addaScan home
Procedure

Machine vision commissioning procedure: device selection, algorithm set-up and testing

Our standard procedure, from requirement to handover, for engineers commissioning a vision station or checking a supplier’s work.

By addaScan engineering teamTechnical review: Frank GuoPublished 10 min read

Discuss commissioning on your line
// Short answer

Commissioning a machine vision station follows eight steps: fix the requirement and samples; select the processing, camera, lens, light, trigger and reject; set up the camera and lighting on real products; build the algorithm by locating first, then measuring, reading or classifying, with limits set on the sample set; connect and test the PLC signals, the reject and fault responses; prove it at FAT; install and prove it at SAT on real production; and hand over with training, backups and documents.

The procedure at a glance

This is the procedure we follow to commission a vision inspection station. It applies to a single smart camera as much as to a multi-camera station; the steps stay the same, only their size changes. Each step ends with a record, so the next step starts from something agreed.

Machine vision commissioning procedure at a glance
StepWhat happensRecord
1. RequirementDefect list, limits, line data, samplesSigned requirement and sample set
2. Devices and opticsCamera or smart camera, lens, light, controller, trigger, rejectBill of materials and layout
3. Camera and lightingMounting, focus, exposure, gain, light, triggerSaved camera settings and reference images
4. AlgorithmLocate, then measure, read or classify; limits from samplesRecipe per product, with limits recorded
5. PLC and rejectSignals, tracking, reject, confirmation, faultsTested I/O list and fault responses
6. FATAgreed samples at agreed speed, before shippingSigned FAT record
7. SATReal production, all products, reject and fault testsSigned SAT report
8. HandoverTraining, backups, documents, spares listSigned handover and support plan

1. Fix the requirement

  • List every defect to catch, with good and defective samples of each, including borderline cases judged by your quality team.
  • Write down the limits: what size of scratch, how much skew, which codes and texts must match what.
  • Record the line: products and variants, speed, product spacing and orientation, space for the station, PLC make and model, existing reject, air supply, cleaning regime.
  • Agree how acceptance is measured: the false reject rate and the missed defect rate, on which samples, at which speed.

See how to prepare samples.

2. Select devices and optics

Processing. A vision sensor for one simple check from one view; a smart camera for one or two checks with a wider tool set; separate cameras with a vision controller or PC for several views, heavy processing or many products. HIKROBOT’s smart camera selection guide works the same way: camera type and lens, focus adjustment, the algorithm capability of each series, algorithm performance against cycle time, and hardware limits such as the number of modules and solutions (HIKROBOT catalogue 2025 Q4). See choosing a HIKROBOT smart camera.

Camera and lens. From the field of view and the smallest defect, which must cover several pixels; area scan for separate products, line scan for continuous material; global shutter for moving products. See camera and lens selection.

Lighting. Chosen on the samples for the defect: backlight for outlines and fill, dark-field for scratches, dome for shiny curved packs, coaxial for flat glossy faces. See the lighting techniques guide.

Code readers. Where the task is only reading codes, a fixed code reader chosen by code size, field of view, material and marking process is often simpler than a camera. See choosing a HIKROBOT code reader.

Trigger and reject. A photoelectric sensor or an encoder for the trigger; an air jet, pusher, diverter or line stop for the reject, with a confirmation sensor. See PLC integration.

3. Set up the camera and lighting

  • Mount camera and light rigidly at the agreed working distance, square to the surface unless the lighting needs an angle; shroud the station from daylight.
  • Connect and configure the camera with the manufacturer’s tool, for example HIKROBOT MVS for its cameras (IP address, firmware, image check) (HIKROBOT camera manual), or SCMVS for its smart cameras (HIKROBOT catalogue 2025 Q4).
  • Set focus and aperture on real products, then lock them.
  • On measurement stations, carry out the machine vision calibration: image a calibration target of known dimensions to convert pixels to millimetres and correct lens distortion, and record the result.
  • Set the exposure short enough to freeze the moving product, then the light intensity or strobing to match; keep gain low.
  • Check the trigger: one image per product, the product in the same place every time, at full line speed.
  • Save the camera settings and a set of reference images of good and defective products.

4. Build the algorithm

  1. Locate the product or feature first, usually with template matching or edge and shape search, so every other tool follows the product when it shifts.
  2. Inspect in regions tied to that position: measure edges and distances, read codes and text (OCR/OCV), check presence, count, or look for surface defects.
  3. Add deep learning only where rules cannot separate good from bad, for example variable surface defects; train it on labelled images and check it on a separate validation set of images not used for training.
  4. Combine the results in logic into one pass or fail, and decide what data goes to the PLC and the records.
  5. Set the limits on the sample set, including borderline samples, not on a single good part; record each limit and why it was chosen.
  6. Build one recipe per product and name it so the PLC can select it.

On HIKROBOT VisionMaster this means a graphical solution of locating, measurement, identification and defect-detection tools, with deep-learning and registration-learning modules where needed (HIKROBOT catalogue 2025 Q4). See VisionMaster algorithm tools and rule-based or deep learning.

5. Connect the PLC and reject

  • Wire or configure the signals: trigger, ready, result valid, pass/fail, alarm, recipe number; confirm NPN/PNP or the network protocol against the PLC.
  • Set up tracking to the reject by time delay or encoder, and test it at the slowest and fastest line speeds.
  • Test the reject and its confirmation on every failed test product.
  • Test every agreed fault response, including restart after a power interruption.

6. Test at FAT

At the factory acceptance test (FAT), before the station leaves the workshop, it runs the agreed samples at the agreed speed. The tests below are the minimum; the proposal sets the numbers and the pass criteria.

Factory and site acceptance tests for a vision station
TestHowPass when
DetectionEach defect type on marked defect samples, fed at known positions at line speedMissed defect rate as agreed
False rejectsA known number of good products, including normal variation, at line speedFalse reject rate as agreed
ChangeoverEvery product: recipe selected from the PLC or HMI, first product checkedCorrect recipe every time
RejectFailed products removed at the right position; confirmation sensor sees each oneEvery failed test product confirmed
CodesNo-read and wrong-content test packsTreated as fails, alarm raised
FaultsCamera offline, no result, air pressure low, bin full, confirmation failed, power lossAgreed response each time
RecordsCounts, results and images stored and retrievable per batchRecords complete

7. Install and test at SAT

  • Install in the agreed shutdown window; re-check mounting, focus and lighting on the line, since site light and vibration differ from the workshop.
  • Run real production for every product, repeat the detection, false reject, changeover, reject and fault tests, and fine-tune limits with the quality team, recording each change.
  • Run a challenge test with marked test packs, the way the site will do it in future.

8. Hand over

  • Train operators (daily use, cleaning, reference checks, alarms) and technicians (new products, backups, first-line fault finding).
  • Hand over recipe and configuration backups, the I/O list, the PLC interface description, operating instructions and the recommended spares list.
  • Agree the support plan: who to call, what is covered, response times.

See support after handover.

Commissioning records

Keep an audit trail of every change to limits and recipes after handover: who changed what, when and why. Keep the signed requirement, the bill of materials, camera settings and reference images, each recipe with its limits and reasons, the tested I/O list, the FAT record, the SAT report and the handover set. When something changes later, these records are what a technician works from. Printable record sheets, including a challenge test record, are in the inspection checklists.

// FAQ

Questions about commissioning

What does commissioning a vision system involve?
Mounting and setting up the camera and lighting on the line, building and tuning the inspection recipe for each product, connecting and testing the PLC signals, the reject and its confirmation, proving detection and false rejects at FAT and SAT, and handing over with training, backups and documents.
How do you choose between a vision sensor, a smart camera and a camera with a controller?
By the checks and views needed: a vision sensor for one simple check, a smart camera for one or two checks with more tools, and cameras with a vision controller or PC for several views, heavy processing such as deep learning, or many products. HIKROBOT’s own selection guide also weighs the algorithm capability of each series and the cycle time.
In what order are inspection tools usually built?
Locate the product first, so the other tools follow it when it moves; then measure, read or check presence in regions tied to that position; then combine the results in logic and send one result to the PLC. Deep-learning tools are added where fixed rules cannot separate good from bad.
How long does commissioning take?
It depends on the number of checks, products and views, and on the access to the line. The schedule is agreed per project after the sample test; most of the tuning is done before site, at FAT, so the shutdown for installation stays short.
Who signs off commissioning?
The plant signs FAT and SAT against the criteria agreed in the proposal: detection on defect samples, false rejects on good samples, changeovers, reject confirmation and fault responses.
// Next step

Commissioning a station, or taking one over?

Tell us what is on the line and what records exist. We will say what the commissioning or a review would involve.

Start an assessment