Key takeaway for buyers
A visual inspection cell is reliable when the photo window, part posture, lighting, robot access, and recheck loop are designed as one control system. The camera is only one component. The real production question is whether every part arrives in a repeatable condition, can be photographed inside a stable window, and can be routed correctly when the inspection result is uncertain.
For procurement and manufacturing engineering teams, the practical approval rule is clear: do not approve a vision cell only because one demo image looks clean. Approve it when the proposal explains how the part is presented, how the camera sees the required features, how false rejects and missed defects are handled, and how operators recover from abnormal states. EVST recommends treating inspection as a closed-loop station, not as a camera add-on.
This article covers robot-assisted visual inspection for automotive parts, machined parts, assembly checks, and similar B2B production cells. It does not cover laboratory metrology, consumer camera inspection, or standalone handheld inspection tools. For broader automation planning context, see the EVST industrial robot blog.
Define the photo window before choosing the camera
The photo window is the controlled moment when the part, camera, lighting, robot, fixture, and controller are all ready for image capture. If this window is unstable, the inspection software receives inconsistent images. A more expensive camera may improve resolution, but it cannot fully compensate for uncontrolled posture, glare, shadow, vibration, or changing distance.
In practice, photo-window problems appear as repeated threshold tuning, unstable pass/fail results, excessive manual review, and unexplained differences between shift conditions. These are process problems before they are software problems. The quotation should therefore show how the photo window is created and maintained.
| Photo-window checkpoint | What to verify | Evidence to request |
|---|---|---|
| Part stop position | The part reaches a repeatable inspection position. | Fixture drawing, locating surfaces, and tolerance assumptions. |
| Camera distance | Working distance and field of view cover the target features. | Camera layout, lens choice, and sample image coverage. |
| Lighting condition | Light angle and intensity control glare, shadow, and texture variation. | Lighting plan and sample images under normal conditions. |
| Motion state | The part is stopped or moving within an acceptable blur limit. | Exposure setting, robot hold time, or trigger timing. |
| Signal timing | The PLC, robot, and camera agree when the image is captured. | Trigger sequence, ready signals, and timeout logic. |
A good design does not need to overcomplicate every station. Some cells can use fixed cameras and simple fixtures. Others need robot repositioning, multiple views, or controlled part rotation. The decision should follow feature visibility, tolerance, surface condition, and takt target.
Part posture is the first source of false results
Part posture means the real position and orientation of the workpiece when it enters the inspection area. If the posture varies too much, the system may fail for reasons unrelated to the actual defect. The part can be good but rejected because the camera sees the wrong angle. The part can also be defective but missed because the defect area is hidden or outside the trained inspection window.
For a visual inspection cell, posture control should be reviewed before algorithm selection. A clear proposal should explain how the part is located, what variation is allowed, and what happens when the part arrives outside the accepted range.
| Posture factor | Good sign | Risk signal |
|---|---|---|
| Locating method | The part has a defined stop, nest, clamp, or datum contact. | The cell assumes the incoming posture is always correct. |
| Allowed variation | The quote states the position and angle tolerance used for inspection. | The proposal only says “vision can recognize it.” |
| Robot presentation | Robot path, gripper, and fixture keep the feature visible. | The robot may cover the feature or create shadow. |
| Part surface | Glare, oil, burrs, texture, and color variation are considered. | Sample images come from only one clean part. |
| Out-of-window handling | The controller detects and routes posture failures. | Operators discover posture problems after repeated rejects. |
EVST usually separates posture control into mechanical control, motion control, and image control. Mechanical control fixes the part where possible. Motion control presents the part at the correct angle. Image control compensates for remaining small variation. This order is more stable than asking software to absorb uncontrolled mechanics.
Lighting and contrast decide inspection stability
Lighting is often the least visible part of a vision proposal, but it is one of the strongest drivers of inspection stability. A defect that is obvious in one light angle may disappear under another. A reflective machined surface may create bright spots that look like defects. A dark casting surface may hide scratches unless the light direction is controlled.
The quote should not only name a light type. It should define what the light is meant to reveal and what variation it must suppress.
| Inspection target | Lighting goal | Common risk |
|---|---|---|
| Missing feature | Create clear contrast between feature and background. | Fixture shadow hides the target area. |
| Surface defect | Reveal scratches, dents, burrs, or contamination. | Glare looks like a defect or hides a defect. |
| Assembly presence | Make fasteners, clips, holes, or labels visible. | Color variation shifts threshold results. |
| Position check | Make datum edges or reference marks stable. | Edge detection changes with part angle. |
| OCR or code check | Keep printed or marked information readable. | Motion blur or surface reflection causes misread. |
According to the International Federation of Robotics, 542,000 industrial robots were installed globally in 2024, with annual installations staying above 500,000 units for the fourth straight year. Source: IFR World Robotics 2025. In this environment, many buyers are no longer asking whether automation is possible. They are asking whether a specific automation station can produce repeatable evidence. Lighting, posture, and trigger logic are central to that evidence.
Robot access and camera access must be planned together
Robot access and camera access can conflict. The robot must grip, move, and release the part. The camera must see the target features. The light must illuminate the same features. The fixture must hold the part without blocking the view. If these requirements are designed separately, the final station may work in a demonstration but fail under production speed or part variation.
A useful layout review asks four questions:
- Can the robot reach the inspection pose without covering the target feature?
- Can the camera see the required surface with enough margin?
- Can lighting be installed and serviced without interfering with motion?
- Can operators clear faults and maintain the station safely?
| Access item | Why it matters | Required quotation detail |
|---|---|---|
| Robot envelope | Prevents path conflict with camera, light, and fixture. | Reach layout, interference check, and service position. |
| Camera mount | Keeps image geometry stable over time. | Mount design, vibration control, and adjustment method. |
| Light mount | Keeps illumination repeatable and serviceable. | Light position, replacement access, and cable route. |
| Fixture clearance | Prevents the fixture from hiding inspection features. | Fixture drawing with camera view lines. |
| Maintenance access | Reduces downtime during cleaning and adjustment. | Safe access point and lockout method. |
In practice, EVST reviews the station from the image outward. First define what must be seen. Then define how the part must be held. Then define how the robot moves without disturbing the view. This sequence reduces late-stage redesign.
Recheck loop is the difference between inspection and production control
A pass/fail signal is not enough. Production needs a recheck loop that defines what happens after a suspicious image, a posture failure, a camera timeout, or a robot handoff issue. Without that loop, operators become the hidden control system. They decide what to recheck, when to override, and how to restart. That creates unstable quality records.
The recheck loop should distinguish at least four states:
- Normal pass.
- Confirmed reject.
- Suspected reject requiring recheck.
- Inspection invalid because the image or posture is not trustworthy.
| Result state | Typical cause | Recommended handling |
|---|---|---|
| Pass | Feature is visible and within rule. | Continue to next process and record result. |
| Reject | Defect or missing feature is confirmed. | Route to reject area and keep traceable evidence. |
| Recheck | Confidence is low or image condition is marginal. | Move to a second view, second trigger, or manual review station. |
| Invalid | Camera, light, trigger, or posture failed. | Stop, alarm, or route to recovery without calling it a quality reject. |
| Timeout | Camera or PLC did not complete within the expected window. | Hold part, log event, and prevent unverified release. |
This distinction matters because an invalid image is not the same as a bad part. A robust visual inspection cell protects both quality and production flow by separating product condition from inspection-condition failure.
Safety and integration scope
Industrial robot safety must be handled at the cell level. ISO 10218-1:2025 covers safety requirements for industrial robots as partly completed machinery, while ISO 10218-2:2025 covers industrial robot applications and robot cells, including integration, commissioning, operation, maintenance, and decommissioning. Sources: ISO 10218-1:2025 and ISO 10218-2:2025.
For a visual inspection cell, safety review should include normal production, camera cleaning, light adjustment, fixture service, part clearing, reject-bin access, and teaching or setup operations. A camera or light may need frequent adjustment in early commissioning. If access is not planned, maintenance becomes slow or unsafe.
| Integration area | Engineering question | Acceptance evidence |
|---|---|---|
| Robot and PLC | Who owns trigger, ready, pass, reject, recheck, and fault signals? | Signal list and sequence chart. |
| Vision controller | How are recipes, thresholds, and image records managed? | Recipe plan and data retention rule. |
| Safety system | How are doors, scanners, interlocks, and recovery positions defined? | Safety layout and reset sequence. |
| Reject handling | Where do suspect parts go, and how are they identified? | Reject path, bin logic, and operator prompt. |
| Commissioning | What abnormal cases are tested before handover? | Trial plan with pass, reject, invalid, timeout, and restart cases. |
What to include in the request for quotation
A weak request for quotation asks for “one robot vision inspection station” and expects the integrator to infer the process. A strong request gives enough information to design the station around the real inspection problem.
Prepare these inputs before comparing suppliers:
- Part drawing or 3D model.
- Photos or videos of real incoming part posture.
- Inspection feature list and defect examples.
- Acceptable tolerance, reject rule, or reference sample.
- Target takt and allowed recheck time.
- Surface condition range, including oil, dust, color, glare, and burrs.
- Upstream and downstream signal list.
- Image record requirement for traceability.
- Reject and recheck routing rule.
- Maintenance access and cleaning requirement.
| RFQ input | Why it improves the proposal | What happens if it is missing |
|---|---|---|
| Defect examples | Helps choose camera, light, and algorithm. | The quote may solve the wrong inspection task. |
| Posture range | Defines fixture and robot presentation needs. | False rejects can appear after commissioning. |
| Takt target | Forces timing separation between motion and image processing. | Cycle claims become difficult to compare. |
| Recheck rule | Turns uncertain images into planned production states. | Operators handle exceptions manually. |
| Traceability rule | Defines image storage, result log, and quality evidence. | Quality records may not support later review. |
Decision matrix for approval
Use the following matrix before approving a supplier proposal. It is intended for procurement, manufacturing engineering, and quality teams that need a practical pass or revise decision.
| Decision area | Approve when | Revise when |
|---|---|---|
| Photo window | Camera, lighting, part, trigger, and hold state are defined. | Only camera model and resolution are specified. |
| Part posture | Locating method and allowed variation are clear. | The proposal assumes perfect incoming posture. |
| Lighting | Light purpose, angle, and service access are explained. | Lighting is only listed as a component. |
| Recheck loop | Pass, reject, recheck, invalid, and timeout states are separated. | Every non-pass event is treated as a simple reject. |
| Commissioning plan | Trial cases include good parts, defect samples, posture variation, timeout, and restart. | Acceptance is based on one clean demonstration. |
This matrix deliberately avoids starting with camera brand. Camera, lens, and algorithm matter, but they should be selected after the inspection window and production loop are defined.
How EVST frames this cell
EVST treats a visual inspection cell as a linked system of robot, gripper, fixture, camera, lighting, PLC logic, safety access, recheck routing, and commissioning evidence. The valuable deliverable is not only a camera specification. It is a station concept that explains how the part is presented, photographed, judged, routed, and recovered.
The practical workflow is four-step. First, define the feature and defect boundary. Second, stabilize posture and lighting. Third, map robot motion, camera trigger, and PLC signals. Fourth, test pass, reject, recheck, invalid, timeout, and restart states before handover. This keeps the quote tied to production reality.
FAQ
What is the first thing to check in a visual inspection cell?
Check the photo window first. If the part, camera, light, trigger, and hold state are not stable at the capture moment, the inspection result will be unstable even with a high-resolution camera.
Does every visual inspection cell need a robot?
No. A fixed camera and fixture may be enough when the part arrives in a repeatable position. A robot becomes useful when the part must be repositioned, presented at multiple angles, transferred between stations, or integrated with handling.
How should recheck logic be designed?
Separate product rejects from inspection-condition failures. A confirmed defect should go to reject handling. A low-confidence image, posture failure, or timeout should trigger recheck, recovery, or invalid-result handling instead of being mixed into normal reject data.
What should be included in the acceptance test?
Test good parts, known defects, marginal posture, glare or shadow variation, camera timeout, robot handoff delay, reject routing, operator recovery, and restart. One clean pass cycle is not enough evidence for production approval.
How can buyers compare two visual inspection proposals?
Compare assumptions first. The stronger proposal defines photo window, posture control, lighting method, trigger timing, recheck routing, safety access, and commissioning cases. After those items are clear, compare equipment, price, delivery time, and service support.
Final checklist
Before approving a visual inspection cell, confirm that the proposal answers these five questions:
- What exact feature or defect must be seen?
- How is the part posture controlled before image capture?
- What makes the photo window repeatable across shifts?
- What happens after pass, reject, recheck, invalid, and timeout states?
- How can operators clean, adjust, recover, and maintain the station safely?
If these answers are clear, camera selection becomes a controlled engineering decision. If they are missing, the inspection task is still under-specified, even if the sample image looks sharp.