Damage identification in plate using wavelet transform and artificial neural network

Abstract

Detection of damage at early stage is necessary to prevent catastrophic collapse of civil structure. The quantification of small damage has however, not received the much needed attention from researchers. Amongst various non-destructive methods studied by researchers to identify structural defects are Wavelet Transform (WT) and Artificial Neural Network (ANN). In this paper, a technique is proposed to quantify damage by combining WT and ANN. The study is divided into two phases. The first phase involves detection and location of damage in a plate numerical model by WT decomposition of the mode shape difference. Due to difficulties in obtaining higher mode shapes in practice, the difference of the first mode shapes of the damaged and undamaged plates are applied. After obtaining the damage location, the coordinates of the damage location and the mean values of the obtained WT moduli are applied as input to the designed neural network. The output of the ANN is the severity of damage in the plate model. This method is demonstrated by using a numerical square steel model with all the four sides fixed. The results indicate the ease of the method and that reliable damage identification can be obtained by combining ANN and WT.