Defect Ontology

Defines a strict parent-child inheritance structure for flaw categories, ensuring logical grouping and drill-down analysis.

• Top-Level Classes: Cosmetic, Functional, Packaging, and Missing Components
• Sub-Classes: Under Cosmetic, you find Scratches, Dents, Discoloration, and Cracks
• Granular Leaf Nodes: Specific instances like 'Hairline Scratch on Screen' or 'Scuff on Bottom Chassis'

This hierarchy prevents the data swamp of free-text descriptions and enables roll-up reporting for root cause analysis.

Each defect node is enriched with structured attributes that quantify severity, location, and dimensions, moving beyond simple labels to parametric data.

• Severity Score: A numerical scale (e.g., 1-5) defining critical, major, or minor impact
• Spatial Coordinates: Bounding box or pixel mask data mapping the flaw's exact location on the product
• Physical Dimensions: Length, width, and depth measurements in metric units

These attributes feed directly into the Automated Disposition Engine to calculate precise recovery costs.

Ontologies map non-hierarchical links between defects, components, and causes to enable reasoning. This is the difference between a flat list and a true knowledge graph.

• Causal Links: 'Impact Damage' causes 'Screen Fracture' and 'Housing Deformation'
• Co-occurrence Rules: 'Battery Swelling' frequently co-occurs with 'Rear Panel Separation'
• Component Associations: Defects are linked to the specific bill of materials (BOM) part they affect

These relationships allow the system to infer hidden damage based on a single visible symptom.

Every leaf node in the ontology is anchored to a set of canonical reference images, providing the ground truth for training and calibrating Computer Vision Grading models.

• Golden Samples: High-resolution images representing the archetype of a specific defect
• Boundary Cases: Examples showing the threshold between acceptable variance and a defect
• Negative Samples: Images of acceptable manufacturing tolerances that must not be flagged

This visual grounding eliminates inter-rater reliability issues between human inspectors and AI.

The internal ontology maps bidirectionally to external industry-standard reason codes to ensure interoperability with carrier and retailer partner systems.

• RMA Reason Codes: Links to standardized return reason codes for customer-facing communication
• UNSPSC Classification: Aligns defects with the United Nations Standard Products and Services Code
• ISO 2859 AQL: Connects defect severity to Acceptable Quality Level sampling standards

This mapping ensures that a 'Minor Cosmetic Anomaly' is translated consistently across every system in the value chain.

The ontology is a living data product requiring strict version control to manage the evolution of product lines and newly discovered failure modes without breaking downstream analytics.

• Backward Compatibility: New defect classes are additive; deprecated classes are soft-retired
• Change Approval Workflow: Modifications require validation against historical inspection data
• Audit Trail: Every schema change is logged with a timestamp and rationale

This governance prevents model drift in production AI systems that depend on a stable label space.