عنوان مقاله [English]
Detection of damage in structures is a main concern in many areas of civil engineering. In recent years, developments in sensor technology and signal processing techniques has attracted researches toward signal-based methods. In this study a damage detection algorithm is proposed for structures based on signal processing and dimensionality reduction. IASC-ASCE health monitoring benchmark structure is utilized for analysis. This building is a 12-DOF frame with braces at each level. Six damage cases are defined which cover a range of extensive to slight damages. It is assumed that each story is instrumented by four sensors. A random excitation is applied on the structure to simulate earthquake load. 10\% noise is considered to model real condition signals. First, wavelet packet decomposition (WPD) is utilized to decompose sensor outputs. In comparison to discrete wavelet transform (DWT), WPD decomposes both approximations and coefficients to form a full tree. Each sensor output is extracted separately. To separate noise from the actual vibration, signal is decomposed and then reconstructed. Best tree is evaluated to estimate
decomposition levels using the Shannon entropy criterion. Since sensor outputs are different, best tree for each signal differs. Therefore extraction is made on one level. Coefficients of the first level are considered as the signal features. Next, several wavelet functions are examined to find the most appropriate one for this study. Twelve functions are compared and Bior 3.3 is selected as the best wavelet function. In the second step, dimensionality reduction is applied to reduce amount of data. Principal component analysis (PCA) is used to reduce data. PCA is an effective technique that has proved its ability when the amount of data is large. It maps linearly correlated data points called `principal components' based on the highest variance. By using PCA the wavelet coefficients of sixteen sensors are reduced to only one signal. Energy of the reduced coefficients is considered as a damage index. Results show that the proposed algorithm has a good capacity to identify damage in the investigated structure.