Line-side inference and machine health, without a PLC rip-out.
The operational problem. High-mix plants are drowning in sensor data that never reaches the maintenance or quality systems that need it. Existing PLCs are oversubscribed, vendor telemetry is locked, and a full rip-and-replace isn't in this year's capex. You need inference at the cell, not in a dashboard five hops away.
Case A · Vibration-conditioned maintenance
Three-axis MEMS + on-edge FFT inference on injection-molding cells. Model retraining runs locally on each gateway; alerts land in the existing CMMS via OPC UA. 8 plants, 1,140 cells.
Case B · Vision-guided defect triage
2D line-scan cameras stream into Pervasive Edge Linux gateways running quantized defect classifiers; suspect parts are physically diverted via a discrete I/O path, not a network round-trip. Food & beverage tier-1 supplier.
Case C · Energy-aware OEE
Per-machine kWh + cycle counter reconciled against MES orders to surface true OEE without a single source of truth migration. Validated at a Tier-1 automotive supplier since 2022.