Portfolio

Defect Detection & Label Validation System

The Industrial Label Validation & Defect Detection System is a machine vision quality control station integrated directly into a packaging conveyor line. A high speed linear scanner inspects each unit as it passes through an inspection chamber, validating the printed label (product name, lot number, expiry date), decoding and checksumming the barcode/2D matrix, measuring label skew against a tolerance envelope, and classifying visual defects (e.g. print bleed smudges) all before a pneumatic diverter physically routes the unit to either the palletizer (pass) or a reject bin (fail), within the line's cycle-time budget.

Like the fatigue monitoring system, this is a hybrid pipeline rather than one model doing everything: deterministic machine vision and pattern matching handle the hard-real-time, must not be wrong checks (OCR extraction, barcode decode, skew geometry), while a AI layer handles the semantic judgment call classifying defect type and severity, and reasoning about root cause that a rules engine alone can't do well.

The Challenge

Hard real time actuation the pneumatic diverter must fire within a precise window tied to physical conveyor position; nothing in that decision path can afford AI model latency.

High-speed, motion blur free capture a 60 FPS line-scan camera running at up to 120 PPM throughput needs tightly synchronized triggering, not just a fast enough model.

Template/SKU binding risk expected value patterns (product name, lot format, expiry format) are per-SKU; if the wrong validation profile loads for the product actually on the line, every field reports a false mismatch.

False reject vs. missed defect trade off overly sensitive defect detection scraps good product and stalls throughput; under-sensitive detection lets a real label defect ship.

Safety critical actuation a vision triggered pneumatic piston operating on a live conveyor near an E-STOP governed motor is a physical safety system, not just a software decision.

Regulatory traceability food/consumer packaging typically requires a defensible, timestamped inspection record per unit, not just a pass/fail count.

Our Solution

The pipeline splits hard-real-time, deterministic work from semantic reasoning rather than routing everything through one model:

Machine vision (edge, deterministic): OCR field extraction, barcode/2D matrix decode with checksum validation, and skew-angle measurement all run as classical, low latency computer vision fast and predictable enough to sit in the actuation critical path.

Pattern/rules validation: each extracted field is checked against its SKU specific expected pattern (regex, date format, checksum) to produce a deterministic per field PASS/MISMATCH status this is what powers the "X/4 VALIDATED" readout.

AI reasoning: invoked for the judgment calls a rules engine can't make well classifying what kind of defect was found (e.g. print bleed smudge), assigning severity (Critical/Major/Minor), and producing the reject confidence score and root cause explanation.

A reject decision engine combines both deterministic validation and AI classification into a single pass/fail call, which then gates the physical diverter itself hard interlocked with the E-STOP, independent of the software decision path.

Technical Architecture Highlights

┌────────────────┐ ┌───────────────────────┐ ┌────────────────────────┐ ┌───────────────────────┐

│ Line-Scan │ │ Stage 1: Machine │ │ Stage 2: Pattern/Rules │ │ Stage 3: AI │

│ Camera (60 FPS) │──▶│ Vision — OCR, Barcode, │──▶│ Validation — expected │──▶│ Defect Classification │

│ @ Inspection │ │ Skew (deterministic, │ │ pattern match, checksum,│ │ & Root-Cause Reasoning │

│ Chamber │ │ edge, real-time) │ │ tolerance envelope │ │ (severity, confidence) │

└────────────────┘ └───────────────────────┘ └────────────────────────┘ └───────────────────────┘

│

▼

┌───────────────────────┐

│ Reject Decision Engine │

│ (rules + AI confidence)│

└───────────────────────┘

│

┌───────────────────────────────────────┴───────────────────┐

▼ ▼

┌───────────────────────┐ ┌───────────────────────┐

│ Pneumatic Diverter │ │ Line Dashboard & │

│ Actuation (Piston A), │ │ Traceability Log │

│ gated by E-STOP │ │ (Inspection ID, lot, │

│ hardware interlock │ │ defect, confidence) │

└───────────────────────┘ └───────────────────────┘

This page failed to load.

Refresh the page or re-upload the theme zip. If this persists, purge LiteSpeed/Cloudflare cache for HTML and theme assets.

Return home