Projects / Magna
Streamed strain-gauge data from a hydraulic press over MQTT into PostgreSQL, with a live web dashboard for operators.
An anomaly detector learns each part's normal stroke envelope; it caught 100% of cracked parts with zero false positives across a 150-part production run. A patent-pending, semaphore-driven multithreaded buffer on a memory-limited PLC raised the data rate from 142 Hz to 500 Hz.
A real-time integrity monitor for an industrial hydraulic press, using strain gauges embedded near the die. I built the pipeline end to end: sensor data streamed over MQTT, re-synchronized and stored in PostgreSQL, plus a web dashboard where operators and engineers compare strain profiles across batches and review anomalies.
I designed an anomaly detection algorithm that finds the characteristic discontinuity in the strain signatures across four sensors when a part cracks during forming. Running in real time, it caught every cracked part with zero false positives over a 150-part validation run, removing defective parts from the line automatically. It recognizes when a different part is being stamped, and a moving-window average stops slow sensor drift from being flagged.
The press's PLC had very little memory. A patent-pending buffering system I engineered, with semaphore-driven multithreading and staged data queues, let it stream far more data, raising collection from about 140 Hz to over 500 Hz for much finer analysis of each stroke.
