Interpretable Intrusion Detection for Automotive Ethernet Networks : A Logical Neural Network Approach for Symbolic Reasoning

Sammanfattning

The rapid integration of Automotive Ethernet as the backbone for modern vehicle communication systems has introduced unprecedented advantages in speed and scalability for Advanced Driver Assistance Systems (ADAS) and autonomous driving technologies in in-vehicle Networks (IVNs). However, this technological advancement has also increased the cybersecurity risks, as the Automotive Ethernet’s uniquearchitecture demands specialized security solutions. Intrusion Detection Systems (IDS) are critical to mitigating these threats, yet the demands of the automotive domain including real-time constraints, reliability, and adherence to safety standards set them apart from IDS inother domains. Traditional IDS approaches often rely on deep learningmodels that, while accurate, lack interpretability, posing challengesfor safety-critical automotive applications where explainability is essential. However, eXplainable artificial intelligence (XAI) provides a pathway to achieve transparency, enabling stakeholders to understand, validate, and trust system outputs. But this thesis studies a novel Neuro-Symbolic approach to intrusion detection by integrating Logical Neural Networks (LNN) ’a form of explainable AI developed by IBM’ into the IDS framework for Automotive Ethernet, by combining the pattern recognition strengths of neural networks with the logical reasoning capabilities of symbolic AI. The proposed framework aims to address the dual challenges of detection performance and explainability. The results demonstrate that LNN achieve competitive detection accuracy while significantly enhancing interpretability, offering a promising pathway toward trustworthy and transparent IDS in modern vehicular networks.