Detection of damage in early stages has always been a primary concern in structural health monitoring. One of the damage detection methods that has attracted more attention today is signal based method. Among these methods, time-frequency methods have widely used in mechanical and aerospace engineering but they are not fully investigated for civil structures. In this study a combined damage detection method is introduced based on signal processing and data mining. 6 damage patterns are defined which is representative of slight to severe damages. First, by implementation of wavelet packet transform, signals are extracted and wavelet coefficients are calculated. For this reason, wavelet tree is established and the best tree is selected. Thus, noise can be extracted from the original signal. Then, dimension of data is reduced to one using dimensionality reduction. By calculating the energy index of the reduced signal damage is detected in the structure. Results show that the proposed algorithm is able to detect all the damage cases.