Machine Learning based Theoretical Framework for Failure Prediction, Detection, and Correction of Mission-Critical Flight Software

Abstract

Mission-critical flight software acts as the control mechanism for autonomous flights and lies at the heart of next-generation developments in the aviation industry. Most state-of-the-art technological evolution is realized through the use of contemporary software which implements the essentially required, novel, innovative, and featuring value additions. Real-time physical exposure and the data-driven flying nature of aerial vehicles make them vulnerable to an ever-evolving new threat spectrum of cyber security. Nation or state-sponsored cyber attacks through sensors’ data corruption, hardware Trojans, or counterfeit wireless signals may exploit dormant and residual software vulnerabilities. It may lead to severe and catastrophic consequences including but not limited to serious injury or death of the crew, extreme damage or loss to equipment and environment. We have proposed a machine learning based theoretical framework for real-time monitoring and failure analysis of autonomous flight software. It has been introduced to protect the mission-critical flight software from run-time data-driven semantic bugs and exploitation that may be caused by missing, jammed, or spoofed data values, due to malicious online cyber activities. The effectiveness of the proposed framework has been demonstrated by the evaluation of a real-world incident of grounding an aerial vehicle by the actors in their vicinity without the intent of the original equipment manufacturer (OEM). The results show that the reported undesired but successful cyber attack may has been avoided by the effective utilization of our proposed cyber defense approach, which is targeted at software failure prediction, detection, and correction for autonomous aerial vehicles.