# AI double-checks what the AOI machine flags on circuit boards and clears the false calls, so operators only review real defects (Siemens Rastatt, USI, FermionX)

> AI reviews AOI flags on PCBs: Siemens Rastatt raised first pass yield 42%, USI cut false alarms 90%, FermionX cut false calls over 40%.

- Canonical URL: https://aiagentsformanufacturing.com/use-cases/ai-aoi-false-call-reduction-smt-pcb-inspection-siemens-rastatt-usi-fermionx/
- Department: Quality (https://aiagentsformanufacturing.com/departments/quality/)
- Evidence level: Verified. Real deployment at a named or precisely described company, with a measured result from a primary source.
- Added: 2026-11-18
- Last updated: 2026-11-18
- Publisher: AI Agents for Manufacturing (https://aiagentsformanufacturing.com)
- Manufacturers named: Siemens Rastatt (https://aiagentsformanufacturing.com/companies/#siemens-rastatt), USI (https://aiagentsformanufacturing.com/companies/#usi), FermionX (https://aiagentsformanufacturing.com/companies/#fermionx)
- AI tools deployed: Siemens Opcenter Intelligence AOI False Call Reduction (https://aiagentsformanufacturing.com/tools/#siemens-opcenter-intelligence-aoi-false-call-reduction), Koh Young Zenith Alpha HS+ 3D AOI (https://aiagentsformanufacturing.com/tools/#koh-young-zenith-alpha-hs-3d-aoi)

## Role

Quality and process engineering on SMT electronics assembly lines (automated optical inspection, first pass yield). Typical users: SMT process engineer, quality manager, AOI programmer, electronics manufacturing (EMS) plant manager, production supervisor

## The problem

Automated optical inspection (AOI) after reflow checks every printed circuit board assembly, but it is tuned to miss nothing, so it flags many good boards. Each flag stops an operator, who zooms in on the image and decides whether the defect is real. The flood of false calls costs time, and it creates alarm fatigue: tired operators start approving boards too quickly, and a real defect can slip through. At Siemens Rastatt (Germany), which makes HVAC controls, burner controls and valves on four lines that each have an AOI machine, first pass yield was only 60%, and of the 40% of boards flagged, 97.5% were false calls; the head of process and technology explains that alarm fatigue made operators more likely to approve a faulty board, and that the AOI vendor's tuning help was limited. At USI, an electronics manufacturer, the team writes that such false alarms can happen thousands or even tens of thousands of times a day on an SMT line; conventional AOI compares images with golden samples, which is very sensitive to lighting, color and position, and one inspection item may need 30 to 80 golden samples, which there is no time to collect in high-mix production with frequent changeovers. At FermionX, a UK electronics manufacturing services provider for aviation, defense and medical technology, the old 2D AOI needed a lot of programming time, and its false calls created unnecessary rework and hurt operator trust.

## How AI helps

A machine learning model learns from the AOI's own measurements or images and from the decisions operators already made on past flagged boards (real defect or false call). For each new flag, it gives a verdict before the operator looks, so clear false calls can be released and the operator's attention goes to the likely real defects. At Siemens Rastatt, Siemens' Opcenter Intelligence AOI False Call Reduction software builds a model from AOI measurement data and operator-classified data; the risk of escapes is addressed in an assessment stage, and models are validated to be at least as accurate as the baseline before going live. The model runs on Siemens Industrial Edge devices next to the line, receives the AOI data in real time and shows its result in the existing repair station, and Siemens keeps monitoring and retraining the models as a service. At USI, the team built a deep learning model to read the text printed on components, combined with a similarity check against known-good images: a part is passed only when both agree, which is the safeguard against escapes; transfer learning lets the model learn new parts from few images. At FermionX, the Koh Young Zenith Alpha HS+ 3D AOI uses the vendor's AI engine to tell real defects from normal process variation, and the KSMART software tracks inspection data with statistical process control.

## Step by step

1. Export three months of AOI flags with the operator verdict (real defect or false call), the defect type, the component, the board and the line.
2. Calculate the false call rate by line, product, component and defect type, and the operator minutes spent reviewing them.
3. Fix what tuning can fix first (library, thresholds, golden samples) and pick the remaining worst categories for AI review.
4. Run the AI in shadow mode for several weeks: it gives a verdict but the operator still decides, and every disagreement is reviewed.
5. Agree in writing with quality (and your customers if required) on the escape criteria before any board is released on the AI's verdict alone.
6. Track false calls per board, review time, escapes found at ICT, functional test or by customers, and retrain when products or suppliers change.

## Tools

- Siemens Opcenter Intelligence AOI False Call Reduction on Siemens Industrial Edge: used at Siemens Rastatt
- In-house deep learning and similarity matching model for component marking checks: built and used at USI
- Koh Young Zenith Alpha HS+ 3D AOI with its AI engine and KSMART: used at FermionX
- Other tools to evaluate: the AI review or false call features of your current AOI vendor, AI verification software that works with several AOI brands, a model trained by your own team on exported AOI images
- To start without new software: your AOI and repair station export and any LLM (ChatGPT, Claude, Copilot, Gemini) with the prompt below

## Result

Siemens Rastatt (Siemens case study; Siemens is both the vendor and the parent company): first pass yield up 42%, manual inspection analysis time down 60%, first-year savings of EUR 160,287, return on investment after eight months, and 12 weeks from first discussions to commissioning. USI (company blog, 2026): an average 90% reduction in false alarms on component marking checks across its factories, up to 92% for one component type, and 0% escapes in blind tests with about 25,000 training images, including tests with unknown wrong parts; the solution runs on its production lines. FermionX (Koh Young case study, the vendor): false calls reduced by over 40%, with shorter programming time for new products.

As reported by the source; most figures are vendor or self-reported and not independently verified.

## Human review: cautions

- All three sources come from the vendor or the manufacturer itself: Siemens Rastatt is a Siemens plant using Siemens software, USI describes its own system, and the FermionX figure comes from Koh Young's case study.
- The Siemens Rastatt yield gain was measured from a very low starting point (60% first pass yield with 97.5% false calls); a line that is already well tuned will see smaller gains.
- An escape costs far more than a false call: test in shadow mode, keep the escape rate as the first measure, and check whether your customers (automotive, medical, aerospace, defense) must approve a change in the inspection method.
- USI's 90% figure covers checks of the text printed on components, not every AOI defect type; results differ by defect type.
- Models drift when parts, suppliers, solder paste or lighting change: plan who retrains them and how often.

## Try it yourself (prompt)

```text
You are an SMT process and quality engineer. I want to cut the false calls on our automated optical inspection (AOI) without letting real defects escape.

I will paste: (1) an export of AOI flags for the last weeks [date, line, product, board serial, component reference, part number, defect type, operator verdict: real defect or false call, repair done]; (2) our escapes found later [date, product, test where found (ICT, functional test, customer), defect, component].

Do this:
1. Calculate the false call rate overall and by line, product, part number and defect type, and estimate operator review time at [X] seconds per flag.
2. List the 15 combinations (part number + defect type) that create the most false calls, and for each say whether it looks like a tuning problem (library, threshold, golden sample) or a good candidate for AI review.
3. Compare the escapes with the AOI flags: which escapes were flagged and released by an operator, and which were never flagged.
4. Propose a shadow-mode test plan for an AI review tool: duration, what to measure, and the escape criteria we should agree on before letting the AI release any board.
5. Draft a one-page summary for the plant manager with the hours at stake and the risks.

Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.
```

## Related use cases

- [AI predicts which boards really need the slow test, so X-ray and end-of-line test stations stop being the bottleneck (Siemens Amberg, ZF, Flex)](https://aiagentsformanufacturing.com/use-cases/ai-predictive-test-selection-skip-xray-end-of-line-tests-siemens-zf-flex/)
- [Vision defect inspection on the edge (Seagate, Procter & Gamble, Coppenrath & Wiese)](https://aiagentsformanufacturing.com/use-cases/vision-defect-inspection-on-the-edge/)
- [AI reads X-ray images of aluminium castings to find porosity and shrinkage, on existing inspection systems (Hengst SE, Fraunhofer IIS)](https://aiagentsformanufacturing.com/use-cases/ai-xray-casting-porosity-automatic-defect-recognition-hengst-fraunhofer/)
- [Deep learning optical sorters spot hard defects in fruit and nuts on the line, so food plants pack cleaner product with fewer hand sorters (Chelan Fruit, Kings River Packing, The Almond Company)](https://aiagentsformanufacturing.com/use-cases/ai-deep-learning-optical-sorting-grading-food-defects-chelan-kings-river-almond-company/)

## Sources

- https://resources.sw.siemens.com/en-US/case-study-siemens-rastatt/
- https://www.usiglobal.com/en/blog/ai-deep-learning-on-smt-production-lines
- https://kohyoungamerica.com/fermionx-and-altus-advance-uk-manufacturing-with-koh-young-3d-aoi/

---
Please cite the canonical URL: https://aiagentsformanufacturing.com/use-cases/ai-aoi-false-call-reduction-smt-pcb-inspection-siemens-rastatt-usi-fermionx/
