Damage Identification Methods Based On Neural Networks And Dynamic Characteristics Of Structure Under Control

Structural damage identification using neural networks and data of dynamic characteristics of structures under control were studied in the thesis. Application of direct damage identification methods for complex structures is difficult as the numbers of components and degrees of freedom (DOFs) in these structures are too large. A multilevel structural damage identification approach is adopted in the study. The structure is divided into several sub-domains and damaged sub-domain is identified using a probabilistic neural network (PNN), then the multiple damage location assurance criterion (MDLAC) or a radial basis function neural network (RBFNN) is employed to identify the location and severity of the specific damaged element in the damaged sub-domain. Damage indicators are very important for the structural damage identification. In the study, the state feedback control method is employed to place the poles of the structure intentionally so that the damage indictors are more sensitive to the damage of structure under control. Locations of control forces in the structure have effects on the accuracy of damage identification and therefore their effects are also investigated in the study. A procedure for multilevel structural identification based on a PNN and MDLAC is described and applied to the damage detection of a cantilever structure and a single-span simply supported beam. Numerical results show that the damaged sub-domain can be well identified using the probabilistic neural network with the dynamic data of the structure under control and diagnosis accuracy using multilevel identification strategy is higher than that of direct identification strategy. A procedure for multilevel structural identification based on a PNN and a RBFNN is also presented and the effectiveness and efficiency of the proposed method is demonstrated by applying the method to a cantilever structure with damage and a three-span continuous beam structure with damage. Numerical results show that the damage location and severity can be identified efficiently and the errors of damage diagnosis are small.