Introduction: Online AI AOI inspection equipment places optical inspection directly inside the SMT conveyor flow, so boards move from placement and soldering into an automated inspection station.
When people first hear the word online in SMT inspection, they often imagine a camera connected to a network or a cloud service. On an SMT line, online means something more physical: the AOI machine is mounted in the conveyor path. Boards enter from upstream equipment, stop or pass through the inspection station, and continue to final assembly. That position changes how the machine is used, how fast it must work, and what kind of yield data it can produce. For SMT process beginners and equipment researchers, understanding this station role is the first step before comparing models or reading detailed specifications.
On a typical SMT line, boards pass through solder paste printing, component placement, reflow soldering, and sometimes wave soldering. An online AI AOI usually sits after one of those soldering steps and before final assembly or test. Its job is to inspect component presence, placement, polarity, and solder joints while the board is still in the production flow. Because the machine is inline, it does not wait for a batch of boards to be collected and carried to a separate room. The conveyor moves each board into the inspection station, and the machine decides whether the board can continue or should be flagged for review. The word online also implies a timing relationship. Placement machines, reflow ovens, and inspection stations all work to a line takt. If inspection takes much longer than the upstream process, it becomes a bottleneck. If it is too fast but unreliable, defects escape. Inline equipment communication standards such as The Hermes Standard describe how SMT machines exchange board-related signals so that conveyor handoffs stay coordinated. An online AI AOI therefore has to fit the conveyor loop, not just perform image processing in isolation. The IRIS ANT T7 is one example of an online AI AOI optical inspection equipment model built for this inline station role, with a conveyor-based design rather than a portable or offline format. Inline installation also brings basic facility needs such as space, floor load, power, and an air line, which shows that this is a fixed production station rather than a movable inspection tool.
A complete cycle looks simple from the outside: a board arrives, the machine captures images, software classifies what it sees, and the board leaves with a result. Inside that cycle, several steps must happen in a fixed order, and each step affects the next one. The cycle is where conveyor motion, optics, lighting, computation, and decision rules meet.
Conveyor handoff and board positioning are easy to overlook because camera specifications are easier to quote. A 5 μm pixel resolution can capture fine detail, and standard 0201 components or optional 01005 support depend on that optical capability. But if the board is not square to the camera, if it vibrates during capture, or if the conveyor transfers it at an inconsistent position, the software sees shifted or blurred images. Those images can look like missing parts, tombstoning, or solder defects when the board is actually fine. Good inline AOI design therefore combines a stable conveyor interface, board clamping or stopping, and repeatable positioning with the optical system. The inspection station must hand off boards smoothly from upstream and downstream equipment so that the line keeps moving.
Image capture starts with controlled lighting and a camera that turns reflected light into digital data. On real boards, solder joints are reflective, silkscreen characters sit close to pads, and board colors vary between suppliers. Traditional rule-based AOI often flags these normal variations as defects, which creates false calls and forces operators to review many good boards. An AI model trained with deep learning can compare many examples and learn which patterns are acceptable and which are true defects. The T7, for example, cites nominal defect detection accuracy of ≥ 99.5% and nominal false-call reduction up to 80% through adaptive learning. Those figures are nominal and depend on board type and process conditions, but they describe the direction AI classification is meant to take. Inspection speed up to 100+ mm/s also helps the station keep pace with the line.
An online AI AOI does more than accept or reject a board. Each inspection creates data: defect type, location, board ID if barcode scanning is installed, and time. When this data is collected over shifts and production runs, engineers can see whether a defect is random or tied to a specific placement head, solder pot, or material batch. Yield monitoring becomes a live view of the process instead of a report written after the batch is finished. Optional MES integration can pass this data to higher-level systems, while barcode scanning can link results to individual boards or panels. These options are not standard on every machine, but they show how inline inspection connects to factory traceability. Rework decisions change because defects are caught earlier. If a board has a missing component or a solder bridge after reflow or wave soldering, the line can divert it before final assembly adds more value. That reduces the cost of rework and prevents a known bad board from being assembled into a finished product. Online AOI also supports fast changeover when the line switches to a different PCB. A good-board guided CAD-free workflow can be configured in about five minutes for standard boards, and operators can learn the basic interface in about thirty minutes. Complex high-density boards may still need engineering adjustment, but the setup direction is toward faster line changes and less dependence on CAD files.
Online AI AOI inspection equipment belongs inside the SMT conveyor flow because that is where it can see boards at the right time, at line speed, and with enough context to support yield decisions. It sits between placement and final assembly, uses conveyor handoff and stable positioning to capture useful images, and applies AI classification to separate real defects from normal board variation. The result is not just inspection; it is a station that feeds rework and yield monitoring with board-level data. For readers who want to see how these ideas appear in a real inline model, the IRIS ANT T7 product documentation provides a concrete example of an online AI AOI with conveyor-based inspection, optional traceability modules, and published nominal performance figures.
A：It inspects boards as they move through the conveyor, checking component placement and solder joints after soldering and before final assembly. It flags defects for review or rework and can send inspection data to yield monitoring systems.
A：The machine receives boards from the upstream conveyor, positions each board at the inspection station, captures images, and releases it downstream. The line may pause briefly for positioning and image capture, but the inspection is integrated into the conveyor flow rather than performed in a separate offline batch.
A：Typical checks include missing components, wrong placement, polarity errors, tombstoning, insufficient or excess solder, bridges, and other visible solder joint problems. Post-wave soldering inspection can focus on through-hole joints and assembly issues, while reflow inspection covers surface-mount joints and component presence.
