Inspection · ACI OS · SOFTWARE

The AI inspection brainthat programs itself in 5 minutes.

Q: What data is needed for the 5-minute automatic programming AI quality inspection brain to build a model?

DaoAI's 5-minute automatic programming AI quality inspection brain is based on the feature recognition of the visual foundation model. It can build a model with just one good sample. No CAD or defect sample library is required. It's simple and efficient, and can be quickly put into use.

Q: What are the advantages of this quality inspection brain compared with traditional rule-based AOI?

Compared with traditional rule-based AOI, DaoAI's quality inspection brain uses the feature recognition of the visual foundation model. It can complete programming in 5 minutes with one good sample, while the traditional method takes 3-5 hours per board. Also, our model can be continuously optimized and becomes more accurate over time.

Q: Can DaoAI ACI OS be deployed flexibly?

Yes. DaoAI ACI supports various deployment methods such as SDK integration, REST API standard interface docking, one-click private deployment in Docker containers, and out-of-the-box use with local all-in-one machines. Data stays 100% in the factory, and edge inference is <12ms, available even offline.

ACI OS is the operating system for every ACI and next-generation AOI machine. It has two halves: visual recognition and cognitive judgement (the brain), and motion control of cameras and rails (the cerebellum). DaoAI ACI OS — built on a visual foundation model with feature cognition; model from one good sample. No CAD, no defect library, 100% on-premise, and embeddable in any 2D / 3D equipment. Next-generation term for this category: ACI, automated cognitive inspection.

Book a Demo Back to Inspection

One auto-programming run · the same panel raw and auto-programmed, every component boxed

SOVEREIGN AI VALUE

Your process know-how never leaves your line.

Yield curves, defect signatures, process parameters — this is a factory's core trade secret. Under IDC's framework, DaoAI's AI inspection brain starts with data sovereignty: flexible SDK / API / Docker deployment, with sample images and results never leaving your site or training any third-party model.

Training a model on the machine itself: the image set, the training run, then the new model applied to DIP solder inspection and catching a defect the previous model missed

Process parameters stay yours
Good-sample templates and defect libraries are decades of production know-how — we help you use it, never help anyone else learn it.
<12ms edge inference, works offline
Core inspection logic in DaoAI ACI runs on the line itself, independent of external networks or cloud availability.
Sharper on your line, every day
Line feedback keeps sharpening detection on your line — the edge you gain is your own competitiveness, not a shared foundation.

5minto program & deploy (as fast as 30 s; was 3–5 hrs)

1Model from one good sample

98%detection accuracy, in one click

−80%false-call rate

1M++real training images

100%On-premise · data stays on-site

Scan one reference board

From one board of the current batch, AI learns component positions and appearance — no CAD, no defect library.

Capture and choose: capture the golden board, then tick the checks you want

AI maps components automatically

Locates components, pins and polarity in seconds; handles rotation, flips and same-color parts — no hand-drawn ROIs.

Auto-program: the AI generates the inspection program in 30 seconds to 5 minutes; for panelized boards it also generates the panel layout

AI derives thresholds

Thresholds are derived from the board so operators can run it, and accuracy improves with every piece of feedback.

Adjusting the inspection threshold against the component score

Feature Cognition

Inspect in feature space, not color space

AOI detection example: Component orientation. Components placed the right way up but turned 90 or 180 degrees

Pre-trained on millions of components · APDT positive-sample learning
Understands a component's "identity" rather than comparing pixel and color similarity — just 1–20 good samples cover the same defect class, no negative-sample library needed.
No more false calls on same-color parts
Black parts on a black board, silver traces and silver parts — same-color scenes separate naturally in feature space.
Any orientation · contamination-resistant
Precise localization despite rotation, flips and surface contamination.

Sharper over time · data loop

Every piece of feedback makes the model smarter

Feedback learning in the review screen: ten alerted components, one reported as no error, an AI update retrains and reloads the model, and the next inspection shows six alerts

Learn and update on the spot
An inspector labels a false call once, the model updates, and the same error never recurs.
Semantic-level false-call filtering
Tells acceptable process variation from real defects; false calls keep falling with feedback.
Building data assets
Continuously optimized for your products and defects, becoming a dedicated model that grows stronger the more you use it.

Flexible Deployment · DEPLOYMENT

SDK integration

Embed into your existing AOI / vision equipment and line software.

REST API

Standard interfaces connect to MES / host systems, with results flowing back in real time.

Docker container

One-click private deployment — isolated environment, easy to operate and maintain.

On-premise all-in-one

Works out of the box with our 2D / 3D equipment.

Private licensing

Data stays 100% inside the factory — secure and compliant.

Edge inference

<12ms on-premise inference — no cloud needed, works offline.

FEATURE DEMOS

Run it: start inspecting production boards; operators review results and correct false calls at the machine

Switching between component samples and their defect scores

Model training: label images, train, run inspection, check results and customize model types

Configure OCR and OCV, test a marking, compare the OCV score with its threshold, apply settings, enable bidirectional inspection and review results

Draw one reference sub-board, automatically detect the panel array, save and align it, then review the inspection regions

Draw a reference sub-board, detect all 25 panel positions, then run Auto Adjust Array and review aligned component regions

Enable foreign object detection, draw the full-panel inspection region, run inspection and review four flagged findings with their board locations

Reporting a reviewed component as No error and confirming the feedback

Training feedback switch: off, on, then off. Feedback is recorded in both states; only on uses it for model training

Starting AI Update, processing the feedback, and completing the model update

Running the next recorded inspection and reviewing the resulting component alerts

DETECTION EXAMPLES

AOI detection example: Components. Missing, wrong, shifted, rotated or damaged components, and tombstoning

AOI detection example: Component markings. Wrong component codes, missing characters, smudged or shifted print

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

AOI detection example: Red glue. Adhesive that has spread onto a pad before wave soldering

DEFECT COVERAGE

Missing / wrong partOffset / skew / polarity / tombstone / billboard / upside-downBody damage / burn / discolorationHeight & coplanarity (3D)Insufficient / excess / cold solderOpen solder jointLifted / bent / missing leadSolder bridgeAdhesive (red glue) on padOCR / OCVBarcode / QR / serial numberThrough-hole lead & joint (pre / post-wave)Through-hole bridgingSolder balls / particlesForeign object debrisSurface contamination (coming soon)

Through-hole lead & joint (pre / post-wave)

INSPECTION COVERAGE · 16 CAPABILITIES

In DaoAI ACI each inspection runs on an automatically generated inspection program. Select the applicable checks for each component package type.

#	Capability	What it covers
01	Component Presence & Identification	Missing or incorrect components.
02	Placement, Orientation & Polarity	X/Y offset, skew or rotation, incorrect polarity, tombstoning, billboarding, and upside-down placement.
03	Component Condition	Package or body damage, damaged chip terminations, burn marks, and discoloration.
04	Component Height & Coplanarity	Measures component height and detects lifted or tilted bodies and coplanarity defects using 3D AOI.
05	Solder Joint Quality	Insufficient solder, excess solder, and cold solder joints.
06	Open Solder Joints	Open or unsoldered connections, available in both 2D and 3D configurations.
07	Lead Condition	Lifted, bent, damaged, missing, or unsoldered component leads.
08	Solder Bridging	Unintended solder connections between adjacent pads or leads.
09	Adhesive Contamination on Pads	Unwanted SMT adhesive (red glue) on solder pads.
10	OCR / OCV	Reads and verifies alphanumeric component markings, symbols, and laser-marked text.
11	Barcode & Serial Number Reading	Reads 1D barcodes, QR codes, and serial numbers for unit-level traceability.
12	Through-Hole Lead & Solder Joint Inspection	Pre-wave: lead/pin presence and insertion. Post-wave: missing, insufficient, or excess solder.
13	Through-Hole Solder Bridging	Solder bridges between adjacent through-hole pads.
14	Solder Balls & Particles	Solder balls or particles anywhere on the assembly, not only adjacent to a solder joint.
15	Foreign Object Debris (FOD)	Debris, wire clippings, dropped components, and other foreign material.
16	Surface Contamination	Unacceptable flux residue, stains, or other surface contamination. Coming soon.

FULL CONTROL

The AI programs it. You stay in full control — four levels of control and twenty-one settings across six areas. Auto-programming and model updates preserve your manual settings; locked parameters stay at the values you choose.

Adjust any threshold
Adjust AI-generated thresholds while reviewing measurements. Define the range available to AI updates, or disable AI updates for that check.
Set the rules for a whole package type
Set inspection rules per package type instead of editing components individually. Save them as presets for other products.
Add a component or inspection region
Draw a box around a component or feature that was not detected; the system learns its parameters from the golden board and adds it.
Train your own model
Train a custom model in DaoAI ACI on your own images for niche defects, then apply it to selected inspection regions.

What it generates · box size / pad, pin & bridge regions / sensitivity / alignment marksHow it judges · any threshold / locked values / which inspections run / judgement criteriaWhere it looks · your own regions / what runs inside / panel layout / region geometryWhat it learned · report a false call / what goes into training / your own model / turn a model offWhat things are called · designators & part numbers / grouping level / ambiguous charactersWho may do it · permission per level / undo

PRODUCTION INTEGRATION · 11 CAPABILITIES

Line handling, quality analysis, data export, and support. Each capability is built into the DaoAI inspection software.

Area	Capability	What it does
Line handling	Capture Path Optimization	The camera stops only where the board requires it. Fewer captures per board with every check enabled.
Line handling	Top/Bottom Side Program Selection	Program each side of a double-sided assembly separately, then run both in one inspection. The AI recognises which side is in front of the camera.
Line handling	Pass/Fail Board Routing	Failed boards are sorted according to the configured strategy, without operator intervention.
Quality analysis	Gage R&R (GR&R)	Repeatability and reproducibility study run in the software, on your boards, with the result exported directly.
Quality analysis	Process Capability (Cpk)	Process capability calculated from the inspection measurements the machine is already taking.
Quality analysis	First Pass Yield	Whole-board and sub-panel FPY. Per-position yield in a panel shows where a stencil or reflow problem sits.
Data & traceability	MES Export	CSV results, defect crop images, and the whole-board image per unit. Defect codes mapped to your MES vendor codes.
Data & traceability	Inspection History & Reports	Every inspection is stored with its images and verdict, so any board can be traced back. Reports and all analysis views export to Excel.
Operations & support	Repair Station	A dedicated screen for the repair operator: the defect call, its location on the board, and the image behind it.
Operations & support	Offline Programming	Program a product on a machine away from the line, export it, and import it on the production machine.
Operations & support	Remote Support	Remote assistance from our engineers when a problem needs a second look.

Criteria	Traditional rule-based AOI	Generic ACI	DaoAI Software
Inspection paradigm	Color / template matching	Generic classification network	Visual foundation model · feature cognition
Modeling data	Manual rules	Thousands of defect images	One good sample
Programming time	3–5 hrs per board	Depends on AI engineer	Operator finishes in 5 min
Optimization	Fixed algorithm	Requires vendor retraining	Learns and updates from line feedback on the spot
Deploy	Locked to dedicated hardware	Often needs the cloud	SDK/API/Docker · 100% on-premise

Book a Demo 2D Equipment 3D Equipment vs Cognex / HALCON

FAQ

What data is needed for the 5-minute automatic programming AI quality inspection brain to build a model?

DaoAI's 5-minute automatic programming AI quality inspection brain is based on the feature recognition of the visual foundation model. It can build a model with just one good sample. No CAD or defect sample library is required. It's simple and efficient, and can be quickly put into use.

What are the advantages of this quality inspection brain compared with traditional rule-based AOI?

Compared with traditional rule-based AOI, DaoAI's quality inspection brain uses the feature recognition of the visual foundation model. It can complete programming in 5 minutes with one good sample, while the traditional method takes 3-5 hours per board. Also, our model can be continuously optimized and becomes more accurate over time.

Can DaoAI ACI OS be deployed flexibly?

Yes. DaoAI ACI supports various deployment methods such as SDK integration, REST API standard interface docking, one-click private deployment in Docker containers, and out-of-the-box use with local all-in-one machines. Data stays 100% in the factory, and edge inference is <12ms, available even offline.

The Q&A above was generated with AI assistance and is for reference only; official specifications and commitments are subject to direct consultation.

Related Cases

FEATURES

Every capability, page by page

This software system is what the ACI cognitive inspection operating system looks like on the software side: one imaging and motion interface upward, adaptation to cameras, lights, motion axes and robot arms downward. That is why it embeds into the equipment you already have through SDK / API / Docker, and why a third-party machine can run it; and perception and control close the loop inside one system — the cognitive result decides what gets captured next.