Super Sorting — Vision Grading
A raw-material mismatch diagnosed, then engineered out with vision grading
01 — Results
Off-spec input
Raw material well under the line's design size window
6
SKU grades classified in-line
Sub-mm
Foreign-material detection spec
| Metric | Before | After | Delta |
|---|---|---|---|
| Grading method | Manual visual grading | In-line vision classification | Spec-driven |
| SKU grades | Judged by eye | 6 spec windows enforced | Deterministic |
| Foreign material | Downstream detection only | Sub-mm in-line spec | Upstream catch |
| Trimming | Double-trim on off-spec | Off-spec diverted pre-portioner | Rework removed |
02 — Challenge
The portioning line was designed for large-format raw input. It was actually being fed smaller, defect-heavy loins and tails.
The mismatch showed up as double trimming, WIP growth and starvation at the portioners, with grading calls being made by hand on the line.
03 — My role
Author of the business case and technical spec
Timeline: December 2025
04 — Scope
- Root-cause analysis: the line was designed for large-format raw input but was actually receiving smaller, defect-heavy loins and tails — driving double trimming, WIP growth and starvation at the portioners
- Six finish SKUs mapped to defined weight and quality windows, including a zero-defect premium grade
- Vendor evaluation across two grading platforms (multi-kicker wide-belt grading with sub-millimeter foreign-material detection), plus a line-analytics dashboard trialed alongside
- Portioner programming studied for tail and knub cuts to eliminate bandsaw steps
- Future-state flow defined from singulated merge output through classification to batching
05 — Execution
- Root-cause analysis linking the input spec gap to trimming, WIP and starvation
- Six finish SKUs mapped to defined weight and quality windows, including a zero-defect premium grade
- Vendor evaluation across two grading platforms — multi-kicker wide-belt grading with sub-millimeter foreign-material detection — plus a line-analytics dashboard trialed alongside
- Portioner programming studied for tail and knub cuts to eliminate bandsaw steps
- Future-state flow defined from singulated merge output through classification to batching
06 — Systems & technology
- Machine vision
- In-line grading
- Portioner programming
- Line analytics
07 — Artifacts & evidence
- Business case and technical specification authored in-house
- Vendor comparison across competing in-line grading platforms
08 — Key takeaways
The line was not underperforming — it was being fed the wrong fish. Specify the input before re-specifying the equipment.