An intelligent intrusion detection system for external communications in autonomous vehicles

Advancements in computing, electronics and mechanical systems have resulted in the creation of a new class of vehicles called autonomous vehicles. These vehicles function using sensory input with an on-board computation system. Self-driving vehicles use an ad hoc vehicular network called VANET. The network has ad hoc infrastructure with mobile vehicles that communicate through open wireless channels. This thesis studies the design and implementation of a novel intelligent intrusion detection system which secures the external communication of self-driving vehicles. This thesis makes the following four contributions: It proposes a hybrid intrusion detection system to protect the external communication in self-driving vehicles from potential attacks. This has been achieved using fuzzification and artificial intelligence. The second contribution is the incorporation of the Integrated Circuit Metrics (ICMetrics) for improved security and privacy. By using the ICMetrics, specific device features have been used to create a unique identity for vehicles. Our work is based on using the bias in on board sensory systems to create ICMetrics for self-driving vehicles. The incorporation of fuzzy petri net in autonomous vehicles is the third contribution of the thesis. Simulation results show that the scheme can successfully detect denial-of-service attacks. The design of a clustering based hierarchical detection system has also been presented to detect worm hole and Sybil attacks. The final contribution of this research is an integrated intrusion detection system which detects various attacks by using a central database in BusNet. The proposed schemes have been simulated using the data extracted from trace files. Simulation results have been compared and studied for high levels of detection capability and performance. Analysis shows that the proposed schemes provide high detection rate with a low rate of false alarm. The system can detect various attacks in an optimised way owing to a reduction in the number of features, fuzzification.