AOI for NPI and Prototypes: Inspect the First Board, Not the 100th

AOI only helps an NPI or prototype build if the inspection program is ready before the first board leaves reflow. With rule-based AOI, programming often takes longer than the build itself, so prototypes get a microscope and a checklist instead. Golden-board AI programming changes that: DaoAI writes an inspection program from one good board in 30 seconds to 5 minutes, with no CAD file required, and DaoAI auto-programming does not require a prebuilt component library, so you inspect the first board of a build rather than the hundredth of a production run.

Why do NPI and prototype builds usually skip AOI?

The problem is lead time, not value. On a rule-based machine, an engineer typically needs 3 to 5 hours to teach a new board variant, and several days for a board with thousands of components. Before that can start, CAD or centroid data has to arrive, and every new package has to be added to the component library.

A prototype build of a few dozen boards can be placed and reflowed in less time than that. When setup takes longer than the run, which we call the NPI paradox, AOI is skipped and boards go to manual visual inspection.

That is an expensive trade. NPI is exactly when you want to find a polarity error in the placement program, a footprint that tombstones, or a package that bridges. Finding it at functional test, or at the customer, costs far more than finding it on the first board.

How does golden-board programming work on a first build?

Golden-board programming uses one known-good assembled board as the whole input. The AI identifies and models each component, generates the inspection boxes and sets thresholds from that board. Our guide to AOI programming without CAD files explains the method in detail.

NPI adds one obvious question: where does the golden board come from when no board has been built yet? The answer is to build the first board or two, verify one thoroughly, and make that board the reference.

1. Build the first boards. Run the first few boards of the lot through placement and reflow as normal.
2. Verify one by hand. Check it carefully under a microscope, confirm polarity and values against the BOM, and use X-ray on BGAs or bottom-terminated parts if your process calls for it.
3. Capture and choose checks. Load the verified board and tick the inspection checks you want for each package type.
4. Auto-program. AI auto-programming generates the program in 30 seconds to 5 minutes, at 98% detection accuracy before any tuning.
5. Inspect the rest of the build. Every remaining board, including the other early boards, now gets the same inspection.

CAD is optional. If a CAD or BOM file arrives later, import it to add designators and part numbers; it does not change programming quality, and your tuning is kept.

What happens when the board revision changes?

NPI boards rarely stay on one revision. A Rev B might move a connector, change two passives and swap a regulator package. You do not want to rebuild a program that is otherwise tuned.

DaoAI lets you draw a box around the part of the board that changed and run auto-programming again inside it. Everything outside the box, including your edits, stays exactly as it was. Parameters you have locked stay fixed, and manual settings are preserved when auto-programming runs again.

Prototype builds also tend to use alternate parts from whichever supplier had stock. A different supplier can mean a different marking or body color for the same part number. Marking that variation OK teaches the AI it is acceptable across the component group, while lifted leads, bridges and missing parts remain inspection targets.

Why put an offline AOI beside the line for NPI?

An inline machine inspects at conveyor speed, but it also ties NPI work to the production line. An offline, manually loaded machine beside the line lets you program and inspect prototype boards without holding the SMT line, which suits irregular job flow, short runs, labs and NPI.

It also makes programming a separate task from production. With DaoAI's offline programming, you program a product on a machine away from the line, export it, and import it on the production machine when the product ramps. Presets for package types can be exported to other machines too.

DaoAI's offline range has three machines, all manual load with auto-clamp and the same core AI software:

All three accept boards 0.5 to 5 mm (≈ 0.0197 to 0.197 in) thick and inspect at 0.238 seconds per field of view.

How do you fit AOI into an NPI build plan?

Treat AOI as a standard step in the NPI traveler rather than an optional extra. A simple plan looks like this:

1. Before the build: choose the checks per package type and load any saved presets from similar products.
2. First boards: verify one board by hand and capture it as the golden board.
3. During the build: inspect every board, and have operators report false calls and real defects at the review screen.
4. After the build: export defects by position and type for the NPI review. Repeated calls on one footprint are design or stencil feedback.
5. At ramp: run an AI Update on the collected feedback, then export the program to the production machine.

None of this is limited to prototypes: the same golden-board programming and feedback learning run on the inline P3 and P5 when the product moves to volume production. The same habits shorten changeovers once the product is in volume. See how to cut AOI changeover time on high-mix SMT lines for that side of the process.

FAQ

Can AOI be used on a prototype build of only a few boards?

Yes, if programming is fast enough. DaoAI auto-programs from one good board in 30 seconds to 5 minutes, which is short enough to inspect even a small prototype lot.

Where does the golden board come from on a first build?

Build the first boards, verify one thoroughly by hand, and use that verified board as the reference. The AI generates the inspection program from it.

Do I need CAD or Gerber files for NPI inspection?

No. One golden board is enough. A CAD or BOM file can be added later for designators and part numbers without losing your tuning.

Can the NPI program be moved to the production line?

Yes. DaoAI supports offline programming: program on one machine, export it, and import it on the production machine.

Which offline DaoAI machine suits NPI?

The P1 covers most NPI and lab work up to 470 × 330 mm. The P1L takes boards up to 680 × 510 mm, and the P2 adds a 12 MP camera at 10 µm per pixel.

Bring your next NPI board to a live session with one of our engineers. Book a demo, or send us one of your own boards and we will test it for you, free of charge.