Manufacturing Part 3: The Data Foundation
04.06.2026
The “AI Adoption” Series: Where We Are
-
Part 1 (Strategy): We defined the outcomes (Asset Maximization, Labor Augmentation).
-
Part 2 (Team): We aligned IT and OT to bridge the “air gap.”
Now we arrive at the plumbing. In manufacturing, “data” is often physically trapped inside a machine that was built before the iPhone existed. It is locked in proprietary PLCs (Programmable Logic Controllers), written on paper clipboards, or stranded in a standalone SCADA system.
To build an AI-ready factory, we must move from Data Silos to a Unified Namespace. We must turn the factory from a collection of isolated machines into a single, cohesive organism.
The Industry Reality: The “Hidden Factory”
The average factory generates massive amounts of data, but very little of it reaches a decision-maker.
-
The Usage Gap: It is estimated that 68% to 73% of industrial data goes unused, often referred to as “Dark Data” (Zuar / Seagate).
-
The Silo Cost: Data silos are not just annoying; they are expensive. Gartner estimates that poor data quality and silos cost the average organization $12.9 million annually in lost productivity and bad decisions (Gartner).
-
The “Purdue” Problem: Traditional factories use the “Purdue Model” (a rigid hierarchy where the Machine talks to the SCADA, which talks to the MES, which talks to the ERP). This creates a game of “Telephone” where data is lost or delayed at every hop. AI needs raw, real-time data, not filtered summaries.
The Strategic Imperative:
Your goal is to flatten the stack. You need an architecture where a vibration sensor on a lathe can talk directly to a cloud analytics model without having to ask permission from three other servers first.
The Strategy Template: The Unified Namespace (UNS)
To fix the plumbing, SMB manufacturers are moving to a Unified Namespace (UNS) architecture.
Think of the UNS as a “Central Nervous System” for your factory. Instead of machines talking to each other one-by-one (a mess of cables), everyone talks to the UNS.
1. The Edge: Wrap, Don’t Replace
You do not need to buy new machines to get data.
-
The Problem: Your 1995 stamping press has no ethernet port.
-
The Fix: Use Edge Gateways. These are small, industrial PCs (costing <$1,000) that physically clamp onto the machine’s existing wiring. They read the electrical signals (e.g., “Motor Current High”) and translate them into digital data.
-
The Outcome: The old machine is now “smart” without touching its internal control logic.
2. The Structure: The Single Source of Truth
-
The Problem: The PLC calls it “Tag_101.” The ERP calls it “Asset_A5.” The Maintenance Log calls it “Line 1 Press.” The AI has no idea these are the same thing.
-
The Fix: The Unified Namespace. This is a centralized directory that organizes all data into a standard hierarchy (usually Enterprise / Site / Area / Line / Cell).
-
The Outcome: When the AI asks for “Line 1 Press Temperature,” it gets the answer instantly, regardless of what the sensor is actually named.
3. The Protocol: Publish/Subscribe (MQTT)
-
The Problem: Traditional systems “poll” machines (asking “Are you okay?” every second). This clogs the network.
-
The Fix: Use MQTT (Message Queuing Telemetry Transport). This is a “Report by Exception” protocol. The machine stays silent until something changes. When the temperature spikes, it “Publishes” the data to the UNS.
-
The Outcome: Network traffic drops by 80-90%, allowing you to connect thousands of sensors without crashing your Wi-Fi.
The Underpinning: Context is King
This is where the Governance underpinning becomes critical.
-
Raw Data is Useless: A number like “400” means nothing. Is it 400 degrees? 400 RPM? Is 400 good or bad?
-
The Rule: You must enforce Contextualization at the Edge.
-
The Execution: You do not send “400” to the cloud. You send a structured packet (Sparkplug B standard) that says:
-
Value: 400
-
Unit: Degrees Celsius
-
Timestamp: 12:01:05 PM
-
Quality: Good
-
Machine State: Running
-
The Direction: Event-Driven Architecture
We are moving from Polling (checking the machine) to Events (listening to the machine).
-
Current State: A manager looks at a report at the end of the shift to see why production was low.
-
Future State: The moment a machine’s temperature deviates from the standard, it publishes an “Event.” This event triggers a text message to the maintenance lead and an update to the scheduling software simultaneously.
Next Step: Making the Data Speak
You now have a “Connected Factory.” Your machines are talking to the UNS, and the data is clean and structured.
But data doesn’t fix machines. Insights fix machines.
In Manufacturing Part 4, we will discuss Analytics & Machine Learning. We will cover how to use this stream of real-time data to predict equipment failures days in advance and automate quality control using computer vision.
Salvatore Magnone is a father, veteran, and a co-founder, a repeat offender in the best way in fact, and a long-time collaborator at DOOR3. Sal builds successful, multinational, technology companies and runs obstacle courses. He teaches business and military strategy at the university level and directly to entrepreneurs and military leaders.
