Data Modeling Strategies for Manufacturing Digital Twins: Edge Processing and Open Standards

When Microsoft began bringing industrial assets to the cloud over a decade ago, they faced a choice that many manufacturing companies still grapple with today: build a proprietary data model or adopt an open standard. The path they chose, and the lessons learned along the way, offer valuable guidance for anyone building digital twin solutions in manufacturing.

Erich Barnstedt, Chief Architect for Standards Consortia and Industrial IoT at Microsoft Azure, has spent the last ten years working directly with manufacturers on this challenge. His team's experience reveals why certain approaches work at scale while others create problems down the road.

‍

‍

Using Open Standards for Industrial Asset Data Models

‍

Microsoft could have created their own data model for industrial assets. Many vendors do. But their customers told them clearly they didn't want vendor lock-in. They needed a data model based on open standards.

OPC UA emerged as the obvious choice. It's now the de facto standard for industrial asset modeling, with a large ecosystem of companies providing tools and connectivity solutions. What started as a technical decision became a market differentiator because it gave customers the freedom they wanted.

For manufacturing data leaders, this points to a broader principle: your data model choice affects not just your current project but your ability to scale across sites and integrate with partners. An open standard gives you options. A proprietary model creates dependencies.

‍

Data Normalization at the Edge for Multi-Site Manufacturing Operations

‍

Here's where Microsoft's experience gets particularly relevant for large manufacturers. They initially tried to bring all the data together in the cloud and create a unified model there. It didn't work well, especially at scale.

The better approach: normalize all data to your data model on the edge. By the time data reaches the cloud, it's already in a consistent format. This lets you run analytics across all assets and sites worldwide without wrestling with data transformation in the cloud.

Think about what this means for your operations. If you have twenty factories running different equipment from different vendors, edge normalization means each site handles its own data translation. Your cloud systems receive standardized data they can immediately use. The alternative is managing twenty different data formats in your cloud analytics, which gets messy fast.

The process works like this: First, send your data model to the cloud and store it. Then send the data along with references to that model. Your systems know exactly what they're working with and can run queries accordingly.

‍

Essential Features of Industrial Data Models

‍

Not all data models are equal for industrial applications. Based on Microsoft's work with manufacturers, here's what you need:

It needs to be easy to work with. This depends partly on the ecosystem. OPC UA has many companies providing data modeling tools, which helps.

It must be serializable and fully featured. Your model should handle tags, alarms, events, methods, and references. This lets you create a graph structure with references in multiple dimensions.

You need good mapping tools. This is critical. Many industrial connectivity providers offer OPC UA servers that can map from other models to OPC UA. Your edge components can do this mapping automatically, which is why the ecosystem matters.

Microsoft is now starting a new working group in the OPC Foundation to standardize not just the data models and telemetry formats, but also the configuration interfaces for mapping software and connectivity tools. Right now, every vendor has their own configuration interface. Standardizing this will make it easier to integrate different tools.

‍

Using OPC Publisher for MQTT PubSub Integration

‍

Most OPC UA servers don't support MQTT PubSub out of the box yet. To bridge this gap, Microsoft created OPC Publisher, an open-source adapter that connects to OPC UA servers, converts data to PubSub format, and sends it to the cloud. It's been around almost ten years and is widely used in production environments.

This type of adapter solves a practical problem: you need to work with existing systems while moving toward newer protocols. Rather than waiting for all your equipment vendors to update their systems, you add a translation layer that works now.

‍

Mapping ISA-95 to Digital Twin Definition Language

‍

Microsoft's Manufacturing Ontologies Reference Solution shows how to apply these concepts in practice. They started with a familiar framework - ISA-95 - that manufacturing people already understand. Then they mapped over 100 different models from ISA-95 to Digital Twin Definition Language (DTDL) and made it all open source.

The reference solution keeps growing based on customer questions. When someone asks how to integrate a specific component, Microsoft adds it to the repository. Recently, they added tools for modeling material properties, process descriptions, and actors that interact with production lines.

This approach makes sense: start with what the industry knows, add modern capabilities, and share it so everyone can build on it. You're not inventing a new way to think about manufacturing. You're using established models in a more flexible format.

‍

Integrating OPC UA Data Models with Mixed Reality Applications

‍

Microsoft's work also extends to mixed reality and the industrial metaverse. Their key insight: don't remodel your assets for augmented or virtual reality applications. Import your existing OPC UA models directly into your metaverse applications.

They've created reference architectures showing how to go from physical assets to digital twins to mixed reality applications using the same data models throughout. This matters for remote maintenance, production planning, and training scenarios.

‍

Practical Guidance for Implementation

‍

Barnstedt's advice for solution architects comes from years of seeing what works and what doesn't:

Look at open source projects first. There are many good open-source projects for industrial applications. Check what exists before building from scratch.

Join existing initiatives. Rather than going alone, build on top of existing assets and data models. You'll save time and money.

Pause before reinventing. When you see something that's almost identical to what you're planning, stop and consider it. Creating incompatibilities takes years to fix later.

Use standards consortia. Groups like the OPC Foundation and Digital Twin Consortium have real experts who've probably already solved problems similar to yours. This reduces risk.

‍

Key Considerations for Digital Twin Implementation in Manufacturing

‍

The pattern here is clear: use open standards, normalize at the edge, build on existing frameworks, and participate in the broader ecosystem. These aren't just theoretical best practices. They're lessons from implementing digital twin solutions with large manufacturers over many years.

Your data model choice sets the foundation for everything that comes after. Choose based on openness, ecosystem support, and proven scale. The time invested in getting this right pays back in every subsequent project.

The tools and standards exist. The reference implementations are open source. The community is active. The question isn't whether you can build effective digital twin solutions in manufacturing. It's whether you'll build them in a way that scales with your business and integrates with your partners.

‍