Cybersecurity and Digital Forensics for Mechatronic Systems

This book includes a graph convolutional network approach to adaptive intrusion detection for water treatment SCADA systems. The convergence of operational and information technology has dissolved a boundary that once did useful work. A modern machine tool, robot cell, or clinical device is commanded through software interfaces, supervised through remote dashboards and updated over the same networks that carry business traffic, and the consequence is that an intrusion into a mechatronic installation is no longer merely an informational event. A compromised supervisory interface does not only leak data; it can move an axis, alter a chemical dose or silence an alarm. This book addresses that exposure across four applied studies. Two concern detection. The first evaluates explainability for machine-learning intrusion detection, comparing SHAP and LIME and treating temporal validation as a first-order concern, since a detector assessed without regard to event ordering appears far more capable than it is. The second applies graph convolutional networks to intrusion detection in water treatment SCADA, exploiting the relational structure of a network in which devices, signals and control loops form a graph rather than independent data streams. The remaining chapters concern response and architecture. One compares traditional security operations centres with an agentic artificial intelligence approach in which an autonomous agent triages, investigates and contains without waiting on human dispatch, grounded in an implemented scenario on the kind of host commonly used as a supervisory node. The other designs a secure, locally deployed electronic records backend with NFC-based identification, treating the backend as the trust anchor for every connected device that depends on it.