How to choose a machine vision platform for a South African production line

What are you asking the station to decide?

Separate reading a code, matching it to an order, checking a missing component and measuring a dimension. A reader may decode a barcode without checking the order data. A presence sensor may find a missing cap without measuring its sealing surface. Define the required output and the consequence of a no-read or an uncertain result.

Which product families are comparable?

The matrix lines up the four vendors by device class, from their own pages: HIKROBOT’s product menu and catalogue (HIKROBOT product menu)(HIKROBOT catalogue 2025 Q4), Cognex’s In-Sight and logistics pages (Cognex In-Sight 3800)(Cognex logistics), KEYENCE’s vision sensor, vision system and code reader pages (KEYENCE vision sensors)(KEYENCE vision systems)(KEYENCE code readers), and Basler’s product documentation (Basler cameras). “Not in the pages we checked” means exactly that; ranges change, so confirm the current model in each quotation.

Software is part of the platform. VisionMaster offers graphical, SDK and operator-design development with deep-learning modules (HIKROBOT VisionMaster); Basler states that most pylon vTools need a licence after demo or evaluation periods (Basler vTool licensing). Ask every supplier for the licence list and how a licence is recovered after a PC or device is replaced.

How should you run a fair comparison?

• Give each candidate the same good and defective products, including borderline examples and normal packaging variation.
• Keep a separate evaluation set that was not used for setup or training. Record misses, false rejects and no-reads separately.
• Test actual exposure, speed changes, product spacing, glare, changeovers and stop/restart behaviour.
• Challenge the complete chain: trigger, image, decision, PLC tracking, reject confirmation and alarm.
• Agree who may adjust recipes and how a software update or replacement camera is revalidated.

How big should the challenge set be? A run with no misses bounds the miss rate; it does not prove zero. Using the Wilson method in NIST’s handbook (NIST e-Handbook, 7.2.4.1), the bounds are:

How do you score the offers?

Agree the weights before the results are in, score each offer on evidence, and keep the scores with the test records. The example uses invented offers to show the arithmetic: B wins on the sample result and price, A wins overall because it demonstrated failure handling and named local support. Your weights may differ; the discipline of fixing them first is what matters.

Which selection mistakes cost the most?

Most of these are not about the camera at all. They are about scope, evidence and responsibility, which is why the scorecard above weights demonstrated behaviour and support alongside the sample result.

What belongs in the South African quotation?

Request an itemised scope for camera, lens, light, mounting, enclosure, controller or PC, runtime and development licences, PLC work, reject mechanism, FAT, SAT, training and backup files. Ask for VAT treatment, freight, exchange-rate assumptions, travel, spare-part availability and the support responsibility in writing. We did not find published, verifiable information on any of the four vendors’ South African service channels, so ask each supplier to name the company that repairs and holds spares. We have not obtained equivalent current quotations from all four suppliers, so this guide gives no brand price ranking.

Our cost guide explains station budgets; PLC and reject integration explains the controls scope.

When should HIKROBOT be on the shortlist?

Our standard proposals centre on HIKROBOT cameras, readers and VisionMaster where the sample test supports them. That can suit a new station whose optics, controls and software are designed together. If your plant mandates another platform, has trained maintenance staff or a validated software estate, include that lifecycle constraint in the decision. We scope new stations; this comparison is not an offer to take over an existing third-party installation.

HIKROBOT vs Cognex · HIKROBOT vs KEYENCE · HIKROBOT vs Basler · HIKROBOT and Hikvision

Test portability and recovery before choosing a platform

Include a recovery exercise in the platform trial. Restore the approved recipe, camera configuration and calibration onto the agreed spare or recovery environment, then run the reference challenge set. The demonstration should include the actual PLC protocol and result acknowledgement. An exported project file is not sufficient if licences, firmware or external dependencies prevent it from running when production needs it.

Separate data portability from recipe portability. Images and result records may be exportable in common formats while proprietary inspection tools, trained models or operator screens still require the original runtime. Define which files the plant will receive and which changes require the integrator. A documented camera interface does not make every algorithm or controller interchangeable.

For a South African plant with a small maintenance team, the practical comparison includes access to backups, fault diagnostics, spare configuration and the agreed support response. Request written evidence from the organisation providing those services. Evaluate these obligations alongside the technical trial rather than assuming that a globally familiar brand automatically supplies local stock, application support or an identical service contract.

Compare platforms on a defined industry task

Start with the inspection contract, sample variation and presentation. Compare systems using the same visible features, operating conditions and acceptance method; device labels and processor specifications alone do not establish a successful station.

Record the selected device class, optics, lighting, licensing and data interfaces in the proposal. Test the hardest normal variation as well as obvious faulty samples.

Machined measurement · Tray inspection