FMCG QA on Edge

The Challenge We Faced

A global Fast-Moving Consumer Goods (FMCG) enterprise - shipping thousands of packaged products per hour - struggled to guarantee consistent quality across an ever-expanding range of box sizes and configurations. Traditional vision-based Quality Assurance (QA) caught glaring defects, but frequent format changes, fast-moving lines, and variable packaging materials allowed subtle dimensional errors and minor surface damage to slip through undetected. Compounding the issue:

1. Conventional off-site data processing risked slowing the line or causing delays if network connectivity lagged.

2. Operators faced manual efforts to update manifests for each box type, introducing the risk of errors and inventory mismatches.

3. Rigid inspection thresholds assumed a single SKU line and had to be recalibrated whenever a new SKU was introduced.

Our customer needed a system capable of precise, real-time shape measurement and surface damage detection - all while integrating seamlessly with their established manufacturing execution software and gaining the capability to work with multiple SKUs on a single conveyor lane seamlessly.

Our Approach

We designed a comprehensive edge-based solution that combined 3D sensors for dimensional verification with RGB cameras for surface inspection. This platform provided:

1. Real-time scanning:

   • Dimension estimation using high-resolution point clouds helped to detect minuscule deviations in length, width, and height - even when switching rapidly between different box sizes.

   • Each package was photographed from multiple angles to spot tears, dents, label misprints, scuffs, or any other damage.

   • Automated reference checks on a known standard minimised manual intervention and calibration, ensuring consistent accuracy as SKU formats changed.

2. Edge processing:

   • All scanning and image analysis occurred locally, reducing latency and eliminating dependence on intermittent network connections. This eliminated the need for expensive, high-spec iPCs (Industrial PC).

   • Packages flagged as out-of-spec triggered automated sorting lines, separating defective or dimensionally incorrect items before they proceeded downstream.

3. Seamless integration with existing systems:

   • An API-based interface updated the facility’s existing software with each package’s exact dimensions and QA results, creating a live “manifest” for every shift.

   • The solution fits onto existing conveyors and tied into the company’s data infrastructure without needing a full-scale re-engineering effort.

A Future-Proof Inspection Architecture

The FMCG giant had strict requirements around throughput, reliability, and extensibility. Our response included:

1. High-fidelity 3D cameras:
   They captured dense point clouds for different conveyor heights and speeds on varying packaging materials. Our internal algorithms filtered noise from movement or inconsistent conditions.

2. Sophisticated yet lightweight AI models:
   Instead of running large, complex vision models that might bog down local hardware, we optimised the damage detection pipeline and models for edge-grade processors, ensuring real-time results without sacrificing accuracy.

3. Automated updates and model refinements:
   The system learned the new “normal” dimensions whenever a new box style was introduced. Operators only needed to confirm the baseline once, after which the software continuously adapted to slight variations.

4. Intuitive dashboards and alerts:
   Operators viewed pass/fail rates and live camera feeds on a centralised interface. If the system detected unusual spikes - eg, an uptick in corner tears - it flagged the line for immediate operator review.

High-Level Technical Approach

To address these challenges, our team designed two parallel inspection pipelines, each focusing on a different aspect of the QA process - surface integrity (local checks) and dimensional accuracy (global checks). Here are the main components:

1. Sensor layer:
   Each conveyor station houses two sensors - one 3D scanner overhead, and several RGB cameras angled for side/top visuals. An industrial-grade microcontroller at each station handles data intake.

2. Edge compute node:
   A compact, specialised onboard processor runs two parallel QA modules - Local Patch Analysis for damage and Global Dimension Analysis for shape verification.
   Each sensor node processes data locally, generating pass/fail flags for surface anomalies and dimensional compliance.
   Alerts (for suspect packages) are triggered without cloud round-trips, keeping conveyor flow uninterrupted.

3. Central orchestrator:
   A supervisory software layer orchestrates the entire line, connecting each edge node to the facility’s manufacturing execution system (MES). It aggregates relevant data (pass/fail counts, dimension stats, error codes) and logs them for real-time dashboards and batch reports.

4. Multi-sensor inputs:
   3D ToF sensor mounted overhead to capture a global view of each passing package’s shape at a very high frequency.
   RGB camera Positioned for high-resolution imagery of package surfaces, complemented by strobe lighting.

5. Local-global analysis:
   Local damage detector splits incoming images into localized patches. Each patch is inspected for any deviation against known “normal” references (via a learned model) and checks each patch for differences. Then, it assigns a local defect score (e.g., how likely a patch is torn, scuffed, or mislabeled).
   Global dimension verifier module pre-filters spurious 3D points using a simple outlier rejection algorithm (e.g., ignoring points too far from the main cluster). We compute estimated package dimensions (length, width, height) from 3D point clouds. Then, it compares them to allowable tolerances against expected SKU specifications.

6. MES Integration:
   Each edge node posts results (damage scores, dimension data, pass/fail flags) to a message queue or a REST endpoint, which the MES monitors.
   If a package fails either the local or global check, the orchestrator triggers a mechanical diverter that removes it from the main flow, sending it for manual review.
   Staff sees immediate status updates on a live dashboard: defect percentages, common types of damage, dimension compliance rates, and any line slowdowns.

The Impact

1. Reduced downtime and rework:
   Instant detection of dimension or surface issues prevented entire batches of incorrect packaging from traveling further. That cut rework times by over 40% compared to the previous system, which only caught major anomalies.

2. Enhanced accuracy across SKUs:
   By automatically tuning each sensor to new box sizes and design features, the solution maintained near 100% detection accuracy. Even subtle shape differences triggered an alert before merging into the final shipping queue.

3. Streamlined manifest management:
   The customer’s legacy MES now updates package dimensions and QA status in real time. Operators no longer needed manual data entry, eliminating a frequent source of human error and saving considerable administrative hours.

4. Future scalability:
   Since the platform was deployed on the edge - using industrial-grade sensors with onboard compute - adding new lines or expanding to other facilities became straightforward. Each inspection node runs autonomously, yet stays synchronized with the central database.

Lessons Learned

1. Edge processing is key:
   Offloading heavy data analysis to local sensor compute removed the bottleneck of external servers and network latency, a critical gain for high-speed packaging lines.

2. Adaptive thresholding beats static rules:
   Packaging changes are inevitable in FMCG, and static models quickly become obsolete. A learning-based approach kept misclassification rates low across fluctuating SKUs.

3. Early intervention minimises waste:
   Instant sorting for defective or dimensionally incorrect items prevented defective packages from mixing with conforming stock, significantly reducing re-sorting and disposal costs.

TLDR

1. Industry: FMCG, high-speed packaging

2. Technologies: 3D sensors, RGB cameras, on-board edge computing

3. Key gains: faster defect detection, seamless MES integration, accurate dimensional data, reduced manual oversight

4. Outcome: 40%+ reduction in rework time, near 100% damage detection accuracy, automated manifest generation

For this large-scale FMCG client, the RGB-D-based quality assessment solution did more than just improve QA metrics - it optimised the way they managed inventory and assured product quality. By unifying shape verification and surface analysis within a single, automated workflow, the plant could easily accommodate new product sizes while achieving near-perfect detection rates for damaged or mislabeled packages. Most importantly, everything tied back into their existing manufacturing systems, ensuring minimal disruption and maximum ROI.

This approach underscores a universal truth in modern manufacturing: when QA systems adapt on the fly, efficiency and product integrity rise to new heights.