Talk · interactive deck

The Unified Namespace: one live picture of your factory

How an event-driven architecture turns shopfloor data into a shared, real-time asset — with a live demo you can play with.

Nitish Singh Data Scientist, KMWE · Assistant Professor (part-time), TU/e — OPAC · AIMM
Live from AIMM Precision’s demo line
The problem

Shopfloor data is trapped in layers

L4 · BusinessERP
L3 · OperationsMES · QMS · CMMS
L2 · SupervisorySCADA · HMI
L1 · ControlPLCs
L0 · ProcessMachines · sensors
The ISA-95 automation pyramid
  • Data climbs one layer at a time, through custom interfaces.
  • Every new question — a dashboard, an AI use case — becomes a new integration project.
  • Nobody has the full, current picture, so people walk to the machine or pick up the phone.
The problem, quantified

Point-to-point integration doesn't scale

Systems 8
Point-to-point · n(n−1)/2 28 interfaces to build, test and maintain
Hub · n 8 connections — one per system

Adding one more system costs 8 new interfaces point-to-point — or exactly one with a hub.

The idea

What is a Unified Namespace?

A single, real-time hub where every system publishes what it knows and subscribes to what it needs — organised the way the business is organised.

Single source of current truth

The latest value of everything, in one place.

Event-driven

Publish when something changes — not when someone polls.

Structured like the business

An ISA-95 hierarchy that humans and machines can both navigate.

Open and lightweight

MQTT and plain payloads — no single-vendor lock-in.

A UNS is an architecture, not a product you buy.

Structure

A topic is an address in your business

aimm-precisionEnterprise/ eindhovenSite/ machiningArea/ line-1Line/ cnc-01Cell/ temperatureData point

Readable by humans

Anyone can guess where a value lives — no tag lists, no lookup tables.

Navigable by machines

Subscribe to a whole line, or one data point across every site, with a single wildcard.

The message (payload)

{
  "value": 47.3,
  "unit": "°C",
  "ts": "2026-11-14T09:41:07.412Z"
}
How it moves

The plumbing: MQTT publish / subscribe

PLC / edge gateway
MES
Quality app
Publishers
Broker routes by topic · keeps last value
Dashboards
Historian / lakehouse
AI agents
Subscribers

Retained messages

Late joiners instantly get the last known value — that is the “current state”.

Report by exception

Publish only on meaningful change. The demo shows the traffic saved, live.

Payloads

Plain JSON (simple, flexible) or Sparkplug B (typed, with birth/death certificates).

Tool · try it

Design your namespace — and watch it come alive

Data points
Naming convention
0Topics
0Messages / s
0%Saved vs polling
Inspector 
  
 
 
 
  
  
 
Live demo · 1
 
Subscribe to exactly what you need
 
 
 
   
 
 
 
 
 
+
exactly one level
 
#
everything below (last level only)
 
 
 
—Matching topics
 
0Messages / s
 
 
 
 
 Messages received by this subscription
 
 
 
  
  
 
Live demo · 2
 
One event, many consumers — zero new integrations
 
 
 Producers · click to inject a fault 
 
 
 UNS broker 0 msg / s 
 
 
 
Andon board0
  
 
 
 
Availability service0
  
 Publishes back to …/line/kpi/availability 
 
 
Maintenance agent0
  
 
 
   
 
 
  
  
 
Data engineering
 
The UNS is the stream, not the store
 
 
 
UNScurrent state, events
 
→
 
IngestMQTT connector / stream
 
→
 
 Historian · lakehouse Bronze — raw events Silver — cleaned, typed Gold — KPIs, features 
 
→
 
UseBI · ML · agents
 
 
 
 
Context travels with the data
 
The topic path is free metadata: site, area, line and machine — no lookup joins.
 
 
 
Replay from the lake
 
Backfills and model training read history from storage, not from the broker.
 
 
 
Payload = data contract
 
temperature:
  value: number   # required
  unit:  "°C"     # fixed per data point
  ts:    ISO-8601 # source timestamp, UTC
  quality: good | uncertain | bad
 
 
 
  
  
 
Agentic AI
 
Agents need context in — and governed actions out
 
 
 
OperatorHumans and agents: roles, competencies, accountability
 
InteractionManufacturing services, tools, authorisation
 
DigitizationReusable factory information — the UNS lives here
 
Cyber-PhysicalMachines, PLCs, sensors
 Layers of the AIMM Framework for Manufacturing Intelligence 
 
 
     
  • Agents subscribe for context — no bespoke integration per agent.
    •  
  • They act by publishing to governed command topics, never by reaching into machines.
    •  
  • Consequential actions need human approval, as in the demo.
    •  
  • Every action is itself an event — auditable by design.
    •  
 
 
  
  
 
Reality check
 
What goes wrong on a real shopfloor
 
 
Naming is governance
Agree the structure and who owns each branch before you scale.
 
It isn't a database
History belongs in a historian or lakehouse, not in the broker.
 
Payload chaos
Units, timestamps and quality flags need a contract from day one.
 
Security per topic
TLS, authentication and access rules for who may publish and subscribe where.
 
Boiling the ocean
Start with one line and one use case; let the namespace grow with value.
 
People, not just tech
Operators, IT and OT must own it together, or it becomes another silo.
 
  
  
 
Takeaways
 
Three things to remember
 
 
1
Integrate once, to the hub — not to each other.
 
2
Structure it like the business, so people and machines can find data.
 
3
Run it as a product: contracts, governance and history alongside.
 
  
  
 
 
 
Thank you
 
Questions?
 
Try the demo yourself, and get in touch:
 
     
  • Webnitishsingh.com
    •  
  • Decknitishsingh.com/slides/uns
    •  
  • LinkedInlinkedin.com/in/nitish-singh
    •