Antoniadis I., Gryllias K., Koukoulis I., Yiakopoulos C., Provatidis C.

Vibration analysis based damage identification in structures has been a subject of intensive research. As a result, a plethora of analytical, numerical and experimental techniques have been developed, involving frequency or mode-shape based methods, wavelet analysis, or higher order statistical signal processing. Parallel, a number of pattern recognition and classification approaches have been used, such as neural networks, either as standalone procedures, or in combination to the above techniques. In this paper, a method for crack detection in structures is presented. As a first step, using the Proper Orthogonal Decomposition (POD) method, a number of Proper Orthogonal Modes (POMs) of a crack cantilever beam are estimated. Proper Orthogonal Decomposition, closely related to Principal Component Analysis, is a powerful and elegant method of data analysis aimed at obtaining low-dimensional approximate descriptions of high-dimensional processes and eliminating noise effects. Then, the POMs are further processed, using basic morphological operators (dilation, erosion, opening, closing). The basic concept of morphological processing is to modify the shape of an object, equivalently considered as a set, by transforming it through its interaction with another object, called the structuring element. Using appropriate morphological processing, the position and the depth of the crack can be estimated by isolating the local sudden change effect of the crack on the space variation of each POM. The influence of the type of the structuring element and of the noise level to the accuracy of the results is examined on cantilever beams, containing a transverse surface opening-closing crack at different lengths and of different depths. The results obtained verify the effectiveness of the method.