Damage detection and classification in composite structure after water-jet cutting using computed tomography and wavelet analysis

The proposed method of processing of X-ray Computed Tomography (CT) data is based on its three-dimensional (3D) wavelet analysis, which allows for detection and localization of sudden changes in voxels of CT data array. The applied wavelet transform can be considered as filtering using sets of low-pass and high-pass filters over directions of a domain. The internal damage, which initially has different spectrum than the healthy regions of a structure, are emphasized after the wavelet analysis and depending on its intensity in the CT data array the wavelet coefficients become different for them. The tests were performed on the carbon fibre reinforced composite plate. The water-jet method was used for cutting of a circular hole inside. During the cutting process several delaminations in various layers occurred. Moreover, during the manufacturing process several air pockets in the matrix appeared. Application of the proposed processing algorithm allows not only for detection and localization of internal damage but also its classification by a type. It is possible due to analysing the magnitude of wavelet coefficients after wavelet-based decomposition and characteristic shape and dimensions of various types of damage. This allows for automation of the examination process. which might be improperly classified as a structural damage. Several studies on extraction and classification of damage features from 3D arrays of reconstructed CT scans have been performed. The earliest studies (Chu & Lee 2004, Chu et al. 2004) were based on a thresholding performed using Otsu’s method extended to the 3D space. This technique solves a problem partially—the data still contain a lot of measurement noise and reconstruction artefacts. Other, more advanced algorithms of feature extraction described e.g. by Lontoc-Roy et al. (2006), are based on 3D neighbourhood analysis, however the procedure was performed on binarized input data, which results in loss of information. The postprocessing algorithm presented by Perret et al. (2007) is based on selection of upper and lower thresholds by the trial-and-error method. The more advanced noise reduction algorithm with use of anisotropic diffusion filter was proposed in (Heinzl et al. 2007), while in (Huang et al. 2003) the authors used 3D Gaussian filter for noise reduction. Another feature extraction algorithm was presented in (Kaestner et al. 2006), where the authors used non-linear diffusion filter to enhance the contrast and then the adaptive thresholding procedure for overall enhancement of measurement data and availability of features extraction from them. Several studies of feature extraction from CT scans were performed using wavelet analysis. The authors of (Roeding &