Damage detection of building structure based on vibration data and hysteretic model

This paper presents a novel approach for damage detection in building structures by using the dissipated energy. In this sense, the hysteretic Bouc-Wen model is introduced as a useful tool for describing the degrading energy, which is directly related to the stiffness loss. Since, parameters and states of this model are unknown, we employ a nonlinear system identification algorithm based on Convolutional Neural Network (CNN) to avoid estimate simultaneously the states and parameters of the model. The used CNN have the sparse connectivity, which ensures that the strong response can be detected by convolution filters. In addition, the shared weights of the CNN reduce the the training complexity and the number of its parameters because the same weights are applied to all inputs. Therefore, the CNN can detect features no matter where they are on the vibration data, also reducing the training complexity. The use of this tool avoids using an adaptive observer, which unlike CNN, the algorithm’s complexity increases with the number of unknown parameters and states. Moreover, the adaptive observer can not guarantee convergence in presence of measurement noise. Experimental results confirmed that the proposed method is promising for real applications.