الفهرس | Only 14 pages are availabe for public view |
Abstract The Controller Area Network (CAN) bus is a critical component in modern vehicles, acting as the backbone for internal communications between various electronic control units. Ensuring the integrity and reliability of CAN bus communications is paramount, especially in the context of increasing vehicle automation and connectivity. This study presents a novel approach to achieving a zero false negative rate in CAN bus systems using a Long Short-Term Memory (LSTM) autoencoder combined with embedded machine learning techniques. We propose a LSTM-based autoencoder model that can be embedded directly into the CAN bus hardware. The autoencoder is trained to recognize normal communication patterns, enabling it to detect anomalies indicative of potential faults or cybersecurity threats. By leveraging the temporal pattern recognition capabilities of LSTM networks, our system excels at identifying subtle and complex anomalies that traditional detection methods might overlook. Our experimental setup involves a simulated CAN bus environment with various types of injected faults and attack scenarios. The LSTM autoencoder model is trained on a dataset of normal operation data, and its performance is evaluated based on its ability to detect anomalies without generating false negatives. The embedded machine learning aspect ensures real-time data processing and analysis, critical for immediate response in vehicular systems. The results demonstrate a remarkable achievement of a zero false negative rate, while maintaining a low false positive rate. This balance is crucial for automotive applications where missing a real threat (false negative) could have severe consequences, but overreacting to normal variations (false positive) could lead to unnecessary disruptions. Our approach signifies a substantial advancement in vehicular communication security and reliability. It not only enhances the safety features of modern vehicles but also serves as a stepping stone towards more advanced applications such as autonomous driving. This research contributes a significant methodological improvement to the field of automotive cybersecurity, paving the way for more secure and reliable vehicle communication systems. |