Numerical Methods On Structural Damage Identification

System health monitoring of structures is important not only for conducting safe operation but also maintaining system performance. Three identification algorithms for assessing structural damages according to the test data of dynamic and static responses have been developed in this thesis.For the practical applications of damage detecting, it is more attractive to avoid employing too much information of the analytical models. A two-stage identification algorithm for the assessment of structural damages using the modal test data has been developed. An important characteristic in the approach is that the employment of the global numerical models (e.g. FEM model) and some important information (e.g. the Young's modulus) of structures are avoided to the great extent. As the first step of the damage identification, an algorithm for the detection of damage location is proposed, which uses the concept of subspace rotation algorithm. Furthermore, a quadratic programming model is set up to predict the damage extent.The second new algorithm is to identify the structural damage using flexibility matrices, which can be constructed from several low orders of natural vibration frequencies and modes more accurately than the stiffness matrices. The damage location in truss and frame structures is detected from the change of the axial strain component or the bending curvature in each truss member or beam element under virtual loads before and after damage. Then the damaged elements and the corresponding damage extents can be identified at the same time. This algorithm is not dependent on the analytical models of the structure as well.Considering the difficulty in accurate measuring the mode shapes of civil structures, an identification method using the changes in natural frequencies and the complementary static test data is presented. To locate damage in the structure, the Damage Signature Matching (DSM) technique is improved through a properdefinition of Measured Damage Signatures (MDS) and Predicted Damage Signatures (PDS). The effect of damage severity can be eliminated effectively as the result that the first-order approximation of changes in static deformation and natural frequencies are employed jointly in the damage signatures. Then the damage location can be detected successfully by matching MDS with PDS. After obtaining the possible damage location, an iterative estimation scheme for solving non-linear optimization programming problems based on the quadratic programming technique, is proposed to predict the damage extent. A remarkable characteristic of the present approach is that it can be directly applied in the cases of incomplete measured data.To demonstrate the capability of the proposed approaches, examples of planar truss or frame structures are analyzed numerically. Furthermore, the experimental data from the vibration test and the static experiment of a beam with two-fixed ends are used directly in the new approaches. The final results show that the present algorithms perform very well in spite of the little structural information and measurement inaccuracies.