Overarching Properties-Based Arguments for the Safety Assurance of an Ai-Assisted Runway Alignment System

Artificial Intelligence/Machine Learning (AI/ML) techniques can be useful for autonomous or semi-autonomous operation of aircraft under diverse circumstances. Such safety-critical applications of AI/ML warrant adequate safety assurance before they are certified for operation. Moreover, because of the varying degrees of opacity and complexity in the design of non-trivial AI/ML-based aerospace applications, safety assurance must holistically consider the low-level behavioral and design elements of AI/ML-based components with respect to system-level hazards. However, existing certification standards and guidelines do not support systems utilizing AI/ML. We present an Overarching Properties (OP)-based approach for developing safety arguments that can be used to claim that an AI/ML-based component will be safe when used in the context of a more complex aircraft system. As a motivating use case for the study, we introduce an AI-Assisted Autonomous Runway Alignment (AARA) System that uses Artificial Neural Networks (ANN) to help align the aircraft nose to the runway center line after touchdown. We use the AARA to identify potential safety concerns and mitigate them through appropriate premises and assumptions in the safety arguments. The results show promise in the applicability of our OP-based approach for the holistic safety assurance of AI/ML-based complex aerospace systems.