Skip to content
LexDataLexData
PlatformIndustriesCustomers
DocsThe Field GuideBlogWhy models drift
AboutCareersSecurityContact
Log inStart now

// MANUFACTURING

Body-in-White Inspection

Body-in-White Inspection. What it takes to run it.

Dents masked for extent on the bare shell under the station's own fixed light, graded against the plant's reportable limit by face.

Start with your footageBook a demo →All of manufacturing →

// 01 · The question

“Is that a dent, or a reflection of the roof lights off bare steel?”

// 02 · Why it is hard

Unpainted steel reflects the whole building, and a swage line or a feature line the designer put there on purpose bends the reflection the same way a dent does. The detector has to be tuned to miss nothing, which means it boxes the shell's own designed contours until it has seen enough of them to learn the difference.

// 03 · What the labels have to be

Instance segmentation

Decided before the first frame

A dent's disposition is read off its size and where it sits on the shell, cosmetic face or hidden face, and only a mask carries that extent. A box says a break in the reflection pattern is there; it does not say how big.

This is the decision that is expensive to reverse. The geometry has to match what the answer contains, and finding out it does not means labelling the set a second time.

// 03a · What it finds

A break in the station's own reflection pattern

a box on every place the fixed stripe or grid light breaks across the shell, tuned to miss nothing, so a swage line is boxed too until enough examples teach the difference

The dent's extent

a mask over the break where the disposition is a severity call, since a three-millimetre mark and a thirty-millimetre mark on the same panel are not the same decision

Where the dent falls on the shell

the mask's position against the body map, a class A face and a face hidden by trim, since the same mark is reportable on one and not the other

A mark the model doubts

a feature line under an odd reflection, or a fixture scuff, held for the inspector with the frame and the body map position before the shell is sorted

// 03b · What it checks

  1. 01

    The mask's extent against the written limit for that face, so reportable is read off a measurement rather than a guess at how a box looked in the reflection.

  2. 02

    The zone the mark sits in, a class A face or a face the trim will hide, before severity is set, since the same mark is not graded the same way on both.

  3. 03

    The second pass's rule, separate from the detector, so a tightened tolerance is a number changed in the rule and not a retrain of what counts as a broken reflection.

  4. 04

    The frame against the model's own doubt. A feature line or a fixture scuff the model is unsure of goes to the inspector before anything ships or trains on it.

// 03c · What the line provides

  • The camera already at the station, with the station's own fixed light arranged so the reflection in the panel is a known stripe or grid pattern; no new camera.
  • Frames from this station, bare shells under this fixed light with the forklifts and the hi-vis going past, not a studio photo set.
  • The plant's own reportable limit in writing, by face: the extent that fails a class A panel and the extent a hidden face tolerates.
  • The inspector at the station, for the shell the model doubts, the way the highlight board decides today.

// 04 · How we run it

Three parts of one loop, on this job.

LexAnnotate

Defect Annotation

Pixel-level defect annotation for surface scratches, cracks, dents, weld faults, and assembly misalignments. Multi-sensor alignment for multi-camera inspection systems.

LexInsight

Line Performance

Ask which shift or which line is producing the most of one defect class, and get an answer with the frames behind it, graded against the plant's own written standard.

LexAlert

Quality Alerts

Monitor defect detection accuracy across changing production conditions. Detect when lighting shifts, new materials, or equipment aging degrades model performance. Auto-retrain to maintain 99%+ accuracy.

How the loop fits together →

// 05 · What breaks it after launch

Concept drift

The spec changed

→Concept drift →

A customer tightens the door-skin tolerance and every shell with a dent at the old limit now fails. The frames are identical and the detector finds the same dents in the same places; what moved is the number in the second pass that decides reportable, not anything the camera sees.

It is not the only one that can get this use case, it is the one that usually gets it first. All ten conditions.

// 06 · In manufacturing

Defect Detection at Scale

Full write-up ↓Every case study →

The Challenge

Vision models performed well in pilots but lost accuracy as factory conditions shifted.

Our Execution

Built pixel-level annotated datasets and multi-sensor alignment for reliable defect detection.

The Result

Maintained 99%+ defect detection accuracy across changing production environments. Reduced unplanned downtime.

99%+

Accuracy maintained in production

// 06a · Questions we get asked

Does this replace the inspector with the highlight board?

No. It boxes every break in the station's own reflection pattern and masks the ones worth sizing, then holds the borderline shell for the inspector with the frame and the body map position. The highlight board stays on its hook for the shells that come back for a second look.

How does it tell a dent from a designed feature line?

From examples. A swage line or a crown in the roof bends the reflection the same way a dent does on the first pass, so the detector is tuned to miss nothing and boxes both until the inspector's corrections on this station's own shells teach it the shell's designed contours.

What happens when the door-skin tolerance tightens?

The frames do not change, and the detector keeps finding the same dents in the same places. What changes is a number in the second pass that decides reportable, written the day the new spec arrives, so the fix is in the rule rather than a retrain of the detector.

Can it read gap and flush the way a CMM does?

No. This finds and sizes dents and surface breaks from a station camera under fixed light; it is not a dimensional measurement system and makes no claim at the tolerances a CMM or a laser scanner reads. A shell flagged for gap or flush still goes to that equipment.

Read next

Body-in-white inspection on bare steel, the full walk-through →Surface defect detection: the same find-then-grade split for any part →What a tightened tolerance does to a label that has not moved →
Start with your footageBook a demo →

// 07 · Also in manufacturing

Surface Defect Detection→Weld Quality Inspection→Assembly Verification→Package & Label Inspection→Date & Lot Code Inspection→Predictive Maintenance→Multi-Sensor Monitoring→Tablet Defect Inspection→Syringe Barrel Defect Detection→Foreign Material Detection→Glass Defect Detection→Semiconductor Wafer Inspection→Strip Surface Defect Inspection→Ceramic Defect Detection→Wood Defect Detection→Contact Lens Defect Inspection→All forty-one→
Surface Defect DetectionWeld Quality InspectionAssembly VerificationPackage & Label InspectionDate & Lot Code InspectionPredictive MaintenanceMulti-Sensor MonitoringTablet Defect InspectionSyringe Barrel Defect DetectionForeign Material DetectionGlass Defect DetectionSemiconductor Wafer InspectionStrip Surface Defect InspectionCeramic Defect DetectionWood Defect DetectionContact Lens Defect InspectionAll forty-one →

Send us a week of this footage. We’ll show you what comes back.

Start with your footageBook a demo →
LexData
LexData

Product

  • Platform
  • Industries
  • Use cases

Resources

  • Docs
  • The Field Guide
  • Blog
  • Why models drift

Industries

  • Energy & utilities
  • Oil & gas
  • Agriculture
  • Manufacturing
  • Insurance
  • Retail
  • Robotics

Company

  • About
  • Customers
  • Careers
  • Contact

Trust

  • Security
  • Privacy
  • Terms

Stay updated

What we learn running vision models in production.

See everything.
Miss nothing.

Stay updated

What we learn running vision models in production.

Terms of use & Privacy policy

© 2026 LexData Labs · All rights reserved