Precision Manufacturing • AI Vision

Visual Defect Detection for Precision Manufacturing

Machine vision systems for makers of precision parts, with models trained on your own defect samples and acceptance criteria agreed before deployment.

Deployed on production lines in mainland China: diamond dicing blades and semiconductor O-rings
Dicing-blade grain analysis · real result
Client's existing tool (coarse segmentation)
Client's existing tool (coarse segmentation)
JunCore model (per-grain delineation)
JunCore model (per-grain delineation)
Up to 15%
Grain-count precision
vs. the client's existing microscope tool
7
Defect types
agreed with the client before development
1.5–100 mm
Seal sizes
most around 30 mm

How a project works

Every project starts from your parts and your defects, not a generic model.

01

Define the standard

We start from your own defect samples and agree the defect list and acceptance criteria with you before any development.

02

Build the system

Industrial camera, custom lighting (including backlight for translucent parts), optics, an edge industrial PC and handling mechanism, and a model trained on your defects.

03

Accept it on your line

The system is tested on site, on your parts, against the criteria agreed up front.

Applications

Precision parts where manual inspection is slow, tiring and inconsistent.

O-rings & seals for semiconductor equipment

1.5–100 mm
  • Foreign matter inside the part and on the surface
  • Cracks, pits and bubbles
  • Tears along the mould line; uneven surfaces
  • Translucent, curved parts that make manual inspection slow and inconsistent
Deployed on the client's production line

Diamond dicing blades

Grit coverage
  • Delineates each diamond grain, even where densely packed
  • Up to 15% more precise grain counts than the client's existing tool
  • Per-grain area and a true coverage metric for process control
Deployed on the client's production line

Real detection results

All images below come from the deployed projects.

Diamond dicing blades: every grain delineated

Raw microscope image
Raw microscope image
Client's existing tool (coarse segmentation)
Client's existing tool (coarse segmentation)
JunCore model (per-grain delineation)
JunCore model (per-grain delineation)

O-rings and seals: foreign matter, fibres, scratches and nicks

Original
Original
Detected
Detected
Original
Original
Detected
Detected
Original
Original
Detected
Detected
Original
Original
Detected
Detected
Original
Original
Detected
Detected
Original
Original
Detected
Detected

Technology

Built for reliability, not demos

We design optics, lighting and AI together, calibrated to your part's material and defects — not a generic model running on generic hardware.
Optics
Industrial cameras with custom lighting, including backlight and transmitted light for translucent parts.
AI models
YOLO, ResNet and EfficientNet, trained and tuned on each client's own defect samples.
Deployment
Edge industrial PC or on-site GPU server, integrated with the line's handling mechanism.
Validation
Accuracy measured on your parts, against criteria agreed before development — not on a demo set.

Contact

Talk to engineering, not sales

For inspection projects, pilot lines and technical evaluation.

Shanghai, China

Outside China: we plan to serve Malaysia and Southeast Asia through our partner VASTLI.

Email
Email Us →Reply within one working day