FMCG QA on Edge

Client

FMCG Leader in EU

Industry

Manufacturing

Location

Confidential

Year

2025

Stack

3D ToF Sensors, RGB Cameras, Edge Computing, Python, MES Integration

A global FMCG manufacturer needed more than traditional QA. As its product lines grew, minor surface damage and dimensional inconsistencies began slipping through legacy systems. We built an edge-first inspection platform that uses 3D vision and learned damage detection to catch flaws in real time, even as packaging formats change constantly. The result is faster lines, fewer errors, and QA that keeps pace with demand.

The problem

A Fast-Moving Consumer Goods enterprise shipping thousands of packaged products an hour struggled to hold consistent quality across expanding box sizes and configurations. Traditional vision-based QA caught obvious defects, but frequent format changes, fast lines, and variable materials let subtle dimensional errors and minor surface damage through.

Three problems defined the brief:

• Off-site processing risked slowing the lines or stalling on network issues.
• Operators updated manifests by hand for each box type, inviting error and inventory mismatch.
• Rigid inspection thresholds built for single-SKU lines needed recalibration whenever a new SKU appeared.

The client needed precise, real-time shape measurement and surface-damage detection that fit into existing manufacturing execution software and supported multiple SKUs on a single conveyor lane.

The approach

The design is edge-first: 3D sensors for dimensional verification and RGB cameras for surface inspection, running two inspection pipelines side by side, surface integrity as local checks and dimensional accuracy as global checks.

All scanning and image analysis happen locally, which removes network dependency and keeps latency low. High-resolution point clouds detect deviations in length, width, and height as box sizes change, while multi-angle photography catches tears, dents, label misprints, and scuffs. Packages flagged out-of-spec trigger automated sorting before downstream processing, and automated reference checks against known standards keep manual calibration to a minimum. API interfaces feed exact dimensions and QA results back to facility software, and the system fits onto existing conveyors without full re-engineering.

The build

Each conveyor station carries a 3D scanner overhead and multiple angled RGB cameras, with an industrial microcontroller handling intake. The multi-angle setup catches damage a single viewpoint would miss, such as underside scuffs or side-panel dents. The 3D cameras capture dense point clouds across varying heights, speeds, and materials, and internal algorithms filter movement noise in real time to produce clean geometry even at line speeds above 100 units per minute.

A compact edge processor runs two parallel QA modules. The local damage detector splits each image into patches and scores every patch against learned “normal” references, so it pinpoints where damage occurs, not just whether it exists. The global dimension verifier pre-filters spurious 3D points, computes package dimensions from the cleaned point clouds, and compares them against SKU tolerances, flagging packages that fall outside bounds by even a few millimetres. Both modules run locally, keeping latency under 100ms from scan to decision, with model inference under 50ms per frame. When a new box style appears, the system learns its baseline dimensions after a single operator confirmation, removing the recalibration downtime of the previous setup.

A supervisory layer connects each edge node to the facility’s manufacturing execution system (MES). Edge nodes post damage scores, dimension data, and pass/fail flags to message queues or REST endpoints the MES monitors, failed packages trigger mechanical diverters for manual review, and a central dashboard shows pass/fail rates, common damage types, dimension compliance, and line slowdowns across every line. A sudden spike, such as a rise in corner tears on one line, raises an immediate review flag so supervisors can step in before defective batches accumulate.

Outcomes

Metric	Result
Rework reduction	40%+ against the previous system
Detection accuracy	Near 100% across all SKUs on each line
Latency	Under 100ms from scan to decision
Manifest handling	Real-time dimensions and QA status to MES, no manual entry
Deployment	Six months, on existing conveyors

Instant detection stops entire batches of incorrect packaging from moving down the line, cutting rework by more than 40% against the previous system, which only caught major anomalies. The faster feedback loop also surfaces root causes, a misaligned conveyor guide or a faulty sealer, before they compound. By tuning automatically to new box sizes, the platform holds near-100% detection accuracy across the full range of SKUs on each line, including seasonal and promotional variants. The MES receives dimensions and QA status in real time, removing manual data entry and the errors that came with it, and because each inspection node runs autonomously on industrial-grade sensors with onboard compute while staying synchronized with the central database, adding lines or facilities does not require rearchitecting.

Two lessons shaped the result. Running inspection on the edge removed the bottleneck of external servers and network latency, which for high-speed lines where milliseconds matter was the single most effective architectural decision. And static thresholds break down fast in FMCG, where packaging changes constantly, so a learning-based approach that adapts as new formats appear kept misclassification low without manual recalibration. Catching defects before they mixed with conforming stock cut re-sorting and disposal costs substantially, since the earlier a defective item leaves the flow, the less labor and material is wasted.

This project applies Algorithmic’s computer vision at industrial scale, real-time detection, edge deployment, and integration with existing manufacturing systems. The backend infrastructure was designed for autonomous edge operation with centralized monitoring, a pattern we apply across factory, logistics, and quality-assurance environments.