Edge ai software for real-time quality prediction and equipment monitoring on existing plcs

Critical quality variables in manufacturing often remain unmeasured until offline lab analysis introduces hours of delay. Deep learning addresses this but requires GPU infrastructure, making it impractical at the production line. Our technology fills this gap: more accurate than statistical methods, lighter than neural networks. The software provides: (1) Soft sensors estimating hard-to-measure variables in real time — e.g., product quality during spray drying or mixing from temperature, pressure, and flow data. (2) Anomaly detection that learns normal behavior and flags equipment deviations, with per-sensor decomposition identifying which component is drifting. (3) Remaining useful life prediction. (4) Sensitivity analysis revealing which variables influence outcomes as interpretable polynomial expressions — auditable and deterministic. Models train in minutes and adapt to new lines with 10–20 data points via transfer learning. Entrox is a DLR (German Aerospace Center) spin-off. We are conducting pilots with DAIKIN (soft sensors deployed on customer microcontrollers) and Toray (materials production). Time to initial results: 2–3 hours from data arrival. We seek a pilot partner in consumer goods manufacturing to validate on processes such as spray drying, mixing, blending, or packaging line monitoring. A pilot would apply our software to existing sensor data, demonstrate prediction accuracy, and evaluate edge deployment. https://entrox-systems.com