CorteX

Sample efficient deep learning & inference system

Superhuman
Accuracy

Achieves superhuman accuracy with 9X lower escapes on high variance defects while not overkilling OK parts

Sample Efficient
AI

Rapidly teach models with as few as 3 images per defect type

Fast
Decisions

Up to 160 MP/s inference speeds quickly pass a decision on OK/NG

Adjustable
Tolerances

Pixel precise defect segmentation and adjustable tolerance

Rapidly train with sample-efficient AI

Manage AI model development through the four core functional centers: AI model labeling, AI training, Thresholds tuning, Production line data

Use intuitive, drag-and-drop interface to develop AI models

As few as 3 images needed to train AI models

High-Resolution Image Support, up to 50MP

High Speed Inference

Up to 160MP/s inference speeds

Maximum detection throughput up to 1800 parts/min

Maximum connect with 8 OptiX

Broader Integration and Compatibility

Expanded PLC Protocol Support: Broader compatibility with all major PLC protocols, including robust trigger integration.

Flexible Camera Options: GigE camera compatibility for versatile deployment.

Open Ecosystem: SDK available for third-party partners to build custom software extensions.

Central Management System for Scale

Manage and optimize models without disrupting production

Deploy AI models across production lines globally with CorteX's scalable architecture

Centrally manage AI training data & labels

Know the yield rate in advance before deploying new models with production simulation

Optimize quality criteria for yield

Instantly tune quality criteria across 6 adjustable attributes

Visualize impact on yield before deploying AI models to production

Use LIMIT-- a separate inference decision from OK or NG– to set aside parts based on certain thresholds to further validate

Not just defect detection. All-in-one inspection that provides:

Detection

Classification

Count

Thresholds

Location

Measurement

Depth

Barcode

The UnitX Advantage

Rule-Based Vision

Other AI / Deep Learning

UnitX CorteX

Effort

Requires engineers to develop complex, bespoke rules for each defect and part type

Requires 100s of images to train

Intuitive training interface requires as few as 5 images to train; no AI experience needed. Quick to adapt to new product and defect types

Accuracy

Only accurate for well-defined defects and parts with consistent environment, leading to escapes and overkill

Can fail on part position and orientation variability, sometimes requiring additional positioning tools

Accurate for defects that are complex and variable in shape, size, location, and presentation, automatically normalizing for position variability

Yield Optimization

Manual effort required to tune quality criteria. Unclear how criteria impact yield, leading to overkill

Easily tune quality criteria specific to each defect across a number of attributes. Visualize impact on yield before pushing changes to production