Element-deletion-enhanced digital image correlation for automated crack detection and tracking in lattice materials

This paper presents a global digital image correlation framework that solves the correlation problem directly on the lattice mesh with automatic element deletion and data-driven damage detection, enabling robust, high-resolution tracking of crack initiation and propagation in architected materials where traditional continuum-based optical methods fail.

The Gist

Imagine you are trying to watch a delicate house of cards collapse. You want to see exactly which card falls first, how the structure crumbles, and where the cracks form. Now, imagine that house of cards is made of thousands of tiny, interconnected beams, and you are trying to film its destruction with a high-speed camera.

This is the challenge scientists face when studying lattice materials—engineered structures that look like complex honeycombs or scaffolding. These materials are incredibly strong yet lightweight, used in everything from airplane wings to medical implants. But when they break, they don't just crack like a piece of glass; individual "beams" snap one by one, creating a chaotic, jagged path of destruction.

The problem? The standard tools used to measure how things stretch and break (called Digital Image Correlation or DIC) are designed for solid, continuous objects like a rubber band or a metal bar. When you try to use them on a lattice, the software gets confused. It tries to force the broken pieces to stay connected in its calculations, leading to "ghost" data that looks like impossible, superhuman stretching. It's like trying to measure the speed of a car by assuming the wheels are still attached to the chassis even after they've fallen off.

The Solution: The "Digital Scissor"

The researchers in this paper invented a clever new way to fix this. They created a "smart" version of the measurement tool that acts like a pair of digital scissors.

Here is how their method works, broken down into simple steps:

1. The "Noise" Detective (The First Pass)
First, the computer watches the video of the material being pulled apart. It looks for tiny, sudden "glitches" in the image.

• The Analogy: Imagine you are listening to a quiet room. Most of the time, there is just a low hum (background noise). Suddenly, you hear a loud CRACK.
• The computer learns to ignore the low hum. It sets a rule: "If the image changes suddenly and sharply—like a loud crack—that's a failure!" It doesn't care why it cracked yet; it just knows a snap happened.

2. The "Digital Scissor" (The Second Pass)
Once the computer identifies a "crack" (a sudden glitch in the image), it performs a magic trick: it deletes that part of the digital map.

• The Analogy: Think of the lattice as a giant digital puzzle. When a piece breaks, the computer physically removes that puzzle piece from the screen. It then re-solves the puzzle using only the pieces that are still standing.
• By removing the broken pieces, the computer stops trying to calculate how they are stretching (which was impossible and wrong). Instead, it focuses entirely on the healthy, intact parts of the structure.

3. The Result: A Clear Picture of Destruction
Because the computer stops looking at the broken bits, the measurements become incredibly accurate.

• It can now track the crack tip (the very front of the breaking line) as it moves, frame by frame.
• It can count exactly how many "beams" have snapped.
• It can even tell you how much force it took to break each beam.

Why This Matters

The researchers tested this on 3D-printed triangular lattices. They found that their "Digital Scissor" method was spot-on.

• Accuracy: When they counted the broken beams on the computer, it matched almost perfectly with what they saw with their own eyes after the test.
• Prediction: They discovered that the "glitch" in the image happened at the exact same moment the machine recorded a drop in force. This means the computer can predict a failure the instant it happens.
• Flexibility: They even tested it on "imperfect" lattices (where they intentionally removed some beams to make it weaker). The method still worked, tracking the crack as it zig-zagged around the missing pieces.

The Big Takeaway

This paper introduces a way to watch complex, fragile structures break without the computer getting confused by the mess. By automatically "cutting out" the broken parts of the digital image, scientists can finally understand exactly how these advanced materials fail.

This is a huge step forward for engineers. If we know exactly how and where these materials break, we can design them to be stronger, safer, and more efficient. It's like teaching a computer to watch a building crumble and say, "Ah, I see exactly which brick fell first, and now I know how to build a better wall."

Technical Summary

1. Problem Statement

Architected materials, such as lattice structures, offer exceptional combinations of stiffness, strength, and toughness but are limited by a lack of understanding regarding how cracks initiate and propagate through their discrete architecture.

• The Challenge: Lattice failure is governed by highly localized deformations of slender beams. Standard optical measurement techniques like Digital Image Correlation (DIC), designed for continuum solids, struggle with these materials.
• Limitations of Existing DIC:
  • Continuity Assumption: Global DIC assumes a continuous displacement field. However, cracks introduce discontinuities, leading to unphysical strain calculations and noise when a crack propagates.
  • Scale Mismatch: Standard fracture mechanics-based DIC methods (e.g., using Williams series expansions) rely on continuum assumptions that fail in lattices where size effects are non-negligible.
  • Element Deletion Difficulty: While element deletion (removing damaged parts of the mesh) is a known concept, defining automated, physically meaningful criteria for deletion in slender beam structures has remained an unsolved challenge.

2. Methodology

The authors propose a global DIC framework tailored to lattice materials that integrates an automated element-deletion mechanism. The methodology involves a two-pass approach and a secondary mechanics-based validation.

A. Experimental Setup

• Specimens: 3D-printed compact tension specimens with triangular lattice topologies (both regular and imperfect with missing struts).
• Material: Gray V4 resin (Formlabs).
• Loading: Mode-I loading at 2 mm/min using an Instron machine.
• Imaging: High-resolution camera (2 fps) capturing speckle patterns on the specimen surface.

B. Mesh Generation

• A custom lattice-based mesh was created where each beam is discretized into 18 triangular elements, and joints are meshed with 6 triangles.
• Backtracking: The mesh is aligned with the reference image to ensure precise node tracking, accounting for the specific geometry of the lattice.

C. The Two-Pass Data-Driven Approach

1. First Pass (Threshold Identification):
   • A standard DIC analysis (no element deletion) is performed.
   • The incremental residual variation ($\Delta\rho$) is calculated between consecutive images.
   • A noise floor is established using the Median Absolute Deviation (MAD).
   • Damage Threshold ($\Delta\rho_{th}$): A statistical threshold is set (e.g., 7.3 gray levels) to distinguish abrupt residual spikes (damage events) from background noise or global brightness shifts.
2. Second Pass (Element Deletion):
   • A second DIC analysis is run where elements are automatically removed if their residual variation exceeds $\Delta\rho_{th}$.
   • The mesh connectivity is updated at each time step, and the correlation is solved only over the remaining intact material.
   • This allows for the tracking of the crack tip (centroid of the most advanced damaged element) and the calculation of physically meaningful strain fields.

D. Alternative Mechanics-Based Approach

• Using the data from the first pass, the authors calculate the critical failure strain ($\epsilon_{crit}$) for each broken strut.
• These strain values are used to define a mechanics-based threshold ($\epsilon_{th}$) for a single-pass DIC analysis.
• In this alternative method, elements are deleted if their calculated strain exceeds $\epsilon_{th}$, removing the need for a preliminary residual-based pass.

3. Key Contributions

• Automated Element Deletion for Lattices: The first successful integration of an automated, data-driven element-deletion criterion into global DIC specifically for discrete lattice architectures.
• Robust Damage Detection: A statistical method to distinguish true damage events from noise using residual fluctuations, enabling reliable tracking of crack initiation and propagation.
• Dual-Strategy Framework: The paper demonstrates two viable pathways:
  1. Residual-based: Highly accurate for detecting the onset of failure and localizing damage.
  2. Strain-based: A physics-based alternative that can be used if failure strain properties are known a priori, reducing the analysis to a single pass.
• Critical Strain Estimation: The method provides an empirical estimate of the critical failure strain of individual struts, which can be directly fed into numerical simulations.

4. Results

• Validation on Regular Lattices:
  • The method successfully tracked crack propagation in regular triangular lattices.
  • Residual Reduction: Progressive element deletion reduced global RMS residuals significantly compared to standard DIC (which showed unphysical strain spikes).
  • Accuracy: The automated method identified 27 broken struts, closely matching the visual inspection count of 24. The slight overestimation was attributed to the mesh definition where a damaged joint counts all connected struts as broken.
  • Correlation with Load: Crack propagation events detected by DIC aligned perfectly with force drops recorded by the testing machine.
• Validation on Imperfect Lattices:
  • The method successfully tracked complex, tortuous crack paths in lattices with intentionally missing struts.
  • It accurately captured the deviation of the crack path caused by the imperfections, demonstrating robustness in non-ideal geometries.
• Comparison of Methods:
  • The residual-based method provided the most pronounced reduction in RMS residuals and the most accurate count of broken struts.
  • Strain-based methods tended to overestimate the number of broken struts because deformation fields are less localized than pixel-level residual changes. However, they still correctly identified the onset of damage and the general crack path.

5. Significance

• Bridging the Gap: This work bridges the gap between continuum-based optical measurement techniques and the discrete nature of architected materials, enabling full-field strain measurements that were previously impossible.
• Design and Simulation: By providing accurate estimates of critical failure strains and crack morphologies, the method offers crucial data for validating and refining numerical models of lattice materials.
• Generalizability: The framework is adaptable to disordered architectures, different loading modes (Mode I–III), and volumetric analyses (Digital Volume Correlation), paving the way for comprehensive experimental characterization of fracture in advanced metamaterials.
• Automation: The fully automated nature of the detection and tracking eliminates the need for manual intervention in identifying crack paths, making it suitable for high-throughput testing and real-time monitoring.