Co-Founder https://xitter.cf/t.co/KwXjxubObe Advisory Board Cleveland State University Computer Eng. #IIoT #AI #ML #CxO #IndustrialTransformation #ICSCybersecurity #AR
United States
Joined January 2009
- Tweets29.2K
- Following4.9K
- Followers31.7K
- Likes24.7K
When operators believe an AI system is measuring them rather than the process, they stop feeding it accurate data and the data foundation degrades. Cybus CEO Peter Sorowka says the fix is keeping agent scope on assets and processes, and making visible exactly what the agent read. Partner content with Cybus. #cybus_ai #barbara_ai #hivemq_iiot
Sorowka also recommends being explicit about what is NOT collected, transparency about scope boundaries is as important as the technical controls themselves.
Cross-document reasoning fails when documents don't share a common structure or vocabulary. Kristen Sauter of Adlib Software: normalizing at ingestion, even imperfectly, reduces that failure surface significantly. Partner content with Adlib. #adlib_iiot
A factory in Spain cut hazardous waste by 64% in film capacitor production. Not by switching materials. By optimizing heat treatment processes and improving resin use efficiency. The waste was engineered out. Partner content with TDK SensEI. #TDK_iioT
Manufacturing waste reduction rarely comes from big capital investments. This case: yield optimization through process tuning eliminated the waste at the source. No new equipment. Just better understanding of the thermal and chemical parameters. Source:
tdk.com/en/sustainability/re…
Solving only one part of the OT media security pipeline creates a new vulnerability. The three controls are: an enforced inspection point at the perimeter, hardware-enforced one-way transfer between zones, and managed transfer that logs every movement. All three are needed, and the order matters. Partner content with @OPSWAT. #opswat_ics
Source: opswat.com/blog/the-usb-to-o…
Servicing equipment every 90 days regardless of condition means failures still happen between scheduled checks. Vibration data with a threshold at 4.2 mm/s triggers service when the machine actually needs it, not when a calendar says so. That takes months of historical baselines alongside live sensor streams, all queryable in one database. Partner content with @TigerDatabase. #tigerdata_ai
When AI cannot establish sufficient confidence in sensor data, degrade gracefully, escalate to a human, or move to safe operating state. Do not continue acting as though inputs are trustworthy. #influxdata_iiot
Sensor health checks AI should run continuously: missing or stale values, sensor drift, impossible readings, disagreement between redundant sensors, and whether a signal is consistent with related process variables.
Unscheduled downtime costs the manufacturing industry roughly $100 billion. Yet most AI systems generate five recommendations where three are right and two are wrong, and operators stop trusting them, which means the alert system itself becomes the reason downtime persists. Partner content with @AI4ProdOutcomes. #infinite_ai
A system that fires anomalies and alerts without specifying what needs to be done and when does not hold operator attention past the first month. Sustained adoption requires clear prescriptions with timelines, not pattern deviation alerts that leave root cause identification to a human investigator.