PSI - Issue 57

Haelie Egbert et al. / Procedia Structural Integrity 57 (2024) 179–190 Haelie Egbert et al. / Structural Integrity Procedia 00 (2019) 000 – 000

184

6

recording speed to provide exactly 25 images per loading cycle. This ensured that the phasing of each image with respect to the loading remained constant between each cycle. In addition, 25 images per cycle allowed for a sufficient resolution of images such that an image was taken reasonably close to the instant of maximum load in the cycle. With a 32 GB camera buffer memory, this resulted in videos 3.45 minutes long corresponding to the last 6,207 cycles of the test, sufficient to capture the visible growth of the crack on the tooth face. Frames corresponding to the maximum load in each loading cycle were stored as 8-bit grayscale, all other frames were discarded. As crack opening is largest under the highest load, these were the only images of relevance for a surface crack length measurement. Figure 4 shows how the image acquisition sequences with the load data.

Figure 4. STBF loading, camera capture and image archival sequencing and phasing

3.3 Digital Image Correlation After performing the image acquisition and sequencing described, images were imported into MATLAB. Each loaded frame corresponded to a matrix of image height, width, and frame number, with the height and width corresponding to the chosen Nova S6 camera resolution of 334 and 386 pixels, respectively. The fatigue test cycle numbercorresponding to each image was known. The 8-bit grayscale images were converted to binary scale (black or white) using the MATLAB command “imbinarize”, such that pixels over a ce rtain threshold were purely black and those below it purely white. This command uses Otsu’s method, in which the threshold is chosen for each image in such a way as to minimize the intra-class variance of pixels. Noise existed in all frames, due to imperfections on the gear tooth. These imperfections or poorly exposed pixels within the tooth face due to lighting restraints resulted in black dots within the rest of the white tooth face, as shown in Figure 5(a). The sensitivity of the binary conversion command needed to be manually adjusted from one image to produce an acceptable binary image showing the tooth profile. By changing the sensitivity, the user increases or decreases the bias of the program towards pixels becoming the foreground or background. An adaptive local threshold was then maintained based upon this sensitivity bias. A result of this change is shown in Figure 5(b).