Detection or measurement
Define which features must be visible and the applicable tolerance, code quality or defect threshold.
Machine vision for production
Machine vision solutions for inspection, identification and robotics
Product variation, short cycle times and manual checks determine which imaging and automation approach is feasible. Start with the production problem below. The camera, lens, lighting, software and machine interface follow from that requirement.
Choose by production problem
What decision must the system make?
Start with the task that is currently manual, unstable or not traceable. The specialist pages explain the technical boundaries and implementation for each application.
Inspect
Quality inspection
Check presence, assembly, print, surfaces or visible defects and send a result to the PLC or line control.
Select by defect definition, contrast, variation and cycle time.
Identify
Code reading, OCR and traceability
Read 1D and 2D codes, DPM marks or text and associate the result with a product, order, PLC, MES or database.
Select by code quality, material, speed and no-read handling.
Measure and locate
Measurement and positioning
Determine dimensions, position, height, shape or volume for process control, sorting or control of a subsequent operation.
Choose 2D or 3D based on tolerance, surface and height variation.
Guide a robot
Robot vision
Locate parts for picking, placing, machine loading or processing and transfer coordinates to the robot in a controlled way.
Choose 2D or 3D based on presentation, height and orientation.
Classify variation
AI inspection
Use a trained model when relevant product or defect variation cannot practically be described with fixed image-processing rules.
Select by dataset, error cost, ownership and behaviour under new variation.
Pick parts from a bin
3D bin picking
Plan a reachable, collision-checked pick for parts lying loose in a bin and partially blocking one another from view.
Select by part geometry, visibility, gripper and robot reach.
Technical architecture
The application does not yet determine the complete system architecture
The same inspection task can be implemented in several ways. Camera count, product motion, data storage, recipe management, robot integration and maintenance determine the suitable architecture.
Smart camera
Camera and image processing are combined in one industrial housing. This suits defined tasks with local I/O and manageable configuration.
Boundary: processing capacity, multiple cameras, data storage and custom logic.
PC-based vision
An industrial PC processes one or more camera streams and provides scope for more extensive software, AI, logging and machine interfaces.
Boundary: broader integration scope, PC management and responsibility for software maintenance.
2D or 3D
2D uses contrast in a flat image. 3D adds height or depth for shape, volume, overlapping parts or variable product height.
Choose 3D when the decision genuinely requires depth information.
Fixed vision rules or AI
Fixed rules suit features that can be described consistently by contrast, geometry or dimensions. AI helps with relevant variation that cannot practically be captured in such rules.
AI requires representative data, version control and monitoring of new variation.
From requirement to production
A solution is tested as a complete imaging chain
The camera is one component. Reliable operation requires a coordinated choice of field of view, lens, lighting, trigger, processing, mechanical installation and machine communication.
Step 1
Define the decision
Describe good, bad and borderline cases and define the action following each result.
Step 2
Test samples
Compare camera, lens and lighting using representative products, motion and ambient light.
Step 3
Select the architecture
Define processing, I/O, PLC or robot integration, recipe management, logging and error handling.
Step 4
Install and validate
Test timing, product variation, changeovers and fault behaviour under agreed production conditions.
Step 5
Transfer ownership
Document settings, spares, cleaning, training and responsibilities for changes.
Production acceptance
Define measurable acceptance before implementation
A demonstration using one good product is insufficient. Define the product variation, speed, error types and fault conditions that the acceptance test must cover.
Product variation
Include approved variants, colour, surface, position, contamination and relevant borderline cases.
Cycle time
Measure trigger, exposure, processing, communication and mechanical action at the highest agreed line load.
Error handling
Test no-read, uncertain result, timeout, connection loss and reject or robot behaviour.
Changeover and recovery
Check recipe changes, restart, calibration and recovery after downtime without uncontrolled production.
Ownership
Define who may change thresholds, models, software versions, spares and maintenance procedures.
Business case
Use process data from your own line
The value depends on the current process, the cost of errors and the actions that can follow a vision result. Use measurements from a representative period and state assumptions separately.
Simple calculation model
Annual process value = avoided manual hours + avoided cost of rejects, rework and claims + contribution from available capacity − additional maintenance, licence and ownership costs. Compare this with engineering, hardware, installation, validation and training.
Required inputs
- Manual inspection time, labour cost, staffing per shift and annual production hours.
- Current quantities of rejects, rework, returns, line stops and operator interventions, including cause.
- Product mix, batch size, changeover time, cycle time and expected volume variation.
- Cost of cameras, optics, lighting, processing, mechanics, cabling, software and integration.
- Validation, training, cleaning, spares, licences, backups and planned software maintenance.
Next step
Test the critical imaging requirement with representative products
Send images, product samples and a short description of cycle time, variation, required decision and machine interface. Sedeco can then determine which questions should be answered first in a feasibility test.