Cooperative adaptive cruise control (CACC) is a promising technology allowing connected and automated vehicles to be driven in close formations known as platoons.  The benefits of platoons include a reduction in fuel consumption due to less drag, improved traffic throughput and less demand on the driver. However, platoons require a reliable communication flow between vehicles to ensure safety. Communication disruption or loss may result in multiple vehicle collisions with catastrophic consequences.

While communication failures may be attributed to system malfunctions, their root cause can frequently lie in the systems flawed operation within unpredictable and uncontrollable environmental and operating conditions. To address these functional insufficiencies the international standard ISO21448:2022 deals with Safety of the Intended Functionality (SOTIF) of Road Vehicles. The safety standard uses a scenario-based approach to identify potentially hazardous triggering conditions. The outputs then contribute to functional modifications (including artificial intelligence) to enhance the overall safety. 

This thesis resolves the critical challenge to ensure safety in a vehicular platoon under communication disruptions. The research problem is focussed on analysing the unpredictable scenarios leading to communication disturbance and thereafter implement safety features to mitigate such occurrences.  Systems Theoretic Process Analysis (STPA) is the preferred hazard and risk assessment tool to identify causal factors leading to unsafe control actions. Corner case simulations subjected on a platoon control system (variable packet loss, network topology and network size) provide useful data to train machine learning models with the aim of mitigating hazardous by assessing and acting upon these operational parameters.         

The primary contribution of this work is a highly interpretable machine learning model demonstrating the potential to reduce platoon collision risks by up to 98%. This research contributes to science by providing a highly interpretable data driven model to monitor the operational environment and apply and additional layer of safety. Beyond this, the research offers a detailed performance evaluation across various interpretable machine learning algorithms and introduces specific recommendations to improve robustness in line with the latest state of the art standard ISO/PAS:8800:2024. By doing so, this work provides a solid platform for subsequent research in the creation of resilient and reliable AI-based safety solutions.