Source-linked AI summary
An Automated and Reproducible Workflow for Crack Identification and Damage Assessment of Fusion Materials
Rinkle Juneja, Viktor Reshniak, Richard K. Archibald, John W. Duggan, Gregory R. Watson, Cory D. Hauck, Gary M. Staebler
TL;DR
Post-exposure crack assessment presents challenges that extend beyond segmentation to standardized downstream products. This paper develops a reproducible Galaxy workflow for automated crack identification and damage descriptors, demonstrating outputs across damage states while noting limits for elongated crack-like features.
Problem
The material-assessment challenge extends beyond segmentation to providing standardized outputs for statistical-learning and physics-simulation tools.
Source: §I · pp. 3–4Method
The authors developed a reproducible Galaxy workflow combining scale-aware data ingestion, automated image enhancement and segmentation, quality control, and descriptor extraction.
Source: §VIII · p. 19Results
The workflow produces crack masks, overlays, skeletons, and crack-density descriptors across highly cracked, weakly damaged, and nominally undamaged surfaces.
Source: §VIII · p. 19Takeaways & Limitations
Crack-density outputs provide standardized inputs for downstream statistical-learning and physics-simulation tools.
Source: §I · p. 4The image-processing pipeline is optimized for elongated cracks, so grain boundaries, polishing lines, or elongated pores can be misidentified when they closely resemble cracks.
Source: Interpretation and limitations · p. 17
Abstract
from arXiv · showhide
Post-exposure microscopy is central to qualification of fusion materials. However, manual analysis does not scale to the volume, heterogeneity, and multiresolution character of modern fusion-materials campaigns. To address this challenge, we present a reproducible workflow, implemented in the Galaxy scientific workflow environment, for automated crack identification and quantitative damage assessment from scanning electron microscopy images. The workflow processes SEM images and experimental metadata to identify cracks, quantify damage, and retain the intermediate products and processing history needed for reproducibility. Outputs include crack masks, skeletonized crack networks, quality-control visualizations, and scalar damage descriptors. The method is designed to operate without image-specific parameter tuning across tungsten grades, microstructures, magnifications, and damage states. We demonstrate the workflow on a sparse electron-beam thermal-shock dataset containing 418 images from 114 experiments spanning five tungsten grades and three microstructural states. We define a crack-density descriptor, which provides standardized inputs for downstream machine-learning prediction and physics-based crack simulation. These predictive components are exposed in the same Galaxy environment and are intentionally treated here as extensible workflow modules. The principal contribution is therefore an end-to-end, shareable, and computationally portable workflow that links experimental characterization, automated image analysis, preliminary damage prediction, and simulation-guided data acquisition for fusion-materials research.
I. INTRODUCTION
SEM crack analysis is difficult to standardize across heterogeneous images, while manual inspection is time-consuming and hard to reproduce. The paper presents a Galaxy-based workflow linking automated crack identification with reproducible processing and downstream analysis.
I. INTRODUCTION
- Crack contrast varies, crack-like features arise from other surface structures, and pixel dimensions differ across images, complicating consistent SEM analysis.Manual inspection remains essential for expert interpretation but is time-consuming, subjective, and difficult to reproduce across analysts or campaigns.Source: §I · p. 3
- Material assessment must extend beyond segmentation to connect images with metadata, preserve processing steps and intermediate outputs, and provide standardized downstream products.The workflow should also support re-execution on new data.Source: §I · pp. 3–4
- The proposed Galaxy workflow accepts SEM images and metadata, extracts cracks, computes descriptors, generates quality-control products, and exposes results to modeling modules.The downstream modules include machine-learning and fracture-simulation tools.Source: §I · p. 4
- Workflow definition, reproducibility, and automated crack identification are the focus; predictive and simulation capabilities are preliminary, with detailed formulation and validation deferred.Those capabilities establish interfaces for the broader material-assessment workflow.Source: §I · p. 4
II. EXPERIMENTAL DATASET AND WORKFLOW DESIGN
The workflow is demonstrated on electron-beam thermal-shock SEM data spanning tungsten materials and unevenly sampled material states. It converts acquisition scales to physical pixel resolution and flags or excludes coarse images from crack-specific analysis.
Electron-beam thermal-shock dataset
- The demonstration includes 418 SEM images from 114 experiments across five tungsten grades, using electron-beam exposure as a thermal-only surrogate condition.The experiments were performed at the JUDITH 1 facility.Source: Electron-beam thermal-shock dataset · pp. 4–5
- The test matrix is sparse and uneven, with fewer transverse and recrystallized observations than longitudinal ones, partly reflecting earlier failures at lower heat flux or base temperature.Testing of some states therefore did not continue extensively at more severe conditions.Source: Electron-beam thermal-shock dataset · p. 5
- The workflow converts acquisition scale to physical pixel resolution, using it to compare images and assess crack-analysis suitability; coarse images may be retained for surface studies but flagged or excluded from crack-specific operations.The derived resolution represents the physical length represented by one pixel.Source: Electron-beam thermal-shock dataset · p. 5
Workflow requirements
The workflow is designed for untuned, scale-consistent analysis with traceable, portable, extensible outputs. Galaxy separates workflow use from computing resources, allowing material-assessment tasks to span different execution environments.
Workflow requirements
- The workflow requirements include automation without image-by-image threshold selection, physical consistency across magnifications, traceability, portability, and extensibility to downstream tools.Standardized image-analysis products are intended for statistical-learning and physics-simulation tools.Source: Workflow requirements · p. 5
Galaxy as the orchestration layer
- Galaxy integrates tools, data, and computing resources while managing provenance and reproducibility, and supports workflows spanning multiple resources and invocation routes.Workflows can be invoked through a browser, custom interfaces, or scripts, with outputs visualized in Galaxy or external software.Source: Galaxy as the orchestration layer · p. 7
- Separating the workflow interface from execution resources allows lightweight image processing and large simulations to run on different kinds of computing systems.The described examples include conventional nodes for image processing and GPU-enabled HPC systems for large simulations.Source: Galaxy as the orchestration layer · p. 7
Tool deployment and data contracts
Galaxy tools pair software builds with XML wrappers that define inputs, parameters, outputs, and dependencies. Composed workflows pass compatible data between steps and generate review products and descriptors for downstream use.
Tool deployment and data contracts
- Galaxy tools use a software build and an XML wrapper that specifies execution, required files, exposed parameters, outputs, dependencies, metadata, and help information.Most MPEX tools are packaged in platform-independent containers; on HPC systems that restrict direct container execution, Galaxy can submit batch jobs and use native builds.Source: Tool deployment and data contracts · p. 7
- The crack-identification tool accepts microscopy images and scale metadata and produces enhanced images, masks, overlays, scalar descriptors, summary tables, and optional skeletons.Standardized outputs can feed downstream tools without manual reformatting or file conversion.Source: Tool deployment and data contracts · pp. 7–8
- Galaxy histories record imported files, tool executions, parameters, and outputs, making processing steps available for sharing and reproducible packaging.The interface displays the original image, enhanced image, selected threshold, detected cracks, mask, and skeleton.Source: Tool deployment and data contracts · p. 8, Figure 2 · p. 8
Workflow composition
- The baseline workflow ingests and validates images and metadata, identifies cracks, generates review visualizations and descriptors, aggregates results, and can pass descriptors to learning or simulation tools.Galaxy transfers outputs to compatible downstream inputs, and saved workflows can be shared, reused, modified, or invoked programmatically.Source: Workflow composition · p. 9
IV. AUTOMATED CRACK IDENTIFICATION
Crack identification is complicated by variation in crack scale, image resolution, and surface morphology, which can make crack-like features difficult to distinguish from actual cracks. The pipeline enhances elongated responses and conditions image intensities before automatic thresholding, using shared parameter values across the analyzed images.
Image processing
- Cracks span length scales and image resolutions, while polishing marks, grain boundaries, and surface modification can produce comparable intensity changes.A fixed global threshold may miss faint cracks or over-segment textured regions.Source: Image processing · p. 9
- The Sato ridge filter enhances elongated crack-like ridges by evaluating Hessian eigenvalues of a Gaussian-smoothed image, after the image is linearly rescaled to [0,1] using its minimum and maximum intensities.Source: Image processing · p. 10
- Gaussian smoothing and unsharp masking condition the intensity distribution before segmentation, reducing image-specific artifacts while preserving crack-to-background contrast.The same parameter values were used for all images analyzed in the study.Source: Image processing · p. 11
Automatic segmentation and morphological refinement.
The workflow segments enhanced images using an automatically selected threshold, then refines the masks to suppress noise and connect small gaps. A directional filter produces skeletons for topology-based analysis.
Automatic segmentation and morphological refinement
- Triangle thresholding automatically selects an intensity cutoff from each enhanced image’s histogram, classifying pixels above it as cracks.The method is suited to images where crack pixels are a small population against a dominant background peak.Source: Automatic segmentation and… · p. 11
- Removing components of 10 pixels or fewer and applying morphological closing suppresses isolated responses and connects small discontinuities.Closing uses a 5×5 square structuring element; the cleaned mask supports crack-area-fraction calculation.Source: Automatic segmentation and… · p. 11
- For topology analysis, the mask is skeletonized and directionally filtered to suppress weakly directional features while retaining elongated crack segments.The structure-tensor criterion retains pixels with γ > 0.8; the resulting skeleton supports length, connectivity, branching, and network-topology measurements.Source: Automatic segmentation and… · pp. 11–12
Scale-aware quantitative outputs
The workflow produces scale-aware crack descriptors and representative crack extractions across alloys and microstructures. It also records intermediate outputs and quality checks so results can be reviewed and processing histories compared.
Scale-aware quantitative outputs
- Crack density is the fraction of analyzed pixels classified as crack, with invalid or excluded regions omitted from the denominator.When physical pixel size is available, the workflow also supports crack-length density and component-level statistics in physical units.Source: Scale-aware quantitative outputs · p. 12
V. WORKFLOW DEMONSTRATION AND QUALITY CONTROL
- The workflow recovers crack networks across alloys and microstructures and returns low crack-area fractions for nominally undamaged images.Figure 4 pairs representative SEM characterization with the corresponding automated crack extractions.Source: Scale-aware quantitative outputs · p. 12, Figure 4 · p. 13, §V · p. 13
- Galaxy applies the same processing logic and parameter values without image-specific threshold tuning and records inputs, parameters, software environment, processing history, and outputs.This record supports reproducibility and comparison of revised tools with earlier executions.Source: §V · pp. 13–14
Scale-aware quantitative outputs
- Patches use a fixed physical field of view to keep morphology comparable across magnifications, while coarse-resolution images are excluded from morphology analysis.The example patches correspond to a fixed physical size of 20 µm; coarse resolution provides too few pixels to resolve cracks and surface morphology reliably.Source: Scale-aware quantitative outputs · p. 12, Figure 5 · p. 14
V. WORKFLOW DEMONSTRATION AND QUALITY CONTROL
- Quality control checks image readability, scale and metadata availability, threshold and component statistics, and degenerate masks, with visual outputs available for expert review.Images can be retained with a quality flag rather than silently discarded, preserving the record and helping identify morphologies outside current algorithm assumptions.Source: §V · pp. 14–15
VI. INTERFACES TO DOWNSTREAM MODELING
The workflow’s crack masks and descriptors can feed downstream morphology representations, including patch embeddings and dimensionality reduction. These predictive components are exploratory, with systematic comparisons and validation deferred to follow-on work.
VI. INTERFACES TO DOWNSTREAM MODELING
- Fixed-physical-size patches are represented with pretrained vision-model features and dimensionality reduction to describe surface morphology.PCA provides a linear low-dimensional baseline, while diffusion coordinates are considered for potentially nonlinear morphology variation.Source: Morphology informed prediction · p. 15
- Diffusion coordinates are exploratory rather than a demonstrated improvement over PCA, and their neighborhood structure may represent relations not captured efficiently by a few linear components.The paper relates morphology representations to material composition, microstructure, and exposure conditions as a possible basis for interpolation and future experiment or simulation selection.Source: Morphology informed prediction · pp. 15–16
- Systematic embedding comparisons, model formulation, validation, and uncertainty quantification are reserved for a follow-on study.The present module demonstrates reuse of crack-identification outputs by downstream predictive tools.Source: Morphology informed prediction · p. 16
Interface to physics-based simulation
The workflow connects physics-based simulations to the same quantitative interface used for experimental characterization. This establishes a shared representation for comparing simulated and experimental damage while leaving model formulation and validation outside this work’s scope.
Interface to physics-based simulation
- Simulation outputs can be converted into image-like or descriptor-compatible products and processed through the same quantitative interface as SEM data.This establishes a common representation for experimental and simulated damage and enables comparison within Galaxy.Source: Interface to physics-based simulation · p. 16
- The interface is designed to accommodate different physics-based models without changing the upstream characterization workflow.CabanaPD is presented as one example of a simulation capability that can be connected through the interface.Source: Interface to physics-based simulation · p. 16
- CabanaPD is a peridynamic fracture code whose nonlocal mechanics accommodates displacement discontinuities and crack evolution without a finite-element mesh.It is performance portable through Kokkos and Cabana and can execute on CPUs and GPUs.Source: Interface to physics-based simulation · p. 16
- Evaluating the interface for tungsten thermal-shock conditions, including model parameterization, development, calibration, and validation, is reserved for separate work.The manuscript focuses on the computational interface rather than formulating or validating a specific fracture model.Source: Interface to physics-based simulation · pp. 16–17
VII. DISCUSSION
The discussion presents Galaxy’s persistent, shareable workflow as a way to connect reproducible image analysis to downstream decisions, while distinguishing the deployed crack-identification tool from planned integration targets. It also bounds the interpretation of crack descriptors and identifies limitations in image recognition, predictive inputs, and simulation representation.
The workflow as the scientific contribution
- Galaxy makes the workflow a persistent scientific object, recording executions and constraining data connections while allowing workflows to be shared and rerun.This addresses risks from lost calibration, changed software versions, unrecorded parameter changes, and manual output transfer.Source: The workflow as the scientific… · p. 17
- The deployed crack-identification tool is the most developed component, while prediction, uncertainty-driven sampling, and automated simulation dispatch remain integration targets.The distinction avoids conflating demonstrated image analysis with a fully validated digital twin.Source: The workflow as the scientific… · p. 17
Interpretation and limitations
- Crack area fraction is useful but incomplete, depends on imaging and segmentation conditions, and can be complemented by skeleton-based length, size, orientation, connectivity, and branching descriptors.Roughening, melting, erosion, and redeposition are distinct but related damage modes.Source: Interpretation and limitations · p. 17
- The elongated-crack pipeline can misidentify similar grain boundaries, polishing lines, or pores and may miss wide spalled regions or diffuse modification.The authors identify multi-class characterization as a future direction.Source: Interpretation and limitations · p. 17
- Prediction inputs lack full grain-size, texture, temperature-dependent property, and anisotropy information; simulation currently represents continuum-level rather than explicit microstructure anisotropy.A learned metric can partially compensate for missing structure but cannot recover unobserved physics.Source: Interpretation and limitations · pp. 17–18
Path toward closed-loop material assessment
The workflow establishes a reproducible characterization layer and interfaces for a future closed-loop material-assessment architecture. It converts microscopy images into provenance-tracked quantitative data and connects experimental characterization with preliminary prediction and physics-based simulation.
Path toward closed-loop material assessment
- The proposed closed loop processes experimental images, estimates damage and uncertainty, selects informative experiments or simulations, dispatches computations or ingests new data, and updates the model.The present work establishes the reproducible characterization layer and interfaces needed for that loop.Source: Path toward closed-loop material… · p. 18
VIII. CONCLUSIONS
- The workflow is presented as an extensible foundation for prediction and simulation-guided assessment, while detailed predictive-model development and simulation evaluation, calibration, and validation are left to follow-on studies.Its longer-term scope includes comparisons with simulated damage across sparsely sampled material and exposure spaces.Source: Path toward closed-loop material… · p. 18, §VIII · p. 19
- Scale-aware ingestion, automated enhancement and segmentation, quality control, and descriptor extraction form the post-exposure assessment workflow.The workflow also captures provenance and provides modular interfaces to data-driven and physics-based models.Source: §VIII · p. 19
- Across 418 images from 114 experiments, the workflow produced crack masks, overlays, skeletons, and crack-density descriptors across highly cracked, weakly damaged, and nominally undamaged surfaces.The dataset covered five tungsten grades and three microstructural states.Source: §VIII · p. 19
- Galaxy deployment makes analysis shareable, re-executable, and portable across computing resources while preserving execution history.The history retains inputs, parameters, software versions, and outputs.Source: §VIII · p. 19