Synthetic Weld Inspection Datasets
Code-aligned synthetic NDT cases for inspector training, curriculum development, and machine-learning research. Generated rather than scanned from physical welds — so they ship without NDA constraints, scale to whatever volume you need, and come with ground-truth labels you can verify.
What's in the Library
Visual-inspection (VT) datasets are the planned first release, given VT's role in shipbuilding and fabrication QC. All methods are currently in development with an early-access waitlist; nothing is shipping yet. Release order beyond VT is set by waitlist demand and customer commitments.
Visual-Inspection Cases
Synthetic visual-inspection imagery for weld VT, fit-up checks, surface-condition documentation, and coatings/corrosion assessment — built around the inspection scenarios that drive shipbuilding and fabrication QC.
- Acceptance grading per code reference (AWS D1.1, ASME, NAVSEA-style criteria)
- Ground-truth labels with bounding boxes and pixel-level masks
- Configurable defect type, size, and surface finish
Radiographic & Ultrasonic Cases
Synthetic RT imagery (butt and fillet welds with labeled discontinuities) and synthetic UT A-scans/B-scans (typical weld geometries with labeled responses from planar and volumetric flaws). Release order is set by waitlist demand.
- RT: severity grading per AWS D1.1, ASME IX, API 1104
- UT: amplitude, time-of-flight, and depth annotations with calibration block / DAC reference traces
- Multiple exposure / IQI variations (RT); configurable flaw geometry (UT)
Surface & Eddy-Current Methods
Subsequent releases extend the library to surface methods (MT/PT imagery) and eddy-current impedance plots, using the same labeling structure across all methods. Priority order is set by waitlist demand and customer commitments.
- Same labeling structure across all methods
- Cross-method case bundles for teams qualifying inspectors in multiple methods
- Custom datasets available now — contact sales
Three Audiences We Build For
Inspector Training
Build a written-practice exam without needing 500 real plates. Run candidates through 200 RT cases drawn from a labeled bank, then check their calls against ground truth. Shippable as a curriculum supplement to any SNT-TC-1A-aligned program.
Curriculum & Education
NDT programs at community colleges, training providers, and corporate Level III programs need consistent, defendable practice material. Synthetic cases give the same scenario to every cohort and remove the customer-confidentiality problem.
Machine Learning Research
Defect-detection model builders need labeled imagery at scale. Real radiographs are tied up in customer NDAs. Our synthetic cases ship with class labels, masks, and provenance metadata suitable for supervised training and evaluation.
Join the Early-Access Waitlist
No fixed launch date. Waitlist members are contacted in priority order when each method ships, and get a 30-day window to evaluate sample cases before general release.