The damage detection method using the extended Kalman filter(EKF) technique has been continuously used since EKF can estimation the responses of the damaged building structure and the stiffness of the structure. However, in the use of EKF, the requirement of setting the initial paramters P, Q, and R has caused the divergence and instability of the state vector, and various researches have been conducted to determine theses parameters. In this paper, adaptive extended Kalman filter(AEKF) method is proposed to solve the problem of setting the values of P, Q, and R, which are important parameters determining the convergence performance of the EKF state vector. By using the AEKF method proposed in this study, the P, Q, and R parameters are updated every k steps. The proposed algorithm is applied for the estimation of stiffness and the damage detection of 3-DOF problem. Based of the verification, it can be found that the selection process for the values of P, Q, and R can improve the convergence performance of EKF.

Recently, as the awareness of safety has become more important, studies on damage assessment techniques for building structures have been actively conducted. The damage of the building structure is caused by the decrease of the stiffness which is inherent dynamic characteristic of the structural system, and the decrease of stiffness acts as a direct variable connected to the collapse of the structure. there have been developed techniques for estimating the inherent dynamics of a structure to identify and evaluate damage to the structure. In this study, we estimate the layer mass due to the modeling error through the optimization algorithm, Genetic Algorithm, and use the optimization algorithm GA to optimize the error covariance matrix, system noise and measured noise covariance matrix We propose an optimal state estimation algorithm. The objective function of the GA algorithm is obtained by the residual which is the difference between the measured values obtained from the EKF calculation and the values obtained from the system model. We verified the feasibility of the algorithm through a 4-DOF system.