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

275

2

The presence of cracks is dependent on both WC composition (Liang et al., 2018) and laser parameters (Eberle and Wegener, 2014), (Calderón Urbina et al., 2013). Currently, the analysis of micro-cracks in laser engineered WC is performed manually via scanning electron microscopy (SEM) to identify the types of cracks present (Pacella, 2019) and to evaluate if the presence of identified cracks will pose a significant risk for the part in use. As WC has little to no plastic deformation once a crack is present, under load, micro-cracks will grow into macro-cracks and cause failure (Smith, 2008). Thus, the ability to identify cracks quickly and objectively after laser processing is pivotal in the prediction of mechanical performance. Advances in image processing and vision-based defection recognition has allowed for greater flexibility in different components and systems (Wang et al., 2020), (Mohan and Poobal, 2018). In industry this has mainly focused on macro-cracks in the construction industry applied to concrete ((Yang et al., 2020), (Bergs et al., 2020), (Kim et al., 2019)), pavements ((Ranjan et al., 2016), (Zou et al., 2012)) and other components e.g., pipelines (Koulali et al., 2021). There are fewer studies where image processing has been employed to identify micro-cracks. Vidal et al. (2016) demonstrated the use of image processing to analyse micro-cracks and corrosion in SEM from chromium electrodeposits. Jin et al. (2021) characterised the texture and pattern of micro-cracks on the sheds of composite insulator used in unmanned aerial vehicles. Schöttl et al. (2020) also used image processing on micro-computed tomography images to characterise cracks in discontinuous fibre-reinforced composites during tensile loading. The application of these methods provides an objective and cost-effective method for damage review.

Nomenclature AI

Artificial intelligence Binary SEM image

BSEM

Bwareafilt MATLAB function to extract objects from binary image by size Bwboundaries MATLAB function to trace region boundaries in a binary image Convhull

MATLAB function to create a matrix of 2D points and calculates the area

CNN CPU Edge GPU IM Plot SEM eps

Convolutional neural network Central processing unit Floating point relative accuracy Graphics processing unit Identification method output S canning electron microscopy Tagged Image File format

MATLAB function that returns binary image of input image

MATLAB function to plot a set a point in graphical space

TIF uint WC

Unsigned integer Tungsten carbide

2. Methodology 2.1. Laser processing and SEM image acquisition

A WC- Co12 % (Z57 C5) blank (mean grain size of 4 μm) was selected to conduct the laser surface engineering experiments. Areas of 1 mm 2 were laser processed using a 1060 nm, 240 ns, 70 W fibre laser. Fluences in the range of 0.050 – 0.200 J/cm 2 were used, lower than the WC ablation threshold of 0.45 J/cm 2 in the ns regime (Dumitru et al., 2002), the laser mechanisms involved are heating and partial binder ablation. The combination of these mechanisms can cause cracking as highlighted by Marimuthu et al.(2020) and See et al. (2017). Other parameter settings include 5 – 100 kHz, 500 – 1500 mm/s, 0.02 – 0.06 mm for frequency, feed speed and hatch distance, respectively. A TM3030 Hitachi backscatter SEM was used to view the surface topography and obtain SEM images. Multiple scanning electron microscope (SEM) images were taken from each processed square. By visual inspection