ENIT
Request demo
ENIT

Module VISION · Metalya Vision

AI Vision & Defect Detection

The factory that sees, follows and recognises: images and video turned into production events.

Request demo

Why it matters

Metalya Vision applies computer vision to the production line: it detects objects and anomalies, follows their movements and recognises the same item across different areas and cameras. Each observation becomes a production event, such as a defect, count, passage, stop or presence, that can trigger an alert, a non-conformity, a traceability update or a task. It is integrated with MES, QCMS, WMS, CMMS, traceability and enterprise systems.

  • Defects stopped early

    Visible defects are flagged on the line before they become waste.

  • The real flow, measured

    Counts, queues, stops and throughput times are observed directly by the cameras.

  • Events, not just images

    Each observation becomes a production event that can trigger alerts, non-conformities or tasks.

Key features

What it does

  • 01

    See: object detection

    Identifies what is in the scene and where: parts, products, people, vehicles, containers, materials, defects and machine states.

  • 02

    Follow: object tracking

    Follows each object with its own ID, counts passages and detects stops, queues, deviations and build-ups.

  • 03

    Recognise: re-identification

    Keeps an object's identity when it disappears, reappears or moves in front of another camera, combining visual features with time and location.

  • 04

    Automated defect detection

    Detects scratches, stains, cracks, missing parts, foreign particles, deformation, discolouration and misalignment, with custom setups possible. User validation of flagged defects feeds back into model retraining.

  • 05

    Flow and productivity

    Counts and measures the real production flow: work in progress between stations, queues, throughput time, conveyor build-ups and items on the wrong route.

  • 06

    Visual traceability

    Maintains identity even when the code is not visible, following pallets, carriers, coils, reels or containers associated with a production lot.

  • 07

    Process monitoring and internal logistics

    Turns visible conditions into events, such as a machine stopped, an empty feeder or an overflow, and tracks what enters, leaves and moves around the plant.

  • 08

    From camera to action

    Near real-time with adequate hardware, or in batch mode. Events flow into Metalya and enterprise systems via APIs, SDKs and webhooks.

How an observation becomes an action

An observation is only useful when it lands on the right object.

A camera on its own produces video. Connected to the manufacturing core, an observation becomes a defect on a lot, a count on a work stage, a stop with a reason or a move of a pallet — attached to the order, lot or machine it concerns. Then the process reacts.

  1. 01 · The camera observesAssembly station ASM-3, frame camera
  2. 02 · The AI understandsBracket missing on housing AX-90detection
  3. 03 · Metalya generates an eventDefect assigned to lot L-24-0917assign_quality_defect_to_batch
  4. 04 · The process reactsLot on hold · quality order createdgenerate_quality_order

Illustrative flows. The event targets are existing Metalya actions; the vision part is configured per use case.

Use cases

The same capabilities apply to discrete and continuous manufacturing, with different objects and events in each case.

  • Quality

    Stop defects before they become waste. Discrete: missing components, incorrect assembly, scratches, defective welds, non-compliant parts. Continuous: colour variations, visible contamination, uneven coating, faulty seals, surface anomalies.

  • Flow and productivity

    Count, follow and measure the real production flow. Discrete: work in progress between stations, queues in front of a machine, throughput time, stopped parts, items on the wrong route. Continuous: conveyor build-ups, empty lines, flow speed, counting cans, bags, rolls or packages.

  • Visual traceability

    Maintain identity even when the code is not visible. Discrete: follow the same frame, pallet, carrier or container across production cells. Continuous: follow coils, reels, big bags, drums, tanks or containers associated with a production batch.

  • Operational safety

    Detect a dangerous situation while it is happening. Discrete: a person inside a robotic cell, a forklift close to an operator, access to a restricted area, missing PPE. Continuous: presence in hot or controlled areas, obstacles in a passageway, vehicles in loading areas, visible spills. For dedicated safety use, see Metalya Safety.

  • Internal logistics

    Know what enters, what leaves and where it goes. Discrete: pallets at loading bays, containers being moved, forklifts, components delivered to the line. Continuous: trucks, tanks, drums, big bags, incoming raw materials and outgoing finished products.

  • Process monitoring

    Turn a visible condition into a production event. Discrete: machine running or stopped, empty feeder, tool present, jammed component, occupied workstation. Continuous: material on a conveyor, visible level, foam, overflow, flame, or an indicator in an abnormal position.

Evaluating your use case

Show us what you need to see. The evaluation starts from a few images or a short video of your production line, to assess the objects, movements, environmental conditions and integration with your processes. Vision is delivered as a project for each use case: accuracy, speed and hardware depend on the scene and are agreed after the feasibility assessment.

  • Sample images

    Good baseline images and representative defective samples help train and tune the model. Our team assists with collecting them.

  • Lighting and surfaces

    Controlled lighting is important, and we provide guidelines and calibration tools. In difficult cases, specialised lighting (such as coaxial or dark-field) or filters can help.

  • Camera setup

    Guidelines on camera position and lighting; existing cameras are reused where image quality allows.

  • Integration

    Observations are mapped to Metalya events and actions — quality, traceability, production, logistics — so the process reacts on the right order, lot or machine.

  • Model maintenance

    Besides retraining based on user validation, periodic recalibration and model versioning are supported.

  • Hardware monitoring

    Health monitoring, fallback inspections and alerts help manage camera or lens issues.

  • Feasibility

    Very glossy or transparent products, or extremely fine textures, can be challenging. Feasibility is assessed in your specific context.

In the platform

One of the modules of Metalya Core. Every module shares one governed data model, so what happens here is immediately visible to planning, quality, materials and maintenance — and Metalya Studio adapts it to the way your plant works, with AI.

Works together with

Buy the foundation. Adapt the experience.

See Metalya on your own production scenario: your routings, your constraints, the screens your people need.

Request demo Talk to us