Building a vision inspection cell: a Cognex + UR walkthrough

Vision inspection cells are one of our most common builds. The reasons are straightforward: customers need 100% inspection rather than sampling, customers need documented traceability for regulatory or customer-audit reasons, or customers need to free a senior operator from a manual inspection station. The technology is mature, the integration is well-understood, and the payback math typically works inside a year.

Here's how we structure a typical Cognex + UR vision cell from feasibility through commissioning.

Feasibility study (1-3 weeks)

We start with a paid feasibility study — usually $4,000 to $8,000, credited against the build if the customer proceeds. The deliverable is a fixed-price quote, a controls architecture, a part-presentation strategy, and a validation approach.

Most of the feasibility work is on the inspection problem itself. We collect representative parts (good and bad), test the inspection logic in our shop on a benchtop In-Sight setup, and quantify the achievable false-pass and false-fail rates. If deep learning is in scope, we build a baseline ViDi model on the customer-supplied training data and report the validation accuracy. The customer sees real numbers on real parts before signing the build PO.

Cell architecture (typical)

Most of our vision cells share a common architecture. A part feeder (usually customer-supplied or specified) presents parts at a known orientation. A UR cobot picks the part with a Robotiq gripper or vacuum end-effector, presents it to one or more vision stations, and routes it to a pass conveyor or a reject chute based on the inspection result.

Vision is anchored on Cognex In-Sight smart cameras — typically 7000- or 9000-series for high-resolution applications, 2800-series for general-purpose work. We program the inspection logic in In-Sight Explorer, with custom scripts for any decision logic the off-the-shelf tools don't cover. For deep-learning applications, we run ViDi on an In-Sight D900 or on a separate edge inference box if the model is too large.

Safety is handled with Banner EZ-SCREEN light curtains or AG4 area scanners, depending on cell geometry. Controls integration is to the customer's existing PLC — usually Allen-Bradley CompactLogix or ControlLogix, occasionally Siemens.

Build and FAT (8-12 weeks)

Cells get built in our Tampa shop. We mock up the upstream parts feeder and the downstream conveyor or reject path, then run real cycles on customer-supplied parts before the cell ships. Factory acceptance testing covers the inspection accuracy targets agreed in feasibility, the cell cycle time, and the safety-system function.

We invite the customer to witness the FAT either in person or via video. About 70% of customers come in person; it usually shortens commissioning by 2-3 days because the customer engineering team has already seen and accepted the cell behavior.

Commissioning (3-7 days)

Site commissioning starts with mechanical install (typically half a day for a standard cell footprint), then PLC integration with the customer's controls (one to two days), then inspection-logic tuning on real production parts (one to three days), then operator training (half a day to a full day).

The most common surprise during commissioning is variation in real production parts versus the parts we ran during feasibility. We expect this; we build inspection-logic margin into the FAT. But occasionally we see a part variation in production that wasn't represented in the feasibility samples — usually a process drift from a previous shift, occasionally a supplier change. We adjust the inspection model and validate the new behavior with the customer's quality team before sign-off.

Validation and handoff

For regulated industries (pharma, aerospace, medical device), the cell delivery includes IQ/OQ/PQ documentation. We've structured packages for FDA-inspected sites and AS9100D-certified suppliers. Documentation is part of the deliverable, not an afterthought.

Operator training is hands-on — two shifts of operators, working through pass and fail scenarios, learning how to clear jams and how to interpret the cell's status display. We leave a one-page laminated runbook at the cell and recorded video walkthroughs for new-hire training. The first 30 days post-commissioning include unlimited remote support; we typically take three to five operator questions in that period.

Typical project metrics

A representative Cognex + UR inspection cell project, end-to-end: 14 weeks from kickoff to commissioning, $180,000 to $350,000 fully-installed depending on complexity, 9-14 months ROI on labor savings or quality cost avoidance. Inspection accuracy on the cells we've shipped: 99.5% to 99.97% first-pass yield depending on the inspection problem, with false-fail rates below 0.5% in production.