Torna al blog

Emerson and Databricks Link OT Data Fabric to Enterprise AI

Emerson and Databricks are connecting AspenTech Inmation OT data with enterprise analytics and AI. The opportunity is stronger context; the engineering challenge is keeping security, safety and con...

Emerson and Databricks have announced a collaboration intended to connect industrial operational data with enterprise analytics and AI. The November 19, 2026 release centers on AspenTech Inmation, described as an operational-technology data fabric, and native access to that data from the Databricks Data Intelligence Platform.

What the collaboration covers

According to Emerson, Inmation can collect, contextualize and govern real-time and historical OT information across field, edge and cloud environments. The new collaboration is designed to make that contextualized information accessible alongside business and enterprise data in Databricks. Emerson also says users will be able to access AspenTech's AVA industrial AI capabilities and Inmation data within the Databricks environment.

The architectural goal is straightforward: avoid building a separate data extraction and contextualization project for every maintenance, performance or AI use case. A common operational data layer could allow an analytics team to combine equipment status, process history, production orders, quality records and maintenance events without repeatedly recreating tag mappings and asset context.

Where plants could see practical value

Predictive maintenance is the obvious example, but the more valuable use cases are usually specific. A team could compare pump vibration or valve behavior with operating mode and product grade, examine whether alarm patterns precede a throughput loss, or join energy consumption with production schedules. The benefit depends on preserving timestamps, engineering units, quality codes and asset relationships—not simply copying large volumes of tags into a lakehouse.

Existing control hardware remains the source of much of that operational context. A process plant may have controllers such as the Emerson DeltaV MQ controller, interface modules such as the DeltaV Virtual I/O Module, and conventional cards such as the KJ3002X1-BC1 analog input module. These are examples of the installed layer from which process information may originate; the announcement does not state that these exact products have a direct native Databricks connection.

The control boundary must remain explicit

The release discusses access, analytics, AI workflows and agents, but it does not define closed-loop control authority. That distinction matters. A recommendation generated from enterprise AI is not equivalent to a validated control action, and a lakehouse is not a substitute for deterministic control, interlocks or an independent safety instrumented system.

Safety functions require an especially firm boundary. A logic solver such as the Emerson VS3202 DeltaV SIS module belongs to a governed safety lifecycle with defined change control, diagnostics and proof-test requirements. Data from a SIS may support analysis, but write paths, credentials and AI-generated recommendations must not bypass the safety design. Nothing in the announcement establishes such authorization.

Questions the announcement leaves open

  • Connector topology: Emerson does not specify which collectors, protocols or deployment patterns provide the native access, or whether an existing Inmation deployment requires additional gateways.
  • Latency and completeness: no figures are given for event-to-query latency, buffering during an outage, timestamp resolution or preservation of source quality flags.
  • Security: plants still need a documented route through OT network zones, identity controls, certificate management, audit logging and one-way or read-only enforcement where required.
  • Data residency: regulated and geographically distributed sites must establish where raw, contextualized and derived data is stored and processed.
  • Lifecycle and cost: the release does not detail licensing, supported versions, upgrade dependencies or the operational ownership of models and agents.
  • Disconnected operations: it is not yet clear what functions are available for air-gapped sites or during loss of cloud connectivity.

A sensible adoption path

Plants should begin with one measurable, read-only use case rather than an enterprise-wide data move. First define the asset hierarchy, source tags, quality rules and expected business outcome. Then measure data latency and gaps through normal operation, maintenance outages and network interruptions. Only after the pipeline is trustworthy should the team train or deploy an AI model.

Model output should initially be advisory and compared with operator and maintenance findings. Any later move toward automated action needs a separate hazard review, authority model, rollback plan and management-of-change process. Procurement teams should also demand a bill of software versions, connector dependencies, data-egress assumptions and restoration procedures so the architecture can be supported beyond a pilot.

What this means for industrial data strategy

The collaboration is important because it acknowledges that enterprise AI is only as useful as its access to correctly contextualized plant data. Inmation can potentially reduce repeated integration work, while Databricks supplies a broader environment for data engineering and AI. The technical value will be determined by the quality of the OT context and the discipline of the control boundary—not by the number of tags made visible to an AI platform.

For operations leaders, the near-term opportunity is better cross-domain analysis with less custom plumbing. The near-term risk is allowing a convenient data path to become an undocumented control path. A successful deployment should make more operational information available without weakening the deterministic, cybersecurity and safety responsibilities of the systems that run the plant.

Source: Emerson, “Emerson and Databricks Bring Industrial and Enterprise AI Together,” published November 19, 2026. Official announcement. Editorial analysis by PLC ProTech, November 20, 2026.

Lascia un commento

Si prega di notare che, prima di essere pubblicati, i commenti devono essere approvati.