Electromagnetic Risk Management for Dependability of Road Vehicles using Discrete Bayesian Networks

The analysis of functional safety, cybersecurity and other risks has become an integral part of the development of modern road vehicles, which are increasingly reliant on the correct functioning of lectrical/electronic/programmable electronic (E/E/PE) systems. Electromagnetic interference (EMI) originating from on-board and off-board radio frequency (RF) sources are well-known common cause of failures or malfunctions in E/E/PE systems. Although vehicle electromagnetic compatibility (EMC) requirements take some account of safety issues, they are mainly concerned with interoperability and their separate and independent development means that they are not directly connected with the more recently developed interests in other dependability aspects e.g., cybersecurity. From the system assessor’s perspective, functional safety issues caused by EMI have mostly been believed to be handled by legislative EMC measures. However, this position is highly questionable for systems that contain new technologies that are not considered in existing EMC standards.

A unified approach that enables EMC risk management for the wider aspects of vehicle resilience as well as for functional safety and cybersecurity would increase the efficiency of the additional analyses by promoting the sharing and reuse of information. Based on the identified challenges for adopting a risk-based EMC approach, the application of probabilistic graphical models called Bayesian network (BN) is proposed in this thesis for two main purposes. First, to graphically model the epistemic uncertainties considered for electromagnetic (EM) immunity risk assessment, second, to overcome the limitations of assessing EMI risks due to multiple disturbances that can be simultaneously present (multitone) in the system EM environment. Two new multi-causal effect prediction models are proposed in this thesis to predict the failure probability of E/E/PE systems due to multitone EMI. The proposed models are verified with experiments to have an enhanced prediction accuracy when compared to the existing models in literature.