PSI - Issue 37

Tomasz Rogala et al. / Procedia Structural Integrity 37 (2022) 187–194 / Structural Integrity Procedia 00 (2019) 000 – 000

188

2

The classification of damage in composite structures based on NDT results was studied by numerous researchers using various approaches and techniques. Numerous studies were focused on the classification of characteristic responses of damage sites during the excitation of a tested structure, see e.g. (Zhou et al., 2009; Larossa et al., 2014). However, the other studies are focused on classification of damage based on its morphology. Jac Fredo et al. (2019; 2021) proposed an approach of classification of surface damage after visual inspection using Zernike moments and support vector machines and other geometrical descriptors, Usamentiaga et al. (2013) applied the approach based on statistical features retrieved from thermal responses of damage in infrared images of damaged composite structures, while Hauffe et al. (2020) presented the classification approach of damage using ultrasonic C-Scans and analyzed the geometrical featured detected in these scans. However, using X-ray computed tomography (XCT), it is possible to obtain results in the form of a 3D array with high spatial resolution, which is excellent source data for precise damage classification. Numerous studies focused on damage classification based on morphology of damaged regions were based on XCT testing results. For example, Sammons et al. (2016) used segmentation of XCT scans, which was then used as an input to a convolutional neural network (CNN) to classify delamination, while Bull et al. (2013) classified the types of damage in composite structure after impact loading based on segmentation of XCT scans and the orientation of the resulting features. Previous studies of the last author of this paper include classification based on geometrical properties of the damage sites as well as their orientation using XCT scans processed with wavelet and Hough transforms (Katunin et al., 2015; Katunin and Wronkowicz, 2015), which determined a promising direction of further research studies. The following study is based on NDT results obtained from glass fiber-reinforced polymeric (GFRP) composite specimens subjected to fatigue loading using XCT. The initially pre-processed tomograms using the developed image processing algorithm consisted of features that represent delamination, cracks, and artifacts resulting from XCT scanning and processing procedures (Katunin and Wronkowicz, 2017). One of the main difficulties of this study is to distinguish mentioned types of damage, even in the cases, when they are merged, and determine a set of features sensitive to their morphology. The authors considered numerous features, both 2D and 3D, including statistical, morphometric, and other, to evaluate their performance in distinguishability of predefined damage types. Further, the pre-selected features were used to construct an input set of parameters to a CNN, being adapted to the investigated problem for automatic classification of damage types. The proposed classification approach might be helpful as the decision supporting tool for NDT inspectors as well as can be used as an input to numerical models for the evaluation of a structural residual The specimens investigated within this study were the glass fiber reinforced polymeric (GFRP) composite coupons manufactured and supplied by Izo-Erg S.A. (Gliwice, Poland). These coupons with the dimensions of 100×10×2.5 mm were loaded in a fatigue bending testing mode until reaching a specific value of the self-heating temperature resulting from mechanical energy dissipation due to viscoelastic nature of the matrix of the considered GFRP composite material. The details on performing fatigue tests can be found in (Katunin and Wronkowicz, 2017). As a result of loading, the mechanical degradation was observed in the tested coupons in the form of cracks and delamination. Depending on the observed maximal self-heating temperature the level of degradation was different, which was reflected in the amount of the observed damaged regions. After fatigue testing, the coupons were inspected using XCT, which made it possible to map both external and internal damaged regions in the form of a 3D array with high spatial resolution (size of a single voxel was 0.025 3 mm 3 ). The scanning was performed the Phoenix v|tome|x m 300 XCT scanner within the region of interest (ROI) of 10×10×2.5 mm, which was limited to these dimensions due to absence of damage in surrounding regions. The details on scanning procedure can be found in (Katunin and Wronkowicz, 2017). life of composite elements. 2. Materials and methods