Towards robust statistical damage localization via model-based sensitivity clustering

摘要: Abstract Damage diagnosis is a fundamental task for structural health monitoring (SHM). With the statistical sensitivity-based damage localization approach, residual vector computed from vibration measurements in reference and damaged state. The analyzed statistically hypothesis tests with respect to change directions defined by sensitivities of parameters associated elements finite element (FE) model investigated structure. If test parameter reacts, then respective structure indicated as damaged. This approach offers very generic theoretically sound framework analyze parametric changes systems, takes into account intrinsic uncertainty related measurement data. Depending on definition parameterization, simple computation statistics directly data system, without need system identification. Since an FE used, applicable arbitrary structures, while no updating required therefore requirements accuracy are less strict. While theoretical has been developed previously, it lacked robustness so far application real structures. purpose this paper development working method that complex To achieve goal, robust sensitivity revisited more precision thanks reduced modal truncation errors, adequate clustering proposed case high-dimensional parameterization Furthermore, several properties proven. Finally, shown first time experimental localization, namely ambient 3D steel frame at University British Columbia.