AI weld analysis
Honest feedback from a photograph — with the limits printed on the label
ArcForge analyzes the visible surface of your practice welds and reports what it sees in cautious, structured language. Here is exactly what that means — and what it can never mean.
What it looks at
Visible bead characteristics
The analysis examines surface evidence — the same things a trained eye scans first during a visual once-over.
- Bead width consistency along the joint
- Ripple pattern and travel-speed rhythm
- Toe transition and blending into the base metal
- Surface profile cues — convexity, concavity, underfill
- Visible undercut or overlap indicators at the toes
- Surface-breaking porosity
- Spatter quantity and distribution
- Start, stop, and crater-fill quality
- Tie-ins between passes or restarts
- Surface color and oxidation cues (especially stainless and TIG work)
- Slag residue, arc strikes, and surface contamination signs
- Bead placement and alignment along the joint line
Hard limits
What a photo can never confirm
These are physical and procedural limits, not feature gaps. No ArcForge result will ever claim any of the following — the language constraints are built into the system itself.
- Code compliance or acceptance to any welding standard
- Structural integrity or fitness for service
- Internal fusion or lack-of-fusion below the surface
- Penetration depth or root quality on a closed joint
- Official acceptance criteria of any kind
- Certification or qualification status
- Readiness or qualification for any employer or job
- Results of destructive or nondestructive testing (bend, tensile, UT, RT, PT, MT)
Educational guidance only
AI-assisted visual feedback is educational guidance only and is not always accurate. A photograph cannot confirm code compliance, structural integrity, internal fusion, penetration, or test results, and it is no substitute for hands-on inspection. Have a qualified instructor or inspector evaluate any weld that matters.
Context matters
How details improve feedback
The photo is half the input. The details you attach are the other half — they change what the analysis expects and how it interprets what it sees.
Process
A textured FCAW surface and a glassy GTAW bead are judged by different visual norms. Naming the process keeps expectations realistic.
Joint & position
Gravity changes everything. A 3F vertical fillet is read differently from a 1F flat one, and pipe positions differently again.
Material
Color tells different stories on carbon steel, stainless, and aluminum. Material context prevents misreading heat tint or soot.
Machine settings
Amps, voltage, wire-feed speed, and polarity let feedback connect a visual symptom to a plausible setup cause worth testing.
Practice objective
Telling the system what you were working on focuses the result — and powers the practice-objective alignment score.
When the photo isn’t enough
If the image is too dark, blurry, or cropped to read responsibly, the system says so and asks for a better photo instead of guessing.
Photo privacy
Your photo is analyzed, then discarded — unless you choose to keep it
Weld photos are analyzed and discarded by default — stored only if you choose to save a photo to your private record, and you can remove it anytime. What we always keep is the structured result and the settings you entered — your history and weld log are built from those, not from images.
- By default the photo travels with your submission, is read by the AI vision model, and is gone when the analysis finishes.
- Want to keep one? Flip “Save this photo to my record (private)” when you submit — saved photos are visible to you alone and removable anytime.
- Results, practice scores, and your full machine setup are saved — that record is yours to revisit and compare.
- If an analysis fails, your allowance is refunded automatically and you simply resubmit the photo — failed runs never store the photo, even if you asked to save it.
- We never use weld photos for marketing and never train our own models on them.
Confidence levels
Every finding says how sure it is
Findings are observations, not verdicts. Each one carries a confidence level so you know what to verify by hand and with your instructor.
The visual indicator is clear in the image — for example, heavy spatter across a well-lit plate. Still a visual observation, never a measurement.
Something appears present but the image leaves room for doubt — lighting, angle, or resolution limits certainty. Worth checking by hand.
A faint or ambiguous cue. Low confidence is deliberately reported rather than hidden, so you know what to verify rather than assume.
Educational scoring
Bands, not bogus percentages
The practice score is an ArcForge educational metric — it describes your visible consistency across seven categories. It is not an industry measurement and was deliberately designed to avoid false precision.
The five levels
- EmergingLevel 1 of 5
- DevelopingLevel 2 of 5
- ConsistentLevel 3 of 5
- ProficientLevel 4 of 5
- RefinedLevel 5 of 5
A score of “Consistent” means your visible results are repeatable; “Refined” means the surface evidence shows deliberate control. No level implies a weld would pass any test.
The seven categories
- Bead consistency
- Profile control
- Toe transition
- Start & stop control
- Surface cleanliness
- Visual uniformity
- Practice-objective alignment
The full limits of AI analysis are documented in the AI analysis disclaimer
Put a photo through it
A couple of AI analyses a day on the free plan — enough to see whether the feedback loop fits how you practice.
AI feedback is educational guidance only and does not replace qualified inspection.