In production we treat AOI as a controlled measurement system rather than a simple defect screen. Under ipc-9716 the factory sets lighting, camera resolution, detection thresholds, and program recipes so that false calls stay low while real defects still get caught. We qualify the machine with known-good and known-defect samples, run measurement system analysis on critical features, and lock the recipe before the line runs. Operators then review only the residual calls instead of every board.
Where Component and Lighting Variation Forces Threshold Compromise
What we typically see is that body color, marking contrast, lead finish, and solder fillet shape all drift between lots. Board color and local warpage change how light reflects. The same algorithm that correctly flags a missing 0402 on one lot starts calling every slightly darker marking as wrong part on the next. Fine-pitch leads and small fillets sit near the resolution limit of the camera, so edge detection becomes sensitive to lighting angle and intensity. Most factories therefore keep a library of golden boards and deliberately tune thresholds against the worst acceptable variation rather than the ideal image.
ipc-9716 requires the system to demonstrate detectability and repeatability for the defects that matter to the product class. In practice that means we cannot leave the machine on default settings. Lighting intensity, camera focus, and threshold windows have to be verified on the actual assemblies we build, not on a generic calibration plate alone.
What Typically Appears When AOI Runs Without Process Control
When the recipe is not controlled, false-call rates climb above 5–10 percent on dense boards. The line stops while operators clear the queue, cycle time stretches, and real defects get buried in the noise. We have seen polarity-reversed tantalum capacitors and solder bridges on 0.4 mm pitch QFNs escape because the threshold was loosened to cut the false-call rate. Escape of those defects shows up later as functional failures or field returns. On the other side, overly tight settings generate continuous stops, operators start bypassing the machine, and the process control value disappears. Either path drops first-pass yield and creates shipment delays while we rework or scrap boards.
IPC-A-610 still defines what is acceptable for the product class. AOI thresholds that drift away from those criteria either reject good product or pass product that should have been a process indicator or defect. The mismatch only becomes visible during customer audits or when reliability failures appear.
How Factories Lock AOI Capability Before Volume Release
Most factories handle this by treating the AOI program as a controlled document. We first capture images under the production lighting set and build the recipe against a set of sample boards that include both good assemblies and deliberately introduced defects (missing parts, wrong orientation, bridges, insufficient solder). Thresholds are set so that Class 2 and Class 3 requirements from IPC-A-610 are respected: process indicators may be logged for review while defects force a stop. After the recipe is complete we run a measurement-system check—repeatability on the same board and reproducibility across shifts or machines—so that the call rate stays stable.
Calibration intervals for cameras and lights follow the machine maker's schedule plus any tighter internal rule. We log false-call and escape rates as process indicators and review them weekly. When a new component lot changes appearance we re-verify the recipe on a small sample before releasing the change. Data from the AOI is tied to the board serial number so that trends can be traced back to a specific feeder, stencil, or reflow profile. This is the practical content of the process-control and verification requirements in ipc-9716.
Where SPI and X-ray Cover What AOI Cannot See
AOI looks at the finished assembly after reflow and catches missing parts, wrong parts, polarity, offset, tombstoning, and visible solder defects. SPI measures paste volume, height, area, and offset before placement and therefore catches stencil or print problems that AOI never sees. X-ray is required for hidden joints—BGA, QFN, or LGA voids and shorts—because optical cameras cannot look under the package. In a controlled line the three systems are sequenced so that each covers its own defect set; we do not expect AOI to replace either of the others.
When the Factory Allows Relaxed AOI Control
On Class 1 products or very simple boards with large components and generous spacing we sometimes accept a higher residual false-call rate and lighter MSA documentation, provided the escape risk stays low and the customer agrees. Prototype or low-volume runs may also run with a manually reviewed recipe rather than full statistical qualification. The trade-off is clear: we gain speed on the front end but lose the ability to prove consistent detection later if a quality problem appears. For Class 3 or high-reliability work the full ipc-9716 controls—calibration records, MSA results, and ongoing false-call tracking—remain in place.
What the Production Checklist Actually Checks
Before releasing a new AOI program we confirm the recipe against sample boards containing the defects that matter for the product class, record the lighting and threshold settings, and complete the MSA. During production we track false-call percentage and escape rate, review any new calls daily, and re-qualify the program when a component or process change occurs. Calibration and maintenance records stay with the machine. These steps keep the system inside the process-control window defined by ipc-9716 and aligned with the acceptability criteria of IPC-A-610.
