Beyond Topic Design
Most discussions of Unified Namespace focus on topic hierarchy - Enterprise / Site / Area / Line / Cell. This is necessary but insufficient. A well-designed topic tree solves addressing. It does not solve data acquisition, contextualization, or reliability.
A functioning UNS requires:
- Protocol-native ingestion that preserves source fidelity
- Automatic topic generation from hierarchy configuration
- Consistent schema with metadata (timestamp, quality, provenance)
- Decoupled producers and consumers with guaranteed delivery
The topic hierarchy is the contract. The data plane underneath it is what makes the contract enforceable.
ISA-95 in Practice
ISA-95 provides the hierarchical model: Enterprise, Site, Area, Line, Cell. In practice, mapping operational assets to this hierarchy requires decisions that are specific to each facility.
A welding robot on Line 3, Station 7 maps cleanly. A shared utility system that serves multiple lines does not. Cross-cutting concerns - energy, environmental, safety - require conventions that ISA-95 does not prescribe.
The hierarchy must be configured, not discovered. And the system that manages this configuration must generate consistent addressing across every protocol and destination.
The Hard Problems
Consistent schema. Every data point must carry the same metadata structure regardless of source protocol. OPC UA provides rich metadata natively. Modbus provides none. The data plane must normalize this.
Automatic topic generation. Manually mapping thousands of tags to topic paths does not scale. Topics should be derived from the hierarchy configuration and source definitions.
Delivery guarantees. A UNS without buffering and replay is a UNS that loses data during network events. Industrial environments have unreliable networks by definition.
Implementation Approach
Implement UNS at the data plane level - not at the broker level. The data plane acquires data natively, applies ISA-95 context, generates topics automatically, and delivers to the broker with full buffering. The broker receives structured, contextualized data. It does not have to create it.