Machine Learning Based Anomaly Detection on ARINC-429 Data Bus: A Simulation-Based Approach

ARINC-429 is an avionics communication protocol widely used in both commercial and military aircraft. While traditional rule-based validation methods used in ARINC-429 are sufficient for detecting large hardware failures, they are inadequate against logical anomalies in the message content. In this study, we present a comprehensive simulation-based machine learning framework for detecting logical errors in ARINC-429 traffic. In the initial phase, we synthesized protocol-compliant ARINC-429 traffic. Subsequently, utilizing a configurable anomaly injection module, we systematically introduced both standard and complex logical anomalies into the data stream. We created a hybrid feature matrix encompassing bit-level features, moving statistical descriptors, and temporal derivatives of the traffic we simulated. On this synthetic dataset, we trained a Random Forest classifier, optimizing the model parameters for maximum performance. Experimental results demonstrate that our approach achieves high accuracy and recall rates across various logical failure scenarios. In conclusion, this study offers a data-driven method that enhances system validation processes by effectively detecting latent logical errors in avionics systems utilizing the ARINC-429 data bus.