Federal Package

Federal Package, a Chanhassen, Minnesota-based contract manufacturer with over 40 years serving leading beauty and personal care brands, faced a compounding quality risk: labor shortages were undermining the reliability of manual inspection on a high-volume deodorant line. In consumer goods, where contract manufacturers are custodians of their customers' brand equity, a defective product reaching retail shelves can damage relationships that took years to build. Two inspection checkpoints required consistent human judgment — identifying drips caused by overfilling and verifying correct label placement, position, and orientation across 30 to 40 distinct deodorant SKUs differentiated by fragrance, color, and naming. Manual inspection at that level of product variety introduced variability that automated throughput could not absorb.

Federal Package deployed two Cognex In-Sight 2D vision systems — one monochrome, one color — each embedded with edge learning technology, a form of AI that runs inference directly on the device without relying on cloud connectivity or centralized compute infrastructure. The monochrome system inspects both the front and back of each deodorant body for drips indicating overfilling, classifying each unit as acceptable or non-conforming. The color system handles label inspection, where the ability to distinguish label boundaries from the container body required chromatic differentiation. Cognex's edge learning classify tool was trained using labeled images of good and defective products, enabling the system to generalize across the full SKU range — filtering out legitimate cosmetic variation in colors and graphics while remaining sensitive to actual defects. The systems were set up by Federal Package's maintenance team following a one-hour initial deployment.

The automated inspection system now covers 100% of product output, a throughput guarantee that manual inspection could not provide. Defect detection accuracy exceeds 99%, minimizing the volume of nonconforming units requiring human remediation. Production throughput is maintained at approximately 80 units per minute on the drip inspection line and 60 units per minute for label inspection. Key operational outcomes include:

• 1-hour deployment from installation to production-ready operation
• 5–10 minute changeover training when switching between deodorant product runs
• Less frequent retraining than anticipated, improving overall line efficiency
• Maintenance staff successfully took ownership of system setup, freeing engineering capacity for other automation projects

• Edge learning is well-suited for high-SKU consumer goods environments: it requires fewer training images than deep learning and can generalize across cosmetic variation without retraining for every product variant.
• Embedding AI inference on-device eliminates cloud dependency and simplifies deployment in production environments where network connectivity or latency is a constraint.
• Rapid deployment timelines (under one hour) make edge AI vision systems viable even for smaller-scale contract manufacturers evaluating automation ROI.
• Assigning system ownership to maintenance rather than engineering staff accelerates adoption and sustains operation without specialized expertise.
• Contract manufacturers should evaluate automated inspection as a brand protection investment, not merely a quality cost — defect escapes carry reputational risk beyond the manufacturer's own margin.