Adversarial CAN Message Generation for Vehicle IDS Evasion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current intrusion detection systems (IDS) in vehicles are vulnerable to adversarial example attacks, which can be evaded by generating and inserting noise into CAN messages, causing the attack to disappear and lose its intended meaning when inserted into a real vehicle.
Innovation Solution
An adversarial attack apparatus and method that collects CAN messages, extracts relevant data, adds type information, and inserts noise based on class and subclass information using a preprocessing module, generating an adversarial CAN message capable of avoiding IDS through an artificial neural network, and restoring the message to its original form for insertion into a vehicle.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If noise is inserted into CAN messages to generate adversarial examples, then the IDS detection accuracy deteriorates, but the attack meaning disappears and the message cannot be effectively inserted into real vehicles
Solution Approach 1:
The patent applies local quality by selectively inserting noise only into specific non-critical data fields of CAN messages while preserving critical fields such as arbitration ID and control fields. This targeted approach allows the adversarial example to evade IDS detection in specific areas without compromising the overall attack meaning or message functionality when inserted into real vehicles.
Solution Approach 2:
The patent segments the CAN message into critical and non-critical fields, applying different noise insertion strategies to each segment. Critical fields remain unchanged to preserve attack meaning, while non-critical fields receive noise to achieve IDS evasion, thus resolving the contradiction between detection accuracy and attack effectiveness.
2Measurement precision
If machine learning-based IDS is used to detect hacking attacks, then the detection accuracy improves, but the system becomes vulnerable to adversarial example attacks
Solution Approach 1:
The patent implements preliminary anti-action by pre-processing CAN messages with targeted noise insertion before they reach the IDS. This anticipatory modification creates adversarial examples that are designed to fool the machine learning-based IDS, thereby demonstrating the system's vulnerability to such attacks before deployment.
Solution Approach 2:
The patent changes the parameters of CAN messages by inserting noise into specific data fields, transforming normal messages into adversarial examples. This parameter modification allows the messages to bypass the machine learning-based IDS while maintaining their ability to execute attacks in real vehicle systems.
3Adaptability or versatility
If noise is modulated when inserting adversarial examples into real vehicles, then the message format adapts to the vehicle system, but the attack meaning disappears and other values change
Solution Approach 1:
The patent applies local quality by identifying and protecting critical message fields from noise modulation while allowing adaptation in non-critical fields. This ensures that the arbitration ID, control fields, and other essential components remain unchanged to preserve attack functionality, while peripheral data fields can be modulated to adapt to the vehicle system's message format requirements.
Solution Approach 2:
The patent performs preliminary analysis of the vehicle's CAN message format requirements before inserting adversarial examples. By pre-identifying which fields must remain unchanged and which can be adapted, the system ensures that noise modulation does not compromise attack functionality while still achieving necessary format compatibility.
Data Source
AI summary
An apparatus and a method for an adversarial attack on vehicle's controller area network (CAN). The apparatus comprises: a data generation module for collecting a plurality of CAN messages, extracting preconfigured some data from the plurality of collected CAN messages, adding type information for the CAN messages to the extracted some data so as to generate CAN message packets, and gathering the generated CAN message packets so as to configure a CAN message packet data set; a pre-processing module for inserting noise into some CAN message packets of the CAN message packet data set, wherein the noise is inserted on the basis of the type information of each CAN message packet; and an adversarial attack generation module for receiving the CAN message packet having noise inserted thereinto so as to generate an adversarial CAN message that can evade an intrusion detection system (IDS) of the vehicle.