PSI - Issue 37

Kafayat Eniola Hazzan et al. / Procedia Structural Integrity 37 (2022) 274–281 Hazzan and Pacella/ Structural Integrity Procedia 00 (2019) 000 – 000

276

3

of SEM images, specific combination of selected parameters caused different extents of surface detects including cracks, splatter, porosity, and balling. 2.2. Development of a novel identification technique MATLAB ® (R2020b v 9.9) was used to develop a manual identification method (IM) for cracks based on image processing from SEM images. The process is referred to as manual as the term automated or automatic typically refers to the use of artificial intelligence (AI) (Wu and Liu, 2021), (Petrich et al., 2017) or convolutional neural networks (CNN) (Kim et al., 2019). SEM images (225 μ m x 175 μ m) in the TIF file format were exported from the TM3030 machine and imported into the MATLAB workspace as uint8 (8-bit) containing 1536 x 2048 pixels. Fig 1. outlines the identification method developed in the study using features and functions in the MATLAB Image Processing Toolbox™. Fig 2. shows output examples at each stage.

Fig. 1. Crack identification steps.

Fig. 2. a) SEM of WC processed by 0.07 J/cm 2 , 100 kHz, 500 mm/s, 0.02 mm. b) Binary image of SEM input. c) Noise filtration. d) Crack identification after pixel analysis.

SEM images were converted to binary images (BSEM) using the Sobel algorithm (Harris and Stephens, 2013). This converts the SEM images into a logical array of zeros and ones (Fig 2. b). Noise was removed from the binary image using the b wareafilt function (Fig 2. c). This allowed objects that are too small and too large to be removed. This made the remaining features (i.e., cracks) independent from the background. To avoid loss of resolution in the remaining regions, the pixel connectivity input variable was adapted to include pixels that touch in vertical, horizontal and the diagonal direction.