Automated visual inspection for manufacturers

Machine vision works when you can define the defect. If two of your own inspectors disagree about what a defect looks like, no camera will settle it for you.

Free, about two minutes, no phone number. You get an honest no when a robot is the wrong answer.

Who this is for

You have people doing visual inspection at a station, escapes are reaching customers, or a customer requires 100 percent inspection you are currently doing by eye.

The signs this fits you

• A dedicated manual visual inspection station.
• Defects escaping to customers despite inspection.
• Inspector fatigue late in a shift, with measurable differences in catch rate.
• A 100 percent inspection requirement you are meeting with people.

What the cell looks like

Either fixed camera stations that the part passes through, or a robot that presents the part to a camera. Lighting is half the project and gets a fraction of the attention it deserves. On the software side there are two families: classic rule-based vision, which measures things you can specify, and trained models, which learn from examples of good and bad. Trained models need a real defect library, which most shops discover they do not have.

What it costs and what it saves

Illustrative, and the widest spread of any application here. A single fixed camera checking presence or absence is a different project from a trained model on a reflective surface.

The return is escapes prevented, containment avoided, and inspection labor. Be honest about false rejects: a system that catches every defect but rejects good parts at a few percent can cost more than the escapes did. Set an acceptable false reject rate before anybody quotes.

When this is the wrong answer

• Your defects cannot be defined consistently. Run the test: have two inspectors grade the same 50 parts and compare.
• Your surfaces are reflective or variable and there is no budget for lighting engineering.
• You do not have, and cannot produce, a library of real defect samples for a trained model.
• The escape rate is low enough that the system cannot pay for itself even if it works perfectly.

What an engineer has to verify

This is the checklist that keeps every number above a range. None of it can be settled from a website, and all of it gets settled before anyone quotes.

• A written defect catalog with real samples of each type.
• Inspector agreement rate on the same parts, measured.
• Acceptable false reject rate, agreed before quoting.
• Part presentation and whether orientation can be controlled.
• Lighting approach, which usually needs its own engineering.
• Data retention requirements if a customer wants inspection records.

Safety determinations are not on this list because they are not ours to make. Cell design, guarding, and the risk assessment belong to an independent certified integrator, every time.

What the research says

Peer reviewed work from our research library on this application. These are studies of other plants and other settings, not a prediction about yours, and the excerpts are the authors' own words. Every title links to the source so you can check us.

• Automated visual inspection in the semiconductor industry: A survey Szu-Hao Huang and Ying-Cheng Pan (2014), Computers in Industry
• Artificial Intelligence-Based Smart Quality Inspection for Manufacturing Sarvesh Sundaram and Abe Zeid (2023), Micromachines

  From the abstract: “In today's era, monitoring the health of the manufacturing environment has become essential in order to prevent unforeseen repairs, shutdowns, and to be able to detect defective products that could incur big losses.”

• How and why we need to capture tacit knowledge in manufacturing: Case studies of visual inspection Teegan Johnson et al. (2018), Applied Ergonomics
• Human Factors in Visual Quality Control Agnieszka Kujawińska and Katarzyna Vogt (2015), Management and Production Engineering Review

  From the abstract: “Visual control (visual inspection) is often used in production because - in comparison to other kinds of control - it is relatively easy to conduct.”

• Moving towards in-line metrology: evaluation of a Laser Radar system for in-line dimensional inspection for automotive assembly systems Ercihan Kiraci et al. (2016), The International Journal of Advanced Manufacturing Technology

  From the abstract: “The increasing interest towards intelligent systems has led to a demand for the development of zero-defect strategies, with a paradigm shift from off-line and dedicated to in-line metrology with integrated robotic systems.”

The full machine vision and inspection on the floor shelf has 14 works.

Real examples

See it run

Two clips of this job actually running, from two different manufacturers. Both are the makers' own marketing footage, showing a cell under good conditions with a part it was set up for, so watch them for what the motion looks like rather than as evidence of what your rate would be.