PROJECT
Procedural Industrial Environment Simulator
A declarative framework for generating realistic, reproducible synthetic industrial environments with process behaviour, operational history, human activity and exact ground truth.
RESEARCH QUESTION
Does procedurally generated operational diversity improve out-of-domain generalisation in industrial control system anomaly detection?
>project notes
Overview
The Procedural Industrial Environment Simulator is an experimental framework for generating synthetic industrial environments that are complicated for the same reasons real industrial environments are complicated.
The aim is not simply to create a network containing simulated industrial devices.
The aim is to generate reproducible industrial worlds containing process behaviour, operational routines, human activity, maintenance, configuration drift, faults, technical debt and exact ground truth.
Why it exists
Industrial and OT machine-learning research often relies on datasets that represent a narrow operational environment.
Models can therefore perform extremely well on familiar data while generalising poorly when the process, topology, implementation, schedule or legitimate engineering behaviour changes.
This project provides the experimental apparatus for investigating whether procedurally generated operational diversity can improve that generalisation.
Current direction
The immediate focus is deterministic, seeded virtual-time behaviour: generating realistic operational activity while retaining complete reproducibility.
The next step is explicit event provenance so that observable behaviour can be traced back to the process, operator, maintenance activity, fault or attack that caused it.