In modern vehicles, the controller area network (CAN) bus is one of the most important means of communication between electronic control units (ECUs). Despite its speed and simplicity, the critical drawback of the CAN bus is the security problem. In t...
In modern vehicles, the controller area network (CAN) bus is one of the most important means of communication between electronic control units (ECUs). Despite its speed and simplicity, the critical drawback of the CAN bus is the security problem. In the CAN bus, the ECUs send and receive messages based on broadcasting and priority mechanisms without authentication and encryption. Hence, the system is vulnerable to various attacks. To address the problem, this thesis introduces three deep learning-based approaches for designing a robust intrusion detection system (IDS) in the CAN bus.
Firstly, this thesis proposes a semi-supervised deep learning model, named convolutional adversarial autoencoder (CAAE), to tackle the issue of lacking labeled attack data, which commonly occurs in the intrusion detection problem. Concretely, the model is trained with unlabeled data to learn the manifolds of normal and attack patterns. Then, only a small number of labeled samples are used in supervised training. The proposed model can detect various kinds of message injection attacks, such as DoS, fuzzy, and spoofing, as well as unknown attacks. The experimental results show that the proposed model achieves the highest F1 score of 0.9984 and a low error rate of 0.1% with limited labeled data compared to other supervised methods. Moreover, the model can meet the real-time requirement by analyzing the model complexity in terms of the number of trainable parameters and inference time. This study successfully reduced the number of model parameters by five times and the inference time by eight times, compared to a state-of-the-art model.
Next, the thesis aims to design a universal IDS which can be applied to all car models. As a result of the different distribution of car models, the goal is difficult to achieve. Therefore, the thesis utilizes transfer learning techniques on a supervised contrastive model. In detail, we proposed a novel deep learning model called supervised contrastive (SupCon) ResNet, which can handle multiple attack classifications on the CAN bus. Furthermore, the model can be used to improve the performance of a limited-size dataset using a transfer learning technique. The capability of the proposed model is evaluated on two real car datasets. When tested with the car hacking dataset, the experiment results show that the SupCon ResNet model improves the overall false-negative rates of four types of attack by five times on average, compared to other models. In addition, the model achieves the highest F1 score on both the vehicle models of the survival dataset by utilizing transfer learning. Finally, the model can adapt to hardware constraints in terms of memory size and running time to be deployed in real devices.
Finally, we extend the transfer learning idea in the previous study by developing a personalized federated learning scheme to aggregate all the datasets from different car models. The proposed model can provide global knowledge that can be transferred to any model’s participant to increase its performance. In comparison to baseline models, the proposed model achieves 4% higher F1 scores overall. Besides, the results also suggest that the proposed model provides significant advantages when the local dataset of each participant is relatively small. According to our experiments, the proposed models can achieve F1 scores of more than 90% with at least 30k training samples on each client.