PSI - Issue 37

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

189

3

3. Data preparation and evaluation 3.1. Preprocessing of XCT scans

The acquired tomograms were pre-processed in the tomograph-dedicated software VGStudio MAX (Volume Graphics, Heidelberg, Germany) to improve the contrast and sharpness of the observed damage in tested coupons as well as to export tomograms in the form of sequences of 2D slices. Next, the sequences were imported to Matlab ® (MathWorks, Natick, MA, USA) for further pre-processing steps, which aimed on segmentation of the particular 2D slices in the sequences to separate the damage sites from the texture of healthy regions a composite. The custom-developed segmentation procedure for this purpose based on the adaptive thresholding was described in detail in (Katunin and Wronkowicz, 2017). In the last step of pre-processing, initial filtering of the very small (classified as measurement noise) and large objects (classified as edges of the tested coupons) and labelling the remaining objects was performed. The application of this procedure to the selected sequence of 2D slices of a tomogram is presented in Fig. 1.

Fig. 1. A 3D view of a tomogram for the selected coupon (a) before; (b) after the last step of pre-processing.

3.2. Determination of features set and transformation For each object resulting from the degradation process, which was a result of cyclic loading and the related self heating effect in the tested samples, a set of features was determined. In the first stage, this set included only the features describing the geometry of the analyzed objects in three-dimensional space. In order to be able to use this method in the future, attention has been focused on scalar features that have at least one resistance to changes in the scale, position and rotation of the described object. Thirty six scalar feature definitions were defined for this purpose and used to describe each of the objects. The number of objects from different samples with varying degrees of degeneracy amounted to over 17 thousands. The features used include: simple features of 3D geometry such as volume as well as measures of similarity e.g. sphericity e.g. Cox-Budhu sphericity (2008), or statistical moments of analysed objects. Figure 2 shows exemplary degradation objects of various types. Fig. 2a shows an exemplary object labeled as crack, Fig. 2b shows an object of the delamination. It should be noted that there are also objects in the dataset that combine crack and delamination damage and appear most often in samples with a much more degenerate structure. Objects of this type, which an example is shown in Fig. 2c, were assigned the damage class that is dominant in a given object. During the research, attention was paid to the great potential of the definition of morphological features intended for the description of two-dimensional images. In order to extend the set of features, a transformation has been defined that allows a relatively lossless transformation of three-dimensional objects into two-dimensional objects while maintaining the most important geometric properties. Due to the fact that experts in the field of the mechanics of degradation of composite materials are able to more or less describe the form of damage resulting from crack or delamination (Katunin and Wronkowicz, 2017), the class for objects with crack damage was initially determined.