Keep each inspection system tied to its camera, lens, lighting and trained image set.

A vision system starts as a question: can it see this defect at this speed? The answer comes from sample parts, a camera, a lens and a lighting angle tried on a bench, and the customer pays for a result on a moving line. Proof comes from the bench, not the brochure.

After the sale, the company installs, trains and supports the system, including later retraining as parts change. The software keeps the sample study, the bill of materials, the image library and the support history under one system number. When a customer asks why false rejects have increased since last quarter, the answer lies in the model version, the lighting hardware and the last parts change, and these should sit together.

A manufacturer sends sample parts with good and bad examples. Application engineers set up a bench with camera, lens, light and trigger, capture images and record what could and could not be detected. A feasibility report states resolution, field of view, cycle time and what remains uncertain. The report sets expectations early. The study may be repeated with a second lens or a different light angle, and each attempt matters, because the customer will later ask why one setup was chosen over the others.

After an order, the company selects cameras, lenses, filters, light controllers and an industrial PC, then builds and tests the unit. Engineers collect images on the customer’s line and train or tune the inspection programme. The system goes live with a reject output, reports and an operator screen. Months later, a new part or a changed material arrives and the models must be retrained. Then the cycle repeats.