| Abstract Scope |
Fracture surfaces include a wealth of information on the fracture mode, site of crack initiation and grain morphology. Unfortunately, engineers often overlook those details, and only take the quantitative number from the experiment for data analysis (e.g., impact energy, fracture toughness, yield strength). Image-based machine learning techniques, however, can be used to predict material properties and identify critical components of the fracture surface. In this study, K1C fracture toughness specimens have been imaged using a standard two-dimensional DSLR camera. These specimens were all from the same lot of material, but had different cooling rates based upon their location in the casting. A convolutional neural network has been trained to predict their ductile or brittle fracture behavior. Since the sample size is relatively small but the image sizes are large, a discussion will also be included on data augmentation by cropping the fracture surface into discrete parts to train the model. |