Annotated digital image correlation displacement fields from fatigue crack growth experiments

Abstract

We present a curated dataset of planar displacement fields from eight fatigue crack growth experiments obtained via full-field digital image correlation (DIC). The dataset covers multiple aerospace-grade aluminium alloys, specimen geometries, material orientations, and load configurations, providing a diverse experimental basis for data-driven fracture mechanics research. Crack tip locations are consistently annotated using an iterative correction procedure applied to all measurements, and fracture mechanical descriptors like stress-intensity factors are provided as additional labels. The dataset comprises 8,794 unique experimentally observed displacement fields and a total of 70,352 supervised samples generated through standardized interpolation and augmentation. DIC data is provided as uniformly interpolated displacement grids at three standardized resolutions 28 x 28, 64 x 64, and 128 x 128 pixels, each available in three dataset sizes to support scalable use cases ranging from educational applications to high-capacity model development. Accompanying metadata and a Python interface facilitate filtering, loading, and integration into reproducible machine learning and fracture mechanics workflows.

Figures (6)

• Figure 1: Overview of the CrackMNIST dataset creation workflow: (a) Experimental setup for fatigue crack growth testing Strohmann24 with I: full-field 3D DIC by a commercial Zeiss Aramis system, II: high-resolution DIC option using a KUKA collaborative robot with a global shutter camera (not used in this dataset release), and III: material specimen (here: MT160). (b) Crack tip annotation process showing initial estimate (yellow cross), refined position after iterative correction (green cross), and the symbolic regression correction formulas applied to Williams-series coefficients. (c) Examples of processed displacement field samples at $28\times28$ pixel resolution with crack tip labels (black crosses) as provided in the CrackMNIST dataset. The dataset is available at pixel resolutions, $p \in \{28, 64, 128\}$, and with various sizes, $s\in \{S,M,L\}$. (d) Data samples consist of input displacement field $(u_x, u_y)$ with corresponding target crack tip masks and stress intensity factors. For each sample metadata such as the applied external force and specimen and material type are provided.
• Figure 2: Distribution of samples across applied normalized loads. Counts are based on non-augmented data.
• Figure 3: CrackMNIST DIC data sample interpolated to different resolutions. Top: $x$-displacement in $mm$. Bottom: $y$-displacement in $mm$. Columns show the same sample at various resolutions.
• Figure 4: Three random CrackMNIST samples at a resolution of $28 \times 28$ pixels. Each input consists of two channels representing the in-plane displacement components, $u_x$ and $u_y$, while the corresponding binary segmentation masks indicate the crack tip location, shown as white pixels. Additional annotations of the stress-intensity factors, $K_I, K_{II}$, both in $\mathrm{MPa}\sqrt{m}$, and the $T$-stress in $\mathrm{MPa}$ are given.
• Figure 5: Prediction failure map for L-28. Within each split, the mosaic shows crack tip occurrence (mask), false negative (FN) pixel and false positive (FP) pixel maps. The maps are normalized over per-pixel absolute crack tip occurrences and over trials and to highlight the spacial density distribution of correctly and incorrectly detected crack tips.