One auto-programming run: the same panel raw and auto-programmed, every component boxed
30s – 5 min
To generate an inspection program from scratch Achieve a 1% false call rate with tuning.
Achieve a 1% false call rate with tuning.
98%
Detection accuracy, in one click Straight out of auto-programming, before any tuning.
Straight out of auto-programming, before any tuning.
CAD optional
One golden board is the whole input A CAD or BOM file fills in designators and part numbers. Nothing else depends on it.
A CAD or BOM file fills in designators and part numbers. Nothing else depends on it.
Hours of teaching, cut to minutes.
Traditional AOI programming takes hours: preparing board data, maintaining component libraries, and tuning inspection settings for each new board.
Rule based teach (per board variant)3 – 5 HRDaoAI AI Auto-Programming30 S – 5 MINDaysNoneNo waiting for dataNothing starts until CAD or a component list arrives, and is correct.3–5 hr5 minLess programming time per boardProgramming a new board takes an engineer 3 to 5 hours, and several days on a board with thousands of components.PrerequisiteNoneNo component library requiredEvery new package added by hand, and kept correct from then on.Hours30 sChangeovers start readyA new board means programming again, with thresholds tuned one at a time.
Eight core detection capabilities.
Our latest AI model covers all eight, with ease and accuracy.
AOI detection example: Components. Missing, wrong, shifted, rotated or damaged components, and tombstoning.AOI detection example: Component polarity. Reversed diodes, LEDs, capacitors, tantalums and ICs.AOI detection example: Component markings. Wrong component codes, missing characters, smudged or shifted print.AOI detection example: Component orientation. Components placed the right way up but turned 90 or 180 degrees.AOI detection example: Solder quality. Insufficient or excess solder, cold joints, and bridges between neighbouring pads.AOI detection example: Pins, pads and solder. Lifted, bent, damaged or missing leads, one verdict per pin.AOI detection example: Through-hole joints. Before the wave: missing, reversed or badly inserted components. After the wave: insufficient or excess solder, bad or bridged joints.AOI detection example: Red glue. Adhesive that has spread onto a pad before wave soldering.
01 / 08Components.
Missing, wrong, shifted, rotated or damaged components, and tombstoning.
01Components02Component polarity03Component markings04Component orientation05Solder quality06Pins, pads and solder07Through-hole joints08Red glue
In just four steps.
Capture and choose: Capture the golden board, then tick the checks you want.
STEP 01
Capture and choose
Capture the golden board, then tick the checks you want.
Auto-program: The AI generates the inspection program in 30 seconds to 5 minutes. For panelized boards, it also generates the panel layout in the same run.
STEP 02
Auto-program
The AI generates the inspection program in 30 seconds to 5 minutes. For panelized boards, it also generates the panel layout in the same run.
Fine-tune: Change anything you are not happy with.
STEP 03
Fine-tune
Change anything you are not happy with.
Run it: Start inspecting production boards. Operators can review results and correct false calls directly at the machine.
STEP 04
Run it
Start inspecting production boards. Operators can review results and correct false calls directly at the machine.
The AI gives every component a score and sets every threshold.
Scores come from a model trained on your board, and thresholds are set against those scores. One AI update retrains the model, regenerates the parameters and re-evaluates the board, whenever you want.
Switching between component samples and their defect scores
A score on every component
Each component gets a score, not just a pass or fail. View its score and detailed inspection results to see how far it deviates from the reference.
Adjusting the inspection threshold against the component score
A threshold you can see
The threshold sits against those scores. You see how the good components and the bad components scored, and set the line between them.
AI-generated. Fully configurable.
Adjust bounding boxes Set detection bounding box sizes for each package type before auto-programming, instead of resizing them component by component afterward.Choose inspection regions For each package type, choose whether to skip a region, generate its geometry only, or include it in inspection. This includes pad, pin and bridge regions.Lock selected parameters Lock a parameter to your chosen value so it stays unchanged when you run auto-programming again.Save as presets Save the whole auto-programming setup under a name, apply it to the next product in one step, or export it to another machine.Re-run auto-programming in a selected area
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. Full Control has the whole list.
Add CAD whenever you want. Or never.
CAD/BOM data adds designators and part numbers to AI-detected components and groups them by part number. You can import it later without losing your tuning.
Board + CAD/ BOMProgramming qualityFull. The file does not change what is detected or how it is judgedWhen you can startAs soon as a golden board is available. Import CAD/BOM data whenever it is available.Designators and part numbersYour real designators and part numbers, matched in automaticallyGrouping across identical componentsGrouped by part number as well, so one edit covers the exact componentMES and traceabilityExports in the vocabulary your MES already uses, no mappingOperator and repair stationDefect named by designator, which is what the repair sheet saysOne Golden Board OnlyProgramming qualityIdentical. Components, leads and pads are all detected from the boardWhen you can startAs soon as a golden board is available.Designators and part numbersAuto-generated identifiers. Positions are correct but unnamedGrouping across identical componentsGrouped by detected component typeMES and traceabilityExports with generated identifiers, mapping needed downstreamOperator and repair stationDefect located by position on the board image
Accepted formats are the component information files SMT equipment already exports, in CSV, TSV or plain text. If one arrives late, or never, inspection quality is unaffected.
No pre-built component library is required for auto-programming. Component templates can be saved, reused and exported to other machines.
How we stand out from the alternatives.
Rule-based teach
TEMPLATE MATCHING · GOLDEN SAMPLE
Generic AI AOI
PRE-TRAINED MODEL · CLOUD PROCESSED
DaoAI
AUTO-PROGRAMMING
What it needs before you startCAD or a component list, plus a package library somebody maintainsA pre-trained model that was never trained on your boardsOne golden boardCAD is accepted and optionalHow a component is judgedPixel by pixel against one golden imageA general model trained on other people's boards, never fitted to yoursComponents, not pixelsAn AI model trained on over a million real production images. The model is adapted to your board to help distinguish defects from acceptable visual variation.A new component typeSomeone adds it to the library by hand firstWorks only if it resembles the training data. Otherwise, back to the vendorNothing to add firstDetected and programmed straight from the boardBringing the false call rate downLoosen the threshold, which also lets real defects throughRaised as a request. The improvement arrives when the vendor retrains and ships itMark it OK onceThe model retrains on it. See Feedback LearningWho is able to change itWhoever holds the threshold knowledgeTheir AI engineer. Cloud processing also carries your board images outside the plantAny operatorCan see what to change and how. Account permissions decide who is allowed to
The questions we get every time.
Does it need CAD or Gerber files?
No. One golden board is enough. CAD is accepted, and it adds designators, part numbers and grouping. It does not change programming quality. See the CAD comparison above.
What happens on a package the system has never seen?
It is detected and modelled from the board in front of it. There is no library entry to create first, which is the point of the whole approach.
If I tune a parameter by hand, will retraining overwrite it?
No. Generation compares when the parameter was last edited by hand against when its region was last generated, and it leaves the newer human edit alone.
How accurate is it before we tune anything?
98% detection accuracy on the program the machine writes itself, in one click. Reporting false calls from the review screen takes it further from there.
Programming gets you a working inspection on day one. Bringing the false call rate down after that is a different mechanism, and it starts with an operator telling it once.
More from this page
Send us a board and we will tell you which machine it needs.
