When drivers have to remove hundreds of harmless alarm systems per change, attention decays, hand-operated confirmation comes to be mechanical, and the allegedly “risk-free” evaluation program may develop the specific defect-escape risk that its conventional thresholds were planned to avoid.
So why do manufacturing facilities keep tightening up limits?
Because incorrect phone calls are visible immediately. Leaves surface later on– often after useful test, often during last setting up, and often via an expensive consumer return. Managers therefore reward sensitivity, while the confirmation terminal silently absorbs the price.
That reasoning is easy to understand. It is likewise incomplete.
The Brief Answer: Eliminate Variation Before Reducing Level Of Sensitivity
The best AOI incorrect call reduction approach is to remove dimension sound and unchecked procedure variant before altering pass/fail thresholds.
In method, that suggests supporting:
- Board positioning, warpage, support and conveyor transfer
- Cam emphasis, direct exposure, lighting angle and shade equilibrium
- Component-library images and bundle definitions
- Fiducial recognition and regional positioning
- Solder-paste printing, positioning and reflow problems
- Driver category policies
- AOI program revision control
Just then ought to designers tune inspection home windows.
I would deny any kind of optimization proposition that promises a lower AOI false phone call rate without showing its confusion matrix, seeded-defect outcomes and downstream escape data. A dropping alarm system matter proves nearly nothing by itself.
Why False Calls Increase– and Escape Threat Can Rise Faster
AOI does not check an abstract “good board.” It contrasts optical measurements versus limits while the genuine production process keeps relocating.
PCB color differs. Element bodies change shade between vendors. Silkscreen changes. Change residue mirrors light. Solder fillets transform with paste quantity, part surface, pad geometry and the reflow account. Also a valid alternative element can look incorrect to a program built around one golden image.
A 2024 production dataset indexed by PubMed followed SMT examination task throughout 132 days. Its writers reported distribution drift from tiny, occasionally unknown production adjustments and acknowledged that human disposition labels might consist of mistakes. That is a hard warning: a classifier educated on yesterday’s boards and yesterday’s operator judgments may not represent next month’s process.
The other catch is alarm fatigue. In a 2023 Siemens Rastatt study covering four AOI-equipped lines, the initial first-pass yield was reported at 60%; among the other 40%, 97.5% of calls were incorrect. Siemens reported a 42% FPY renovation, 60% much less manual-analysis time, an eight-month ROI and first-year financial savings of EUR160,287 after implementing its false-call decrease software program. Those are impressive numbers– however this is a vendor-published, single-site case, not a global criteria.
The difficult reality? Excessive level of sensitivity can come to be harmful when it conditions operators to anticipate that virtually every alarm is wrong.
Specify the Denominators Before Claiming Enhancement
“Incorrect call price” is consistently presented without a denominator. That makes comparisons unsafe.
One plant may separate false phone calls by evaluated boards. One more uses components. One more counts solder joints, assessment home windows or complete alarms. A board consisting of 2,000 assessment opportunities can not be compared honestly with a board including 200 unless the system is normalized.
I advise tracking at least these steps:
- False-call PPM: false-positive phone calls ÷ known-good assessment possibilities × 1,000,000
- Phone calls per board: complete AOI calls ÷ boards evaluated
- Alarm precision: validated problems ÷ total AOI alarm systems
- Flaw sensitivity: real flaws spotted ÷ all understood true flaws
- Escape PPM: verified issues missed by AOI ÷ checked possibilities × 1,000,000
- Verification problem: hands-on testimonial secs per board
- Repeatability: agreement across repeated assessments of the same control panel
- First-pass yield: boards proceeding without repair or unneeded disposition
Do not mix board-level and opportunity-level numbers. And never report “99% precision” when excellent systems stand for 99.8% of production; a classifier that calls everything good would certainly show up exact while missing every issue.
IPC-9716, provided in December 2024, explicitly organizes AOI procedure control around subjects including illumination, calibration, measurement-system evaluation, examination specifications, threshold values, false-call reduction and retreat price. That framework supports the main point: incorrect phone calls and leaves need to be regulated as a linked measurement trouble, not as different dashboard decors.
Take Care Of Upstream Process Variant Prior To Reprogramming AOI
AOI often gets criticized for exposing unpredictable production.
Beginning with a Pareto chart by line, item, recipe modification, bundle, recommendation designator, defect code, feeder, component whole lot and hour. If 61% of annoyance calls originated from one 0402 capacitor family, do not unwind every capacitor guideline. Find out what transformed.
Solder-paste printing
Inspect pattern tidiness, paste age, aperture design, bottom cleaning, squeegee pressure, separation speed and board support. If paste down payments wander, AOI shows optimization can camouflage the sign while the soldering procedure remains unpredictable.
A regulated print procedure utilizing equipment such as the Meraif 5151 high-precision printer can minimize the picture variation getting to downstream assessment, but devices alone is not the control plan. Dimension data, maintenance intervals and dish self-control still determine the result.
Positioning
Inspect nozzle wear, feeder indexing, pickup centering, part teaching and local fiducial technique. Repeated positional alarm systems at the very same referral designators might show positioning bias, not bad AOI logic.
Reflow
Reflow can both create flaws and transform how acceptable solder joints look. Verify ramp rate, soak, time over liquidus, height temperature and cooling versus the solder-paste requirements.
The JT TRV accuracy reflow oven is placed around uniform heating and flexible temperature level profiles. Whatever oven is mounted, accounts must be gauged on representative settings up– declined due to the fact that the maker’s area setpoints look secure.
Make use of a Wickon reflow thermal profiler or equivalent instrument to catch actual board temperatures. A thermocouple analysis is evidence. A setpoint is an intent.
Song AOI by Problem Family, Not by Global Level of sensitivity
The most effective AOI false call decrease methods are medical.
Different missing parts, polarity, offset, rotation, raised leads, inadequate solder, excess solder, bridges, headstones and character acknowledgment into unique recognition populaces. Each flaw has various optical physics and different getaway effects.
Utilize this adjusting order:
- Deal with the photo: clean optics, maintain lighting, appropriate emphasis and verify board assistance.
- Right placement: examine international fiducials, neighborhood fiducials and element search areas.
- Deal with the library: separate visually various bundle vendors, surfaces and markings.
- Deal with the region of passion: exclude silkscreen, surrounding pads, vias and variable board attributes.
- Select the best dimension: use geometry, comparison, shade, height or multi-angle evidence suitable to the issue.
- Tune the threshold: readjust just after the measurement itself is stable.
- Challenge the change: test both known-good variant and representative issues.
For a 2D application, a system such as the Saki 2Di-LU1 2D AOI system can support high-resolution, real-time assessment. But 2D contrast can not accurately recoup elevation information that was never ever caught.
Where raised leads, coplanarity, solder quantity or trailing dominate, examine a system such as the Pemtron ZEUS 3D AOI platform. Three-dimensional data can divide height from color variant, although it still requires repeatability studies, package-specific shows and self-displined recognition.
A lot more dimensions aid. They do not rescind data.
Show False Alarm System Reduction Without Wagering on Escapes
A qualified trial needs frozen baselines and controlled contrasts.
Develop 3 datasets:
- Known-good collection: typical production variation across vendors, great deals, changes and board coatings
- Known-defect collection: confirmed natural defects plus managed difficulty examples
- Locked recognition collection: boards and pictures excluded from programming, training and threshold option
Run the baseline and candidate dish on the same populations. Record every changed category. Do not allow the programmer relabel awkward examples after seeing the prospect result; that infects the test.
For rare, safety-significant flaws, a small recognition example is weak evidence. If zero leaves occur in only 100 issue opportunities, the discriminatory 95% top confidence bound continues to be about 3%. “Absolutely no observed” does not indicate “no risk.”
And inspect downstream evidence. Useful examination, in-circuit examination, X-ray evaluation, repair documents and customer returns can disclose blind spots that an AOI-only study can not see.
A 2024 peer-reviewed meeting research study from Binghamton University analyzed 30,000 manufacturing data points from SPI and pre-reflow AOI, generating an 89.3% category precision result. Useful? Yes. Adequate to release boards immediately? Not without class-specific false-negative information, because accumulated precision does not subject the costliest blunders.
When AI Is Worth Using– and When It Is Expensive Movie theater
AI can rate or reclassify AOI alarm systems utilizing measurement vectors, images and historic driver choices. It is especially useful when appropriate variation is multivariate and can not be represented cleanly by one fixed threshold.
Yet poor labels toxin the design.
If one operator calls a limited solder joint appropriate and another calls it faulty, the design finds out incongruity with mathematical confidence. If element suppliers change, illumination is serviced or a new PCB solder mask is introduced, circulation drift can deteriorate efficiency.
Before deployment, require:
- Different training, validation and locked examination data
- Recall and false-negative outcomes by flaw household
- Performance by item, line, package and supplier
- Confidence-based routing for unsure predictions
- Wander checking for input functions and outcomes
- A safe backup when the version or connection stops working
- Variation control, approval records and fast rollback
- Periodic tasting of AI-passed calls for human audit
My viewpoint is blunt: AI should minimize review initiative prior to it gets authority to subdue alarm systems. Advisory setting first. Automatic disposition later on, after the getaway evidence makes that opportunity.
A 30-Day AOI False Call Decrease Strategy
Throughout days 1– 5, ice up the energetic dishes and develop the baseline. Normalize phone calls by evaluation opportunity, validate defect-code interpretations and audit operator contract.
During days 6– 10, construct a false-call Pareto. Photograph reoccuring examples and trace them to part whole lot, area, illumination channel, procedure time and tools state.
During days 11– 15, remove assignable causes. Tidy optics, repair components, proper collections, profile the stove and maintain printing or positioning.
Throughout days 16– 20, change one problem household at once. Videotape old and new parameters, engineering reasoning and affected bundles.
Throughout days 21– 25, challenge the candidate recipe with known-good boards, all-natural problems and seeded defects. Keep the recognition set secured.
Throughout days 26– 30, run a secured manufacturing launch. Experience AOI-passed boards, integrate downstream failings daily and keep a one-click rollback path.
Then hold the gain. A dish that does well for 5 days and wanders for 5 months is not enhanced; it is temporarily quiet.
AOI Optimization Choice Table
| Intervention | Anticipated false-call impact | Escape-risk result | Evidence needed before launch |
|---|---|---|---|
| Clean optics and stabilize illumination | Trung bình đến cao | Normally decreases | Repeatability research and before/after picture contrast |
| Appropriate fiducials and board support | Trung bình | Typically lowers | Positional circulation and repeated-board examination |
| Split element libraries by distributor or coating | High for concentrated issues | Low when validated | Lot-specific known-good and defect examples |
| Repair printing, placement or reflow variant | High and resilient | Decreases | SPI, placement or account information tied to AOI calls |
| Boost areas of rate of interest | Tool to high | Moderate | Changed-call review by flaw family |
| Loosen up global thresholds | High initially | Cao | Typically decline; examination every impacted problem course |
| Include 3D dimension | High for height-related calls | May reduce | GR&R, cycle-time and challenge-board results |
| Include AI reclassification | Potentially high | Model-dependent | Locked examination set, drift controls and rollback strategy |
Often Asked Questions
What is AOI incorrect call reduction?
AOI incorrect call reduction is the regulated removal of false-positive examination alarms while preserving tried and tested sensitivity to genuine flaws, confirmed through classified boards, challenge examples, getaway audits, and stable measurement problems as opposed to by merely widening every pass/fail limit in the evaluation program.
The goal is not fewer alarm systems at any price. It is a higher proportion of significant alarms, less hand-operated confirmation and no statistically considerable deterioration in problem discovery.
What is an appropriate AOI false phone call rate?
An appropriate AOI false call rate is a site-specific limit linked to item class, assessment possibility matter, verification capacity, and escape danger; no truthful global percent works across a low-mix customer board, a dense automotive PCBA, and a high-reliability Course 3 setting up.
Specify the denominator, sector the outcome by issue family members and establish a restriction based on standard ability. “Below 5%” may sound attractive, but it is useless unless we understand whether that percent describes boards, parts, chances or alarms.
Just how can AOI thresholds be altered without boosting defect escapes?
The safest method to tune AOI limits is to alter one defect-family regulation at a time, freeze all unconnected specifications, run known-good and seeded-defect boards, and authorize the modification only when incorrect phone calls fall while problem sensitivity, repeatability, and downstream getaway proof remain at or over the verified baseline.
Maintain the previous recipe readily available for rollback. Continue sampling passed boards throughout the secured launch because laboratory validation can not duplicate every vendor, lot and process interaction.
When should a manufacturing facility relocation from 2D AOI to 3D AOI?
Three-dimensional AOI is warranted when height, coplanarity, raised leads, solder quantity, or tailing can not be separated accurately in 2D images, offered the included measurement capacity is demonstrated on the factory’s actual packages, coatings, board colors, line rate, and defect collection as opposed to approved from a brochure.
A 3D system is not immediately far better for every single defect. Character recognition, polarity markings and visibility checks may remain largely 2D issues.
Should AI be used for AOI false alarm decrease?
AI should be contributed to AOI when the plant has constant human labels, enough depictive manufacturing data, a secured recognition set, drift monitoring, rollback control, and a company case based upon minimized testimonial labor; it needs to not become an unvalidated second threshold that conceals unclear issues.
Beginning by ranking alarms or recognizing likely incorrect ask for faster driver evaluation. Automatic bypass need to come only after defect-family recall and downstream getaway audits show sustained efficiency.
Cut Sound, Keep the Safety Net
Do not begin by asking, “Which limit can we loosen?”
Ask which procedure variant, optical problem or programs decision caused the good board to look faulty. After that confirm the modification against understood defects and protected production information.
If your AOI terminal is eating labor, slowing down throughput or training operators to overlook alarms, demand a line-specific testimonial covering assessment innovation, printing, positioning, reflow profiling and validation. Bring the raw telephone call background. That is where reliable AOI incorrect call reduction begins.
