TrustConnect: An In-Vehicle Anomaly Detection Framework through Topology-Based Trust Rating

TL;DR

TrustConnect introduces a topology-based trust framework to assess in-vehicle network reliability by computing per-ECU and overall network trust from ECU dependencies and outside-exposure risks, addressing remote injection threats without relying on data signatures or frequency analyses. The method defines a directed graph of ECUs, with propagated influence quantified via distance-based weights and a decay factor, and combines these with an exposure-driven trust aggregation to yield ECU and network scores. Experimental simulations using NetworkX demonstrate how exposure and inter-ECU relationships shape trust dynamics, revealing which ECUs are more robust and how parameters like $\alpha$ influence trust propagation. The work suggests a practical, resource-light approach for IDS design and reliability assessment in modern connected vehicles, where data integrity can be compromised through remote injections.

Abstract

Modern vehicles are equipped with numerous in-vehicle components that interact with the external environment through remote communications and services, such as Bluetooth and vehicle-to-infrastructure communication. These components form a network, exchanging information to ensure the proper functioning of the vehicle. However, the presence of false or fabricated information can disrupt the vehicle's performance. Given that these components are interconnected, erroneous data can propagate throughout the network, potentially affecting other components and leading to catastrophic consequences. To address this issue, we propose TrustConnect, a framework designed to assess the trustworthiness of a vehicle's in-vehicle network by evaluating the trust levels of individual components under various network configurations. The proposed framework leverages the interdependency of all the vehicle's components, along with the correlation of their values and their vulnerability to remote injection based on the outside exposure of each component, to determine the reliability of the in-vehicle network. The effectiveness of the proposed framework has been validated through programming simulations conducted across various scenarios using a random distribution of an in-vehicle network graph generated with the Networkx package in Python.

Figures (4)

• Figure 1: In-Vehicle ECU Dependency Graph: Each ECU is dependent on other ECUs, with a corresponding weight representing the strength of the dependency.
• Figure 2: Simulated In-Vehicle Network Dependency Graph
• Figure 3: Simulated Remote Attack Difficulty of Individual ECU- 1: Low chance of attack, 0: high chance of attack
• Figure 4: Comparison of the Baseline Trust Value, Trust Value, and ECU Adjusted Trust Value under different combinations of the k and $alpha$ (denoted as a) values