University Lectures Track
Physically-based process invariants for the identification of process anomalies and detection of cyber-physical attacks
Lecturer: Ombretta Paladino (Universita di Genova)
Thursday 19th March, 15:30 – 17:30
The correct approach to detect and reduce cyber-attacks must involve not only control system engineers and sw/hw/communication engineers, but also the process engineers. Process and plant design should actually include the dynamic design of the process, i.e. the definition of the main rules that holds during the correct operation of a plant in each given state.
These rules must contain the measurable state variables, also combined among them by first principles (mass and energy balances) to produce invariants.
Methods that use state-based invariants can then be implemented into the control system to identify the deviation of the plant from its correct behavior. In this way distributed attack detection is more feasible.
Training course content:
– Type of invariants (thresholds, error based, predicted values, mixed state-variables + state of the actuators, physically based);
– Local and global invariants (steady-state invariants, start-up /shut-down invariants, stage invariants);
– Classical methods to define invariants: design centric (P&ID and physically-based) or data-centric (time series from pilot / dynamic simulation and black-box models, NN, machine learning)
A new proposed approach:
– Use of both dimensionless numbers and response times as invariants;
– Use of Lyapunov exponents to detect anomalies due to attacks during transient process behavior.