SEC-04 / 7 guides
Data foundations: governed data readiness for industrial AI
Most industrial AI projects that fail, fail on data. This section treats plant data as the control surface for AI risk: who owns it, where it flows, how good it is, and who can touch it.
Most industrial AI projects do not fail because the model was wrong. They fail because the data feeding it was wrong, late, missing context, or owned by nobody. A vibration model trained on a historian with aggressive compression. A scrap predictor that never knew which product was running. A vision QC system labeled by one inspector on day shift and judged by a different inspector on nights. None of these are algorithm problems. They are data problems, and they are governance problems.
That is why data foundations sit inside an AI governance site and not next to it.
Governed data is the control surface for AI risk
When you write an AI policy or run an AI risk assessment, most of the controls you can actually enforce live in the data layer:
- Ownership. Someone signs off on what a tag means, what a label means, and who may use it.
- Lineage. You can trace a model's input back to the sensor, PLC, historian and transformation that produced it.
- Quality. Timestamps, units, calibration and missing context are checked before a model sees them, not after it misbehaves.
- Access. Read and write rights are explicit. Nobody publishes a setpoint into the plant because a dashboard made it easy.
- OT exposure. Every new data path out of Level 2 or Level 3 is a new conduit, and it gets reviewed like one.
- Retention. You know how long raw and derived data are kept, and why.
If those six things are weak, no amount of model documentation will make the AI system trustworthy. If they are strong, most AI risk becomes manageable with the small team you already have.
What this section covers
The pages here walk through the data decisions that mid-market plants actually face:
- Moving OT data to a cloud historian without opening inbound paths into the control network.
- The Unified Namespace in plain terms, and who should be allowed to publish to it.
- Predictive maintenance programs, where the gap is usually failure history, not sensors.
- Digital twins that earn their keep, and how to avoid vanity twins.
- Industrial data quality: drift, deadbands, timestamps and labels.
- Edge vs cloud AI as a governance and security choice, not just an architecture choice.
- The industrial data readiness checklist to score where you stand today.
Each page stays vendor-neutral and assumes a lean IT/OT team, brownfield equipment and OEM-owned assets. Where a product category matters, we name well-known examples without endorsing them.
Where to start
If you have one afternoon, run the industrial data readiness assessment and pick the two lowest-scoring dimensions. Then cross-check them against your AI use case inventory. The use cases that depend on your weakest data are the ones to pause, fix or scope down first.
Guides in this section
SEC-04.01
Digital twins for mid-market manufacturers: data, governance and security
A digital twin is only useful when it answers a specific operational question at the fidelity that question needs. Govern it like any engineering model: versioned…
6 min read
SEC-04.02
Edge vs cloud AI in manufacturing: a governance and security decision guide
Run AI at the edge when it needs fast, deterministic responses, high-volume raw data, or must keep working without the internet. Use the cloud for training, cross-site…
6 min read
SEC-04.03
Industrial data quality for analytics and AI: what breaks and who fixes it
Industrial AI fails most often on data quality problems that are specific to plants: clocks that disagree, sensors that drift, historians that discard detail, missing…
6 min read
SEC-04.04
Industrial data readiness for AI: a 40-point checklist with scoring
Before you fund an industrial AI use case, score your data against 40 checks in eight dimensions. The lowest-scoring dimension, not the average, tells you what will…
6 min read
SEC-04.05
Moving OT data to a cloud historian: architecture, security and governance
Move plant data to the cloud through a one-way, outbound-only path from a DMZ, add context before it leaves the site, and treat every new data flow as a governed conduit…
7 min read
SEC-04.06
Predictive maintenance programs: data, validation and governance
Predictive maintenance works when you have trustworthy failure history, the right sensors on the right assets, a validated model, and a named person who acts on each…
7 min read
SEC-04.07
Unified Namespace (UNS) explained for mid-market manufacturers
A Unified Namespace is a single, structured, real-time view of plant data that every system publishes to and reads from, usually over MQTT. It works when topic structure…
7 min read
TEMPLATES
Free templates for this topic
AI use case register for manufacturers (Excel template)
A ready-to-use Excel register for listing every AI system in your plants, including AI hidden inside vendor equipment and SaaS tools, with…
Free download
AI risk triage worksheet (Excel)
Answer twelve yes, no or partial questions per use case and the worksheet computes a risk tier, flags possible EU AI Act high-risk or…
Free download
OT and AI vendor security questionnaire (Excel)
A structured questionnaire to send to OEMs, integrators and AI vendors before purchase or renewal, covering OT security, remote access…
Free download
Not sure where to start?
Take a 10-minute self-assessment and get a prioritised 30/60/90-day plan for your plants.