| Abstract: | A reliable computational model is necessary for evaluating the state and predicting the future performance of existing structures, especially after exposure to damaging effects such as an earthquake. A major problem with the existing iterative‐based model updating methods is that the search might be trapped in local optima. The genetic algorithms (GAs) offer a desirable alternative because of their ability in performing a robust search for the global optimal solution. This paper presents a GA‐based model updating approach using a real‐coding scheme for global model updating based on dynamic measurement data. An eigensensitivity method is employed to further fine‐tune the GA updated results in case the sensitivity problem arises due to restricted measurement information. The application on shear‐type frames reveals that with a limited amount of modal data, namely the lowest three natural frequencies and the first mode shape, it is possible to achieve satisfactory updating by the GA alone for cases involving a limited number of parameters (storey stiffness herein). With the incorporation of the eigensensitivity algorithm, the updating capability is extended to a sufficiently large number of parameters. In case the modal data contain errors, the GA is also shown to be able to update the model to a satisfactory accuracy, provided the required amount of modal data is available. An example is given in which a 6‐DOF stick model for an actual six‐storey RC frame is updated using the measured dynamic properties. The effectiveness of the updating is evaluated by comparing the measured and predicted seismic response using the updated model. Copyright © 2005 John Wiley & Sons, Ltd. |
