Manufacturing plants rarely have a shortage of data. Historians capture years of operating information. SCADA and DCS systems know what is happening on the plant floor. ERP platforms hold business and maintenance information. Engineering teams work with P&IDs and thousands of technical drawings. Sensors and cameras create new streams every second.
The problem is that these sources often tell different parts of the same story.
A maintenance engineer looking at a pump may need its tag, current operating values, historical trends, engineering drawing, OEM documentation and maintenance history. When those pieces sit across different systems, finding the answer can take longer than acting on it. The CoreShift session noted that a typical factory can have more than 10,000 technical drawings, while disconnected tag registries, drawings and engineering systems can consume significant maintenance time and even introduce plant safety risk.
That is the context behind Industrial AI for Smarter Manufacturing Operations, a CoreShift 2026 session exploring how manufacturers can connect operational data, enterprise information, AI, ontologies and digital twins rather than treating each as a separate initiative.
Industrial AI Is an Architecture Problem Before It Is an AI Problem
Much of the discussion around AI in manufacturing starts with the model: predictive maintenance, machine vision, anomaly detection or production optimization.
In practice, the model is only one layer. The session presents Industrial AI as an end-to-end architecture. Data can originate from historians, sensors, cameras, business applications, engineering information and external sources. Some information needs real-time processing at the edge. Other data belongs in an enterprise Data & AI platform where it can be cleaned, contextualized and combined with business information. AI, machine learning, machine vision and GenAI then sit on top of that foundation, with results delivered through Power BI, Copilot, agents, alerts or digital twins.
This distinction matters because an isolated predictive model may answer one narrow question. A connected architecture allows the enterprise to reuse the same data foundation across predictive maintenance, quality, OEE, engineering, planning and other manufacturing operations.
That is also the principle behind Acuvate’s broader approach to Industrial AI for manufacturing: connect operational and enterprise information first, then apply AI where it can influence a real business process.
Build the Data Foundation Before Scaling Industrial AI
The first question is not simply, “What AI model should we build?” It is, “Do we have the right data, at the right speed, with enough context to make the model useful?”
The session identifies several possible inputs: PI Historian time-series data, additional plant sensors, cameras, robotics and drones. Different inputs also create different architectural requirements.
Real-time does not mean everything belongs in the same place
High-speed sensor messages and high-volume camera imagery are not identical workloads. The session positions Microsoft Fabric Real-Time Intelligence for high-speed streams of relatively small sensor messages. For larger inputs such as camera images, Azure IoT Operations and edge-based machine vision can bring processing closer to the equipment when faster action is required. PI data can meanwhile stream from plant gateways into an enterprise platform for immediate AI and analytics.
This edge-and-cloud split is important for real-time data analytics. The objective is not to move every byte into a centralized platform as quickly as possible. It is to decide where processing should happen based on latency, data volume and the operational decision being supported.
Acuvate’s real-time industrial solutions and AcuNow edge platform provide additional context on this edge-to-enterprise pattern.
The enterprise platform provides consistency
Once data leaves isolated systems, it still needs quality, structure and governance. Within the session, AcuPrism – the enterprise Data & AI layer where organizations can align asset-tag naming conventions, address data-quality issues, bring multiple sources together, prepare information for ontology services and develop or run models for scenarios such as overall equipment effectiveness and predictive maintenance. Security is treated as an end-to-end requirement rather than a separate final step.
For enterprises building this layer, the AcuPrism Enterprise Data Platform provides further detail on how industrial and enterprise information can be managed within a common foundation.
Make Asset Context Usable Across the Plant
Connecting data technically does not automatically make it useful operationally. Asset identity is a good example. A manufacturing site may contain pumps, engines, compressors, turbines, conveyor belts and production lines. Each has a Tag ID, but organizations can end up with multiple registries, spreadsheets and drawing systems that are incomplete or out of alignment. Engineers may know the equipment they need, but not which drawing contains the latest information.
This is where the asset management process becomes a data-context problem. The session shows DiagramIQ using an Enterprise Tag Registry as the governed asset reference, with AI scanning engineering drawings for tag information. P&IDs can then be digitized and represented in a graph database, preserving not only the individual tags but also their relationships. A Copilot-based interface can query information held across the Tag Registry and P&ID graph using natural language.
That changes P&ID digitization from a document-conversion exercise into a foundation for connected asset intelligence.
Ontologies Give Industrial Data Business Meaning
A connected plant still needs a common way to describe what its data means. The session positions ontology as the layer that defines business structures and dependencies so users can work with familiar business concepts rather than underlying system structures. It supports questions across past, current and future performance and can be consumed by business users, dashboards, digital twins, Copilot and agents.
Microsoft Fabric IQ extends this idea through entities, relationships, rules and metrics organized into a governed ontology. The session also connects ontology to broader enterprise context through Foundry IQ, Work IQ and Web IQ.
For enterprise asset management, this matters because a pump should not exist merely as Tag P-101. The organization needs to understand what production line it belongs to, which work orders relate to it, what documents describe it, which measurements represent its condition and which business processes depend on it.
A Digital Twin Becomes Useful When It Connects the Full Asset Story
A digital twin should do more than visualize equipment. The session describes a practical model in which operations, engineering and maintenance work from aligned information. Manufacturers can begin with a 2D experience based on smart P&IDs and then connect the selected asset to PI data, ERP information, OEM documentation, camera images and other relevant sources.
The integrated PNID.IO–AcuPrism view shown in the deck takes this further: ERP, OT historians, SharePoint or Microsoft 365, engineering drawings, business applications and other plant sources can feed a connected data layer and surface live information directly against the P&ID.
That is where digital twins become valuable to maintenance operations. Instead of searching separately through systems, the user starts with the physical asset and moves outward to the information needed to understand or act on it.
What Manufacturers Should Do Next
The most useful takeaway from the session is that Industrial AI does not need to begin with a plant-wide transformation.
The experts explicitly recommend starting where the business has the greatest problem and expanding step by step.
A practical way to apply that guidance is to:
- Choose a measurable operational problem such as predictive maintenance, OEE, quality or slow access to asset information.
- Identify the OT, engineering and enterprise data required to solve it.
- Establish trusted asset identities and business context before adding more AI.
- Decide which processing belongs at the edge and which belongs in the enterprise platform.
- Put data quality, access, security and governance into the architecture from the beginning.
This creates a reusable foundation instead of another isolated proof of concept. The same connected plant context can then support new AI models, agents, digital twins and decision workflows as requirements grow.
Industrial AI becomes genuinely useful when manufacturers stop treating data platforms, OT connectivity, asset management systems, AI and digital twins as separate technology projects. The value comes from connecting them around the decisions plant teams need to make.
Industrial AI for Manufacturing - FAQs
Industrial AI combines operational and enterprise data with machine learning, machine vision, GenAI, edge processing and analytics to support manufacturing decisions across areas such as maintenance, quality, OEE and asset performance.
Predictive maintenance uses historical and streaming equipment data to develop and run AI models that identify operating patterns, detect abnormal behavior and support earlier maintenance decisions.
OT systems contain the real operating information of the plant, including historian, SCADA, DCS and sensor data. Connecting this information with enterprise context gives AI a more complete view of equipment and processes.
Edge processing is useful when decisions require low latency or when high-volume data such as camera imagery needs to be analyzed close to the equipment. Other analytics and enterprise workloads can run centrally in the cloud data platform.
Ontologies organize business entities, relationships, rules and terminology so AI, agents, dashboards and digital twins can interpret manufacturing information within its business context rather than as disconnected data points.
P&ID digitization extracts asset tags and relationships from engineering drawings, making the information searchable and allowing it to be connected with Tag Registries, graph databases and other operational systems.
Connected digital twins bring engineering, operational and enterprise information together around an asset, helping operations, engineering and maintenance teams work from a more complete and aligned view.
Manufacturers should begin with a clearly defined operational problem, connect the data needed for that use case, establish trusted context and governance, and then expand the architecture step by step as value is demonstrated.
