测

为提高环境和运营变化（Environmental and Operational Variations, EOV）影响下的桥梁损伤检测可靠性，结合逆非线性主成分分析（Inverse Nonlinear Principal Component Analysis, INLPCA）和极值理论提出一种新的桥梁损伤检方法。该方法采用INLPCA对桥梁损伤特征进行建模，利用不完备健康监测数据的估计误差和添加神经网络训练惩罚项控制INLPCA的非线性程度。采用INLPCA对损伤特征的重构误差和马氏平方距离（Mahalanobis Squared Distance, MSD）建立损伤指标（Damage Indicator, DI），最后基于DI的广义极值（Generalized Extreme Value, GEV）分布建立损伤检测阈值。以比利时KW51铁路桥和天津永和斜拉桥为例，验证所提方法的有效性。结果表明，所提方法能准确检测EOV影响下的桥梁损伤且对不同桥型和不同损伤特征均有良好的适用性。

Bridge Damage Detection Based on INLPCA and Extreme Value Theory

To improve the reliability of bridge damage detection under the influence of environmental and operational changes (EOV), a new bridge damage detection method is proposed by combining inverse nonlinear principal component analysis (INLPCA) and extreme value theory. This method uses INLPCA to model the damage features of bridges. The estimation error of incomplete health monitoring data and the addition of neural network training penalty terms are used to adjust the nonlinearity of INLPCA. Damage Indicator (DI) is established using INLPCA reconstruction error of damage features and Mahalanobis Squared Distance (MSD). Finally, a damage detection threshold is established based on the generalized extreme value (GEV) distribution of DI. Taking the KW51 railway bridge in Belgium and the Yonghe cable-stayed bridge in Tianjin as examples, the effectiveness of the proposed method was verified. The results indicate that the proposed method can accurately detect bridge damage under the influence of EOV and has good applicability to different bridge types and damage features.