AI Enabled Intrusion Detection System in a Digital Twin Based Vehicular Cyber-Physical System

Dr. Sunitha’s project, "AI Enabled Intrusion Detection System in a Digital Twin Based Vehicular Cyber-Physical System (VCPS)", harnesses advanced artificial intelligence to secure autonomous vehicular networks from sophisticated cyber threats, particularly False Data Injection (FDI) attacks that conventional systems cannot detect.

This innovative model integrates an Asynchronous Federated Learning (AFL) framework with a Gated Recurrent Unit (GRU) neural network optimized for sequential data processing. The asynchronous AFL enables adaptive, real-time learning by continuously updating the global model based on device responses, ensuring timely detection and management of threats within the VCPS.

Central to this approach is the Digital Twin Model (DTM), a virtual replica of the VCPS that simulates real-world driving scenarios and potential cyber-attacks. The AI-powered DTM generates adversarial samples of synthetic data representing possible attack patterns to train the system, enhancing its resilience to evolving threats. This adversarial training process strengthens the model’s robustness, allowing it to accurately identify FDI attacks across diverse vehicular data sets.

The model has demonstrated significant success, achieving high accuracy in detecting cyber threats while preserving user privacy by sharing only model parameters, not raw data. This privacy-preserving design fosters collaborative learning among distributed network entities, enhancing security without compromising sensitive information.

The impact of this research is underscored by the filing of a patent by Howard University, reflecting its innovative contributions to the field of cybersecurity. Additionally, this work was partly supported by VMware Inc., Meta Inc., Microsoft Corp. Research Gift Funds, and the NSF, highlighting a strong commitment from the industry to advance cybersecurity in autonomous vehicular systems. Through its AI-driven architecture, Dr. Sunitha’s project combines adaptive learning, robust intrusion detection, and privacy preservation, setting a new standard for securing digital twin-enabled cyber-physical systems.

This model has been reviewed and accepted by international professors as part of the International Conference on Computer, Software, and Modeling (ICCSM) peer review process, and this was presented at the Paris conference in July 2024.