Why Visual Inspection Becomes a Machine Task
On any line running faster than 60 parts per minute, relying on human eyes to sort defects becomes an expensive decision. Human inspector accuracy realistically sits between 70% and 85% under the best conditions, and it drops noticeably after the third hour of a shift and with repetition of the same task. A machine vision system, by contrast, holds steady performance above 99% across all three shifts and logs every rejected part with its image and rejection reason, giving quality management real data instead of verbal estimates.
When you import a production line from China, the vision system is usually offered as an add-on line item with a short spec that is not enough to judge it. Understanding the system's components lets you assess whether it genuinely fits your product or is simply an item that inflates the quotation without operational benefit.
The Four Components Behind Any Vision System
- Camera and sensor: they set how many pixels cover the part, and therefore the smallest defect that can be detected.
- Lens: it sets the field of view, working distance and depth of field, and it is the component most often chosen incorrectly in packaged offers.
- Lighting: it makes the defect stand out from the background, and it is usually the difference between a system that works and one switched off after a month.
- Software and processing unit: they run the measurement, matching and code-reading tools and send the accept or reject decision to the controller.
How to Calculate the Resolution You Need Before Choosing a Camera
The calculation takes three steps, and you should run it yourself before accepting any offer:
- Define the field of view width in millimetres: the largest part dimension plus a positioning margin of at least 15%.
- Divide the field of view width by the sensor's horizontal pixel count to get the resolution per pixel in millimetres.
- Make sure the smallest defect you want to catch spans at least three pixels, and preferably five for low-contrast defects.
A practical example: a 170 mm wide part plus a positioning margin gives a 200 mm field of view. A camera with 2448 horizontal pixels yields 0.082 mm per pixel. If the smallest defect to detect is a 0.3 mm crack, it spans about 3.6 pixels, which is acceptable. If the defect is 0.1 mm, the camera is not sufficient and you must move to a higher-resolution sensor or split the field of view across two cameras.
Comparing the Available Types of Vision Systems
| Type | Best Use | Approximate Speed | Cost Level |
|---|---|---|---|
| Simple vision sensor | Presence or absence of a component, cap and label checks | Up to 60 parts/min | Low |
| Integrated smart camera | Dimensional measurement and code reading on a single line | 60 to 300 parts/min | Medium |
| Industrial PC-based system | Multi-camera inspection and complex surface defects | 300 parts/min and above | High |
| Line-scan camera | Continuous surfaces such as rolls, sheets and pipes | Depends on belt speed | High |
Lighting Is the Most Neglected Element
Most vision systems that end up idle inside factories fail because of lighting, not the camera. Ring lighting suits flat, non-reflective surfaces; backlighting is the most accurate choice for dimensional and edge measurement because it produces a sharp silhouette; and low-angle grazing light reveals scratches and burrs on metal surfaces. The inspection station should also be shielded from daylight and ceiling lamps with a simple enclosure, because ambient light changing between day and night is a common cause of false rejects rising on the night shift.
Line Integration and Reject Rate
An inspection decision is worthless unless it reaches a real ejection mechanism. Ask the supplier to clarify three points in the offer: the trigger signal that fires the image capture, usually a proximity sensor or an encoder; the response time from capture to sending the decision to the controller; and the rejection mechanism, whether an air jet, a push arm or a diverter gate. Also set a contractually accepted false-reject rate: 2% on a line producing 20,000 parts a day means 400 good parts wasted every day.
Golden rule: never accept a vision system before running it on a realistic sample of your own production containing both good parts and known defective parts, and documenting the detection rate and false-reject rate in the pre-shipment inspection report.
Common Mistakes When Buying a Vision System with a Chinese Line
- Accepting a megapixel figure in the quotation without tying it to the field of view and the smallest required defect.
- Not sending real product samples to the supplier before the lens and lighting are selected.
- Overlooking the software interface language; some systems are delivered in Chinese only, making later parameter changes difficult.
- Failing to agree that your own team gets rights to adjust acceptance thresholds instead of the supplier holding them.
- Neglecting optical spare parts such as the lens and lighting unit, which have long lead times once the line is running.
A vision system pays for itself more through fewer returns and customer complaints than through saved sorting labour, but only if the specification is built around your product rather than a generic package. If you are planning to add a visual inspection station to your line or evaluating an offer from a Chinese supplier, contact our team to review the technical specification and choose the right configuration for your product before you sign.