Extracting ductile cast iron microstructure parameters from fracture surfaces: A deep learning based instance segmentation approach

Abstract

This study investigates the deep-learning based microstructural analysis from SEM images of ductile cast iron fracture surfaces. A Mask R-CNN model was trained, achieving 70% precision and 75% recall in graphite particle detection. Combined with a fracture surface reconstruction using the.

4-quadrant backscattered electron signal, key parameters, including the particle size, shape and distance were extracted accurately. Compared to micrograph analysis, following probabilistic simulations showed the impact of the higher microstructural variance for the fracture surfaces on crack initiation, leading to higher scatter and elevated crack resistance curves. This highlights the potential of deep-learning based analysis for comprehensive microstructural characterization.