The challenge
Unplanned failures on induction furnace auxiliaries and moulding line motors were costing the foundry roughly eleven production days a year, with each stoppage risking a solidified melt.
What we did
Battery-powered vibration and current-signature nodes on 84 rotating assets, reporting into Pulse over a LoRaWAN backhaul chosen for the plant's RF environment. Baseline models were learned per asset over six weeks, then anomaly scoring drove maintenance work orders. The nodes were specified for a foundry's thermal and dust conditions first, sensitivity second.