Product photo coming soon
ASUS IoTASUS AISVision 2.0
AI vision tool for training custom recognition models on site and deploying them fast.
- Classification, segmentation, detection, anomaly
- No-code labelling and training
- On-premises inference
Edge AI Computing
Edge AI Computing Solutions
Image recognition and data analysis run right at the equipment — fast response without sending data to the cloud.
ASUS IoT solutions
AISVision 2.0 × AIEHS 2.0
INIKI sells and deploys ASUS AI solutions with hands-on experience — on-site installation, model training and integration with existing equipment.
Product photo coming soon
ASUS IoTASUS AIEHS 2.0
AI workplace safety platform that recognises safety events and notifies the right people.
- Smoke, fire, intrusion, missing PPE
- Instant event notifications
- Event logs and visual reports
What it does
Edge AI on the factory floor
Image coming soon
AI visual inspection
Classification, segmentation, detection and anomaly models recognise defects and work states.
Image coming soon
Workplace safety alerts
Detect smoke, fire, intrusion and missing PPE and notify automatically.
Image coming soon
Event notification & logs
Integrate warning lights and PA systems and record every event for audits.
Image coming soon
AI maintenance guidance
Help technicians find the right troubleshooting steps and cut downtime.
Image coming soon
Vibration AI analysis
Analyse vibration at the machine to catch rotating-equipment faults early.
Why the edge
Benefits of analysing on site
Instant response
Low-latency inference on site for immediate action.
Data stays in the plant
Images and process data remain on your network.
Fast deployment
No-code model training adapts quickly to site needs.
AI vision inspection
Traditional AOI vs AI deep-learning defect detection
Each has its place. Classic AOI still suits fixed specifications and precise dimensional measurement; AI shines when defects vary and backgrounds are complex.
| Traditional AOI | AI deep-learning inspection | |
|---|---|---|
| How it decides | Rules and fixed parameters | Neural networks learn features and infer |
| Best for | High volume, fixed specs, absolute dimensions | High mix (with anomaly detection trained on good parts), unstructured defects, complex backgrounds |
| False rejects (overkill) | High; sensitive to lighting | Lower; tolerant of variation |
| Escapes (underkill) | Misses defects that were never defined | Catches complex, non-standard anomalies |
| Changeover | Days to weeks of re-tuning and programming | Hours to days with transfer learning |
Keys to success
Good AI inspection starts with good images
Garbage in, garbage out. Optical design often matters more than the model itself.
Lighting
Bright-field for outlines and presence checks; dark-field makes scratches and fine marks light up.
Shutter
Global shutter for high-speed lines to avoid distortion; rolling shutter for stop-and-go.
Lens
Chosen by field of view, working distance and depth of field to fit the machine.
Resolution
The smallest defect should span 3–5 pixels for stable AI detection; camera resolution is sized from that.
Acceptance metrics
Don’t judge by accuracy alone
On a line with 98% yield, an AI that passes everything still scores 98% accuracy while every defect escapes. We accept systems on two line-relevant metrics.
Never miss
Recall
The share of defects caught. Critical defects usually require 99% or more.
Don’t waste labour
False alarm rate
The share of good parts rejected. Too high overwhelms manual review; below 5% is a common target.
Few defect samples?
Rare defects are not a blocker
Anomaly detection
Train on good parts only and flag anything that deviates, without collecting many defects.
Data generation & augmentation
Generative AI and rotation, contrast and noise augmentation make the most of rare defects.
Continuous learning
Borderline images from the line are fed back for retraining, so the model keeps improving.
Case studies
All case studiesAI vision case studies
AI Vision Inspection
Wafer Offset AI Detection for Plating Equipment
Two cameras and AI image recognition track each wafer's path inside a plating tool in real time. When the offset exceeds the limit, a stack light, buzzer and relay trigger instantly, and every event's video and track data are kept.
AI Vision Inspection
AI Print Inspection for Coloured Foil Pouches
AI image recognition replaced the existing reader to inspect lot and expiry printing on every foil pouch inline, across light and dark packaging, triggering existing machine signals on failures with no re-tuning after installation.
Bringing AI to your production line?
Tell us what you want to recognise and we will help plan the deployment.