Physical-based parametrization and local damage identification for frame-type structures using response sensitivity approach in time domain

摘要

Structures are usually subjected to some specific local damages or weaknesses, which cannot be explained by uniformly distributed stiffness or mass changes, especially when the local damages or weaknesses have slight influences on global natural periods or vibration modes. Furthermore, large amounts of measurement positions are in need to capture the damage location and severity in traditional damage identification, and results easily depend on the initial estimates of the unknown parameters for the iterative update. To circumvent the above problems, a novel framework is proposed, involving strategies for physical-based parametrization and local damage identification for frame-type structures. The novelty of this research lies in the careful integration of the physical-based parameterization and the identification techniques to robustly quantify commonly encountered local damages, namely, the weakness in exposed column base and partially fractured beam end. In this framework, the guidance for the application of physical-based parametrization is provided, and thus, the above-mentioned local damages of structures could be parametrized in an appropriate and convenient way. The response sensitivity approach is employed to derive the parameters by directly using time domain data. Derivatives of the stiffness matrix with respective to the newly introduced physical parameters are directly proposed so that no finite difference method is involved. Furthermore, trust-region restriction is adopted to enhance the efficiency and stability of the identification approach when facing strong convergence difficulty or way-out initial estimates. Two case studies are presented to validate the effectiveness and accuracy of the proposed framework. The shell element-based models with intentional damage and weakness are built in ABAQUS to provide the data source. Two corresponding beam element-based models are developed in MATLAB as physical-based parametrized models. Results show that damages are accurately and efficiently estimated by the proposed framework, and the original dynamic responses are well reproduced by the identified models. Comparisons to the results by Bayesian estimation are also discussed.