Machine vision / Article
What AOI Actually Does: Six Inspection Jobs on One Line
AOI inspection, short for automated optical inspection, is a camera and software system that checks every part on your line for defects without a person looking at it. The textbook answer stops there. The useful answer is that AOI does six concrete jobs: it detects defects, measures dimensions, verifies the right parts are present, reads codes and text, guides robots to a part, and sorts pass from fail. This article walks each job with a real defect example, so you can see exactly what an inspection camera is doing on the floor.
The numbers at a glance
01 / The definition
AOI is a camera that inspects every part
Automated optical inspection is, at its core, a non-contact method that uses camera systems and image-processing algorithms to examine manufactured components for surface defects, dimensional deviations, and assembly errors without human visual intervention. In electronics it has become a standard in-line step, verifying printed circuit boards and surface-mount assemblies before anything moves downstream. The point is simple: a machine looks at every unit at line speed, where a person can only look at a sample.
But "inspects for defects" flattens what the system really does. A modern AOI line is asked to answer several different questions about the same part in a single pass, and each question is a distinct job with its own logic. Naming those jobs is the fastest way to understand what you are actually buying, and where a camera helps your line most. So here they are, the six jobs CODETRACE builds an inspection cell around.
02 / The six jobs
Detect, measure, verify, read, guide, sort
Every inspection task on your floor maps to one of six verbs. The same camera can do several of them on one part in one pass. Each one catches a different kind of problem:
Not every line needs all six. A bare-board maker leans on detect and measure. A final-assembly line leans on verify, read, and sort. The value of the framework is that it lets you say precisely which jobs your process needs, instead of buying a generic "inspection machine" and hoping it fits.
03 / How it decides
Every job compares against a known-good reference
Under all six jobs sits one mechanism. The system captures a high-resolution image under controlled lighting, then compares each measured feature against a reference derived from the part's design data or a golden sample. Captured images are analyzed by comparing them against CAD data, Gerber files, or a known-good reference, using classical methods like template matching, thresholding, and blob analysis to segment and classify each region.
The limitation of pure rule-based comparison is false calls, where a slightly different but perfectly good part gets flagged. That is why modern systems add a learned layer. AI models trained on your defect images significantly reduce false positives and negatives, reaching accuracy rates of up to 97 percent and higher, and adapt to variation that would trip a fixed threshold. We cover that trade-off in depth in our companion piece on reducing AOI false calls without letting defects escape.
04 / The accuracy gap
Six jobs, done faster and more consistently
The reason to hand these jobs to a camera is not novelty, it is the gap against manual inspection. A comparative study of electronics inspection found AOI delivers superior accuracy of 98 to 99 percent versus 85 to 90 percent for manual inspection, and efficiency of 5,000 or more components per hour versus 500 to 800. A person doing the same six jobs by eye slows down, gets less consistent as the shift wears on, and inspects a sample rather than every unit.
Speed matters as much as accuracy, because inspection sits in the critical path of the line. AOI combines machine vision hardware, lighting, and pattern recognition to reach defect detection rates and throughput speeds that manual inspection cannot match at production scale. That is what lets a single camera cover detect, measure, verify, read, guide, and sort without becoming the bottleneck it was meant to protect.
05 / On your parts
Tuned to your tolerances, not a generic dataset
An inspection system is only as good as its fit to your part. A model tuned to someone else's components in another country will over-reject on yours, drowning operators in false calls until they start ignoring it. This is where CODETRACE machine vision and AOI is built differently. The models are trained on your components and your defect classes, so the six jobs run to your spec, not a template. CODETRACE is a member of the NVIDIA Inception program, and the systems run on the floor at line speed.
The hardware behind the jobs is built in-house. JOVIS handles robotic vision inspection and micron-level geometry analysis for the measure and detect jobs, and AONIA is our optical inspection machine for in-line verify, read, and sort. Both are deployed and tuned on-site across Selangor and the Klang Valley, so the headline accuracy becomes a number you actually see on your own line. Bring us a defect you keep missing, and we will tell you honestly whether vision can catch it.
06 / Where to start
Pick the one job that hurts most
You do not automate all six jobs on day one. Start with the one that is costing you now: the defect that keeps escaping to customers, the measurement your operators cannot hold, the mislabelled lot that triggered a recall. Put a camera on that single job, run it in parallel with your existing check, and measure the escape rate on both. The data tells you plainly where the system is stronger before you widen its remit.
Once one job is trusted, the others fold in on the same cell, because the camera and the lighting are already there. Inspection is also one station in a larger picture, sitting inside factory automation alongside handling, packing, and palletizing. Get the inspection job right first, and the rest of the line has a quality gate it can trust.
Detect, measure, verify, read, guide, sort. Name the job, then automate it.
FAQ / AOI inspection
Questions, answered.
01What is AOI inspection?
AOI stands for automated optical inspection. It is a camera and software system that examines every part on a production line for surface defects, dimensional deviations, and assembly errors without a person looking at it. Instead of one textbook task, AOI does six real jobs on the line: detect defects, measure dimensions, verify the right parts are present, read codes and text, guide robots to a part, and sort pass from fail.
02What can AOI detect?
Scratches, cracks, contamination, missing or wrong components, solder bridges and voids, misalignment, and dimensional deviations that fall outside tolerance. Beyond finding defects, AOI also measures features against spec, verifies component presence and polarity, reads date codes and barcodes, and classifies parts into pass or fail bins. What it looks for is defined by your own spec and defect library, not a generic template.
03How is AOI different from machine vision?
Machine vision is the broad field of using cameras and software to make decisions on a factory floor. AOI is machine vision applied specifically to quality inspection, checking that a manufactured part is built correctly and free of defects. Every AOI system is a machine vision system, but machine vision also covers robot guidance, code reading, and sorting that are not strictly inspection.
04Is AOI more accurate than human inspection?
Yes. Comparative studies of electronics inspection put AOI accuracy at 98 to 99 percent against 85 to 90 percent for manual inspection, at throughput of 5,000 or more components per hour versus 500 to 800 for a person. A camera does not fatigue, inspects every unit rather than a sample, and holds an image record of every decision.
05Where does AOI fit in a Malaysian factory?
For semiconductor, electronics, automotive, and food producers in Selangor and the Klang Valley, AOI is the in-line quality gate that catches defects before they ship to an export customer. CODETRACE deploys optical inspection on-site with models trained on your own parts and defect classes, and is a member of the NVIDIA Inception program. The result is a quality check that runs at line speed rather than a lab bench.