Machine Learning Classifiers for Surface Crack Detection in Fracture Experiments

Correctly determining the onset of fracture is crucial when performing mechanical experiments. Commonly carried out by visual inspection, here an image-based machine learning approach proposed to classify cracked and un-cracked specimens. It yields potential objectify automate crack detection, thereby removing sources uncertainty error from post-processing More than 30’000 speckle-pattern images obtained 77 experiments on three specimen geometries are evaluated. They comprise uniaxial tension, notched tension as well axisymmetric V-bending Statistical texture features extracted all images. include both first-order (variance, skewness, kurtosis) higher-order statistical features, i.e. Haralick features. The discriminatory power information evaluated based Fisher's Discriminant Ratio feature correlations identified quantified. Image subsets high used parse neural network architectures different complexities simple perceptron feed-forward cascade networks. found that a small subset investigated highly significant for Using this in conjunction with multi-layer, non-linear low complexity classification accuracies order 99% obtained. At same time, it shown linear classifiers not sufficient robustly distinguish state specimens, even used. Graphical Abstract: Download : high-res image (198KB)Download full-size