Return to SCADA
Returning to an old SCADA security project, building reproducible industrial datasets, and testing what useful analysis can be done with small local models.
A few years ago I spent a considerable amount of time looking at the security of industrial control systems.
This project is a return to some of that work: rebuilding the experiments with better datasets, more reproducible tooling, and a particular interest in what can be achieved using relatively small local models.
Why revisit it?
The original work answered some questions, but it also left plenty of interesting ones behind.
In particular:
- Can industrial protocol datasets be built in a more reproducible way?
- How useful are small models when the problem is tightly constrained?
- How little hardware can we realistically get away with?
- What changes when we test against real equipment rather than purely synthetic data?
The plan
The first stage is deliberately simple.
Build clean baseline traffic, capture it consistently, introduce controlled changes, and create datasets that can be inspected and reproduced.
From there, the interesting part begins: seeing what useful classification and analysis can be performed locally without requiring enormous models or expensive hardware.
More soon.