Close the coverage gap that AOI and functional test leave behind
Willowark builds the inspection, test, and traceability systems that sit around an SMT line: AI-assisted verification of AOI calls, solder paste inspection data correlated to downstream defects, conformal coating coverage checks, custom functional test fixtures, and component traceability down to reel and date code. The problem it removes is escapes hiding inside noise — real defects lost among false calls operators have learned to click through.
We start from your dispositioned defect images, because that is ground truth nobody else has. A classifier trained on your boards, your paste, and your operators' actual accept and reject decisions runs as a second pass behind AOI, clearing obvious false calls and routing anything ambiguous to a person. Fixtures and handling hardware are designed ESD-safe from the start, with grounding, materials, and handling paths specified rather than assumed.
A first project here usually starts with data you already have and are not using: the AOI call log, SPI measurements, and test results that live in separate machines and never meet. Pulling them into one database keyed by board serial and reference designator is a modest build that often changes how the line is run before any new hardware arrives. For contract manufacturers the constraint is mix — dozens of customers, each with its own acceptance class and traceability demands — so anything we build is configured per assembly rather than coded per assembly.
Reviewed
Sound familiar?
If you've said any of these, we should talk.
“AOI flags so many false calls that operators just click through the screen.”
We collect the flagged images together with the operator's verified disposition and train a classifier on that history. It runs as a second pass that clears obvious false calls and escalates the rest, so the review queue shrinks to defects that deserve human attention.
“We chase solder defects at AOI when they actually started at the printer.”
We correlate SPI measurements to AOI and test results by reference designator, so paste volume, area, and height trends tie to the defects they eventually cause. Once that link is visible, stencil cleaning frequency and print parameters get set on evidence instead of a fixed schedule.
“A functional test fixture takes months and costs more than the product.”
We build fixtures and the software behind them: pogo or bed-of-nails interfaces, instrument control, test sequencing, limits, and result logging to a database rather than a text file. For high-mix work a common chassis with swappable interface plates makes each new assembly a plate, not a project.
“A customer asked which reel a part came from and we had no answer.”
We capture serialized traceability linking a board serial to placement records, feeder position, reel and date code, reflow profile, and test results. When a component issue surfaces, containment is scoped to the boards that actually received the suspect reel.
“Our test data is a folder of text files and nobody can tell me first-pass yield by assembly.”
We parse what the testers already write — log files, CSV exports, instrument output — into a database keyed by serial, assembly, revision, and station. First-pass yield, retest rates, and failing test steps become a query by assembly and date range, and the same records feed the traceability your customers ask for.
How this industry actually runs
The operation as we usually find it.
An SMT line runs stencil print, solder paste inspection, placement, reflow, then AOI and often X-ray for BGAs and bottom-terminated components whose joints cannot be seen optically. Most solder defects are born at the printer: insufficient or bridged paste, poor aperture release, inadequate board support. Reflow profiles are validated per assembly with thermocouples, and a profile that drifts shows up later as tombstoning or voiding. Downstream come selective or wave solder for through-hole, cleaning, conformal coating with UV tracer for coverage verification, in-circuit and functional test, and burn-in. All of it happens under ESD controls with wrist-strap and ionizer verification, against IPC-A-610 acceptance classes, and for medical, aerospace, and automotive work, with traceability down to reel, date code, and moisture-sensitive device floor life.
Machine signals to the people who decide
Components:
- PLCs & sensors (counts, states, current)
- Legacy machine (dry contact / clamp)
- Edge gateway (normalize, buffer)
- Production dashboard (downtime, OEE)
- Alerts & reports (who acts, when)
Connections:
- PLCs & sensors to Edge gateway (EtherNet/IP, Modbus)
- Legacy machine to Edge gateway
- Edge gateway to Production dashboard (MQTT)
- Edge gateway to Alerts & reports
What we build
Starting projects that fit Electronics Manufacturing.
- Second-pass AOI verification using models trained on your own dispositioned defect images
- SPI to AOI correlation with paste volume, area, and height trending by aperture and refdes
- Conformal coating coverage inspection under UV with edge and keep-out verification
- Custom functional and in-circuit test fixtures with test software, limits, and result logging
- Board-level serialized traceability linking serial to feeder, reel, date code, and reflow profile
- Moisture-sensitive device floor-life tracking with dry cabinet monitoring and expiry alarms
- ESD-safe automated handling, part presence verification, and label and barcode grading
- Test data aggregation parsing existing logs into first-pass yield, retest, and failure reporting by assembly
Capabilities we bring
What we'd build first
Three places a first project usually starts.
01
Second-pass AOI verification using models trained on your own dispositioned defect images
02
SPI to AOI correlation with paste volume, area, and height trending by aperture and refdes
03
Conformal coating coverage inspection under UV with edge and keep-out verification
Scoped in writing before any work starts — how engagements work.
Working in Electronics Manufacturing?
Tell us the line.
What runs by hand, what is not connected, what you are trying to build. An engineer replies within one business day with whether and how we would approach it.
Common questions
What Electronics Manufacturing teams ask first.
Can AI reduce AOI false calls without letting real defects through?
It can when the model is used as a filter rather than a judge. We tune the operating point conservatively, so anything the model is unsure about still reaches an operator, and we measure escape rate on a held-out set of confirmed defects before it goes live. Any claim of zero human review deserves suspicion.
How do we prove conformal coating coverage rather than assume it?
Most coatings carry a UV tracer, so imaging the board under controlled ultraviolet illumination reveals coverage, thin areas, bridging, and coating where it does not belong. We take keep-out zones from your assembly drawing, check each one, and archive the image against the board serial.
Does custom test make sense for low-volume, high-mix production?
Often more than for high volume, because high-mix shops pay the fixture engineering cost repeatedly. The way to make it work is a reusable platform: one instrumentation chassis and software framework, with per-product interface plates and test sequences defined as configuration.
How many defect images do we need before an AI second-pass classifier is useful?
Fewer than most people expect for the false-call side, and more than they expect for the rare real defects. False calls are usually abundant, so the model learns to clear them quickly. Confirmed defects are scarce by definition, which is why we keep the operating point conservative and route anything uncertain to a person. We typically start from your existing AOI archive and disposition history, and the model improves as reviewed images accumulate.
Can you pull data from our existing AOI, SPI, and pick-and-place machines?
Usually. Most current SMT equipment exports results as files, over a database connection, or through a vendor interface, and we work with whatever each machine provides. The practical limits are vendor-specific: some interfaces are documented and open, others require a license from the manufacturer or only export summaries. We confirm what each machine on your line will give us during the survey, because that decides how deep the traceability can go.
Strategy. Software. Systems.
Engineering for Electronics Manufacturing.
Describe the problem in your own words. An engineer reads it — not a sales script — and tells you plainly what it would take.