The operational and information technology divide is a governance divide
Plant systems and enterprise systems are usually governed by two different models, run by two different teams, with a firewall and a spreadsheet between them. The moment a regulator, an insurer or an incident asks a question that spans both, the answer has to be assembled by hand, and the part of the estate with the weakest identity story is the part physically closest to the process.
Meanwhile the assets that matter most sit on links you do not control: a substation, a rig, a vessel, a remote pumping station. A platform that treats a partition as an error cannot operate there, so the workaround is a second stack, and the gap between the two is permanent.
One substrate, reaching the plant with its own protocols and its own governance, closes the divide rather than bridging it.
What reaches the plant
Protocol reach into the process
OPC-UA, MTConnect, Modbus, DNP3, IEC 61850, BACnet, PROFINET, EtherCAT and CAN bus, plus historian and controller integrations, seismic and drilling formats, smart-meter and substation automation feeds. Industry stream packs carry parsers and monitors for aviation, energy, maritime, logistics, manufacturing and oil-and-gas telemetry.
Prognostics that show their reasoning
Asset prognostics on the open condition-monitoring model: condition-indicator featurization across spectral, motor-current-signature and bearing-defect families, state detection, remaining-useful-life bounds with backtesting, and fleet health boards. Advisories show their rationale rather than emitting a score, which is the difference between something an engineer acts on and something an engineer overrides.
The same features, computed on the device
The condition-indicator feature set runs on the edge agent as well as the core, so an asset can be scored where it lives rather than waiting for a link. The device carries a governed sandbox, local inference, on-board and hardware sensor drivers, and a hash-chained journal that keeps recording through an outage.
Identity for things that cannot log in
A substation device gets a cryptographic identity bound to its physical hardware, and a delegation the control centre can revoke in one call, effective on the next request, with no certificate rotation and no site visit. A rogue device cannot inject control signals, and a compromised endpoint is isolated without anyone driving to it.
Detection that spans both halves
Industrial protocols are inbound connector types, so a plant's sensors land in the same event plane as its endpoints and are correlated by the same engine. A cross-environment pattern that is faint at any single vantage point becomes visible when observed from many at once.
Sensing beyond the historian
Radar, sonar, thermal and infrared, laser and seismic-acoustic contact management with covariance-intersection fusion; LiDAR point-cloud processing; image, video and audio analysis; RF and spectrum sensing; satellite tracking and positioning. These exist because of the mission work, and an industrial estate gets them on the same governed substrate as its historian data.
Two things usually missing from multi-sensor work
Clock synchronisation across sensors, without which fusion is guesswork rather than analysis, and time-difference-of-arrival aperture synthesis across distributed receivers, which is what turns several cheap sensors into one useful one.
Both are in the substrate rather than in an application on top of it, which is why they apply to a defense sensing problem and a rotating-equipment problem in the same way.
The crossover is the differentiator. Very few vendors can hold both ends of it, and the ones that can usually charge you twice.