Ductile fracture occurs with large plastic deformation prior to fracture and is primarily characterized by voids of varying sizes. In this work, a new approach based on machine learning is proposed for the characterization of ductile fractures. The method is based on the application of a pre-processing of the fractographies, followed by a processing that includes the generation of a mask through a mathematical morphology, a classification is made according to their size in micrometers to later make a histogram of the size of the cavity in relation respect to its frequency of repetition in each fractography. Besides, the histogram is then normalized to generate a Probability Density Function (PDF). Based on the PDF, a third-order statistical metric known as Fractography Skewness (FS) is used to evaluate fracture susceptibility and is related to the Degree of Fracture Susceptibility (DFS). If the FS is negative (indicating a left-skewed PDF), the fractography is characterized by large voids and is assigned a DFS of 1, indicating low susceptibility. Conversely, a positive FS (indicating a right-skewed PDF) indicates the presence of small voids, resulting in a DFS of 3, indicating high susceptibility. The proposed quantitative methodology demonstrates a higher accuracy than the most advanced current techniques based on deep learning and the subjectivity of the specialist.
Read full abstract