What is AOI beyond PCB? A 2026 guide for non-SMT lines
Per IPC's 2025 industry report, 92 percent of automated optical inspection installations still sit on surface-mount technology lines. The other 8 percent is quietly where the fastest AOI growth is happening: automotive welds, pharma blister packs, food bottling, and furniture finishing. This post is about that other 8 percent.
AOI beyond PCB is not a rebrand of the SMT toolkit. The optics, the training approach, and the commercial model are all different. If you are evaluating AOI for a non-electronics line, most of the assumptions from the SMT playbook will lead you into a deployment that is overbuilt and under-delivering.
AOI outside electronics: where it is quietly winning
Three verticals account for most of the growth. Food and beverage bottling uses AOI for label registration, fill level, and cap seating. Pharma uses it for blister pack completeness, print legibility on unit doses, and tamper-evidence. Automotive uses it post-weld and at the paint-booth exit for cosmetic surface inspection.
The pattern across these three is consistent. Each one moved from manual visual checks to AOI because their traditional rule-based machine vision setups either could not handle product variation or cost too much to reconfigure when the line changed over.
Three things SMT-era AOI got wrong for non-PCB lines
First, fixed lighting. SMT AOI assumes a green solder mask, a known board orientation, and consistent component geometry. Food labels ship in 40 SKUs per line. Automotive BIW welds sit under factory lighting that shifts all day. Fixed, bolted-down lighting rigs that work for PCBs break immediately off-electronics.
Second, programmatic defect rules. SMT-era AOI inspects by asking whether a component is within tolerance boxes. That works when the defect taxonomy is closed, like tombstoning or solder bridging. It falls apart when the defect is a wrinkled label or a paint run, where the shape of the defect is the variable.
Third, expensive recipe switching. Every new product variant on an SMT line costs engineering time to re-program the AOI recipe. Most non-PCB lines have more SKUs per shift than an SMT line sees in a month.
What modern AOI looks like
Modern AOI beyond PCB uses AI models that learn from images of good and bad parts, not from hand-coded geometric rules. The optics are closer to consumer camera specs than industrial. And the deployment model is under €1,000 of off-the-shelf hardware rather than a two-year line retrofit. For more detail on the underlying tech, see what AI visual inspection actually is.
At Enao Vision we build AOI stations on consumer-grade iPhone hardware in food-grade stainless housings. The iPhone runs on-device Core ML inference so the line does not wait for a server round trip. Sites without wired ethernet run on 5G hotspots. A line lead can stand one up in four hours and onboard the first defect class in five days. No ML team required.
When AOI beats conventional machine vision, and when it does not
AOI beats conventional machine vision when the defect set is open-ended or when the product variant count is high. Examples are bottling lines with 20 label SKUs, snack packaging with seasonal artwork, or pharma lines that change over between batches hourly.
Conventional machine vision still wins when the defect set is closed, the geometry is tight, and the throughput is above 2,000 parts per minute. Cap torque measurement, SMT pick-and-place verification, and micro-dimensional checks under 50 microns are still the home turf of traditional systems.
If you want a head-to-head view, our own vendor comparison walks through the tradeoffs by vendor.
Picking an AOI vendor for your first non-PCB line
Four questions sort most vendors in one conversation.
One, can they show a live deployment in a non-SMT context? If all the references are electronics lines, the commercial pricing and the optics will be wrong for you.
Two, do they price the hardware separately from the model licensing? Vendors that bundle a EUR 20,000 camera box into a EUR 30,000 annual software fee are pricing like the old AOI world.
Three, can you run a paid pilot on one line in under six weeks? If the pilot timeline is three months or longer, the vendor is building custom rather than deploying a platform.
Four, is the system usable by the line lead, or does it require a data scientist? The point of modern AOI is to move model tuning onto the shop floor. If the vendor demo requires their engineer to click through every retrain, you have bought a service contract, not a platform. See the 20 ways computer vision shows up in manufacturing for the breadth of lines where this pattern now works, and defect categories AI catches that humans miss for what the uplift actually looks like.
The AOI category outside PCB is about six times larger by part count than SMT. The tools to serve it finally exist. Most manufacturers just do not know the old SMT-era playbook is the wrong starting point.