Why it matters
Without AI grading, inconsistent sizes, bruising, and colour defects lead to rejected retailer loads and failed audits. Manual checks miss subtle defects and slow throughput, increasing risk across quality and safety records.
From design to safety
Design
AI guides grading system concept
AI readiness defines grading camera placement, lighting, target cap size classes, and defect taxonomy feeding the system for early integration into line design and processing flow.
- grading camera
- cap size classes
- discolouration sensors
Build
System integration and calibration
During build, calibrations align grading camera outputs with reference samples, set defect thresholds, validate throughput against target weekly pack counts, and document baseline performance in the commissioning dossier.
- grading camera
- open veils sensor
- discolouration sensors
- bruising severity sensor
Operation
Real-time grading and sorting
Live inference reads cap size, bruising, open veils, and colour, then assigns acceptance, rework, or discard, logging batch IDs and KGs per lot with live operator prompts.
- grading camera
- cap size classes
- bruising sensors
- discolouration sensors
Maintenance
Vision system upkeep
Routine cleaning, recalibration after lamp changes, firmware checks, and sensor health monitoring maintain accuracy, with drift flags and maintenance tickets linked to line run logs.
- grading camera
- lighting system
- discolouration sensors
Safety
Traceability and HACCP compliance
System-generated records support HACCP, GLOBALG.A.P., and third-party audits, with access controls and signed-off change requests by QA and certifications officer.
- grading camera
- discolouration sensors
- cap size classes
Part by part
- 1
grading camera
- Watch for
- Drift in color grading thresholds and misclassification of cap sizes.
- AI action
- Read real-time defect signals; trigger re-calibration or operator rework alerts.
- 2
cap size classes
- Watch for
- Mislabeling of cap size categories or shifts in average cap diameter.
- AI action
- Update grading tiers and flag outliers for manual check.
- 3
open veils sensor
- Watch for
- Veil opening not consistent, affecting appearance scoring.
- AI action
- Flag infeed for re-inspection and adjust lighting angle.
- 4
bruising severity sensor
- Watch for
- Bruise patterns beyond threshold levels in batches.
- AI action
- Trigger batch hold and allocate rework line slot.
- 5
discolouration sensors
- Watch for
- Unusual darkening or uneven colour distribution.
- AI action
- Adjust classification bins and alert QA for sampling.
- 6
lighting system
- Watch for
- Shifts in luminance affecting defect detection.
- AI action
- Run quick calibration check and notify operator.
Worked example
Catching a grading anomaly before a shipment
Illustrative scenario with sample data.
A routine packline inspection identifies a persistent drift in cap size classification and an uptick in bruising scores over two shifts. This triggers a closer QA review and immediate revalidation of camera calibration, lighting settings, and reference samples.
Signal in
Line: PackLine-3; Time: 09:12; BatchID: B-3421; CapSizeTarget: 38–48mm; DefectThreshold: bruising 1.8; VeilStatus: normal; Temperature: 3C; Lighting: 650 lux
System out
Alert: grading drift detected; Action: reclassify Batch B-3421; Hold 2 pallets; QA review required; Calibration queued; Rework queue opened; Disposition: buffer for further assessment.
Human sign-off: QA supervisor approves hold and commands corrective calibration; packline supervisor verifies downstream sorting before release to shipping, then updates the baseline reference for future batches.
How to roll it out
Phase 1
Weeks 1 to 4
Baseline data capture, camera calibration, and initial threshold tuning on one packing line to establish a stable reference.
Phase 2
Weeks 5 to 8
Wider rollout to other lines, standardized operator alerts, QA sign-off procedures, and initial trend reporting.
Phase 3
Weeks 9 to 12
Full operation with integrated reporting, traceability, and continuous improvement loops across all lines.
Where people stay in charge
- Final acceptance decisions must stay with the QA manager or packhouse supervisor and be signed off.
- Operators cannot override AI grading without documented justification and immediate supervisor authorization at all times.
- Calibration changes or threshold updates require a formal change request with QA sign-off before deployment.
Questions owners and crews ask
What is AI grading in a mushroom packing line?
AI grading uses vision sensors to assess size, surface defects, and colour, classifies lots, and directs flow for acceptance, rework, or discard with traceable records.
How do you handle bruising detection on the line?
Bruising sensors feed the AI, which flags affected trays, routes to rework, and logs incidents for QA review and corrective action.
How is cap size accuracy maintained?
Cap size classes are calibrated against reference samples, with real-time inference updating bins and triggering operator checks when drift occurs.
What records are produced for audits?
Defect logs, batch IDs, disposition outcomes, calibration history, and access-controlled QA sign-offs are stored for audits and certification.