Project Outcomes Report

Title:   CPS: Synergy: Security of Distributed Cyber-Physical Systems with Connected Vehicle Applications;

Our research focuses on improving the resiliency of cyber-physical systems (CPSs). In particular we focused on two different aspects: network communication attacks and malicious agents.

As part of the first aspect, sampled data, packet losses, and variable network delays are taken into account as network imperfections that are naturally in networked control systems and that are influenced and accentuated by the presence of cyber-attacks. We developed two different resilient approaches. The first approach is based on the development of separate types of resilient control strategies, one for each type of attack, that are available and selected by a supervisory decision maker based on the real-time solution of an optimization-like problem whose objective is to maintain an acceptable system performance level. The second approach instead is based on the modeling of the CPS as a hybrid dynamical system and the design of a hybrid controller that is robust to all the above attacks while maintaining a prescribed performance.

We considered the vehicle platooning application as a case study employed to validate our theoretical contributions. In particular, we proposed new design approaches capable of making Cooperative Adaptive Cruise Controls more resilient to packet dropouts, variable network delays, and Denial-of-Service (DOS) attacks. To determine the effectiveness of the proposed approaches, the resilient control strategies were evaluated for different driving cycles against a set of metrics such as fuel consumption loss, number of collisions, loss of comfort, percent of performance loss of distributed controller versus theoretical centralized controller, and false positive/false negative detection rate. The analysis shows that our mitigation strategies designed for each specific attack perform quite well if the correct attack is identified by completely avoiding collision and improving comfort and fuel consumption. The decentralized controller compared to the centralized one causes a percentage loss in fuel economy that is well within our targets. The hybrid controller instead focuses on a design that maximizes the resiliency to DOS attacks. Such a resiliency is measured through the number of consecutive packet dropouts that the control strategy can overcome without losing string stability. When transmission intervals of 50 ms are considered, vehicle platooning with hybrid control is resilient to 5 consecutive dropouts compared to standard control resilient to only one dropout. The hybrid controller performs better in general than the switching controller both from comfort and fuel consumption aspects.

The outcomes of our research further contribute to advance intelligent transportation systems with new control algorithms for improved resiliency for Connected and Automated Vehicles. To this end, our research is expected to impact the security and reliability of CPSs by helping accelerate the adoption of resilient controllers. Given the general-purpose nature of the control theory approach, the contributions developed in this research will have a vast spectrum of applications, especially those with limited computational and communication capabilities.

The second aspect of the research dealt with agents in the CPS that purposely behave maliciously. Within this context, game theory analysis based robust controller design was considered. A comparison was conducted between centralized and decentralized control strategies under the presence of different types and levels of measurement noise. For the validation of the approach, we developed a simulation for autonomous vehicle platooning. The platoon simulation is used to simulate user-defined test cases. Our game theory-based analysis shows that actuator noise has more impact on the fuel consumption and number of crashes than the sensor noise. The centralized controller provides better fuel efficiency, whereas decentralized controller performs well in collision avoidance. Kalman filter performs well compared to other countermeasures. For any malicious behavior, there is at least one control and filtering strategy that can mitigate or avert influence and minimize the probability of collisions. The developed approach has a general-purpose nature and can be used in other engineering applications for robust control design such as smart grid security and flight controllers.

Last Modified: 02/28/2020
Modified by: Pierluigi Pisu