Connecting Legacy SCADA Systems to Modern AI
Nobody is ripping out a SCADA system that's been running for 15 years. Here's how to layer modern AI on top of existing industrial infrastructure.
The legacy reality
Let's be honest: most industrial control systems are older than the smartphones in our pockets. Siemens WinCC installations from 2008, ABB 800xA systems running on Windows Server 2012, OSIsoft PI historians that nobody wants to touch because they "just work."
And they should keep working. The question isn't whether to replace them — it's how to extract more value from the data they already collect.
The integration approach
SUPi connects to legacy systems using industry-standard protocols that these systems already support:
| System | Protocol | Connection Method | |--------|----------|-------------------| | Siemens WinCC / PCS 7 | OPC-UA, OPC-DA | Native connector | | ABB 800xA | OPC-UA | Native connector | | Emerson DeltaV | OPC-UA, OPC-DA | Native connector | | Yokogawa CENTUM | OPC-UA | Native connector | | Honeywell Experion | OPC-UA | Native connector | | OSIsoft PI | PI SDK / PI Web API | Historian connector |
All connections are read-only. SUPi never writes back to your control systems.
Security considerations
Connecting AI to OT networks raises legitimate security concerns. Here's how SUPi addresses them:
- DMZ deployment — SUPi can run in a network DMZ between IT and OT
- Encrypted protocols — OPC-UA with X.509 certificates
- Minimum permissions — Read-only service accounts
- Network segmentation — Supports unidirectional data diodes
- Audit logging — Every data request is logged
What you need to get started
- A list of tag names for the assets you want to monitor
- Network access from the SUPi instance to your data sources (OPC-UA port 4840 or historian API)
- Service account credentials with read-only access
- At least 30 days of historical data for initial model training
Most connections are live within 1–2 days. The historical backfill for model training typically takes another 1–2 days depending on data volume.