Use case · Quality

Visual quality inspection

Cameras and computer vision models inspect parts on the line to catch surface, assembly and dimensional defects consistently.

Maturity
Proven, scaling
Complexity
Medium (our assessment)
Department
Quality
Technology
Computer vision, Machine learning
Solves
Defects reaching customers, Scrap & rework, Hard-to-fill roles, Compliance paperwork

What it is

AI visual inspection uses industrial cameras, controlled lighting and deep learning models to find defects such as scratches, porosity, cracks, missing components, mislabeling and contamination. Unlike traditional rule-based machine vision, which needs every defect described in explicit measurements, learned models are trained on example images and can handle natural variation in texture, color and part orientation.

It matters because manual visual inspection is tiring, hard to staff and inconsistent between people and across a shift. Rule-based vision works well for simple gauging but struggles with cosmetic defects on variable surfaces. Learned models fill that gap and give a consistent, recorded decision for every part, which also produces data that points back to where defects originate.

How it works

  1. Define the defects and the stakes

    Quality and production agree on a defect catalog, which defects are critical versus cosmetic, and what happens to a rejected part. This sets how strict the model must be and where a human stays in the loop.

  2. Get the optics right

    Camera resolution, lens, lighting angle and part presentation are chosen so defects are actually visible in the image. Most inspection projects succeed or fail here, before any model is trained.

  3. Collect and label images

    Images of good and defective parts are gathered from production and labeled by experienced inspectors. Some approaches train mostly on good parts and flag anything unusual, which helps when defects are rare.

  4. Deploy at the line

    Models run on edge hardware next to the station and send pass, fail or review decisions to the PLC for diverting parts. Borderline cases can go to a human reviewer whose decisions feed back into training.

  5. Monitor and retrain

    Lighting changes, new suppliers and design revisions shift what images look like. Tracking false rejects and escapes per shift shows when the model needs fresh examples.

Benefits

Consistent decisions

The same part gets the same call on first shift and third shift, because the model does not fatigue or drift with attention the way human inspection does over a long shift.

Inspectors redeployed

Experienced inspectors move from repetitive checking to reviewing edge cases, improving the defect catalog and working problems upstream.

Defect data for root cause

Every image and decision is stored with a timestamp, so quality engineers can tie defect types to specific tools, shifts, lots or machines instead of relying on tally sheets.

Earlier detection

Inspecting at intermediate stations rather than only at final inspection catches problems before more value is added to a part that will be scrapped.

Data & prerequisites

Data you will need

  • Labeled images of good parts and each defect type from production
  • Defect catalog with severity classes and acceptance criteria
  • Part identifiers, lot and timestamp to link images to production records
  • PLC signals for part presence, triggering and reject actuation
  • Inspector review decisions on borderline cases

Before you start

  • Stable part presentation and controlled lighting at the inspection point
  • Agreed, written acceptance criteria that experienced inspectors apply consistently
  • Access to enough defect examples, or acceptance of an anomaly-based approach while examples are collected
  • Controls engineering support to integrate triggers and reject mechanisms

Software for visual quality inspection

Companies whose published products address this use case. Listings are editorial, not paid. Filter the directory.

Avathon

Asset performance & predictive maintenance · Pleasanton, CA

Avathon, formerly SparkCognition, offers an industrial AI platform that combines an operational knowledge graph with AI agents for asset-heavy operations. For manufacturers it covers predictive maintenance through...

Predictive maintenanceProcess anomaly detectionVisual quality inspection
Founded 2013

Cognex

Vision inspection & quality AI · Natick, MA

Cognex is a long-established machine vision and barcode reading company whose product line now includes deep-learning inspection tools built partly on its acquisitions of ViDi Systems and SUALAB. Its AI offerings range...

Visual quality inspection
Founded 1981

Cybord

Vision inspection & quality AI · Tel Aviv, Israel

Cybord applies visual AI to images of electronic components captured on SMT lines to detect damaged, counterfeit or unapproved parts before they end up on circuit boards. It also builds component-level traceability and...

Visual quality inspectionSupply chain risk & visibility
Founded 2018

Eigen Innovations

Vision inspection & quality AI · Fredericton, Canada

Eigen Innovations combines thermal and optical imaging with AI to inspect parts and monitor heat-driven processes such as plastic welding, molding, adhesive application and furnace operation. Manufacturers use it for...

Visual quality inspectionQuality prediction & virtual metrologyPredictive maintenance
Founded 2012

Elementary

Vision inspection & quality AI · South Pasadena, CA

Elementary sells machine vision inspection systems for production lines, pairing cameras, lighting and edge controllers with self-training AI for defect detection, label verification and presence checks. A cloud...

Visual quality inspection
Founded 2017

Instrumental

Vision inspection & quality AI · Palo Alto, CA

Instrumental builds an AI platform for electronics assembly that captures images of every unit on the line and combines them with factory test and process data. Engineering and operations teams use it to catch defects...

Visual quality inspectionRoot-cause analysis for quality & yield
Founded 2015

LandingAI

Vision inspection & quality AI · Mountain View, CA

LandingAI, founded by Andrew Ng, offers LandingLens, a no-code platform for labeling images, training and deploying computer vision models, originally launched for manufacturing visual inspection. The company's main...

Visual quality inspectionDocument automation for supplier, compliance & QMS
Founded 2017

Siemens

Industrial copilots & AI agents · Munich, Germany

Siemens, the German industrial automation and software company, sells a family of industrial AI tools under Siemens Xcelerator. They include the generative-AI Industrial Copilot for automation engineering in TIA Portal,...

Copilots for PLC & automation engineeringGenAI knowledge assistants & copilotsPredictive maintenance
Founded 1847

Real-world examples

Each example is labeled by who reported the result. How we grade evidence.

Supporting research

The Rise of Artificial Intelligence (AI) in U.S. Manufacturing

NIST Manufacturing Extension Partnership (MEP) · Report · 2025

A NIST MEP infographic summarizing AI adoption, barriers, core technologies and use cases in U.S. manufacturing, with MEP Center case examples (GaMEP and CJB Industries, NY MEP and zBeats, Ohio MEP and Magellan Aerospace). Its adoption and investment figures are compiled from Manufacturing Leadership Council surveys cited in its footnotes.

Government

Common pitfalls

  • Skimping on lighting and optics and expecting the model to compensate for defects the camera cannot see
  • Inspectors disagreeing on what counts as a defect, which produces contradictory labels and a confused model
  • Validating only on a curated test set and not on a full production run with real variation
  • No plan for retraining when suppliers, materials or product revisions change

Questions to ask vendors

  1. How many labeled defect images do you need per defect type, and what do you do when defects are rare?
  2. Who designs and is accountable for the camera, lens and lighting setup?
  3. How do you measure false rejects and escapes during acceptance testing, and on what volume of parts?
  4. Does inference run at the edge within our cycle time, and what happens if the network goes down?
  5. Can our own engineers add new defect types and retrain, or does every change need your team?

What to measure

  • Defect escape rate to downstream stations and customers
  • False reject rate (good parts flagged as bad)
  • First pass yield at the inspected station
  • Inspection labor hours per unit
  • Customer complaints and returns tied to visual defects

Funding that can help

Frequently asked questions

How is AI visual inspection different from traditional machine vision?

Traditional machine vision relies on hand-written rules such as edge positions, pixel thresholds and measurements, which works well for gauging and presence checks. AI inspection learns from example images, so it handles defects with variable appearance, like scratches on textured surfaces or irregular contamination. Many lines use both, with rules for measurement and learned models for cosmetic and complex defects.

How many images do you need to train a defect detection model?

It depends on how varied the parts and defects are, but the bigger constraint is usually defect examples rather than total images. Anomaly detection approaches can start from mostly good-part images and flag anything unusual, then add defect-specific classification as examples accumulate. Consistent labeling by experienced inspectors matters more than raw volume.

Can AI inspection replace human inspectors?

It typically replaces repetitive checking at a station, not the inspection function. People are still needed to define acceptance criteria, review borderline parts, label new defect types and investigate why defects occur. Regulated industries also need documented validation and human oversight of the system.

Related use cases

Quality

Root-cause analysis for quality & yield

Machine learning sifts process, material and equipment data to narrow down which factors are actually driving defects and yield loss.

Machine learningGenerative AI & LLMs
Proven, scaling8 software companies

Quality

Quality prediction & virtual metrology

Predict product quality from live process data so problems are caught before parts are measured, tested or shipped.

Machine learningIndustrial IoT & edge
Proven, scaling2 software companies

Operations & production

Manual assembly analytics & operator guidance

Cameras and sensors guide operators step by step and verify each manual assembly step, catching errors before the part moves on.

Computer visionMachine learningIndustrial IoT & edge
Proven, scaling3 software companies

Quality

Document automation for supplier, compliance & QMS

AI reads, checks and drafts the documents behind quality and compliance, from certificates and specs to deviations and audit records.

Generative AI & LLMsMachine learningComputer vision
Proven, scaling6 software companies

Put AI to work

Find the AI opportunities hiding in your plant.

Genvo, our implementation partner, works with manufacturers to pick the right first project, connect the systems you already run and put AI agents and tools into daily use, with people approving what goes out.

Assessments and consultations are provided by Genvo. How our partnership works.