Sensor and Process Fingerprinting in Industrial Control Systems
summary
Experiments run on the SWaT water treatment testbed use ultrasonic, flow and pressure sensors and show sensor and process noise can fingerprint devices and flag spoofing. A mathematical process model produces residuals; supervised learning (SVM) classifies those residual chunks of about 2-minute chunks (≈120 samples). Tests across up to 60 identical sensors reach about 99% accuracy for static conditions and ~96% when using residuals on dynamic processes, detecting flat and many stealthy injections in live runs.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:45Presentation
- 7:47Why do we need defenseindepth
- 13:57Virtual Reality
- 14:16Demonstration
- 15:27Cryptography
- 18:56Stealth
- 19:56Invariants
- 20:49Sensor Noise
- 26:09Noise Detection
- 27:58Machine Learning
- 30:58Summary
- 32:18Residual
- 33:20Flat Noise
- 37:41Security Evaluation
- 38:22Architecture
- 41:39Takeaways
related talks
07mNot Quite Yourself Today: Behaviour-Based Continuous Authentication in IoT Environments
04mToward Physical-Layer Fingerprinting of Production Automotive...
27mSpecification Mining for Intrusion Detection in Networked Control Systems
47mMiniaturization
45mAnalogy Cyber Security—From 0101 to Mixed Signals
11mIs Your ECU Really Yours?
01h 05mTheory and Security
30mThe Motion Sensor Western: The Good, the Bad, and the Ugly