Data-driven models based on deep learning revolutionized classical computer vision tasks and have recently made their way into materials science as well. However, the absence of domain knowledge in their inherent design significantly hinders their understanding. Nevertheless, explainability is crucial to raise acceptance among experts and to justify their use in safety-relevant applications such as aircraft component design, service, and inspection.

In this work, we train convolutional neural networks (CNNs) for crack tip detection in fatigue crack growth experiments using full-field displacement data obtained by digital image correlation. For this, we introduce the novel architecture ParallelNets – a network which combines segmentation and regression of the crack tip coordinates – and compare it with a classical U-Net-based architecture. Aiming for explainability, we use the Grad-CAM interpretability method to visualize the neural attention of several models. The attention heatmaps show that ParallelNets learns to focus on physically relevant areas like the crack tip field, which explains its superior performance in terms of accuracy, robustness, and stability.