Automatic identification of brittle, elongated and equiaxed ductile fracture modes in weld joints through machine learning

Identification of the fracture mechanism in weld joints is essential for ensuring weld quality, reduce weld defects, along with enhancing the welding process.Therefore, an attempt is made to use image processing techniques such as wavelet transformation, Gray-Level Co-occurrence Matrix (GLCM), and Local Binary Pattern (LBP) to develop an automatic identification system for brittle, elongated, and equiaxed ductile fracture modes in weld joints.The GLCM technique employ Haralick functions, while the LBP and wavelet transform techniques use histograms and Gabor filters, respectively for extracting features in the fracture images.Classification based on textural features (granular or fiberous) was performed using support vector machine.LBP achieved superior accuracy of 96%, followed by GLCM.Further research could explore real-time implementation and expand the dataset to enhance the system's robustness and applicability.