Veteran engineers’ sensor know-how turned into 24/7 operations
“The sensor data was there. What no one could do was keep interpreting every signal.”
Problem
The equipment emits dozens of sensor signals every second, each meaning something different. Signals inside a defined range can be checked against the manual, but the changes and combined signals the manual does not cover could only be read by veteran engineers.
- Unreadable signals — changes and combinations the manual does not cover
- No continuous watch — no one can watch time series at that volume
- Personnel dependency — judgment rests on the few who can read the signals
What BeSir did
Build 24/7 sensor monitoring so that even a non-expert can immediately see an abnormal state and what to do about it.
- Structures the relationships between sensor states and signals as an ontology, analyzing the time-series stream 24/7 to detect signals that deviate from normal
- Encodes how veteran engineers judge and respond to anomalies as the response criteria
- Monitors every sensor signal on a single dashboard
- On an anomaly, presents the related sensors, the likely cause, and the relevant manual and action
- Sends an automatic alert the moment a sensor value crosses its threshold
Results
- PoC
Every sensor signal structured into one real-time model
- PoC
Response procedures and manuals linked by anomaly type
- PoC
Non-experts can read equipment state and the required action
The data preprocessing foundation for 24/7 real-time detection was built. The project ran at a national security facility, so site identifiers, system names, screens and raw sensor data are excluded here; signal types and sampling rates are outlined only.
Verified confirmed in the operating environment · PoC confirmed within PoC scope · Target a target set on work in progress · Estimate calculated from the stated assumptions