Fig. 1

We demonstrate a stealthy attack scenario on a reactor’s temperature sensor (XMEAS(9)). Here, PASAD framework is delayed in detecting the attack because of the projection of attacked measurements towards the loose side of decision boundary. Subfigure a shows the sensor-generated measurements. The green and black measurements are the normal measurements used for training and validation, respectively, while the red measurements are captured under a stealthy attack (SA3). Subfigure b represents the departure score of corresponding measurements generated by PASAD frameworks. Subfigure c demonstrates the projections of each normal and attack measurement on the signal subspace (we consider a \(2\)-dimensional signal subspace for better visualization) and the PASAD’s Decision Boundary (PDB)